Photovoltaic power CNN (Convolutional Neural Network) prediction method and system considering multivariable nonlinear spatial relationship
By using a convolutional neural network model to process photovoltaic power data, the problems of insufficient complexity and accuracy in photovoltaic power prediction in existing technologies are solved. This enables high-precision prediction of photovoltaic power, supports accurate load management of low-cycle load shedding systems, and improves the stability of the power system.
Patent Information
- Application Number
- CN202511801112.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-02
- Publication Date
- 2026-03-03
AI Technical Summary
Existing photovoltaic power prediction methods suffer from problems such as complex modeling, high parameter accuracy requirements, and poor generalization ability when dealing with multivariate and high-dimensional data. This makes it difficult to accurately predict distributed photovoltaic output, leading to false tripping or failure to trip of the under-cycle load shedding system, which affects the safety of the power system.
By employing a convolutional neural network (CNN) model, through data preprocessing, feature extraction, and multi-layer convolution operations, the nonlinear relationship between photovoltaic power and various influencing factors is captured. Combined with the sliding window method and normalization processing, accurate prediction of photovoltaic power is achieved.
This improves the accuracy and reliability of photovoltaic power prediction, helps low-cycle load shedding systems accurately grasp downstream loads of feeders, reduces malfunctions, and ensures power system stability.
Smart Images
Figure CN121602908A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of smart grids, and specifically to a CNN prediction method and system for photovoltaic power considering multivariable nonlinear spatial relationships. Background Technology
[0002] With the increasing global demand for clean energy, photovoltaic (PV) power generation, as an important component of the renewable energy system, is experiencing rapid and continuous growth in installed capacity and application scenarios. However, PV power output is influenced by a variety of internal and external factors. External factors mainly include meteorological conditions such as solar radiation intensity, ambient temperature, humidity, wind speed, and cloud cover, while internal factors involve inherent attributes such as the geographical location of the PV power plant, the tilt and azimuth angle of the PV panels, and the aging of the equipment. The complex interactions and uncertain changes among these factors result in significant fluctuations and intermittent characteristics in PV power output, posing a severe challenge to the power balance, dispatching, and stability control of the power system.
[0003] Especially in low-frequency load shedding applications, with the large-scale integration of distributed photovoltaic (PV) power into the distribution network and its increasing penetration rate, traditional protection and control methods are facing significant challenges. Centralized low-frequency load shedding systems require accurate assessment of the actual load levels downstream of each feeder to avoid overestimation of the load due to localized power supply from distributed sources, which could lead to malfunctions or failures of the load shedding devices, threatening system safety. Therefore, accurate prediction of the distributed PV output downstream of each feeder has become a crucial prerequisite for the correct operation of low-frequency load shedding systems.
[0004] Currently, photovoltaic power prediction methods can be mainly divided into three categories: physical model methods, statistical model methods, and artificial intelligence model methods. Physical model methods are based on the physical equations of photovoltaic cells and meteorological parameters. Although the mechanism is clear, the calculations are complex, sensitive to parameter accuracy, and the prediction effect is limited by the simplification of the model. Statistical model methods (such as time series analysis and regression models) construct statistical relationships based on historical data, making it difficult to effectively characterize the complex nonlinear dynamic characteristics between variables, and their generalization ability is limited. Among artificial intelligence methods, traditional artificial neural networks, while possessing certain nonlinear fitting capabilities, are prone to structural limitations when handling multivariate, high-dimensional data, resulting in problems such as overfitting, slow convergence speed, and insufficient prediction accuracy. Therefore, developing a photovoltaic power prediction method that can integrate multi-source information, effectively capture complex nonlinear relationships, achieve high prediction accuracy, and is easy to implement in engineering is of significant practical importance. Summary of the Invention
[0005] The technical problem to be solved by this invention is to provide a photovoltaic power convolutional neural network (CNN) prediction method that considers multivariate nonlinear spatial relationships, thereby providing the feeder photovoltaic output power of centralized underclock load shedding systems and facilitating the centralized underclock load shedding systems to grasp the actual load output of the downstream power grid.
[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention includes the following steps: S1 collects meteorological data for the time to be predicted and the preceding 12 consecutive hours, photovoltaic power station characteristic data, and photovoltaic power data. The collected data is preprocessed by imputing missing values, correcting outliers, and normalizing based on the same extreme value parameters. S2, the sliding window method is used to extract features from the collected data, with a window length of 12 hours, and the collected data is composed of dimensions. The two-dimensional feature matrix, where N is the number of selected input variables; S3, the two-dimensional feature matrix is input into a pre-trained convolutional neural network model. The model sequentially performs convolution operations, non-linear activation, pooling dimensionality reduction, and fully connected mapping, and outputs the predicted value in the normalized domain. ; S4. Perform inverse normalization on the predicted values to obtain the photovoltaic active power prediction results. The formula is:
[0007] in , These are the historical maximum and minimum values for the photovoltaic power, respectively.
[0008] Furthermore, the meteorological data mentioned in step S1 includes solar radiation intensity, temperature, humidity, wind speed, and cloud cover, while the photovoltaic power station's own characteristic data includes geographical location and the degree of aging of the photovoltaic panels.
[0009] Furthermore, the normalization formula in step S1 is:
[0010] in, and These are the minimum and maximum values of variable x in the dataset, respectively. It is the normalized value.
[0011] Furthermore, the preprocessing in step S1 also includes analyzing the correlation between the meteorological data and photovoltaic power using the Pearson correlation coefficient method, calculating the Pearson correlation coefficient between each meteorological data and photovoltaic power, and selecting the data with a strong correlation. Furthermore, the convolutional neural network model structure described in step S3 includes multiple convolutional layers, pooling layers, and fully connected layers, wherein: Convolutional layers are used to extract local features from the input data. Convolution operations are performed by sliding convolution kernels of different sizes across the input data. Pooling layers are used to reduce the dimensionality of data and reduce computation while retaining important feature information. Max pooling is used to take the maximum value within a fixed-size pooling window as the output. The fully connected layer integrates the features extracted by the preceding convolutional and pooling layers and outputs the final prediction result.
[0012] Furthermore, the convolutional neural network model training method described in step S3 includes: Data partitioning: The dataset after feature selection and extraction is divided into training set, validation set and test set according to the ratio of 70%, 15% and 15% respectively; Training preparation: Define the loss function as mean squared error (MSE), use the stochastic gradient descent (SGD) algorithm for parameter updates, and set the initial learning rate to 0.01.
[0013] Iterative training and monitoring: During the training iteration, the loss function value on the training set is calculated, and the weights and biases of the model are updated according to the gradient of the loss function and the learning rate. Adjustment and continuation: If the loss function value on the validation set does not decrease significantly in the last 100 iterations, adjust the learning rate and continue training.
[0014] Furthermore, the evaluation and optimization method for the convolutional neural network model described in step S3 includes: Model evaluation: The trained model is evaluated using a test set, and the prediction accuracy of the model is evaluated using mean absolute error (MAE) and root mean square error (RMSE). Model optimization: When the model's evaluation metrics are not ideal, increasing the number of convolutional layers, changing the size of the convolutional kernels, and adjusting the number of neurons in the fully connected layers can help the model converge faster.
[0015] The present invention also provides a photovoltaic power CNN prediction system considering multivariable nonlinear spatial relationships, comprising a processor and a computer-readable storage medium interconnected with each other, wherein the computer-readable storage medium stores a computer program, which is executed by the processor to implement the steps of the photovoltaic power CNN prediction method considering multivariable nonlinear spatial relationships described above.
[0016] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described photovoltaic power CNN prediction method considering multivariable nonlinear spatial relationships.
[0017] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described photovoltaic power CNN prediction method considering multivariable nonlinear spatial relationships.
[0018] Compared with the prior art, the advantages of the present invention are as follows: This invention effectively addresses the problems of existing photovoltaic power prediction methods, such as complex modeling and calculation processes, high parameter accuracy requirements, poor generalization ability, and difficulty in handling multivariate high-dimensional data. It provides support for accurately determining downstream loads in under-frequency load shedding systems. Specifically, this invention fully utilizes the advantages of convolutional neural networks in processing multivariate data, effectively capturing the complex nonlinear relationship between photovoltaic power and various influencing factors. This improves the accuracy and reliability of photovoltaic power prediction and solves the problem of significant discrepancies between the actual downstream load and the power detected by local devices on the feeders due to the large-scale distributed photovoltaic grid integration, leading to the shedding of a large amount of load on the feeders by under-frequency load shedding devices. This method is easy to implement in the field and has good application prospects. Attached Figure Description
[0019] Figure 1 This is a schematic diagram of the photovoltaic power CNN prediction model structure that considers multivariable nonlinear spatial relationships in this invention. Figure 2 This is a schematic diagram comparing the training data results of the model in an embodiment of the present invention; Figure 3 This is a schematic diagram showing the comparison between predicted and actual values in an embodiment of the present invention. Detailed Implementation
[0020] The present invention will be further described below with reference to the accompanying drawings and specific preferred embodiments, but this does not limit the scope of protection of the present invention.
[0021] The technical solution adopted in this embodiment includes the following steps: S1 collects meteorological data for the time to be predicted and the preceding 12 consecutive hours, photovoltaic power station characteristic data, and photovoltaic power data. The collected data is preprocessed by imputing missing values, correcting outliers, and normalizing based on the same extreme value parameters. S2, the sliding window method is used to extract features from the collected data, with a window length of 12 hours, and the collected data is composed of dimensions. The two-dimensional feature matrix, where N is the number of selected input variables; S3, the two-dimensional feature matrix is input into a pre-trained convolutional neural network model. The model sequentially performs convolution operations, non-linear activation, pooling dimensionality reduction, and fully connected mapping, and outputs the predicted value in the normalized domain. ; S4. Perform inverse normalization on the predicted values to obtain the photovoltaic active power prediction results. The formula is:
[0022] in , These are the historical maximum and minimum values for the photovoltaic power, respectively.
[0023] Preferably, the meteorological data in step S1 includes solar radiation intensity, temperature, humidity, wind speed, and cloud cover, and the photovoltaic power station's own characteristic data includes geographical location and the degree of aging of photovoltaic panels.
[0024] Preferably, the normalization formula in step S1 is:
[0025] in, and These are the minimum and maximum values of variable x in the dataset, respectively. It is the normalized value.
[0026] Preferably, the preprocessing in step S1 further includes analyzing the correlation between the meteorological data and photovoltaic power using the Pearson correlation coefficient method, calculating the Pearson correlation coefficient between each meteorological data and photovoltaic power, and selecting data with a strong correlation. Choosing highly correlated variables as input features and reducing data dimensionality can improve model training efficiency and computation speed.
[0027] Preferably, the convolutional neural network model structure described in step S3 is as follows: Figure 1 As shown, it includes multiple convolutional layers, pooling layers, and fully connected layers, wherein: Convolutional layers are used to extract local features from the input data. Convolution operations are performed by sliding convolution kernels of different sizes across the input data. Pooling layers are used to reduce the dimensionality of data and reduce computation while retaining important feature information. Max pooling is used to take the maximum value within a fixed-size pooling window as the output. The fully connected layer integrates the features extracted by the preceding convolutional and pooling layers and outputs the final prediction result.
[0028] The convolutional neural network model constructed in this embodiment includes 3 convolutional layers, 2 pooling layers, and 2 fully connected layers. The first convolutional layer uses 16 3×3 convolutional kernels; the second convolutional layer uses 32 3×3 convolutional kernels; and the third convolutional layer uses 64 3×3 convolutional kernels. Each convolutional layer is followed by a ReLU activation function to introduce nonlinearity. The first pooling layer uses a 2×2 max pooling window, and the second pooling layer also uses a 2×2 max pooling window.
[0029] In the fully connected layers, the first fully connected layer has 128 neurons, and the second fully connected layer has 1 neuron, which is used to output the final photovoltaic power prediction value.
[0030] Preferably, the convolutional neural network model training method described in step S3 includes: Data partitioning: The dataset after feature selection and extraction is divided into training set, validation set and test set according to the ratio of 70%, 15% and 15% respectively; Training preparation: Define the loss function as mean squared error (MSE), use the stochastic gradient descent (SGD) algorithm for parameter updates, and set the initial learning rate to 0.01.
[0031] Iterative training and monitoring: During the training iteration, the loss function value on the training set is calculated, and the weights and biases of the model are updated according to the gradient of the loss function and the learning rate. Adjustment and continuation: If the loss function value on the validation set does not decrease significantly in the last 100 iterations, adjust the learning rate and continue training.
[0032] Preferably, the evaluation and optimization method for the convolutional neural network model described in step S3 includes: Model evaluation: The trained model is evaluated using a test set, and the prediction accuracy of the model is evaluated using mean absolute error (MAE) and root mean square error (RMSE). Model optimization: When the model's evaluation metrics are not ideal, increasing the number of convolutional layers, changing the size of the convolutional kernels, and adjusting the number of neurons in the fully connected layers can help the model converge faster.
[0033] like Figure 2 , Figure 3 The diagram shows a comparison of the model training results and the effect of the predicted value versus the true value in this embodiment. It can be seen that the prediction results provided by the photovoltaic power CNN prediction method that considers the nonlinear spatial relationship of multiple variables are basically close to the true value, and the error between most of the predicted values and the true values is within 5%.
[0034] This embodiment also provides a photovoltaic power CNN prediction system that considers multivariable nonlinear spatial relationships, including a processor and a computer-readable storage medium interconnected with each other. The computer-readable storage medium stores a computer program, which is executed by the processor to implement the steps of the photovoltaic power CNN prediction method that considers multivariable nonlinear spatial relationships described above.
[0035] This embodiment also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the photovoltaic power CNN prediction method considering multivariable nonlinear spatial relationships described above.
[0036] This embodiment also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the photovoltaic power CNN prediction method that considers multivariable nonlinear spatial relationships described above.
[0037] Compared with the prior art, the advantages of this embodiment are: This embodiment effectively addresses the problems of existing photovoltaic power prediction methods, such as complex modeling and calculation processes, high parameter accuracy requirements, poor generalization ability, and difficulty in handling multivariate high-dimensional data. It provides support for accurately assessing downstream load in low-frequency load shedding systems. Specifically, this invention fully utilizes the advantages of convolutional neural networks in processing multivariate data, effectively capturing the complex nonlinear relationship between photovoltaic power and various influencing factors. This improves the accuracy and reliability of photovoltaic power prediction and solves the problem of significant discrepancies between the actual downstream load and the power detected by local devices on the feeders due to the large-scale distributed photovoltaic grid integration, leading to the shedding of a large amount of load on the feeders by low-frequency load shedding devices. This method is easy to implement in the field and has good application prospects.
[0038] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0039] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.
Claims
1. A CNN prediction method for photovoltaic power considering multivariable nonlinear spatial relationships, characterized in that, include: S1 collects meteorological data for the time to be predicted and the preceding 12 consecutive hours, photovoltaic power station characteristic data, and photovoltaic power data. The collected data is preprocessed by imputing missing values, correcting outliers, and normalizing based on the same extreme value parameters. S2, the sliding window method is used to extract features from the collected data, with a window length of 12 hours, and the collected data is composed of dimensions. The two-dimensional feature matrix, where N is the number of selected input variables; S3, the two-dimensional feature matrix is input into a pre-trained convolutional neural network model. The model sequentially performs convolution operations, non-linear activation, pooling dimensionality reduction, and fully connected mapping, and outputs the predicted value in the normalized domain. ; S4. Perform inverse normalization on the predicted values to obtain the photovoltaic active power prediction results. The formula is: in , These are the historical maximum and minimum values for this photovoltaic power, respectively.
2. The photovoltaic power CNN prediction method considering multivariable nonlinear spatial relationships according to claim 1, characterized in that, The meteorological data mentioned in step S1 includes solar radiation intensity, temperature, humidity, wind speed, and cloud cover, while the photovoltaic power station's own characteristic data includes geographical location and the degree of aging of the photovoltaic panels.
3. The photovoltaic power CNN prediction method considering multivariable nonlinear spatial relationships according to claim 1, characterized in that, The formula for normalization in step S1 is: in, and These are the minimum and maximum values of variable x in the dataset, respectively. It is the normalized value.
4. The photovoltaic power CNN prediction method considering multivariable nonlinear spatial relationships according to claim 1, characterized in that, The preprocessing described in step S1 also includes analyzing the correlation between the meteorological data and photovoltaic power using the Pearson correlation coefficient method, calculating the Pearson correlation coefficient between each meteorological data point and photovoltaic power, and selecting data with a strong correlation from them.
5. The photovoltaic power CNN prediction method considering multivariable nonlinear spatial relationships according to claim 1, characterized in that, The convolutional neural network model structure described in step S3 includes multiple convolutional layers, pooling layers, and fully connected layers, wherein: Convolutional layers are used to extract local features from the input data. Convolution operations are performed by sliding convolution kernels of different sizes across the input data. Pooling layers are used to reduce the dimensionality of data and reduce computation while retaining important feature information. Max pooling is used to take the maximum value within a fixed-size pooling window as the output. The fully connected layer integrates the features extracted by the preceding convolutional and pooling layers and outputs the final prediction result.
6. The photovoltaic power CNN prediction method considering multivariable nonlinear spatial relationships according to claim 1, characterized in that, The convolutional neural network model training method described in step S3 includes: Data partitioning: The dataset after feature selection and extraction is divided into training set, validation set and test set according to the ratio of 70%, 15% and 15% respectively; Training preparation: Define the loss function as mean squared error (MSE), use the stochastic gradient descent (SGD) algorithm for parameter updates, and set the initial learning rate to 0.01; Iterative training and monitoring: During the training iteration, the loss function value on the training set is calculated, and the weights and biases of the model are updated according to the gradient of the loss function and the learning rate. Adjustment and continuation: If the loss function value on the validation set does not decrease significantly in the last 100 iterations, adjust the learning rate and continue training.
7. The photovoltaic power CNN prediction method considering multivariable nonlinear spatial relationships according to claim 6, characterized in that, The evaluation and optimization method for the convolutional neural network model described in step S3 includes: Model evaluation: The trained model is evaluated using a test set, and the prediction accuracy of the model is evaluated using mean absolute error (MAE) and root mean square error (RMSE). Model optimization: When the model's evaluation metrics are not ideal, increasing the number of convolutional layers, changing the size of the convolutional kernels, and adjusting the number of neurons in the fully connected layers can help the model converge faster.
8. A photovoltaic power CNN prediction system considering multivariable nonlinear spatial relationships, characterized in that, The device includes an interconnected processor and a computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and the computer program is executed by the processor to implement the steps of the photovoltaic power CNN prediction method considering multivariable nonlinear spatial relationships as described in any one of claims 1 to 7.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that is executed by a processor to implement the steps of the photovoltaic power CNN prediction method considering multivariable nonlinear spatial relationships as described in any one of claims 1 to 7.
10. A computer program product, characterized in that, The method includes a computer program, which is executed by a processor to implement the steps of the photovoltaic power CNN prediction method considering multivariable nonlinear spatial relationships as described in any one of claims 1 to 7.