JP2024155698A - A reservoir area landslide deformation monitoring and early warning method based on sliding zone strain evolution - Google Patents

Abstract

The present invention relates to landslide monitoring and early warning, and in particular to a reservoir area landslide deformation monitoring and early warning method based on sliding zone strain evolution.
A reservoir area landslide deformation monitoring and early warning method based on sliding zone strain progression is disclosed, which includes the steps of: installing a high spatial resolution strain sensing fiber cable through a vertical borehole in the landslide subzone, remotely acquiring the underground strain distribution of the entire landslide borehole in real time, identifying the sliding zone position, and taking the sliding zone strain as a parameter for characterizing the overall landslide deformation; and considering four influencing factors such as daily precipitation, etc., to establish a prediction and warning model based on landslide strain progression. The four influencing factors are input variables, and a prediction rule corresponding to the characteristic deformation cluster whose target output variable is acceleration deformation is obtained, which is the hydrometeorological warning threshold of the landslide acceleration deformation.
[Selected Figure] Figure 1

Description

The present invention relates to landslide monitoring and early warning in the field of engineering geology, and in particular to a reservoir area landslide deformation monitoring and early warning method based on sliding zone strain evolution.

Existing landslide deformation monitoring techniques include non-contact scanning (e.g., airborne InSA)
R, ground-based InSAR) can provide time series of surface deformation at regional or hillslope scales, but deformation results are generally used for trend analysis.
It is difficult to quantify the deformation. GNSS displacement monitoring stations provide highly accurate surface deformation information at the monitoring points, but the monitoring points are discrete.
Large-area landslides in super-large reservoir areas are challenging, and therefore, we need to rely on borehole monitoring to understand the landslide kinematic evolution and provide efficient and reliable early warning.
Among the deformation monitoring technologies that use boreholes, conventional sliding tiltmeters provide accurate deep deformation but often fail within 3 to 5 years due to excessive deformation during flood seasons.
Fixed tiltmeters have obvious advantages in terms of durability and automation, but due to technical and cost constraints, typically an array of three fixed tiltmeter sensors is installed in a borehole in series, which provides a very limited insight into the relative deformation at depth and is unable to quantitatively characterize deformation information at important geological interfaces such as the crust. Distributed fiber optics (DFOS) and quasi-dispersive fiber Bragg gratings (FBG) are
It has shown great potential in geotechnical health monitoring to reflect deep deformation by monitoring strain. The former provides high spatial resolution (sub-meter level) but low level of automatic monitoring and warning, while the latter allows remote real-time monitoring but has low spatial resolution to obtain accurate strain distribution (e.g. F in borehole).
BG sensors are limited in number (usually less than 10 sensors). Therefore, in large-scale landslides involving the development and progression of multiple sliding zones, where the thickness of the sliding zone ranges from tens of centimeters to several meters, the spatial resolution of the sensor is the key to identifying the sliding zone.
Prior art lacks methods that can characterize the true kinematic behavior of global landslides, while traditional remote real-time landslide monitoring not only cannot comprehensively reveal the underground evolution process of landslides, but also lacks the timeliness and effectiveness of early warning prediction based on surface displacement.

ここで、ＧＲ（Ｓ,Ｔ）はデータセットＳ中の特徴属性Ｔの情報利得率であり、ＩＶ（Ｔ
）は属性Ｔのエントロピーであり、ＩＧ（Ｓ,Ｔ）は属性Ｔの情報利得である。属性Ｔの
値が大きいほど、ＩＶ（Ｔ）の純度が低くなり、すなわちＩＶ（Ｔ）の値が大きくなり、
ＧＲ（Ｓ,Ｔ）の値が小さくなる。これは、ある属性の値が多く取られた場合懲罰メカニ
ズムを与える。同時に、正しく分類されなかった属性により高い重みを割り当て、後続の
訓練モデルがそれらを優先するようになる。予測警報モデルの成長過程は出力変数に基づ
く入力変数の連続的なグループ化過程である。予測警報モデルの各枝は入力データセット
の連続的なグループ化過程で徐々に成長したものであり、データセットの継続的なグルー
プ化が意味をなさなくなった時点で予測警報モデルの成長過程は終了する。
サンプルサブゾーンを７：３または８：２の比率で訓練集合とテスト集合に分け、反復回
数をＮに設定することでサンプルを繰り返して訓練してモデル予測の安定性を高め、最終
的にテストサンプルにより生成された予測警報モデル精度が約６０％～７０％に達したと
き新しいデータ対象（テストデータセット）の分類予測を達成する。
生成された予測警報モデルは新しいデータ入力変数の値に基づいてその出力変数の値、す
なわち歪み速度集合結果、（例えばＣｌｕｓｔｅｒ １）を推論する。以上のように、生
成された予測警報モデルに基づいて地すべり地下歪み進展と外部駆動因子（降雨量、貯水
池水位）の閾値基準を確立する。
全体として、本発明のスライディングゾーン歪み進展に基づく貯水池地域の地すべり変形
監視および早期警報方法は、活動的な地すべりサブゾーンに垂直ボーリング孔を通じて高
い空間分解能の超微弱反射型格子歪みセンシングファイバーケーブルを取り付け、遠隔で
地すべりボーリング孔全体の地中歪みをリアルタイムで取得して地すべり変形を特徴付け
、スライディングゾーン位置を識別してスライディングゾーン歪みを地すべり全体的な変
形を特徴付けるパラメータとして使用し、日降雨量Ｒ_ｉ、日降雨強度Ｉ_ｉ、貯水池水位標
高Ｌ_ｉおよび日水位変動ｆ_ｉの４つの影響因子を考慮し、地すべり歪み進展に基づく予測
警報モデルを確立する。前記の４つの影響因子を入力変数とし、スライディングゾーンの
時系列歪みをその進展特徴集合に基づいて近似安定状態、加速変形および安定状態という
３つの特徴変形クラスタＣｌｕｓｔｅｒ ｊに分解して出力変数とし、目標出力変数が加
速変形である特徴変形クラスタに対応する予測規則を見つけ、すなわち地すべり加速変形
の水文気象学的警報閾値とする。本発明で採用された高空間分解能歪みセンシングファイ
バーケーブルはスライディングゾーンの正確な位置決めおよびスライディングゾーン歪み
変化の捕捉を実現し、スライディングゾーン歪み進展に基づく予測警報モデルはより信頼
性の高い水文気象学的警報閾値を取得し、貯水池地域の地すべり深部大変形精准監視の耐
久性が不十分であり、地面変形に基づく警報誤警報率が高いという問題を解決することが
できる。
先行技術と比較すると、本発明は以下の有益な効果を有する。
（１）サブメートルスケールの高空間分解能歪みセンシングファイバーケーブルは、潜在
的なスライディングゾーンを正確に特定し、スライディングゾーン変形を捕捉することが
でき、スライディングゾーン歪み進展に基づく予測警報モデルはより信頼性の高い水文気
象学的警報閾値を取得でき、貯水池地域の地すべり深部大変形精准監視の耐久性が不十分
であり、地面変形に基づく警報誤警報率が高いという問題を解決することができる。
（２）本発明で採用された超微弱反射型格子歪みセンシングファイバーケーブルは、遠隔
で地すべり地下全ボーリング孔深さの歪み分布および進展過程をリアルタイムで監視でき
、特に複数の薄い層スライディングゾーンの識別に適し、せん断作用下でのすべり加速度
運動がすべりゾーンでの歪みピーク特性として現れる。
（３）ＦＢＧと比較し、本発明は超微弱反射型格子に基づいて取得された歪みデータ精度
が１～２μεに達し、空間分解能が０．５ｍを超え、ＦＢＧよりも高く、長距離、サブメ
ートルスケールの高空間分解能の正確な監視を実現し、ＤＦＯＳと比較すると、本発明の
自動化程度が高く、システムコストが低く、遠隔でリアルタイムな監視を実現する。
（４）本発明のスライディングゾーン歪みの変形監視は貯水池地域の地すべり変形の長期
間監視に適用でき、地すべりの実の運動状態を反映し、地面変形の警報に比べて、地下変
形（歪み）の進展に基づく早期リスク警報方法は精度が高く、実用性が高く、誤警報率を
大幅に低減し、頻繁な「過剰警報」を回避することができる。
（５）本発明が提供する貯水池地域の地すべり警報方法は複数の影響因子（降雨量、降雨
強度、貯水池水位標高および水位変動）を同時に考慮し、従来の貯水池地域の地すべり早
期リスク警報閾値判定よりも合理的である。 Objective of the invention: In order to solve the problems of insufficient durability of accurate monitoring of deep large deformation of landslides in reservoir areas, low level of remote real-time automation, and high false alarm rate in existing technologies, the present invention proposes a reservoir area landslide deformation monitoring and early warning method based on sliding zone strain evolution, and identifies potential sliding zones by characterizing the actual deformation of the entire landslide by time series strain of the entire underground depth (i.e., the depth to bedrock), and further establishes hydrometeorological alarm thresholds based on sliding zone strain evolution by considering multiple influencing factors, thereby improving the long-term monitoring and effective alarm level of landslides in reservoir areas; the high spatial resolution strain sensing fiber cable adopted in the present invention can accurately identify and capture the sliding zone dynamics, and the alarm model based on sliding zone strain evolution can obtain more reliable hydrometeorological alarm thresholds, which can solve the problems of unclear multi-level sliding zone identification of landslides in reservoir areas, insufficient durability of deep large deformation refinement monitoring, and high false alarm rate based on ground deformation.
Technical solution: The landslide deformation monitoring and early warning method for reservoir areas based on sliding zone strain evolution of the present invention includes the following steps:
S1, set up a monitoring station:
First, drill a number of holes extending below the stable rock layer in the landslide subzone (a landslide subzone is a landslide divided into multiple subzones based on the spatial distribution of deformation and risk at the mountainside scale), and install an ultra-weak reflective grating strain sensing fiber cable, an ultra-weak reflective grating demodulator and a data transmission unit on the ground at the hole mouth, the data transmission unit is remotely connected to a user access unit, the data transmission unit is a wireless transmission module in the prior art, and is integrated into the ground monitoring station protection box together with the ultra-weak reflective grating demodulator, and the user access unit is a computer;
Next, install an inverted "convex"-shaped ultra-weak reflection type lattice strain sensing fiber cable in the hole, connect a gravity guide hammer to the ultra-weak reflection type lattice strain sensing fiber cable, place the gravity guide hammer at the bottom of the hole, introduce one end of the ultra-weak reflection type lattice strain sensing fiber cable into a protective box and connect it to an ultra-weak reflection type lattice demodulator;
the other end of the ultra-weak reflection type grating strain sensing fiber cable as a backup collection end;
The gravity guide hammer is fixed to the bottom of the hole to pretension the optical fiber, and is redundantly placed at the bottom of the hole by 0.5 m to keep the fiber cable always in a vertical pretensioned state.
The other end of the ultra-weak reflection grating strain sensing fiber cable is led out from the hole mouth, the hole bottom is the boundary, and the ultra-weak reflection grating demodulator is connected to the ultra-weak reflection grating demodulator. Section A is a section that extends redundantly from the hole mouth by a certain length at the hole bottom. Section B is a section that extends redundantly from the hole mouth by a certain length at the hole bottom.
The borehole is backfilled, and the gravity guide hammer connected to the middle of the ultra-weak reflection type lattice strain sensing fiber cable is lowered to the bottom of the borehole, and then both ends of the ultra-weak reflection type lattice strain sensing fiber cable extended from the borehole mouth are fixed and kept in tension, and then the borehole is backfilled with different rock and soil materials according to the ground distribution characteristics revealed by the borehole,
S2, remote real-time monitoring: debugging the ultra-weak reflection grating demodulator, the data transmission unit and the user access unit, and collecting time series strain data sets of rock and soil bodies at various depths underground through the user access unit;
Debugging refers to connecting each module, powering it on, and debugging whether data can be collected.
In step S1, the backfill material is backfilled up to the hole mouth and then compacted; the end of section A of the ultra-weak reflection grating strain sensing fiber cable extending from the hole mouth is introduced into a protective box on the ground near the hole mouth, and connected to an optical fiber jumper to connect the collection end of the ultra-weak reflection grating demodulator; then the ultra-weak reflection grating demodulator, data transmission unit and user access unit are debugged; after debugging, the backfill material is left to stand for one month to solidify and stabilize, which is the initial monitoring period; and then long-term remote real-time monitoring activities are started to establish a time series strain data set of rock and soil bodies at different depths underground;
S3. Sliding zone identification and time series distortion set analysis:
In step S2, based on the time series strain data set obtained by the user access unit, a cross-sectional strain evolution map of the entire borehole is drawn to obtain a strain distribution, the area with the highest strain value in the strain distribution is the strain peak zone, the strain peak zone is defined as a sliding zone, the average value of each time series strain data for each day for the sliding zone is defined as a daily strain average value, the daily strain average value is subjected to differential processing (the differential processing refers to obtaining the difference between the cumulative strain of the next day and the cumulative strain of the previous day, i.e., the strain rate), and a daily average data set of strain rate is obtained, the sliding zone has a characteristic that the strain increases rapidly during slope acceleration deformation, and the daily average data of strain rate is classified into three classification results based on the change characteristics: an approximate stable state, an accelerated (strain fluctuates rapidly) deformation, and a stable state;
At this time, the numerical variables are converted into type variables (Cluster j, j=1,2,3), i.e., the output variables of the alarm model, and the method adopted in the set analysis is "data backtracking".
According to the principle, noise errors in the strain rate time series are eliminated, that is, discontinuous strain changes (singular values) are designated as possible noise data, and the zero region strain is the steady state, and is within ±300μ
ε is a critical value between near-steady state and accelerated deformation, above which the set is an accelerated deformation section, and below which the set is a near-steady state section;
S4, Establish a forecasting and warning model and threshold criteria: The three classification results obtained in step S3 are output variables, and the daily value datasets of rainfall R _i , rainfall intensity I _i , reservoir water level elevation L _i and water level fluctuation f _i are input variables to establish a forecasting and warning model, where the forecasting model is a decision tree model (prior art), and infers the value of the output variable according to the value of the input variable, establishes the threshold criteria of the landslide underground strain progression and external driving factors, and provides early warning of the landslide deformation in the reservoir area;
The external drivers include rainfall amount, rainfall intensity, reservoir water level elevation, and reservoir water level fluctuation;
The threshold criteria is:
Criterion 1: The water level drop rate exceeds 0.88 m/d and the reservoir water level elevation is between 146.45 m and 14
9.505m and the daily rainfall exceeds 57.9mm, accelerated deformation will occur, that is, the reservoir area landslide is an accelerated deformation section.
The accuracy reached 90.68%.
Criterion 2: When the reservoir water level elevation is below 169.7m and the rainfall intensity is above 24.4mm/h, accelerated deformation occurs, that is, the reservoir area landslide is an accelerated deformation section. After verification by actual example, the accuracy rate reaches 74.27%;
Criterion 3: When the reservoir water level elevation is below 169.7m, the daily rainfall exceeds 1.8mm, and the rainfall intensity exceeds 24.4mm/h, accelerated deformation occurs, that is, the reservoir area landslide is an accelerated deformation section. After verification by actual example, the accuracy reaches 70.48%;
Criterion 4: When the reservoir water level elevation is below 169.7m, the rainfall intensity is more than 0.8mm/h, and the daily rainfall is more than 57.9mm, accelerated deformation occurs, that is, the reservoir area landslide is an accelerated deformation section. After verification by actual example, the accuracy rate reaches 61.3%;
Criterion 5: When the water level rising speed exceeds 0.14 m/d, the reservoir water level elevation is less than or equal to 150.195 m, and the daily rainfall exceeds 57.9 mm, accelerated deformation occurs, that is, the reservoir area landslide is an accelerated deformation section. After verification by actual example, the accuracy is 80.34%.
reaches.
Specifically, the sample data (i.e., the input variables and output variables) are divided into a training set and a test set in a ratio of 7:3, and the decision tree model is repeatedly trained by the input of the training set. When the accuracy of the decision tree model generated by the test set reaches 60%-70%, the new data object (test set) is classified and predicted. The generated forecast and warning model predicts the output variable value Cluster j corresponding to the input variable based on the value of the input variable, and is shown in the form of a series of "if-then" statements for understanding and application. The "if-then" statements are prediction rules, including deformation categories, multiple influencing factors and specific hydrometeorological thresholds. As described above, the threshold criteria for landslide acceleration deformation and different hydrometeorological conditions are set based on the correlation between the developing sliding zone distortion and external driving factors (such as rainfall amount, rainfall intensity, reservoir water level elevation and reservoir water level fluctuation). For example, under what reservoir water level elevation L _i and what rainfall intensity I _i conditions, the acceleration deformation Cluster j This provides an early warning of whether (j=2) will occur.
Divide the sample data in the forecasting and warning model into a training set and a test set, set the number of iterations,
The predictive warning model is repeatedly trained using a training set as input, and a classification prediction is made for the test set when the accuracy of the decision tree model generated using the test set reaches a set range, and the generated decision tree model infers the value of the output variable based on the value of the input variable.
In one aspect of the present invention, the hole is divided into an intact rock section, a crushed rock section, and a crushed soil section, from bottom to top, and the height of the intact rock section is 3 to 5 m;
In step S2, the complete rock section is backfilled with cement quick-setting agent, the crushed rock section is backfilled with fine sand, and the crushed soil section is backfilled with fine sand and crushed clay in a volume ratio of 1:1. In principle, the actual soil is filled in sections according to the type of rock-soil body in order to approximate the actual soil distribution.
In one aspect of the present invention, the ultra-weak reflective grating strain sensing fiber cable comprises an ultra-weak reflective grating and an optical fiber transmission section.
In one aspect of the present invention, the sampling interval d of the ultra-weak reflection type grating strain sensing fiber cable is d≦1.0 m.
In step S1, the collecting end and the backup collecting end are redundantly arranged by 2 to 5 m near the mouth of the well to assist in accurate positioning of the underground depth, and to facilitate identification of the sliding zone, especially in thin layers (thickness of several tens of cm);
In step S1, the ultra-weak reflective lattice strain sensing fiber cable is pre-tensioned, and the lattice spacing between the collection end and the backup collection end is 0.5 m when projected onto the borehole cross section.
It is.
In one aspect of the present invention, in step S3, the sliding zone strain rate is qualitatively classified based on the characteristics of the time-varying sliding zone strain rate, the characteristics of the time-varying sliding zone strain rate including a slight change in strain rate, a sudden change in strain rate, and no change in strain rate;
The classification standards are as follows:
When the strain rate fluctuates slightly, i.e., when the strain rate ranges from 50 to 200 με/d, the movement state of the entire landslide is approximately stable.
When the strain rate changes rapidly, i.e., when the strain rate exceeds 200 με/d, the movement state of the entire landslide is accelerating deformation.
When the strain rate does not fluctuate, that is, when the strain rate ranges from 0 to 50 με/d, the movement state of the entire landslide is stable.
Working principle: The predictive warning model adopted in step S4 of the present invention uses an improved decision tree model to learn from historical data with clear results, explore the features of the data, and further infer simple decision-making rules, and then classify and predict new data based on this. Starting from the root node to the leaf nodes, test the feature attributes of the variables of the classification items;
The value is selected according to its characteristics and output to the corresponding branch, and the category stored in the leaf node is the decision result. A rule is constructed for each path from the root node to the leaf node of the decision tree, and the characteristics of the internal nodes on the path correspond to the condition of the rule, i.e., the prediction rule/alarm condition. The category label of the leaf node corresponds to the conclusion of the rule, i.e., the strain rate aggregate result Cluster j.
The predictive warning model uses a branch-growing algorithm, a variable value splitting algorithm, and a pruning algorithm. The branch-growing algorithm is used to determine the currently optimal grouping variables, with the maximum information gain rate as the standard:

where GR(S,T) is the information gain rate of feature attribute T in dataset S, and IV(T
) is the entropy of attribute T, and IG(S,T) is the information gain of attribute T. The larger the value of attribute T, the lower the purity of IV(T), i.e., the larger the value of IV(T),
The value of GR(S,T) becomes smaller. This provides a punishment mechanism when an attribute is over-valued. At the same time, it assigns higher weights to attributes that are incorrectly classified so that subsequent training models will prioritize them. The growth process of the predictive warning model is a process of successive grouping of input variables based on the output variables. Each branch of the predictive warning model is gradually grown through a process of successive grouping of the input dataset, and the growth process of the predictive warning model ends when the continued grouping of the dataset no longer makes sense.
The sample subzones are divided into a training set and a test set in a ratio of 7:3 or 8:2, and the number of iterations is set to N to repeatedly train the samples to improve the stability of model prediction, and finally, when the accuracy of the predictive warning model generated by the test samples reaches about 60% to 70%, the classification prediction of the new data object (test data set) is achieved.
The generated forecasting and warning model infers its output variable values, i.e., strain rate aggregation results, (e.g., Cluster 1) based on the values of new data input variables. As described above, the threshold criteria of landslide underground strain progression and external driving factors (rainfall, reservoir water level) are established based on the generated forecasting and warning model.
In general, the method for monitoring and early warning of landslide deformation in reservoir area based on sliding zone strain evolution of the present invention installs a high spatial resolution ultra-weak reflective lattice strain sensing fiber cable through a vertical borehole in the active landslide subzone, remotely obtains the underground strain of the entire landslide borehole in real time to characterize the landslide deformation, identifies the sliding zone position and uses the sliding zone strain as a parameter to characterize the overall landslide deformation, considers four influencing factors, namely daily rainfall R _i , daily rainfall intensity I _i , reservoir water level elevation L _i and daily water level fluctuation f _i , and establishes a prediction and warning model based on landslide strain evolution. The above four influencing factors are input variables, and the time series strain of the sliding zone is decomposed into three characteristic deformation clusters, namely approximate stable state, accelerated deformation and stable state, based on its evolution feature set as output variables, and finds a prediction rule corresponding to the characteristic deformation cluster whose target output variable is accelerated deformation, i.e., the hydrometeorological warning threshold of landslide accelerated deformation. The high spatial resolution strain sensing fiber cable adopted in the present invention can realize accurate positioning of the sliding zone and capture the sliding zone strain changes, and the predictive alarm model based on the sliding zone strain evolution can obtain a more reliable hydrometeorological alarm threshold, thereby solving the problems of insufficient durability of precision monitoring of deep large deformation of landslides in reservoir areas and high false alarm rate of ground deformation-based alarms.
Compared with the prior art, the present invention has the following beneficial effects:
(1) The sub-meter-scale high spatial resolution strain sensing fiber cable can accurately identify potential sliding zones and capture sliding zone deformation, and the predictive warning model based on the sliding zone strain evolution can obtain more reliable hydrometeorological warning thresholds, thereby solving the problems of insufficient durability of precision monitoring of deep large deformation landslides in reservoir areas and high false alarm rates based on ground deformation.
(2) The ultra-weak reflective lattice strain sensing fiber cable adopted in the present invention can remotely monitor the strain distribution and progression process of landslides throughout the entire borehole depth in real time, and is particularly suitable for identifying multiple thin-layer sliding zones, where the sliding acceleration movement under shear action appears as strain peak characteristics in the sliding zones.
(3) Compared with FBG, the strain data acquired by the present invention based on ultra-weak reflection grating has an accuracy of 1-2 με, and the spatial resolution exceeds 0.5 m, which is higher than that of FBG, and realizes accurate monitoring with long distance and high spatial resolution on a sub-meter scale. Compared with DFOS, the present invention has a higher degree of automation, a lower system cost, and realizes remote real-time monitoring.
(4) The sliding zone strain deformation monitoring of the present invention can be applied to long-term monitoring of landslide deformation in reservoir areas, and can reflect the actual movement state of landslides. Compared with ground deformation warning, the early risk warning method based on the progression of underground deformation (strain) has higher accuracy and practicability, and can significantly reduce the false alarm rate and avoid frequent "excessive alarms".
(5) The landslide warning method for reservoir areas provided by the present invention simultaneously considers multiple influencing factors (rainfall amount, rainfall intensity, reservoir water level elevation and water level fluctuation), which is more reasonable than the traditional landslide early risk warning threshold judgment for reservoir areas.

FIG. 1 is a schematic diagram showing a landslide monitoring system structure for a reservoir area used in the present invention. FIG. 1 is a schematic diagram of the borehole-mounted in-situ real-time automatic strain monitoring system of the present invention. 1 is a flow chart of a landslide deformation monitoring and warning method for a reservoir area according to the present invention; 1 is a flow chart of data processing during the landslide deformation monitoring and warning process of the reservoir area according to the present invention; Fig. 1 shows the results of monitoring landslide ground displacement and underground strain in a reservoir area under the synergistic action of rainfall and reservoir water level in an experimental example of the present invention, (a) is the reservoir water level and rainfall record, (b) is the ground displacement and sliding zone strain curve. FIG. 13 is a diagram showing the accumulated strain set result of the deep sliding zone SS2 of the embodiment of the present invention. 1 is a diagram showing an example of landslide deformation monitoring and early warning for three hydrological years from 2021 to 2023 using the method for landslide deformation monitoring and early warning in a reservoir area based on the sliding zone strain evolution of the present invention. From top to bottom, the spatiotemporal strain distribution generated by high-resolution optical fiber strain sensing nerve, three significant strain changes (i.e., I, II and III), daily rainfall and water level elevation, daily rainfall intensity peak value and water level fluctuation are shown.

[Explanation of symbols]
1 Landslide subzone 2 River 3 Hole 4 Weak reflection grating strain sensing fiber cable 5 Ultra-weak reflection grating demodulator 6 Data transmission unit 7 Protection box 8 User access unit

表２ 超微弱反射型格子復調器５の技術パラメータ

次に、孔３内で逆「凸」字形の超微弱反射型格子歪みセンシングファイバーケーブル４を
取り付け、図２に示すように、超微弱反射型格子歪みセンシングファイバーケーブル４に
重力ガイドハンマー４３を接続させて前記重力ガイドハンマー４３を孔３の底部に配置し
、超微弱反射型格子歪みセンシングファイバーケーブル４の一端を保護ボックス７内に導
入して超微弱反射型格子復調器５に接続させ、収集端とし、超微弱反射型格子歪みセンシ
ングファイバーケーブル４の他端をバックアップ収集端とし、
超微弱反射型格子歪みセンシングファイバーケーブル４は高分解能の１．ｍサンプリング
間隔の超微弱反射型格子歪みセンシングファイバーケーブルであり、超微弱反射型格子歪
みセンシングファイバーケーブル４は超微弱反射型格子４１および光ファイバー伝送セク
ション４２から構成され、
超微弱反射型格子歪みセンシングファイバーケーブル４のサンプリング間隔ｄ≦１．０ｍ
であり、
収集端とバックアップ収集端が孔口付近に２ｍだけ冗長に配置されて地下深さの正確な位
置決めを補助し、特に薄い層（厚さ数十ｃｍ）のスライディングゾーンの識別に適し、
超微弱反射型格子歪みセンシングファイバーケーブル４を予め引っ張り、収集端とバック
アップ収集端の格子間隔がボーリング孔断面投影において０．５ｍである。
超微弱反射型格子歪みセンシングファイバーケーブル４の構造は裸ファイバ、被覆層、金
属ベースのケーブル状補強材、および外側シースからなる。本実施例中の重力ガイドハン
マー４３は直径５０～６０ｍｍ、長さ６００～１０００ｍｍ、重量２０～３５ｋｇの金属
製である。超微弱反射型格子歪みセンシングファイバーケーブル４において、４１は超微
弱反射型格子であり、４２は光ファイバー伝送セクションであり、
次に孔３を埋め戻しし、超微弱反射型格子歪みセンシングファイバーケーブル４の中部に
接続された重力ガイドハンマー４３を孔底位置に下げた後、孔口から伸びた超微弱反射型
格子歪みセンシングファイバーケーブル４の両端を固定して常に引っ張り状態とし、孔３
を直ちに埋め戻し、孔３は下から上へ順次完全岩盤セクション３１、破砕岩層セクション
３２、砕石土砂セクション３３に分けられ、前記完全岩盤セクション３１の高さは３ｍで
あり、
完全岩盤セクション３１をセメント急結剤（先行技術）で埋め戻して硬質地盤とし、破砕
岩層セクション３２を細砂で埋め戻し、砕石土砂セクション３３を質量比１：１の細砂と
粉砕粘土で埋め戻し、
具体的に、細砂は通常の川砂であり、粉砕粘土はボーリング孔３を掘削した粘土を使用し
、
Ｓ２、遠隔リアルタイム監視：超微弱反射型格子復調器５、データ伝送ユニット６および
ユーザアクセスユニット８をデバッグし、ユーザアクセスユニット８によって地下各深さ
岩石土壌体の時系列歪みデータセットを収集し、
デバッグとは各モジュールを接続した後電源を投入し、データを収集できるかどうかをデ
バッグすることを指し、
ステップＳ２において、孔口まで埋め戻しした後締め固め、孔口から伸びた超微弱反射型
格子歪みセンシングファイバーケーブル４の一端を孔口付近の地面の保護ボックス７内に
導入し、光ファイバージャンパー接続して超微弱反射型格子復調器５の収集端口を接続し
、超微弱反射型格子復調器５と無線伝送モジュール６およびデータユーザアクセスユニッ
ト８をデバッグする。孔３の取付埋め戻しと監視システムのソフトウエアハードウェアデ
バッグ終了あと埋め戻し材料の固化および安定化のために１ヶ月待って、監視初期とする
。その後長期間の遠隔リアルタイム監視活動を開始する。本実施例では、監視データサン
プリング周波数は最大５ｓ／回であり、変形が大きい雨季または洪水期（例えば毎年５～
９月）にサンプリング周波数が５ｓであり、洪水期以外にサンプリング周波数が６０～６
００ｓであり、
Ｓ３、スライディングゾーン識別と時系列歪み集合分析：
ステップＳ２においてユーザアクセスユニット８によって取得された遠隔でリアルタイム
で記録された地下累積時系列歪みデータセットに基づいて、全ボーリング孔歪み断面進展
図を描いて歪み分布を得、前記歪み分布中の歪み値の最も高い領域が歪みピークゾーンで
あり、前記歪みピークゾーンをスライディングゾーンとし、前記スライディングゾーンに
ついて毎日各時系列歪みデータの平均値を歪み日平均値とし、前記歪み日平均値を差分処
理し（前記差分処理とは、翌日の累積歪みから前日の累積歪みを指し差、すなわち歪み速
度を指す）、歪み速度の日平均値データセットを取得し、スライディングゾーンが斜面で
加速変形すると歪みが急激に増加する特徴があり、歪み速度の日平均値データをその変化
特徴に基づいて近似安定状態、加速（歪み急激な変動）変形および安定状態の３つの分類
結果に分類し、（図４に示す）
具体的に、スライディングゾーン歪み速度の経時変化の特徴に基づいて、スライディング
ゾーン歪み速度を定性的に分類し、前記スライディングゾーン歪み速度の経時変化の特徴
は、歪み速度わずかな変動、歪み速度急激な変動および歪み速度不変動を含み、分類標準
は以下のとおりであり：
歪み速度がわずかに変動する場合、すなわち歪み速度範囲が５０～２００με／ｄにある
場合、地すべり全体の運動状態が近似安定状態であり、
歪み速度が急激に変動する場合、すなわち歪み速度が２００με／ｄを超える場合、地す
べり全体の運動状態が加速変形であり、
歪み速度が変動しない場合、すなわち歪み速度範囲が０～５０με／ｄにある場合、地す
べり全体の運動状態が安定状態であり、
なお、集合過程中、高い歪み速度（＞３００με／ｄ）または累積歪み増加＞２００με
である状態が２日以上連続すると加速変形となり、歪み速度が５０～２００με／ｄにあ
り、または累積歪みが急激に増加して急激にリバウンド（単一不連続な高歪み速度孤立点
）すると近似安定状態となり、歪み速度が０～５０με／ｄまたは累積歪みがほとんど変
化しないと安定状態となる。累積歪み集合数は手動で指定するか、または一般的な集合ア
ルゴリズムによって自動的に決定され、
この時点で、数値型変数がタイプ変数（Ｃｌｕｓｔｅｒ ｊ、ｊ=１,２,３）、すなわち
警報モデルの出力変数に変換され、該集合分析で採用される方法は「データバックトラッ
ク」原則に従って歪み速度時系列のノイズ誤差を除去し、すなわち不連続な歪み変化（特
異値）を可能なノイズデータとして指定され、ゼロ領域歪みを安定状態とし、±３００μ
εが近似安定状態と加速変形の臨界値であり、該臨界値を超えると集合が加速変形セクシ
ョンになり、該臨界値未満であると近似安定状態セクションとなり、
Ｓ４、予測警報モデルと閾値基準の確立：
ステップＳ３で得られた３つの分類結果（スライディングゾーン歪み速度の集合結果Ｃｌ
ｕｓｔｅｒ ｊ）を出力変数とし、降雨量Ｒ_ｉ、降雨強度Ｉ_ｉ、貯水池水位標高Ｌ_ｉおよ
び水位変動ｆ_ｉの日値データセットを入力変数として予測警報モデルを確立し、前記予測
モデルは意思決定ツリーモデルであり（先行技術）、入力変数の値に基づいて出力変数の
値を推論し、地すべり地下歪み進展と外部駆動因子の閾値基準を確立し、貯水池地域の地
すべり変形を早期警報し、前記外部駆動因子は降雨量、降雨強度、貯水池水位標高および
貯水池水位変動を含み、
閾値基準は、以下の基準を含み：
基準１：水位降下速度が０．８８ｍ／ｄを超え、貯水池水位標高が１４６．４５ｍ～１４
９．５０５ｍ間にあり、日降雨量が５７．９ｍｍを超える場合、加速変形が発生し、すな
わち該貯水池地域の地すべりが加速変形セクションであり、実例によって検証された結果
、精度が９０．６８％に達し、
基準２：貯水池水位標高が１６９．７ｍ以下で降雨強度が２４．４ｍｍ／ｈを超える場合
、加速変形が発生し、すなわち該貯水池地域の地すべりが加速変形セクションであり、実
例によって検証された結果、精度が７４．２７％に達し、
基準３：貯水池水位標高が１６９．７ｍ以下で、日降雨量が１．８ｍｍを超え、降雨強度
が２４．４ｍｍ／ｈを超える場合、加速変形が発生し、すなわち該貯水池地域の地すべり
が加速変形セクションであり、実例によって検証された結果、精度が７０．４８％に達し
、
基準４：貯水池水位標高が１６９．７ｍ以下で、降雨強度が０．８ｍｍ／ｈを超え、日降
雨量が５７．９ｍｍを超える場合、加速変形が発生し、すなわち該貯水池地域の地すべり
が加速変形セクションであり、実例によって検証された結果、精度が６１．３％に達し、
基準５：水位上昇速度が０．１４ｍ／ｄを超え、貯水池水位標高が１５０．１９５ｍ以下
で、日降雨量が５７．９ｍｍを超える場合、加速変形が発生し、すなわち該貯水池地域の
地すべりが加速変形セクションであり、実例によって検証された結果、精度が８０．３４
％に達し、
具体的に以下の表３に示され：
表３ 閾値基準

ここで、ｆ_ｉは水位変動値であり、Ｌ_ｉは貯水池水位標高であり、Ｒ_ｉは降雨量であり、
Ｉ_ｉは降雨強度であり、ＤＰは変形階段であり、ＩＩは前記３つの集合結果中の加速変形
を示し、
まず、降雨Ｒ_ｉ、降雨強度Ｉ_ｉ、貯水池水位標高Ｌ_ｉおよび水位変動ｆ_ｉの日値データセ
ット予測警報モデルの入力変数、歪み速度集合結果Ｃｌｕｓｔｅｒ ｊを出力変数として
設定し、次に、すべてのサンプルを７：３で訓練集合とテスト集合に分け、適応強化反復
回数を１０とし、次に、予測警報モデルは計算を開始し、グループ化変数および分割閾値
を決定し、各枝結果が図７に示される。この意思決定ツリーの深さは８であり、ツリーの
第１層では、最大情報利得率の標準に基づいて、貯水池水位標高Ｌ_ｉが最適なグループ化
変数として選択され、ＭＤＬＰスプリッターボックスから得られた最小サブリミット値１
６９．７ｍに基づいて、貯水池水位標高を２組に分割し、ＲＷＬ＞１６９．７ｍの場合、
剪定アルゴリズムに基づいて、グループ化を継続する意味がなくなったと判定され、した
がって、この枝はもはや成長しない。その後、このように、この標準に従って第ｋ層意思
決定ツリーの成長および剪定を行う。
モデル中の意思決定ツリーの成長過程は出力変数に従って入力変数をグループ化する過程
である。意思決定ツリーの各枝は降雨量Ｒ_ｉ、降雨強度Ｉ_ｉ貯水池水位標高Ｌ_ｉおよび水
位変動ｆ_ｉデータセットを連続的にグループ化する過程で徐々に成長したものであり、デ
ータセットの継続的なグループ化の意味がなくなった時点で、意思決定ツリーの成長過程
が終了する。
予測警報モデルは新しいデータ入力変数の値に基づいてその出力変数の値、すなわち歪み
速度集合結果、例えばＣｌｕｓｔｅｒ ２を推論する。以上のように、生成された予測警
報モデルに基づいて地すべり地下歪み進展と外部駆動因子（降雨量、貯水池水位）の閾値
基準を確立し、外部駆動因子に基づく早期リスク警報を実現する。 As shown in FIG. 1 to FIG. 7, the landslide deformation monitoring and early warning method for reservoir area based on sliding zone strain evolution of the present invention is realized by a landslide monitoring system for reservoir area, and the deformation monitoring is based on underground strain monitoring of a vertical borehole;
Potential sliding zones are identified from the strain distribution of the whole borehole, i.e., when the landslide accelerates, the strain in the sliding zone increases rapidly. The early warning is a hydrometeorological warning of significant accelerated deformation of the landslide body during the monitoring period. The method for monitoring and early warning of landslide deformation in a reservoir area based on the sliding zone strain evolution, as shown in Figure 3, includes the following steps:
S1, set up a monitoring station:
First, an active landslide subzone 1 is selected, and a monitoring station is set up in a vertical borehole 3 at the flattened site, as shown in Figure 1, 2 in Figure 1 is a river, the diameter of the vertical borehole is 110 mm, and it is drilled to 3 m below the full rock layer. In the hole 3, an ultra-weak reflective grating strain sensing fiber cable 4, an ultra-weak reflective grating demodulator 5, and a data transmission unit 6 are installed, and the data transmission unit 6 is remotely connected to a user access unit 8, which is a wireless transmission module in the prior art and is integrated in a protection box of the ground monitoring station, and the user access unit 8 is a computer;
In this embodiment, the ultra-weak reflection grating strain sensing fiber cable 4 and the ultra-weak reflection grating demodulator 5 are manufactured by Suzhou Nanzhi Sensing Science and Technology Co., Ltd., China. The technical parameters of both are shown in Tables 1 and 2.
Table 1 Technical parameters of the ultra-weak reflection type grating strain sensing fiber cable 4

Table 2 Technical parameters of the ultra-weak reflection grating demodulator 5

Next, an inverted "convex"-shaped ultra-weak reflection type lattice strain sensing fiber cable 4 is installed in the hole 3, and as shown in FIG. 2, a gravity guide hammer 43 is connected to the ultra-weak reflection type lattice strain sensing fiber cable 4 and the gravity guide hammer 43 is placed at the bottom of the hole 3, one end of the ultra-weak reflection type lattice strain sensing fiber cable 4 is introduced into a protection box 7 and connected to an ultra-weak reflection type lattice demodulator 5 as a collection end, and the other end of the ultra-weak reflection type lattice strain sensing fiber cable 4 is a backup collection end;
The ultra-weak reflection type grating strain sensing fiber cable 4 is a high-resolution ultra-weak reflection type grating strain sensing fiber cable with a sampling interval of 1.m, and the ultra-weak reflection type grating strain sensing fiber cable 4 is composed of an ultra-weak reflection type grating 41 and an optical fiber transmission section 42;
Sampling interval d of the ultra-weak reflection type grating strain sensing fiber cable 4 is d≦1.0 m
and
The collecting end and the backup collecting end are redundantly placed 2 m near the mouth of the hole to assist in accurate positioning of the underground depth, and are particularly suitable for identifying sliding zones in thin layers (several tens of centimeters thick).
The ultra-weak reflective lattice strain sensing fiber cable 4 is pre-tensioned, and the lattice spacing between the collection end and the backup collection end is 0.5 m when projected onto the borehole cross section.
The structure of the ultra-weak reflection grating strain sensing fiber cable 4 is composed of a bare fiber, a coating layer, a metal-based cable-like reinforcement material, and an outer sheath. The gravity guide hammer 43 in this embodiment is made of metal, has a diameter of 50-60 mm, a length of 600-1000 mm, and weighs 20-35 kg. In the ultra-weak reflection grating strain sensing fiber cable 4, 41 is an ultra-weak reflection grating, 42 is an optical fiber transmission section,
Next, the hole 3 is backfilled, and the gravity guide hammer 43 connected to the middle of the ultra-weak reflection type lattice strain sensing fiber cable 4 is lowered to the hole bottom position, and then both ends of the ultra-weak reflection type lattice strain sensing fiber cable 4 extending from the hole mouth are fixed and kept in tension.
The hole 3 is immediately backfilled, and the hole 3 is divided into a complete rock section 31, a crushed rock section 32, and a crushed stone and sand section 33 from bottom to top, and the complete rock section 31 is 3 m high.
The complete rock section 31 is backfilled with a cement quick-setting agent (prior art) to form a hard foundation, the crushed rock section 32 is backfilled with fine sand, and the crushed soil section 33 is backfilled with fine sand and crushed clay in a mass ratio of 1:1.
Specifically, the fine sand is ordinary river sand, and the crushed clay is clay excavated from a borehole 3.
S2, remote real-time monitoring: debugging the ultra-weak reflection grating demodulator 5, the data transmission unit 6 and the user access unit 8, and collecting time-series strain data sets of rock and soil bodies at various depths underground through the user access unit 8;
Debugging refers to connecting each module, turning on the power, and debugging to see if data can be collected.
In step S2, the hole mouth is backfilled and then compacted, one end of the ultra-weak reflection grating strain sensing fiber cable 4 extending from the hole mouth is introduced into a protection box 7 on the ground near the hole mouth, and an optical fiber jumper is connected to connect the collection end of the ultra-weak reflection grating demodulator 5, and the ultra-weak reflection grating demodulator 5, the wireless transmission module 6, and the data user access unit 8 are debugged. After the hole 3 is installed and backfilled and the software and hardware debugging of the monitoring system are completed, one month is waited for the backfill material to solidify and stabilize, and the monitoring initial stage is reached. After that, long-term remote real-time monitoring activities are started. In this embodiment, the monitoring data sampling frequency is a maximum of 5 s/time, and the data sampling frequency is set to 10 s/time during the rainy season or flood season (e.g., 5 to 10 s per year) when deformation is large.
The sampling frequency is 5s during the flood season (September 2019), and the sampling frequency is 60-60s during the non-flood season (September 2019).
00s,
S3. Sliding zone identification and time series distortion set analysis:
In step S2, based on the underground cumulative time series strain data set recorded remotely in real time and acquired by the user access unit 8, a cross-sectional development map of the entire borehole strain is drawn to obtain a strain distribution, the highest strain value area in the strain distribution is the strain peak zone, the strain peak zone is the sliding zone, the average value of each time series strain data for each day for the sliding zone is the daily strain average value, the daily strain average value is subjected to differential processing (the differential processing refers to the difference between the cumulative strain of the next day and the cumulative strain of the previous day, i.e., the strain rate), and a daily average data set of the strain rate is obtained, and the sliding zone has the characteristic that the strain increases rapidly when the accelerating deformation occurs on the slope, and the daily average data of the strain rate is classified into three classification results based on the change characteristics: an approximate stable state, an accelerating (sudden change in strain) deformation, and a stable state (as shown in FIG. 4).
Specifically, the sliding zone strain rate is qualitatively classified based on the characteristics of the change in the sliding zone strain rate over time, the characteristics of the change in the sliding zone strain rate over time include slight strain rate fluctuation, sudden strain rate fluctuation and no strain rate fluctuation, and the classification standard is as follows:
When the strain rate fluctuates slightly, i.e., when the strain rate ranges from 50 to 200 με/d, the movement state of the entire landslide is approximately stable.
When the strain rate changes rapidly, i.e., when the strain rate exceeds 200 με/d, the movement state of the entire landslide is accelerating deformation.
When the strain rate does not fluctuate, i.e., when the strain rate ranges from 0 to 50 με/d, the movement state of the entire landslide is stable.
During the assembly process, high strain rates (>300 με/d) or cumulative strain increases >200 με
Accelerated deformation occurs when the strain rate is between 50 and 200 με/d or the accumulated strain increases rapidly and then rebounds rapidly (a single discontinuous high strain rate isolated point), and the stable state occurs when the strain rate is between 0 and 50 με/d or the accumulated strain changes little. The number of accumulated strain sets can be manually specified or automatically determined by a common set algorithm.
At this point, the numeric variables are converted into type variables (Cluster j, j=1,2,3), i.e., the output variables of the alarm model. The method adopted in the set analysis is to eliminate noise errors in the strain rate time series according to the "data backtracking" principle, i.e., the discontinuous strain changes (singular values) are designated as possible noise data, the zero-area strain is the stable state, and the ±300μ
ε is a critical value between near-steady state and accelerated deformation, above which the set is an accelerated deformation section, and below which the set is a near-steady state section;
S4. Establishment of predictive warning model and threshold criteria:
The three classification results obtained in step S3 (the set result of the sliding zone strain rate Cl
A forecasting and warning model is established by using the rainfall amount R _i , the rainfall intensity I _i , the reservoir water level elevation L _i and the water level fluctuation f _i as output variables, and the daily value data set of the rainfall amount R i , the rainfall intensity I i , the reservoir water level elevation L i and the water level fluctuation f i as input variables, the forecasting model is a decision tree model (prior art), and the value of the output variable is inferred according to the value of the input variable, and the threshold criterion of the landslide underground strain development and the external driving factor is established, and the early warning of the landslide deformation in the reservoir area is provided, and the external driving factors include the rainfall amount, the rainfall intensity, the reservoir water level elevation and the reservoir water level fluctuation;
The threshold criteria include the following criteria:
Criterion 1: The water level drop rate exceeds 0.88 m/d and the reservoir water level elevation is between 146.45 m and 14
9.505m apart, and when the daily rainfall exceeds 57.9mm, accelerated deformation will occur, that is, the landslide in the reservoir area is an accelerated deformation section. After verification by an actual example, the accuracy rate reaches 90.68%.
Criterion 2: When the reservoir water level elevation is below 169.7m and the rainfall intensity is above 24.4mm/h, accelerated deformation occurs, that is, the landslide in the reservoir area is an accelerated deformation section. After verification by actual example, the accuracy rate reaches 74.27%;
Criterion 3: When the reservoir water level elevation is below 169.7m, the daily rainfall exceeds 1.8mm, and the rainfall intensity exceeds 24.4mm/h, accelerated deformation occurs, that is, the landslide in the reservoir area is an accelerated deformation section. After verification by actual example, the accuracy reaches 70.48%;
Criterion 4: When the reservoir water level elevation is below 169.7m, the rainfall intensity is more than 0.8mm/h, and the daily rainfall is more than 57.9mm, accelerated deformation occurs, that is, the landslide in the reservoir area is an accelerated deformation section. After verification by actual example, the accuracy rate reaches 61.3%;
Criterion 5: When the water level rising speed exceeds 0.14 m/d, the reservoir water level elevation is less than or equal to 150.195 m, and the daily rainfall exceeds 57.9 mm, accelerated deformation occurs, that is, the landslide in the reservoir area is an accelerated deformation section. After verification by an actual example, the accuracy is 80.34.
reached %
Specifically, as shown in Table 3 below:
Table 3 Threshold criteria

where f _i is the water level fluctuation value, L _i is the reservoir water level elevation, and R _i is the amount of rainfall.
_Ii is the rainfall intensity, DP is the deformation step, and II indicates the accelerated deformation in the three sets of results;
First, the daily value dataset of rainfall R _i , rainfall intensity I _i , reservoir water level elevation L _i and water level fluctuation f _i are set as input variables of the forecast warning model, and the strain rate set result Cluster j is set as the output variable, then all samples are divided into training set and test set in a ratio of 7:3, and the number of adaptive reinforcement iterations is 10, then the forecast warning model starts to calculate and determine the grouping variables and splitting thresholds, and each branch result is shown in Figure 7. The depth of this decision tree is 8, and in the first layer of the tree, based on the standard of maximum information gain rate, the reservoir water level elevation L _i is selected as the optimal grouping variable, and the minimum sublimit value obtained from the MDLP splitter box is 1.
Based on 69.7m, the reservoir water level elevation is divided into two sets, and when RWL>169.7m,
Based on the pruning algorithm, it is determined that it is no longer meaningful to continue grouping, and therefore this branch is no longer grown. The k-th layer decision tree is then thus grown and pruned according to this standard.
The process of growing the decision tree in the model is a process of grouping the input variables according to the output variables. Each branch of the decision tree is gradually grown in the process of successively grouping the rainfall amount R _i , rainfall intensity I _{i ,} reservoir water level elevation L _{i ,} and water level fluctuation f _i data sets, and the process of growing the decision tree ends when the continued grouping of the data sets makes no sense.
The forecasting and warning model infers its output variable values, i.e., the strain rate aggregation result, for example, Cluster 2, based on the values of new data input variables. As described above, the threshold criteria of landslide underground strain progression and external driving factors (rainfall, reservoir water level) are established based on the generated forecasting and warning model, and early risk warning based on external driving factors is realized.

Test Example As shown in Figures 3 to 6, the study targets landslides in large reservoir areas widely distributed in the Three Gorges Reservoir Region of China, and their deformation is mainly governed by the annual cycle of reservoir water level fluctuations (145-175m fluctuations) and seasonal rainfall (annual average precipitation of 1000mm or more). In-situ monitoring showed that the surface displacements (D1-D5) obtained by the GNSS installed near the borehole and the subsurface strains of the present invention (here, SS1 and SS2 are the temporal strains of the shallow secondary slip zone and the deep primary slip zone, respectively) showed accelerated movement on July 6 (153.3mm daily precipitation). According to the existing weather warning index, heavy rains (less than 50mm daily precipitation) occurred four times during the monitoring period, but not all heavy rains caused landslide movement, and the heavy rain of 97.7mm on August 26 is the best evidence of this (Figure 5).
As shown in Figure 6, the most representative deep primary sliding zone SS2 strain set analysis is performed, and the acceleration stage set is Cluster 2. Standardized rainfall daily value dataset,
The reservoir water level elevation data set is used as the input variable, and the cumulative strain set result is used as the output variable to establish a forecasting and warning model. The sample data is divided into training samples and test samples in a ratio of 7:3.
The overall accuracy of the training sample reaches 92.3%, and the overall accuracy of the test sample reaches 90.6%, which shows that the prediction model of the present invention has higher accuracy. According to the early warning method based on underground strain evolution of the present invention, the warning threshold is set at the reservoir water level RWL altitude of 146.5-149.5 m.
and is obtained as daily rainfall > 57.9 mm or daily rainfall intensity > 24.4 mm/h with reservoir water level RWL below 169.7 m.
To verify the validity of the proposed hydrometeorological threshold, we analyzed the subsurface strain evolution of the real landslide kinematics over the entire flood season of the three hydrological years from 2021 to 2023. As shown in Fig. 7, during this period, two significant increases in the sliding zone strain were observed;
The two landslide acceleration deformation events occurred on July 6 (and July 7) 2021 and June 22, 2022, respectively, and the entire landslide was significantly deformed. Both landslide acceleration deformation events were hydrometeorological conditions indicating that there was a large amount of heavy rainfall (especially extreme rainfall of less than one day duration) during a period when the reservoir water level was low. Specifically, both events were observed when the reservoir water level RWL elevation obtained in the examples was 169
. 7 m (especially 146.45-149.505 m), which fully meets the hydrometeorological warning threshold conditions of daily rainfall > 57.9 mm and daily rainfall intensity exceeding 24.4 mm/h.

Claims (7)

Step S1, set up a monitoring station:
First, a number of holes (3) are drilled in the landslide subzone, extending below the stable rock layer, and the holes (
An ultra-weak reflection type grating strain sensing fiber cable (4), an ultra-weak reflection type grating demodulator (5), and a data transmission unit (6) are installed in the sensor (3), and the data transmission unit (6) is remotely connected to a user access unit (8);
Next, an inverted "convex"-shaped ultra-weak reflection type lattice strain sensing fiber cable (4) is installed in the hole (3), and a gravity guide hammer (43) is connected to the ultra-weak reflection type lattice strain sensing fiber cable (4), and the gravity guide hammer (43) is placed at the bottom of the hole (3). One end of the ultra-weak reflection type lattice strain sensing fiber cable (4) is introduced into a protection box (7) and connected to an ultra-weak reflection type lattice demodulator (5) as a collection end, and the other end of the ultra-weak reflection type lattice strain sensing fiber cable (4) is a backup collection end;
Thereafter, backfilling the hole (3);
Step S2, remote real-time monitoring: debugging the ultra-weak reflection grating demodulator (5), the data transmission unit (6) and the user access unit (8), and collecting time-series strain data sets of rock and soil bodies at various depths underground through the user access unit (8);
Step S3, sliding zone identification and time series distortion set analysis: based on the time series distortion data set acquired by the user access unit (8) in step S2,
A strain distribution is obtained, and a region in the strain distribution with the highest strain value is a strain peak zone;
a step of setting the strain peak zone as a sliding zone, setting the average value of each time-series strain data of the sliding zone as a daily strain average value, performing differential processing on the daily strain average value to obtain a daily average strain rate data set, and classifying the daily average strain rate data into three classification results into an approximate stable state, an accelerating deformation, and a stable state based on its change characteristics;
Step S4, establishing a forecasting and warning model and threshold criteria: taking the three classification results obtained in step S3 as output variables, and the daily value datasets of rainfall _Ri , rainfall intensity _Ii , reservoir water level elevation _Li and water level fluctuation _fi as input variables to establish a forecasting and warning model, said forecasting model is a decision tree model, which infers the value of the output variable according to the value of the input variable, and establishes threshold criteria of landslide underground strain progression and external driving factors for early warning of reservoir area landslide deformation, said external driving factors including rainfall, rainfall intensity, reservoir water level elevation and reservoir water level fluctuation;
The threshold criteria include the following criteria:
Criterion 1: The water level drop rate exceeds 0.88 m/d and the reservoir water level elevation is between 146.45 m and 14
9.505 m apart, and the daily rainfall is more than 57.9 mm, accelerated deformation occurs, i.e., the reservoir area landslide is an accelerated deformation section;
Criterion 2: When the reservoir water level elevation is below 169.7 m and the rainfall intensity is above 24.4 mm/h, accelerated deformation occurs, i.e., the reservoir area landslide is an accelerated deformation section;
Criterion 3: When the reservoir water level elevation is below 169.7 m, the daily rainfall exceeds 1.8 mm, and the rainfall intensity exceeds 24.4 mm/h, accelerated deformation occurs, that is, the reservoir area landslide is an accelerated deformation section;
Criterion 4: When the reservoir water level elevation is below 169.7 m, the rainfall intensity is more than 0.8 mm/h, and the daily rainfall is more than 57.9 mm, accelerated deformation occurs, that is, the reservoir area landslide is an accelerated deformation section;
Criterion 5: when the water level rising speed is more than 0.14 m/d, the reservoir water level elevation is less than or equal to 150.195 m, and the daily rainfall is more than 57.9 mm, accelerated deformation occurs, that is, the reservoir area landslide is an accelerated deformation section;
A reservoir area landslide deformation monitoring and early warning method based on sliding zone strain evolution, comprising: The hole (3) is divided into a complete rock section (31), a fractured rock section (32), and a fractured rock section (33).
2) a crushed rock section (33), and the height of the complete rock section (31) is 3-5 m;
In step S2, the complete rock section (31) is backfilled with a cement quick-setting agent, the crushed rock section (32) is backfilled with fine sand, and the crushed soil section (33) is backfilled with a volume ratio of 1:1.
The reservoir area landslide deformation monitoring and early warning method based on sliding zone strain evolution according to claim 1, characterized in that the ... The ultra-weak reflection type grating strain sensing fiber cable (4) is a ultra-weak reflection type grating (
The reservoir area landslide deformation monitoring and early warning method based on sliding zone strain evolution according to claim 1, characterized in that it is composed of a landslide detection section (41) and an optical fiber transmission section (42). The sampling interval of the ultra-weak reflection type grating strain sensing fiber cable (4) is d≦
The reservoir area landslide deformation monitoring and early warning method based on sliding zone strain evolution as claimed in claim 1, wherein the sliding zone strain evolution is 1.0 m. The reservoir area landslide deformation monitoring and early warning method based on sliding zone strain evolution as claimed in claim 1, characterized in that in step S1, the collecting end and the backup collecting end of the ultra-weak reflective lattice strain sensing fiber cable (4) are redundantly arranged at the mouth of the hole (3) by 2 to 5 m. The reservoir area landslide deformation monitoring and early warning method based on sliding zone strain evolution, as described in claim 5, characterized in that the ultra-weak reflective lattice strain sensing fiber cable (4) is pre-tensioned so that the lattice spacing between the collection end and the backup collection end is 0.5 m in the borehole cross-sectional projection. In step S4, the sliding zone strain rate is qualitatively classified based on the characteristics of the time-varying sliding zone strain rate, the characteristics of the time-varying sliding zone strain rate including: slight strain rate fluctuation, sudden strain rate fluctuation, and no strain rate fluctuation;
The classification standards are as follows:
When the strain rate fluctuates slightly, i.e., when the strain rate ranges from 50 to 200 με/d, the movement state of the entire landslide is approximately stable.
When the strain rate changes rapidly, i.e., when the strain rate exceeds 200 με/d, the movement state of the entire landslide is accelerating deformation.
The reservoir area landslide deformation monitoring and early warning method based on sliding zone strain evolution according to claim 1, characterized in that when the strain rate does not fluctuate, i.e., when the strain rate ranges from 0 to 50 με/d, the overall landslide motion state is stable.