A photovoltaic power prediction method

Abstract

The present invention discloses a photovoltaic power prediction method, which relates to the technical field of smart grids. The method includes collecting and preprocessing data from multiple data sources to form a multi-modal data set; constructing a radiation transfer model to calculate the solar radiation intensity on the surface of photovoltaic modules; constructing a photovoltaic power prediction model based on historical data to capture the non-linear relationship between weather patterns and photovoltaic power output, and outputting the predicted photovoltaic power; training the photovoltaic power prediction model using the historical data set, and taking the solar radiation intensity as an auxiliary input to the model; collecting a real-time multi-modal data set, combining the real-time solar radiation intensity and inputting it into the trained photovoltaic power prediction model to perform real-time photovoltaic power prediction; dynamically adjusting the prediction strategy according to the error of the real-time prediction value and optimizing the model parameters; according to the photovoltaic power prediction result, combining the real-time demand of the power grid and the state of the energy storage system, efficiently allocating and managing electric power resources.

Description

Technical Field

[0001] The present invention relates to the technical field of smart grids, and particularly to a photovoltaic power prediction method. Background Art

[0002] As a clean and sustainable power generation method, photovoltaic energy has developed rapidly worldwide in recent years. With the continuous increase in photovoltaic installed capacity, how to accurately predict the output power of photovoltaic power plants has become one of the key challenges in power system planning, dispatching, and operation. The short-term and ultra-short-term prediction of photovoltaic power prediction technology is of great significance for optimizing power system operation, improving power market efficiency, and ensuring the safe and stable operation of the power system. Traditional photovoltaic power prediction methods mainly rely on single meteorological data, including temperature, humidity, and wind speed, and combine the characteristic parameters of photovoltaic modules for prediction. However, these methods often ignore the comprehensive influence of various factors, including cloud occlusion, air pollution, and changes in surface reflectivity, resulting in limited prediction accuracy.

[0003] In recent years, with the progress of big data, artificial intelligence, and Internet of Things technologies, multi-source data fusion and deep learning algorithms have been widely used in the field of photovoltaic power prediction. The construction of multi-modal data sets not only covers traditional meteorological parameters but also incorporates multi-source information such as satellite remote sensing, radar monitoring, and social media, improving the comprehensiveness and accuracy of prediction models. However, existing technologies still have limitations in dealing with non-linear relationships, real-time data fusion, and dynamic parameter adjustment. For example, most prediction models fail to fully consider the dynamic changes in solar radiation intensity, especially in the case of rapid cloud movement, resulting in poor real-time performance and adaptability of prediction results. In addition, the lack of an effective mechanism for real-time optimization of model parameters and the immature application in power dispatching and energy storage management are all bottlenecks restricting the further development of photovoltaic power prediction technology. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a photovoltaic power prediction method to solve the problems that existing technologies still have limitations in dealing with non-linear relationships, real-time data fusion, and dynamic parameter adjustment. For example, most prediction models fail to fully consider the dynamic changes in solar radiation intensity, especially in the case of rapid cloud movement, resulting in poor real-time performance and adaptability of prediction results. In addition, the lack of an effective mechanism for real-time optimization of model parameters and the immature application in power dispatching and energy storage management are all bottlenecks restricting the further development of photovoltaic power prediction technology.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a photovoltaic power prediction method, which includes:

[0008] Collect and preprocess data from multiple data sources to form a multi-modal data set;

[0009] Construct a radiation transfer model to calculate the solar radiation intensity on the surface of photovoltaic modules;

[0010] Construct a photovoltaic power prediction model based on historical data to capture the non-linear relationship between weather patterns and photovoltaic power output, and output the predicted photovoltaic power;

[0011] Train the photovoltaic power prediction model using a historical data set, and use the solar radiation intensity as an auxiliary input to the model;

[0012] Collect a real-time multi-modal data set, combine the real-time solar radiation intensity and input it into the trained photovoltaic power prediction model to perform real-time photovoltaic power prediction;

[0013] According to the error of the real-time prediction value, dynamically adjust the prediction strategy and optimize the model parameters;

[0014] According to the photovoltaic power prediction result, combine the real-time demand of the power grid and the state of the energy storage system to efficiently allocate and manage power resources.

[0015] As a preferred solution of the photovoltaic power prediction method of the present invention, wherein: the step of collecting and preprocessing data from multiple data sources to form a multi-modal data set is specifically as follows:

[0016] Collect meteorological data and preprocess the collected data;

[0017] Form the preprocessed data set into a multi-modal data matrix X, where X is an n×p matrix, n is the number of observations, and p is the number of variables;

[0018] Fuse the temperature and wind speed in the meteorological data to form a wind chill index, and the expression is:

[0019] WCI = 13.12 + 0.6215·T - 11.37·v 0.16 + 0.3965·T·v 0.16 ;

[0020] Wherein, WCI is the wind chill index, which represents the temperature perceived by the human skin under given temperature and wind speed conditions, T represents the air temperature, v represents the wind speed, and the specific values in the formula are constants;

[0021] The multi-modal matrix is X, and the expression is:

[0022] X = (WCI, I t));

[0023] Among them, I t represents the solar radiation intensity;

[0024] Use principal component analysis (PCA) to reduce the dimensionality of the data. The expression is:

[0025] Y = X·W;

[0026] Among them, X represents the multi-modal data matrix, W is the weight matrix composed of principal components, and Y represents the fused feature vector;

[0027] Organize the preprocessed data in time series to construct a multi-modal data set containing historical weather conditions and corresponding photovoltaic power outputs.

[0028] As a preferred solution of the photovoltaic power prediction method described in the present invention, among them: the steps of constructing the radiation transfer model and calculating the solar radiation intensity on the surface of the photovoltaic module are specifically as follows:

[0029] Adopt the two-stream approximation radiation transfer model as the radiation transfer model;

[0030] Adopt the Pearson-Blankman formula to calculate the direct radiation intensity. The expression is:

[0031] I d = I0·cos(δ) / z;

[0032] Among them, I0 represents the solar constant, δ represents the solar radiation angle, z represents the air mass number, and cos represents the cosine value;

[0033] Adopt the Ångström formula to calculate the diffuse radiation intensity. The expression is:

[0034] I s = π·I0·ρ·sin(δ)·cos(L) / (1 - Cos(δ));

[0035] Among them, π represents the pi, ρ represents the surface albedo, δ represents the solar radiation angle, L represents the latitude of the geographical location, and sin represents the sine value;

[0036] The calculation expression of the solar radiation intensity is:

[0037] I t = I d ·T a ·(1 - R a ) + I s ·(1 - R s )(1 - ρ);

[0038] Among them, I t represents the total solar radiation intensity on the surface of the photovoltaic module, Id Denotes the direct radiation intensity, I s Denotes the scattered radiation intensity, R a Denotes the atmospheric reflectivity, T a Denotes the atmospheric transmittance R s Denotes the surface reflectivity.

[0039] As a preferred solution of the photovoltaic power prediction method described in the present invention, wherein: the photovoltaic power prediction model is constructed based on historical data to capture the non-linear relationship between weather patterns and photovoltaic power output, and the predicted photovoltaic power is output. The specific steps are as follows:

[0040] An LSTM network is used as the core model of the photovoltaic power prediction model;

[0041] Extract the deep features of the meteorological data and convert these data into a form suitable for input to the photovoltaic power prediction model. The expression is:

[0042]

[0043] where F represents the deep features, y i represents the i-th input feature, w i represents the weight value of y i , μ i represents the average value of the i-th feature, is a regulation factor, and p represents the number of input features;

[0044] Based on LSTM, construct the structure of the photovoltaic power prediction model. The expression is:

[0045] h t = LSTM(h t-1 , F);

[0046] where h t represents the hidden state at the current moment, which is used to predict the photovoltaic power at the current time point, and h t -1 respectively represent the hidden states at the previous moment;

[0047] The photovoltaic power prediction value is obtained through the activation function ReLU, which is expressed as:

[0048]

[0049] where represents the photovoltaic power prediction value at time point t, W out represents the weight matrix of the output layer, and b out represents the bias term of the output layer.

[0050] As a preferred embodiment of the photovoltaic power prediction method of the present invention, the steps of training the photovoltaic power prediction model using the historical dataset with solar radiation intensity as an auxiliary input to the model are as follows:

[0051] Collect a large amount of historical data, preprocess and extract features to form a historical dataset;

[0052] Divide the historical dataset into a training set, a validation set, and a test set. Input the training set, the validation set, and the test set, along with the input to the photovoltaic power prediction model, to complete the training, validation, and testing of the model;

[0053] Based on the solar radiation intensity and the depth features, construct a comprehensive feature fusion function. The expression of the comprehensive feature fusion function is:

[0054]

[0055] where H t represents the comprehensive feature at time point t, F i represents the i-th depth feature, f represents the hyperbolic tangent function Tanh, g(I t ) represents the non-linear transformation of the solar radiation intensity I t , w i and γ are the weights of the depth feature and the solar radiation intensity respectively, μ i and μ I represent the average values of the i-th feature and the solar radiation intensity respectively, and α and β are adjustment factors;

[0056] Use the training set data to input into the photovoltaic power prediction model for training, and introduce a loss function to calculate the mean square error L between the predicted photovoltaic power and the actual photovoltaic power;

[0057] Use the validation set data to validate the photovoltaic power prediction model and adjust the model parameters to prevent overfitting;

[0058] Input the comprehensive feature vector of the test set into the model to generate the photovoltaic power prediction value

[0059] As a preferred embodiment of the photovoltaic power prediction method of the present invention, the steps of collecting a real-time multi-modal dataset, combining the real-time solar radiation intensity, and inputting them into the trained photovoltaic power prediction model for real-time photovoltaic power prediction are as follows:

[0060] Real-time collect the latest data from national meteorological stations, international meteorological satellite organizations, local radar monitoring networks, and social media;

[0061] Using the Z-score method to detect and process outliers in real-time data, using the K-nearest neighbor interpolation method to fill in missing values, and then normalizing the data to form a real-time multi-modal data matrix;

[0062] Fusing the temperature and wind speed in the real-time meteorological data to form the wind chill index;

[0063] Using principal component analysis (PCA) to reduce the dimension of the real-time data to form a real-time fusion feature vector;

[0064] Calculating the direct radiation intensity and diffuse radiation intensity in real-time through the Pearson-Brackmann formula and the Ångström formula;

[0065] Inputting the real-time fusion feature vector and the total solar radiation intensity calculated in real-time into the photovoltaic power prediction model to obtain the comprehensive feature fusion function;

[0066] Through the current moment hidden state and the comprehensive feature fusion function, obtaining the real-time photovoltaic power prediction value, and the expression is:

[0067] P tt =W out ·h t (H tt )+b out ;

[0068] Among them, P tt represents the real-time photovoltaic power prediction value, W out represents the matrix of the output layer, H tt represents the comprehensive feature fusion function at time point t, and b out represents the bias term of the output layer;

[0069] Comparing the real-time photovoltaic power prediction value with the actual photovoltaic power, and using the mean square error (MSE) to evaluate the prediction accuracy.

[0070] As a preferred solution of the photovoltaic power prediction method described in the present invention, wherein: according to the error of the real-time prediction value, dynamically adjusting the prediction strategy and optimizing the model parameters, the specific steps are as follows:

[0071] Calculating the root mean square error of the predicted photovoltaic power and the actual photovoltaic power, and the expression is:

[0072]

[0073] Among them, RMSE represents the root mean square error value, represents the summation of a series of terms with subscript t from 1 to N, represents the predicted photovoltaic power at time point t, and P t represents the actual photovoltaic power at the same time point;

[0074] Calculate the mean absolute percentage error between the predicted photovoltaic power and the actual photovoltaic power. The expression is as follows:

[0075]

[0076] Among them, MAPE represents the mean absolute percentage error value;

[0077] Update the model parameters using the adaptive learning rate algorithm based on the Adam optimizer;

[0078] Construct a comprehensive performance index based on RMSE and MAPE. The expression is as follows:

[0079]

[0080] Among them, η is the comprehensive performance index, RMSE t represents the preset RMSE threshold, and MAPE t represents the preset MAPE threshold;

[0081] When the prediction error is lower than RMSE t and MAPE t , η will be greater than 0, otherwise less than 0;

[0082] Input η into the adjustment formula. The expression is as follows:

[0083] α = α · (1 + η · γ);

[0084] Among them, α represents the learning rate, and γ represents the learning rate adjustment factor;

[0085] Continuously iterate and update until the prediction error reaches the standard.

[0086] As a preferred solution of the photovoltaic power prediction method described in the present invention, wherein: according to the photovoltaic power prediction result, combined with the real-time demand of the power grid and the state of the energy storage system, efficiently allocate and manage the power resources. The specific steps are as follows:

[0087] Construct a comprehensive performance index based on the power dispatch efficiency Ξ, the health degree Ω of the energy storage system, and the system operation cost Λ. The expression is as follows:

[0088]

[0089] Among them, Ψ represents the comprehensive performance index;

[0090] Adopt a questionnaire survey to collect users' satisfaction scores on power dispatch, energy storage system, and operation cost;

[0091] Construct a user feedback index based on user feedback and model prediction results. The expression is as follows:

[0092]

[0093] Among them, Φ represents the user feedback index, and S sc represents the satisfaction score of users with the power dispatching efficiency, and S st represents the satisfaction score of users with the health and usage experience of the energy storage system, and S co represents the satisfaction score of users with the system operation cost, and S max represents the score when users are completely satisfied, and S min represents the score when users are completely dissatisfied;

[0094] Dynamically adjust the model parameters according to Φ and Ψ, and the expression is:

[0095]

[0096] Among them, α new represents the updated new learning rate, and α old represents the current learning rate, Ψ t and Φ t respectively represent the preset comprehensive performance index target value and the preset user feedback index target value.

[0097] In a second aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the photovoltaic power prediction method described in the first aspect of the present invention is implemented.

[0098] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, any step of the photovoltaic power prediction method described in the first aspect of the present invention is implemented.

[0099] The beneficial effects of the present invention are as follows: By constructing a multi-modal data set and using a radiative transfer model and deep learning technology, the present invention realizes high-precision real-time prediction of photovoltaic power. It not only covers traditional meteorological data, but also integrates remote sensing, radar, and social data to form a comprehensive multi-modal data set, enhancing the robustness and generalization ability of the prediction model. By accurately calculating the solar radiation intensity on the surface of photovoltaic modules and combining with the photovoltaic power prediction model trained with historical data, it can capture the complex non-linear relationship between weather patterns and photovoltaic power output, improving the prediction accuracy. In addition, the use of real-time multi-modal data sets, combined with a dynamic parameter adjustment mechanism, ensures the rapid response and adaptation of the model to real-time environmental changes. More importantly, this method combines the prediction results with power dispatching, energy storage management, and system operation optimization, realizing the closed-loop optimization of photovoltaic power prediction in the practical application of power systems, and providing strong support for the efficient allocation and management of power resources. Brief Description of the Drawings

[0100] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0101] Figure 1 It is a flowchart of the photovoltaic power prediction method in Embodiment 1.

[0102] Figure 2 It is a flowchart of the solar radiation intensity calculation in Embodiment 1. Detailed Embodiments

[0103] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific embodiments of the present invention in conjunction with the drawings of the specification.

[0104] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0105] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selectively exclusive embodiment from other embodiments.

[0106] Embodiment 1, referring to Figure 1 and Figure 2 , is the first embodiment of the present invention. This embodiment provides a photovoltaic power prediction method, including the following steps:

[0107] S1 Collect and preprocess data from multiple data sources to form a multi-modal data set;

[0108] Collect meteorological data from the channels of national meteorological stations. In order to ensure the timeliness and accuracy of the data, an API interface is used to capture the latest data in real time;

[0109] Preprocess the collected data

[0110] Form a multi-modal data matrix X from the preprocessed data set. X is an n×p matrix, where n is the number of observations and p is the number of variables;

[0111] Fuse the temperature and wind speed in meteorological data to form the wind chill index, and the expression is:

[0112] WCI = 13.12 + 0.6215·T - 11.37·v 0.16 + 0.3965·T·v 0.16 ;

[0113] Among them, WCI is the wind chill index, which represents the temperature perceived by the human skin under given temperature and wind speed conditions. T represents the air temperature, and v represents the wind speed. The specific values in the formula are constants obtained through experiments and data analysis, and are used to accurately reflect the relationship between temperature, wind speed and the degree of cold felt by the human body;

[0114] The multi-modal matrix is X, and the expression is:

[0115] X = (WCI, I t );

[0116] Among them, I t represents the solar radiation intensity;

[0117] Use principal component analysis PCA to reduce the dimension of the data and reduce redundant information. The expression is:

[0118] Y = X·W;

[0119] Among them, X represents the multi-modal data matrix, W is the weight matrix composed of principal components, W is a p×k weight matrix, k is the number of principal components selected, which is equal to p, Y represents the fused feature vector, and Y is an n×k matrix, representing the projection of the original data in the principal component space, that is, the dimensionality reduction representation of the data;

[0120] Organize the preprocessed data in a time series to construct a multi-modal dataset containing historical weather conditions and corresponding photovoltaic power outputs;

[0121] Fill in the missing values, and the expression is:

[0122]

[0123] Among them, represents the filling result of the missing value, k represents the k value in the k-nearest neighbor interpolation method, and x j represents the values of k nearest neighbor data points;

[0124] Meteorological data includes temperature, wind speed and solar radiation.

[0125] S2 Construct a radiation transfer model to calculate the solar radiation intensity on the surface of the photovoltaic module;

[0126] The two-stream approximate radiative transfer model is used as the radiative transfer model, which has high accuracy in dealing with atmospheric scattering and absorption processes and good computational efficiency;

[0127] The direct radiation intensity is calculated using the Pearson-Blackman formula, which is expressed as:

[0128] I d =I0·cos(δ) / z;

[0129] Among them, I0 represents the solar constant, δ represents the solar radiation angle, z represents the air mass number, and cos represents the cosine value;

[0130] The Assmann formula is used to calculate the scattered radiation intensity, and the expression is:

[0131] I s =π·I0·ρ·sin(δ)·cos(L) / (1-cos(δ));

[0132] Among them, π represents pi, ρ represents the surface albedo, which varies with the surface type and the sun angle, δ represents the solar radiation angle, L represents the latitude of the geographical location, and sin represents the sine value;

[0133] The calculation expression of solar radiation intensity is:

[0134] I t =I d ·T a ·(1-R a )+I s ·(1-R s )·(1-ρ);

[0135] Among them, I t Represents the total solar radiation intensity on the surface of the photovoltaic module, I d Indicates the intensity of direct radiation, which is affected by the solar altitude angle, solar azimuth angle, atmospheric pressure, temperature and altitude. s Represents the intensity of scattered radiation, which is affected by air mass number, surface albedo, latitude and observation date. a Represents the atmospheric reflectivity, which is related to the aerosol and water vapor content in the atmosphere. a Represents atmospheric transmittance, which is affected by the absorption and scattering of gas molecules in the atmosphere. s Represents the reflectivity of the surface, which is affected by the type of surface cover and moisture.

[0136] S3 builds a PV power prediction model based on historical data, captures the nonlinear relationship between weather patterns and PV power output, and outputs the predicted PV power;

[0137] The LSTM network is adopted as the core model of the photovoltaic power prediction model because LSTM can handle the long-term dependence problem in time series data, as the photovoltaic power is affected by the periodic and seasonal changes of weather conditions;

[0138] Extract the deep features of meteorological data and convert these data into a form suitable for input to the photovoltaic power prediction model. The expression is:

[0139]

[0140] where F represents the deep feature, y i represents the i-th input feature, w i represents the weight value of y i , μ i represents the average value of the i-th feature, is a regulation factor used to control the influence degree of the feature value deviating from the average value on the model output, and p represents the number of input features;

[0141] Construct the photovoltaic power prediction model structure based on LSTM. The expression is:

[0142] h t = LSTM(h t-1 , F);

[0143] where h t represents the hidden state at the current moment, which is used to predict the photovoltaic power at the current time point, and h t -1 represents the hidden state at the previous moment respectively;

[0144] Obtain the photovoltaic power prediction value through the activation function ReLU, which is expressed as:

[0145]

[0146] where, represents the photovoltaic power prediction value at time point t, W out represents the weight matrix of the output layer, and b out represents the bias term of the output layer.

[0147] In S4, use the historical data set to train the photovoltaic power prediction model, and use the solar radiation intensity as the auxiliary input of the model;

[0148] Collect a large amount of historical data, and after preprocessing and feature extraction, combine them into a historical data set;

[0149] Divide the historical dataset into a training set, a validation set, and a test set. In this embodiment, the proportions are 70%, 15%, and 15% respectively to ensure the generalization ability of the model and the accuracy of evaluation. Input the training set, the validation set, and the test set, as well as the input photovoltaic power prediction model to complete the training, validation, and testing of the model;

[0150] Construct a comprehensive feature fusion function based on solar radiation intensity and depth features to enhance the prediction ability of the model. The expression of the comprehensive feature fusion function is:

[0151]

[0152] where H t represents the comprehensive feature at time point t, F i represents the i-th depth feature, f represents the hyperbolic tangent function Tanh, g(I t ) represents the non-linear transformation performed on the solar radiation intensity I t , w i and γ are the weights of the depth feature and the solar radiation intensity respectively, μ i and μ I represent the average values of the i-th feature and the solar radiation intensity respectively, and α and β are adjustment factors used to control the influence degree of the feature value deviating from the average value on the model output;

[0153] Use the training set data to input the photovoltaic power prediction model for training, and introduce a loss function to calculate the mean square error L between the predicted photovoltaic power and the actual photovoltaic power;

[0154] Use the validation set data to validate the photovoltaic power prediction model, and adjust the model parameters to prevent overfitting;

[0155] Input the comprehensive feature vector of the test set into the model to generate the photovoltaic power prediction value

[0156] The values of the two adjustment factors α and β are directly set according to the experience of domain experts, or can also be set according to actual needs.

[0157] S5 Collect the real-time multimodal dataset, combine the real-time solar radiation intensity, input the trained photovoltaic power prediction model, and perform real-time photovoltaic power prediction;

[0158] Collect the latest data from national meteorological stations, international meteorological satellite organizations, local radar monitoring networks, and social media in real time;

[0159] Use the Z-score method to detect and process outliers in the real-time data, use the K-nearest neighbor interpolation method to fill in missing values, and then perform standardization processing on the data to form a real-time multimodal data matrix;

[0160] Fuse the temperature and wind speed in the real-time meteorological data to form the wind chill index;

[0161] Use principal component analysis (PCA) to reduce the dimension of the real-time data to form a real-time fusion feature vector;

[0162] Calculate the direct radiation intensity and diffuse radiation intensity in real time through the Pearson-Brackmann formula and the Angstrom formula;

[0163] Input the real-time fusion feature vector and the total solar radiation intensity calculated in real time into the photovoltaic power prediction model to obtain the comprehensive feature fusion function;

[0164] Obtain the real-time photovoltaic power prediction value through the current moment hidden state and the comprehensive feature fusion function. The expression is:

[0165] P tt =W out ·h t (H tt )+b out ;

[0166] Among them, P tt represents the real-time photovoltaic power prediction value, W out represents the matrix of the output layer, H tt represents the comprehensive feature fusion function at time point t, and b out represents the bias term of the output layer;

[0167] Compare the real-time photovoltaic power prediction value with the actual photovoltaic power, and use the mean square error (MSE) to evaluate the prediction accuracy to ensure the reliability and accuracy of the model in the real-time prediction scenario.

[0168] S6 Dynamically adjust the prediction strategy and optimize the model parameters according to the error of the real-time prediction value;

[0169] Calculate the root mean square error of the predicted photovoltaic power and the actual photovoltaic power. The expression is:

[0170]

[0171] Among them, RMSE represents the root mean square error value, represents the sum of a series of terms with subscript t from 1 to N, represents the predicted photovoltaic power at time point t, and P t represents the actual photovoltaic power at the same time point;

[0172] Calculate the mean absolute percentage error of the predicted photovoltaic power and the actual photovoltaic power. The expression is:

[0173]

[0174] Among them, MAPE represents the mean absolute percentage error value;

[0175] Based on the Adam optimizer, the adaptive learning rate algorithm is used to update the model parameters. Let the model parameters be θ and the learning rate be α. The update expression is:

[0176] m t = β1m t-1 + (1 - β1)g t (θ);

[0177] v t = β2v t-1 + (1 - β2)g t (θ) 2 ;

[0178]

[0179] Among them, β1 and β2 are the decay rates of the first and second matrices respectively, ε represents a small number to prevent division by zero, gt(θ) represents the gradient of the parameter θ at time point t, m t represents the exponentially weighted moving average at time point t, v t represents the exponentially weighted moving average at time point t, represents the updated exponentially weighted moving average at time point t, represents the updated exponentially weighted moving average at time point t;

[0180] Based on RMSE and MAPE, a comprehensive performance index is constructed. The expression is:

[0181]

[0182] Among them, η is the comprehensive performance index, that is, the error adjustment coefficient. The larger the value, the better the prediction performance of the model. RMSE t represents the preset RMSE threshold. When the RMSE of the photovoltaic power prediction model is lower than the preset threshold, the model performs well. MAPE t represents the preset MAPE threshold. When the MAPE of the photovoltaic power prediction model is lower than the preset threshold, the model performs well;

[0183] When the prediction error is lower than RMSE t and MAPE t , η will be greater than 0, otherwise less than 0;

[0184] Input η into the adjustment formula to adjust the learning rate. The expression is:

[0185] α = α·(1 + η·γ);

[0186] Among them, α represents the learning rate, and γ represents the learning rate adjustment factor, which is used to control the increase and decrease speed of the learning rate;

[0187] Keep iterating and updating until the prediction error reaches the standard.

[0188] S7 efficiently allocates and manages power resources according to the photovoltaic power prediction results, combined with the real-time demand of the power grid and the state of the energy storage system;

[0189] Construct a comprehensive performance index based on the power dispatching efficiency Ξ, the health degree Ω of the energy storage system, and the system operation cost Λ. The expression is:

[0190]

[0191] Among them, Ψ represents the comprehensive performance index. The larger its value, the more efficient, economical, and durable the overall system operation is;

[0192] Use a questionnaire survey to collect users' satisfaction ratings on power dispatching, energy storage systems, and operation costs

[0193] ratings;

[0194] Construct a user feedback index based on user feedback and model prediction results. The expression is:

[0195]

[0196] Among them, Φ represents the user feedback index, S sc represents the user's satisfaction rating on power dispatching efficiency, S st represents the user's satisfaction rating on the health degree and usage experience of the energy storage system, S co represents the user's satisfaction rating on the system operation cost, S max represents the rating when the user is completely satisfied, S min represents the rating when the user is completely dissatisfied;

[0197] Dynamically adjust the model parameters according to Φ and Ψ. The expression is:

[0198]

[0199] Among them, α new represents the updated new learning rate, α old represents the current learning rate, Ψ t and Φ t respectively represent the preset target value of the comprehensive performance index and the preset target value of the user feedback index;

[0200] Power dispatching efficiency, which represents the dispatching response speed and matching degree of the power system, is determined by the real-time response time of the power dispatching center and the execution deviation of the dispatching plan. The expression is:

[0201]

[0202] Among them, t re represents the actual response time at time point t, indicating t ta represents the target response time, and N represents the total number of time points;

[0203] The health of the energy storage system, which represents the number of charge and discharge cycles and battery life loss of the energy storage device under predicted power fluctuations. The expression is:

[0204] Ω = e^(-λ·C);

[0205] Among them, Ω represents the health of the energy storage system. The value range of Ω is between 0 and 1. The closer the value is to 1, the better the health of the energy storage system. e is a data constant, -λ is a positive real number representing the attenuation coefficient of the energy storage device. The larger the value of λ, the faster the health of the energy storage system decreases in each charge and discharge cycle, and C represents the number of charge and discharge cycles;

[0206] The system operation cost, including electricity trading cost, energy storage cost, and system maintenance cost. The expression is:

[0207] Λ = α·P mt + β·P se + γ·P me ;

[0208] Among them, Λ represents the overall index of the system operation cost, α, β, and γ are the weight values of the electricity trading cost, energy storage cost, and system maintenance cost respectively, and P mt 、P se and P me are the electricity trading cost, energy storage cost, and system maintenance cost respectively.

[0209] This embodiment also provides a computer device applicable to the case of the photovoltaic power prediction method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the photovoltaic power prediction method proposed in the above embodiment.

[0210] The computer device may be a terminal, which includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or may be a button, a trackball, or a touchpad provided on the housing of the computer device, or may also be an external keyboard, touchpad, or mouse, etc.

[0211] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the photovoltaic power prediction method proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (abbreviated as SRAM), electrically erasable programmable read-only memory (abbreviated as EEPROM), erasable programmable read-only memory (abbreviated as EPROM), programmable read-only memory (abbreviated as PROM), read-only memory (abbreviated as ROM), magnetic memory, flash memory, a magnetic disk, or an optical disc.

[0212] In summary, the present invention realizes high-precision real-time prediction of photovoltaic power by: constructing a multi-modal dataset and utilizing the radiative transfer model and deep learning technology. It not only covers traditional meteorological data but also integrates remote sensing, radar, and social data to form a comprehensive multi-modal dataset, enhancing the robustness and generalization ability of the prediction model. By accurately calculating the solar radiation intensity on the surface of photovoltaic modules and combining with the photovoltaic power prediction model trained with historical data, it can capture the complex non-linear relationship between weather patterns and photovoltaic power output, improving the prediction accuracy. In addition, the application of the real-time multi-modal dataset, combined with the dynamic parameter adjustment mechanism, ensures the rapid response and adaptation of the model to real-time environmental changes. More importantly, this method combines the prediction results with power dispatch, energy storage management, and system operation optimization, realizing the closed-loop optimization of photovoltaic power prediction in the practical application of the power system and providing strong support for the efficient allocation and management of power resources.

[0213] Example 2, referring to Table 1, is the second example of the present invention. To further verify the advancement of the present invention, experimental simulation data of the photovoltaic power prediction method is given.

[0214] To verify the effectiveness and superiority of the photovoltaic power prediction method, a series of experiments were carried out. A typical photovoltaic power station was selected as the research object. The power station is located at 30°N latitude and 120°E longitude and has a total installed capacity of 10 MW. During the experiment, from July 1, 2023, to July 31, 2023, for a total of 31 days, the time interval for collecting data was once per hour, that is, 24 groups of data were collected in a day, and a total of 744 groups of data samples were obtained.

[0215] During the data collection phase, real-time meteorological data, satellite remote sensing data, and radar data were obtained from multi-source channels such as national meteorological stations, international meteorological satellite organizations, local radar monitoring networks, and social media platforms. These data were subjected to Z-score outlier identification and processing, as well as missing value filling using the K-nearest neighbor interpolation method, and a multi-modal dataset was formed through standardization processing.

[0216] When constructing the radiative transfer model, the two-stream approximation radiative transfer model and the Pearson-Blankman formula were used to calculate the direct radiation intensity, and the Angstrom formula was used to calculate the scattered radiation intensity. Finally, the total solar radiation intensity on the surface of the photovoltaic module was obtained.

[0217] In the stage of constructing the photovoltaic power prediction model, the LSTM network was selected as the core model, and at the same time, the deep features of the wind chill index, meteorological data, satellite remote sensing data, and radar data were combined. Feature fusion was carried out through formulas to establish a photovoltaic power prediction model. The model was trained using the historical dataset, and the solar radiation intensity was used as an auxiliary input.

[0218] In the real-time prediction phase, a real-time multi-modal dataset was collected, and the same data preprocessing process as the historical data was carried out. Real-time photovoltaic power prediction was performed through the model. Subsequently, according to the error of the real-time prediction value, the prediction strategy was dynamically adjusted and the model parameters were optimized until the prediction error reached the predetermined standard.

[0219] Finally, based on the photovoltaic power prediction results, combined with the real-time demand of the power grid and the state of the energy storage system, the electric power resources were efficiently allocated and managed. The model parameters were dynamically adjusted through comprehensive performance indicators and user feedback indicators to optimize the allocation of electric power resources.

[0220] Specifically, as shown in Table 1:

[0221] Table 1 Experimental Record Table

[0222]

[0223]

[0224] Through the analysis of the experimental data, it can be clearly seen that the proposed photovoltaic power prediction method has significant advantages in terms of accuracy and stability. For example, at around 12:00 noon on July 15th, when the solar radiation reached the peak, the actual photovoltaic power was 8500 kW, while the predicted photovoltaic power was 8520 kW, and the prediction error was only 0.24%. This indicates that the model can accurately capture the non-linear relationship between weather patterns and photovoltaic power output, especially in the case of rapidly changing lighting conditions.

[0225] In addition, by calculating the root mean square error (RMSE) and mean absolute percentage error (MAPE) during the entire experimental period, it was found that the RMSE was 150 kW and the MAPE was 1.7%, which were much lower than 300 kW and 3.5% of the traditional prediction methods. This shows that the photovoltaic power prediction method in the present invention not only improves the prediction accuracy, but also performs excellently in real-time prediction and dynamic parameter adjustment, and can more effectively cope with the uncertainty of the weather.

[0226] Compared with the prior art, the method of the present invention significantly improves the accuracy and real-time performance of photovoltaic power prediction through the construction of a multi-modal dataset, the fusion of deep features, and the application of the LSTM network. Especially in dynamically adjusting the prediction strategy and optimizing the model parameters, the present invention can self-correct according to the real-time prediction error to ensure the reliability of the prediction results, and thus promote the efficient allocation and management of electric power resources, providing strong support for the stable operation and economic benefits of the power system.

[0227] In summary, the photovoltaic power prediction method of the present invention has remarkable creativity and novelty in improving prediction accuracy, enhancing real-time performance and adaptability, can effectively solve the deficiencies of the prior art in photovoltaic power prediction, and has important practical significance for promoting the development of smart grid technology and renewable energy.

[0228] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A photovoltaic power prediction method, characterized in that: including, collecting and preprocessing data from multiple data sources to form a multimodal dataset; constructing a radiative transfer model to calculate the solar radiation intensity on the surface of the photovoltaic module; constructing a photovoltaic power prediction model based on historical data to capture the non-linear relationship between weather patterns and photovoltaic power output, and outputting the predicted photovoltaic power; training the photovoltaic power prediction model using the historical dataset, with the solar radiation intensity as an auxiliary input to the model; collecting a real-time multimodal dataset, combining the real-time solar radiation intensity and inputting it into the trained photovoltaic power prediction model to perform real-time photovoltaic power prediction; dynamically adjusting the prediction strategy and optimizing the model parameters according to the error of the real-time prediction value; efficiently allocating and managing power resources according to the photovoltaic power prediction results, in combination with the real-time demand of the power grid and the state of the energy storage system; The specific steps of collecting and preprocessing data from multiple data sources to form a multimodal dataset are as follows: collecting meteorological data and preprocessing the collected data; forming a multimodal data matrix X from the preprocessed data set. X is an n×p matrix, where n is the number of observations and p is the number of variables; fusing the temperature and wind speed in the meteorological data to form the wind chill index, with the expression: ； where WCI is the wind chill index, representing the temperature perceived by the human skin under given temperature and wind speed conditions, T represents the air temperature, v represents the wind speed, and the specific values in the formula are constants; The multimodal data matrix is X, with the expression: ； Among them, I t represents the solar radiation intensity; using principal component analysis (PCA) to reduce the dimensionality of the data, with the expression: ； where X represents the multimodal data matrix, W is the weight matrix composed of principal components, and Y represents the fused feature vector; organizing the preprocessed data in a time series to construct a multimodal dataset containing historical weather conditions and corresponding photovoltaic power outputs; The specific steps of constructing a radiative transfer model to calculate the solar radiation intensity on the surface of the photovoltaic module are as follows: adopting the two-stream approximation radiative transfer model as the radiative transfer model; using the Pearson-Blankman formula to calculate the direct radiation intensity, with the expression: ； where I0 represents the solar constant, δ represents the solar radiation angle, z represents the air mass number, and cos represents the cosine value; using the Ångström formula to calculate the scattered radiation intensity, with the expression: ； where π represents the pi, represents the surface albedo, δ represents the solar radiation angle, L represents the latitude of the geographical location, and sin represents the sine value; The expression for calculating the solar radiation intensity is: ； Among them, I t represents the total solar radiation intensity on the surface of the photovoltaic module, I d represents the direct radiation intensity, I s represents the diffuse radiation intensity, R a represents the atmospheric reflectivity, T a represents the atmospheric transmittance, R s represents the surface reflectivity.

2. The photovoltaic power prediction method according to claim 1, wherein: The specific steps of constructing a photovoltaic power prediction model based on historical data to capture the non-linear relationship between weather patterns and photovoltaic power output, and outputting the predicted photovoltaic power are as follows: adopting the LSTM network as the core model of the photovoltaic power prediction model; extracting the deep features of the meteorological data and converting these meteorological data into a form suitable for input to the photovoltaic power prediction model, with the expression: ； Among them, F represents the depth feature, and y i represents the i-th input feature, and w i represents the weight value of y i , μ i represents the average value of the i-th feature, is the adjustment factor, and p represents the number of input features; constructing the structure of the photovoltaic power prediction model based on LSTM, with the expression: ； where h t represents the hidden state at the current moment, which is used to predict the photovoltaic power at the current time point, and h t -1 respectively represent the hidden states at the previous moment; obtaining the photovoltaic power prediction value through the activation function ReLU, expressed as: ； Among them, represents the predicted value of photovoltaic power at time point t, in W out represents the weight matrix of the output layer, b out represents the bias term of the output layer.

3. The photovoltaic power prediction method according to claim 2, wherein: The specific steps of training the photovoltaic power prediction model using the historical dataset, with the solar radiation intensity as an auxiliary input to the model are as follows: collecting a large amount of historical data, preprocessing and feature extraction to form a historical dataset; Divide the historical dataset into a training set, a validation set, and a test set, and input the training set, validation set, and test set into the photovoltaic power prediction model to complete the training, validation, and testing of the model; Construct a comprehensive feature fusion function based on solar radiation intensity and depth features. The expression of the comprehensive feature fusion function is: ； Among them, H t represents the comprehensive feature at time point t, F i represents the i-th depth feature, f represents the hyperbolic tangent function Tanh, g(I t ) represents the non-linear transformation performed on the solar radiation intensity I t , and γ are the weights of the depth feature and the solar radiation intensity respectively, μ i and μ I represent the average values of the i-th feature and the solar radiation intensity respectively, and α and β are adjustment factors; Use the training set data to input into the photovoltaic power prediction model for training, and introduce a loss function to calculate the mean square error L between the predicted photovoltaic power and the actual photovoltaic power; Use the validation set data to verify the photovoltaic power prediction model, and adjust the model parameters to prevent overfitting; Input the comprehensive feature vector of the test set into the model to generate the predicted value of photovoltaic power .

4. The photovoltaic power prediction method according to claim 3, wherein: Collect the real-time multimodal dataset, combine the real-time solar radiation intensity, and input it into the trained photovoltaic power prediction model for real-time photovoltaic power prediction. The specific steps are as follows: Collect the latest data from national meteorological stations, international meteorological satellite organizations, local radar monitoring networks, and social media in real time; Use the Z-score method to detect and process outliers in the real-time data, use the K-nearest neighbor interpolation method to fill in missing values, and then perform standardization processing on the data to form a real-time multimodal data matrix; Use principal component analysis PCA to reduce the dimension of the real-time data to form a real-time fusion feature vector; Calculate the direct radiation intensity and diffuse radiation intensity in real time through the Pearson-Brackmann formula and the Angstrom formula; Input the real-time fusion feature vector and the total solar radiation intensity calculated in real time into the photovoltaic power prediction model to obtain the comprehensive feature fusion function; Through the current moment hidden state and the comprehensive feature fusion function, obtain the real-time photovoltaic power prediction value. The expression is: ； Among them, P tt represents the predicted value of real-time photovoltaic power, W out represents the weight matrix of the output layer, H tt represents the comprehensive feature fusion function at time point t, b out represents the bias term of the output layer; Compare the real-time photovoltaic power prediction value with the actual photovoltaic power, and use the mean square error MSE to evaluate the prediction accuracy.

5. The photovoltaic power prediction method according to claim 4, wherein: According to the error of the real-time prediction value, dynamically adjust the prediction strategy and optimize the model parameters. The specific steps are as follows: Calculate the root mean square error between the predicted photovoltaic power and the actual photovoltaic power. The expression is: ； Among them, RMSE represents the root mean square error value, represents the sum of a series of terms with subscript t ranging from 1 to N, represents the predicted photovoltaic power at time point t, represents the actual photovoltaic power at the same time point; Calculate the mean absolute percentage error between the predicted photovoltaic power and the actual photovoltaic power. The expression is: ； Where, MAPE represents the mean absolute percentage error value; Update the model parameters using the adaptive learning rate algorithm based on the Adam optimizer; Construct a comprehensive performance index based on RMSE and MAPE. The expression is: ； where η is the comprehensive performance index, RMSE t represents the preset RMSE threshold, MAPE t represents the preset MAPE threshold; When the prediction error is lower than RMSE t and MAPE t η will be greater than 0, and vice versa less than 0; Input η into the adjustment formula. The expression is: ； where α represents the learning rate, represents the learning rate adjustment factor; Continuously iterate and update until the prediction error reaches the standard.

6. The photovoltaic power prediction method according to claim 5, wherein: According to the photovoltaic power prediction result, combine the real-time demand of the power grid and the state of the energy storage system to efficiently allocate and manage power resources. The specific steps are as follows: Construct a comprehensive performance index based on the power dispatch efficiency Ξ, the health degree Ω of the energy storage system, and the system operation cost Λ. The expression is: ； Where, Ψ represents the comprehensive performance index; Adopt a questionnaire survey to collect users' satisfaction scores on power dispatch, energy storage system, and operation cost; Construct a user feedback index based on user feedback and model prediction results. The expression is: ； Among them, Φ represents the user feedback index, represents the satisfaction score of users with the power dispatching efficiency, represents the satisfaction score of users with the health and usage experience of the energy storage system, represents the satisfaction score of users with the system operation cost, S max represents the score when users are completely satisfied, S min represents the score when users are completely dissatisfied; Dynamically adjust the model parameters according to Φ and Ψ. The expression is: ； Among them, α new represents the updated new learning rate, and α old represents the current learning rate. Ψ t and Φ t respectively represent the preset target value of the comprehensive performance index and the preset target value of the user feedback index.

7. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the photovoltaic power prediction method according to any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the photovoltaic power prediction method according to any one of claims 1 to 6.

Images