Photovoltaic power short-term prediction method and system based on fusion of physical model and LSTM
By combining photovoltaic power station configuration information and historical meteorological data, a DC power prediction physical model and an LSTM model were established, which solved the problems of accuracy and stability in short-term photovoltaic power prediction, achieved more accurate photovoltaic power prediction, and improved the management efficiency of microgrids and integrated energy systems.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HENAN YUANWANG HECHU ELECTRIC RES INST CO LTD
- Filing Date
- 2024-12-02
- Publication Date
- 2026-06-05
AI Technical Summary
Existing short-term photovoltaic power prediction technologies suffer from low prediction accuracy, weak stability, and insufficient integration with physical models.
By combining photovoltaic power station configuration information, standard experimental parameters of photovoltaic modules, and historical meteorological data, a physical model for DC power prediction is established and fused with an LSTM model. Derivative features are generated by filtering the correlation between solar position models and meteorological data, and the LSTM model is trained to finally make short-term predictions of photovoltaic power.
It improves the accuracy and stability of short-term photovoltaic power forecasting, reduces the impact of system parameter deviations, provides more accurate forecasting capabilities, assists in energy management of microgrids and integrated energy systems, and enhances economic efficiency.
Smart Images
Figure CN122153265A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of photovoltaic power prediction technology, and in particular to a method and system for short-term photovoltaic power prediction based on the fusion of physical models and LSTM. Background Technology
[0002] As the proportion of photovoltaic power generation in the power grid structure continues to increase, and as scenarios such as integrated photovoltaic, energy storage, and charging microgrids develop, the ability to accurately predict the power generation of photovoltaic power plants becomes particularly important.
[0003] Photovoltaic power forecasting is generally divided into short-term power forecasting, ultra-short-term power forecasting, and minute-level power forecasting. Currently, due to limitations in application scenarios, data collection conditions, and the maturity of forecasting methods, we mainly consider short-term and ultra-short-term power forecasting functions. Short-term power forecasting refers to predicting the power generation of a photovoltaic power plant over the next 72 hours, typically at 15-minute intervals; ultra-short-term power forecasting, on the other hand, predicts the power generation over the next 4 hours in a rolling fashion, also at 15-minute intervals.
[0004] Numerous forecasting methods have been developed both domestically and internationally for photovoltaic (PV) power plant power prediction. These methods can be broadly categorized into two types: physical methods based on physical models and data-driven mathematical methods. Physical methods primarily rely on PV module performance parameters, plant system configuration information, and the physical mechanisms of PV power generation to establish physical formula models that estimate the plant's output power step-by-step. Mathematical modeling methods, on the other hand, use historical operating data of PV power plants and numerical weather prediction (NWP) data to perform mathematical modeling and prediction. Typical methods include linear regression, time series analysis, backpropagation (BP) neural network models, and LSTM networks.
[0005] Because the structure and steps of physical methods are relatively fixed, existing research has focused relatively little on improving them, generally concentrating on simulation studies of the computational process. While existing models combine elements of physical and statistical models, they do not deeply integrate the key components of the physical model and fail to explore the correlations of important features, resulting in low prediction accuracy and weak stability. Furthermore, the methods used to build the predictive mathematical models, such as K-Means clustering and SVM regression, also suffer from insufficient integration with the physical model. Summary of the Invention
[0006] (I) Purpose of the Invention
[0007] The purpose of this invention is to provide a method and system for short-term photovoltaic power prediction based on the fusion of physical models and LSTM, which deeply integrates physical models and LSTM models to solve the problems of low prediction accuracy, weak stability and insufficient integration of physical models in existing short-term photovoltaic power prediction technologies.
[0008] (II) Technical Solution
[0009] To address the aforementioned problems, a first aspect of this invention provides a method for short-term photovoltaic power prediction based on the fusion of a physical model and LSTM, comprising:
[0010] Based on the configuration information of photovoltaic power stations, standard experimental parameters of various types of photovoltaic modules, and historical meteorological data, a physical model for predicting the DC power of various types of photovoltaic modules in the photovoltaic power station is established.
[0011] Based on the historical power generation, historical meteorological data, and DC power prediction of the photovoltaic power station, an LSTM model is established; the DC power prediction is obtained through the DC power prediction physical model.
[0012] The DC power prediction physical model and the LSTM model are fused to obtain a short-term photovoltaic power prediction model;
[0013] The aforementioned photovoltaic power short-term prediction model is used for photovoltaic power short-term prediction.
[0014] Furthermore, the establishment of a physical model for predicting the DC power of various types of photovoltaic modules within the photovoltaic power station includes:
[0015] Establish solar position models for various types of photovoltaic modules;
[0016] Based on the solar position model, a physical model for predicting the DC power of each type of photovoltaic module is established.
[0017] Furthermore, the establishment of solar position models for various types of photovoltaic modules includes:
[0018]
[0019] Among them, G e G represents the effective irradiance that various types of photovoltaic modules can receive. dn G represents diffuse irradiance. dif G represents direct solar irradiance. t θ is the total solar irradiance; θ is the angle of incidence of sunlight relative to the tilted plane of the photovoltaic panel, that is, the angle between the direction of sunlight incidence and the normal of the tilted plane of the photovoltaic panel. θ changes continuously with the position of the sun; β is the tilt angle of various types of photovoltaic modules; ρ is the solar irradiance reflection coefficient.
[0020] Furthermore, the step of establishing a physical model for predicting the DC power of various types of photovoltaic modules based on the solar position model includes:
[0021] Based on the effective irradiance G that can be received by various types of photovoltaic modules eBased on the temperature and standard experimental parameters of each type of photovoltaic module, a physical model for predicting the DC power of each type of photovoltaic module is established; the physical model for predicting the DC power is expressed as follows:
[0022]
[0023] U MPP =U mref ln(e+bΔG)(1-cΔT)
[0024] P dc =U MPP I MPP
[0025] I MPP For the estimated output current of photovoltaic modules, U MPP For the estimated output voltage of the photovoltaic module, P dc For the estimated DC output power of photovoltaic modules, G ref I represents the solar irradiance under the stated standard experimental conditions. mref U represents the optimal output current under the stated standard experimental conditions. mref ΔT is the optimal output voltage under the standard experimental conditions; ΔT is the difference between the actual temperature of the photovoltaic panel and the temperature under the standard experimental conditions; ΔG is the difference between the actual solar irradiance and the solar irradiance under the standard experimental conditions; and a, b, and c are the respective compensation correction coefficients.
[0026] Furthermore, the LSTM model is established based on the historical power generation of the photovoltaic power station, historical meteorological data, and the estimated DC power, including:
[0027] Based on the historical power generation and meteorological data of the photovoltaic power station, the correlation between each meteorological indicator in the historical meteorological data and the power generation of the photovoltaic power station is calculated.
[0028] Meteorological indicators are screened based on the aforementioned correlation;
[0029] Derivative features are generated based on the screening results;
[0030] The LSTM model is trained by using the DC power prediction, the screened meteorological indicators and the derived features as inputs, and the photovoltaic power generation of the station as the output, to obtain a trained LSTM model.
[0031] Furthermore, the fusion of the DC power prediction physical model and the LSTM model to obtain the photovoltaic power short-term prediction model, and the use of the photovoltaic power short-term prediction model to predict photovoltaic power in the short term, includes: using the solar position model to obtain the effective irradiance that each type of photovoltaic module can receive;
[0032] Using the effective irradiance as input, the DC power of various types of photovoltaic modules is estimated using the DC power prediction physical model.
[0033] Using the DC power as input, the trained LSTM model is used to predict the short-term power generation of the photovoltaic power station.
[0034] Furthermore, the step of calculating the correlation between various meteorological indicators in the historical meteorological data and the power generation of the photovoltaic power station based on the historical power generation and meteorological data of the photovoltaic power station includes:
[0035] Acquire historical power generation and meteorological data of photovoltaic power stations;
[0036] Calculate the Pearson correlation coefficient between each meteorological indicator in the historical meteorological data and the historical power generation.
[0037] Furthermore, the screening of meteorological indicators based on the correlation includes:
[0038] The meteorological indicators are filtered based on the set threshold.
[0039] Generate a subset of features whose Pearson correlation coefficients fall within a set threshold range.
[0040] Furthermore, the generation of derived features based on the screening results includes:
[0041] The feature subset within the set threshold range is used to generate a sequence of equal steps as a list of powers. The solar irradiance is then transformed by the corresponding power calculation to generate the first derived feature.
[0042] Calculate the Pearson correlation coefficient between the derived features after power transformation and the historical power generation, and find the point with the highest correlation coefficient (p). ex ,R max ), with {p ex -δ, p ex p ex +δ} is a set of powers, and the corresponding power-transformed indicators are calculated to generate three second derived features;
[0043] Among them, R max p is the maximum correlation coefficient. ex δ is the power corresponding to the maximum correlation coefficient, where p is the power of p. ex The step size within a specific neighborhood.
[0044] According to another aspect of the present invention, a short-term photovoltaic power prediction system based on the fusion of a physical model and LSTM is provided, comprising:
[0045] The first model building module is used to build a physical model for predicting the DC power of various types of photovoltaic modules in the photovoltaic station based on the configuration information of the photovoltaic station, the standard experimental parameters of various types of photovoltaic modules and historical meteorological data.
[0046] The second model building module establishes an LSTM model based on the historical power generation, historical meteorological data, and DC power prediction of the photovoltaic power station; the DC power prediction is obtained through the DC power prediction physical model.
[0047] The model fusion module is used to fuse the DC power prediction physical model and the LSTM model to obtain a short-term photovoltaic power prediction model.
[0048] The power prediction module is used to perform short-term photovoltaic power prediction using the photovoltaic power short-term prediction model.
[0049] (III) Beneficial Effects
[0050] The above-described technical solution of the present invention has the following beneficial technical effects:
[0051] This invention provides a method and system for short-term photovoltaic (PV) power prediction based on the fusion of a physical model and an LSTM (Laser-Based Array Transformer) model. The method includes: establishing a physical model for DC power prediction of various types of PV modules within the PV site, based on PV site configuration information, standard experimental parameters of various PV modules, and historical meteorological data; establishing an LSTM model based on the historical power generation, historical meteorological data, and DC power prediction value of the PV site; obtaining the DC power prediction value through the DC power prediction physical model; fusing the DC power prediction physical model and the LSTM model to obtain a short-term PV power prediction model; and using the short-term PV power prediction model for short-term PV power prediction. The technical solution provided by this invention deeply integrates the physical model and the LSTM model, solving the problems of low prediction accuracy, weak stability, and insufficient integration of the physical model in existing short-term PV power prediction technologies. While fully considering the physical mechanism, it reduces the impact of system parameter deviations on prediction accuracy. In terms of prediction accuracy and stability, it shows a significant improvement compared to standalone physical models or LSTM network models, providing more stable and accurate short-term PV power prediction capabilities, assisting microgrids and integrated energy systems in energy management, better achieving energy conservation and carbon reduction, and improving economic efficiency. Attached Figure Description
[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0053] Figure 1 This is a flowchart of the photovoltaic power short-term prediction method based on the fusion of physical model and LSTM of the present invention.
[0054] Figure 2 This is a schematic diagram illustrating the correlation between the power conversion of irradiance and power generation of two different photovoltaic power stations using the short-term photovoltaic power prediction method provided in this embodiment of the invention.
[0055] Figure 3 This is a data input structure diagram of the LSTM model of the photovoltaic power short-term prediction method provided in this embodiment of the invention;
[0056] Figure 4 This is a flowchart of the short-term photovoltaic power prediction method provided in the embodiments of the present invention. Detailed Implementation
[0057] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.
[0058] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in one or more embodiments of the present invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the element or object listed following the word and its equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.
[0059] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings. Embodiments of the present invention provide a short-term photovoltaic power prediction method based on the fusion of a physical model and LSTM. Figure 1 Here is a flowchart of the method, such as Figure 1 As shown, the method includes:
[0060] Based on the photovoltaic (PV) station configuration information, standard experimental parameters of various types of PV modules, and historical meteorological data, a physical model for predicting the DC power of each type of PV module within the PV station is established. This establishment of the DC power prediction physical model includes: establishing a solar position model for each type of PV module, and based on the solar position model, establishing a physical model for predicting the DC power of each type of PV module. The details are as follows:
[0061]
[0062] Among them, G e G represents the effective irradiance that various types of photovoltaic modules can receive. dn G represents diffuse irradiance. dif G represents direct solar irradiance. t θ is the total solar irradiance; θ is the angle of incidence of sunlight relative to the tilted plane of the photovoltaic panel, that is, the angle between the direction of sunlight incidence and the normal of the tilted plane of the photovoltaic panel. θ changes continuously with the position of the sun; β is the tilt angle of various types of photovoltaic modules; ρ is the solar irradiance reflection coefficient.
[0063] Based on the effective irradiance G that can be received by various types of photovoltaic modules e Based on temperature and standard experimental parameters for various types of photovoltaic modules, a physical model for predicting the DC power of each type of photovoltaic module is established. The physical model for predicting the DC power is expressed as follows:
[0064]
[0065] U MPP =U mref ln(e+bΔG)(1-cΔT)
[0066] P dc =U MPP I MPP
[0067] Among them, I MPP For the estimated output current of photovoltaic modules, U MPP For the estimated output voltage of the photovoltaic module, P dc For the estimated DC output power of photovoltaic modules, G ref I represents the solar irradiance under standard experimental conditions. mref U is the optimal output current under standard experimental conditions. mref ΔT is the optimal output voltage under standard experimental conditions; ΔT is the difference between the actual temperature of the photovoltaic panel and the temperature under standard experimental conditions; ΔG is the difference between the actual solar irradiance and the solar irradiance under standard experimental conditions; and a, b, and c are the respective compensation correction coefficients.
[0068] Among them, the LSTM (Long Short-Term Memory) model is the Long Short-Term Memory network model.
[0069] Based on the historical power generation, historical meteorological data, and DC power prediction of the photovoltaic power station, an LSTM model is established, including the following steps:
[0070] Based on the historical power generation and meteorological data of photovoltaic power stations, the correlation between various meteorological indicators in the historical meteorological data and the power generation of photovoltaic power stations is calculated.
[0071] Meteorological indicators are screened based on correlation;
[0072] Derivative features are generated based on the screening results;
[0073] The LSTM model is trained by using DC power prediction, screened meteorological indicators and derived features as inputs, and photovoltaic power generation as output.
[0074] The specific implementation is as follows:
[0075] Historical power generation and historical meteorological data of photovoltaic power stations are obtained. The Pearson correlation coefficient between each meteorological indicator and historical power generation in the historical meteorological data is calculated. Based on the set threshold, each meteorological indicator is screened to generate a feature subset with Pearson correlation coefficient within the set threshold range. The feature subset within the set threshold range is generated into an equal step size sequence as a power list. The solar irradiance is transformed by the corresponding power calculation to generate the first derived feature.
[0076] Calculate the Pearson correlation coefficient between the derived characteristics after power transformation and historical power generation, and find the point with the highest correlation coefficient (p). ex ,R max ), with {p ex -δ, p ex p ex +δ} is a set of powers, and the corresponding power-transformed indicators are calculated to generate three second derived features;
[0077] Among them, R max p is the maximum correlation coefficient. ex δ is the power corresponding to the maximum correlation coefficient, where p is the power of p. ex The step size within a specific neighborhood.
[0078] The specific value of δ can be determined based on the correlation coefficient in R. max The rate of change in the vicinity is determined.
[0079] Before training the LSTM model, records during periods of no irradiance are filtered out. The DC power estimate, the filtered meteorological indicators, and the derived features are used as the input dataset. The input dataset is split into a training set and a test set. The training set is used to train the LSTM model, and the test set is used to test the performance of each trained model. The power generation of the photovoltaic power station is used as the output result. Based on the output result, the LSTM model parameters are tuned and the fitting is optimized to obtain the trained LSTM model.
[0080] The Pearson product-moment correlation coefficient is a statistic used to measure the linear correlation between two variables X and Y.
[0081] Figure 2 This diagram illustrates the correlation between the power conversion of irradiance and power generation for two different photovoltaic power plants. Using a grid search method with a power list of equal step sizes (0.1 to 3.5), the irradiance of the two different photovoltaic power plants is converted to the corresponding power values. The optimal power values are then determined based on the Pearson correlation coefficient between the derived characteristics after the power conversion and the power generation. The results show that the optimal power values for the two photovoltaic power plants are 0.3 and 2.4, respectively.
[0082] According to some optional embodiments, if the magnitude of the derived feature values after power transformation is too large, the min-max method or the stand-scaler method can be used to compress the numerical range.
[0083] The min-max method, or minimum-maximum normalization, standardizes data by linearly transforming the original data to a specified range. It normalizes data of different dimensions or ranges to facilitate further analysis and processing.
[0084] `stand-scaler` is a preprocessing function in the scikit-eamn library used to standardize data, setting the mean to 0 and the standard deviation to 1. This ensures that different features contribute equally to the model, leading to more accurate training and prediction.
[0085] Figure 3 This is a diagram of the data input structure for an LSTM model, where seq_len = 3, indicating that the feature sequence length used in a single training iteration is 3. Each layer of this LSTM model has 3 neurons. The feature vector X at time t... t Then by P t W t W t+1 GF t GFt+1 It consists of five parts, of which P t The estimated DC power of various types of photovoltaic modules at time t is W. t Let W be the meteorological forecast index data at time t. t+1 For meteorological forecast index data at time t+1, GF t GF is the derived feature generated by the power transformation of irradiance at time t. t+1 The derived feature is generated by power transformation of irradiance at time t+1.
[0086] Among them, the meteorological forecast index data is the NWP forecast data.
[0087] By fusing the DC power prediction physical model and the LSTM model, a short-term photovoltaic power prediction model is obtained.
[0088] Figure 4 The process of using a short-term photovoltaic power forecasting model to predict short-term photovoltaic power is shown, including:
[0089] Using a solar position model, the effective irradiance that each type of photovoltaic module can receive is obtained;
[0090] Using effective irradiance as input, the DC power of various types of photovoltaic modules is estimated using a DC power prediction physical model.
[0091] Using DC power as input, a trained LSTM model is used to predict the short-term power generation of the photovoltaic power station.
[0092] The specific implementation is as follows:
[0093] Using meteorological forecast data and photovoltaic station configuration information as inputs, the effective irradiance that each type of photovoltaic module can receive is obtained using the solar position model.
[0094] Using meteorological forecast data, the effective irradiance that each type of photovoltaic module can receive, and the standard experimental parameters of each type of photovoltaic module as inputs, the DC power prediction value of each type of photovoltaic module is obtained by using a DC power prediction physical model.
[0095] Using meteorological forecast data, estimated DC power of various types of photovoltaic modules, and standard experimental parameters as inputs, the estimated short-term power generation of the photovoltaic power station is obtained through a trained LSTM model.
[0096] An embodiment of the present invention also provides a short-term photovoltaic power prediction system based on the fusion of a physical model and LSTM, the system comprising:
[0097] The first model building module is used to build a physical model for predicting the DC power of various types of photovoltaic modules in the photovoltaic station based on the configuration information of the photovoltaic station, the standard experimental parameters of various types of photovoltaic modules and historical meteorological data.
[0098] The second model building module establishes an LSTM model based on the historical power generation, historical meteorological data, and DC power prediction of the photovoltaic power station; the DC power prediction is obtained through the DC power prediction physical model.
[0099] The model fusion module is used to fuse the DC power prediction physical model and the LSTM model to obtain a short-term photovoltaic power prediction model.
[0100] The power prediction module is used to make short-term predictions of photovoltaic power using a short-term photovoltaic power prediction model.
[0101] In one embodiment,
[0102] The prediction process of the short-term photovoltaic power prediction model can be set to be fully automated and executed on a timed basis. It can be embedded into the back-end of the photovoltaic power plant management system, the photovoltaic power plant monitoring system, the microgrid EMS system, or the integrated energy management system.
[0103] In summary, the present invention relates to a method and system for short-term photovoltaic power prediction based on the fusion of a physical model and an LSTM model. The method includes: establishing a physical model for DC power prediction of various types of photovoltaic modules within the photovoltaic power station based on photovoltaic station configuration information, standard experimental parameters of various types of photovoltaic modules, and historical meteorological data; establishing an LSTM model based on the historical power generation, historical meteorological data, and DC power prediction value of the photovoltaic power station; obtaining the DC power prediction value through the DC power prediction physical model; fusing the DC power prediction physical model and the LSTM model to obtain a short-term photovoltaic power prediction model; and using the short-term photovoltaic power prediction model to perform short-term photovoltaic power prediction. The technical solution provided by the present invention deeply integrates the physical model and the LSTM model, solving the problems of low prediction accuracy, weak stability, and insufficient integration of the physical model in existing short-term photovoltaic power prediction technologies. While fully considering the physical mechanism, it reduces the impact of system parameter deviations on prediction accuracy. In terms of prediction accuracy and stability, it shows a significant improvement compared to standalone physical models or LSTM network models, providing more stable and accurate short-term photovoltaic power prediction capabilities. This helps the grid better absorb photovoltaic power, thereby reducing economic losses to photovoltaic owners due to power curtailment and increasing the return on investment of photovoltaic power plants. It can also help microgrids and integrated energy systems to better manage energy, achieve energy conservation and carbon reduction, and improve economic efficiency.
[0104] It should be understood that the specific embodiments described above are merely illustrative or explanatory of the principles of the invention and do not constitute a limitation thereof. Therefore, any modifications, equivalent substitutions, improvements, etc., made without departing from the spirit and scope of the invention should be included within the protection scope of the invention. Furthermore, the appended claims are intended to cover all variations and modifications falling within the scope and boundaries of the appended claims, or equivalent forms of such scope and boundaries.
Claims
1. A method for short-term photovoltaic power prediction based on the fusion of a physical model and LSTM, characterized in that, The method includes: Based on the configuration information of photovoltaic power stations, standard experimental parameters of various types of photovoltaic modules, and historical meteorological data, a physical model for predicting the DC power of various types of photovoltaic modules in the photovoltaic power station is established. Based on the historical power generation, historical meteorological data, and DC power prediction of the photovoltaic power station, an LSTM model is established; the DC power prediction is obtained through the DC power prediction physical model. The DC power prediction physical model and the LSTM model are fused to obtain a short-term photovoltaic power prediction model; The aforementioned photovoltaic power short-term prediction model is used for photovoltaic power short-term prediction.
2. The method according to claim 1, characterized in that, The establishment of a physical model for predicting the DC power of various types of photovoltaic modules within the photovoltaic power station includes: Establish solar position models for various types of photovoltaic modules; Based on the solar position model, a physical model for predicting the DC power of each type of photovoltaic module is established.
3. The method according to claim 2, characterized in that, The establishment of solar position models for various types of photovoltaic modules includes: Among them, G e G represents the effective irradiance that various types of photovoltaic modules can receive. dn G represents diffuse irradiance. dif G represents direct solar irradiance. t θ is the total solar irradiance; θ is the angle of incidence of sunlight relative to the tilted plane of the photovoltaic panel, that is, the angle between the direction of sunlight incidence and the normal of the tilted plane of the photovoltaic panel. θ changes continuously with the position of the sun; β is the tilt angle of various types of photovoltaic modules; ρ is the solar irradiance reflection coefficient.
4. The method according to claim 2, characterized in that, The step of establishing a physical model for predicting the DC power of various types of photovoltaic modules based on the solar position model includes: Based on the effective irradiance G that can be received by various types of photovoltaic modules e Based on the temperature and standard experimental parameters of each type of photovoltaic module, a physical model for predicting the DC power of each type of photovoltaic module is established; the physical model for predicting the DC power is expressed as follows: The MPP =U mref ln(e+bΔG)(1-cΔT) P dc =U MPP AND MPP I MPP For the estimated output current of photovoltaic modules, U MPP For the estimated output voltage of the photovoltaic module, P dc For the estimated DC output power of photovoltaic modules, G ref I represents the solar irradiance under the stated standard experimental conditions. mref U represents the optimal output current under the stated standard experimental conditions. mref ΔT is the optimal output voltage under the standard experimental conditions; ΔT is the difference between the actual temperature of the photovoltaic panel and the temperature under the standard experimental conditions; ΔG is the difference between the actual solar irradiance and the solar irradiance under the standard experimental conditions; and a, b, and c are the respective compensation correction coefficients.
5. The method according to claim 1, characterized in that, The LSTM model is established based on the historical power generation, historical meteorological data, and the estimated DC power of the photovoltaic power station, including: Based on the historical power generation and meteorological data of the photovoltaic power station, the correlation between each meteorological indicator in the historical meteorological data and the power generation of the photovoltaic power station is calculated. Meteorological indicators are screened based on the aforementioned correlation; Derivative features are generated based on the screening results; The LSTM model is trained by using the DC power prediction, the screened meteorological indicators and the derived features as inputs, and the photovoltaic power generation of the station as the output, to obtain a trained LSTM model.
6. The method according to claim 1, characterized in that, The DC power prediction physical model and the LSTM model are fused to obtain a short-term photovoltaic power prediction model. The short-term photovoltaic power prediction model is used to predict photovoltaic power in the short term, including: using the solar position model to obtain the effective irradiance that each type of photovoltaic module can receive. Using the effective irradiance as input, the DC power of various types of photovoltaic modules is estimated using the DC power prediction physical model. Using the DC power as input, the trained LSTM model is used to predict the short-term power generation of the photovoltaic power station.
7. The method according to claim 5, characterized in that, The step of calculating the correlation between various meteorological indicators in the historical meteorological data and the power generation of the photovoltaic power station based on the historical power generation and meteorological data of the photovoltaic power station includes: Acquire historical power generation and meteorological data of photovoltaic power stations; Calculate the Pearson correlation coefficient between each meteorological indicator in the historical meteorological data and the historical power generation.
8. The method according to claim 7, characterized in that, The screening of meteorological indicators based on the correlation includes: The meteorological indicators are filtered based on the set threshold. Generate a subset of features whose Pearson correlation coefficients fall within a set threshold range.
9. The method according to claim 8, characterized in that, The generation of derived features based on the screening results includes: The feature subset within the set threshold range is used to generate a sequence of equal steps as a list of powers. The solar irradiance is then transformed by the corresponding power calculation to generate the first derived feature. Calculate the Pearson correlation coefficient between the derived features after power transformation and the historical power generation, and find the point with the highest correlation coefficient (p). ex ,R max ), with {p ex -δ, p ex p ex +δ} is a set of powers, and the corresponding power-transformed indicators are calculated to generate three second derived features; Among them, R max p is the maximum correlation coefficient. ex δ is the power corresponding to the maximum correlation coefficient, where p is the power of p. ex The step size within a specific neighborhood.
10. A short-term photovoltaic power prediction system based on the fusion of a physical model and LSTM, characterized in that, The system includes: The first model building module is used to build a physical model for predicting the DC power of various types of photovoltaic modules in the photovoltaic station based on the configuration information of the photovoltaic station, the standard experimental parameters of various types of photovoltaic modules and historical meteorological data. The second model building module establishes an LSTM model based on the historical power generation, historical meteorological data, and DC power prediction of the photovoltaic power station; the DC power prediction is obtained through the DC power prediction physical model. The model fusion module is used to fuse the DC power prediction physical model and the LSTM model to obtain a short-term photovoltaic power prediction model. The power prediction module is used to perform short-term photovoltaic power prediction using the photovoltaic power short-term prediction model.