A method and apparatus for short-term prediction of photovoltaic power generation based on satellite inversion data

Abstract

The application discloses a photovoltaic power generation short-time prediction method and device based on satellite inversion data, and belongs to the photovoltaic power generation prediction technical field.The method comprises the following steps: step S1, establishing a photovoltaic power generation prediction data set; step S2, preprocessing the photovoltaic power generation prediction data set; step S3, constructing a photovoltaic power generation short-time prediction hybrid model; step S4, training the photovoltaic power generation short-time prediction hybrid model based on the photovoltaic power generation prediction data set, and iterating until the loss function converges; step S5, acquiring satellite inversion data, meteorological data and photovoltaic power generation historical data of a target photovoltaic power station; and step S6, obtaining a photovoltaic power generation short-time prediction result of the target photovoltaic power station according to the trained photovoltaic power generation short-time prediction hybrid model.The application can fully utilize the relevant information of satellite inversion data, avoid the overfitting defects of a single model, and improve the accuracy of the photovoltaic power generation prediction result.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic power generation prediction technology, and in particular relates to a method and apparatus for short-term prediction of photovoltaic power generation based on satellite inversion data. Background Technology

[0002] With the widespread application of renewable energy, photovoltaic (PV) power generation, as a crucial component, presents challenges to the stable operation of the power grid due to its intermittent and unpredictable nature. PV power generation is significantly affected by weather, season, and time, leading to fluctuations in power supply. This necessitates accurate forecasting and management by grid operators to ensure grid stability and power continuity. Current research primarily focuses on short-term and long-duration PV power generation forecasting, using predictions at different time scales to help grid operators respond quickly to fluctuations in PV power generation.

[0003] Photovoltaic power generation prediction methods can be divided into physical methods and machine learning methods. Physical methods, based on the physical characteristics of photovoltaic cells and the photoelectric conversion principle, predict photovoltaic power generation by simulating the transfer of solar irradiance and the physical processes of photovoltaic module power generation. These methods typically require accurate meteorological data, detailed geographical information, and photovoltaic system parameters such as temperature, solar radiation, and solar radiation distribution. The accuracy of this prediction method largely depends on the accuracy of the meteorological data, leading to difficulties in practical deployment. Machine learning methods primarily utilize artificial intelligence algorithms. Currently, methods such as Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and combinations of CNN and LSTM are used for photovoltaic power generation prediction. However, increasing literature indicates that different meteorological data, model algorithms, and hyperparameter settings can lead to different prediction results, meaning the accuracy of model predictions needs improvement. For example, LSTM alone cannot fully utilize spatial features such as geographical location and terrain, which significantly affect photovoltaic power generation efficiency. CNN alone typically lacks the ability to process time-series data, while photovoltaic power generation data has a significant time dependency; using CNN alone may not be effective for prediction. While the method combining CNN and LSTM achieves simultaneous consideration of spatial and temporal features, actual data shows that the predicted values ​​output by this method fluctuate less and have larger errors compared to the actual values. This is because the input data only considers terrain effects, atmospheric effects, and sky conditions, without making full use of the relevant parameters of photovoltaic power generation.

[0004] Therefore, there is an urgent need for a new method and device for short-term prediction of photovoltaic power generation. This method and device not only considers both spatial and temporal characteristics to avoid overfitting of a single model, but also fully explores the data characteristics of photovoltaic power generation, thereby achieving accurate prediction of photovoltaic power generation. Summary of the Invention

[0005] To address the shortcomings of the existing technology, this invention provides a short-term prediction method for photovoltaic power generation based on satellite inversion data, characterized in that the method includes:

[0006] Step S1: Establish a photovoltaic power generation prediction dataset, which includes satellite inversion data, meteorological data, and historical photovoltaic power generation data;

[0007] Step S2: Preprocess the photovoltaic power generation prediction dataset;

[0008] Step S3: Construct a hybrid model for short-term prediction of photovoltaic power generation. The hybrid model for short-term prediction of photovoltaic power generation includes a convolutional neural network-long short-term memory network (CNN-LSTM) module, an LSTM module, and a feature fusion module. The output of the CNN-LSTM module and the output of the LSTM module are both connected to the feature fusion module.

[0009] The CNN-LSTM module processes the preprocessed satellite inversion data to obtain a first short-term prediction result for photovoltaic power generation; the LSTM module processes the preprocessed meteorological data and historical photovoltaic power generation data to obtain a second short-term prediction result for photovoltaic power generation; and the feature fusion module fuses the first and second short-term prediction results for photovoltaic power generation to obtain the final prediction result.

[0010] Step S4: Train the photovoltaic power generation short-term prediction hybrid model based on the photovoltaic power generation prediction dataset, iterating until the loss function converges;

[0011] Step S5: Obtain satellite inversion data, meteorological data, and historical photovoltaic power generation data of the target photovoltaic power station;

[0012] Step S6: After preprocessing the satellite inversion data, meteorological data and historical photovoltaic power generation data of the target photovoltaic power station, input them into the trained photovoltaic power generation short-term prediction hybrid model to obtain the photovoltaic power generation short-term prediction result of the target photovoltaic power station.

[0013] Furthermore, the preprocessing includes missing value handling, data normalization, enhanced Dickie-Fuller test, and autocorrelation test.

[0014] Furthermore, the enhanced Dickie-Fuller test specifically includes:

[0015] S21: Test the photovoltaic power generation prediction dataset y according to the following formula. t unit root:

[0016]

[0017] Where, Δy t The photovoltaic power generation prediction dataset y t The first difference; α is the parameter to be tested. If α = 0, it indicates that the photovoltaic power generation prediction dataset y t It has a unit root and is non-stationary; Here, p is the lag coefficient, p is the quantity of the lag term, and βt is the time trend term. t It is an error term;

[0018] S22: Estimate α, β, and β in the above equation using the least squares method.

[0019] S23: Calculate the statistic of the enhanced Dickie-Fuller test, as shown in the following formula:

[0020]

[0021] in, This is an estimate of α. yes The standard error;

[0022] S24: Calculate the above statistic ADF Sta Compared with the critical value, if ADF Sta If the value is less than 0.05, then the photovoltaic power generation prediction dataset y t For stationary data; if ADF Sta If the value is greater than 0.05, then the photovoltaic power generation prediction dataset y t The data is non-stationary.

[0023] S25: For non-stationary data, the difference method is used to eliminate the photovoltaic power generation prediction dataset y. t The time series dependency is determined, and then the process proceeds to step S21 to continue calculating the ADF. Sta This continues until the aforementioned dataset is confirmed to be stationary.

[0024] Furthermore, after performing the enhanced Dickie-Fuller test, an autocorrelation test is then performed, specifically as follows:

[0025] S26: Perform the autocorrelation test according to the following formula:

[0026]

[0027] Where h is the lag time, cov is the covariance calculation, and var(y) t ) is the photovoltaic power generation prediction dataset y t Variance in a time series;

[0028] S27: Determine whether the above calculation result ACF(h) exceeds the preset confidence interval threshold. If it exceeds the preset confidence interval threshold, consider the photovoltaic power generation prediction dataset y to be... t It meets the correlation requirements of time series.

[0029] Furthermore, the CNN-LSTM module first extracts the spatial features of the satellite inversion data through the CNN layer, and maps the obtained feature map to the LSTM layer. The LSTM layer is used to extract time information for time series prediction. Finally, the first result of the short-term prediction of photovoltaic power generation is obtained through the fully connected layer.

[0030] Furthermore, the specific process by which the CNN-LSTM module obtains the first result of the short-term prediction of photovoltaic power generation is as follows:

[0031] Step S311: The CNN layer processes the satellite inversion data through convolution and activation functions:

[0032] O CNN =ReLU(W CNN *X+b CNN );

[0033] Where X represents the satellite inversion data, and W... CNN and b CNN These are the weights and biases of the CNN layer, * indicates a convolution operation, and ReLU indicates a modified linear activation unit function.

[0034] Step S312: Reduce the feature dimension using pooling operations, i.e.:

[0035] P = Pooling(O CNN );

[0036] Pooling represents pooling operations;

[0037] Step S313: Use the output P of the CNN layer as the input of the LSTM layer. The dimension of the output P is D×T, where D is the feature depth and T is the time series step size. The LSTM layer processes the data and outputs the hidden state h at each time step. t :

[0038] h t =LSTM(P, h)t-1 );

[0039] Among them, h t It represents the hidden state at time step t, LSTM represents the operation of the LSTM unit, and h t-1 It is the hidden state of the previous time step;

[0040] Step S314: Set the hidden state h of the last time step of the LSTM layer. T The final output is obtained by processing through a fully connected layer. Specifically:

[0041] Z = W FC H+b FC ;

[0042]

[0043] Among them, W FC and b FC These are the bias and weights of the fully connected layer, respectively, and softmax is the activation function.

[0044] Furthermore, after receiving the meteorological data and the historical photovoltaic power generation data, the LSTM module first combines the meteorological data and the historical photovoltaic power generation data using timestamp information, and then performs short-term photovoltaic power generation prediction on the combined data to obtain the second result of the short-term photovoltaic power generation prediction; wherein, the meteorological data includes radiation, DHI, DNI, GHI, relative humidity, wind direction, dew point, pressure, temperature and solar zenith angle, and the historical photovoltaic power generation data includes historical photovoltaic power generation.

[0045] The present invention also provides a short-term photovoltaic power generation prediction device based on satellite inversion data, characterized in that the device comprises:

[0046] A dataset construction module is used to build a photovoltaic power generation prediction dataset, which includes satellite inversion data, meteorological data, and historical photovoltaic power generation data.

[0047] A preprocessing module is used to preprocess the photovoltaic power generation prediction dataset;

[0048] The model building module is used to build a hybrid model for short-term prediction of photovoltaic power generation. The hybrid model for short-term prediction of photovoltaic power generation includes a CNN-LSTM module, an LSTM module and a feature fusion module. The outputs of the CNN-LSTM module and the LSTM module are both connected to the feature fusion module.

[0049] The CNN-LSTM module processes the preprocessed satellite inversion data to obtain a first short-term prediction result for photovoltaic power generation; the LSTM module processes the preprocessed meteorological data and historical photovoltaic power generation data to obtain a second short-term prediction result for photovoltaic power generation; and the feature fusion module fuses the first and second short-term prediction results for photovoltaic power generation to obtain the final prediction result.

[0050] The training module is used to train the photovoltaic power generation short-term prediction hybrid model based on the photovoltaic power generation prediction dataset, iterating until the loss function converges;

[0051] The acquisition module is used to acquire satellite inversion data, meteorological data, and historical photovoltaic power generation data of the target photovoltaic power station;

[0052] The prediction module is used to preprocess the satellite inversion data, meteorological data and historical photovoltaic power generation data of the target photovoltaic power station, and then input them into the trained photovoltaic power generation short-term prediction hybrid model to obtain the photovoltaic power generation short-term prediction result of the target photovoltaic power station.

[0053] The present invention also provides a computer-readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements the aforementioned short-term prediction method for photovoltaic power generation based on satellite inversion data.

[0054] The present invention also provides an electronic device, the electronic device including a memory and a processor, characterized in that the memory stores a computer program, and the processor is used to execute the computer program to implement the aforementioned method for short-term prediction of photovoltaic power generation based on satellite inversion data.

[0055] This invention first utilizes a CNN-LSTM module to process satellite inversion data, fully leveraging its spatial and temporal series characteristics to obtain a first result for short-term photovoltaic (PV) power generation prediction. Simultaneously, an LSTM module is used to process meteorological data and historical PV power generation data, fully utilizing their temporal series characteristics to obtain a second result for short-term PV power generation prediction. Then, a fully connected layer fuses the first and second results to obtain the final short-term PV power generation prediction result. This approach fully utilizes the relevant information from the satellite inversion data while avoiding the overfitting defects of a single model, and strengthens the weight of the temporal series characteristics of meteorological data and historical PV power generation data in the PV power generation prediction process, thereby improving the accuracy of the PV power generation prediction results. Attached Figure Description

[0056] The above and other objects, features, and advantages of exemplary embodiments of the present disclosure will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the present disclosure are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein:

[0057] Figure 1 This is a flowchart illustrating a short-term prediction method for photovoltaic power generation based on satellite inversion data according to an embodiment of the present invention.

[0058] Figure 2 This illustrates a hybrid model architecture for short-term photovoltaic power generation prediction according to an embodiment of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0060] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “said,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0061] It should be understood that although the terms first, second, third, etc., may be used to describe... in the embodiments of the present invention, these... should not be limited to these terms. These terms are only used to distinguish... For example, first... may also be referred to as second... without departing from the scope of the embodiments of the present invention, and similarly, second... may also be referred to as first...

[0062] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0063] Depending on the context, the words “if” or “suppose” as used here can be interpreted as “when”, “when”, “in response to determination”, or “in response to detection”. Similarly, depending on the context, the phrases “if determination” or “if detection (of the stated condition or event)” can be interpreted as “when determination”, “in response to determination”, “when detection (of the stated condition or event)”, or “in response to detection (of the stated condition or event)”.

[0064] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device comprising a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device that includes said element.

[0065] To achieve short-time power prediction of photovoltaic power generation, such as Figure 1 As shown in the figure, an embodiment of the present invention discloses a short-term prediction method for photovoltaic power generation based on satellite inversion data, the method comprising:

[0066] Step S1: Establish a photovoltaic power generation prediction dataset, which includes satellite inversion data, meteorological data, and historical photovoltaic power generation data.

[0067] Satellite inversion data typically refers to data on various physical parameters of the Earth's surface or atmosphere acquired through satellite remote sensing technology. This data is derived by analyzing electromagnetic wave information received by satellite sensors to invert characteristics of the Earth's surface or atmosphere, such as temperature, humidity, vegetation index, and atmospheric component concentration. In photovoltaic power generation prediction, satellite inversion data can provide high spatiotemporal resolution solar radiation data, helping to more accurately simulate the output of photovoltaic systems.

[0068] Meteorological data includes various parameters such as solar radiation (W / m²), temperature (°C), dew point (°C), relative humidity (%), pressure (mBar), and solar zenith angle (degrees), all of which are related to photovoltaic power generation. In this embodiment of the invention, the Pearson correlation coefficient is used to determine whether the above meteorological data are significantly correlated with photovoltaic power generation. According to the inventors' research, factors with significant correlations include: radiation, GHI, DHI, DNI, relative humidity, wind direction, dew point, pressure, temperature, and solar zenith angle. Furthermore, although solar zenith angle shows a negative correlation, this factor still exhibits relatively good linear correlation. DHI (Direct Horizontal Irradiance) represents the amount of direct solar radiation on a horizontal plane, i.e., radiation directly from the sun that is not scattered by the atmosphere. DNI (Direct Normal Irradiance) represents the amount of direct solar radiation on a vertical plane, typically used in concentrating photovoltaic systems that track the sun. GHI (Global Horizontal Irradiance) represents the total solar radiation on a horizontal plane, including direct and diffuse radiation, but excluding reflected radiation.

[0069] Historical photovoltaic power generation data includes historical power generation data, photovoltaic panel operating status and maintenance records, etc. The above historical data can reflect the power generation trend of photovoltaic system performance.

[0070] Step S2: Preprocess the photovoltaic power generation prediction dataset.

[0071] As described above, the embodiments of this invention involve using satellite inversion data, meteorological data, and historical photovoltaic power generation data from different sources for short-term photovoltaic power generation forecasting. Even small changes in the input data can lead to significant changes in the forecast output, affecting the accuracy of the short-term photovoltaic power generation forecast. Therefore, it is necessary to preprocess the photovoltaic power generation forecast dataset to improve the accuracy of the short-term photovoltaic power generation forecast.

[0072] In this embodiment of the invention, the preprocessing includes missing value handling, data normalization, enhanced Dickie-Fuller test, and autocorrelation test. The satellite inversion data, meteorological data, and historical photovoltaic power generation data may all contain missing data. To avoid poor learning performance of the subsequent short-term photovoltaic power generation prediction hybrid model, interpolation methods known in the art are used to fill in the missing values. Simultaneously, since the input data have different scales, the original data needs to be transformed using a specific ratio. In this embodiment of the invention, the Min-Max normalization method can be used for the above transformation. The above method is a commonly used method for data normalization in the art and will not be elaborated further here.

[0073] In the hybrid photovoltaic power generation short-term prediction model of this invention, both the CNN-LSTM module and the LSTM module involve the application of time series models. Therefore, during data preprocessing, it is necessary to first confirm whether the data is stationary. Stationary data means that the mean and variance of the exponent series do not differ significantly over time. If the data is not stationary, the prediction model will estimate inaccurate results. In this embodiment of the invention, an enhanced Dickie-Fuller test is used to confirm the photovoltaic power generation prediction dataset y. t Is the data stationary? If it is non-stationary, the difference method is used to eliminate the time series dependency and transform it into stationary data.

[0074] In this embodiment of the invention, the enhanced Dickie-Fuller test specifically includes:

[0075] S21: Test the photovoltaic power generation prediction dataset y according to the following formula. t unit root:

[0076]

[0077] Where, Δy t The photovoltaic power generation prediction dataset y t The first difference; α is the parameter to be tested. If α = 0, it indicates that the photovoltaic power generation prediction dataset y t It has a unit root and is non-stationary; Here, p is the lag coefficient, p is the quantity of the lag term, and βt is the time trend term. t It is an error term;

[0078] S22: Estimate α, β, and β in the above equation using the least squares method.

[0079] S23: Calculate the statistic of the enhanced Dickie-Fuller test, as shown in the following formula:

[0080]

[0081] in, This is an estimate of α. yes The standard error;

[0082] S24: Calculate the above statistic ADF Sta Compared with the critical value, if ADF Sta If the value is less than 0.05, then the photovoltaic power generation prediction dataset y t For stationary data; if ADF Sta If the value is greater than 0.05, then the photovoltaic power generation prediction dataset y t The data is non-stationary.

[0083] S25: For non-stationary data, the difference method is used to eliminate the photovoltaic power generation prediction dataset y. t The time series dependency is determined, and then the process proceeds to step S21 to continue calculating the ADF. Sta This continues until the aforementioned dataset is confirmed to be stationary.

[0084] After confirming the photovoltaic power generation prediction dataset y t After stabilizing the data, it is also necessary to determine the predictor factors for the hybrid model of short-term photovoltaic power generation prediction. That is, after performing the enhanced Dickie-Fuller test, an autocorrelation test is then conducted, specifically:

[0085] S26: Perform the autocorrelation test according to the following formula:

[0086]

[0087] Where h is the lag time, cov is the covariance calculation, and var(y) t ) is the photovoltaic power generation prediction dataset y t Variance in a time series;

[0088] S27: Determine whether the above calculation result ACF(h) exceeds the preset confidence interval threshold. If it exceeds the preset confidence interval threshold, consider the photovoltaic power generation prediction dataset y to be... t It meets the correlation requirements of time series.

[0089] In this embodiment of the invention, y is obtained through the above formula. t and y t-1 The correlation is 0.9, very close to 1, therefore, y is chosen. t-1 As y t Predictors.

[0090] Step S3: Construct a hybrid model for short-term prediction of photovoltaic power generation. The hybrid model for short-term prediction of photovoltaic power generation includes a CNN-LSTM module, an LSTM module, and a feature fusion module. The outputs of the CNN-LSTM module and the LSTM module are both connected to the feature fusion module.

[0091] The CNN-LSTM module is used to process the preprocessed satellite inversion data to obtain a first short-term prediction result for photovoltaic power generation based on the satellite inversion data; the LSTM module is used to process the preprocessed meteorological data and the historical photovoltaic power generation data to obtain a second short-term prediction result for photovoltaic power generation based on the meteorological data and the historical photovoltaic power generation data; the feature fusion module is used to fuse the first short-term prediction result for photovoltaic power generation and the second short-term prediction result for photovoltaic power generation to obtain the final prediction result.

[0092] In this embodiment of the invention, the hybrid model for short-term photovoltaic power generation prediction is divided into two branches to handle different data types. Finally, the outputs of each branch are fused to obtain the final photovoltaic power generation prediction. The main reason for this is that this approach avoids the defects such as overfitting that may occur when using a single model. At the same time, designing a separate LSTM module can increase the influence weight of meteorological data and historical photovoltaic power generation data during the model prediction process, thereby improving the accuracy and stability of the model prediction.

[0093] For machine learning models, the setting of hyperparameters is crucial because they affect the performance of the training algorithm and prevent overfitting. Therefore, in this embodiment, grid search is used to optimize both the CNN-LSTM and LSTM modules to find the optimal hyperparameters. Although the grid search process is time-consuming, the training and testing time is significantly reduced after determining the ideal hyperparameters. In this embodiment, the final hyperparameters of the CNN-LSTM and LSTM modules are as follows:

[0094] CNN-LSTM:

[0095]

[0096] LSTM:

[0097]

[0098] In the hybrid model for short-term photovoltaic power generation prediction in this embodiment of the invention, the CNN-LSTM module first extracts the spatial features of the satellite inversion data through the CNN layer, and maps the obtained feature map to the LSTM layer. The LSTM layer is used to extract time information for time series prediction. Finally, the first result of the short-term photovoltaic power generation prediction is obtained through the fully connected layer.

[0099] Specifically, the process by which the CNN-LSTM module obtains the first short-time prediction result of photovoltaic power generation is as follows:

[0100] Step S311: The CNN layer processes the satellite inversion data through convolution and activation functions:

[0101] O CNN =ReLU(W CNN *X+b CNN );

[0102] Where X represents the satellite inversion data, and W... CNN and b CNN These are the weights and biases of the CNN layer, * indicates a convolution operation, and ReLU indicates a modified linear activation unit function.

[0103] Step S312: Reduce the feature dimension using pooling operations, i.e.:

[0104] P = Pooling(O CNN );

[0105] Pooling represents pooling operations;

[0106] Step S313: Use the output P of the CNN layer as the input of the LSTM layer. The dimension of the output P is D×T, where D is the feature depth and T is the time series step size. The LSTM layer processes the data and outputs the hidden state h at each time step. t :

[0107] h t =LSTM(P, h) t-1 );

[0108] Among them, h t It represents the hidden state at time step t, LSTM represents the operation of the LSTM unit, and h t-1 It is the hidden state of the previous time step;

[0109] Step S314: Set the hidden state h of the last time step of the LSTM layer. T The final output is obtained by processing through a fully connected layer. Specifically:

[0110] Z = W FC H+b FC ;

[0111]

[0112] Among them, W FC and b FC These are the bias and weights of the fully connected layer, respectively, and softmax is the activation function.

[0113] In this embodiment of the invention, after receiving the meteorological data and the historical photovoltaic power generation data, the LSTM module first combines the meteorological data and the historical photovoltaic power generation data using timestamp information, and then performs short-term photovoltaic power generation prediction on the combined data to obtain the second result of the short-term photovoltaic power generation prediction; wherein, the meteorological data includes radiation, DHI, DNI, GHI, relative humidity, wind direction, dew point, pressure, temperature and solar zenith angle, and the historical photovoltaic power generation data includes historical photovoltaic power generation.

[0114] In one possible implementation, after determining the correspondence between meteorological data and historical photovoltaic power generation data using timestamps, as a simple method, the historical photovoltaic power generation data can be directly appended to the end of the meteorological data to achieve the combination of the two.

[0115] After obtaining the first result of the short-term prediction of photovoltaic power generation using the CNN-LSTM module and the second result of the short-term prediction of photovoltaic power generation using the LSTM module, the first result and the second result are input into the feature fusion module of the hybrid model of short-term prediction of photovoltaic power generation to obtain the final prediction result.

[0116] In this feature fusion module, feature fusion is performed first:

[0117]

[0118] in, This is the output of the CNN-LSTM module. This is the output of the LSTM module; Concat indicates the feature fusion operation.

[0119] Then, the fully connected layer of this feature fusion module maps the fused features to the final prediction result:

[0120]

[0121] in, It is the predicted output, W f and b f These are the weights and biases of the fully connected layer, respectively.

[0122] Step S4: Train the photovoltaic power generation short-term prediction hybrid model based on the photovoltaic power generation prediction dataset, iterating until the loss function converges;

[0123] Step S5: Obtain satellite inversion data, meteorological data, and historical photovoltaic power generation data of the target photovoltaic power station;

[0124] Step S6: After preprocessing the satellite inversion data, meteorological data and historical photovoltaic power generation data of the target photovoltaic power station, input them into the trained photovoltaic power generation short-term prediction hybrid model to obtain the photovoltaic power generation short-term prediction result of the target photovoltaic power station.

[0125] After obtaining accurate short-term forecasts of photovoltaic power generation, power operators can use this information to conduct power dispatch and resource optimization, ensure the stable operation of the power grid, and further improve the grid's ability to absorb photovoltaic energy.

[0126] like Figure 2 As shown, according to another embodiment of the present invention, the present invention also provides a short-term photovoltaic power generation prediction device based on satellite inversion data, the device comprising:

[0127] A dataset construction module is used to build a photovoltaic power generation prediction dataset, which includes satellite inversion data, meteorological data, and historical photovoltaic power generation data.

[0128] A preprocessing module is used to preprocess the photovoltaic power generation prediction dataset;

[0129] The model building module is used to build a hybrid model for short-term prediction of photovoltaic power generation. The hybrid model for short-term prediction of photovoltaic power generation includes a CNN-LSTM module, an LSTM module and a feature fusion module. The outputs of the CNN-LSTM module and the LSTM module are both connected to the feature fusion module.

[0130] The CNN-LSTM module processes the preprocessed satellite inversion data to obtain a first short-term prediction result for photovoltaic power generation; the LSTM module processes the preprocessed meteorological data and historical photovoltaic power generation data to obtain a second short-term prediction result for photovoltaic power generation; and the feature fusion module fuses the first and second short-term prediction results for photovoltaic power generation to obtain the final prediction result.

[0131] The training module is used to train the photovoltaic power generation short-term prediction hybrid model based on the photovoltaic power generation prediction dataset, iterating until the loss function converges;

[0132] The acquisition module is used to acquire satellite inversion data, meteorological data, and historical photovoltaic power generation data of the target photovoltaic power station;

[0133] The prediction module is used to preprocess the satellite inversion data, meteorological data and historical photovoltaic power generation data of the target photovoltaic power station, and then input them into the trained photovoltaic power generation short-term prediction hybrid model to obtain the photovoltaic power generation short-term prediction result of the target photovoltaic power station.

[0134] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0135] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0136] Computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0137] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0138] The units described in the embodiments of this disclosure can be implemented in software or hardware. The names of the units are not, in some cases, intended to limit the specific unit.

[0139] The preferred embodiments of the present invention have been described above to make the spirit of the present invention clearer and easier to understand, and are not intended to limit the present invention. All modifications, substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope summarized by the appended claims.

Claims

1. A photovoltaic power generation short-time prediction method based on satellite inversion data, characterized in that, The method comprises: Step S1, establishing a photovoltaic power generation prediction data set, the photovoltaic power generation prediction data set comprising satellite inversion data, meteorological data and photovoltaic power generation historical data; Step S2, preprocessing the photovoltaic power generation prediction data set; Step S3, constructing a photovoltaic power generation short-time prediction hybrid model, the photovoltaic power generation short-time prediction hybrid model comprising a CNN-LSTM module, an LSTM module and a feature fusion module, the output end of the CNN-LSTM module and the output end of the LSTM module being both connected to the feature fusion module; The CNN-LSTM module is used for processing the preprocessed satellite inversion data, and obtaining a photovoltaic power generation short-time prediction first result based on the satellite inversion data; the LSTM module is used for processing the preprocessed meteorological data and photovoltaic power generation historical data, and obtaining a photovoltaic power generation short-time prediction second result based on the meteorological data and the photovoltaic power generation historical data; and the feature fusion module is used for fusing the photovoltaic power generation short-time prediction first result and the photovoltaic power generation short-time prediction second result to obtain a final prediction result; Step S4, training the photovoltaic power generation short-time prediction hybrid model based on the photovoltaic power generation prediction data set, and iterating until the loss function converges; Step S5, obtaining satellite inversion data, meteorological data and photovoltaic power generation historical data of a target photovoltaic power station; Step S6, preprocessing the satellite inversion data, meteorological data and photovoltaic power generation historical data of the target photovoltaic power station, and inputting the preprocessed data into the trained photovoltaic power generation short-time prediction hybrid model to obtain a photovoltaic power generation short-time prediction result of the target photovoltaic power station.

2. The method for short-term photovoltaic power generation prediction based on satellite inversion data according to claim 1, characterized in that, The preprocessing comprises missing value processing, data normalization, enhanced Dickey-Fuller test and autocorrelation test.

3. The method for short-term photovoltaic power generation prediction based on satellite inversion data according to claim 2, characterized in that, The enhanced Dickey-Fuller test is specifically as follows: S21 : The photovoltaic power generation prediction dataset y is tested according to the following formula t Unit root: where Δy t is the first difference of the photovoltaic power generation prediction dataset y t ; α is a parameter to be tested, and if α = 0, it indicates that the photovoltaic power generation prediction dataset y t has a unit root and is non-stationary; is the lag term coefficient, p is the number of lag terms, β t is the time trend term, ∈ t is the error term; S22: Estimate a, β and γ in the above equation using the least square method S23: calculating the statistic quantity of the enhanced Dickey-Fuller test, as shown in the following formula: wherein is an estimate of a, is the standard error of S24: comparing the calculated statistical quantity ADF with a critical value Sta If ADF Sta is less than 0.05, the photovoltaic power generation prediction data set y t is stationary data; if ADF Sta is greater than 0.05, the photovoltaic power generation prediction data set y t is non-stationary data. S25: For non-stationary data, a difference method is used to eliminate the time series dependence of the photovoltaic power generation prediction data set y t , and then step S21 is entered to continue calculating ADF Sta until it is confirmed that the data set is stationary data.

4. The method for short-term photovoltaic power generation prediction based on satellite inversion data according to claim 3, characterized in that, After the enhanced Dickey-Fuller test, the autocorrelation test is continued, which is specifically as follows: S26: performing autocorrelation test according to the following formula: where h is the lag time, cov is the covariance calculation, var(y t ) is the variance of the photovoltaic power generation prediction dataset y t time series; S27: judging whether the above calculation result ACF(h) exceeds a preset confidence interval threshold, and in the case of exceeding the preset confidence interval threshold, considering that the photovoltaic power generation prediction data set y t satisfies the correlation requirement of the time series.

5. The method for short-term photovoltaic power generation prediction based on satellite inversion data according to claim 1, characterized in that, The CNN-LSTM module first extracts spatial features of the satellite inversion data through a CNN layer, and maps the obtained feature map to an LSTM layer, the LSTM layer being used for extracting time information for time series prediction, and finally obtaining the photovoltaic power generation short-time prediction first result through a fully connected layer.

6. The method for short-term photovoltaic power generation prediction based on satellite inversion data according to claim 5, characterized in that, The specific process of the CNN-LSTM module obtaining the photovoltaic power generation short-time prediction first result is as follows: Step S311: the CNN layer processes the satellite inversion data through convolution and an activation function: O CNN = ReLU(W CNN * X + b CNN ); where X is the satellite inversion data, W CNN and b CNN are the weights and biases of the CNN layers, * denotes the convolution operation, and ReLU denotes the rectified linear activation unit function. Step S312: reducing the feature dimension by using a pooling operation, i.e. P = Pooling(O CNN ); Wherein, Pooling represents the pooling operation; Step S313: taking the output P of the CNN layer as the input of the LSTM layer, the dimension of the output P being D x T, where D is the feature depth and T is the time sequence step, the LSTM layer processing and outputting the hidden state h of each time step t t : h t = LSTM(P, h t-1 ); where h t is the hidden state at time step t, LSTM denotes the operation of an LSTM unit, h t-1 is the hidden state of the previous time step; Step S314: obtaining the hidden state h T The final output is obtained by processing through a fully connected layer Specifically: Z = W FC H + b FC ; where W FC and b FC are the bias and weights of the fully connected layer, respectively, and softmax is the activation function.

7. The method for short-term photovoltaic power generation prediction based on satellite inversion data according to claim 1, characterized in that, The LSTM module, after receiving the meteorological data and the photovoltaic power generation historical data, firstly combines the meteorological data and the photovoltaic power generation historical data by using timestamp information, and then performs photovoltaic power generation short-time prediction on the combined data to obtain the photovoltaic power generation short-time prediction second result; wherein the meteorological data comprises radiation, DHI, DNI, GHI, relative humidity, wind direction, dew point, pressure, temperature and solar zenith angle, and the photovoltaic power generation historical data comprises photovoltaic historical power generation.

8. A photovoltaic power generation short-time prediction device based on satellite inversion data, characterized by, The device comprises: a dataset construction module, configured to establish a photovoltaic power generation prediction dataset, wherein the photovoltaic power generation prediction dataset comprises satellite inversion data, meteorological data and photovoltaic power generation historical data; a preprocessing module, configured to preprocess the photovoltaic power generation prediction dataset; a model construction module, configured to construct a photovoltaic power generation short-time prediction hybrid model, wherein the photovoltaic power generation short-time prediction hybrid model comprises a CNN-LSTM module, an LSTM module and a feature fusion module, and the output end of the CNN-LSTM module and the output end of the LSTM module are both connected to the feature fusion module; wherein the CNN-LSTM module is configured to process the preprocessed satellite inversion data, and obtain a photovoltaic power generation short-time prediction first result based on the satellite inversion data; the LSTM module is configured to process the preprocessed meteorological data and photovoltaic power generation historical data, and obtain a photovoltaic power generation short-time prediction second result based on the meteorological data and the photovoltaic power generation historical data; and the feature fusion module is configured to fuse the photovoltaic power generation short-time prediction first result and the photovoltaic power generation short-time prediction second result to obtain a final prediction result; a training module, configured to train the photovoltaic power generation short-time prediction hybrid model based on the photovoltaic power generation prediction dataset, and iterate until a loss function converges; an acquisition module, configured to acquire satellite inversion data, meteorological data and photovoltaic power generation historical data of a target photovoltaic power station; a prediction module, configured to input the preprocessed satellite inversion data, meteorological data and photovoltaic power generation historical data of the target photovoltaic power station into the trained photovoltaic power generation short-time prediction hybrid model to obtain a photovoltaic power generation short-time prediction result of the target photovoltaic power station.

9. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 8. The computer program, when executed by a processor, implements the photovoltaic power generation short-time prediction method based on satellite inversion data according to any one of claims 1 to 7.

10. An electronic device, the electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the computer program to implement the photovoltaic power generation short-time prediction method based on satellite inversion data according to any one of claims 1 to 7.

Images