A method, system, device and storage medium for joint prediction of new energy in power market based on SA-GAN

Abstract

The present invention belongs to the technical field of new energy prediction, and discloses a method and system for joint prediction of new energy in the power market based on SA-GAN, the method comprising: obtaining historical data of the power market; adding the attention mechanism to the generator network and the discriminator network in the generative adversarial network respectively, and establishing an attention-generation adversarial prediction model; inputting historical data of the power market into the attention-generation adversarial prediction model for training, and continuously optimizing the parameters of the attention-generation adversarial prediction model through adversarial learning of the generator and the discriminator; and performing joint prediction using the attention-generation adversarial prediction model after training optimization to obtain future power generation data of new energy. The present invention effectively solves the limitations of traditional methods in complex and nonlinear data processing, realizes joint prediction of multivariate data, fully considers the interaction between variables, and generates more comprehensive and accurate prediction results.

Description

Technical Field

[0001] The present invention relates to the field of new energy forecasting technology, and in particular to a SA-GAN-based new energy joint forecasting method and system for the power market. Background Art

[0002] With the development of the Global Energy Internet, the operating mechanisms of the power market are becoming increasingly complex. The behavior of market participants and the uncertainty of renewable energy generation have increased the difficulty of power system operation. Renewable energy generation in the power market is a key parameter for power system scheduling and market operations. Accurately predicting this data is crucial to ensuring the stable operation and economic benefits of the power system. Traditional power market forecasting methods are mostly based on statistics and time series analysis, such as the autoregressive integrated moving average (ARIMA) model and the generalized autoregressive conditional heteroskedasticity (GARCH) model. These methods perform well when dealing with linear relationships and stable time series, but their forecasting accuracy and adaptability are often insufficient when faced with nonlinear and highly volatile data in the power market. In addition, traditional methods are generally only able to handle single-variable forecasts and have difficulty in simultaneously considering the interactions between multiple variables, which limits the improvement of forecasting results.

[0003] With the advancement of computing power and the increase in data volumes, deep learning technology has been widely used in renewable energy forecasting in power markets. Neural network-based forecasting methods, such as long short-term memory (LSTM) networks and convolutional neural networks (CNN), can capture complex nonlinear relationships in data and have achieved good results in single-variable forecasting. However, these methods mostly focus on single-variable forecasting and struggle to achieve joint forecasting of multivariable renewable energy data, such as wind and solar power. This limits the comprehensiveness and accuracy of forecast results. Summary of the Invention

[0004] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract and title of this application to avoid obscuring the purpose of this section, the abstract and the title of the invention, and such simplifications or omissions should not be used to limit the scope of the present invention.

[0005] In view of the above existing problems, the present invention is proposed. Therefore, the present invention provides a SA-GAN-based new energy joint forecasting method for the power market to solve the problem of insufficient accuracy in the joint forecasting of renewable energy power generation in the power market.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a method for joint prediction of new energy in the power market based on SA-GAN, comprising:

[0008] Obtain historical electricity market data and preprocess it;

[0009] The attention mechanism is added to the generator network and discriminator network in the generative adversarial network respectively to establish an attention-generative adversarial prediction model;

[0010] Inputting the electricity market historical data into the attention-generation adversarial prediction model for training, and continuously optimizing the parameters of the attention-generation adversarial prediction model through adversarial learning between the generator and the discriminator;

[0011] The historical data of the electricity market is jointly predicted using the trained and optimized attention-generation adversarial prediction model to obtain future power generation data of new energy.

[0012] As a preferred solution of the SA-GAN-based power market new energy joint forecasting method described in the present invention, the generator network that adds the attention mechanism to the generative adversarial network includes:

[0013] The generator network adopts a deep residual network structure, combined with self-attention layers and fully connected layers, and introduces a dynamic weight adjustment mechanism to adaptively adjust the depth and width of the deep residual network structure according to the distribution characteristics of historical data in the power market;

[0014] The number of self-attention layer heads of the generator network is set to 8, the dimension of the attention head is set to 64, the number of fully connected layers is set to 3, the number of neurons in each layer is set to 256, 128 and 64, and the activation function is ReLU.

[0015] As a preferred solution of the SA-GAN-based power market new energy joint forecasting method described in the present invention, the discriminator network of adding the attention mechanism to the generative adversarial network includes:

[0016] The discriminator network consists of multi-scale convolutional layers and self-attention layers, which captures micro- and macro-patterns of historical electricity market data through multi-scale feature extraction;

[0017] The number of heads of the self-attention layer of the discriminator network is set to 8, the dimension of the attention head is set to 64, the convolution kernel size of the convolution layer is set to 3×3, and the number of convolution kernels is set to 32, 64, and 128.

[0018] As a preferred solution of the SA-GAN-based power market new energy joint forecasting method of the present invention, the inputting of the power market historical data into the attention-generation adversarial prediction model for training includes:

[0019] Initialize the parameters of the generator and discriminator. In each training iteration, use the historical electricity market data to train the discriminator network so that the discriminator network can distinguish between the historical electricity market data and the data samples generated by the generator network.

[0020] After training the discriminator network using historical electricity market data, the data samples generated by the generator network are input into the discriminator network together with the historical electricity market data, and the parameters of the generator network are updated according to the recognition results of the discriminator network.

[0021] As a preferred solution of the SA-GAN-based power market new energy joint forecasting method of the present invention, it also includes:

[0022] During the training process of the attention-generation adversarial prediction model, an adaptive learning rate adjustment strategy is adopted, combined with a quantum-inspired algorithm to optimize the parameters of the attention-generation adversarial prediction model;

[0023] The introduction of sparse connections reduces the complexity of the attention-generation adversarial prediction model and improves generalization ability.

[0024] As a preferred solution of the SA-GAN-based power market new energy joint forecasting method described in the present invention, the future power generation data of new energy includes:

[0025] The predicted future power generation data of new energy sources are evaluated, and the mean square error and mean absolute error indicators are used to evaluate the performance of the attention-generative adversarial prediction model. Based on the evaluation results, the attention-generative adversarial prediction model is optimized and adjusted.

[0026] As a preferred solution of the SA-GAN-based power market new energy joint forecasting method described in the present invention, the optimization and adjustment of the attention-generation adversarial prediction model includes:

[0027] Analyze the error distribution in future power generation data of new energy and obtain error analysis results;

[0028] According to the error analysis results, the structure and parameters of the attention-generation adversarial prediction model are adjusted to reduce the prediction error;

[0029] The adjustment method is a trial-and-error method, which uses the trial-and-error method to try possible combinations multiple times to obtain the optimal adjustment solution for the attention-generation adversarial prediction model;

[0030] The adjusted attention-generation adversarial prediction model is retrained until the prediction results reach the expected accuracy.

[0031] In a second aspect, the present invention provides a system for joint forecasting of new energy in the power market based on SA-GAN, comprising:

[0032] The data acquisition module is used to obtain historical data of the electricity market and perform preprocessing; the model prediction module is used to add the attention mechanism to the generator network and the discriminator network in the generative adversarial network respectively, and establish an attention-generation adversarial prediction model; the optimization module is used to input the historical data of the electricity market into the attention-generation adversarial prediction model for training, and continuously optimize the parameters of the attention-generation adversarial prediction model through adversarial learning between the generator and the discriminator; the joint prediction module is used to jointly predict the historical data of the electricity market using the trained and optimized attention-generation adversarial prediction model to obtain future power generation data of new energy.

[0033] In a third aspect, the present invention provides an electronic device, comprising:

[0034] memory and processor;

[0035] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the SA-GAN-based electricity market new energy joint forecasting method are implemented.

[0036] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the SA-GAN-based electricity market new energy joint forecasting method.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] 1. This invention significantly improves the forecasting accuracy of renewable energy data in the power market by introducing a self-attention mechanism. The self-attention mechanism captures the global characteristics and long-range dependencies of the data within the prediction model, enabling the model to not only focus on local features but also understand the inherent connections across the entire data sequence. This mechanism effectively addresses the limitations of traditional methods in processing complex, nonlinear data. In the power market, data such as renewable energy generation often exhibits significant temporal and spatial correlations. The self-attention mechanism can more accurately capture these correlations, improving the model's overall forecasting performance.

[0039] 2. Traditional forecasting methods typically focus on predicting a single variable, while ignoring the interactions and influences between different variables. This paper combines self-attention with a generative adversarial network (SA-GAN) to achieve joint forecasting of multivariate data, such as renewable energy power generation. By simultaneously forecasting multiple variables, the model can fully consider the interrelationships between them, resulting in more comprehensive and accurate forecasts. This multivariate joint forecasting not only improves the prediction accuracy of a single variable but also better reflects the combined impact of various factors in the power market, facilitating intelligent scheduling and optimized operation of the power system.

[0040] 3. Generative Adversarial Networks (GANs) can effectively learn and simulate the complex characteristics of data distributions through adversarial training of generators and discriminators. This invention incorporates a self-attention mechanism, enabling the generator to focus on the global information of the data when generating samples, thereby generating samples that more closely approximate the real data distribution. This approach not only improves the model's performance on training data, but also enhances the model's generalization ability on unknown data. Specifically, by continuously adjusting the parameters of the generator and discriminator, the generated prediction samples can maintain high credibility and accuracy in various power market environments, making them suitable for different power market scenarios.

[0041] 4. The data in the electricity market usually has large volatility and uncertainty. When faced with these complex data, traditional prediction methods are prone to large fluctuations in prediction results. The present invention uses the design of a self-attention generative adversarial network to enable the model to converge more stably during the training process, thereby improving the stability of the prediction results. The self-attention mechanism enables the model to better handle noise and outliers in the data and reduce prediction errors by calculating the dependencies between different positions. In addition, through the adversarial learning of the generator and the discriminator, the model can maintain a high level of robustness when faced with new data, reducing the prediction uncertainty caused by data fluctuations, thereby providing more stable and reliable prediction results for the scheduling and optimization of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0043] Figure 1 This is a schematic diagram of the overall process of a SA-GAN-based new energy joint forecasting method for the power market according to an embodiment of the present invention;

[0044] Figure 2 This is a schematic diagram of the framework of a SA-GAN-based new energy joint forecasting system for the power market according to an embodiment of the present invention;

[0045] Figure 3 This is a diagram illustrating the practical application of a SA-GAN-based method for joint forecasting of new energy in the power market according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0047] Example 1, reference Figure 1~Figure 2 , which is an embodiment of the present invention, provides a new energy joint forecasting method for the power market based on SA-GAN, including:

[0048] S100: Obtain historical electricity market data and perform preprocessing.

[0049] In the embodiment of the present application, the electricity market historical data includes data such as installed capacity, weather conditions, and renewable energy power generation.

[0050] After obtaining the historical data of the electricity market, preprocessing is performed, including data cleaning, missing value filling, data normalization and other operations to ensure the quality and consistency of the data.

[0051] S102: Add the attention mechanism to the generator network and discriminator network in the generative adversarial network respectively to establish an attention-generative adversarial prediction model.

[0052] Preferably, the generator network adopts a deep residual network structure, combined with a self-attention layer and a fully connected layer, and introduces a dynamic weight adjustment mechanism to adaptively adjust the depth and width of the deep residual network structure according to the historical data distribution characteristics of the electricity market.

[0053] Specifically, the generator network is initialized and the parameters are set as follows: the number of self-attention layer heads of the generator network is set to 8, the dimension of the attention head is set to 64, the number of fully connected layers is set to 3, the number of neurons in each layer is set to 256, 128 and 64, and the activation function is ReLU.

[0054] The deep residual network structure is constructed through residual blocks. Each residual block contains two convolutional layers and passes the input directly to the output through the residual connection, thereby retaining the original information and allowing gradient information to flow, thereby alleviating the gradient disappearance problem in deep networks.

[0055] After each residual block of the deep residual network, a self-attention layer is integrated. This layer assigns weights to each feature by calculating the correlation between input features, enabling the generator to capture and emphasize features in the data that have a greater impact on the prediction task, thereby improving the model's understanding of historical data patterns in the electricity market.

[0056] The output of the self-attention layer is then fed into the fully connected layer, which is responsible for mapping the features to a higher or lower dimensional space, performing nonlinear transformations, further refining the features, and providing the final feature representation for the output of the generator network.

[0057] In an embodiment of the present application, in order to enable the generator network to adaptively adjust according to the distribution characteristics of historical data of the electricity market, a dynamic weight adjustment mechanism is introduced to adaptively adjust the learning rate, number of network layers, and number of neurons to better adapt to data characteristics and improve prediction performance.

[0058] Preferably, the discriminator network includes multi-scale convolutional layers and self-attention layers to capture micro and macro patterns of historical electricity market data through multi-scale feature extraction.

[0059] In an embodiment of the present application, the number of heads of the self-attention layer of the discriminator network is set to 8, the dimension of the attention head is set to 64, the convolution kernel size of the convolution layer is set to 3×3, and the number of convolution kernels is set to 32, 64 and 128.

[0060] Specifically, the discriminator receives data generated by the generator, uses multi-scale convolutional layers to process the data in parallel, and analyzes the data from multiple angles by setting different numbers of convolution kernels 32, 64, and 128 to capture micro and macro patterns; each convolution layer uses a 3×3 convolution kernel, which is specially designed to capture the fine features of the data.

[0061] The self-attention layer then processes the output of the multi-scale convolutional layer. Each self-attention layer contains 8 heads, each with a dimension of 64, to calculate the correlation between features and assign weights. The self-attention layer can identify and emphasize the features that are most critical to the discrimination task, thereby improving the discrimination accuracy; the output of the self-attention layer is input into the fully connected layer for nonlinear transformation to extract the features that are critical to the discrimination decision.

[0062] The discriminator network makes judgments based on the integrated features and outputs a similarity score between the generated data and the real data. The similarity score is used to guide the optimization of the generator network. Through adversarial learning, the parameters of the generator network are continuously adjusted to generate more realistic data samples.

[0063] It should be noted that in the attention-generative adversarial prediction model of this invention, the discriminator network design integrates multi-scale convolutional layers and self-attention layers to enhance in-depth understanding and prediction capabilities of historical electricity market data. By employing multi-scale convolutional layers, the discriminator can process features of different scales in parallel, using different numbers of convolution kernels (32, 64, and 128) to analyze data from multiple perspectives, effectively capturing both microscopic details and macroscopic trends in historical electricity market data. Each convolutional layer uses a 3×3 convolution kernel, specifically designed to capture fine-grained features of the data, enhancing understanding of complex data structures.

[0064] The integration of self-attention layers further enhances the discriminator's performance. Each self-attention layer contains eight heads, each with a dimension of 64. By calculating the correlation between features and assigning weights, the discriminator can identify and emphasize the features most critical to the task, significantly improving accuracy. The fully connected layer then performs a nonlinear transformation on the output of the self-attention layer to extract the features that are crucial for the discriminative decision.

[0065] Furthermore, the discriminator network's dynamic weight adjustment mechanism allows the network to adaptively adjust its learning rate, number of network layers, and number of neurons based on the distribution characteristics of historical electricity market data, to better adapt to data characteristics and improve prediction performance. This design enables the discriminator network to accurately assess the similarity between data samples generated by the generator and real data, thereby continuously optimizing the generator network's parameters through adversarial learning and generating more realistic data samples.

[0066] It should also be noted that traditional discriminators cannot fully capture the multi-scale characteristics of data and may have limitations when processing complex data structures; the present invention improves the discriminator's understanding and prediction capabilities of historical electricity market data through the combination of multi-scale convolutional layers and self-attention layers, as well as a dynamic weight adjustment mechanism, thereby improving the accuracy and reliability of renewable energy power generation forecasts.

[0067] S104: Input the historical data of the electricity market into the attention-generation adversarial prediction model for training, and continuously optimize the parameters of the attention-generation adversarial prediction model through adversarial learning between the generator and the discriminator.

[0068] Preferably, the parameters of the generator and the discriminator are initialized, and in each training iteration, the discriminator network is trained using the electricity market historical data so that the discriminator network can distinguish between the electricity market historical data and the data samples generated by the generator network.

[0069] In this embodiment of the present application, the number of training rounds is set to 10,000 rounds, and the batch size used in each round is set to 64.

[0070] Preferably, after the discriminator network is trained using the electricity market historical data, the data samples generated by the generator network are input into the discriminator network together with the electricity market historical data, and the parameters of the generator network are updated according to the recognition results of the discriminator network.

[0071] Preferably, during the training process of the attention-generation adversarial prediction model, an adaptive learning rate adjustment strategy is adopted, and a quantum heuristic algorithm is combined to optimize the parameters of the attention-generation adversarial prediction model.

[0072] In an embodiment of the present application, in order to further optimize the model parameters, a quantum heuristic algorithm is combined to search for a better parameter combination, such as a quantum particle swarm optimization algorithm or a quantum genetic algorithm.

[0073] Quantum-inspired algorithms exploit the parallelism and randomness of quantum computing to quickly find better solutions in a larger search space.

[0074] Specifically, the optimization process is as follows: set the initial learning rate and other hyperparameters; in each round of iteration, use the adaptive learning rate adjustment strategy to update the learning rate and model parameters based on the current gradient information and other training status, including loss value, accuracy, etc.; suspend the adaptive learning rate adjustment during the iteration cycle, and use the quantum heuristic algorithm to perform a global search for the model parameters; use the parameter combination found by the quantum heuristic algorithm as a new starting point, and continue to use the adaptive learning rate adjustment strategy for local optimization.

[0075] In the embodiment of the present application, the search process can be expressed as:

[0076] ,

[0077] in, is a better parameter combination found by the quantum heuristic algorithm, is the current parameter, is the current learning rate, is the training data.

[0078] The parameter update formula is expressed as:

[0079] ,

[0080] in, represents the model parameters at the t+1th iteration, Adam represents the update rule of the Adam optimizer, represents the gradient at the tth iteration, represents the learning rate at the tth iteration, It is a tuning factor that controls the degree of influence of the parameter combination found by the quantum heuristic algorithm on the current parameter update.

[0081] In an embodiment of the present application, when the attention-generation adversarial prediction model is initialized, the learning rate is set to 0.0002, β1 is set to 0.5, and β2 is set to 0.999.

[0082] Preferably, sparse connections are introduced to reduce the complexity of the attention-generation adversarial prediction model and improve the generalization ability.

[0083] In the attention-generation adversarial prediction model, the present invention preferably introduces a strategy of sparse connections, aiming to reduce the complexity of the model and improve the generalization ability; sparse connections make the network structure more streamlined by setting some connections in the model to zero, reducing redundant parameters, thereby reducing computational costs and storage requirements.

[0084] Furthermore, in the generator and discriminator, the present invention adopts a sparse network layer, and achieves effective control of the model complexity by adjusting the connection sparsity of the network layer.

[0085] During the training of the attention-generation adversarial prediction model, the parameters of the generator and discriminator were first initialized, and the number of training rounds was set to 10,000, with a batch size of 64 per round. At the beginning of training, the discriminator network was trained using historical electricity market data, enabling it to distinguish between historical electricity market data and data samples generated by the generator network. Subsequently, data samples generated by the generator network were input into the discriminator network along with the historical electricity market data. The parameters of the generator network were updated based on the discriminator network's recognition results. To further optimize the model parameters, quantum-inspired algorithms such as quantum particle swarm optimization or quantum genetic algorithms were incorporated to search for optimal parameter combinations.

[0086] In each iteration, an adaptive learning rate adjustment strategy (such as the Adam optimizer) is used to update the learning rate and model parameters based on the current gradient information and other training status (such as loss and accuracy). At specific iterations, the adaptive learning rate adjustment is paused, and a quantum-inspired algorithm is used to perform a global search of the model parameters. The found parameter combination is used as a new starting point for continued local optimization. This process continuously optimizes the parameters of the attention-generative adversarial prediction model to improve its predictive performance.

[0087] Through the aforementioned training process, the attention-generative adversarial prediction model is able to fully leverage information from historical electricity market data. Through adversarial learning between the generator and the discriminator, it continuously optimizes model parameters, thereby improving the accuracy of electricity market data forecasts. Incorporating a quantum-inspired algorithm for parameter optimization further enhances the model's search capabilities and optimization efficiency, enabling the model to quickly find optimal parameter combinations within a larger search space. Furthermore, the introduction of sparse connections reduces model complexity and improves its generalization capabilities, enabling it to better adapt to new data and tasks. Ultimately, the trained attention-generative adversarial prediction model can accurately predict changing trends in the electricity market, providing strong support for renewable energy generation forecasting, electricity market scheduling, and other applications.

[0088] S106: Use the trained and optimized attention-generation adversarial prediction model to jointly predict the historical data of the electricity market and obtain the future power generation data of new energy.

[0089] Preferably, the predicted future power generation data of new energy is evaluated, and the performance of the attention-generation adversarial prediction model is evaluated using mean square error and mean absolute error indicators. Based on the evaluation results, the attention-generation adversarial prediction model is optimized and adjusted.

[0090] The formulas for mean square error and average error are expressed as:

[0091] ,

[0092] ,

[0093] in, is the sample size, is the true value, is the predicted value;

[0094] Preferably, the error distribution in the future power generation data of new energy is analyzed to obtain error analysis results; based on the error analysis results, the structure and parameters of the attention-generation adversarial prediction model are adjusted to reduce the prediction error.

[0095] Preferably, the adjustment method is a trial-and-error method, which uses the trial-and-error method to try possible combinations multiple times to obtain the optimal adjustment solution for the attention-generation adversarial prediction model.

[0096] The adjusted attention-generation adversarial prediction model is retrained until the prediction results reach the expected accuracy.

[0097] In another possible embodiment, to evaluate the performance of the attention-generative adversarial prediction model, R² (coefficient of determination) can be used as an additional evaluation metric. This metric not only considers the difference between the predicted and true values ​​but also measures the improvement of the model over a simple average prediction, providing deeper insight into the model's predictive capabilities. Furthermore, in addition to traditional trial-and-error methods, gradient descent-based optimization algorithms, such as Adam and RMSprop, can be introduced for model adjustment. These algorithms can automatically adjust the learning rate based on the model's performance during training, accelerating convergence and potentially finding more optimal model parameters. The present invention does not impose strict restrictions on specific evaluation metrics and adjustment methods, but encourages flexible selection based on actual needs and data characteristics to achieve optimal prediction results and model performance.

[0098] It should be noted that the present invention realizes the joint prediction of historical data of the electricity market by training the optimized attention-generative adversarial prediction model, effectively obtains the future power generation data of new energy, and provides accurate data support for energy planning and management. By adopting mean square error and mean absolute error as evaluation indicators, the present invention can comprehensively evaluate the performance of the prediction model, and accordingly optimize and adjust the model in a targeted manner, significantly improving the prediction accuracy. Furthermore, through the analysis of the distribution of prediction errors, the present invention can deeply understand the shortcomings of the model performance, and optimize the model structure and parameters through flexible adjustment methods such as trial and error, effectively reducing the prediction error. Finally, after repeated training and adjustment, the present invention ensures that the prediction results achieve the expected accuracy, brings better technical effects to the prediction and management of new energy power generation, and improves energy utilization efficiency and economic benefits.

[0099] The above is a schematic scheme of a method for joint prediction of new energy in the power market based on SA-GAN in this embodiment. It should be noted that the technical solution of the system for joint prediction of new energy in the power market based on SA-GAN and the technical solution of the method for joint prediction of new energy in the power market based on SA-GAN above are of the same concept. For details not described in detail in the technical solution of the system for joint prediction of new energy in the power market based on SA-GAN in this embodiment, please refer to the description of the technical solution of the method for joint prediction of new energy in the power market based on SA-GAN above.

[0100] In this embodiment, the SA-GAN-based electricity market new energy joint forecasting system is as follows: Figure 2 Shown, including:

[0101] The data acquisition module 100 is used to acquire historical power market data and perform pre-processing.

[0102] The model prediction module 200 is used to add the attention mechanism to the generator network and the discriminator network in the generative adversarial network respectively, and establish an attention-generative adversarial prediction model.

[0103] The optimization module 300 is used to input the historical data of the electricity market into the attention-generation adversarial prediction model for training, and continuously optimize the parameters of the attention-generation adversarial prediction model through adversarial learning between the generator and the discriminator.

[0104] The joint prediction module 400 is used to perform joint prediction on the historical data of the electricity market using the trained and optimized attention-generative adversarial prediction model to obtain future power generation data of new energy.

[0105] This embodiment further provides an electronic device suitable for SA-GAN-based joint forecasting of new energy sources in the power market, including:

[0106] Memory and processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the SA-GAN-based new energy joint forecasting method for the power market as proposed in the above embodiment.

[0107] This embodiment also provides a storage medium having a computer program stored thereon. When the program is executed by a processor, the method for joint prediction of new energy in the power market based on SA-GAN as proposed in the above embodiment is implemented.

[0108] The storage medium proposed in this embodiment and the method for implementing the joint prediction of new energy in the power market based on SA-GAN proposed in the above embodiment belong to the same inventive concept. For technical details not described in detail in this embodiment, please refer to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0109] Through the above description of the embodiments, those skilled in the art will clearly understand that the present invention can be implemented using software and necessary general-purpose hardware. Of course, it can also be implemented using hardware, but in many cases the former is a more preferred embodiment. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk, or optical disk, and includes a number of instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various embodiments of the present invention.

[0110] Example 2, reference Figure 3 ,This embodiment provides a SA-GAN based joint forecast method for ,new energy in the electricity market. ,In order to verify its beneficial effects, the ,comparison results of several schemes are provided.

[0111] Comprehensive historical data on the electricity market was collected, covering key information such as installed capacity, weather patterns, and renewable energy generation. Through data cleaning, outliers and noise were removed, and missing values ​​were cleverly corrected using interpolation or mean imputation, ensuring data integrity and accuracy. Furthermore, the data was normalized to the range [0, 1], effectively eliminating the potential impact of different dimensions on data analysis and laying a solid foundation for subsequent model construction.

[0112] The self-attention generative adversarial network of the present invention is applied to perform predictions. Through comparative experiments with traditional time series prediction methods, such as the ARIMA model and the standard generative adversarial network, the performance of different models in renewable energy power generation prediction is obtained.

[0113] Using MSE and MAE as preliminary evaluation metrics, we comprehensively assess the forecasting performance of different models. To more intuitively demonstrate forecast quality, we define a method for calculating forecast accuracy, determining whether a forecast is correct based on whether the relative error between the predicted and actual values ​​is within a set threshold. Compared to traditional time series forecasting methods, such as the ARIMA model and standard generative adversarial networks, our method significantly improves forecast accuracy for photovoltaic power generation, wind power generation, and other renewable energy types.

[0114] Specifically, the proposed method achieved an accuracy of 83.5% for photovoltaic power generation forecasts, 73.8% for wind power generation forecasts, and 75.4% for other energy sources, with an average accuracy of 77.6%. This result demonstrates that the proposed method, through its unique combination of self-attention mechanism and generative adversarial networks, effectively improves its ability to understand and predict complex electricity market data.

[0115] like Figure 3 As shown, by forecasting renewable energy output over a period of time, power retailers can adjust their day-ahead reporting strategies. When forecasting high renewable energy output on a given day, they can reduce load reporting; when forecasting limited renewable energy output on a given day, they can increase load reporting; and when renewable energy output fluctuates significantly, they can report based on actual proxy load. This is expected to increase returns by 5% in the medium- and long-term markets and by 12% in the spot market.

[0116] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A new energy joint forecasting method for power market based on SA-GAN, characterized in that: include: Obtain historical electricity market data and pre-process it; The attention mechanism is added to the generator network and the discriminator network in the generative adversarial network respectively to establish an attention-generative adversarial prediction model; The generator network adopts a deep residual network structure, combined with a self-attention layer and a fully connected layer, and introduces a dynamic weight adjustment mechanism to adaptively adjust the depth and width of the deep residual network structure according to the distribution characteristics of historical data in the power market; The number of self-attention layer heads of the generator network is set to 8, the dimension of the attention head is set to 64, the number of fully connected layers is set to 3, the number of neurons in each layer is set to 256, 128 and 64, and the activation function is ReLU; The discriminator network includes multi-scale convolutional layers and self-attention layers to capture micro- and macro-patterns of historical power market data through multi-scale feature extraction; The number of heads of the self-attention layer of the discriminator network is set to 8, the dimension of the attention head is set to 64, the convolution kernel size of the convolution layer is set to 3×3, and the number of convolution kernels is set to 32, 64, and 128; Inputting the electricity market historical data into the attention-generation adversarial prediction model for training, and continuously optimizing the parameters of the attention-generation adversarial prediction model through adversarial learning between the generator and the discriminator; During the training process of the attention-generation adversarial prediction model, an adaptive learning rate adjustment strategy is adopted, and the quantum heuristic algorithm is combined to optimize the parameters of the attention-generation adversarial prediction model; Introducing sparse connections reduces the complexity of the attention-generation adversarial prediction model and improves generalization ability; The historical data of the electricity market is jointly predicted using the trained and optimized attention-generation adversarial prediction model to obtain future power generation data of new energy.

2. The SA-GAN-based new energy joint forecasting method for the power market as claimed in claim 1, characterized in that: Inputting the electricity market historical data into the attention-generation adversarial prediction model for training includes: Initialize the parameters of the generator and the discriminator. In each training iteration, use the historical data of the electricity market to train the discriminator network so that the discriminator network can distinguish the historical data of the electricity market from the data samples generated by the generator network. After training the discriminator network with the historical data of the electricity market, the data samples generated by the generator network are input into the discriminator network together with the historical data of the electricity market, and the parameters of the generator network are updated according to the recognition results of the discriminator network.

3. The SA-GAN-based new energy joint forecasting method for the power market as claimed in claim 2, characterized in that: New energy future power generation data include: The predicted future power generation data of new energy is evaluated, and the mean square error and mean absolute error indicators are used to evaluate the performance of the attention-generative adversarial prediction model. According to the evaluation results, the attention-generative adversarial prediction model is optimized and adjusted.

4. The SA-GAN-based new energy joint forecasting method for the power market as claimed in claim 3, characterized in that: Optimization adjustments to the attention-generation adversarial prediction model include: Analyze the error distribution in the future power generation data of new energy and obtain the error analysis results; According to the error analysis results, adjusting the structure and parameters of the attention-generation adversarial prediction model to reduce the prediction error; The adjustment method is a trial-and-error method, which uses the trial-and-error method to try possible combinations multiple times to obtain the optimal adjustment solution for the attention-generation adversarial prediction model; The adjusted attention-generation adversarial prediction model is retrained until the prediction results reach the expected accuracy.

5. A system for joint prediction of new energy in the power market based on SA-GAN, characterized in that: include: A data acquisition module (100), used for acquiring historical data of the power market and performing pre-processing; A model prediction module (200) is used to add the attention mechanism to the generator network and the discriminator network in the generative adversarial network respectively, and establish an attention-generative adversarial prediction model; An optimization module (300) is used to input the electricity market historical data into the attention-generation adversarial prediction model for training, and continuously optimize the parameters of the attention-generation adversarial prediction model through adversarial learning between the generator and the discriminator; The joint prediction module (400) is used to perform joint prediction on the electricity market historical data using the trained and optimized attention-generation adversarial prediction model to obtain future power generation data of new energy.

6. An electronic device, characterized in that: include: Memory and processor; The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the SA-GAN-based new energy joint prediction method for the power market as described in any one of claims 1-4 are implemented.

7. A computer-readable storage medium, characterized in that: Computer executable instructions are stored, and when the computer executable instructions are executed by a processor, the steps of the SA-GAN-based new energy joint prediction method for the power market as described in any one of claims 1-4 are implemented.

Images