A photovoltaic power prediction method

Abstract

This application relates to a photovoltaic power prediction method. The method includes: acquiring training data, which includes multiple photovoltaic training parameters and multiple training photovoltaic powers; the photovoltaic training parameters include a first photovoltaic parameter and a second photovoltaic parameter; the second photovoltaic parameter is a photovoltaic parameter that is temporally continuous with the first photovoltaic parameter but later in time; the training photovoltaic power is the actual photovoltaic power within the time corresponding to the second photovoltaic parameter; inputting the first photovoltaic training parameter into an initial prediction model to obtain a first predicted photovoltaic power; the first predicted photovoltaic power is the predicted photovoltaic power within the time corresponding to the second photovoltaic parameter; determining a loss value based on the first predicted photovoltaic power and the first photovoltaic power; and optimizing the parameters of the initial prediction model based on the loss value to obtain a target prediction model. This method can train a target prediction model with high accuracy in photovoltaic power prediction.

Description

Technical Field

[0001] This application relates to the field of photovoltaic technology, and in particular to a method for predicting photovoltaic power. Background Technology

[0002] With the popularization of photovoltaic power generation technology, photovoltaic power prediction technology has emerged. Photovoltaic power prediction refers to the prediction of the output power of photovoltaic power generation systems through various technical means, in order to improve the accuracy of power system resource scheduling and ensure the stable operation of the power grid.

[0003] With the acceleration of the global energy transition and the continuous expansion of photovoltaic power generation capacity, the randomness and intermittency of photovoltaic power generation technology have brought severe challenges to power grid dispatch. If the accuracy of photovoltaic power prediction technology cannot be improved, it will be difficult for power grid dispatching departments to ensure the operational stability of the power grid. Summary of the Invention

[0004] Therefore, it is necessary to provide a photovoltaic power prediction method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can train a target prediction model with high accuracy in photovoltaic power prediction to address the above-mentioned technical problems.

[0005] In a first aspect, this application provides a photovoltaic power prediction method, including:

[0006] Acquire training data, which includes multiple photovoltaic training parameters and multiple training photovoltaic powers; the photovoltaic training parameters include a first photovoltaic parameter and a second photovoltaic parameter; the second photovoltaic parameter is a photovoltaic parameter that is continuous with the first photovoltaic parameter in time but later than the first photovoltaic parameter; the training photovoltaic power is the actual photovoltaic power within the time corresponding to the second photovoltaic parameter;

[0007] The first photovoltaic training parameter is input into the initial prediction model to obtain the first predicted photovoltaic power; the first photovoltaic training parameter is one of multiple photovoltaic training parameters, and the first predicted photovoltaic power is the photovoltaic power within the time period corresponding to the predicted second photovoltaic parameter;

[0008] The loss value is determined based on the first predicted photovoltaic power and the first photovoltaic power; the first photovoltaic power is one photovoltaic power that corresponds to the first photovoltaic training parameter among multiple training photovoltaic powers;

[0009] The parameters of the initial prediction model are optimized based on the loss value to obtain the target prediction model.

[0010] Secondly, this application also provides a photovoltaic power prediction device, comprising:

[0011] The acquisition module is used to acquire training data, which includes multiple photovoltaic training parameters and multiple training photovoltaic powers. The photovoltaic training parameters include a first photovoltaic parameter and a second photovoltaic parameter. The second photovoltaic parameter is a photovoltaic parameter that is continuous with the first photovoltaic parameter in time but later than the first photovoltaic parameter. The training photovoltaic power is the actual photovoltaic power within the time corresponding to the second photovoltaic parameter.

[0012] The first determining module is used to input the first photovoltaic training parameters into the initial prediction model to obtain the first predicted photovoltaic power; the first photovoltaic training parameter is one of a plurality of photovoltaic training parameters, and the first predicted photovoltaic power is the photovoltaic power within the time period corresponding to the predicted second photovoltaic parameter;

[0013] The second determining module is used to determine the loss value based on the first predicted photovoltaic power and the first photovoltaic power; the first photovoltaic power is a photovoltaic power corresponding to the first photovoltaic training parameter among multiple training photovoltaic powers;

[0014] The third determining module is used to optimize the parameters of the initial prediction model based on the loss value to obtain the target prediction model.

[0015] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement some or all of the steps described in any method of the first aspect of the embodiments of this application.

[0016] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements some or all of the steps described in any method of the first aspect of the embodiments of this application.

[0017] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements some or all of the steps described in any method of the first aspect of the embodiments of this application.

[0018] The aforementioned photovoltaic power prediction method, apparatus, computer equipment, computer-readable storage medium, and computer program product acquire training data, which includes multiple photovoltaic training parameters and multiple training photovoltaic powers. The photovoltaic training parameters include a first photovoltaic parameter and a second photovoltaic parameter. The second photovoltaic parameter is a photovoltaic parameter that is temporally continuous with the first photovoltaic parameter but later in time. The training photovoltaic power is the actual photovoltaic power within the time corresponding to the second photovoltaic parameter. The first photovoltaic training parameter is input into an initial prediction model to obtain a first predicted photovoltaic power. The first photovoltaic training parameter is one of multiple photovoltaic training parameters, and the first predicted photovoltaic power is the photovoltaic power within the time corresponding to the predicted second photovoltaic parameter. A loss value is determined based on the first predicted photovoltaic power and the first photovoltaic power. The first photovoltaic power is one of the multiple training photovoltaic powers corresponding to the first photovoltaic training parameter. The parameters of the initial prediction model are optimized based on the loss value to obtain a target prediction model. Using the photovoltaic power prediction method provided in this application, since the photovoltaic training parameters include a first photovoltaic parameter and a second photovoltaic parameter with temporal continuity, by optimizing the parameters of the initial prediction model based on the loss value determined from the first predicted photovoltaic power and the first photovoltaic power, a target prediction model with high accuracy in photovoltaic power prediction can be trained. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a diagram illustrating the application environment of the photovoltaic power prediction method in one embodiment;

[0021] Figure 2 This is a flowchart illustrating a photovoltaic power prediction method in one embodiment;

[0022] Figure 3 This is a schematic diagram of the structure of the initial prediction model in one embodiment;

[0023] Figure 4 This is a structural block diagram of a photovoltaic power prediction device in one embodiment;

[0024] Figure 5 This is an internal structural diagram of a computer device in one embodiment;

[0025] Figure 6 This is a diagram of the internal structure of a computer device in another embodiment. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0027] The photovoltaic power prediction method provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed on the cloud or other network servers. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can be smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, projection devices, etc. Portable wearable devices can be smartwatches, smart bracelets, head-mounted devices, etc. Head-mounted devices can be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. Server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. It is easily understood that, in addition to obtaining the target prediction model using the photovoltaic power prediction method provided in this application embodiment, the application of the target prediction model can also be applied to, for example... Figure 1 The application environment shown.

[0028] In one exemplary embodiment, such as Figure 2 As shown, a photovoltaic power prediction method is provided, which is applied to... Figure 1 Taking the terminal in the example, the explanation includes the following steps 202 to 208. Wherein:

[0029] Step 202: Obtain training data, which includes multiple photovoltaic training parameters and multiple training photovoltaic powers. The photovoltaic training parameters include a first photovoltaic parameter and a second photovoltaic parameter. The second photovoltaic parameter is a photovoltaic parameter that is continuous with the first photovoltaic parameter in time but later than the first photovoltaic parameter. The training photovoltaic power is the actual photovoltaic power within the time corresponding to the second photovoltaic parameter.

[0030] The training data can be obtained from historical data of photovoltaic power generation systems. Optionally, a photovoltaic power generation system can refer to a photovoltaic power station.

[0031] Optionally, the first photovoltaic parameters may include a first weather variable, a first theoretical irradiance, a first photovoltaic module temperature, and a second photovoltaic power.

[0032] Optionally, the second photovoltaic parameters may include the second weather variable, the second theoretical irradiance, and the second photovoltaic module temperature.

[0033] Because during model training, the parameters of the initial prediction model need to be optimized based on the loss value between the first predicted photovoltaic power output by the initial prediction model and the first photovoltaic power in the training photovoltaic power to obtain the final target prediction model after training, the photovoltaic power parameter is not included in the second photovoltaic parameter.

[0034] The second photovoltaic parameter is continuous with the first photovoltaic parameter in time, meaning that the start time of the second photovoltaic parameter immediately follows the end time of the first photovoltaic parameter.

[0035] Optionally, the time length corresponding to the second photovoltaic parameter is shorter than the time length corresponding to the first photovoltaic parameter.

[0036] For example, assuming the first photovoltaic parameter corresponds to the period from June 1, 2024 to August 31, 2024, the second photovoltaic parameter can correspond to the period from September 1, 2024 to September 3, 2024. By ensuring that the second photovoltaic parameter is continuous with the first photovoltaic parameter in time, it is beneficial to ensure that the first predicted photovoltaic power obtained by the initial prediction model based on the first photovoltaic training parameters can fully consider the photovoltaic power characteristics within the time length corresponding to both the first and second photovoltaic parameters. In other words, it can more fully consider the temporal continuity of photovoltaic power characteristics, thus achieving higher accuracy.

[0037] Step 204: Input the first photovoltaic training parameter into the initial prediction model to obtain the first predicted photovoltaic power; the first photovoltaic training parameter is one of multiple photovoltaic training parameters, and the first predicted photovoltaic power is the photovoltaic power within the time period corresponding to the predicted second photovoltaic parameter.

[0038] The initial prediction model refers to the prediction model that is in a non-converged state, meaning that the parameters have not yet been optimized.

[0039] In a straightforward manner, the first photovoltaic training parameter includes its corresponding first photovoltaic parameter and second photovoltaic parameter.

[0040] The first predicted photovoltaic power refers to the result of the initial prediction model predicting the photovoltaic power within the time corresponding to the second photovoltaic parameter.

[0041] Step 206: Determine the loss value based on the first predicted photovoltaic power and the first photovoltaic power; the first photovoltaic power is one photovoltaic power that corresponds to the first photovoltaic training parameter among multiple training photovoltaic powers.

[0042] The loss value refers to a quantitative indicator used to measure the difference between the first predicted photovoltaic power and the first photovoltaic power.

[0043] The first photovoltaic power refers to the photovoltaic power within the time period corresponding to the first photovoltaic training parameter.

[0044] Step 208: Optimize the parameters of the initial prediction model based on the loss value to obtain the target prediction model.

[0045] In an exemplary embodiment, the above-mentioned optimization of the parameters of the initial prediction model based on the loss value to obtain the target prediction model includes: optimizing the parameters of the initial prediction model based on the target loss function and the loss value to obtain the target prediction model.

[0046] Optionally, the target loss function can be a rectified linear unit (ReLU) function.

[0047] In a straightforward manner, the parameters of the initial prediction model are iteratively optimized based on the loss value until the loss value converges, thus obtaining the target prediction model.

[0048] In the aforementioned photovoltaic power prediction method, training data is acquired, including multiple photovoltaic training parameters and multiple training photovoltaic powers. The photovoltaic training parameters include a first photovoltaic parameter and a second photovoltaic parameter. The second photovoltaic parameter is a photovoltaic parameter that is temporally continuous with the first photovoltaic parameter but occurs later. The training photovoltaic power is the actual photovoltaic power within the time corresponding to the second photovoltaic parameter. The first photovoltaic training parameter is input into the initial prediction model to obtain the first predicted photovoltaic power. The first photovoltaic training parameter is one of the multiple photovoltaic training parameters, and the first predicted photovoltaic power is the photovoltaic power within the time corresponding to the predicted second photovoltaic parameter. A loss value is determined based on the first predicted photovoltaic power and the first photovoltaic power. The first photovoltaic power is one of the multiple training photovoltaic powers corresponding to the first photovoltaic training parameter. The parameters of the initial prediction model are optimized based on the loss value to obtain the target prediction model. Using the photovoltaic power prediction method provided in this application, since the photovoltaic training parameters include a first photovoltaic parameter and a second photovoltaic parameter with temporal continuity, by optimizing the parameters of the initial prediction model based on the loss value determined from the first predicted photovoltaic power and the first photovoltaic power, a target prediction model with high accuracy in photovoltaic power prediction can be trained.

[0049] In one exemplary embodiment, the method further includes:

[0050] Obtain the data to be predicted;

[0051] Based on the data to be predicted, the photovoltaic power is predicted using a target prediction model to obtain the target photovoltaic power.

[0052] Since the first photovoltaic training parameters input to the initial prediction model include the first photovoltaic parameter and the second photovoltaic parameter, the data to be predicted also includes photovoltaic parameters within a preset historical time period and photovoltaic parameters within a preset future time period. The photovoltaic parameters within the preset future time period are continuous in time with the photovoltaic parameters within the preset historical time period, and the time period is later than that of the photovoltaic parameters within the preset historical time period.

[0053] Target photovoltaic power refers to the photovoltaic power within a predicted future time period corresponding to the photovoltaic parameters.

[0054] In this embodiment, since the target prediction model has high accuracy in photovoltaic power prediction, by inputting the data to be predicted into the target prediction model, the target photovoltaic power corresponding to the preset future time can be obtained with high accuracy.

[0055] In an exemplary embodiment, obtaining the data to be predicted includes:

[0056] Acquire historical weather variables, historical photovoltaic module temperature, historical photovoltaic power within a preset historical time period, as well as predicted weather variables within a preset future time period;

[0057] Historical theoretical irradiance is determined based on historical weather variables;

[0058] Determine the theoretical irradiance based on predicted weather variables;

[0059] Based on historical weather variables, predicted weather variables, static heat loss coefficient, wind cooling coefficient, and the absorption efficiency of solar radiation energy by photovoltaic modules, the predicted temperature of photovoltaic modules is determined.

[0060] Historical weather variables, historical theoretical irradiance, historical photovoltaic module temperature, historical photovoltaic power, predicted weather variables, predicted theoretical irradiance, and predicted photovoltaic module temperature are identified as the data to be predicted.

[0061] The time length corresponding to the preset future time is less than the time length corresponding to the preset historical time.

[0062] Optionally, the preset historical time can be 1 month, 3 months, 6 months or other historical time lengths.

[0063] Optionally, a future time can be preset, which can be 1 day, 3 days, 5 days or other future time lengths.

[0064] Optionally, weather variables can be obtained from meteorological bureau platforms, meteorological stations where photovoltaic power generation systems are located, or other meteorological agencies.

[0065] Theoretical irradiance and photovoltaic module temperature are variables that are manually calculated and added to reduce model bias caused by sensor errors or outliers and to enhance the generalization ability of the target prediction model to cope with different climatic conditions. In other words, the historical theoretical irradiance and predicted theoretical irradiance involved in the initial prediction model training process and the application process of the target prediction model are manually calculated, while the historical photovoltaic module temperature can be obtained by directly retrieving historical data, and the predicted photovoltaic module temperature needs to be calculated.

[0066] Theoretical irradiance is used to provide an accurate irradiance benchmark, which helps to remove random fluctuations in the weather environment caused by the atmosphere and weather, strengthens the physical consistency of the weather environment, improves the adaptability of the final trained target prediction model to different weather and time periods, and reduces the fitting bias of the target prediction model to climate change.

[0067] Optionally, the clear sky model (Ineichen model) can be used to calculate the historical theoretical irradiance and the predicted theoretical irradiance of the photovoltaic power generation system.

[0068] The static heat loss coefficient is a parameter used to describe the heat loss capacity of photovoltaic modules in a photovoltaic power generation system under static conditions, where there is no wind cooling or other convection effects.

[0069] The wind cooling coefficient is a parameter used to describe the cooling effect of wind on photovoltaic modules.

[0070] The temperature of a photovoltaic module can be understood as the temperature of a photovoltaic cell.

[0071] The solar radiation absorption efficiency of a photovoltaic (PV) module refers to the ratio between the portion of solar radiation energy that can be converted into electrical energy by the PV module and the solar radiation energy absorbed by the PV module.

[0072] Photovoltaic module temperature is a crucial factor affecting photovoltaic (PV) power output. However, when predicting PV power output over future time periods from the current point in time, it is usually impossible to directly obtain the future PV module temperature. Therefore, alternatively, a wind speed-coupled thermal balance model can be used to calculate and predict the PV module temperature.

[0073] Optionally, the predicted photovoltaic module temperature can be determined based on the historical average temperature in historical weather variables, the predicted total irradiance in predicted weather variables, the predicted wind speed in predicted weather variables, the temperature rise adjustment term in predicted weather variables, the static heat loss coefficient, the wind cooling coefficient, and the absorption efficiency of the photovoltaic module for solar radiation energy.

[0074] For example, the historical temperature average in historical weather variables is represented as The predicted total irradiance in the predicted weather variables is expressed as The predicted wind speed in the weather forecast variables is represented as The temperature rise adjustment term in the weather forecast variables is represented as The static heat loss coefficient is expressed as The air-cooling coefficient is expressed as The absorption efficiency of photovoltaic modules for solar radiation energy is expressed as... The predicted temperature of a photovoltaic module can be expressed by the following formula: .

[0075] In this embodiment, the data to be predicted is determined by historical weather variables, historical theoretical irradiance, historical photovoltaic module temperature, historical photovoltaic power, predicted weather variables, predicted theoretical irradiance, and predicted photovoltaic module temperature. Among them, the historical theoretical irradiance, predicted theoretical irradiance, and predicted photovoltaic module temperature are all variables that have been calculated and determined. Therefore, the data quality of the data to be predicted can be improved, thereby improving the accuracy and reliability of the target photovoltaic power predicted by the target prediction model in the future time.

[0076] In an exemplary embodiment, the first photovoltaic parameters include a first weather variable, a first theoretical irradiance, a first photovoltaic module temperature, and a second photovoltaic power, and the second photovoltaic parameters include a second weather variable, a second theoretical irradiance, and a second photovoltaic module temperature.

[0077] In this regard, it is easy to understand that whether in the training process of the initial prediction model or in the application process of the target prediction model, the photovoltaic power in the future time relative to the historical time needs to be predicted. Therefore, in the training process of the initial prediction model, although the second photovoltaic parameter, which is used as historical data, actually includes the first photovoltaic power, the first photovoltaic power will not be input into the initial prediction model.

[0078] It should be noted that since the training process of the initial prediction model corresponds to the application process of the target prediction model, the calculation methods of the first theoretical irradiance in the first photovoltaic parameter and the second theoretical irradiance in the second photovoltaic parameter are the same as the calculation methods of the historical theoretical irradiance and the predicted theoretical irradiance in the data to be predicted, so they will not be elaborated here.

[0079] In one exemplary embodiment, such as Figure 3 As shown, the initial prediction model 30 includes a time series module 302, a non-time series module 304, and a prediction module 306;

[0080] The above-mentioned input of the first photovoltaic training parameters into the initial prediction model 30 to obtain the first predicted photovoltaic power includes:

[0081] The time-series module 302 is used to extract features from the first photovoltaic parameters in the first photovoltaic training parameters to obtain time-series features;

[0082] The non-time-series module 304 is used to extract features from the second photovoltaic parameters in the first photovoltaic training parameters to obtain non-time-series features.

[0083] Based on time-series and non-time-series characteristics, the prediction module 306 is used to predict the photovoltaic power to obtain the first predicted photovoltaic power.

[0084] Since environmental factors such as illumination and temperature over historical periods have temporal dynamic characteristics, a time-series module 302 with autoregressive temporal feature extraction capabilities is particularly suitable for the first photovoltaic training parameters. Optionally, a time-series module 302 including a Temporal Convolutional Network (TCN) layer can be used to extract features from the first photovoltaic parameters in the first photovoltaic training parameters; that is, the time-series module 302 may include a TCN layer.

[0085] The TCN layer consists of causal convolutions, residual connections, and dilated convolutions. Causal convolutions ensure that when predicting the first photovoltaic power in future timeframes, the TCN layer relies only on the first photovoltaic parameters from historical timeframes. Residual connections address the vanishing gradient problem in deep networks through skip connections, improving the training efficiency of the initial prediction model 30. Dilated convolutions expand the receptive field of the convolution kernel through interval sampling, capturing the dependencies of the first photovoltaic parameters over a longer timeframe without increasing computational complexity. It can be seen that, compared to traditional Recurrent Neural Networks (RNNs) or Long Short-Term Memory (LSTM) networks, the TCN layer exhibits better parallelism and training stability during the training of the initial prediction model 30.

[0086] Since the most direct factor affecting photovoltaic power in the future is the weather conditions at the corresponding time point, in addition to the time-series module 302 used to extract the time-series features of the first photovoltaic parameter in historical time, a non-time-series module 304 is needed to extract the impact of other variables on photovoltaic power in the future. Optionally, other variables in the future include weather variables, manually calculated theoretical irradiance, and photovoltaic cell temperature, etc. Optionally, the non-time-series module 304, which includes a fully convolutional neural network (FCNN) layer, can be used to extract high-order features of weather variables in the future to obtain non-time-series features corresponding to the weather variables in the future.

[0087] It should be noted that during the training of the initial prediction model 30, the second photovoltaic parameter is a photovoltaic parameter in the future relative to the first photovoltaic parameter. However, relative to the current time point, both the first and second photovoltaic parameters are photovoltaic parameters in the historical time period.

[0088] For example, when the non-temporal module 304 includes an FCNN layer, the FCNN layer includes multiple convolutional layers, the number of which is expressed as... The weights of the l-th convolutional layer are represented as The bias term of the l-th convolutional layer is represented as Representing non-temporal features as Non-temporal characteristics As a high-level abstract representation of weather variables, it can highlight non-time-series weather environmental attributes to generate high-level, abstract weather environmental features for representing weather environmental states. Based on this, non-time-series features can be represented by the following calculation formula: It can be seen that by extracting features from the second photovoltaic parameters in the first photovoltaic training parameters through the multi-layer convolutional layer of the non-time-series module 304, the non-time-series weather variable attribute information can be highlighted. Thus, high-level, abstract non-time-series features can be generated to represent the state of weather variables in the future time through non-time-series features.

[0089] Optionally, the weights of different convolutional layers in the FCNN layer are continuously adjusted during the training of the initial prediction model 30. Ultimately, after multiple transformations of the FCNN layer, the overall mapping relationship from input to output gradually approaches, that is, the mapping relationship from the second photovoltaic parameter to the non-temporal feature gradually becomes clear and explicit.

[0090] In this embodiment, a time-series module is used to extract features from the first photovoltaic parameter to obtain time-series features, and a non-time-series module is used to extract features from the second photovoltaic parameter to obtain non-time-series features. Further, based on the time-series and non-time-series features, a prediction module is used to predict the photovoltaic power to obtain the first predicted photovoltaic power. Based on this, effective complementarity of information based on time-series and non-time-series features is achieved in the process of predicting photovoltaic power in the future. The second photovoltaic parameter, with its non-time-series features, fully considers the weather variable state in the future, thereby ensuring that the trained target prediction model has high accuracy in photovoltaic power prediction.

[0091] In one exemplary embodiment, such as Figure 3 As shown, the time-series module 302 includes a time-series feature extraction layer 3022; the time-series module 302 is used to extract features from the first photovoltaic parameters in the first photovoltaic training parameters to obtain time-series features, including:

[0092] The temporal feature extraction layer 3022 is used to extract features from the first photovoltaic parameters in the first photovoltaic training parameters to obtain temporal features.

[0093] In this embodiment, when the temporal feature extraction layer is a TCN layer, the temporal characteristics of the first photovoltaic parameter can be fully captured through the TCN layer, thereby extracting richer and more accurate temporal features. These temporal features provide accurate temporal information for the initial prediction model training process, ensuring that the target prediction model trained based on these temporal features has high accuracy in photovoltaic power prediction.

[0094] In one exemplary embodiment, such as Figure 3 As shown, the timing module 302 also includes a decomposition layer 3024;

[0095] The aforementioned time-series feature extraction layer 3022 is used to extract features from the first photovoltaic parameters in the first photovoltaic training parameters to obtain time-series features, including:

[0096] Variational mode decomposition of the first photovoltaic parameters is performed using decomposition layer 3024 to obtain multiple intrinsic mode functions; the values ​​of the multiple intrinsic mode functions satisfy preset value conditions;

[0097] The temporal feature extraction layer 3022 is used to extract features from multiple intrinsic mode functions to obtain temporal features.

[0098] Variational Mode Decomposition (VMD) is an adaptive signal processing method used to decompose complex signals into multiple Intrinsic Mode Functions (IMFs) with different center frequencies and bandwidths. VMD achieves multi-scale decomposition of signals by constructing a variational problem and using optimization algorithms to find the optimal IMFs.

[0099] Variational mode decomposition of the first photovoltaic parameter yields multiple intrinsic mode functions, each corresponding to a different frequency scale. These intrinsic mode functions can reflect the changes of the first photovoltaic parameter in different frequency bands, which is helpful in revealing the hidden structural relationships in the first photovoltaic parameter.

[0100] The purpose of performing variational mode decomposition on the first photovoltaic parameter is to find a set of optimal eigenmode functions and center frequencies that minimize the objective function.

[0101] The preset value condition refers to the condition that makes the objective function of the variational mode decomposition reach its minimum value. The preset value condition means that the values ​​of multiple eigenmode functions satisfy the condition that makes the objective function of the variational mode decomposition reach its minimum value.

[0102] For example, the number of intrinsic mode functions is expressed as The intrinsic mode function is expressed as The center frequency is expressed as The partial derivative with respect to time t is expressed as The Dirac function representing the unit impulse is expressed as: The kernel of the Hilbert transform used to generate the analytic signal is represented as Then the complex analytic signal can be expressed as Based on this, the objective function can be expressed as: Based on mathematical principles, the objective function is an energy function that sums up all intrinsic mode functions to obtain the minimum value.

[0103] Optionally, the decomposition layer 3024 is used to perform variational mode decomposition on the first photovoltaic parameters to obtain multiple intrinsic mode functions. This is achieved by using the decomposition layer 3024 to perform variational mode decomposition on the historical photovoltaic power (i.e., the second photovoltaic power) in the first photovoltaic parameters to obtain multiple intrinsic mode functions corresponding to historical photovoltaic power with different center frequencies. Then, the multiple intrinsic mode functions are combined with other parameter variables in the first photovoltaic parameters other than historical photovoltaic power (i.e., the first weather variable, the first theoretical irradiance, and the first photovoltaic module temperature) to obtain multiple variable functions. These multiple variable functions are then input to the time series feature extraction layer 3022 for feature extraction to obtain time series features.

[0104] When the temporal module 302 includes a TCN layer, the temporal feature extraction layer 3022 can be the TCN layer. In this case, the temporal feature extraction layer 3022 can be composed of multiple residual units stacked together. Each residual unit contains a causal dilation convolution component, a weight normalization component, an activation function component, a regularization component, and a 1*1 convolution component.

[0105] The causal dilated convolution component is used to perform convolution operations that combine the characteristics of causal convolution and dilated convolution. Dilated convolution expands the receptive field of the convolution kernel through interval sampling, which can capture the dependence of the first photovoltaic parameter over a longer time range without increasing computational complexity. Based on this, the causal dilated convolution component can both maintain causality and expand the receptive field, making it suitable for temporal feature extraction.

[0106] Weight normalization is a method used to normalize the weights of convolutional kernels, which improves the stability and efficiency of the training process by reducing the problems of gradient vanishing and gradient exploding.

[0107] Optionally, the activation function is the ReLU function. Activation functions are used to enable neural networks to learn complex patterns and relationships, and can also help reduce the vanishing gradient problem.

[0108] The regularization component is used to randomly discard a portion of the neuron outputs during the training of the initial prediction model 30, in order to prevent overfitting in the final target prediction model and improve the generalization ability of the target prediction model.

[0109] A 1x1 convolutional component is used to adjust the number of channels of the first photovoltaic parameter without changing the spatial dimension of the first photovoltaic parameter, so as to ensure that the first photovoltaic parameter can be residually connected.

[0110] For example, the eigenmode function of the first photovoltaic parameter is expressed as: The temporal feature extraction layer 3022 is used to extract features from multiple intrinsic mode functions, realizing feature encoding through multi-layer dilated convolution. The number of layers is represented as... The dilated convolution operation is represented as , Finally, the feature encoding obtained through multi-layer dilated convolution is: By fusing the features of all intrinsic mode functions, the resulting time-series features can be expressed as follows: , where i is the number of intrinsic mode functions.

[0111] In this embodiment, the time series module includes a decomposition layer and a time series feature extraction layer. Thus, through variational mode decomposition and multi-layer convolution operations, it is ensured that the target prediction model trained in the end can effectively capture the time series features of photovoltaic parameters in historical time periods, thereby improving the target prediction model's ability to understand and predict time series data, and also improving the generalization ability of the target prediction model.

[0112] In one exemplary embodiment, such as Figure 3 As shown, the timing module 302 also includes a dynamic window layer 3026;

[0113] The above-mentioned variational mode decomposition of the first photovoltaic parameters using decomposition layer 3024 yields multiple intrinsic mode functions, including:

[0114] The first photovoltaic parameter is filtered using a dynamic window layer 3026 to obtain partial training data, which includes the partial training data.

[0115] Variational mode decomposition was performed on a portion of the training data using decomposition layer 3024 to obtain multiple intrinsic mode functions.

[0116] The first photovoltaic parameter includes a portion of the training data, meaning the amount of the training data is less than that of the first photovoltaic parameter.

[0117] The dynamic window layer 3026 has different window lengths at different time points, that is to say, the dynamic window layer 3026 is a dynamic window that adaptively adjusts with time points.

[0118] Optionally, when the first photovoltaic parameter includes weather variables (i.e., the first weather variable) within its corresponding historical time period, and the weather variables within the historical time period include irradiance data, the dynamic window layer 3026 used for data filtering of the first photovoltaic parameter has a corresponding dynamic window length determined by the minimum window length, the maximum window length, the irradiance variance at the corresponding time node, the response sensitivity, the response threshold, and the median of the historical rolling maximum variance.

[0119] Optionally, the corresponding time node can be the transition time node between the first photovoltaic parameter and the second photovoltaic parameter, that is, the time intersection point between the time node of the first photovoltaic parameter and the time node of the second photovoltaic parameter. During the data filtering of the first photovoltaic parameter using the dynamic window layer 3026, the dynamic window length corresponding to the dynamic window layer 3026 is the dynamic window length under the corresponding time node.

[0120] Optionally, the minimum window length and the maximum window length are preset lengths.

[0121] Optionally, the response sensitivity and response threshold are preset parameters that can be adjusted.

[0122] For example, the minimum window length is represented as The maximum window length is expressed as The variance of irradiance at the corresponding time point is expressed as The response sensitivity is expressed as The response threshold is expressed as The median of the historical rolling maximum variance is expressed as The size of the scroll window is represented as Then the median of the historical rolling maximum variance can be expressed as: The median of the historical rolling maximum variance can improve the robustness of the dynamic window length against anomalies. The dynamic window length corresponding to dynamic window layer 3026 is expressed as... The length of the dynamic window corresponding to dynamic window layer 3026 can be expressed by the following formula: The formula for calculating the dynamic window length corresponding to dynamic window layer 3026 incorporates a sigmoid non-linear mapping, ensuring that the range of dynamic window length corresponding to dynamic window layer 3026 can be adaptively adjusted.

[0123] It is easy to understand that, based on the calculation formula of the dynamic window length corresponding to the dynamic window layer 3026, it can be seen that the dynamic window length is smaller in weather conditions such as rainy or foggy weather with large changes in irradiance, and larger in weather conditions such as clear skies with small changes in irradiance.

[0124] It should be noted that after adaptively obtaining the dynamic window length, the lengths of the various dynamic windows are inconsistent, making them unsuitable for subsequent training of the initial prediction model 30. Therefore, a mask is constructed here to unify the input length. For the dynamic window layer 3026 corresponding to For earlier All time steps are set to 0.

[0125] In this embodiment, based on a dynamic window layer, a shorter dynamic time window is used to capture rapid changes in the first photovoltaic parameter during short-term fluctuating weather conditions such as rain and dense fog, while a longer dynamic time window is used to extract the stable, long-term trend of the first photovoltaic parameter during clear weather. Therefore, this embodiment avoids the traditional coarse method of classifying weather types on a daily basis, allowing the dynamic window layer to adapt to different weather changes within the same day. In other words, the dynamic window layer can dynamically adapt to short-term weather changes, adjusting the granularity of feature extraction for the first photovoltaic parameter using an adaptive time window method to cope with different scales of short-term weather changes. This improves the temporal sensitivity of the trained target prediction model and enhances its targeting of temporal features.

[0126] In one exemplary embodiment, obtaining training data as described above includes:

[0127] Obtain initial training data;

[0128] The initial training data is divided into multiple initial training data segments according to the preset time window length.

[0129] Determine the mean and standard deviation of each initial training data segment within its corresponding preset time window length;

[0130] Based on each initial training data segment, the mean of each initial training data segment, and the standard deviation of each initial training data segment, outliers in the initial training data are identified.

[0131] The training data is obtained by removing outliers from the initial training data.

[0132] Initial training data refers to training data that has not yet undergone outlier removal.

[0133] Optionally, the preset time window length is set by the user. Optionally, the preset time window length can be a 1-week time window, a 10-day time window, a 1-month time window, or a time window of other lengths.

[0134] In a straightforward manner, the preset time window length is shorter than the time length corresponding to the first photovoltaic parameter.

[0135] The time window length of each initial training data segment in multiple initial training data segments is the same.

[0136] Optionally, determining the mean and standard deviation of each initial training data segment within its corresponding preset time window length involves determining the mean and standard deviation of different types of parameters within the preset time window length in each initial training data segment. This allows for the removal of abnormal data from the initial training data, eliminating abnormal data corresponding to different types of parameters, thus obtaining training data in which no abnormal data exists for any type of parameter.

[0137] In an exemplary embodiment, the above-described method for determining outlier data in the initial training data based on each initial training data segment, the mean of each initial training data segment, and the standard deviation of each initial training data segment includes: determining the product between the adjustment parameter and the standard deviation of each initial training data segment; determining the minimum value of the observed data based on the difference between the mean and the product of each initial training data segment; determining the maximum value of the observed data based on the sum of the mean and the product of each initial training data segment; and determining the observed data in each initial training data segment that are outside the minimum and maximum values ​​of the observed data, thereby obtaining outlier data in the initial training data.

[0138] Among them, the adjustment parameters can be set based on the experience of model training, that is, they can be adjusted and optimized based on the loss value.

[0139] For example, each initial training data segment is represented as The mean value corresponding to each initial training data segment is expressed as: The standard deviation of each initial training data segment is expressed as: The adjustment parameter is expressed as The minimum value of the initial training data segment is represented as The maximum value of the initial training data segment is represented as The formula for calculating the minimum value of the initial training data segment is expressed as: The formula for calculating the maximum value of the initial training data segment is expressed as: .

[0140] In this embodiment, abnormal data is identified from the initial training data and removed to obtain training data. This avoids the interference of abnormal data caused by noise in the initial training data on the model training process, and ensures that a target prediction model with high accuracy in photovoltaic power prediction can be trained based on the training data after removing abnormal data.

[0141] In an exemplary embodiment, the above-described process of removing outlier data from the initial training data to obtain training data includes:

[0142] Outlier data is removed from the initial training data to obtain the initial training data after removing outlier data; the missing values ​​in the initial training data after removing outlier data are filled by the K-Nearest Neighbors (KNN) algorithm to obtain the training data.

[0143] In this embodiment, since there may be continuous abnormal data in the initial training data, if a simple interpolation method is used to fill in the missing values, the filled values ​​will have a large error. Therefore, the KNN algorithm is used to predict the missing values ​​using the initial training data after removing the abnormal data, so as to replace the missing values ​​with the predicted filled values. This can significantly improve the data quality of the training data, thereby enabling the target prediction model to have high accuracy in photovoltaic power prediction.

[0144] In one exemplary embodiment, such as Figure 3 As shown, the prediction module 306 includes a feature fusion layer 3062 and a fully connected layer 3064; the above-mentioned prediction module 306 predicts photovoltaic power based on temporal and non-temporal features to obtain the first predicted photovoltaic power, including:

[0145] The feature fusion layer 3062 is used to fuse temporal features and non-temporal features to obtain fused features.

[0146] Based on the fusion characteristics, the photovoltaic power is predicted using the fully connected layer 3064 to obtain the first predicted photovoltaic power.

[0147] For the fully connected layer 3064, the input data is the fused features, and the output data is the first predicted photovoltaic power.

[0148] In an exemplary embodiment, the above-described feature fusion of temporal and non-temporal features to obtain fused features includes:

[0149] The first weight is determined based on non-temporal features;

[0150] The second weight is determined based on the first weight;

[0151] Based on the first weight, temporal features, second weight, and non-temporal features, a fusion feature is obtained; the first weight corresponds to the temporal features, and the second weight corresponds to the non-temporal features.

[0152] The first weight refers to the attention weight learned based on non-temporal features.

[0153] Optionally, the sum of the first weight and the second weight is 1, that is, the second weight = 1 - the first weight.

[0154] Optionally, the fused feature can be the weighted sum of the second weight, the temporal feature, the first weight, and the non-temporal feature. That is, it can be expressed as: fused feature = first weight * temporal feature + second weight * non-temporal feature.

[0155] Alternatively, the first weight can be determined based on non-temporal features using the Sigmoid function.

[0156] For example, the non-temporal characteristics of the t-th time node are represented as: The temporal characteristics of time node t are represented as follows: The fusion features are represented as The first weight is represented as If the first weight is determined based on non-temporal features using the Sigmoid function, then the first weight can be further expressed as: Based on this, in order to map the first weight to temporal features and the second weight to non-temporal features, the fused features can be represented by the following formula: .

[0157] In this embodiment, a fused feature is obtained based on a first weight, a temporal feature, a second weight, and a non-temporal feature. The first weight corresponds to the temporal feature, and the second weight corresponds to the non-temporal feature. Thus, attention weights are used to dynamically adjust the feature weights corresponding to the temporal and non-temporal features, so that the fused feature can deeply explore the feature characteristics of both temporal and non-temporal features. It also ensures that the temporal and non-temporal features can form effective information complementarity, thereby ensuring that the target prediction model trained in the end has high accuracy in photovoltaic power prediction.

[0158] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0159] Based on the same inventive concept, this application also provides a photovoltaic power prediction device for implementing the photovoltaic power prediction method described above. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations in one or more photovoltaic power prediction device embodiments provided below can be found in the limitations of the photovoltaic power prediction method described above, and will not be repeated here.

[0160] In one exemplary embodiment, such as Figure 4 As shown, a photovoltaic power prediction device is provided, including: an acquisition module 402, a first determination module 404, a second determination module 406, and a third determination module 408, wherein:

[0161] The acquisition module 402 is used to acquire training data, which includes multiple photovoltaic training parameters and multiple training photovoltaic powers. The photovoltaic training parameters include a first photovoltaic parameter and a second photovoltaic parameter. The second photovoltaic parameter is a photovoltaic parameter that is continuous with the first photovoltaic parameter in time and is later than the first photovoltaic parameter. The training photovoltaic power is the actual photovoltaic power within the time corresponding to the second photovoltaic parameter.

[0162] The first determining module 404 is used to input the first photovoltaic training parameters into the initial prediction model to obtain the first predicted photovoltaic power; the first photovoltaic training parameters are one of a plurality of photovoltaic training parameters, and the first predicted photovoltaic power is the photovoltaic power within the time period corresponding to the predicted second photovoltaic parameters.

[0163] The second determining module 406 is used to determine the loss value based on the first predicted photovoltaic power and the first photovoltaic power; the first photovoltaic power is one photovoltaic power that corresponds to the first photovoltaic training parameter among multiple training photovoltaic powers.

[0164] The third determining module 408 is used to optimize the parameters of the initial prediction model based on the loss value to obtain the target prediction model.

[0165] In an exemplary embodiment, the initial prediction model includes a time-series module, a non-time-series module, and a prediction module; the first determining module 404 is specifically used to use the time-series module to extract features from the first photovoltaic parameters in the first photovoltaic training parameters to obtain time-series features; use the non-time-series module to extract features from the second photovoltaic parameters in the first photovoltaic training parameters to obtain non-time-series features; and use the prediction module to predict photovoltaic power based on the time-series and non-time-series features to obtain the first predicted photovoltaic power.

[0166] In an exemplary embodiment, the time-series module includes a time-series feature extraction layer; the first determining module 404 is specifically used to use the time-series feature extraction layer to extract features from the first photovoltaic parameters in the first photovoltaic training parameters to obtain time-series features.

[0167] In an exemplary embodiment, the timing module further includes a decomposition layer; the first determining module 404 is specifically used to perform variational mode decomposition on the first photovoltaic parameters using the decomposition layer to obtain multiple intrinsic mode functions; the values ​​of the multiple intrinsic mode functions satisfy preset value conditions; and the timing feature extraction layer is used to extract features from the multiple intrinsic mode functions to obtain timing features.

[0168] In an exemplary embodiment, the timing module further includes a dynamic window layer; the first determining module 404 is specifically used to use the dynamic window layer to filter the data of the first photovoltaic training parameters to obtain partial training data, wherein the first photovoltaic parameters include partial training data; and to use the decomposition layer to perform variational mode decomposition on the partial training data to obtain multiple intrinsic mode functions.

[0169] In an exemplary embodiment, the acquisition module 402 is specifically used to acquire initial training data; divide the initial training data into multiple initial training data segments according to a preset time window length; determine the mean and standard deviation of each initial training data segment within its corresponding preset time window length; determine abnormal data in the initial training data based on each initial training data segment, the mean of each initial training data segment, and the standard deviation of each initial training data segment; and remove abnormal data from the initial training data to obtain training data.

[0170] In an exemplary embodiment, the prediction module includes a feature fusion layer and a fully connected layer; the first determination module 404 is specifically used to use the feature fusion layer to perform feature fusion on time-series features and non-time-series features to obtain fused features; based on the fused features, the fully connected layer is used to predict photovoltaic power to obtain a first predicted photovoltaic power.

[0171] In an exemplary embodiment, the first determining module 404 is specifically used to determine a first weight based on non-temporal features; determine a second weight based on the first weight; and obtain a fused feature based on the first weight, temporal features, second weight, and non-temporal features; the first weight corresponds to the temporal features, and the second weight corresponds to the non-temporal features.

[0172] In an exemplary embodiment, the third determining module 408 is further configured to acquire data to be predicted; based on the data to be predicted, the photovoltaic power is predicted using a target prediction model to obtain the target photovoltaic power.

[0173] In an exemplary embodiment, the third determining module 408 is specifically used to acquire historical weather variables, historical photovoltaic module temperature, historical photovoltaic power, and predicted weather variables within a preset historical time period; determine historical theoretical irradiance based on historical weather variables; determine predicted theoretical irradiance based on historical weather variables and predicted weather variables; determine predicted photovoltaic module temperature based on historical weather variables, predicted weather variables, static heat loss coefficient, wind cooling coefficient, and the photovoltaic module's absorption efficiency of solar radiation energy; and determine historical weather variables, historical theoretical irradiance, historical photovoltaic module temperature, historical photovoltaic power, predicted weather variables, predicted theoretical irradiance, and predicted photovoltaic module temperature as data to be predicted.

[0174] Each module in the aforementioned photovoltaic power prediction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0175] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 5 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores training data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a photovoltaic power prediction method.

[0176] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 6As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When executed by the processor, the computer program implements a photovoltaic power prediction method. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0177] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0178] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0179] In one exemplary embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above-described method embodiments.

[0180] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0181] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0182] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0183] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0184] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A photovoltaic power prediction method, characterized in that, The method includes: Acquire training data, which includes multiple photovoltaic training parameters and multiple training photovoltaic powers; the photovoltaic training parameters include a first photovoltaic parameter and a second photovoltaic parameter; the second photovoltaic parameter is a photovoltaic parameter that is temporally continuous with the first photovoltaic parameter but later in time; the training photovoltaic power is the actual photovoltaic power within the time corresponding to the second photovoltaic parameter; the second photovoltaic parameter does not include a photovoltaic power parameter; The first photovoltaic training parameter is input into the initial prediction model to obtain the first predicted photovoltaic power; the first photovoltaic training parameter is one of a plurality of photovoltaic training parameters, and the first predicted photovoltaic power is the photovoltaic power predicted within the time period corresponding to the second photovoltaic parameter; The loss value is determined based on the first predicted photovoltaic power and the first photovoltaic power; the first photovoltaic power is one of the multiple trained photovoltaic powers that corresponds to the first photovoltaic training parameter. The parameters of the initial prediction model are optimized based on the loss value to obtain the target prediction model; The method further includes: The system acquires historical weather variables, historical photovoltaic module temperatures, historical photovoltaic power, and predicted weather variables for a preset future time period. The preset future time period is temporally continuous with the preset historical time period, but later than the preset historical time period. Based on the historical weather variables, the system determines the historical theoretical irradiance. Based on the predicted weather variables, the system determines the predicted theoretical irradiance. Based on the historical weather variables, the predicted weather variables, the static heat loss coefficient, the wind cooling coefficient, and the photovoltaic module's absorption efficiency of solar radiation, the system determines the predicted photovoltaic module temperature. The system uses the historical weather variables, the historical theoretical irradiance, the historical photovoltaic module temperature, the historical photovoltaic power, the predicted weather variables, the predicted theoretical irradiance, and the predicted photovoltaic module temperature as data to be predicted. Based on this data, the system uses the target prediction model to predict the photovoltaic power, obtaining the target photovoltaic power.

2. The method according to claim 1, characterized in that, The initial prediction model includes a time-series module, a non-time-series module, and a prediction module; The step of inputting the first photovoltaic training parameters into the initial prediction model to obtain the first predicted photovoltaic power includes: The time-series module is used to extract features from the first photovoltaic parameters in the first photovoltaic training parameters to obtain time-series features; The non-time-series module is used to extract features from the second photovoltaic parameters in the first photovoltaic training parameters to obtain non-time-series features. Based on the time-series features and the non-time-series features, the prediction module is used to predict the photovoltaic power to obtain the first predicted photovoltaic power.

3. The method according to claim 2, characterized in that, The time-series module includes a time-series feature extraction layer; the time-series module is used to extract features from the first photovoltaic parameters in the first photovoltaic training parameters to obtain time-series features, including: The time-series feature extraction layer is used to extract features from the first photovoltaic parameters in the first photovoltaic training parameters to obtain time-series features.

4. The method according to claim 3, characterized in that, The timing module also includes a decomposition layer; The time-series feature extraction layer is used to extract features from the first photovoltaic parameters in the first photovoltaic training parameters to obtain time-series features, including: The first photovoltaic parameter is subjected to variational mode decomposition using the decomposition layer to obtain multiple intrinsic mode functions; the values ​​of the multiple intrinsic mode functions satisfy preset value conditions; The temporal feature extraction layer is used to extract features from the multiple intrinsic mode functions to obtain temporal features.

5. The method according to claim 4, characterized in that, The timing module also includes a dynamic window layer; The first photovoltaic parameters are subjected to variational mode decomposition using the decomposition layer to obtain multiple intrinsic mode functions, including: The dynamic window layer is used to filter the data of the first photovoltaic training parameters to obtain partial training data. The first photovoltaic training parameters include partial training data. The decomposition layer is used to perform variational mode decomposition on the portion of the training data to obtain multiple intrinsic mode functions.

6. The method according to claim 1, characterized in that, The acquisition of training data includes: Obtain initial training data; The initial training data is divided into multiple initial training data segments according to the preset time window length. Determine the mean and standard deviation of each initial training data segment within its respective preset time window length; Based on each of the initial training data segments, the mean of each of the initial training data segments, and the standard deviation of each of the initial training data segments, abnormal data in the initial training data are determined. The abnormal data is removed from the initial training data to obtain the training data.

7. The method according to claim 2, characterized in that, The prediction module includes a feature fusion layer and a fully connected layer; based on the time-series features and the non-time-series features, the prediction module is used to predict photovoltaic power to obtain a first predicted photovoltaic power, including: The feature fusion layer is used to fuse the temporal features and the non-temporal features to obtain fused features. Based on the fusion characteristics, the photovoltaic power is predicted using the fully connected layer to obtain the first predicted photovoltaic power.

8. The method according to claim 7, characterized in that, The temporal features and the non-temporal features are fused to obtain fused features, including: Based on the aforementioned non-temporal features, a first weight is determined; Based on the first weight, determine the second weight; The fused feature is obtained based on the first weight, the temporal feature, the second weight, and the non-temporal feature; the first weight corresponds to the temporal feature, and the second weight corresponds to the non-temporal feature.

9. The method according to any one of claims 1-8, characterized in that, The step of optimizing the parameters of the initial prediction model based on the loss value to obtain the target prediction model includes: The target prediction model is obtained by optimizing the parameters of the initial prediction model based on the target loss function and the loss value.

10. The method according to claim 9, characterized in that, The target loss function is a modified linear unit function.

Images