CN112149883A - Photovoltaic power prediction method based on FWA-BP neural network - Google Patents

Photovoltaic power prediction method based on FWA-BP neural network Download PDF

Info

Publication number
CN112149883A
CN112149883A CN202010927605.0A CN202010927605A CN112149883A CN 112149883 A CN112149883 A CN 112149883A CN 202010927605 A CN202010927605 A CN 202010927605A CN 112149883 A CN112149883 A CN 112149883A
Authority
CN
China
Prior art keywords
neural network
fwa
data
photovoltaic power
firework
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.)
Withdrawn
Application number
CN202010927605.0A
Other languages
Chinese (zh)
Inventor
张洁
郝倩男
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.)
Nanjing University of Posts and Telecommunications
Original Assignee
Nanjing University of Posts and Telecommunications
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 Nanjing University of Posts and Telecommunications filed Critical Nanjing University of Posts and Telecommunications
Priority to CN202010927605.0A priority Critical patent/CN112149883A/en
Publication of CN112149883A publication Critical patent/CN112149883A/en
Withdrawn legal-status Critical Current

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
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Computational Linguistics (AREA)
  • Development Economics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Game Theory and Decision Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a photovoltaic power prediction method based on a FWA-BP neural network, and belongs to the technical field of photovoltaic power generation. The randomness and the indirectness of the photovoltaic power generation power can cause certain influence on large-scale photovoltaic grid connection, so the invention provides a photovoltaic power prediction method based on an FWA-BP neural network, the weight and the threshold of the BP neural network are optimized by using a firework algorithm (FWA), the BP neural network is prevented from falling into local optimization, the convergence speed of the BP neural network is accelerated, meanwhile, the model prediction precision is improved through data preprocessing, and the high-precision prediction of the photovoltaic power can be realized.

Description

Photovoltaic power prediction method based on FWA-BP neural network
Technical Field
The invention relates to a photovoltaic power prediction method based on a FWA-BP neural network, and belongs to the field of power prediction.
Background
With the rapid increase of traditional energy consumption and the continuous deterioration of climate environment, photovoltaic power generation technology has been rapidly developed in recent years. However, due to the influence of meteorological conditions, randomness and indirection of photovoltaic power generation power can have certain influence on large-scale photovoltaic grid connection. In order to ensure the stable operation of the photovoltaic power station and the safe dispatching of the power grid, the method has very important significance in accurately and timely predicting the photovoltaic power.
The conventional photovoltaic power prediction method mainly comprises a physical method and a statistical method. The physical method needs meteorological data, geographic information of a photovoltaic power station and a photovoltaic module ground parameter, photovoltaic power generation amount is obtained through calculation according to an output characteristic curve, the prediction accuracy depends on the structure of an object to be measured and the precision of selected parameters, but the method involves many links and is complex in process; the statistical rule is to find out the intrinsic rule by performing statistical analysis on historical data.
The BP neural network is a multilayer feedforward neural network based on error back propagation, has strong robustness, can infinitely approximate nonlinearity, has strong learning capability, and is widely applied to the field of photovoltaic power generation power prediction. However, the BP neural network also has the phenomena of easy falling into a local minimum value, slow convergence rate and overfitting, and meanwhile, the BP neural network model has large error and low prediction precision, and the ideal prediction state is still difficult to achieve by optimizing the BP neural network for many times by technical personnel in the field.
In the process of model training of the traditional BP neural network, network parameters in the neural network are easy to fall into local optimization, the network parameters stop changing after the network parameters fall into the local optimization, and errors in a training set are not reduced even if the neural network is continuously trained.
In view of the above, there is a need for an improvement of the existing BP neural network to solve the above problems.
Disclosure of Invention
The invention aims to provide a photovoltaic power prediction method based on an FWA-BP neural network, which can improve the accuracy of photovoltaic power generation power prediction.
In order to achieve the purpose, the invention provides the following technical scheme:
the photovoltaic power prediction method based on the FWA-BP neural network mainly comprises the following steps:
s1: acquiring historical meteorological data and historical output power data of a photovoltaic power station;
s2: preprocessing historical meteorological data and historical output power data;
s3: selecting data influencing short-term prediction of photovoltaic power from historical meteorological data as an input vector;
s4: normalizing the input vector;
s5: performing off-line training by using a BP neural network optimized by a Firework algorithm (FWA) to create a prediction model based on the FWA-BP neural network;
s6: and predicting the photovoltaic power generation power according to the prediction model.
As a further improvement of the present invention, step S1 is specifically to obtain historical meteorological data and historical output power data which affect the photovoltaic output power and form a data set, where the historical meteorological data includes irradiance, temperature, humidity and wind speed, and 80% of the data in the data set is selected as a training set and 20% of the data is selected as a testing set.
As a further improvement of the present invention, step S2 is specifically to detect abnormal data and fill abnormal values, the detection of abnormal data adopts 3 σ principle, the fill of abnormal values adopts K-nearest neighbor method, and the K-nearest neighbor algorithm formula is:
Figure BDA0002668976660000021
in the formula, Xi-kIs the k-th data preceding the outlier, Xi+kIs the k-th data after the outlier.
As a further improvement of the present invention, step S3 specifically includes calculating a correlation coefficient between historical meteorological data and historical output power by using a Pearson similarity analysis method, selecting two values with the highest correlation coefficient as input vectors, where the calculation formula of the Pearson similarity analysis method is:
Figure BDA0002668976660000031
wherein X, Y are historical meteorological data and historical output power, respectively.
As a further improvement of the present invention, step S4 is specifically a normalization process for normalizing the input vector to the interval of [0, 1], where the calculation formula is:
Figure BDA0002668976660000032
wherein max is the maximum value of the data in the input vector, min is the minimum value of the data in the input vector, x is the numerical value of the current point, and x is the numerical value after the normalization conversion calculation.
As a further improvement of the present invention, step S5 specifically includes taking irradiance, temperature, and historical generated power (including but not limited to irradiance and temperature) in the training set as input quantities, taking generated power (including but not limited to generated power) as output quantities, optimizing the weight and threshold of the BP neural network model by using a firework algorithm, and updating the weight and threshold of the BP neural network to obtain a prediction model based on the FWA-BP neural network.
As a further improvement of the present invention, the optimizing step of the weight and the threshold based on the FWA-BP neural network specifically comprises:
a. initializing the position of a population, and setting the upper limit and the lower limit of the number of fireworks explosion sparks;
b. calculating the fitness value of each firework individual, and according to a formula:
Figure BDA0002668976660000033
and
Figure BDA0002668976660000034
calculating the number of fireworks generated by each fireworks explosion and the explosion radius;
wherein, ymaxAnd yminRespectively as the maximum fitness value and the minimum fitness value in the current population,
Figure BDA0002668976660000035
is a constant for adjusting the size of the explosion radius, MIs also a constant used for adjusting the number of fireworks produced by explosion, is a minimum amount of a machine and is used for avoiding zero operation.
c. Performing explosion difference operation according to the formula exik＝xik+ h and mxik＝xikBy x e for generating individual fireworksThe position offset of the spark and the Gaussian variant spark after Gaussian variation is expressed by the formula h as AiX rand (1, -1) is obtained by calculation; wherein h is a position offset, xikIs the kth dimension, ex, of the ith firework individual in the populationikFor the spark after the explosion of the ith individual firework, mxikIs xikE is a random number which follows Gaussian distribution with the mean value of 1 and the variance of 1;
d. selecting N firework individuals from the firework, the explosion spark and the Gaussian variation spark population to form a candidate population;
e. judging whether a termination condition is met, if so, executing the next step, otherwise, skipping to the step b;
f. outputting the global optimal individual and the adaptive value, and ending the algorithm;
g. and substituting the obtained optimal weight and the threshold into a BP neural network for training to obtain an optimal prediction model.
As a further improvement of the present invention, the method for selecting the candidate population in step d comprises: selecting min (f (x) with minimum fitness valuei) ) individual xkDirectly adopting a roulette mode for the next-generation firework population individuals and the rest N-1 firework individuals.
As a further improvement of the present invention, the update formula of the weight and the threshold of the BP neural network in step S5 is specifically:
Figure BDA0002668976660000041
ωjk＝ωjk+ηHjek,j＝1,2,...,l；k＝1,2,...,m
Figure BDA0002668976660000042
bk＝bk+ηek,k＝1,2,...,m
wherein, ω isijA is the connection weight between the input layer and the hidden layer and the threshold value of the hidden layer,ωjkb is the connection weight and threshold between the hidden layer and the output layer, HjFor hidden layer output, η is the learning rate.
As a further improvement of the present invention, step S6 is to save the model file after the training is finished, test the data in the test set, and use the formula ek＝Yk-OkK 1, 2.. m, are checked for errors and the model is repeatedly validated and optimized.
The invention has the beneficial effects that: according to the method, after various influencing factors are compared with the correlation coefficient of the photovoltaic power generation power, a proper model is selected for input, sample data is preprocessed, and a prediction model is established by adopting the FWA-BP neural network, so that the short-term power prediction of the photovoltaic power station is realized, the problems that the traditional BP neural network is easy to fall into local optimum and the convergence speed is low are solved, and the high-precision prediction can be realized.
Drawings
FIG. 1 is an overall flow diagram of the present invention;
FIG. 2 is a prediction flow chart for step 5;
FIG. 3 is a diagram of a BP neural network topology;
FIG. 4 is a comparison of different algorithm power prediction results.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in detail with reference to the accompanying drawings and specific embodiments.
Compared with the BP neural network model and the GA-BP neural network model, the photovoltaic power prediction method based on the FWA-BP neural network has the advantages of high precision, high convergence speed and difficulty in falling into a local minimum value. As shown in fig. 1, the photovoltaic power prediction method of the FWA-BP neural network specifically includes:
step 1: acquiring historical meteorological data and historical output power data of a photovoltaic power station;
and intercepting data of the year 2018 in the whole year, and collecting irradiance, temperature, humidity and wind speed of 8: 00-17: 00 every 15min in one day and historical generated power of a corresponding photovoltaic power station every 15min as a data set. And selecting the first 80% of data in the data set as a training set for training the prediction model, and selecting the second 20% of data in the data set as a test set for testing and verifying the model.
Step 2: preprocessing historical meteorological data and historical output power data;
the preprocessing comprises the detection of abnormal data and the filling of abnormal values, the detection of the abnormal data adopts a 3 sigma principle, the filling of the abnormal values adopts a K neighbor method, and the calculation formula of the K neighbor method is as follows:
Figure BDA0002668976660000061
in the formula, Xi-kIs the k-th data preceding the outlier, Xi+kIs the k-th data after the outlier.
And step 3: selecting data influencing short-term prediction of photovoltaic power from historical meteorological data as an input vector;
calculating a correlation coefficient of historical meteorological data and historical output power data by using a Pearson similarity analysis method, selecting two values with the highest correlation coefficient as input vectors, wherein the calculation formula of the Pearson similarity analysis method is as follows:
Figure BDA0002668976660000062
x, Y, wherein r is a correlation coefficient, and r is a term of the historical meteorological data and the historical output power.
And 4, step 4: normalizing the input vector;
the normalization process normalizes the input vector to the interval of [0, 1], and the calculation formula is as follows:
Figure BDA0002668976660000063
wherein max is the maximum value of the data in the input vector, min is the minimum value of the data in the input vector, x is the numerical value of the current point, and x is the numerical value after the normalization conversion calculation.
And 5: performing off-line training by using a BP neural network optimized by a Firework algorithm (FWA) to create a prediction model based on the FWA-BP neural network;
as shown in fig. 2, the step 5 specifically includes the following steps:
1. determining the topological structure of the BP network; the input quantity of the prediction model comprises irradiance, temperature and historical photovoltaic power of two days before the prediction day, and irradiance and temperature of the day after the prediction day; the sunlight voltage power is predicted to be used as output quantity, so that the number of input layer nodes of the BP neural network model is 8, the number of output layer nodes is 1, and the number of hidden layer nodes is represented by a formula
Figure BDA0002668976660000064
(m and n represent the number of nodes of an input layer and the number of nodes of an output layer of the neural network respectively, and a is a constant between 0 and 10) determining the range, and setting the number of the hidden nodes as 4 through repeated tests. The topology of the BP neural network model is 8-4-1, as shown in FIG. 3.
2. Setting an activation function of the neural network model; in a prediction model of a neural network, sigmoid functions are adopted for activation functions of an input layer and an output layer, and a gradient descent method is mainly adopted for training of the neural network.
3. The firework algorithm optimizes the initial weight and the threshold of the BP neural network, and the method specifically comprises the following steps:
a. initializing the position of the population, and setting the upper limit and the lower limit of the number of fireworks explosion sparks. The population size determines the diversity of the sparks, the larger the population, the more the spark types, but the longer the running time, and the population size is generally selected to be 5. The spark upper and lower limits are used to limit the number of sparks, so that each spark can explode a new number of sparks, the value of which is related to the spark adjustment constant.
b. Calculating the fitness value of each firework individual, and according to a formula:
Figure BDA0002668976660000071
and
Figure BDA0002668976660000072
and calculating the number of fireworks generated by each fireworks explosion and the explosion radius.
Figure BDA0002668976660000073
The method is used for adjusting the size of the explosion radius, the radius adjusting parameter determines the range and the vibration amplitude of spark explosion, the larger the radius adjusting parameter is, the stronger the global exploration capacity is, but the weakened local searching capacity is, and generally 20 is taken. MThe number of the sparks is used for adjusting the number of the fireworks generated by explosion, the number of the sparks determines the number of the sparks generated by explosion, the greater the number is, the greater the possibility of finding the optimal solution is, but the longer the time is required, and in order to reduce the running time, the value of the number of the sparks is smaller.
c. Performing explosion difference operation according to the formula exik＝xik+ h and mxik＝xikX e generating the sparks of the fireworks individuals and the Gaussian variation sparks after Gaussian variation, and the position offset is calculated by the formula h ═ Ai×rand(1,-1)；
Wherein h is a position offset, xikIs the kth dimension, ex, of the ith firework individual in the populationikFor the spark after the explosion of the ith individual firework, mxikIs xikThe Gaussian variant spark after Gaussian variation, wherein e is a random number which follows Gaussian distribution with a mean value of 1 and a variance of 1, has the value of e-N (1, 1). Find exikAnd mxikThe optimal position of the weight is the optimal parameter of the weight of the neural network. The number of variant sparks is mainly used to enhance the global exploration capability.
d. And selecting N firework individuals from the firework, the explosion sparks and the Gaussian variation spark population to form a candidate population. The selection method comprises the following steps: selecting min (f (x) with minimum fitness valuei) ) individual xkDirectly adopting a roulette mode for the next-generation firework population individuals and the rest N-1 firework individuals.
e. And (c) judging whether a termination condition is met, if so, executing the next step, and otherwise, skipping to the step (b).
f. And outputting the global optimal individual and the adaptive value, and finishing the algorithm.
g. And substituting the obtained optimal weight and the threshold into a BP neural network for training to obtain an optimal prediction model.
In the comparison experiment with the BP neural network model and the GA-BP neural network model, the iteration times of all models are 100, and the key parameters of the smoke algorithm are shown in the table 1:
TABLE 1
Parameter name Description of the parameters Parameter value
N Size of fireworks group 5
d Fireworks explosion radius regulating constant 20
m Firework explosion spark number regulating constant 8
Im Upper bound value of number of fireworks explosion sparks 5
bm Lower bound value of number of fireworks explosion sparks 1
g Number of sparks of Gaussian variation 5
T Maximum number of iterations 100
4. The update formula of the weight and the threshold of the BP neural network is as follows:
Figure BDA0002668976660000081
ωjk＝ωjk+ηHjek,j＝1,2,...,l；k＝1,2,...,m
Figure BDA0002668976660000091
bk＝bk+ηek,k＝1,2,...,m
wherein, ω isijA is the connection weight between the input layer and the hidden layer and the threshold value of the hidden layer, omegajkB is the connection weight and threshold between the hidden layer and the output layer, HjFor hidden layer output, η is the learning rate.
The network learning rate is one of key parameters of the neural network, the learning rate is too low, the change speed of the loss function is slower, the convergence time is longer, and the local optimization is easy to fall into; the learning rate is too high, learning can be accelerated in the early stage of algorithm optimization, so that the model is easier to approach to a local or global optimal solution, but large fluctuation exists in the later stage, and even the situation that the value of the loss function is wandering near the optimal solution occurs. The momentum factor is mainly used for accelerating the convergence speed of the network, and the network learning rate and the momentum factor are selected and debugged for many times in the research. The iteration times mainly consider whether the running time and the network are sufficiently updated in an iterative mode, the running time is too long due to too large iteration times, the iteration times are too small, the running is fast, the iteration is insufficient, the network has an optimized space, and the iteration times are generally 100-500 times according to the running condition. The parameter settings are shown in table 2:
TABLE 2
Parameter name Description of the parameters Parameter value
Ir Network learning rate 0.02
mc Accessory momentum factor 0.09
epochs Training maximum number of iterations 200
goal Minimum error of training target 0.001
Step 6: and predicting the photovoltaic power generation power according to the prediction model.
After the training in the step 5 is finished, saving the model file, testing the data in the test set, and performing error index check on the prediction result, wherein the check formula is as follows:
ek＝Yk-Ok,k＝1,2,...,m
wherein, YkAs a predictor of the network, OkAnd (4) checking the error index for the actual value, and if the specified error performance index is not met and the iteration number does not reach the specified upper limit value, repeating the step (4) until a termination condition is met.
As shown in fig. 4: and respectively carrying out experiments on the BP neural network model, the GA-BP neural network model and the FWA-BP neural network model, wherein the experiments adopt the same experiment environment and iteration times. Wherein the model prediction error index pair is shown in table 3 below:
TABLE 3
BP GA-BP FWA-BP
RMSE (root mean square error) 0.80 0.73 0.59
MAE (mean absolute error) 0.65 0.58 0.51
MSE (mean square error) 0.64 0.53 0.35
MAPE% (mean absolute percentage error) 118.14 33.80 6.01
It can be seen from the comparison of the power prediction results of the different algorithms in fig. 4 that: compared with the traditional BP neural network and GA-BP neural network, the FWA optimized BP neural network model provided by the invention has the advantages that the prediction result of the photovoltaic power generation power is closer to the actual value, the prediction of the photovoltaic power generation power can be better realized, and further through the specific prediction error index analysis, as listed in the table 3, the error of the FWA-BP neural network is the minimum under the four error indexes, and the model performance is the best.
Although the present invention has been described in detail with reference to the preferred embodiments, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted for elements thereof without departing from the spirit and scope of the present invention.

Claims (10)

1. A photovoltaic power prediction method based on an FWA-BP neural network is characterized by comprising the following steps:
step 1, acquiring historical meteorological data and historical output power data of a photovoltaic power station;
step 2, preprocessing the historical meteorological data and the historical output power data;
step 3, selecting data influencing short-term prediction of photovoltaic power from the historical meteorological data as an input vector;
step 4, normalizing the input vector;
step 5, performing off-line training by using the BP neural network optimized by the Firework algorithm (FWA) to create a prediction model based on the FWA-BP neural network;
and 6, predicting the photovoltaic power generation power according to the prediction model.
2. The FWA-BP neural network-based photovoltaic power prediction method according to claim 1, wherein: step 1 specifically includes acquiring historical meteorological data and historical output power data which affect photovoltaic output power and forming a data set, wherein the historical meteorological data comprises irradiance, temperature, humidity and wind speed.
3. The FWA-BP neural network-based photovoltaic power prediction method according to claim 1, wherein: the preprocessing in the step 2 comprises detection of abnormal data and filling of abnormal values, the detection of the abnormal data adopts a 3 sigma principle, the filling of the abnormal values adopts a K neighbor method, and a calculation formula of the K neighbor method is as follows:
Figure FDA0002668976650000011
wherein, Xi-kIs the k-th data preceding the outlier, Xi+kIs the k-th data after the outlier.
4. The FWA-BP neural network-based photovoltaic power prediction method according to claim 1, wherein: in step 3, a Pearson similarity analysis method is adopted to calculate correlation coefficients of the historical meteorological data and the historical output power data, two values with the highest correlation coefficients are selected as input vectors, and a calculation formula of the Pearson similarity analysis method is as follows:
Figure FDA0002668976650000021
wherein X, Y are the historical meteorological data and the historical output power, respectively.
5. The FWA-BP neural network-based photovoltaic power prediction method according to claim 1, wherein: normalizing the input vector to the interval of [0, 1] by the normalization processing in the step 4, wherein the calculation formula of the normalization processing is as follows:
Figure FDA0002668976650000022
wherein max is the maximum value of the data in the input vector, min is the minimum value of the data in the input vector, x is the numerical value of the current point, and x is the numerical value after the normalization conversion calculation.
6. The FWA-BP neural network-based photovoltaic power prediction method according to claim 2, wherein: and 5, specifically, taking the irradiance, the temperature and the historical generated power in the training set as input quantities, taking the generated power as output quantities, optimizing the weight and the threshold of the BP neural network model by using a firework algorithm, updating the weight and the threshold of the BP neural network, and obtaining the prediction model based on the FWA-BP neural network.
7. The photovoltaic power prediction method based on the FWA-BP neural network according to claim 6, wherein the optimization steps of the weight and the threshold based on the BP neural network specifically include:
a. initializing the position of a population, and setting the upper limit and the lower limit of the number of fireworks explosion sparks;
b. calculating the fitness value of each firework individual and calculating the fitness value according to a formula
Figure FDA0002668976650000023
And
Figure FDA0002668976650000024
calculating the number of fireworks generated by each fireworks explosion and the explosion radius;
wherein, ymaxAnd yminRespectively as the maximum fitness value and the minimum fitness value in the current population,
Figure FDA0002668976650000025
is a constant for adjusting the size of the explosion radius, MIs a constant for adjusting the number of fireworks generated by explosion, is a machine minimum amount and is used for avoiding zero operation;
c. performing explosion difference operation according to the formula exik＝xik+ h and mxik＝xikXe generating the sparks of the fireworks and the Gaussian variant sparks after Gaussian variant, and the position offset is AiX rand (1, -1), wherein h is the position offset, xikIs the kth dimension, ex, of the ith firework individual in the populationikFor the spark after the explosion of the ith individual firework, mxikIs xikE is a random number which follows Gaussian distribution with the mean value of 1 and the variance of 1;
d. selecting N firework individuals from the firework, the explosion spark and the Gaussian variation spark population to form a candidate population;
e. judging whether a termination condition is met, if so, executing the next step, otherwise, skipping to the step b;
f. outputting the global optimal individual and the adaptive value, and ending the algorithm;
g. and substituting the obtained optimal weight and the threshold into a BP neural network for training to obtain an optimal prediction model.
8. The FWA-BP neural network-based photovoltaic power prediction of claim 7The method is characterized in that the selection method of the candidate population in the step d is as follows: selecting min (f (x) with minimum fitness valuei) ) individual xkDirectly adopting a roulette mode for the next-generation firework population individuals and the rest N-1 firework individuals.
9. The FWA-BP neural network-based photovoltaic power prediction method according to claim 6, wherein: the update formula of the weight and the threshold of the BP neural network is as follows:
Figure FDA0002668976650000031
ωjk＝ωjk+ηHjek,j＝1,2,...,l；k＝1,2,...,m
Figure FDA0002668976650000032
bk＝bk+ηek,k＝1,2,...,m
wherein, ω isijA is the connection weight between the input layer and the hidden layer and the threshold value of the hidden layer, omegajkB is the connection weight and threshold between the hidden layer and the output layer, HjFor hidden layer output, η is the learning rate.
10. The photovoltaic power prediction method based on the FWA-BP neural network according to claim 2, wherein the step 6 specifically comprises: after the training is finished in the step 5, saving the model file, testing the data in the test set and passing a formula ek＝Yk-OkK 1, 2.. m, are checked for errors and the model is repeatedly validated and optimized.
CN202010927605.0A 2020-09-07 2020-09-07 Photovoltaic power prediction method based on FWA-BP neural network Withdrawn CN112149883A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010927605.0A CN112149883A (en) 2020-09-07 2020-09-07 Photovoltaic power prediction method based on FWA-BP neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010927605.0A CN112149883A (en) 2020-09-07 2020-09-07 Photovoltaic power prediction method based on FWA-BP neural network

Publications (1)

Publication Number Publication Date
CN112149883A true CN112149883A (en) 2020-12-29

Family

ID=73889348

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010927605.0A Withdrawn CN112149883A (en) 2020-09-07 2020-09-07 Photovoltaic power prediction method based on FWA-BP neural network

Country Status (1)

Country Link
CN (1) CN112149883A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112733720A (en) * 2021-01-12 2021-04-30 上海理工大学 Face recognition method based on firework algorithm improved convolutional neural network
CN113034310A (en) * 2021-04-16 2021-06-25 国网黑龙江省电力有限公司电力科学研究院 Photovoltaic power generation output power prediction method based on optimized BP neural network
CN113151842A (en) * 2021-01-29 2021-07-23 河北建投新能源有限公司 Method and device for determining conversion efficiency of wind-solar complementary water electrolysis hydrogen production
CN116451824A (en) * 2022-12-16 2023-07-18 西北工业大学 Photovoltaic power prediction method based on neural network and honeypot optimization
CN117638929A (en) * 2024-01-26 2024-03-01 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) Photovoltaic power generation prediction method based on clustering algorithm fusion

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112733720A (en) * 2021-01-12 2021-04-30 上海理工大学 Face recognition method based on firework algorithm improved convolutional neural network
CN113151842A (en) * 2021-01-29 2021-07-23 河北建投新能源有限公司 Method and device for determining conversion efficiency of wind-solar complementary water electrolysis hydrogen production
CN113034310A (en) * 2021-04-16 2021-06-25 国网黑龙江省电力有限公司电力科学研究院 Photovoltaic power generation output power prediction method based on optimized BP neural network
CN116451824A (en) * 2022-12-16 2023-07-18 西北工业大学 Photovoltaic power prediction method based on neural network and honeypot optimization
CN116451824B (en) * 2022-12-16 2024-08-20 西北工业大学 Photovoltaic power prediction method based on neural network and honeypot optimization
CN117638929A (en) * 2024-01-26 2024-03-01 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) Photovoltaic power generation prediction method based on clustering algorithm fusion

Similar Documents

Publication Publication Date Title
CN109117951B (en) BP neural network-based probability load flow online calculation method
CN112149883A (en) Photovoltaic power prediction method based on FWA-BP neural network
Ge et al. A hybrid model for short-term PV output forecasting based on PCA-GWO-GRNN
CN108205717A (en) A kind of photovoltaic generation power Multiple Time Scales Forecasting Methodology
CN111815033A (en) Offshore wind power prediction method based on RCNN and meteorological time sequence characteristics
CN110942205B (en) Short-term photovoltaic power generation power prediction method based on HIMVO-SVM
CN110852902A (en) Photovoltaic power generation power prediction method based on BAS-BP
CN113468817B (en) Ultra-short-term wind power prediction method based on IGOA (insulated gate bipolar transistor) optimized ELM (ELM)
CN103105246A (en) Greenhouse environment forecasting feedback method of back propagation (BP) neural network based on improvement of genetic algorithm
CN106503867A (en) A kind of genetic algorithm least square wind power forecasting method
CN110675278A (en) Photovoltaic power short-term prediction method based on RBF neural network
CN113722980B (en) Ocean wave height prediction method, ocean wave height prediction system, computer equipment, storage medium and terminal
CN108182500A (en) Ammunition Storage Reliability Forecasting Methodology based on accelerated life test
CN115275991A (en) Active power distribution network operation situation prediction method based on IEMD-TA-LSTM model
CN115099296A (en) Sea wave height prediction method based on deep learning algorithm
CN113095477B (en) Wind power prediction method based on DE-BP neural network
CN115995810A (en) Wind power prediction method and system considering weather fluctuation self-adaptive matching
CN113435595A (en) Two-stage optimization method for extreme learning machine network parameters based on natural evolution strategy
CN116341717A (en) Wind speed prediction method based on error compensation
CN113392972B (en) Photovoltaic short-term power prediction model training method, prediction method and device
Xu et al. Short-term wind speed prediction based on GRU
CN110276478B (en) Short-term wind power prediction method based on segmented ant colony algorithm optimization SVM
CN113033898A (en) Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network
CN118428754A (en) New energy power generation input power prediction method and system
CN111950696A (en) Short-term power load prediction method based on dimension reduction and improved neural network

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
WW01 Invention patent application withdrawn after publication
WW01 Invention patent application withdrawn after publication

Application publication date: 20201229