Photovoltaic station power prediction system and method based on irradiation environment weighting algorithm

Abstract

The invention relates to the technical field of photovoltaic station power prediction, in particular to a photovoltaic station power prediction system and method based on an irradiation environment weighting algorithm, and the method comprises the steps: obtaining wider meteorological information through a multi-dimensional meteorological data collection module, extracting generated power data from the historical operation record of a photovoltaic station, and carrying out the calculation of the generated power data; then, according to the influence degree of different meteorological factors on photovoltaic power generation, an initial weight is distributed for each feature, then a deep learning model is established, fine adjustment is conducted on the deep learning model regularly, and the structure and parameters of the model are adjusted according to latest operation data and feedback information; the method comprises the steps that firstly, a deep learning model is constructed, then a prediction result is output, structural parameter setting of the deep learning model is continuously iterated and improved according to an evaluation result, finally, application scene adaptability adjustment is conducted, the weight is adjusted according to the importance or representativeness of a sample by adopting a dynamic weight distribution strategy, and therefore the overall performance of the model is improved, and prediction accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic station power prediction, and in particular relates to a photovoltaic station power prediction system and method based on an illumination environment weighted algorithm. Background Art

[0002] The photovoltaic power station power prediction system is a technology based on data analysis and model prediction. It predicts the power generation of photovoltaic power stations in the future through comprehensive analysis of historical data, weather forecasts, power station parameters and other information.

[0003] For example, the patent application number is 202010518479.3, which discloses a method and system for predicting the power generation of a photovoltaic station, including constructing a training sample set based on the historical power data of the photovoltaic station and the meteorological data of the corresponding time period, and assigning sample weights to each training sample in the training sample set; using the sub-training sample set generated by the self-service sampling method to train the random forest model, and calculating the error rate and weight coefficient of the random forest model under the current sample weight according to the adaptive enhancement algorithm; updating the sample weight according to the error rate and weight coefficient under the preset number of iterations, training the random forest model in turn, and weighting it according to the weight coefficient to obtain a weighted random forest prediction model; using the weighted random forest prediction model to predict the meteorological data of the prediction time period to obtain the power generation of the photovoltaic station. Fully explore the information in the multi-dimensional features, solve the uncertainty problem in photovoltaic power generation, and improve the reliability and accuracy of photovoltaic station power prediction.

[0004] However, the above prediction method adopts a fixed sample weight allocation strategy, which means that all training samples are given the same weight without considering their importance or representativeness in the overall data set. This method ignores the differences and importance between samples and may cause the model to favor certain specific samples, thereby affecting the accuracy of the prediction. Therefore, we need to propose a photovoltaic station power prediction system and method based on an irradiation environment weighted algorithm to solve the above problems, so that it can adopt a dynamic weight allocation strategy to adjust the weight of the sample according to its importance or representativeness, thereby improving the overall performance of the model and improving the accuracy of the prediction. Summary of the invention

[0005] In view of the above problems, the present invention provides a photovoltaic station power prediction system based on an illumination environment weighted algorithm, comprising: a multi-dimensional meteorological data acquisition module, which collects multi-dimensional meteorological data including light intensity, temperature, wind speed, humidity, cloud coverage and air quality index in real time, and extracts power generation data from the historical operation records of the photovoltaic power station;

[0006] A feature extraction module, which extracts key features from the raw data according to the physical characteristics and operating rules of the photovoltaic power station. The key features include sunshine duration, solar altitude angle and photovoltaic panel tilt angle;

[0007] A dynamic weight allocation module, wherein the dynamic weight allocation module allocates an initial weight to each feature according to the degree of influence of different meteorological factors on photovoltaic power generation;

[0008] A deep model building module, wherein the deep model building module is used to build a deep learning model capable of capturing long-term dependencies and local features in the data;

[0009] An adaptive adjustment module, which is used to introduce an adaptive adjustment mechanism during the deep learning model training process;

[0010] A prediction result output module, which generates real-time power generation prediction results based on current and future meteorological data and power station status information;

[0011] An error evaluation and iterative update module, which continuously and iteratively improves the structural parameter settings of the deep learning model according to the evaluation results;

[0012] The application scenario adaptive adjustment module first customizes the model parameters according to the unique environmental and climatic conditions faced by photovoltaic power stations in different geographical locations, and then adopts different data processing methods and model structures for photovoltaic power stations of different sizes and types, and then customizes the development of corresponding prediction functions and services according to the specific needs and characteristics of the power stations.

[0013] Furthermore, the process of the multi-dimensional meteorological data acquisition module for performing multi-dimensional meteorological data acquisition is as follows:

[0014] A1. Equip the photovoltaic power station with meteorological sensors and remote sensing instruments to monitor the light intensity, temperature, wind speed and humidity data in real time through meteorological sensors, and obtain cloud coverage and air quality index data over a large area through remote sensing instruments;

[0015] A2. Transmit meteorological data, cloud coverage data and air quality index data to the cloud server for storage and management through the wireless network module;

[0016] A3. Integrate with the existing monitoring system of the photovoltaic power station and directly extract historical operation records from the monitoring system;

[0017] A4. Process the collected raw meteorological data, including data cleaning and missing value filling and interpolation.

[0018] Furthermore, the process of extracting key features by the feature extraction module is as follows:

[0019] B1. Check the sunrise and sunset timetable provided by the local weather station and calculate the sunrise and sunset times of each day;

[0020] B2. According to weather conditions and the installation angle of photovoltaic panels, select the effective time period that can receive sunlight, and dynamically adjust the start and end time of the effective light period by combining ground measured data or satellite remote sensing image analysis;

[0021] B3. Input the latitude and longitude coordinates of the location of the photovoltaic power station and the current time to calculate the solar altitude angle;

[0022] B4. Analyze the changes in the incident angle of light at different tilt angles, evaluate its impact on the intensity of light received by the photovoltaic panel, and then combine historical data and simulation results to determine the optimal tilt angle range;

[0023] B5. Integrate the calculated features of sunshine duration, solar altitude angle and photovoltaic panel tilt angle to form a set of comprehensive feature vectors;

[0024] B6. Calculate the correlation between each feature and photovoltaic power generation, and evaluate the effect of feature selection through cross-validation.

[0025] Furthermore, when calculating the solar altitude angle, the solar inclination angle is calculated first, and then the solar hour angle is calculated. Then, the solar altitude angle is calculated based on the solar inclination angle and the solar hour angle. The specific operation is as follows:

[0026] B3.1. Calculate the solar inclination using the solar inclination formula , the formula for the solar inclination is:

[0027] ,in, To convert a Gregorian calendar date to a day of the year;

[0028] B3.2. Calculate the solar hour angle using the solar hour angle calculation formula , the solar hour angle calculation formula is as follows:

[0029] ,T=UTC time+ Where T is the local solar time;

[0030] B3.3. Use the solar altitude angle calculation formula to calculate the solar altitude angle h. The solar altitude angle calculation formula is:

[0031] =arc ,in is the latitude of the observation site;

[0032] B3.4. Convert the solar altitude angle h from radians to degrees using the following conversion formula:

[0033] h.

[0034] Furthermore, the process of the dynamic weight allocation module to allocate initial weights to each feature is as follows:

[0035] C1. Standardize the original data matrix. The standardization formula is as follows:

[0036] , where x is the original data and u is the mean of the data. is the standard deviation of the data, is the standardized data matrix;

[0037] C2. Calculate the jth indicator The proportion of samples , the calculation formula is:

[0038] , where m is the number of samples;

[0039] C3. Calculate the entropy value of the jth indicator , the calculation formula is as follows:

[0040] , where k is a constant, i.e. k= ;

[0041] C4. Calculate the coefficient of difference of the jth indicator and the weight of the jth item , the calculation formula is as follows:

[0042] , ,in is the number of indicators;

[0043] C5. Calculate the correlation between indicators , the calculation formula is as follows:

[0044] ,in, is the minimum value of the standardized data matrix, is the maximum value of the data matrix after standardization, and m is the number of samples;

[0045] C6. Calculate indicator weights , and then the indicator weight and the weight of the jth item Combine and calculate the comprehensive weight , the calculation formula is as follows:

[0046] , ;

[0047] C7. Based on the calculated comprehensive weight , select the element features that are important to the prediction model, use the element features to build the prediction model, and train the prediction model;

[0048] C8. Evaluate model performance using the test dataset.

[0049] Furthermore, the process of establishing a deep learning model by the deep model building module is as follows:

[0050] D1. Select the appropriate network architecture based on the characteristics of the data and task requirements; network architectures include recurrent neural networks, long short-term memory networks, and convolutional neural networks;

[0051] D2. Based on the selected network architecture, design a specific network structure; the network structure design process includes determining the number of network layers, the number of neurons in each layer, and the type of activation function;

[0052] D3. Preprocess the input data and use the principal component analysis method to extract the features of the prediction model;

[0053] D4. Initialize the weights by random initialization, and then use the SGD algorithm to update the network weights;

[0054] D5. Select the appropriate loss function according to the task type;

[0055] D6. Find the best model configuration by adjusting the hyperparameters of learning rate, batch size, and number of iterations, and use grid search or Bayesian optimization methods for hyperparameter tuning;

[0056] D7. Use an independent test set to evaluate the deep learning model and calculate the accuracy index. At the same time, use cross-validation to more comprehensively evaluate the performance of the deep learning model.

[0057] D8. Deploy the trained deep learning model to actual applications, continuously collect feedback data, and adjust and optimize the model based on actual performance.

[0058] Furthermore, the process of introducing the adaptive adjustment mechanism into the adaptive adjustment module is as follows:

[0059] E1. Select a suitable adaptive learning rate algorithm based on task requirements and quantity characteristics;

[0060] E2. Set initial parameters for the selected adaptive learning rate algorithm, including initial learning rate, decay factor and momentum coefficient;

[0061] E3. During the deep learning model training process, monitor the changes in training loss and validation set accuracy in real time;

[0062] E4. Adjust the learning rate according to the changes in the monitored indicators. For example, if it is found that the deep learning model performs well on the training set but poorly on the validation set, reduce some learning rates to prevent overfitting. If it is found that the model performs poorly on the training set but well on the validation set, increase some learning rates to speed up convergence.

[0063] E5. Add a regularization term to the loss function;

[0064] E6. When the accuracy of the validation set no longer improves, set an accuracy threshold. When the accuracy of the validation set does not improve significantly after several consecutive iterations, stop training.

[0065] Furthermore, when generating power generation prediction results, the prediction result output module first collects current and future meteorological data and power plant status information and inputs them into the trained deep learning model to generate real-time power generation prediction results. This power generation prediction result can help companies or power plant operators make better decisions. The generated prediction results are then evaluated and the accuracy index is calculated to measure the performance of the model. If it is found that the prediction results are not accurate or there are deviations, it is necessary to review the steps of data collection, feature extraction and model training, and make corresponding adjustments and optimizations. The prediction results are then presented to relevant personnel or departments in the form of charts or reports.

[0066] Furthermore, the error evaluation and iterative update module calculates the error value according to the prediction result of the prediction result output module. ,error The calculation formula is as follows:

[0067] ,in, is the true value, is the predicted value, n is the number of samples;

[0068] Then analyze the source of the error and confirm whether the error belongs to overfitting, underfitting and uneven data distribution. If it is overfitting, it means that the deep learning model is too complex and the deep learning model structure needs to be simplified or regularized; if it is underfitting, it means that the deep learning model is too simple and the complexity of the deep learning model needs to be increased or more features need to be introduced; when the data is unevenly distributed, it means that the deep learning model may have poor prediction performance for minority categories and it is necessary to balance the data distribution through resampling or weighting methods; finally, use visualization tools to show the difference between the model prediction results and the true value.

[0069] Further, based on the photovoltaic station power prediction system based on the radiation environment weighted algorithm described above, the present invention also provides a photovoltaic station power prediction method based on the radiation environment weighted algorithm, comprising the following steps:

[0070] S1. Collect multi-dimensional meteorological data such as light intensity, temperature, wind speed, humidity, cloud coverage and air quality index in real time through a multi-dimensional meteorological data acquisition module, and extract power generation data from the historical operation records of the photovoltaic power station;

[0071] S2. Extract key features from the original data through the feature extraction module according to the physical characteristics and operation rules of the photovoltaic power station, and process the extracted key features;

[0072] S3, using a dynamic weight allocation module to assign initial weights to each feature according to the degree of influence of different meteorological factors on photovoltaic power generation;

[0073] S4, establish a deep learning model that can capture long-term dependencies and local features in the data through the deep model building module;

[0074] S5. Introduce an adaptive adjustment mechanism in the deep learning model training process through the adaptive adjustment module, and continuously adjust the deep learning model parameters and structure according to the error evaluation results;

[0075] S6. Generate real-time power generation prediction results through the prediction result output module, and customize the deep learning model parameters and data processing methods for photovoltaic power stations of different geographical locations, scales and types;

[0076] S7, according to the evaluation results of the prediction result output module, the structural parameter setting of the deep learning model is continuously iterated and improved through the error evaluation and iterative update module;

[0077] S8. Customize and adjust the parameters of the deep learning model through the application scenario adaptive adjustment module, deploy the prediction system to the photovoltaic power station site, and continuously monitor the system operation status and the accuracy of the prediction results.

[0078] The beneficial effects of the present invention are:

[0079] 1. The present invention obtains more extensive meteorological information through a multi-dimensional meteorological data acquisition module, extracts power generation data from the historical operation records of the photovoltaic power station, and then assigns initial weights to each feature according to the degree of influence of different meteorological factors on photovoltaic power generation. A deep learning model is then established, and the deep learning model is regularly fine-tuned. The model structure and parameters are adjusted according to the latest operation data and feedback information. The prediction results are then output and the structural parameter settings of the deep learning model are continuously iterated and improved according to the evaluation results. Finally, the application scenario is adaptively adjusted. The dynamic weight allocation strategy is adopted to adjust the weight of the sample according to its importance or representativeness, thereby improving the overall performance of the model and improving the accuracy of the prediction.

[0080] 2. The present invention can allocate initial weights to each feature according to the degree of influence of different meteorological factors on photovoltaic power generation through a dynamic weight allocation module, so that the model can adapt to environmental changes and improve the accuracy of prediction.

[0081] 3. The present invention can introduce an adaptive adjustment mechanism in the deep learning model training process through an adaptive adjustment module to prevent overfitting and accelerate the convergence speed.

[0082] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0084] Figure 1 A block diagram of a photovoltaic station power system according to an embodiment of the present invention is shown;

[0085] Figure 2 A flowchart of multi-dimensional meteorological data collection according to an embodiment of the present invention is shown;

[0086] Figure 3 A flowchart of key feature extraction according to an embodiment of the present invention is shown;

[0087] Figure 4 A flowchart of establishing a deep learning model according to an embodiment of the present invention is shown;

[0088] Figure 5A flowchart of introducing an adaptive adjustment mechanism according to an embodiment of the present invention is shown;

[0089] Figure 6 A flowchart of photovoltaic power prediction according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0090] 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.

[0091] The embodiment of the present invention provides a photovoltaic power prediction system based on an illumination environment weighted algorithm, such as Figure 1-5 As shown, it includes: a multi-dimensional meteorological data acquisition module, which collects multi-dimensional meteorological data including light intensity, temperature, wind speed, humidity, cloud coverage and air quality index in real time, and extracts power generation data from the historical operation records of the photovoltaic power station, wherein the power generation data includes information on the power generation and equipment status at each time point. The multi-dimensional meteorological data will serve as an important input of the prediction model, directly affecting the accuracy and reliability of the prediction.

[0092] like Figure 2 As shown, the process of the multi-dimensional meteorological data acquisition module for multi-dimensional meteorological data acquisition is as follows:

[0093] A1. Equip the photovoltaic power station with meteorological sensors and remote sensing instruments to monitor the light intensity, temperature, wind speed and humidity data in real time through meteorological sensors, and obtain cloud coverage and air quality index data over a large area through remote sensing instruments;

[0094] A2. Transmit meteorological data, cloud coverage data and air quality index data to the cloud server for storage and management through the wireless network module;

[0095] A3. Integrate with the existing monitoring system of the photovoltaic power station and directly extract historical operation records from the monitoring system;

[0096] A4. Process the collected raw meteorological data. Data processing includes data cleaning, missing value filling and interpolation. Data cleaning is used to remove obvious erroneous values ​​and outliers. Missing data caused by equipment failure are filled by interpolation. For continuously changing variables such as temperature and humidity, time series analysis is used for interpolation. For discrete variables such as cloud cover, classification interpolation is required.

[0097] The feature extraction module extracts key features from the original data according to the physical characteristics and operating rules of the photovoltaic power station. The key features include sunshine duration, solar altitude angle and photovoltaic panel tilt angle. The key features are used to reflect the power generation potential and influencing factors of the photovoltaic power station.

[0098] like Figure 3 As shown, the process of extracting key features by the feature extraction module is as follows:

[0099] B1. Check the sunrise and sunset timetable provided by the local weather station and calculate the sunrise and sunset times of each day;

[0100] B2. According to weather conditions and the installation angle of photovoltaic panels, select the effective time period that can receive sunlight, combine ground measured data or satellite remote sensing image analysis, and dynamically adjust the start and end time of the effective light period to improve the accuracy of calculation;

[0101] B3. Input the latitude and longitude coordinates of the location of the photovoltaic power station and the current time to calculate the solar altitude angle;

[0102] When calculating the solar altitude angle, first calculate the solar inclination angle, then calculate the solar hour angle, and then calculate the solar altitude angle based on the solar inclination angle and the solar hour angle. The specific operations are as follows:

[0103] B3.1. Calculate the solar inclination using the solar inclination formula , the formula for the solar inclination is:

[0104] ,in, To convert a Gregorian calendar date to a day of the year;

[0105] B3.2. Calculate the solar hour angle using the solar hour angle calculation formula , the solar hour angle calculation formula is as follows:

[0106] ,T=UTC time+ Where T is the local solar time;

[0107] B3.3. Use the solar altitude angle calculation formula to calculate the solar altitude angle h. The solar altitude angle calculation formula is:

[0108] =arc ,in is the latitude of the observation site;

[0109] B3.4. Convert the solar altitude angle h from radians to degrees using the following conversion formula:

[0110] h.

[0111] The above calculation method can accurately calculate the solar altitude angle of the location of the photovoltaic power station at any given time, which not only improves the efficiency and accuracy of data processing, but also provides a solid foundation for the subsequent photovoltaic power generation prediction model.

[0112] B4. Analyze the changes in the incident angle of light at different tilt angles, evaluate its impact on the intensity of light received by the photovoltaic panel, and then combine historical data and simulation results to determine the optimal tilt angle range to facilitate adjustment in actual operation;

[0113] B5. Integrate the calculated features of sunshine duration, solar altitude angle and photovoltaic panel tilt angle to form a set of comprehensive feature vectors;

[0114] B6. Calculate the correlation between each feature and photovoltaic power generation, evaluate the effect of feature selection by cross-validation, and ensure that the selected features can effectively improve the prediction ability of the prediction model;

[0115] The correlation calculation formula between each feature and photovoltaic power generation is as follows:

[0116] ,in and represent the observed values ​​of two variables, and Respectively represent the average values ​​of two variables;

[0117] The above formula is used to calculate the Pearson correlation coefficient between the characteristics (such as sunshine duration, solar altitude angle, photovoltaic panel tilt angle, etc.) as the independent variable X and the photovoltaic power generation as the dependent variable Y. By comparing the correlation coefficients between different characteristics and photovoltaic power generation, it is possible to preliminarily determine which characteristics are more closely related to photovoltaic power generation.

[0118] When evaluating the effect of feature selection through cross-validation, we start with an empty model and gradually add features. Each time we add features, we select the current optimal feature until the stopping condition is met (such as reaching the preset maximum number of features or performance no longer improving). This method can effectively avoid the overfitting problem.

[0119] Then use a specific decision tree machine learning algorithm to evaluate the importance of each feature and perform feature selection based on it. These algorithms usually have a built-in feature importance evaluation mechanism that can help quickly identify the most useful features for prediction tasks. The original data set is divided into a training set and a test set. The k-fold cross-validation method is used, that is, the data set is divided into k subsets, and one subset is set aside as the test set each time, and the remaining k-1 subsets are used as training sets. Repeat this process k times.

[0120] In each cross-validation process, the training set data is used to train the prediction model, and the model performance is evaluated on the test set. The performance indicators of each cross-validation (such as accuracy, mean square error, etc.) are recorded for subsequent analysis. Based on the results of cross-validation, the model performance under different feature combinations is analyzed. If a feature can significantly improve the model performance in multiple cross-validations, this feature is considered to be effective.

[0121] A dynamic weight allocation module allocates initial weights to each feature according to the degree of influence of different meteorological factors on photovoltaic power generation, so that the model can adapt to environmental changes and improve the accuracy of prediction.

[0122] The process of the dynamic weight assignment module assigning initial weights to each feature is as follows:

[0123] C1. Standardize the original data matrix to eliminate the influence of different indicators. The standardization formula is as follows:

[0124] , where x is the original data and u is the mean of the data. is the standard deviation of the data, is the standardized data matrix;

[0125] C2. Calculate the jth indicator The proportion of samples , the calculation formula is:

[0126] , where m is the number of samples;

[0127] C3. Calculate the entropy value of the jth indicator , the calculation formula is as follows:

[0128] , where k is a constant, i.e. k= ;

[0129] C4. Calculate the coefficient of difference of the jth indicator and the weight of the jth item , the calculation formula is as follows:

[0130] , ,in is the number of indicators;

[0131] C5. Calculate the correlation between indicators , the calculation formula is as follows:

[0132] ,in, is the minimum value of the standardized data matrix, is the maximum value of the data matrix after standardization, and m is the number of samples;

[0133] C6. Calculate indicator weights , and then the indicator weight and the weight of the jth item Combine and calculate the comprehensive weight , the calculation formula is as follows:

[0134] , ;

[0135] C7. Based on the calculated comprehensive weight , select the element features that are important to the prediction model, use the element features to build the prediction model, and train the prediction model;

[0136] C8. Evaluate model performance through test data sets to ensure that the selected features and weights can effectively improve prediction accuracy;

[0137] A deep model building module is used to build a deep learning model that can capture long-term dependencies and local features in data.

[0138] like Figure 4 As shown, the process of establishing a deep learning model by the deep model building module is as follows:

[0139] D1. Select the appropriate network architecture based on the characteristics of the data and task requirements; network architectures include recurrent neural networks (RNNs), long short-term memory networks (LSTMs), and convolutional neural networks (CNNs). For sequence data, use recurrent neural networks or long short-term memory networks; for image data, use convolutional neural networks;

[0140] D2. Based on the selected network architecture, design a specific network structure; the network structure design process includes determining the number of network layers, the number of neurons in each layer, and the type of activation function;

[0141] D3. Preprocess the input data and use the principal component analysis (PCA) method to extract the features of the prediction model. Principal component analysis (PCA) is an existing technology and will not be described in detail here.

[0142] D4. Initialize the weights by random initialization, and then use the SGD algorithm to update the network weights;

[0143] D5. Select an appropriate loss function based on the task type, such as using cross entropy loss for classification tasks and mean square error loss for regression tasks;

[0144] D6. Find the best model configuration by adjusting the hyperparameters of learning rate, batch size, and number of iterations, and use grid search or Bayesian optimization methods for hyperparameter tuning;

[0145] D7. Use an independent test set to evaluate the deep learning model and calculate the accuracy index. At the same time, use cross-validation to more comprehensively evaluate the performance of the deep learning model.

[0146] D8. Deploy the trained deep learning model to actual applications, and continuously collect feedback data, adjust and optimize the model based on actual performance to ensure that it can maintain good performance in real environments.

[0147] An adaptive adjustment module, which is used to introduce an adaptive adjustment mechanism during the deep learning model training process to prevent overfitting and speed up convergence;

[0148] like Figure 5 As shown, the process of introducing the adaptive adjustment mechanism by the adaptive adjustment module is as follows:

[0149] E1. Select a suitable adaptive learning rate algorithm based on task requirements and quantity characteristics. The adaptive learning rate algorithms include AdaGrad, RMSprop and Adam. The adaptive learning rate algorithm can dynamically adjust the learning rate according to the performance of the deep learning model during the training process, thereby improving the performance and convergence speed of the deep learning model.

[0150] E2. Set initial parameters for the selected adaptive learning rate algorithm. The initial parameters include initial learning rate, decay factor and momentum coefficient. Setting the initial parameters helps improve the training effect of the deep learning model.

[0151] E3. During the deep learning model training process, monitor the changes in training loss and validation set accuracy in real time to detect overfitting and take corresponding measures;

[0152] E4. Adjust the learning rate according to the changes in the monitored indicators. For example, if it is found that the deep learning model performs well on the training set but poorly on the validation set, reduce some learning rates to prevent overfitting. If it is found that the model performs poorly on the training set but well on the validation set, increase some learning rates to speed up convergence.

[0153] E5. Add regularization terms to the loss function to reduce the deep learning model’s dependence on training data and improve its generalization ability;

[0154] E6. When the accuracy of the validation set no longer improves, set an accuracy threshold. When the accuracy of the validation set does not improve significantly after several consecutive iterations, stop training.

[0155] A prediction result output module, which generates real-time power generation prediction results based on current and future meteorological data and power station status information;

[0156] When generating the power generation prediction result, the prediction result output module first collects current and future meteorological data and power station status information and inputs them into the trained deep learning model to generate real-time power generation prediction results. This power generation prediction result can help companies or power station operators make better decisions, such as adjusting power generation plans, optimizing resource allocation, etc., and then evaluate the generated prediction results, calculate the accuracy index to measure the performance of the model. If it is found that the prediction results are not accurate or biased, it is necessary to review the steps of data collection, feature extraction and model training, and make corresponding adjustments and optimizations. The prediction results are then presented to relevant personnel or departments in the form of charts or reports, so that they can more intuitively understand the changing trends and potential risks of power generation, and make more informed decisions:

[0157] An error evaluation and iterative update module, which continuously and iteratively improves the structural parameter settings of the deep learning model according to the evaluation results;

[0158] The error evaluation and iterative update module calculates the error value according to the prediction result of the prediction result output module ,error The calculation formula is as follows:

[0159] ,in, is the true value, is the predicted value, n is the number of samples;

[0160] Then analyze the source of the error and confirm whether the error belongs to overfitting, underfitting and uneven data distribution. If it is overfitting, it means that the deep learning model is too complex and the deep learning model structure needs to be simplified or regularized; if it is underfitting, it means that the deep learning model is too simple and the complexity of the deep learning model needs to be increased or more features need to be introduced; when the data is unevenly distributed, it may cause the deep learning model to have poor prediction performance for minority categories and it is necessary to balance the data distribution through resampling or weighting methods; finally, use visualization tools to show the difference between the model prediction results and the true value to help understand the error distribution more intuitively.

[0161] The application scenario adaptive adjustment module first customizes the model parameters according to the unique environmental and climatic conditions faced by photovoltaic power stations in different geographical locations, and then adopts different data processing methods and model structures for photovoltaic power stations of different sizes and types, and then customizes the development of corresponding prediction functions and services according to the specific needs and characteristics of the power stations.

[0162] Through the multi-dimensional meteorological data acquisition module, a wider range of meteorological information is obtained, and then the power generation data is extracted from the historical operation records of the photovoltaic power station. Then, according to the degree of influence of different meteorological factors on photovoltaic power generation, an initial weight is assigned to each feature, and then a deep learning model is established. The deep learning model is regularly fine-tuned, and the model structure and parameters are adjusted according to the latest operation data and feedback information. Then, the prediction results are output and the structural parameter settings of the deep learning model are continuously iterated and improved according to the evaluation results. Finally, adaptive adjustments are made to the application scenarios. By adopting a dynamic weight allocation strategy to adjust the weights of samples according to their importance or representativeness, the overall performance of the model is improved to improve the accuracy of the prediction.

[0163] Based on the photovoltaic power station power prediction system based on the radiation environment weighted algorithm described above, the present invention also provides a photovoltaic power station power prediction method based on the radiation environment weighted algorithm, such as Figure 6 As shown, the following steps are included:

[0164] S1. Collect multi-dimensional meteorological data such as light intensity, temperature, wind speed, humidity, cloud coverage and air quality index in real time through a multi-dimensional meteorological data acquisition module, and extract power generation data from the historical operation records of the photovoltaic power station;

[0165] S2. Extract key features from the original data through the feature extraction module according to the physical characteristics and operation rules of the photovoltaic power station, and process the extracted key features;

[0166] S3, using a dynamic weight allocation module to assign initial weights to each feature according to the degree of influence of different meteorological factors on photovoltaic power generation;

[0167] S4, establish a deep learning model that can capture long-term dependencies and local features in the data through the deep model building module;

[0168] S5. Introduce an adaptive adjustment mechanism in the deep learning model training process through the adaptive adjustment module, and continuously adjust the deep learning model parameters and structure according to the error evaluation results;

[0169] S6. Generate real-time power generation prediction results through the prediction result output module, and customize the deep learning model parameters and data processing methods for photovoltaic power stations of different geographical locations, scales and types;

[0170] S7, according to the evaluation results of the prediction result output module, the structural parameter setting of the deep learning model is continuously iterated and improved through the error evaluation and iterative update module;

[0171] S8. Customize and adjust the parameters of the deep learning model through the application scenario adaptive adjustment module, deploy the prediction system to the photovoltaic power station site, and continuously monitor the system operation status and the accuracy of the prediction results.

[0172] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A photovoltaic power prediction system based on an irradiation environment weighted algorithm, characterized in that: include: A multi-dimensional meteorological data acquisition module, which collects multi-dimensional meteorological data including light intensity, temperature, wind speed, humidity, cloud cover and air quality index in real time, and extracts power generation data from the historical operation records of the photovoltaic power station; A feature extraction module, which extracts key features from the raw data according to the physical characteristics and operating rules of the photovoltaic power station. The key features include sunshine duration, solar altitude angle and photovoltaic panel tilt angle; A dynamic weight allocation module, wherein the dynamic weight allocation module allocates an initial weight to each feature according to the degree of influence of different meteorological factors on photovoltaic power generation; A deep model building module, wherein the deep model building module is used to build a deep learning model capable of capturing long-term dependencies and local features in the data; An adaptive adjustment module, which is used to introduce an adaptive adjustment mechanism during the deep learning model training process; A prediction result output module, which generates real-time power generation prediction results based on current and future meteorological data and power station status information; An error evaluation and iterative update module, which continuously and iteratively improves the structural parameter settings of the deep learning model according to the evaluation results; The application scenario adaptive adjustment module first customizes the model parameters according to the unique environmental and climatic conditions faced by photovoltaic power stations in different geographical locations, and then adopts different data processing methods and model structures for photovoltaic power stations of different sizes and types, and then customizes the development of corresponding prediction functions and services according to the specific needs and characteristics of the power stations.

2. The photovoltaic power prediction system based on the radiation environment weighted algorithm according to claim 1 is characterized in that: The process of the multi-dimensional meteorological data acquisition module for multi-dimensional meteorological data acquisition is as follows: A1. Equip the photovoltaic power station with meteorological sensors and remote sensing instruments to monitor the light intensity, temperature, wind speed and humidity data in real time through meteorological sensors, and obtain cloud coverage and air quality index data over a large area through remote sensing instruments; A2. Transmit meteorological data, cloud coverage data and air quality index data to the cloud server for storage and management through the wireless network module; A3. Integrate with the existing monitoring system of the photovoltaic power station and directly extract historical operation records from the monitoring system; A4. Process the collected raw meteorological data, including data cleaning and missing value filling and interpolation.

3. The photovoltaic power prediction system based on the radiation environment weighted algorithm according to claim 2 is characterized in that: The process of extracting key features by the feature extraction module is as follows: B1. Check the sunrise and sunset timetable provided by the local weather station and calculate the sunrise and sunset times of each day; B2. According to weather conditions and the installation angle of photovoltaic panels, select the effective time period that can receive sunlight, and dynamically adjust the start and end time of the effective light period by combining ground measured data or satellite remote sensing image analysis; B3. Input the latitude and longitude coordinates of the location of the photovoltaic power station and the current time to calculate the solar altitude angle; B4. Analyze the changes in the incident angle of light at different tilt angles, evaluate its impact on the intensity of light received by the photovoltaic panel, and then combine historical data and simulation results to determine the optimal tilt angle range; B5. Integrate the calculated features of sunshine duration, solar altitude angle and photovoltaic panel tilt angle to form a set of comprehensive feature vectors; B6. Calculate the correlation between each feature and photovoltaic power generation, and evaluate the effect of feature selection through cross-validation.

4. The photovoltaic power prediction system based on the radiation environment weighted algorithm according to claim 3 is characterized by: When calculating the solar altitude angle, first calculate the solar inclination angle, then calculate the solar hour angle, and then calculate the solar altitude angle based on the solar inclination angle and the solar hour angle. The specific operations are as follows: B3.

1. Calculate the solar inclination using the solar inclination formula , the formula for the solar inclination is: ,in, To convert a Gregorian calendar date to the ordinal number of the day of the year; B3.

2. Calculate the solar hour angle using the solar hour angle calculation formula , the solar hour angle calculation formula is as follows: ,T=UTC time+ Where T is the local solar time; B3.

3. Use the solar altitude angle calculation formula to calculate the solar altitude angle h. The solar altitude angle calculation formula is: =arc ,in is the latitude of the observation site; B3.

4. Convert the solar altitude angle h from radians to degrees using the following conversion formula: h。 5. The photovoltaic power prediction system based on the radiation environment weighted algorithm according to claim 4 is characterized in that: The process of the dynamic weight assignment module assigning initial weights to each feature is as follows: C1. Standardize the original data matrix. The standardization formula is as follows: , where x is the original data and u is the mean of the data. is the standard deviation of the data, is the standardized data matrix; C2. Calculate the jth index. The proportion of samples , the calculation formula is: , where m is the number of samples; C3. Calculate the entropy value of the jth indicator , the calculation formula is as follows: , where k is a constant, i.e. k= ; C4. Calculate the coefficient of difference of the jth indicator and the weight of the jth item , the calculation formula is as follows: , ,in is the number of indicators; C5. Calculate the correlation between indicators , the calculation formula is as follows: ,in, is the minimum value of the standardized data matrix, is the maximum value of the data matrix after standardization, and m is the number of samples; C6. Calculate indicator weights , and then the indicator weight and the weight of the jth item Combine and calculate the comprehensive weight , the calculation formula is as follows: ， ； C7. Based on the calculated comprehensive weight , select the element features that are important to the prediction model, use the element features to build the prediction model, and train the prediction model; C8. Evaluate model performance using the test dataset.

6. The photovoltaic power prediction system based on the radiation environment weighted algorithm according to claim 5 is characterized in that: The process of establishing a deep learning model by the deep model building module is as follows: D1. Select the appropriate network architecture based on the characteristics of the data and task requirements; network architectures include recurrent neural networks, long short-term memory networks, and convolutional neural networks; D2. Based on the selected network architecture, design a specific network structure; the network structure design process includes determining the number of network layers, the number of neurons in each layer, and the type of activation function; D3. Preprocess the input data and use the principal component analysis method to extract the features of the prediction model; D4. Initialize the weights by random initialization, and then use the SGD algorithm to update the network weights; D5. Select the appropriate loss function according to the task type; D6. Find the best model configuration by adjusting the hyperparameters of learning rate, batch size, and number of iterations, and use grid search or Bayesian optimization methods for hyperparameter tuning; D7. Use an independent test set to evaluate the deep learning model and calculate the accuracy index. At the same time, use cross-validation to more comprehensively evaluate the performance of the deep learning model. D8. Deploy the trained deep learning model to actual applications, continuously collect feedback data, and adjust and optimize the model based on actual performance.

7. The photovoltaic power prediction system based on the radiation environment weighted algorithm according to claim 6 is characterized by: The process of introducing the adaptive adjustment mechanism into the adaptive adjustment module is as follows: E1. Select a suitable adaptive learning rate algorithm based on task requirements and quantity characteristics; E2. Set initial parameters for the selected adaptive learning rate algorithm, including initial learning rate, decay factor and momentum coefficient; E3. During the deep learning model training process, monitor the changes in training loss and validation set accuracy in real time; E4. Adjust the learning rate according to the changes in the monitored indicators. For example, if it is found that the deep learning model performs well on the training set but poorly on the validation set, reduce some learning rates to prevent overfitting. If it is found that the model performs poorly on the training set but well on the validation set, increase some learning rates to speed up convergence. E5. Add a regularization term to the loss function; E6. When the accuracy of the validation set no longer improves, set an accuracy threshold. When the accuracy of the validation set does not improve significantly after several consecutive iterations, stop training.

8. The photovoltaic power prediction system based on the radiation environment weighted algorithm according to claim 7 is characterized in that: When generating power generation prediction results, the prediction result output module first collects current and future meteorological data and power station status information and inputs them into the trained deep learning model to generate real-time power generation prediction results. This power generation prediction result can help companies or power station operators make better decisions. The generated prediction results are then evaluated and the accuracy index is calculated to measure the performance of the model. If it is found that the prediction results are not accurate or there are deviations, it is necessary to review the steps of data collection, feature extraction and model training, make corresponding adjustments and optimizations, and then present the prediction results to relevant personnel or departments in the form of charts or reports.

9. The photovoltaic power prediction system based on the radiation environment weighted algorithm according to claim 8 is characterized in that: The error evaluation and iterative update module calculates the error value according to the prediction result of the prediction result output module ,error The calculation formula is as follows: ,in, is the true value, is the predicted value, n is the number of samples; Then analyze the source of the error and confirm whether the error belongs to overfitting, underfitting and uneven data distribution. If it is overfitting, it means that the deep learning model is too complex and the deep learning model structure needs to be simplified or regularized; if it is underfitting, it means that the deep learning model is too simple and the complexity of the deep learning model needs to be increased or more features need to be introduced; when the data is unevenly distributed, it means that the deep learning model may have poor prediction performance for minority categories and it is necessary to balance the data distribution through resampling or weighting methods; finally, use visualization tools to show the difference between the model prediction results and the true value.

10. A photovoltaic station power prediction method based on an illumination environment weighted algorithm, based on a photovoltaic station power prediction system based on an illumination environment weighted algorithm according to any one of claims 1 to 9, characterized in that: The steps include: S1. Collect multi-dimensional meteorological data such as light intensity, temperature, wind speed, humidity, cloud coverage and air quality index in real time through a multi-dimensional meteorological data acquisition module, and extract power generation data from the historical operation records of the photovoltaic power station; S2. Extract key features from the original data through the feature extraction module according to the physical characteristics and operation rules of the photovoltaic power station, and process the extracted key features; S3, using a dynamic weight allocation module to assign initial weights to each feature according to the degree of influence of different meteorological factors on photovoltaic power generation; S4, establish a deep learning model that can capture long-term dependencies and local features in the data through the deep model building module; S5. Introduce an adaptive adjustment mechanism in the deep learning model training process through the adaptive adjustment module, and continuously adjust the deep learning model parameters and structure according to the error evaluation results; S6. Generate real-time power generation prediction results through the prediction result output module, and customize the deep learning model parameters and data processing methods for photovoltaic power stations of different geographical locations, scales and types; S7, according to the evaluation results of the prediction result output module, the structural parameter setting of the deep learning model is continuously iterated and improved through the error evaluation and iterative update module; S8. Customize and adjust the parameters of the deep learning model through the application scenario adaptive adjustment module, deploy the prediction system to the photovoltaic power station site, and continuously monitor the system operation status and the accuracy of the prediction results.

Images