New energy high-uncertainty scene generation method and device based on CWGAN-GP model

Abstract

The invention relates to the technical field of novel power system optimization operation, in particular to a new energy high-uncertainty scene generation method and device based on a CWGAN-GP model, and the method comprises the steps: correcting an initial generative adversarial network through employing a Wasserstein distance and a gradient penalty term, so as to obtain a final generative adversarial network, and the target convolutional neural network is introduced into the final generative adversarial network to obtain a conditional deep convolutional generative adversarial network of improved deep convolution, so that WGAN-GP is introduced into the conditional deep convolutional generative adversarial network of improved deep convolution to obtain a CWGAN-GP model of improved deep convolution. And a target L-ISODATA clustering algorithm is applied to cluster the generated wind and light scenes so as to generate a high-uncertainty new energy power generation output scene. Therefore, the problems that in the related technology, when the GAN processes high-dimensional data, complex space-time characteristics are difficult to effectively capture, and due to lack of prior information, the training direction is prone to deviation, and the scene generation effect and stability are affected are solved.

Description

Technical Field

[0001] The present application relates to the technical field of optimized operation of new power systems, and in particular to a method and device for generating high-uncertainty scenarios of new energy based on a CWGAN-GP model. Background Art

[0002] With the advancement of deep learning technology, especially the development of GAN (generative adversarial networks), a new and powerful tool has been provided to more accurately simulate the uncertainty of renewable energy generation. In related technologies, new samples with high similarity to the distribution law of historical data are generated through the unique game process of the internal structure to model the active output of wind and solar power and generate scenarios.

[0003] However, in related technologies, GAN has difficulty in effectively capturing complex spatiotemporal characteristics when processing high-dimensional data, and due to the lack of prior information, the training direction is easily deviated, thus affecting the scene generation effect and stability, which urgently needs to be improved. Summary of the invention

[0004] The present application provides a method and device for generating high-uncertainty scenarios of new energy based on the CWGAN-GP model, in order to solve the problems in the related art that GAN is difficult to effectively capture complex spatiotemporal characteristics when processing high-dimensional data, and the training direction is easily deviated due to the lack of prior information, thereby affecting the scene generation effect and stability.

[0005] The first aspect of the present application provides a method for generating high-uncertainty scenarios of new energy based on the CWGAN-GP model, comprising the following steps: using the Wasserstein distance and gradient penalty terms to correct the initial generative adversarial network to obtain a final generative adversarial network; introducing the target convolutional neural network into the final generative adversarial network to obtain a conditional deep convolutional generative adversarial network of improved deep convolution; introducing WGAN-GP into the conditional deep convolutional generative adversarial network of improved deep convolution to obtain a CWGAN-GP model of improved deep convolution, and applying the target L-ISODATA clustering algorithm to cluster the generated wind and solar scenes to generate a high-uncertainty new energy power generation output scenario.

[0006] The embodiment of the present application can use the Wasserstein distance and gradient penalty term to correct the initial generative adversarial network, solve the gradient vanishing problem that is easy to occur in traditional GAN ​​when measuring the similarity of generated data and real data distribution, improve the stability of the generative adversarial network, introduce the target convolutional neural network into the final generative adversarial network, and obtain a conditional deep convolutional generative adversarial network with improved deep convolution, which can better capture the complex features in the data, thereby improving the accuracy and practicality of the generated scenario, and introduce WGAN-GP into the conditional deep convolutional generative adversarial network with improved deep convolution, combining multiple technical advantages to generate samples that are closer to real data, further improving the model's adaptability and generation ability to renewable energy power generation output scenarios under different conditions.

[0007] Optionally, in one embodiment of the present application, the calculation formula of the Wasserstein distance is:

[0008]

[0009] Where W(P(x), P(z)) is the Wasserstein distance between the real data and the generated data, Sup is the minimum upper bound, and L≤1 represents the upper bound of the expected value difference of 1-Lipschitz;

[0010] The calculation formula of the gradient penalty term is:

[0011]

[0012] in, is the gradient of D(x) with respect to x, and ||.||2 is the norm.

[0013] The embodiments of the present application can use the Wasserstein distance to more accurately characterize the similarity between the two distributions of real data and generated data, and use the gradient penalty term to ensure that the weight size is within a certain range, thereby maintaining the 1-Lipschitz constraint of the function, effectively preventing gradient vanishing and gradient explosion, and enabling the model to more effectively learn the characteristics and laws of the data.

[0014] Optionally, in one embodiment of the present application, the objective function of the CWGAN-GP model of the improved deep convolution is:

[0015]

[0016] in, is the optimal parameter of the generator G that minimizes the objective function V(D,G), and the optimal parameter of the discriminator D that maximizes the objective function V(D,G), E[·] is the mathematical expectation, z is the random noise, x is the real data, x~P(x) is the data distribution of the real sample x, z~P(z) is the data distribution of the generated sample, logD(x) is the judgment of the discriminator on the real sample, and λ is the gradient penalty coefficient.

[0017] The embodiments of the present application can promote the generator and the discriminator to compete and learn from each other during the training process by setting the minimax optimization problem, and eventually tend to an equilibrium state. The generator continuously improves the quality of the generated samples to make them closer to the real samples, and the discriminator also continuously improves its discrimination ability, thereby improving the accuracy of the entire model.

[0018] Optionally, in one embodiment of the present application, the application target L-ISODATA clustering algorithm is used to cluster the generated scenery scenes, including: calculating the kernel function between each sample in the sample set and storing them in a kernel function matrix; during the calculation process of the clustering algorithm, querying the corresponding kernel function from the kernel function matrix for clustering.

[0019] The embodiment of the present application can avoid repeatedly calculating the kernel function in each calculation process of the clustering algorithm by calculating the kernel function between each sample in the sample set and storing it in the kernel function matrix.

[0020] The overall computational efficiency of the clustering algorithm is improved. The corresponding kernel function is queried from the kernel function matrix for clustering, ensuring that the kernel function used in the entire clustering process is consistent, avoiding errors that may be caused by multiple calculations of the kernel function, and improving the accuracy and stability of clustering calculations.

[0021] The second aspect of the present application provides a new energy high-uncertainty scenario generation device based on the CWGAN-GP model, including: a first generation module, used to use Wasserstein distance and gradient penalty terms to correct the initial generation adversarial network to obtain a final generation adversarial network; a second generation module, used to introduce the target convolutional neural network into the final generation adversarial network to obtain a conditional deep convolution generation adversarial network of improved deep convolution; a third generation module, used to introduce WGAN-GP into the conditional deep convolution generation adversarial network of improved deep convolution to obtain a CWGAN-GP model of improved deep convolution, and apply the target L-ISODATA clustering algorithm to cluster the generated wind and solar scenes to generate a high-uncertainty new energy power generation output scenario.

[0022] The embodiment of the present application can use the Wasserstein distance and gradient penalty term to correct the initial generative adversarial network, solve the gradient vanishing problem that is easy to occur in traditional GAN ​​when measuring the similarity of generated data and real data distribution, improve the stability of the generative adversarial network, introduce the target convolutional neural network into the final generative adversarial network, and obtain a conditional deep convolutional generative adversarial network with improved deep convolution, which can better capture the complex features in the data, thereby improving the accuracy and practicality of the generated scenario, and introduce WGAN-GP into the conditional deep convolutional generative adversarial network with improved deep convolution, combining multiple technical advantages to generate samples that are closer to real data, further improving the model's adaptability and generation ability to renewable energy power generation output scenarios under different conditions.

[0023] Optionally, in one embodiment of the present application, the calculation formula of the Wasserstein distance is:

[0024]

[0025] Where W(P(x), P(z)) is the Wasserstein distance between the real data and the generated data, Sup is the minimum upper bound, and L≤1 represents the upper bound of the expected value difference of 1-Lipschitz;

[0026] The calculation formula of the gradient penalty term is:

[0027]

[0028] in, is the gradient of D(x) with respect to x, and ||.||2 is the norm.

[0029] The embodiments of the present application can use the Wasserstein distance to more accurately characterize the similarity between the two distributions of real data and generated data, and use the gradient penalty term to ensure that the weight size is within a certain range, thereby maintaining the 1-Lipschitz constraint of the function, effectively preventing gradient vanishing and gradient explosion, and enabling the model to more effectively learn the characteristics and laws of the data.

[0030] Optionally, in one embodiment of the present application, the objective function of the CWGAN-GP model of the improved deep convolution is:

[0031]

[0032] in, is the optimal parameter of the generator G that minimizes the objective function V(D,G), and the optimal parameter of the discriminator D that maximizes the objective function V(D,G), E[·] is the mathematical expectation, z is the random noise, x is the real data, x~P(x) is the data distribution of the real sample x, z~P(z) is the data distribution of the generated sample, logD(x) is the judgment of the discriminator on the real sample, and λ is the gradient penalty coefficient.

[0033] The embodiments of the present application can promote the generator and the discriminator to compete and learn from each other during the training process by setting the minimax optimization problem, and eventually tend to an equilibrium state. The generator continuously improves the quality of the generated samples to make them closer to the real samples, and the discriminator also continuously improves its discrimination ability, thereby improving the accuracy of the entire model.

[0034] Optionally, in one embodiment of the present application, the third generation module includes: a first calculation unit, used to calculate the kernel function between each sample in the sample set and store them in a kernel function matrix; a clustering unit, used to query the corresponding kernel function from the kernel function matrix for clustering during the calculation process of the clustering algorithm.

[0035] The embodiment of the present application can avoid repeatedly calculating the kernel function in each calculation process of the clustering algorithm by calculating the kernel function between each sample in the sample set and storing it in the kernel function matrix.

[0036] The overall computational efficiency of the clustering algorithm is improved. The corresponding kernel function is queried from the kernel function matrix for clustering, ensuring that the kernel function used in the entire clustering process is consistent, avoiding errors that may be caused by multiple calculations of the kernel function, and improving the accuracy and stability of clustering calculations.

[0037] The third aspect of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for generating high-uncertainty scenarios of new energy based on the CWGAN-GP model as described in the above embodiment.

[0038] The fourth aspect embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the program is executed by a processor, it implements the above-mentioned method for generating high-uncertainty scenarios of new energy based on the CWGAN-GP model.

[0039] The fifth aspect of the present application provides a computer program product, including a computer program, which, when executed, is used to implement the above-mentioned method for generating new energy high-uncertainty scenarios based on the CWGAN-GP model.

[0040] The embodiment of the present application can use Wasserstein distance and gradient penalty terms to correct the initial generative adversarial network, solve the gradient vanishing problem of traditional GAN, improve network stability, and at the same time, more accurately characterize distribution similarity, maintain function constraints, prevent gradient problems, and enable the model to learn data feature laws more effectively; secondly, the conditional deep convolutional generative adversarial network with improved deep convolution obtained by introducing the target convolutional neural network can better capture complex features and improve the accuracy and practicality of the generated scenarios. After introducing WGAN-GP, it combines multiple advantages to generate samples that are closer to real data, and enhances the model's adaptability and generation ability to renewable energy power generation output scenarios under different conditions; in addition, by setting the minimax optimization problem, the competitive learning of the generator and the discriminator is prompted to tend to equilibrium, thereby improving the accuracy of the model. In terms of clustering, by pre-calculating and storing the kernel function, repeated calculations are avoided, the overall computational efficiency of the clustering algorithm is improved, and the consistency of the kernel function is ensured, thereby improving the accuracy and stability of the clustering calculation.

[0041] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0043] Figure 1 A flowchart of a method for generating a high-uncertainty scenario of new energy based on a CWGAN-GP model according to an embodiment of the present application;

[0044] Figure 2 A schematic diagram of a CWGAN-GP network structure with improved convolution according to a specific embodiment of the present application;

[0045] Figure 3 It is a flowchart of a method for generating a new energy high uncertainty scenario based on a CWGAN-GP model according to a specific embodiment of the present application;

[0046] Figure 4 A schematic diagram of the structure of a new energy high-uncertainty scenario generation device based on a CWGAN-GP model provided according to an embodiment of the present application;

[0047] Figure 5 The figure is a structural example diagram of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0048] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0049] The following describes the method and device for generating high-uncertainty scenarios of new energy based on the CWGAN-GP model of the embodiment of the present application with reference to the accompanying drawings. In view of the related technologies mentioned in the above background technology center, GAN is difficult to effectively capture complex spatiotemporal characteristics when processing high-dimensional data, and due to the lack of prior information, the training direction is easily deviated, thereby affecting the effect and stability of scene generation. The present application provides a method for generating high-uncertainty scenarios of new energy based on the CWGAN-GP model, in which the initial generative adversarial network can be corrected by using the Wasserstein distance and the gradient penalty term, the target convolutional neural network is introduced into the final generative adversarial network, and the conditional deep convolution generative adversarial network of improved deep convolution is obtained, and WGAN-GP is introduced into the conditional deep convolution generative adversarial network of improved deep convolution to obtain the CWGAN-GP model of improved deep convolution. Combining multiple technical advantages, the characteristics and laws of the data can be better learned during the training process, and the accuracy of generating various high-uncertainty new energy power generation output scenarios can be improved, providing more reliable support for the optimized scheduling and safe operation of the power system. 。 This solves the problem in related technologies that GAN has difficulty in effectively capturing complex spatiotemporal characteristics when processing high-dimensional data, and that the training direction is easily deviated due to the lack of prior information, thus affecting the scene generation effect and stability.

[0050] Specifically, Figure 1 A flowchart of a method for generating high-uncertainty scenarios of new energy based on the CWGAN-GP model provided in an embodiment of the present application.

[0051] like Figure 1 As shown, the new energy high uncertainty scenario generation method based on the CWGAN-GP model includes the following steps:

[0052] In step S101, the initial generative adversarial network is corrected using the Wasserstein distance and the gradient penalty term to obtain the final generative adversarial network.

[0053] It is understandable that the GAN (generative adversarial networks) structure consists of two interacting neural networks, G (Generator) and D (Discriminator). The main function of the generator is to simulate the distribution of real data samples and generate new data samples that are as close to the real data as possible; the discriminator is committed to identifying the source of data samples and judging whether they come from actual samples or imitations made by the generator. The two networks compete against each other during the training process and eventually tend to the Nash equilibrium state, at which time the data samples generated by the generator are no different from the real samples, and the discriminator has difficulty distinguishing the authenticity of these samples.

[0054] In some embodiments, the traditional GAN ​​network structure generator (G) and discriminator (D) are designed as fully connected neural networks. The generator continuously adjusts parameters to generate new data samples through mapping, and the discriminator accepts real samples and generated samples, and outputs probability values ​​for discrimination. A value close to 1 represents a real data sample, and a value close to 0 represents a generated sample. Through mutual adversarial training, the accuracy is continuously improved. According to the model of the G and D networks, G receives data samples generated by random noise z that meets a specific distribution. Through D's judgment of the generated samples log(1-D(G(z)), the distribution of the generated samples is constantly close to the distribution of the real samples. The corresponding loss functions lossG and lossD are as follows:

[0055]

[0056] Where E[·] is the mathematical expectation, z is random noise, x is the real data, x~P(x) is the data distribution of the real sample x, z~P(z) is the data distribution of the generated sample, and logD(x) is the judgment of the discriminator on the real sample.

[0057] Based on the idea of ​​the above game, the target loss function of the GAN optimization process is given as follows:

[0058]

[0059] The game relationship between the two networks G and D can be expressed as a minimal-maximum optimization problem. Under the premise of sharing losses, G and D achieve confrontation through mutual competition.

[0060] Optionally, in one embodiment of the present application, the calculation formula of the Wasserstein distance is:

[0061]

[0062] Where W(P(x), P(z)) is the Wasserstein distance between the real data and the generated data, Sup is the minimum upper bound, and L≤1 represents the upper bound of the 1-Lipschitz expected value difference.

[0063] The calculation formula of the gradient penalty term is:

[0064]

[0065] in, is the gradient of D(x) with respect to x, and ||.||2 is the norm.

[0066] Specifically, in optimizing the GAN network, in order to avoid the vanishing gradient problem caused by the Jensen-Shannon divergence (JS) and Kullback-Leibler divergence (KL) used by traditional GAN ​​to measure the similarity between the generated data and the real data distribution, the embodiment of the present application proposes to determine a continuously available gradient based on the Wasserstein distance, which effectively solves the mode collapse problem in traditional GAN ​​training. Wasserstein GAN (WGAN) introduces the RMSProp optimization algorithm to replace the Adam gradient descent method in traditional GAN, significantly improving the stability of training. The Wasserstein distance can more accurately characterize the similarity between two distributions, so its application in WGAN helps to improve the accuracy of the generated scene; in order to ensure that the weight size is within a certain range, thereby maintaining the 1-Lipschitz constraint of the function and preventing gradient vanishing and gradient explosion, GP (Gradient Penalty) is introduced to replace the weight clipping strategy in WGAN.

[0067] The embodiment of the present application uses Wasserstein distance to solve the traditional GAN ​​gradient vanishing and mode collapse problems, improve training stability and more accurately characterize distribution similarity. The gradient penalty term maintains function constraints, prevents gradient problems and optimizes parameters, retains the GAN adversarial mechanism, promotes the performance of the generator and discriminator, optimizes the structure and loss function based on the original GAN, enables the model to stably learn data features, improves sample quality, enhances practicality and performance, and better adapts to needs.

[0068] In step S102, the target convolutional neural network is introduced into the final generative adversarial network to obtain a conditional deep convolutional generative adversarial network with improved deep convolution.

[0069] It is understandable that traditional GANs often find it difficult to effectively capture complex spatiotemporal characteristics when processing high-dimensional data. DCGAN (Deep Convolutional Generative Adversarial Network) is an improved model that introduces convolutional neural networks into the GAN structure. It makes full use of the excellent feature extraction capabilities of convolutional networks and can generate samples that are closer to real data.

[0070] In the actual data processing process, the regularized data contains negative values. Therefore, the Parametric ReLU activation function is selected. A parameter that can be dynamically adjusted through training is introduced in the gradient value of the negative semi-axis of the function, so as to dynamically optimize the network parameters and improve the training effect. In order to better understand how past outputs affect future outputs, self-attention mechanisms are added to the generator and discriminator respectively to ensure the operation speed and better context capture. Month conditional labels are added to the improved DCGAN to generate scenery scenes in any month. WGAN-GP is introduced into the conditional DCGAN of improved deep convolution, and finally the CWGAN-GP model of improved deep convolution is obtained.

[0071] Specifically, DCGAN is an improved model that introduces convolutional neural networks into the GAN structure. It makes full use of the excellent ability of convolutional networks in feature extraction and can generate more realistic data samples, thereby significantly improving the quality of generated data. In wind and solar power generation scenarios, the output characteristics are significantly dynamic and complex, so the design of DCGAN can effectively extract the complex dynamic information in the wind and solar output curve. At present, the derivative GAN network based on the GAN network, G / D generally uses the DCGAN model, so as to effectively learn the characteristics of different wind and solar scenarios and a variety of scenario types.

[0072] On the one hand, G uses transposed convolution for upsampling, and D uses stride convolution layer for downsampling. During the training process, G gradually adjusts the weights to generate data closer to the real sample, and D improves the discrimination ability through learning. In DCGAN, as the number of network layers increases, the difficulty of training increases. In order to prevent the activation function from entering the saturation zone or dead zone during training, a regularization layer is introduced after the convolution layer in the generator network to adjust the data to the standard normal distribution, thereby enhancing the training efficiency of the network. In order to output scene diversity, the regularization layer is not introduced in the output layer.

[0073] On the other hand, during the data processing, some of the regularized data contains negative values, which is unrealistic in some application scenarios such as renewable energy output, because the output value cannot be negative. The original DCGAN was originally used for image generation. In its generator, except for the output layer which is Tanh, the rest of the layers are ReLU. However, in order to prevent the data from entering the dead zone of the ReLU activation function too early and losing the training effect, the LeakyReLU activation function is used in the hidden layer to maintain a certain training potential for negative data, and the output layer is activated with a ReLU-type activation function to ensure that the output values ​​of all scenarios are non-negative. However, the gradient of the negative semi-axis of the LeakyReLU activation function is fixed (always fixed to 0.01y), so it is not very flexible and may perform poorly in some cases. The Parametric ReLU activation function introduces a parameter that can be dynamically adjusted through training in the gradient value of the function on the negative semi-axis. The analytical expressions of the two functions are as follows:

[0074]

[0075] Through the above analysis, it can be observed that the function introduces a parameter that can be dynamically adjusted through training for the gradient value of the negative semi-axis. This mechanism allows sparsity to be retained or neurons to be inhibited as needed during the training process, thereby dynamically optimizing network parameters, which helps to achieve parameter optimization of the neural network in the training stage, thereby improving the training effect. Therefore, the embodiment of the present application selects ParametricReLU as the activation function of the hidden layer of the generator, and to ensure that the final output scenery scene result is a positive value, the output layer selects ReLU for activation.

[0076] In the actual implementation process, the self-attention mechanism (Self-Attention) can help the model capture long-distance dependencies, which is very effective in understanding and simulating complex patterns in time series data. In the generation of wind and solar output scenarios, we can better understand how past outputs affect future outputs, thereby improving the overall prediction accuracy and model performance. In the embodiment of the present application, the self-attention mechanism is added to the generator and the discriminator respectively. In order to ensure the operation speed and better capture the context, the self-attention layer is added to the second convolutional layer, which can help the model to understand and integrate a wider range of contextual information at a smaller scale of the image, and help subsequent layers to maintain these contextual relationships when enlarging the image, which is beneficial for generating complex scenes. The use of self-attention in the discriminator can also improve its sensitivity to global information, thereby more accurately judging the authenticity of the generated scene.

[0077] In some cases, due to the lack of prior information, the training direction of traditional GAN ​​is easy to deviate, resulting in poor results and instability. CGAN (conditional generative adversarial networks) introduces conditional information c as part of the input of the generator G and the discriminator D, so that data of a specified category can be generated. In this model, the conditional input provides information related to the target features. The model can capture the key dynamic and spatial relationships in the wind and solar power generation scenario, so that the generator can generate samples under specific conditions more specifically. Combined with the category label information c added by the conditional GAN, the objective function of the improved conditional DCGAN is adjusted accordingly on the basis of step S102 as follows:

[0078]

[0079] From this formula, we can see that the training goal of conditional DCGAN is a two-party minimax game problem with conditional input.

[0080] In step S103, WGAN-GP is introduced into the conditional deep convolutional generative adversarial network of the improved deep convolution to obtain the CWGAN-GP model of improved deep convolution, and the target L-ISODATA clustering algorithm is applied to cluster the generated wind and solar scenes to generate a high-uncertainty new energy power generation output scenario.

[0081] It can be understood that in order to better characterize the uncertainty relationship between wind and solar output, the embodiment of the present application combines the characteristics of WGAN-GP and conditional DCGAN to propose an improved deep convolution CWGAN-GP wind and solar joint output scene generation method. The model structure is as follows: Figure 2 shown.

[0082] Optionally, in one embodiment of the present application, the objective function of the CWGAN-GP model of the improved deep convolution is:

[0083]

[0084] in, is the optimal parameter of the generator G that minimizes the objective function V(D,G), and the optimal parameter of the discriminator D that maximizes the objective function V(D,G), E[·] is the mathematical expectation, z is the random noise, x is the real data, x~P(x) is the data distribution of the real sample x, z~P(z) is the data distribution of the generated sample, logD(x) is the judgment of the discriminator on the real sample, and λ is the gradient penalty coefficient.

[0085] In the embodiment of the present application, the model引用深度卷积网络在特征提取方面的卓越能力,并改进卷积网络激活函数和加入自适应模块,能够生成更加接近于真实数据的样本,并引入梯度惩罚(GP)来代替GAN训练过程中的权值剪裁问题,以提高模型的性能,依据学习风光真实数据,通过生成器G和判别器D两者博弈过程,有效生成与真实数据无限逼近的海量数据集,同时,在训练过程中可以加入不同的标签y(如包括但不限于月份、天气等),利用CGAN的优点充分挖掘不同特征与风光出力的相关性,生成不同标签下的风光出力数据集。

[0086] Optionally, in an embodiment of the present application, applying the target L-ISODATA clustering algorithm to cluster the generated wind-solar scenarios includes: calculating the kernel function between each pair of samples in the sample set and storing it in the kernel function matrix; during the calculation of the clustering algorithm, querying the corresponding kernel function from the kernel function matrix for clustering.

[0087] It can be understood that the classical K-means clustering algorithm needs to determine the number K of clusters in advance and can only try multiple times to find the optimal k value. However, for the Iterative Selforganizing Data Analysis Techniques Algorithm (ISODATA), its k value can change during the clustering process. The L-ISODATA (Load curve-ISODATA) algorithm uses the kernel method to map the load curve to a high-dimensional feature space for clustering, enabling the capture of high-dimensional features of the load curve and improving the clustering effect and speed. In the embodiment of the present application, the L-ISODATA clustering algorithm is applied to cluster the generated wind-solar scenarios, and the probability of each typical scenario is obtained.

[0088] On the one hand, for the problem of randomly selecting the initial clustering centers, the optimization strategy adopted is as follows: when selecting the first initial clustering center, it is still selected randomly. After selecting the subsequent clustering centers, assuming that n initial clustering centers (0 < n < k) have been selected, when selecting the (n + 1)-th initial clustering center, the points farther away from the current n clustering centers have a higher probability of being selected. The clustering centers are more likely to belong to different clusters, the convergence speed of the clustering algorithm is faster, and the clustering effect is better. The steps of the optimization algorithm for selecting the initial clustering centers include: randomly selecting a sample from the sample set as the first initial clustering center; for each remaining sample x i , first calculate the distances between it and the existing clustering centers, record the shortest distance among them, and use d(x i It should be noted that the content in seems to be incomplete or contains some incorrect expressions in the original Chinese. The translation is based on the existing text as accurately as possible.) means; for each sample x i , the probability of being selected as the next cluster center is The cluster center of this round is selected according to probability, and the sample d(x i ) is larger, the higher the probability of being selected; iterative steps are performed until k initial cluster centers are selected.

[0089] On the other hand, the kernel function provides a convenient solution for the mapping relationship of the kernel method. Assuming X is the input space and H is a high-dimensional space, there is a mapping function Map a point x in X space to a point h in H space: If there exists a function K(x,z) such that for all xz∈X Then K(x,z) is called the kernel function, where for and Therefore, if the kernel function is found, the inner product of the two points after mapping can be solved directly in the original space.

[0090] Furthermore, kernel methods can bypass the mapping function Solving through the kernel function K(x,z) makes the calculation more convenient. If you first map it to a high-dimensional space and then perform the algorithm calculation, the amount of calculation is much greater than directly using the kernel function in the input space. In the clustering algorithm, it is necessary to calculate the similarity of two samples, which can be represented by the inner product of the two samples or by the kernel function. Therefore, the kernel method can be perfectly applied to the clustering algorithm. In the clustering algorithm, it is necessary to calculate the distance between two samples. The distance between two samples in high-dimensional space can be represented by the kernel function:

[0091]

[0092] In addition, it is necessary to calculate each sample x i To each cluster center μ j The distance in high-dimensional space is:

[0093]

[0094] The cluster center in high-dimensional space can be expressed as:

[0095]

[0096] Substitute the cluster center formula in the high-dimensional space into each sample x i To each cluster center μ j The distance formula is further solved as:

[0097]

[0098] It can be seen from the derivation process that in the calculation process of the clustering algorithm, it is not necessary to explicitly know the mapping function The expression of can be completely calculated by kernel function. In order to speed up the algorithm calculation and avoid repeated calculation, the kernel function between each sample in the sample set can be calculated in advance and stored in the kernel function matrix. When used, it can be directly obtained by table lookup. Commonly used kernel functions include:

[0099] Linear kernel function:

[0100] K(x,z)=x T z+c

[0101] Polynomial kernel function:

[0102] K(x,z)=(ax T z+c) d

[0103] Gaussian kernel function:

[0104] K(x,z)=exp(-γ||xz|| 2 )

[0105] Using different kernel functions will result in different high-dimensional feature spaces after mapping, and the clustering results will also be different.

[0106] The embodiments of the present application can combine the characteristics of WGAN-GP and conditional DCGAN to better characterize the uncertainty relationship of wind and solar power output, and can also add label mining feature correlation to generate different label data sets. The L-ISODATA clustering algorithm can change the k value, can adaptively adjust the number of clusters, and use the kernel method to improve the effect and speed, and conduct a comprehensive and scientific analysis of wind and solar power output.

[0107] The following is further described in conjunction with specific embodiments. Figure 3 As shown in the figure, the steps of the new energy high uncertainty scenario generation method based on the CWGAN-GP model are as follows:

[0108] Step S301: Start.

[0109] Step S302: Processing historical scenery scene set data.

[0110] Historical wind and solar power output scene data is an important basis for studying the output characteristics of renewable energy power generation. These data include but are not limited to the output records of wind power generation and photovoltaic power generation in different geographical locations and environmental conditions in different time periods in the past.

[0111] Step S303: Data preprocessing, determining real data, noise data, and conditional data.

[0112] To determine the real data, it is necessary to screen out the data that conforms to the actual situation and is representative from a large number of historical wind and solar power output scene sets; to determine the noise data, it is necessary to identify the data corresponding to the random factors in the original data that may interfere with model training and scene generation; to determine the conditional data, it is necessary to extract the data sets corresponding to various conditional information related to wind and solar power generation output from the original data based on the research purpose and model requirements;

[0113] Step S303: Divide into training set and validation set.

[0114] The main purpose of the training set is to train the generator and discriminator networks in the model so that they can learn the characteristics of wind and solar power data and the relationship between different data. The validation set is used to evaluate the performance and effect of the model during the training process.

[0115] Step S303: Design of G and D network parameters.

[0116] When designing the parameters of the generator (G) network, we must first consider its network structure. Based on the idea of ​​deep convolution, we need to determine parameters such as the number of convolutional layers, the size of the convolution kernel, and the step size. The generator network (D) also needs to determine the activation function and its related parameters.

[0117] Step S306: Optimize the network parameters of G and D based on the noise data and conditional data training set.

[0118] Optimizing the training set based on noisy and conditional data to adjust the network parameters of G and D is a key step, which is essential to improve model performance and generate accurate scenarios.

[0119] Step S307: Determine whether the training process has converged.

[0120] If converged, execute step S308; otherwise, execute step S303.

[0121] Step S308: Analyze the performance of the CWGAN-GP model based on the validation set.

[0122] By comprehensively analyzing the performance of the CWGAN-GP model based on the validation set, including but not limited to accuracy, stability, diversity, and responsiveness to conditional data, the pros and cons of the model can be accurately evaluated, providing a basis for further optimization of the model, thereby improving the quality and practicality of the model in generating high-uncertainty scenarios for new energy, and providing strong support for the optimized dispatch and safe operation of new power systems.

[0123] Step S309: Determine whether the performance meets the requirements.

[0124] If the performance meets the requirements, execute step S310, otherwise, execute step S303.

[0125] Step S310: Output G to generate diversified scenes.

[0126] Through the generation of diversified scenarios, rich reference data is provided for the optimized dispatching and safe operation of new power systems, enabling dispatchers to better cope with the high uncertainty output of new energy sources, thereby improving the operating efficiency and stability of the power system.

[0127] Step S311: End.

[0128] According to the new energy high-uncertainty scenario generation method based on the CWGAN-GP model proposed in the embodiment of the present application, the Wasserstein distance and gradient penalty terms can be used to correct the initial generative adversarial network, and the target convolutional neural network can be introduced into the final generative adversarial network to obtain a conditional deep convolution generative adversarial network with improved deep convolution, and WGAN-GP can be introduced into the conditional deep convolution generative adversarial network with improved deep convolution to obtain a CWGAN-GP model with improved deep convolution. By combining multiple technical advantages, the characteristics and laws of the data can be better learned during the training process, and the accuracy of generating various high-uncertainty new energy power generation output scenarios can be improved, providing more reliable support for the optimized scheduling and safe operation of the power system.

[0129] Secondly, the new energy high-uncertainty scenario generation device based on the CWGAN-GP model proposed in an embodiment of the present application is described with reference to the accompanying drawings.

[0130] Figure 4 It is a block diagram of a new energy high-uncertainty scenario generation device based on a CWGAN-GP model in an embodiment of the present application.

[0131] like Figure 4 As shown, the new energy high-uncertainty scenario generation device 10 based on the CWGAN-GP model includes: a first generation module 100, a second generation module 200 and a third generation module 300.

[0132] Specifically, the first generation module 100 is used to modify the initial generative adversarial network using the Wasserstein distance and the gradient penalty term to obtain the final generative adversarial network.

[0133] The second generation module 200 is used to introduce the target convolutional neural network into the final generative adversarial network to obtain a conditional deep convolutional generative adversarial network with improved deep convolution.

[0134] The third generation module 300 is used to introduce WGAN-GP into the conditional deep convolution generative adversarial network of the improved deep convolution to obtain the CWGAN-GP model of the improved deep convolution, and apply the target L-ISODATA clustering algorithm to cluster the generated wind and solar scenes to generate a high-uncertainty new energy power generation output scene.

[0135] Optionally, in one embodiment of the present application, the calculation formula of the Wasserstein distance is:

[0136]

[0137] Where W(P(x), P(z)) is the Wasserstein distance between the real data and the generated data, Sup is the minimum upper bound, and L≤1 represents the upper bound of the 1-Lipschitz expected value difference.

[0138] The calculation formula of the gradient penalty term is:

[0139]

[0140] in, is the gradient of D(x) with respect to x, and ||.||2 is the norm.

[0141] Optionally, in one embodiment of the present application, the objective function of the CWGAN-GP model of the improved deep convolution is:

[0142]

[0143] in, is the optimal parameter of the generator G that minimizes the objective function V(D,G), and the optimal parameter of the discriminator D that maximizes the objective function V(D,G), E[·] is the mathematical expectation, z is the random noise, x is the real data, x~P(x) is the data distribution of the real sample x, z~P(z) is the data distribution of the generated sample, logD(x) is the judgment of the discriminator on the real sample, and λ is the gradient penalty coefficient.

[0144] Optionally, in one embodiment of the present application, the third generating module 300 includes: a first calculating unit and a clustering unit.

[0145] The first calculation unit is used to calculate the kernel function between each sample in the sample set and store them in the kernel function matrix.

[0146] The clustering unit is used to query the corresponding kernel function from the kernel function matrix during the calculation process of the clustering algorithm to perform clustering.

[0147] It should be noted that the aforementioned explanation of the embodiment of the method for generating high-uncertainty scenarios of new energy based on the CWGAN-GP model is also applicable to the device for generating high-uncertainty scenarios of new energy based on the CWGAN-GP model of this embodiment, and will not be repeated here.

[0148] According to the new energy high-uncertainty scenario generation device based on the CWGAN-GP model proposed in the embodiment of the present application, the Wasserstein distance and gradient penalty terms can be used to correct the initial generative adversarial network, and the target convolutional neural network can be introduced into the final generative adversarial network to obtain a conditional deep convolution generative adversarial network with improved deep convolution, and WGAN-GP can be introduced into the conditional deep convolution generative adversarial network with improved deep convolution to obtain a CWGAN-GP model with improved deep convolution. By combining multiple technical advantages, the characteristics and laws of the data can be better learned during the training process, and the accuracy of generating various high-uncertainty new energy power generation output scenarios can be improved, providing more reliable support for the optimized scheduling and safe operation of the power system.

[0149] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may include:

[0150] A memory 501 , a processor 502 , and a computer program stored in the memory 501 and executable on the processor 502 .

[0151] When the processor 502 executes the program, the method for generating new energy high-uncertainty scenarios based on the CWGAN-GP model provided in the above embodiment is implemented.

[0152] Furthermore, the electronic device further comprises:

[0153] The communication interface 503 is used for communication between the memory 501 and the processor 502 .

[0154] The memory 501 is used to store computer programs that can be executed on the processor 502 .

[0155] The memory 501 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0156] If the memory 501, the processor 502 and the communication interface 503 are implemented independently, the communication interface 503, the memory 501 and the processor 502 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0157] Optionally, in a specific implementation, if the memory 501, the processor 502 and the communication interface 503 are integrated on a chip, the memory 501, the processor 502 and the communication interface 503 can communicate with each other through an internal interface.

[0158] The processor 502 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0159] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned method for generating new energy high-uncertainty scenarios based on the CWGAN-GP model.

[0160] An embodiment of the present application also provides a computer program product, including a computer program, which, when executed, is used to implement the above-mentioned method for generating new energy high-uncertainty scenarios based on the CWGAN-GP model.

[0161] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0162] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0163] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0164] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or N wirings (electronic devices), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways as necessary and then storing it in a computer memory.

[0165] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above embodiment, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0166] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0167] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0168] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for generating high-uncertainty scenarios of new energy based on the CWGAN-GP model, characterized in that: The following steps are involved: Use Wasserstein distance and gradient penalty term to modify the initial generative adversarial network to obtain the final generative adversarial network; Introducing the target convolutional neural network into the final generative adversarial network to obtain a conditional deep convolutional generative adversarial network with improved deep convolution; The WGAN-GP is introduced into the conditional deep convolutional generative adversarial network of the improved deep convolution to obtain the CWGAN-GP model of the improved deep convolution, and the target L-ISODATA clustering algorithm is applied to cluster the generated wind and solar scenes to generate high-uncertainty renewable energy power generation output scenarios.

2. The method according to claim 1, characterized in that in, The calculation formula of the Wasserstein distance is: Where W(P(x), P(z)) is the Wasserstein distance between the real data and the generated data, Sup is the minimum upper bound, and L≤1 represents the upper bound of the expected value difference of 1-Lipschitz; The calculation formula of the gradient penalty term is: in, is the gradient of D(x) with respect to x, and ||.||2 is the norm.

3. The method according to claim 1, characterized in that The objective function of the CWGAN-GP model with improved deep convolution is: in, is the optimal parameter of the generator G that minimizes the objective function V(D,G), and the optimal parameter of the discriminator D that maximizes the objective function V(D,G), E[·] is the mathematical expectation, z is the random noise, x is the real data, x~P(x) is the data distribution of the real sample x, z~P(z) is the data distribution of the generated sample, logD(x) is the judgment of the discriminator on the real sample, and λ is the gradient penalty coefficient.

4. The method according to claim 1, characterized in that: The application target L-ISODATA clustering algorithm is used to cluster the generated scenery scene, including: Calculate the kernel function between each sample in the sample set and store it in the kernel function matrix; During the calculation process of the clustering algorithm, the corresponding kernel function is queried from the kernel function matrix to perform clustering.

5. A new energy high uncertainty scenario generation device based on the CWGAN-GP model, characterized in that: include: The first generation module is used to modify the initial generation adversarial network using the Wasserstein distance and the gradient penalty term to obtain the final generation adversarial network; A second generation module is used to introduce the target convolutional neural network into the final generative adversarial network to obtain a conditional deep convolutional generative adversarial network with improved deep convolution; The third generation module is used to introduce WGAN-GP into the conditional deep convolutional generative adversarial network of the improved deep convolution to obtain the CWGAN-GP model of improved deep convolution, and apply the target L-ISODATA clustering algorithm to cluster the generated wind and solar scenes to generate high-uncertainty new energy power generation output scenarios.

6. The device according to claim 5, characterized in that in, The calculation formula of the Wasserstein distance is: Where W(P(x), P(z)) is the Wasserstein distance between the real data and the generated data, Sup is the minimum upper bound, and L≤1 represents the upper bound of the expected value difference of 1-Lipschitz; The calculation formula of the gradient penalty term is: in, is the gradient of D(x) with respect to x, and ||.||2 is the norm.

7. The device according to claim 5, characterized in that The third generation module comprises: A first calculation unit is used to calculate the kernel function between each sample in the sample set and store it in a kernel function matrix; The clustering unit is used to query the corresponding kernel function from the kernel function matrix during the calculation process of the clustering algorithm to perform clustering.

8. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for generating new energy high-uncertainty scenarios based on the CWGAN-GP model as described in any one of claims 1 to 4.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the new energy high-uncertainty scenario generation method based on the CWGAN-GP model as described in any one of claims 1 to 4.

10. A computer program product, characterized in that, It comprises a computer program, characterized in that the computer program is executed to implement the new energy high uncertainty scenario generation method based on the CWGAN-GP model as described in any one of claims 1-4.

Images