Photovoltaic energy storage generation power prediction method
The hyperparameters of the photovoltaic power generation prediction model are optimized through the niche bacteria optimization algorithm, and combined with real-time sky images and ambient temperature, the problem of low accuracy of photovoltaic power generation prediction in the existing technology is solved, achieving better prediction capabilities and Training efficiency.
Patent Information
- Application Number
- CN202411212910.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-30
- Publication Date
- 2024-12-06
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing photovoltaic power prediction methods have low accuracy on short time scales, making it difficult to set appropriate model hyperparameters, resulting in poor prediction and generalization capabilities.
The microbe flora optimization algorithm is used to optimize the hyperparameters of the power generation power prediction model, and combine real-time sky images and ambient temperature to calculate the real-time position information of the clouds and the sun, and input the trained power generation power prediction model for prediction.
It improves the prediction and generalization capabilities of the model, and can more accurately predict photovoltaic power fluctuations in a short period of time, improving model training efficiency.
Smart Images

Figure CN119093354A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to power generation prediction, and in particular to a method for predicting power generation from photovoltaic energy storage. Background Art
[0002] In the existing photovoltaic power prediction methods, meteorological factor vectors are mainly established through numerical weather forecasts, and the solar irradiation distribution or photovoltaic power output on similar days is obtained through vector calculation and other methods. However, numerical weather forecast data cannot describe environmental changes on a smaller time scale. The change in solar irradiation caused by cloud movement may cause drastic power fluctuations within a few seconds, and conventional solar irradiation prediction technology is difficult to describe such short-term solar irradiation changes, resulting in low accuracy in photovoltaic power prediction.
[0003] The use of photovoltaic power prediction methods based on deep learning can be used as a way to solve the above problems. However, this type of method has the problem of difficulty in setting appropriate model hyperparameters, resulting in poor prediction and generalization capabilities of the model, as well as low model training efficiency. Summary of the invention
[0004] 1. Technical issues to be resolved
[0005] In view of the above-mentioned shortcomings of the prior art, the present invention provides a method for predicting photovoltaic energy storage power generation, which can effectively overcome the defects of the prior art, such as low accuracy of photovoltaic power generation prediction in a short period of time due to failure to consider cloud movement, and difficulty in setting appropriate model hyperparameters.
[0006] (II) Technical solution
[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions:
[0008] A photovoltaic energy storage power generation prediction method, which constructs a power generation prediction model, optimizes the hyperparameters of the power generation prediction model using a microhabitat bacterial community optimization algorithm, and performs model training on the optimized power generation prediction model to obtain a trained power generation prediction model;
[0009] The sky image and ambient temperature above the photovoltaic energy storage power station are collected in real time. The moving speed and direction of the clouds above the photovoltaic energy storage power station are calculated based on the sky image, and the real-time position information of the clouds is estimated. At the same time, the corresponding real-time position information of the sun is calculated. The real-time position information of the clouds, the real-time position information of the sun and the real-time ambient temperature are input into the trained power generation prediction model to obtain the power generation prediction result.
[0010] Preferably, the optimization of the hyperparameters of the power generation prediction model using the microhabitat flora optimization algorithm includes:
[0011] S11, randomly generating a certain number of bacterial individuals in the search space, each bacterial individual has its own initial position and initial velocity vector, and these bacterial individuals together constitute an initial bacterial population;
[0012] S12, obtaining initial hyperparameters of the power generation prediction model, using the initial hyperparameters as initial positions of bacterial individuals in the initial bacterial community, and initializing parameters of the microhabitat bacterial community optimization algorithm;
[0013] S13, calculating the distance between individual bacteria, and dividing the initial bacterial population into multiple microhabitat groups according to the distance between individual bacteria and the preset microhabitat radius;
[0014] S14, performing local optimization;
[0015] S15, performing global optimization;
[0016] S16, judging whether the iteration termination condition is met, if the iteration termination condition is not met, returning to S14, otherwise taking the current global optimal solution as the hyperparameter of the optimized power generation prediction model;
[0017] Among them, the power generation prediction model is built based on BP neural network.
[0018] Preferably, the local optimization is performed in S14, including:
[0019] S141. In each microhabitat population, the position of each bacterial individual is updated, and the fitness value of each bacterial individual is calculated;
[0020] S142, by modifying the original cost function and introducing a penalty function, the bacterial individuals try to attract the surrounding neighbor individuals to gather in the area with higher fitness values during the search process, so as to encourage the bacterial individuals to move closer to the local optimal solution;
[0021] S143, repeating S141 and S142, after reaching the preset number of times, in each microhabitat population, sorting according to the fitness value, and selecting bacterial individuals with higher fitness values for reproduction, while eliminating bacterial individuals with lower fitness values.
[0022] Preferably, in S141, in each microhabitat population, the position of each bacterial individual is updated, including:
[0023] The following formula is used to update the position of bacterial individuals:
[0024]
[0025] in, is the position of bacterial individual i at the tth iteration, is the position of bacterial individual i at the t+1th iteration, is the speed of bacterial individual i at the tth iteration, is the swimming direction of bacterial individual i at the tth iteration;
[0026] The following formula is used to update the speed of individual bacteria:
[0027]
[0028] in, is the speed of bacterial individual i at the t+1th iteration, ε v is the speed variation rate during swimming, χ F (t) is the difference in fitness values of neighboring individuals around the bacterial individual at the tth iteration;
[0029] The following formula is used to update the swimming direction of individual bacteria:
[0030]
[0031] in, is the swimming direction of bacterial individual i at the t+1th iteration, ε D is the directional mutation rate during the swimming process, r is a random number, r∈[0,1], and R is a neighbor individual randomly selected in the neighborhood of bacterial individual i.
[0032] Preferably, the global optimization is performed in S15, including:
[0033] S151. After each microhabitat group completes reproduction, the fitness value of each bacterial individual is adjusted using the fitness value sharing mechanism to limit the large increase of individual bacterial individuals, maintain group diversity, and form a microhabitat evolutionary environment;
[0034] S152. In each microhabitat population, a penalty function is used to penalize the bacterial individual with the lowest fitness value to avoid falling into a local optimal solution.
[0035] Preferably, in S151, after each microhabitat population completes reproduction, the fitness value of each bacterial individual is adjusted using a fitness value sharing mechanism, including:
[0036] S1511, determine the shared function S(d):
[0037]
[0038] Among them, d is the distance between bacterial individuals in the microhabitat population, and σ is the preset microhabitat radius;
[0039] S1512, calculating the sharing degree:
[0040]
[0041] Among them, S i is the sharing degree of bacterial individual i, d ij is the distance between bacterial individual i and bacterial individual j, and N is the number of bacterial individuals in the microhabitat group where bacterial individual i is located;
[0042] S1513. Adjust the fitness value of individual bacteria:
[0043]
[0044] Among them, F i is the original fitness value of bacterial individual i, F i ' is the adjusted fitness value of bacterial individual i.
[0045] Preferably, the performing model training on the optimized power generation prediction model to obtain a trained power generation prediction model includes:
[0046] S21, dividing the historical data set into a training set, a validation set, and a test set according to a preset ratio;
[0047] S22, setting the loss function and optimizer of the power generation prediction model, and inputting the training set into the power generation prediction model for model training;
[0048] S23, calculating the loss value based on the loss function, and the optimizer updates the model parameters according to the loss value and the network gradient information;
[0049] S24. If the loss value is less than the preset threshold, the model training ends, and the current power generation prediction model is the trained power generation prediction model. Otherwise, the process returns to S22 and continues to use the training set for model training.
[0050] S25. Input the validation set into the trained power generation prediction model, evaluate the generalization ability of the model by observing the performance on the validation set, and tune the hyperparameters and structure of the model;
[0051] S26. Input the test set into the optimized power generation prediction model to evaluate the model performance.
[0052] Preferably, before dividing the historical data set into a training set, a validation set, and a test set according to a preset ratio in S21, the following steps are included:
[0053] S211, collecting historical sky images, historical ambient temperatures, and corresponding historical power generation above the photovoltaic energy storage power station within a preset historical time period;
[0054] S212, calculating the moving speed and direction of the clouds above the photovoltaic energy storage power station based on the historical sky images, estimating the cloud position information, and calculating the corresponding sun position information;
[0055] S213: Use the historical power generation to set labels for the corresponding cloud position information, sun position information and historical ambient temperature to form a historical data set.
[0056] Preferably, the step of calculating the moving speed and direction of clouds above the photovoltaic energy storage power station based on the sky image and estimating the real-time position information of the clouds includes:
[0057] S31, after performing distortion correction on the sky image, scaling processing is performed through a feature pyramid;
[0058] S32, performing feature point detection on the preprocessed sky image, and matching the feature points with the sky image at the previous moment, removing the feature points that are mismatched, and obtaining multiple groups of feature point pairs;
[0059] S33, calculating the number of moving pixels of a set of feature point pairs projected in the same sky image as the cloud movement distance, and calculating the cloud movement speed based on the image sampling frequency, and taking the angle formed by the line connecting the set of feature points as the cloud movement direction;
[0060] S34, averaging the calculated cloud movement speed and direction for all feature points, and finally obtaining the cloud movement speed and direction above the photovoltaic energy storage power station, and estimating the real-time position information of the cloud.
[0061] Preferably, the calculation to obtain the corresponding real-time sun position information includes:
[0062] The real-time position information of the sun is calculated using the following formula:
[0063]
[0064] Among them, h is the solar altitude angle, A is the solar azimuth angle, is the latitude of the observation site, δ is the solar declination, and α is the solar hour angle.
[0065] (III) Beneficial effects
[0066] Compared with the prior art, the photovoltaic energy storage power generation power prediction method provided by the present invention has the following beneficial effects:
[0067] 1) Construct a power generation prediction model and use the microhabitat microbial community optimization algorithm to optimize the hyperparameters of the power generation prediction model, which can quickly find the model hyperparameters that are compatible with the photovoltaic power generation prediction target, so that the model has better prediction and generalization capabilities, and can also improve the model training efficiency;
[0068] 2) Collect sky images and ambient temperature above the photovoltaic energy storage power station in real time, calculate the moving speed and direction of the clouds above the photovoltaic energy storage power station based on the sky images, estimate the real-time position information of the clouds, and calculate the corresponding real-time position information of the sun. The real-time position information of the clouds and the real-time position information of the sun can reflect the direct impact of cloud movement on solar irradiation changes in real time, so that the power generation prediction model can better learn the relationship between the short-term solar irradiation changes caused by cloud movement and photovoltaic power generation, and combine the real-time ambient temperature to realize the accurate prediction of photovoltaic power generation in a short time. BRIEF DESCRIPTION OF THE DRAWINGS
[0069] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0070] Figure 1 It is a schematic diagram of the process of the present invention;
[0071] Figure 2 It is a schematic diagram of the process of optimizing the hyperparameters of the power generation prediction model using the microhabitat flora optimization algorithm in the present invention;
[0072] Figure 3 It is a schematic diagram of the process of using the trained power generation prediction model to perform real-time power generation prediction in the present invention. DETAILED DESCRIPTION
[0073] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0074] A photovoltaic energy storage power generation power prediction method, such as Figure 1 As shown, a power generation prediction model is constructed, a microhabitat flora optimization algorithm is used to optimize the hyperparameters of the power generation prediction model, and the optimized power generation prediction model is trained to obtain a trained power generation prediction model.
[0075] ① Such as Figure 2As shown in the figure, the microhabitat bacterial community optimization algorithm is used to optimize the hyperparameters of the power generation prediction model, including:
[0076] S11, randomly generating a certain number of bacterial individuals in the search space, each bacterial individual has its own initial position and initial velocity vector, and these bacterial individuals together constitute an initial bacterial population;
[0077] S12, obtaining initial hyperparameters of the power generation prediction model, using the initial hyperparameters as initial positions of bacterial individuals in the initial bacterial community, and initializing parameters of the microhabitat bacterial community optimization algorithm;
[0078] S13, calculating the distance between individual bacteria, and dividing the initial bacterial population into multiple microhabitat groups according to the distance between individual bacteria and the preset microhabitat radius;
[0079] S14, performing local optimization;
[0080] S15, performing global optimization;
[0081] S16, judging whether the iteration termination condition is met, if the iteration termination condition is not met, returning to S14, otherwise taking the current global optimal solution as the hyperparameter of the optimized power generation prediction model;
[0082] Among them, the power generation prediction model is built based on BP neural network.
[0083] 1) Local optimization is performed in S14, including:
[0084] S141. In each microhabitat population, the position of each bacterial individual is updated, and the fitness value of each bacterial individual is calculated;
[0085] S142, by modifying the original cost function and introducing a penalty function, the bacterial individuals try to attract the surrounding neighbor individuals to gather in the area with higher fitness values during the search process, so as to encourage the bacterial individuals to move closer to the local optimal solution;
[0086] S143, repeating S141 and S142, after reaching the preset number of times, in each microhabitat population, sorting according to the fitness value, and selecting bacterial individuals with higher fitness values for reproduction, while eliminating bacterial individuals with lower fitness values.
[0087] Specifically, in S141, the position of each bacterial individual in each microhabitat population is updated, including:
[0088] The following formula is used to update the position of bacterial individuals:
[0089]
[0090] in, is the position of bacterial individual i at the tth iteration, is the position of bacterial individual i at the t+1th iteration, is the speed of bacterial individual i at the tth iteration, is the swimming direction of bacterial individual i at the tth iteration;
[0091] The following formula is used to update the speed of individual bacteria:
[0092]
[0093] in, is the speed of bacterial individual i at the t+1th iteration, ε v is the speed variation rate during swimming, χ F (t) is the difference in fitness values of neighboring individuals around the bacterial individual at the tth iteration;
[0094] The following formula is used to update the swimming direction of individual bacteria:
[0095]
[0096] in, is the swimming direction of bacterial individual i at the t+1th iteration, ε D is the directional mutation rate during the swimming process, r is a random number, r∈[0,1], and R is a neighbor individual randomly selected in the neighborhood of bacterial individual i.
[0097] 2) Global optimization is performed in S15, including:
[0098] S151. After each microhabitat group completes reproduction, the fitness value of each bacterial individual is adjusted using the fitness value sharing mechanism to limit the large increase of individual bacterial individuals, maintain group diversity, and form a microhabitat evolutionary environment;
[0099] S152. In each microhabitat population, a penalty function is used to penalize the bacterial individual with the lowest fitness value to avoid falling into a local optimal solution.
[0100] Specifically, in S151, after each microhabitat group completes reproduction, the fitness value of each bacterial individual is adjusted using the fitness value sharing mechanism, including:
[0101] S1511, determine the shared function S(d):
[0102]
[0103] Among them, d is the distance between bacterial individuals in the microhabitat population, and σ is the preset microhabitat radius;
[0104] S1512, calculating the sharing degree:
[0105]
[0106] Among them, S i is the sharing degree of bacterial individual i, d ij is the distance between bacterial individual i and bacterial individual j, and N is the number of bacterial individuals in the microhabitat group where bacterial individual i is located;
[0107] S1513. Adjust the fitness value of individual bacteria:
[0108]
[0109] Among them, F i is the original fitness value of bacterial individual i, F i ' is the adjusted fitness value of bacterial individual i.
[0110] ② Perform model training on the optimized power generation prediction model to obtain a trained power generation prediction model, including:
[0111] S21, dividing the historical data set into a training set, a validation set, and a test set according to a preset ratio;
[0112] S22, setting the loss function and optimizer of the power generation prediction model, and inputting the training set into the power generation prediction model for model training;
[0113] S23, calculating the loss value based on the loss function, and the optimizer updates the model parameters according to the loss value and the network gradient information;
[0114] S24. If the loss value is less than the preset threshold, the model training ends, and the current power generation prediction model is the trained power generation prediction model. Otherwise, the process returns to S22 and continues to use the training set for model training.
[0115] S25. Input the validation set into the trained power generation prediction model, evaluate the generalization ability of the model by observing the performance on the validation set, and tune the hyperparameters and structure of the model;
[0116] S26. Input the test set into the optimized power generation prediction model to evaluate the model performance.
[0117] Specifically, before dividing the historical data set into a training set, a validation set, and a test set according to a preset ratio in S21, it includes:
[0118] S211, collecting historical sky images, historical ambient temperatures, and corresponding historical power generation above the photovoltaic energy storage power station within a preset historical time period;
[0119] S212, calculating the moving speed and direction of the clouds above the photovoltaic energy storage power station based on the historical sky images, estimating the cloud position information, and calculating the corresponding sun position information;
[0120] S213: Use the historical power generation to set labels for the corresponding cloud position information, sun position information and historical ambient temperature to form a historical data set.
[0121] The above technical solution constructs a power generation prediction model and uses a microhabitat bacterial community optimization algorithm to optimize the hyperparameters of the power generation prediction model. It can quickly find the model hyperparameters that are compatible with the photovoltaic power generation prediction target, so that the model has better prediction and generalization capabilities, and at the same time can improve the model training efficiency.
[0122] like Figure 1 and Figure 3 As shown, the sky image and ambient temperature above the photovoltaic energy storage power station are collected in real time, the moving speed and direction of the clouds above the photovoltaic energy storage power station are calculated based on the sky image, the real-time position information of the clouds is estimated, and the corresponding real-time position information of the sun is calculated at the same time. The real-time position information of the clouds, the real-time position information of the sun and the real-time ambient temperature are input into the trained power generation prediction model to obtain the power generation prediction result.
[0123] ① Calculate the moving speed and direction of the clouds above the photovoltaic energy storage power station based on the sky image, and estimate the real-time position information of the clouds, including:
[0124] S31, after performing distortion correction on the sky image, scaling processing is performed through a feature pyramid;
[0125] S32, performing feature point detection on the preprocessed sky image, and matching the feature points with the sky image at the previous moment, removing the feature points that are mismatched, and obtaining multiple groups of feature point pairs;
[0126] S33, calculating the number of moving pixels of a set of feature point pairs projected in the same sky image as the cloud movement distance, and calculating the cloud movement speed based on the image sampling frequency, and taking the angle formed by the line connecting the set of feature points as the cloud movement direction;
[0127] S34, averaging the calculated cloud movement speed and direction for all feature points, and finally obtaining the cloud movement speed and direction above the photovoltaic energy storage power station, and estimating the real-time position information of the cloud.
[0128] ② Calculate the corresponding real-time sun position information, including:
[0129] The real-time position information of the sun is calculated using the following formula:
[0130]
[0131] Among them, h is the solar altitude angle, A is the solar azimuth angle, is the latitude of the observation site, δ is the solar declination, and α is the solar hour angle.
[0132] The above technical solution collects the sky image and ambient temperature above the photovoltaic energy storage power station in real time, calculates the movement speed and direction of the clouds above the photovoltaic energy storage power station based on the sky image, estimates the real-time position information of the clouds, and simultaneously calculates the corresponding real-time position information of the sun. The real-time position information of the clouds and the real-time position information of the sun can reflect the direct impact of cloud movement on solar irradiation changes in real time, so that the power generation prediction model can better learn the relationship between the short-term solar irradiation changes caused by cloud movement and photovoltaic power generation, and realize accurate prediction of photovoltaic power generation in a short period of time in combination with the real-time ambient temperature.
[0133] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A photovoltaic energy storage power generation power prediction method, characterized by: Constructing a power generation prediction model, optimizing the hyperparameters of the power generation prediction model using a microhabitat microbial community optimization algorithm, and training the optimized power generation prediction model to obtain a trained power generation prediction model; The sky image and ambient temperature above the photovoltaic energy storage power station are collected in real time. The moving speed and direction of the clouds above the photovoltaic energy storage power station are calculated based on the sky image, and the real-time position information of the clouds is estimated. At the same time, the corresponding real-time position information of the sun is calculated. The real-time position information of the clouds, the real-time position information of the sun and the real-time ambient temperature are input into the trained power generation prediction model to obtain the power generation prediction result.
2. The photovoltaic energy storage power generation power prediction method according to claim 1 is characterized in that: The method of optimizing the hyperparameters of the power generation prediction model using the microhabitat bacterial community optimization algorithm includes: S11, randomly generating a certain number of bacterial individuals in the search space, each bacterial individual has its own initial position and initial velocity vector, and these bacterial individuals together constitute an initial bacterial population; S12, obtaining initial hyperparameters of the power generation prediction model, using the initial hyperparameters as initial positions of bacterial individuals in the initial bacterial community, and initializing parameters of the microhabitat bacterial community optimization algorithm; S13, calculating the distance between individual bacteria, and dividing the initial bacterial population into multiple microhabitat groups according to the distance between individual bacteria and the preset microhabitat radius; S14, performing local optimization; S15, performing global optimization; S16, judging whether the iteration termination condition is met, if the iteration termination condition is not met, returning to S14, otherwise taking the current global optimal solution as the hyperparameter of the optimized power generation prediction model; Among them, the power generation prediction model is built based on BP neural network.
3. The photovoltaic energy storage power generation power prediction method according to claim 2 is characterized in that: Local optimization is performed in S14, including: S141. In each microhabitat population, the position of each bacterial individual is updated, and the fitness value of each bacterial individual is calculated; S142, by modifying the original cost function and introducing a penalty function, the bacterial individuals try to attract the surrounding neighbor individuals to gather in the area with higher fitness values during the search process, so as to encourage the bacterial individuals to move closer to the local optimal solution; S143, repeating S141 and S142, after reaching the preset number of times, in each microhabitat population, sorting according to the fitness value, and selecting bacterial individuals with higher fitness values for reproduction, while eliminating bacterial individuals with lower fitness values.
4. The photovoltaic energy storage power generation power prediction method according to claim 3 is characterized by: In S141, the position of each bacterial individual in each microhabitat group is updated, including: The following formula is used to update the position of bacterial individuals: in, is the position of bacterial individual i at the tth iteration, is the position of bacterial individual i at the t+1th iteration, is the speed of bacterial individual i at the tth iteration, is the swimming direction of bacterial individual i at the tth iteration; The following formula is used to update the speed of individual bacteria: in, is the speed of bacterial individual i at the t+1th iteration, ε v is the speed variation rate during swimming, χ F (t) is the difference in fitness values of neighboring individuals around the bacterial individual at the tth iteration; The following formula is used to update the swimming direction of individual bacteria: in, is the swimming direction of bacterial individual i at the t+1th iteration, ε D is the directional mutation rate during the swimming process, r is a random number, r∈[0,1], and R is a neighbor individual randomly selected in the neighborhood of bacterial individual i.
5. The photovoltaic energy storage power generation prediction method according to claim 3 is characterized by: Global optimization is performed in S15, including: S151. After each microhabitat group completes reproduction, the fitness value of each bacterial individual is adjusted using the fitness value sharing mechanism to limit the large increase of individual bacterial individuals, maintain group diversity, and form a microhabitat evolutionary environment; S152. In each microhabitat population, a penalty function is used to penalize the bacterial individual with the lowest fitness value to avoid falling into a local optimal solution.
6. The photovoltaic energy storage power generation prediction method according to claim 5 is characterized by: In S151, after each microhabitat group completes reproduction, the fitness value of each bacterial individual is adjusted using the fitness value sharing mechanism, including: S1511, determine the shared function S(d): Among them, d is the distance between bacterial individuals in the microhabitat population, and σ is the preset microhabitat radius; S1512, calculating the sharing degree: Among them, S i is the sharing degree of bacterial individual i, d ij is the distance between bacterial individual i and bacterial individual j, and N is the number of bacterial individuals in the microhabitat group where bacterial individual i is located; S1513. Adjust the fitness value of individual bacteria: Among them, F i is the original fitness value of bacterial individual i, F i ' is the adjusted fitness value of bacterial individual i.
7. The photovoltaic energy storage power generation power prediction method according to claim 1 is characterized by: The step of training the optimized power generation prediction model to obtain a trained power generation prediction model includes: S21, dividing the historical data set into a training set, a validation set, and a test set according to a preset ratio; S22, setting the loss function and optimizer of the power generation prediction model, and inputting the training set into the power generation prediction model for model training; S23, calculating the loss value based on the loss function, and the optimizer updates the model parameters according to the loss value and the network gradient information; S24. If the loss value is less than the preset threshold, the model training ends, and the current power generation prediction model is the trained power generation prediction model. Otherwise, the process returns to S22 and continues to use the training set for model training. S25. Input the validation set into the trained power generation prediction model, evaluate the generalization ability of the model by observing the performance on the validation set, and tune the hyperparameters and structure of the model; S26. Input the test set into the optimized power generation prediction model to evaluate the model performance.
8. The photovoltaic energy storage power generation power prediction method according to claim 7 is characterized in that: Before S21 divides the historical data set into a training set, a validation set, and a test set according to a preset ratio, it includes: S211, collecting historical sky images, historical ambient temperatures, and corresponding historical power generation above the photovoltaic energy storage power station within a preset historical time period; S212, calculating the moving speed and direction of the clouds above the photovoltaic energy storage power station based on the historical sky images, estimating the cloud position information, and calculating the corresponding sun position information; S213: Use the historical power generation to set labels for the corresponding cloud position information, sun position information and historical ambient temperature to form a historical data set.
9. The photovoltaic energy storage power generation power prediction method according to claim 1 is characterized by: The method of calculating the moving speed and direction of clouds above the photovoltaic energy storage power station based on the sky image and estimating the real-time position information of the clouds includes: S31, after performing distortion correction on the sky image, scaling processing is performed through a feature pyramid; S32, performing feature point detection on the preprocessed sky image, and matching the feature points with the sky image at the previous moment, removing the feature points that are mismatched, and obtaining multiple groups of feature point pairs; S33, calculating the number of moving pixels of a set of feature point pairs projected in the same sky image as the cloud movement distance, and calculating the cloud movement speed based on the image sampling frequency, and taking the angle formed by the line connecting the set of feature points as the cloud movement direction; S34, averaging the calculated cloud movement speed and direction for all feature points, and finally obtaining the cloud movement speed and direction above the photovoltaic energy storage power station, and estimating the real-time position information of the cloud.
10. The photovoltaic energy storage power generation power prediction method according to claim 9 is characterized in that: The calculation to obtain the corresponding real-time sun position information includes: The real-time position information of the sun is calculated using the following formula: Among them, h is the solar altitude angle, A is the solar azimuth angle, is the latitude of the observation site, δ is the solar declination, and α is the solar hour angle.
Citation Information
Patent Citations
Ultra-short-term irradiation prediction method based on typical cloud layer shielding irradiation difference
CN111738327A
Second-level ultra-short-term photovoltaic power prediction method
CN112507613A
Model parameter adaptive GRU new energy short-term generation power prediction method
CN114330934A
Energy storage photovoltaic off-grid micro-grid control method and system aiming at thin cloud layer disturbance
CN118432176A
Method and system for power prediction of photovoltaic power station based on operating data of grid-connected inverters
US20210194424A1
