CN114048928A - Building short-term load prediction method with high migratability - Google Patents

Building short-term load prediction method with high migratability Download PDF

Info

Publication number
CN114048928A
CN114048928A CN202210029375.5A CN202210029375A CN114048928A CN 114048928 A CN114048928 A CN 114048928A CN 202210029375 A CN202210029375 A CN 202210029375A CN 114048928 A CN114048928 A CN 114048928A
Authority
CN
China
Prior art keywords
prediction
building
load
data
model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202210029375.5A
Other languages
Chinese (zh)
Other versions
CN114048928B (en
Inventor
陈超
鄢烈祥
周力
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hangu Yunzhi Wuhan Technology Co ltd
Original Assignee
Hangu Yunzhi Wuhan Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Hangu Yunzhi Wuhan Technology Co ltd filed Critical Hangu Yunzhi Wuhan Technology Co ltd
Priority to CN202210029375.5A priority Critical patent/CN114048928B/en
Publication of CN114048928A publication Critical patent/CN114048928A/en
Application granted granted Critical
Publication of CN114048928B publication Critical patent/CN114048928B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/08—Construction

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • General Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Evolutionary Computation (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Operations Research (AREA)
  • Medical Informatics (AREA)
  • Quality & Reliability (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a building short-term load forecasting method with high migratability, which comprises the following steps: reading data in minutes, reading the data from a secondary network real-time operation database table every 1min, cleaning the data, judging the integrity and consistency of the data, calculating hour characteristic data: updating an hour characteristic data table once every hour, and calculating characteristic values of all parameters in the hour; screening the data of the load prediction training set according to criteria; analyzing the correlation between the alternative input variables and the prediction variables of the load prediction model; using an autoencoder to perform feature recognition and extraction of model input variables; calculating the building heat supply load by combining a plurality of prediction models with an SVR support vector regression algorithm; the method has the advantages of backtracking of prediction results and optimization of model parameters, and good migratable capability and prediction precision.

Description

Building short-term load prediction method with high migratability
Technical Field
The invention relates to a building short-term load forecasting method, in particular to a building short-term load forecasting method with high migratability.
Background
With the rapid development of natural science and the continuous emergence of new technologies, the prediction algorithm of machine learning is also widely applied to different fields of various industries. For the heating load of the urban heating system, the buildings and the heat medium have thermal inertia, and are closely related to the heating area, the living habits, the outdoor temperature, the humidity, the solar radiation, the wind direction, the wind speed and other climatic conditions. Making it relatively difficult to build a mathematical model of a building load prediction from a physical model.
Many existing algorithms build machine learning training sets based on historical relationships of heat supply to weather data. In the historical data, there are a large amount of unreasonable heating data, including excessive heating, insufficient heating and the like, which can interfere with the accuracy of the load prediction result.
The prediction of the building heating load is a key technology for realizing the intellectualization of a building heating system. Through carrying out accurate prediction to building load, carry out automatically regulated to building heating system's water supply temperature and flow, can make heat supply enterprise reach the target of heat supply as required, accurate regulation to improve and improve quality of service, reduce operation energy consumption and carbon and discharge, improve system economy nature. In different areas and different types of buildings, when the heat supply load is predicted, the applicable prediction models, model input variables and model parameters are different, and how to improve the adaptivity of the prediction models and the prediction algorithms is realized, so that the buildings have good mobility.
Disclosure of Invention
In order to solve the defects of the prior art, the invention aims to solve the problem that the establishment of a mathematical model for building load prediction through a physical model in the prior art is relatively difficult.
In order to achieve the above object, the present invention relates to a method for predicting short-term load of a building with high migratability, comprising the steps of:
step 1, reading data every 1min, reading the data from a secondary network real-time operation database table, cleaning the data, and judging the integrity and consistency of the data;
step 2, calculating hour characteristic data: updating an hour characteristic data table once every hour, and calculating characteristic values of all parameters in the hour;
step 3, screening the data of the load prediction training set according to criteria;
step 4, analyzing the correlation between the alternative input variables of the load prediction model and the prediction variables;
step 5, using an autoencoder to perform feature recognition and extraction of model input variables;
step 6, calculating the heat supply load by combining a plurality of prediction models with an SVR support vector regression algorithm;
and 7, backtracking a prediction result and optimizing model parameters.
Further, the data read in step 1 includes a time stamp, a water supply temperature of the building, a water return temperature, a building flow rate, and an indoor temperature of a typical room in the building.
Further, the characteristic values of the parameters in the step 2 include supply and return water temperature, flow, heat consumption, typical indoor temperature of each building, and average value, maximum value, minimum value and standard deviation of various meteorological parameters.
Further, the criterion in step 3 is:
criterion I:
more than 80% of all building indoor temperature measuring points in the heating range of the heating station simultaneously meet the following two sub-criteria:
1. average temperature of temperature test value in the hourT indoor,aveIn the range of the heating temperature requirementT CL, T CH]Inner, TCL, TCHIs the upper and lower allowable temperature limits of heat supply, determined by the heating company, lower than TCLThe heat supply is considered to be insufficient and higher than TCHExcess heat supply is considered; the parameters are consistent with the corresponding parameter set values in the prediction model, and the parameters are defaulted to 20 ℃, 22 DEG C]；
2. In this hour, the real-time collection value of the indoor temperature is in the range [, ]T CL, T CH]The accumulated time in the process is more than 45 minutes;
criterion II:
the residents in the building simultaneously meet the following four sub-criteria:
1. the proportion of the residents with windowing data records in the building to the total residents is less than or equal to 10 percent;
2. more than or equal to 85 percent of the residents without windows meet the criterion I;
3. the complaint rate of the users is lower than 2 percent;
4. the deviation of the accumulated overheating temperature and the accumulated insufficient temperature of the building which does not meet the criterion I and is not windowed is less than 20 percent;
criterion III:
the data of the hour and the first two hours of the hour meet the criterion I and the criterion II;
when the hour data meets the criterion III, adding the hour building heat supply load into a training sample database of a building heat supply load prediction model; the method for calculating the heat supply load comprises the following steps:
Figure 425643DEST_PATH_IMAGE001
Figure 882032DEST_PATH_IMAGE002
rho is the density of water, c is the specific heat capacity of water,
Figure 840761DEST_PATH_IMAGE003
is the volume flow of the thermal power station,
Figure 433417DEST_PATH_IMAGE004
the temperature of the water supply of the heating power station,
Figure 265237DEST_PATH_IMAGE005
the temperature of the return water of the heating power station;
Figure 412185DEST_PATH_IMAGE006
representing the accumulation of data collected within one hour, calculated in seconds, so N is 3600;
the building energy consumption data are screened based on the criterion, and the influence of factors such as building thermal inertia and unreasonable heat supply caused by manual operation on the load prediction accuracy can be avoided.
Further, the method for analyzing the correlation between the candidate input variables and the predicted variables of the load prediction model in the step 4 comprises the following steps: analyzing the relation between different input variables and prediction variables based on the system operation historical data, wherein the model alternative input variables comprise but are not limited to:
outdoor dry-bulb temperature, humidity, wind direction, wind speed, horizontal plane total irradiance, scattering irradiance, illuminance, atmospheric pressure, total cloud cover, water air pressure and precipitation;
calculating the correlation degree between each alternative variable and the building heating load by adopting the following formula:
Figure 971342DEST_PATH_IMAGE007
in the formula y i Data set representing building loadiA sample, x i First in the input variable data setiThe number of the samples is one,
Figure 949663DEST_PATH_IMAGE008
represents the average of the input variable data set,
Figure 670494DEST_PATH_IMAGE009
represents the average of the load data set.
Further, the method for performing feature recognition and extraction of the model input variable by using the self-encoder in the step 5 comprises: after the selection of the model input variable is determined, an automatic encoder based on an artificial neural network algorithm is used for feature extraction of the model input parameter;
the number of units of the self-encoder is the same, the self-encoder comprises a plurality of hidden layers, and the number of layers of the hidden layers is smaller than that of the output layers; input parameters by modelxAs output of the modely(ii) a The output from the encoder is used as input to the prediction model.
Further, the method for performing the calculation in step 6 is as follows: calculating the building heat supply load by using various prediction models, and in a load prediction result backtracking stage (step (7)), evaluating the optimal prediction effect of each model and automatically calling an optimal prediction algorithm; all the prediction models run simultaneously, and large system resources are occupied; therefore, learning is carried out by adopting an SVR support vector regression algorithm as a default algorithm so as to obtain a preliminary prediction result; after historical data of a period of time is established, backtracking is carried out, the prediction accuracy of other alternative algorithms is evaluated one by one based on the historical data, and the backtracking and learning are carried out asynchronously; the method ensures that excessive calculation workload cannot be caused, and can gradually improve the accuracy of model prediction;
the accuracy of the support vector regression algorithm prediction mainly depends on the selection of a kernel function, a Gaussian kernel function is selected according to the reasons of the particularity of a city heat supply system, the complexity of the system and the like, the feature data of samples can be uniformly changed according to a certain rule to obtain new samples, the new samples can be better classified according to the new feature data, and the feature data of the new samples and the feature data of original samples are in a corresponding relation of a certain rule, so that the classification condition of the original samples is obtained according to the distribution and the classification condition of the new samples, and a new space which is more beneficial to a classification task is found.
Further, the plurality of predictive models includes, but is not limited to: regression analysis, artificial neural network, grey prediction, time series, and autoregressive moving.
Further, the method for backtracking the prediction result and optimizing the model parameters in step 7 comprises: and (3) backtracking the operation result every time the algorithm operates for a period of time, screening the building load data based on the criterion in the step (3), comparing the building load data meeting the criterion with a predicted value thereof, and evaluating the deviation between the predicted result and the actual load by using a coefficient, wherein the calculation method comprises the following steps:
Figure 304738DEST_PATH_IMAGE010
calculating the prediction results of various prediction models, and calculating the optimal parameters of various models based on a pattern search algorithm; in the formula, r2To evaluate the goodness of fit of the predictive model,
Figure 916853DEST_PATH_IMAGE011
the closer to 1 the value of (A) indicates a better fit of the regression to the historical data, whereas r2The smaller the value of (a), the worse the fitting degree of regression to historical data;
the denominator is the sum of squared deviations of the actual heating load,
Figure 749680DEST_PATH_IMAGE012
represents the average actual thermal load;
the numerator is the sum of the prediction errors of the heating load,
Figure 844675DEST_PATH_IMAGE013
indicating the predicted thermal load at time i,
Figure 28532DEST_PATH_IMAGE014
representing the actual thermal load at time i;
and finally, selecting an algorithm with the highest prediction precision under the condition of the optimal model parameters for predicting the building heating load in the next stage until the next prediction result is backtracked.
In general, compared with the prior art, the above technical solution contemplated by the present invention can achieve the following beneficial effects:
(1) the building short-term load online prediction method with high migratability has good migratability, adaptively improves prediction accuracy, and is used for building heat supply load prediction in different areas and different use scenes.
(2) The building short-term load online prediction method with high migratability judges the rationality of building heat supply energy consumption based on real-time building room temperature monitoring and heat supply management platform complaint data, screens the data, and eliminates the influence of human factors and a complex heat transfer process in a building on the building load prediction accuracy; three data screening criteria (such as criteria I, II and III in the technical route diagram of the invention shown in FIG. 1) are provided, and based on the criteria, data screening can be automatically realized without manual intervention.
(3) The building short-term load online prediction method with high migratability analyzes the correlation between the prediction model candidate input variables and building heat supply energy consumption data based on historical data such as building external environment, human behaviors and the like, realizes automatic screening of the model input variables, periodically checks and evaluates results, and corrects a model training method.
(4) According to the building short-term load online prediction method with high migratability, the effectiveness of a historical prediction result is backtracked and analyzed, model parameters are optimized based on a pattern search algorithm, so that the prediction precision of a model is continuously improved, the global optimum value of a target function can be quickly obtained, and the convergence success rate is high and the cost is low.
(5) The building short-term load online prediction method with high migratability uses a self-encoder for noise reduction and feature extraction of model input parameters, so that the generalization capability of a model is improved;
description of the drawings:
FIG. 1 is a logic flow diagram of a preferred embodiment of the present invention;
FIG. 2 is a diagram illustrating feature extraction of model input parameters according to a preferred embodiment of the present invention;
FIG. 3 is a diagram illustrating an optimization process of SVM model parameters according to a preferred embodiment of the present invention;
FIG. 4 is a logic diagram of the pattern search algorithm of the preferred embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
Referring to fig. 1, a method for predicting short-term load of a building with high migratability includes the following steps:
step 1, reading data every 1min, reading the data from a secondary network real-time operation database table, cleaning the data, and judging the integrity and consistency of the data;
step 2, calculating hour characteristic data: updating an hour characteristic data table once every hour, and calculating characteristic values of all parameters in the hour;
step 3, screening the data of the load prediction training set according to criteria;
step 4, analyzing the correlation between the alternative input variables of the load prediction model and the prediction variables;
step 5, using an autoencoder to perform feature recognition and extraction of model input variables;
step 6, calculating the heat supply load by combining a plurality of prediction models with an SVR support vector regression algorithm;
and 7, backtracking a prediction result and optimizing model parameters.
The data read in the step 1 comprise a time stamp, the water supply temperature of the building, the water return temperature, the building flow and the indoor temperature of a typical room in the building.
The characteristic values of the parameters in the step 2 comprise the supply and return water temperature, the flow, the heat consumption, the typical indoor temperature of each building and the average value, the maximum value, the minimum value and the standard deviation of various meteorological parameters.
The criterion in the step 3 is as follows:
criterion I:
more than 80% of all building indoor temperature measuring points in the heating range of the heating station simultaneously meet the following two sub-criteria:
1. average temperature of temperature test value in the hourT indoor,aveIn the range of the heating temperature requirementT CL, T CH]Inner, TCL, TCHIs the upper and lower allowable temperature limits of heat supply, determined by the heating company, lower than TCLThe heat supply is considered to be insufficient and higher than TCHExcess heat supply is considered; the parameters are consistent with the corresponding parameter set values in the prediction model, and the parameters are defaulted to 20 ℃, 22 DEG C]；
2. In this hour, the real-time collection value of the indoor temperature is in the range [, ]T CL, T CH]The accumulated time in the process is more than 45 minutes;
criterion II:
the residents in the building simultaneously meet the following four sub-criteria:
1. the proportion of the residents with windowing data records in the building to the total residents is less than or equal to 10 percent;
2. more than or equal to 85 percent of the residents without windows meet the criterion I;
3. the complaint rate of the users is lower than 2 percent;
4. the deviation of the accumulated overheating temperature and the accumulated insufficient temperature of the building which does not meet the criterion I and is not windowed is less than 20 percent;
criterion III:
the data of the hour and the first two hours of the hour meet the criterion I and the criterion II;
when the hour data meets the criterion III, adding the hour building heat supply load into a training sample database of a building heat supply load prediction model; the method for calculating the heat supply load comprises the following steps:
Figure 929492DEST_PATH_IMAGE015
Figure 554508DEST_PATH_IMAGE016
rho is the density of water, c is the specific heat capacity of water,
Figure 899033DEST_PATH_IMAGE017
is the volume flow of the thermal power station,
Figure 39027DEST_PATH_IMAGE018
the temperature of the water supply of the heating power station,
Figure 946940DEST_PATH_IMAGE019
the temperature of the return water of the heating power station;
Figure 754359DEST_PATH_IMAGE020
representing the accumulation of data collected over an hour, in seconds, so N is taken to be 3600；
The building energy consumption data are screened based on the criterion, and the influence of factors such as building thermal inertia and unreasonable heat supply caused by manual operation on the load prediction accuracy can be avoided.
The method for analyzing the correlation between the alternative input variables of the load prediction model and the prediction variables in the step 4 comprises the following steps: analyzing the relation between different input variables and prediction variables based on the system operation historical data, wherein the model alternative input variables comprise but are not limited to:
outdoor dry-bulb temperature, humidity, wind direction, wind speed, horizontal plane total irradiance, scattering irradiance, illuminance, atmospheric pressure, total cloud cover, water air pressure and precipitation;
calculating the correlation degree between each alternative variable and the building heating load by adopting the following formula:
Figure 456736DEST_PATH_IMAGE021
in the formula y i Data set representing building loadiA sample, x i First in the input variable data setiThe number of the samples is one,
Figure 84027DEST_PATH_IMAGE022
represents the average of the input variable data set,
Figure 838706DEST_PATH_IMAGE023
represents the average of the load data set.
Referring to fig. 2, the method for performing feature recognition and extraction of the model input variables in step 5 by using the self-encoder includes: after the selection of the model input variable is determined, an automatic encoder based on an artificial neural network algorithm is used for feature extraction of the model input parameter;
the number of units of the self-encoder is the same, the self-encoder comprises a plurality of hidden layers, and the number of layers of the hidden layers is smaller than that of the output layers; input parameters by modelxAs output of the modely(ii) a The output from the encoder is used as input to the prediction model.
The calculation method in the step 6 comprises the following steps: calculating the building heat supply load by using various prediction models, and in a load prediction result backtracking stage (step (7)), evaluating the optimal prediction effect of each model and automatically calling an optimal prediction algorithm; all the prediction models run simultaneously, and large system resources are occupied; therefore, learning is carried out by adopting an SVR support vector regression algorithm as a default algorithm so as to obtain a preliminary prediction result; after historical data of a period of time is established, backtracking is carried out, the prediction accuracy of other alternative algorithms is evaluated one by one based on the historical data, and the backtracking and learning are carried out asynchronously; the method ensures that excessive calculation workload cannot be caused, and can gradually improve the accuracy of model prediction;
the accuracy of the support vector regression algorithm prediction mainly depends on the selection of a kernel function, a Gaussian kernel function is selected according to the reasons of the particularity of a city heat supply system, the complexity of the system and the like, the feature data of samples can be uniformly changed according to a certain rule to obtain new samples, the new samples can be better classified according to the new feature data, and the feature data of the new samples and the feature data of original samples are in a corresponding relation of a certain rule, so that the classification condition of the original samples is obtained according to the distribution and the classification condition of the new samples, and a new space which is more beneficial to a classification task is found.
The plurality of predictive models includes, but is not limited to: regression analysis, artificial neural network, grey prediction, time series, and autoregressive moving.
The method for backtracking the prediction result and optimizing the model parameters in the step 7 comprises the following steps: and (3) backtracking the operation result every time the algorithm operates for a period of time, screening the building load data based on the criterion in the step (3), comparing the building load data meeting the criterion with a predicted value thereof, and evaluating the deviation between the predicted result and the actual load by using a coefficient, wherein the calculation method comprises the following steps:
Figure 235053DEST_PATH_IMAGE024
denominator is actual heating loadThe sum of squared deviations of (a) and (b),
Figure 108331DEST_PATH_IMAGE012
represents the average actual thermal load;
the numerator is the sum of the prediction errors of the heating load,
Figure 488496DEST_PATH_IMAGE013
indicating the predicted thermal load at time i,
Figure 269371DEST_PATH_IMAGE014
representing the actual thermal load at time i;
referring to fig. 4, the prediction results of various prediction models are calculated, and the optimal parameters of various models are calculated based on a pattern search algorithm; in the formula, r2To evaluate the goodness of fit of the prediction model, the closer the value is to 1, which indicates that the regression fits better to the historical data, whereas r2The smaller the value of (a), the worse the fitting degree of regression to historical data;
the denominator is the sum of squared deviations of the actual heat supply load, representing the average actual heat load;
the numerator is the sum of the prediction errors of the heating load,
Figure 989065DEST_PATH_IMAGE025
indicating the predicted thermal load at time i,
Figure 846294DEST_PATH_IMAGE026
representing the actual thermal load at time i;
taking an SVM model as an example, the optimization process of the model parameters (C, sigma) (C represents different coefficients; sigma represents the width of the kernel function) is shown in FIG. 3;
and finally, selecting an algorithm with the highest prediction precision under the condition of the optimal model parameters for predicting the building heating load in the next stage until the next prediction result is backtracked.
Building heating loads are affected by a number of complex factors. The main factors affecting the thermal load of different areas and types of buildings are different. Therefore, the prediction models are different from the appropriate prediction methods. The algorithm analyzes and predicts the correlation between the model alternative input variables and the building heat supply energy consumption data on line based on the historical data of the building external environment, human behaviors and the like, can realize the automatic screening of the model input variables, regularly checks and evaluates the result, and corrects the model training method. By the method, the algorithm has self-adaptability, and can adaptively screen data and switch the model and the model input parameters without special setting aiming at different regions and different types of buildings, and the accuracy of prediction calculation can be gradually improved without manual modification and configuration. Because the algorithm has good adaptivity, the algorithm can be conveniently migrated to other projects for use. Therefore, has good mobility.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (9)

1. A building short-term load forecasting method with high migratability is characterized by comprising the following steps:
step 1, reading data every 1min, reading the data from a secondary network real-time operation database table, cleaning the data, and judging the integrity and consistency of the data;
step 2, calculating hour characteristic data: updating an hour characteristic data table once every hour, and calculating characteristic values of all parameters in the hour;
step 3, screening the data of the load prediction training set according to criteria;
step 4, analyzing the correlation between the alternative input variables of the load prediction model and the prediction variables;
step 5, using an autoencoder to perform feature recognition and extraction of model input variables;
step 6, calculating the building heat supply load by combining a plurality of prediction models with an SVR support vector regression algorithm;
and 7, backtracking a prediction result and optimizing model parameters.
2. The method for predicting the short-term load of the building with high migratability as recited in claim 1, wherein the data read in step 1 comprises a time stamp, a water supply temperature of the building, a water return temperature, a building flow rate, and an indoor temperature of a typical room in the building.
3. The method for predicting the short-term load of a building with high migratability as recited in claim 1, wherein the characteristic values of the parameters in step 2 include the supply and return water temperature, flow rate, heat consumption, typical indoor temperature, and the average, maximum, minimum and standard deviation of various meteorological parameters of each building.
4. The method for predicting the short-term load of the building with high migratability as recited in claim 1, wherein the criterion in the step 3 is:
criterion I:
more than 80% of all building indoor temperature measuring points in the heating range of the heating station simultaneously meet the following two sub-criteria:
1. average temperature of temperature test value in the hourT indoor,aveIn the range of the heating temperature requirementT CL, T CH]Inner, TCL, TCHIs the upper and lower allowable temperature limits of heat supply, determined by the heating company, lower than TCLThe heat supply is considered to be insufficient and higher than TCHExcess heat supply is considered; the parameters are consistent with the corresponding parameter set values in the prediction model, and the parameters are defaulted to 20 ℃, 22 DEG C]；
2. In this hour, the real-time collection value of the indoor temperature is in the range [, ]T CL, T CH]The accumulated time in the process is more than 45 minutes;
criterion II:
the residents in the building simultaneously meet the following four sub-criteria:
the proportion of the residents with windowing data records in the building to the total residents is less than or equal to 10 percent;
more than or equal to 85 percent of the residents without windows meet the criterion I;
the complaint rate of the users is lower than 2 percent;
the deviation of the accumulated overheating temperature and the accumulated insufficient temperature of the building which does not meet the criterion I and is not windowed is less than 20 percent;
criterion III:
the data of the hour and the first two hours of the hour meet the criterion I and the criterion II;
when the hour data meets the criterion III, adding the hour building heat supply load into a training sample database of a building heat supply load prediction model; the method for calculating the heat supply load comprises the following steps:
Figure 435267DEST_PATH_IMAGE001
Figure 19832DEST_PATH_IMAGE002
rho is the density of water, c is the specific heat capacity of water,
Figure 141503DEST_PATH_IMAGE003
is the volume flow of the thermal power station,
Figure 356584DEST_PATH_IMAGE004
the temperature of the water supply of the heating power station,
Figure 445763DEST_PATH_IMAGE005
the temperature of the return water of the heating power station;
Figure 834019DEST_PATH_IMAGE006
representing the accumulation of data collected within one hour, calculated in seconds, so N is 3600;
the building energy consumption data are screened based on the criterion, and the influence of factors such as building thermal inertia and unreasonable heat supply caused by manual operation on the load prediction accuracy can be avoided.
5. The method for predicting the short-term load of the building with the high migratability capability according to the claim 1, wherein the method for analyzing the correlation between the candidate input variables and the predicted variables of the load prediction model in the step 4 is as follows: analyzing the relation between different input variables and prediction variables based on the system operation historical data, wherein the model alternative input variables comprise but are not limited to:
outdoor dry-bulb temperature, humidity, wind direction, wind speed, horizontal plane total irradiance, scattering irradiance, illuminance, atmospheric pressure, total cloud cover, water air pressure and precipitation;
calculating the correlation degree between each alternative variable and the building heating load by adopting the following formula:
Figure 262726DEST_PATH_IMAGE007
in the formula y i Data set representing building loadiA sample, x i First in the input variable data setiThe number of the samples is one,
Figure 711025DEST_PATH_IMAGE008
represents the average of the input variable data set,
Figure 490762DEST_PATH_IMAGE009
represents the average of the load data set.
6. The method for predicting the short-term load of the building with high migratability as recited in claim 1, wherein the step 5 of using the self-encoder to perform the feature recognition and extraction of the model input variables comprises: after the selection of the model input variable is determined, an automatic encoder based on an artificial neural network algorithm is used for feature extraction of the model input parameter;
sheets from encodersThe number of the elements is the same, the elements comprise a plurality of hidden layers, and the number of the hidden layers is smaller than that of the output layers; input parameters by modelxAs output of the modely(ii) a The output from the encoder is used as input to the prediction model.
7. The method for predicting the short-term load of the building with high migratability as recited in claim 1, wherein the calculation in step 6 is performed by: calculating the building heat supply load by using various prediction models, evaluating the optimal prediction effect of each model in the backtracking stage of the load prediction result, and automatically calling an optimal prediction algorithm; because all the prediction models run simultaneously, larger system resources are occupied; therefore, learning is carried out by adopting an SVR support vector regression algorithm as a default algorithm so as to obtain a preliminary prediction result; after historical data of a period of time is established, backtracking is carried out, the prediction accuracy of other alternative algorithms is evaluated one by one based on the historical data, and the backtracking and learning are carried out asynchronously; the excessive calculation workload is not caused, and the model prediction precision can be gradually improved;
the accuracy of the support vector regression algorithm prediction mainly depends on the selection of a kernel function, a Gaussian kernel function is selected according to the particularity of a city heat supply system and the complexity of the system, the Gaussian kernel function can uniformly change the feature data of samples according to rules to obtain new samples, the new samples are better classified according to the new feature data, and the feature data of the new samples and the feature data of original samples are in a corresponding relation, so that the classification condition of the original samples is obtained according to the distribution and classification condition of the new samples, and a new space which is more beneficial to a classification task is found.
8. The method of claim 7, wherein the plurality of predictive models includes, but is not limited to: regression analysis, artificial neural network, grey prediction, time series, and autoregressive moving.
9. The method for predicting the short-term load of a building with high migratability as recited in any of claims 1-8, wherein the method for backtracking the prediction result and optimizing the model parameters in step 7 comprises: and (3) backtracking the operation result every time the algorithm operates for a period of time, screening the building load data based on the criterion in the step (3), comparing the building load data meeting the criterion with a predicted value thereof, and evaluating the deviation between the predicted result and the actual load by using a coefficient, wherein the calculation method comprises the following steps:
Figure 669327DEST_PATH_IMAGE010
calculating the prediction results of various prediction models, and calculating the optimal parameters of various models based on a pattern search algorithm; in the formula, r2To evaluate goodness of fit of the prediction model, r2The closer to 1 the value of (A) indicates a better fit of the regression to the historical data, whereas r2The smaller the value of (a), the worse the fitting degree of regression to historical data;
the denominator is the sum of squared deviations of the actual heating load,
Figure 14858DEST_PATH_IMAGE011
represents the average actual thermal load;
the numerator is the sum of the prediction errors of the heating load,
Figure 837320DEST_PATH_IMAGE012
indicating the predicted thermal load at time i,
Figure 901091DEST_PATH_IMAGE013
representing the actual thermal load at time i;
and finally, selecting an algorithm with the highest prediction precision under the condition of the optimal model parameters for predicting the building heating load in the next stage until the next prediction result is backtracked.
CN202210029375.5A 2022-01-12 2022-01-12 Building short-term load prediction method with high migratability Active CN114048928B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210029375.5A CN114048928B (en) 2022-01-12 2022-01-12 Building short-term load prediction method with high migratability

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210029375.5A CN114048928B (en) 2022-01-12 2022-01-12 Building short-term load prediction method with high migratability

Publications (2)

Publication Number Publication Date
CN114048928A true CN114048928A (en) 2022-02-15
CN114048928B CN114048928B (en) 2022-04-19

Family

ID=80196363

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210029375.5A Active CN114048928B (en) 2022-01-12 2022-01-12 Building short-term load prediction method with high migratability

Country Status (1)

Country Link
CN (1) CN114048928B (en)

Citations (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160305678A1 (en) * 2015-04-20 2016-10-20 Alexandre PAVLOVSKI Predictive building control system and method for optimizing energy use and thermal comfort for a building or network of buildings
US20170300599A1 (en) * 2016-04-18 2017-10-19 University Of Southern California System and method for calibrating multi-level building energy simulation
CN107704875A (en) * 2017-09-30 2018-02-16 山东建筑大学 Based on the building load Forecasting Methodology and device for improving IHCMAC neutral nets
CN108197404A (en) * 2018-01-22 2018-06-22 河北工业大学 A kind of building load Forecasting Methodology based on time hereditary capacity
CN109933850A (en) * 2019-02-02 2019-06-25 天津大学 A kind of residential architecture thermic load model step calibration method
CN111503718A (en) * 2020-03-09 2020-08-07 华电电力科学研究院有限公司 Cogeneration heating load prediction method based on multi-factor influence and heating system
CN111626783A (en) * 2020-04-30 2020-09-04 贝壳技术有限公司 Offline information setting method and device for realizing event conversion probability prediction
US20200379423A1 (en) * 2009-06-22 2020-12-03 Johnson Controls Technology Company Systems and methods for detecting changes in energy usage in a building
CN112926795A (en) * 2021-03-22 2021-06-08 西安建筑科技大学 SBO (statistical analysis) -based CNN (continuous casting) optimization-based high-rise residential building group heat load prediction method and system
CN113237135A (en) * 2021-05-18 2021-08-10 天津理工大学 SD-LSTMs-based apartment independent heating system heat demand prediction and energy-saving control method
CN113671923A (en) * 2021-08-30 2021-11-19 东北大学 Continuous rolling process fault detection method based on data driving
CN113757781A (en) * 2020-10-30 2021-12-07 浙江正泰聚能科技有限公司 Heat supply load prediction energy-saving control method and system based on BP neural network prediction model
WO2022000700A1 (en) * 2020-07-03 2022-01-06 南京东博智慧能源研究院有限公司 Backup configuration method for park comprehensive energy system in which gas and thermal inertia backups participate

Patent Citations (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200379423A1 (en) * 2009-06-22 2020-12-03 Johnson Controls Technology Company Systems and methods for detecting changes in energy usage in a building
US20160305678A1 (en) * 2015-04-20 2016-10-20 Alexandre PAVLOVSKI Predictive building control system and method for optimizing energy use and thermal comfort for a building or network of buildings
US20170300599A1 (en) * 2016-04-18 2017-10-19 University Of Southern California System and method for calibrating multi-level building energy simulation
CN107704875A (en) * 2017-09-30 2018-02-16 山东建筑大学 Based on the building load Forecasting Methodology and device for improving IHCMAC neutral nets
CN108197404A (en) * 2018-01-22 2018-06-22 河北工业大学 A kind of building load Forecasting Methodology based on time hereditary capacity
CN109933850A (en) * 2019-02-02 2019-06-25 天津大学 A kind of residential architecture thermic load model step calibration method
CN111503718A (en) * 2020-03-09 2020-08-07 华电电力科学研究院有限公司 Cogeneration heating load prediction method based on multi-factor influence and heating system
CN111626783A (en) * 2020-04-30 2020-09-04 贝壳技术有限公司 Offline information setting method and device for realizing event conversion probability prediction
WO2022000700A1 (en) * 2020-07-03 2022-01-06 南京东博智慧能源研究院有限公司 Backup configuration method for park comprehensive energy system in which gas and thermal inertia backups participate
CN113757781A (en) * 2020-10-30 2021-12-07 浙江正泰聚能科技有限公司 Heat supply load prediction energy-saving control method and system based on BP neural network prediction model
CN112926795A (en) * 2021-03-22 2021-06-08 西安建筑科技大学 SBO (statistical analysis) -based CNN (continuous casting) optimization-based high-rise residential building group heat load prediction method and system
CN113237135A (en) * 2021-05-18 2021-08-10 天津理工大学 SD-LSTMs-based apartment independent heating system heat demand prediction and energy-saving control method
CN113671923A (en) * 2021-08-30 2021-11-19 东北大学 Continuous rolling process fault detection method based on data driving

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
HAI-XIANG ZHAO等: "Feature selection for support vector regression in the application of building energy prediction", 《 2011 IEEE 9TH INTERNATIONAL SYMPOSIUM ON APPLIED MACHINE INTELLIGENCE AND INFORMATICS (SAMI)》 *
JIHAOGU,JINWANG等: "Medium-term heat load prediction for an existing residential building based on a wireless on-off control system", 《ENERGY》 *
X.J.LUO,LUKUMON ,O.OYEDELE,ANUOLUWAPO等: "Comparative study of machine learning-based multi-objective prediction framework for multiple building energy loads", 《SUSTAINABLE CITIES AND SOCIETY》 *
刘晓娟: "基于气象因素分析的办公建筑动态短期负荷预测研究", 《中国优秀博硕士学位论文全文数据库(硕士) 工程科技Ⅱ辑》 *
张强: "办公建筑冷热负荷短期和超短期预测方法及应用研究", 《万方期刊》 *
李凯: "基于机器学习的短期建筑供热负荷预测", 《中国优秀博硕士学位论文全文数据库(硕士)工程科技Ⅱ辑》 *

Also Published As

Publication number Publication date
CN114048928B (en) 2022-04-19

Similar Documents

Publication Publication Date Title
CN110363347B (en) Method for predicting air quality based on neural network of decision tree index
CN107730031B (en) Ultra-short-term peak load prediction method and system
CN112801388B (en) Power load prediction method and system based on nonlinear time series algorithm
CN110046743A (en) Energy Consumption of Public Buildings prediction technique and system based on GA-ANN
CN106651036A (en) Air quality forecasting system
CN113240184B (en) Building space unit cold load prediction method and system based on federal learning
CN111191854A (en) Photovoltaic power generation prediction model and method based on linear regression and neural network
CN109858700A (en) BP neural network heating system energy consumption prediction technique based on similar screening sample
CN112365056A (en) Electrical load joint prediction method and device, terminal and storage medium
CN117540938B (en) Integrated building energy consumption prediction method and system based on TD3 reinforcement learning optimization
CN117669895A (en) Highway engineering environment influence evaluation system
CN114169254A (en) Abnormal energy consumption diagnosis method and system based on short-term building energy consumption prediction model
CN110728409A (en) Flood process type similarity mining and rapid prediction method
CN115018200A (en) Power load prediction method and system based on deep learning and considering multiple influence factors
CN115470862A (en) Dynamic self-adaptive load prediction model combination method
CN117150677A (en) Method and device for determining type selection parameters of electric dust collector
CN111723085B (en) Method and system for cleaning electric energy consumption data of public building air conditioner
CN117829374A (en) Prediction method and system for power consumption of iron and steel enterprises
Abdellatif et al. A thermal control methodology based on a machine learning forecasting model for indoor heating
CN114048928B (en) Building short-term load prediction method with high migratability
CN114861555A (en) Regional comprehensive energy system short-term load prediction method based on Copula theory
CN116826745B (en) Layered and partitioned short-term load prediction method and system in power system background
Fathi et al. Campus energy use prediction (CEUP) using artificial intelligence (AI) to study climate change impacts
CN118485267A (en) Energy-saving management platform for carbon neutralization energy consumption
CN117892655B (en) Building networking virtualization management platform and management method and system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant