Clean energy multi-time scale power generation capability evaluation model training method and device

Abstract

The invention relates to a clean energy multi-time scale power generation capability evaluation model training method and device. The method comprises the steps that sample power generation data of clean energy under different time scales are collected, the sample power generation load data are input into a pre-trained power generation load scene generation model, a power generation load scene set is obtained, and the power generation load scene set comprises multiple sample power generation load scenes; the sample power generation load scenes are used for representing environment state information of the sample power generation load data in any time scale, and scene reduction is performed on each sample power generation load scene by using a clustering algorithm to obtain a target sample power generation load scene set, and inputting the target sample power generation load scene set into a pre-constructed power generation capability evaluation model, and performing model training on the power generation capability evaluation model to obtain a trained power generation capability evaluation model. By adopting the method, the evaluation accuracy of the trained model can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of energy generation, and particularly to a method, device, computer equipment, computer-readable storage medium, and computer program product for evaluating the power generation capacity of clean energy on multiple time scales. Background Art

[0002] With the development of energy generation technology, the methods for evaluating the power generation capacity of clean energy rely on complex models. Moreover, due to the instability and volatility of clean energy, problems that do not conform to the actual situation are likely to occur in the evaluation of the power generation capacity of clean energy.

[0003] However, in the current method for training the power generation capacity evaluation model of clean energy, there is a problem of low model evaluation accuracy. Summary of the Invention

[0004] Based on this, in view of the above technical problems, it is necessary to provide a method, device, computer equipment, computer-readable storage medium, and computer program product for training a clean energy multi-time scale power generation capacity evaluation model that can improve the model evaluation accuracy.

[0005] In a first aspect, the present application provides a method for training a clean energy multi-time scale power generation capacity evaluation model, including:

[0006] Collecting sample power generation load data of clean energy at different time scales;

[0007] Inputting the sample power generation load data into a pre-trained power generation load scenario generation model to obtain a set of sample power generation load scenarios; the set of sample power generation load scenarios includes multiple sample power generation load scenarios, and the sample power generation load scenarios are used to characterize the environmental state information of the sample power generation load data within any one time scale;

[0008] Using a clustering algorithm to reduce the scenarios of each sample power generation load scenario to obtain a set of target sample power generation load scenarios;

[0009] Inputting the set of target sample power generation load scenarios into a pre-constructed clean energy multi-time scale power generation capacity evaluation model to train the clean energy multi-time scale power generation capacity evaluation model, and obtaining a trained clean energy multi-time scale power generation capacity evaluation model.

[0010] In one embodiment, the power generation load scenario generation model includes a generator and a discriminator, and the discriminator is trained through the following steps:

[0011] Collecting the power generation load data of clean energy and inputting the power generation load data into the generator to generate corresponding predicted power generation load scenario data; the power generation load data includes power generation load scenario data;

[0012] Input the power generation load scenario data and the predicted power generation load scenario data into the discriminator to be trained, and obtain the first expected probability information and the second expected probability information. The first expected probability information is used to characterize the scoring information of the discriminator to be trained for the power generation load scenario data, and the second expected probability information is used to characterize the scoring information of the discriminator to be trained for the predicted power generation load scenario data.

[0013] Based on the first expected probability information, the second expected probability information, and the gradient penalty term, obtain the discriminator loss value corresponding to the discriminator to be trained.

[0014] According to the discriminator loss value, update the discriminator parameters of the discriminator to be trained until the discriminator loss value reaches the minimum value or the update times of the discriminator parameters reach the preset number threshold, and then determine that the discriminator training is completed.

[0015] In one embodiment, the generator is trained through the following steps:

[0016] Update the generator parameters of the generator according to the second expected probability information until the second expected probability information reaches the minimum value or the update times of the generator parameters reach the preset number threshold, and then determine that the generator training is completed.

[0017] In one embodiment, the clean energy multi-time scale power generation capacity evaluation model includes a causal convolution module, a residual module, and an evaluation module. Input the target sample power generation load scenario set into the pre-constructed clean energy multi-time scale power generation capacity evaluation model, and perform model training on the clean energy multi-time scale power generation capacity evaluation model, including:

[0018] Input the target sample power generation load scenario set into the causal convolution module, and obtain the first predicted power generation load data corresponding to the target sample power generation load scenario set through the causal convolution module.

[0019] Input the first predicted power generation load data into the residual module, obtain the second predicted power generation load data through the residual module, and obtain the predicted power generation load data according to the first predicted power generation load data and the second predicted power generation load data.

[0020] Input the predicted power generation load data and the sample power generation load data into the evaluation module, and obtain the evaluation accuracy information corresponding to the clean energy multi-time scale power generation capacity evaluation model through the evaluation module.

[0021] In the case where the evaluation accuracy information does not meet the preset conditions, update the model parameters of the clean energy multi-time scale power generation capacity evaluation model to train the clean energy multi-time scale power generation capacity evaluation model.

[0022] In an exemplary embodiment, a clustering algorithm is used to reduce the scenarios of each sample power generation load scenario, and a set of target sample power generation load scenarios is obtained, including:

[0023] Obtain the distance information between each sample power generation load scenario, and obtain the set variance of the sample power generation load scenario set and the object variance of each sample power generation load scenario according to the distance information; the set variance is used to measure the similarity between each sample power generation load scenario, and the object variance is used to characterize the distribution of the sample power generation load scenario;

[0024] Determine the candidate scenario clusters corresponding to the sample power generation load scenario set according to the set variance, the object variance, and a preset stretching factor;

[0025] Based on the distance information between each sample power generation load scenario and the candidate scenario clusters, add each sample power generation load scenario to the corresponding candidate scenario cluster;

[0026] Update the candidate scenario clusters according to the added candidate scenario clusters, and obtain and record the total cluster cost corresponding to the updated candidate scenario clusters. Return to execute updating the candidate scenario clusters based on the distance information between each sample power generation load scenario and the candidate scenario clusters and obtaining the total cluster cost corresponding thereto until the total cluster cost is the same as the previously recorded total cluster cost;

[0027] Construct a set of target sample power generation load scenarios based on the current candidate scenario clusters.

[0028] In one embodiment, the clean energy includes wind power energy and photovoltaic energy. Sample power generation load data of the clean energy at different time scales is collected, including:

[0029] Obtain the actual output power of the wind turbine, the rated capacity of the wind turbine, the wind speed at the hub height of the wind turbine, the rated wind speed, the cut-in wind speed, and the cut-out wind speed, and obtain the actual output power of the photovoltaic array, the rated capacity of the photovoltaic array, the sunshine information and temperature information on the surface of the photovoltaic array;

[0030] Obtain the first sample output power of the wind power energy according to the actual output power of the wind turbine, the rated capacity of the wind turbine, the wind speed at the hub height of the wind turbine, the rated wind speed, the cut-in wind speed, and the cut-out wind speed;

[0031] Based on the actual output power of the photovoltaic array, the rated capacity of the photovoltaic array, the sunshine information and temperature information on the surface of the photovoltaic array, obtain the second sample output power of the photovoltaic energy;

[0032] Obtain the sample power generation load data according to the first sample output power and the second sample output power.

[0033] In a second aspect, the present application also provides a training device for an evaluation model of clean energy's multi-time-scale power generation capacity, including:

[0034] A data acquisition module, configured to acquire sample power generation load data of clean energy at different time scales;

[0035] A scenario construction module, configured to input the sample power generation load data into a pre-trained power generation load scenario generation model to obtain a set of sample power generation load scenarios; the set of sample power generation load scenarios includes multiple sample power generation load scenarios, and the sample power generation load scenarios are used to characterize the environmental state information of the sample power generation load data within any one time scale;

[0036] A scenario reduction module, configured to perform scenario reduction on each sample power generation load scenario by using a clustering algorithm to obtain a set of target sample power generation load scenarios;

[0037] A model training module, configured to input the set of target sample power generation load scenarios into a pre-constructed evaluation model of clean energy's multi-time-scale power generation capacity, and perform model training on the evaluation model of clean energy's multi-time-scale power generation capacity to obtain a trained evaluation model of clean energy's multi-time-scale power generation capacity.

[0038] In a third aspect, the present application also provides a computer device, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0039] Acquire sample power generation load data of clean energy at different time scales;

[0040] Input the sample power generation load data into a pre-trained power generation load scenario generation model to obtain a set of sample power generation load scenarios; the set of sample power generation load scenarios includes multiple sample power generation load scenarios, and the sample power generation load scenarios are used to characterize the environmental state information of the sample power generation load data within any one time scale;

[0041] Perform scenario reduction on each sample power generation load scenario by using a clustering algorithm to obtain a set of target sample power generation load scenarios;

[0042] Input the set of target sample power generation load scenarios into a pre-constructed evaluation model of clean energy's multi-time-scale power generation capacity, and perform model training on the evaluation model of clean energy's multi-time-scale power generation capacity to obtain a trained evaluation model of clean energy's multi-time-scale power generation capacity.

[0043] In a fourth aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0044] Acquire sample power generation load data of clean energy at different time scales;

[0045] Input the sample power generation load data into the pre-trained power generation load scenario generation model to obtain a set of sample power generation load scenarios; the set of sample power generation load scenarios contains multiple sample power generation load scenarios, and the sample power generation load scenarios are used to characterize the environmental state information of the sample power generation load data within any time scale.

[0046] Use a clustering algorithm to perform scenario reduction on each sample power generation load scenario to obtain a set of target sample power generation load scenarios.

[0047] Input the set of target sample power generation load scenarios into the pre-constructed clean energy multi-time scale power generation capacity evaluation model, and perform model training on the clean energy multi-time scale power generation capacity evaluation model to obtain a trained clean energy multi-time scale power generation capacity evaluation model.

[0048] In a fifth aspect, the present application also provides a computer program product, including a computer program, which when executed by a processor implements the following steps:

[0049] Collect sample power generation load data of clean energy at different time scales.

[0050] Input the sample power generation load data into the pre-trained power generation load scenario generation model to obtain a set of sample power generation load scenarios; the set of sample power generation load scenarios contains multiple sample power generation load scenarios, and the sample power generation load scenarios are used to characterize the environmental state information of the sample power generation load data within any time scale.

[0051] Use a clustering algorithm to perform scenario reduction on each sample power generation load scenario to obtain a set of target sample power generation load scenarios.

[0052] Input the set of target sample power generation load scenarios into the pre-constructed clean energy multi-time scale power generation capacity evaluation model, and perform model training on the clean energy multi-time scale power generation capacity evaluation model to obtain a trained clean energy multi-time scale power generation capacity evaluation model.

[0053] The above-mentioned training method, device, computer equipment, computer-readable storage medium and computer program product for the clean energy multi-time scale power generation capacity evaluation model collect the sample power generation data of clean energy at different time scales, input the sample power load data into the pre-trained power load scenario generation model to obtain a set of power load scenarios. The set of power load scenarios contains multiple sample power load scenarios, and the sample power load scenario is used to characterize the environmental state information of the sample power load data within any one time scale. Use the clustering algorithm to reduce the scenarios of each sample power load scenario to obtain a set of target sample power load scenarios, input the set of target sample power load scenarios into the pre-constructed clean energy multi-time scale power generation capacity evaluation model, and train the clean energy multi-time scale power generation capacity evaluation model to obtain the trained clean energy multi-time scale power generation capacity evaluation model. By collecting the sample power load data of clean energy at different time scales, generating corresponding sample power load scenarios using the sample power load data, and then using the clustering algorithm to determine representative scenarios from them to form a set of target sample power load scenarios, it reduces the possibility that the evaluation accuracy of the model decreases due to the input of invalid sample power load scenarios into the clean energy multi-time scale power generation capacity evaluation model for model training. And through multi-source data collection and corresponding data processing, the evaluation accuracy of the trained model is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for the description of the embodiments of the present application or related technologies. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0055] Figure 1 It is an application environment diagram of the training method for the clean energy multi-time scale power generation capacity evaluation model in an embodiment;

[0056] Figure 2 It is a flowchart of the training method for the clean energy multi-time scale power generation capacity evaluation model in an embodiment;

[0057] Figure 3 It is a flowchart of the training method for the clean energy multi-time scale power generation capacity evaluation model in another embodiment;

[0058] Figure 4 It is a structural block diagram of the training device for the clean energy multi-time scale power generation capacity evaluation model in an embodiment;

[0059] Figure 5Internal structure diagram of a computer device in an embodiment. Detailed implementation manners

[0060] In order to make the objectives, technical solutions and advantages of the present application more clear and understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0061] The method for training an evaluation model of clean energy multi-time scale power generation capacity provided by an embodiment of the present application can be applied to, for example Figure 1 the application environment shown in the figure. Among them, the power grid system communicates with the server 102 through a network. This power grid system includes electrical equipment and power generation equipment. The power generation equipment mainly includes clean energy, which is composed of biomass energy, solar energy, wind power generation, geothermal energy, tidal energy, and hydroelectric power, etc. Clean energy does not produce harmful substances during the power generation process, so it is called clean energy. The data storage system can store the data that the server 102 needs to process. The data storage system can be integrated on the server 102, or placed in the cloud or other network servers. Collect the sample power generation load data of clean energy at different time scales, input the sample power generation load data into a pre-trained power generation load scenario generation model to obtain a set of sample power generation load scenarios. The set of sample power generation load scenarios contains multiple sample power generation load scenarios, and the sample power generation load scenarios are used to characterize the environmental state information of the sample power generation load data within any one time scale. Use a clustering algorithm to reduce the scenarios of each sample power generation load scenario to obtain a set of target sample power generation load scenarios. Finally, input the set of target sample power generation load scenarios into a pre-constructed evaluation model of clean energy multi-time scale power generation capacity to train the evaluation model of clean energy multi-time scale power generation capacity, and obtain a trained evaluation model of clean energy multi-time scale power generation capacity. Among them, the server 102 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0062] In an exemplary embodiment, as Figure 2 shown, a method for training an evaluation model of clean energy multi-time scale power generation capacity is provided. Taking the method applied to Figure 1 the server 102 in the figure as an example, it includes the following steps S201 to S204. Among them:

[0063] Step S201: Collect sample power generation load data of clean energy at different time scales.

[0064] Among them, clean energy can be understood as a substance that can convert the energy it possesses into electrical energy without generating harmful substances during this process, such as solar energy, wind energy, etc. Different time scales can be understood as different lengths of time, such as time scales under different time measurement units like one hour, half an hour, one day, or several days, etc.

[0065] Optionally, the server 102 can collect sample power generation load data corresponding to a time scale from clean energy within a set time length, and uniformly convert the collected sample power generation load data into power output, laying a data foundation for various subsequent data processing, avoiding the need for data conversion during subsequent processing, and thus accelerating the training progress of model training.

[0066] Step S202: Input the sample power generation load data into a pre-trained power generation load scenario generation model to obtain a set of sample power generation load scenarios; the set of sample power generation load scenarios contains multiple sample power generation load scenarios, and the sample power generation load scenarios are used to represent the environmental state information of the sample power generation load data within any one time scale.

[0067] Among them, the sample power generation load scenario can be understood as the environmental state information in the power grid system when the server collects the sample power generation load data corresponding to a time scale.

[0068] Exemplarily, the server 102 inputs the sample power generation load data into a pre-trained power generation load scenario generation model, generates a sample power generation load scenario for representing the environmental state information of the sample power generation load data within any one time scale through this power generation load scenario generation model, combines each sample power generation load scenario to form a set of sample power generation load scenarios, and simultaneously pays attention to the power generation information of clean energy under multiple scenarios, avoiding inaccurate power generation evaluation caused by only paying attention to the power generation information under a single scenario, and thus improving the evaluation accuracy of the trained model.

[0069] Step S203: Use a clustering algorithm to reduce the scenarios of each sample power generation load scenario to obtain a set of target sample power generation load scenarios.

[0070] Among them, the clustering algorithm can be understood as an unsupervised learning method used to divide the samples in a dataset into several clusters, so that the samples within the same cluster are similar to each other, while the samples between different clusters are quite different.

[0071] Optionally, the server 102 uses a clustering algorithm to perform scenario reduction on each sample power generation load scenario included in the sample power generation load scenario set, removes the unrepresentative sample power generation load scenarios, and constructs a corresponding target sample power generation load scenario set using the remaining sample power generation load scenarios. Using the clustering algorithm for scenario reduction to select more representative scenario data from multiple scenarios increases the quantity of data input into the model, thereby accelerating the training speed of the model training and also improving the evaluation accuracy of the model trained with more representative data.

[0072] Step S204: Input the target sample power generation load scenario set into a pre-constructed clean energy multi-time scale power generation capacity evaluation model, perform model training on the clean energy multi-time scale power generation capacity evaluation model, and obtain the trained clean energy multi-time scale power generation capacity evaluation model.

[0073] Exemplarily, the server 102 inputs the target sample power generation load scenario set into a pre-constructed clean energy multi-time scale power generation capacity evaluation model, obtains corresponding evaluation accuracy information through the clean energy multi-time scale power generation capacity evaluation model. In the case where the evaluation accuracy information does not meet the preset conditions, update the model parameters of the clean energy multi-time scale power generation capacity evaluation model to achieve model training of the clean energy multi-time scale power generation capacity evaluation model until the power generation capacity evaluation information meets the preset conditions, at which point it is determined that the training is complete. Using the accuracy information as the condition for judging whether the model training is complete ensures the accuracy of the trained model in terms of power generation capacity evaluation, which is beneficial for helping users perform corresponding energy planning based on the model output.

[0074] In the above method for training the multi-time scale power generation capacity evaluation model of clean energy, sample power generation data of clean energy at different time scales is collected, and the sample power generation load data is input into the pre-trained power generation load scenario generation model to obtain a set of power generation load scenarios. The set of power generation load scenarios contains multiple sample power generation load scenarios, and the sample power generation load scenario is used to characterize the environmental state information of the sample power generation load data within any one time scale. The clustering algorithm is used to reduce the scenarios of each sample power generation load scenario to obtain a set of target sample power generation load scenarios. The set of target sample power generation load scenarios is input into the pre-constructed multi-time scale power generation capacity evaluation model of clean energy to train the multi-time scale power generation capacity evaluation model of clean energy, and a trained multi-time scale power generation capacity evaluation model of clean energy is obtained. By collecting sample power generation load data of clean energy at different time scales, generating corresponding sample power generation load scenarios using the sample power generation load data, and then using the clustering algorithm to determine representative scenarios from them to form a set of target sample power generation load scenarios, the possibility of the evaluation accuracy of the model decreasing due to the input of invalid sample power generation load scenarios into the multi-time scale power generation capacity evaluation model of clean energy for model training is reduced, and the evaluation accuracy of the trained model is improved through multi-source data collection and corresponding data processing.

[0075] In one embodiment, the power generation load scenario generation model includes a generator and a discriminator, and the discriminator is trained through the following steps:

[0076] Collect the power generation load data of clean energy, and input the power generation load data into the generator to generate corresponding predicted power generation load scenario data; the power generation load data includes power generation load scenario data; input the power generation load scenario data and the predicted power generation load scenario data into the discriminator to be trained to obtain first expected probability information and second expected probability information; the first expected probability information is used to characterize the scoring information of the discriminator to be trained for the power generation load scenario data, and the second expected probability information is used to characterize the scoring information of the discriminator to be trained for the predicted power generation load scenario data; based on the first expected probability information, the second expected probability information, and the gradient penalty term, obtain the discriminator loss value corresponding to the discriminator to be trained; update the discriminator parameters of the discriminator to be trained according to the discriminator loss value until the discriminator loss value reaches the minimum value or the number of updates of the discriminator parameters reaches the preset number threshold, and then determine that the training of the discriminator is completed.

[0077] Among them, the first expected probability information can be understood as the score of the discriminator for the input power generation load scenario data. The higher this first expected probability information is, the better the discriminator can distinguish between real data and the data generated by the generator. Similarly, the second expected probability information can be understood as the score of the discriminator for the input predicted power generation load scenario data. The closer this second expected probability information is to 0, the more it indicates that the discriminator can recognize that the predicted power generation load scenario data is false.

[0078] Optionally, the server 102 collects power generation load data including power generation load scenario data of clean energy, inputs the power generation load data into the generator to generate corresponding predicted power generation load scenario data, inputs the power generation load scenario data and the predicted power generation load scenario data into the discriminator to be trained, and obtains the first expected probability information and the second expected probability information. Based on the first expected probability information, the second expected probability information, and the gradient penalty term, the discriminator loss value corresponding to the discriminator to be trained is obtained, and the discriminator parameters of the discriminator to be trained are updated according to the discriminator loss value until the discriminator loss value reaches the minimum value or the update times of the discriminator parameters reach the preset number threshold, and then it is determined that the training of the discriminator is completed. By training a high-performance discriminator, the accuracy of predicting the power generation load can be improved, the allocation of power resources can be optimized, the stability of the power system can be enhanced, and ultimately the utilization of clean energy and sustainable development can be promoted. The key lies in the training process of the discriminator. By combining the first expected probability information, the second expected probability information, and the gradient penalty term, the performance of the discriminator can be effectively improved.

[0079] In one embodiment, the generator is trained through the following steps: The generator parameters of the generator are updated according to the second expected probability information until the second expected probability information reaches the minimum value or the update times of the generator parameters reach the preset number threshold, and then it is determined that the training of the generator is completed.

[0080] Exemplarily, the server 102 updates the generator parameters of the generator according to the second expected probability information until the second expected probability information reaches the minimum value, that is, the second expected probability information is closest to 0, or the update times of the generator parameters reach the preset number threshold. At this time, the training of the generator is completed. By minimizing the second expected probability information, this solution can effectively train the generator to generate more realistic predicted power generation load scenario data, thereby improving the accuracy of power generation load prediction and the operating efficiency of the power system.

[0081] In an exemplary embodiment, the clean energy multi-time scale power generation capacity evaluation model includes a causal convolution module, a residual module, and an evaluation module; inputting the target sample power generation load scenario set into the pre-constructed clean energy multi-time scale power generation capacity evaluation model for model training of the clean energy multi-time scale power generation capacity evaluation model includes:

[0082] The target sample power generation load scenario set is input into the causal convolution module, and the first predicted power generation load data corresponding to the target sample power generation load scenario set is obtained through the causal convolution module; the first predicted power generation load data is input into the residual module, and the second predicted power generation load data is obtained through the residual module, and the predicted power generation load data is obtained according to the first predicted power generation load data and the second predicted power generation load data; the predicted power generation load data and the sample power generation load data are input into the evaluation module, and the evaluation accuracy information corresponding to the clean energy multi-time scale power generation capacity evaluation model is obtained through the evaluation module; when the evaluation accuracy information does not meet the preset conditions, the model parameters of the clean energy multi-time scale power generation capacity evaluation model are updated to train the clean energy multi-time scale power generation capacity evaluation model.

[0083] Optionally, since the clean energy power generation load data at different time scales is large in scale and has a long time span, a temporal convolutional network model is constructed through causal convolution and residual modules. The formula of causal convolution is:

[0084]

[0085] Among them, X represents the expansion coefficient, K represents the size of the convolution kernel, f(i) represents the i-th data in the convolution kernel, and T−di means that the convolution operation is only performed on the data from time T to time di.

[0086] As the depth of TCN increases, its ability to mine complex correlation features between time series information is enhanced, but it also brings problems such as gradient explosion and gradient disappearance. In order to solve the degradation problem of deep learning networks, the residual module is introduced for error correction, which is defined as follows:

[0087]

[0088] Among them, x represents the input sequence of the residual module, R(x) represents the residual term, and Activation(·) represents the activation function.

[0089] Each residual module consists of two nonlinear, dilated causal convolutional layers. After each dilated causal convolutional layer, a batch normalization layer is added to normalize the input of each layer network. After the normalization layer, the ReLU activation function is used to improve the model's ability to fit nonlinear data, and Dropout is introduced to reduce the risk of model overfitting.

[0090] Mean absolute percentage error and RMS error As an error evaluation indicator (i.e. the aforementioned evaluation accuracy information).

[0091]

[0092]

[0093] The smaller the value of the above evaluation index, the higher the evaluation accuracy.

[0094] Use the coefficient of determination to evaluate the effectiveness of the model (i.e., the aforementioned evaluation accuracy information). The larger the value, the more significant the fitting effect of the evaluation model on the data. The calculation formula is:

[0095]

[0096] where and respectively represent the predicted value and the actual value of the clean energy power generation load within the t time period, represents the average power generation load of clean energy, and N represents the number of samples.

[0097] Through the above model construction and model training methods, the evaluation efficiency of the power generation capacity is improved. At the same time, by calculating the mean absolute percentage error, root mean square error, and coefficient of determination, as the basis for whether the model continues to be trained, the evaluation accuracy of the model processed by training is guaranteed.

[0098] In one embodiment, the clustering algorithm is used to reduce the scenarios of each sample power generation load scenario, and the target sample power generation load scenario set is obtained, including:

[0099] Obtain the distance information between each sample power generation load scenario, and obtain the set variance of the sample power generation load scenario set and the object variance of each sample power generation load scenario according to the distance information; the set variance is used to measure the similarity between each sample power generation load scenario, and the object variance is used to characterize the distribution of the sample power generation load scenario; according to the set variance, object variance, and a preset stretching factor, determine the candidate scenario clusters corresponding to the sample power generation load scenario set; based on the distance information between each sample power generation load scenario and the candidate scenario clusters, add each sample power generation load scenario to the corresponding candidate scenario cluster; update the candidate scenario cluster according to the added candidate scenario cluster, and obtain and record the total cluster cost corresponding to the updated candidate scenario cluster, return to execute updating the candidate scenario cluster based on the distance information between each sample power generation load scenario and the candidate scenario cluster and obtaining the total cluster cost corresponding to it until the total cluster cost is the same as the previously recorded total cluster cost; construct the target sample power generation load scenario set based on the current candidate scenario cluster.

[0100] Exemplarily, the K-medoids (K-center clustering) algorithm is used to reduce the scenarios, and the power generation load scenarios with typical characteristics are selected from the scenario set, so as to obtain fine wind power and photovoltaic power generation load scenarios;

[0101] Specifically, the number of cluster centers in the preset k-means algorithm is set, and optimized cluster centers are obtained during the clustering process. According to the principle of the closest cluster center, the remaining points are assigned to the class represented by the current best cluster center. Finally, when all cluster centers no longer change, the scenario reduction is completed.

[0102] Embodiment: S31. Input: dataset D, number of clusters K, stretching factor λ. Output: clusters .

[0103] S32. Calculate the distance between each pair of objects using formula (5):

[0104] (5)

[0105] where m is the attribute number of the object.

[0106] S33. Calculate the variance σ of dataset D using formula (6), calculate the object variance using formula (7), and then determine the candidate medoids subset using formula (8) , and then use formula (8) to determine the candidate medoids subset :

[0107] (6)

[0108] where is the object mean;

[0109] (7)

[0110] (8)

[0111] where λ is the stretching factor.

[0112] S34. Select two initial medoids using formula (9) and formula (10) :

[0113] (9)

[0114] (10)

[0115] S35. Assign each object to the nearest medoid and calculate the total cluster cost E using formula (11).

[0116] (11)

[0117] S36. For k from 2 to k - 1, calculate the newly added medoids using formula (12) , and generate a new medoids set 。

[0118] (12)

[0119] S37. Repeat.

[0120] S38. Assign each object to the nearest medoid according to the nearest distance principle.

[0121] S39. Update the medoid set O.

[0122] S310. Calculate the total clustering cost E using formula (11).

[0123] S311. When the total clustering cost E no longer changes, end.

[0124] By using the K-medoids algorithm to reduce scenarios, typical power generation load scenarios are selected from the scenario set, so as to obtain refined wind power and photovoltaic power generation load scenarios, improve the effectiveness of the scenario data input into the model, and further improve the training effect of the model.

[0125] In one embodiment, the clean energy includes wind power energy and photovoltaic energy. Sample power generation load data of clean energy at different time scales are collected, including: obtaining the actual output power of the wind turbine, the rated capacity of the wind turbine, the wind speed at the hub height of the wind turbine, the rated wind speed, the cut-in wind speed and the cut-out wind speed, and obtaining the actual output power of the photovoltaic array, the rated capacity of the photovoltaic array, the sunshine information and temperature information on the surface of the photovoltaic array; according to the actual output power of the wind turbine, the rated capacity of the wind turbine, the wind speed at the hub height of the wind turbine, the rated wind speed, the cut-in wind speed and the cut-out wind speed, obtaining the first sample output power of the wind power energy; based on the actual output power of the photovoltaic array, the rated capacity of the photovoltaic array, the sunshine information and temperature information on the surface of the photovoltaic array, obtaining the second sample output power of the photovoltaic energy; according to the first sample output power and the second sample output power, obtaining the sample power generation load data.

[0126] Optionally, the clean energy power generation loads at different time scales are collected and unified into power output.

[0127] The clean energy power generation loads at different time scales include: the power generation loads of wind power at different time scales and the power generation loads of photovoltaic at different time scales.

[0128] The output power of wind power is:

[0129]

[0130] Where: 、 are the actual output power and rated capacity of the wind turbine respectively; v, , , are the wind speed, rated wind speed, cut-in wind speed, and cut-out wind speed at the hub height of the wind turbine, respectively;

[0131] The output power of the photovoltaic is:

[0132]

[0133] where and are the actual power output and rated capacity of the photovoltaic array, respectively; is the power attenuation coefficient; and are the sunlight and temperature on the surface of the photovoltaic array; and are the sunlight and temperature under standard test conditions; is the power temperature coefficient.

[0134] Collecting the power generation load data of clean energy at different time scales, uniformly converting it into power output, and processing the numerical units of the data to convert it into the same measurement metric is beneficial for subsequent data processing using the data and speeds up the progress of model training.

[0135] In an exemplary embodiment, as Figure 3 shown, a specific implementation manner of a training method for an evaluation model of the multi-time scale power generation capacity of clean energy is provided. The overall process includes the following 5 steps, where:

[0136] Step 1: Collect the clean energy power generation load at different time scales and unify it into power output.

[0137] The clean energy power generation load at different time scales includes: the power generation load of wind power at different time scales and the power generation load of photovoltaic at different time scales.

[0138] The output power of the wind power is:

[0139]

[0140] In the formula: , are the actual output power of the wind turbine and the rated capacity, respectively; v, , , are the wind speed, rated wind speed, cut-in wind speed, and cut-out wind speed at the hub height of the wind turbine, respectively;

[0141] The output power of the photovoltaic is:

[0142]

[0143] where and are the actual power output and rated capacity of the photovoltaic array, respectively; is the power attenuation coefficient; and are the sunlight and temperature on the surface of the photovoltaic array; and are the sunlight and temperature under standard test conditions; is the power temperature coefficient.

[0144] Step 2: Input the power generation load data, use the Wasserstein generative adversarial network - gradient penalty algorithm to generate a set of power generation load scenarios, and conduct multi - scenario power generation capacity evaluation to avoid inaccurate evaluation in a single scenario.

[0145] The input of the generator is a set of random noise data , which is used to represent 's probability distribution, and the output is the generated data sample. The input of the discriminator is the scenario data at different time scales and the data generated by the generator , and the output is the probability value for determining whether the data comes from a real data sample.

[0146] The training model of the generative adversarial network is:

[0147] (1)

[0148] where E(·) represents the expected value, D(x) represents the probability of judging real data as true in the discriminator, and D(G(z)) represents the probability that the input data conforms to the data distribution at different time scales .

[0149] Taking the Wasserstein distance as the training objective of the model and introducing the gradient penalty term is beneficial to measuring the distribution differences of different data, and solving problems such as gradient explosion, unstable training, and difficult convergence in the training process of traditional generative adversarial networks.

[0150] The Wasserstein distance is defined as:

[0151] (2)

[0152] where represents the Wasserstein distance between the real data distribution and the generated data distribution, K represents the Lipschitz constant of f(x), ‖f‖L represents the function f(x) that satisfies Lipschitz continuity, L represents Lipschitz, and sup represents the least upper bound.

[0153] The gradient penalty term is defined as:

[0154] (3)

[0155] where λ represents the penalty coefficient, represents the random interpolation sampling between the generated samples and the real samples, represents the gradient of the discriminator.

[0156] The training objective function of the gradient penalty is defined as:

[0157] (4)

[0158] Step 3: Use the K-medoids algorithm to reduce the scenarios, and select the power generation load scenarios with typical features from the scenario set, so as to obtain fine wind power and photovoltaic power generation load scenarios;

[0159] Specifically, preset the number of cluster centers in the k-means algorithm, and obtain the optimized cluster centers during the clustering process. According to the principle of the closest cluster center, assign the remaining points to the class represented by the current best cluster center. Finally, when all the cluster centers no longer change, the scenario reduction is completed.

[0160] Example: S31. Input: dataset D, number of clusters K, stretching factor λ. Output: clusters 。

[0161] S32. Calculate the distance between each pair of objects using formula (5):

[0162] (5)

[0163] where m is the attribute number of the object.

[0164] S33. Calculate the variance σ of dataset D using formula (6), calculate the object variance using formula (7), and then determine the candidate medoids subset using formula (8) ,then use formula (8) to determine the candidate medoids subset :

[0165] (6)

[0166] where, is the object mean;

[0167] (7)

[0168] (8)

[0169] where λ is the stretching factor.

[0170] S34. Select two initial medoids using formula (9) and formula (10). :

[0171] (9)

[0172] (10)

[0173] S35. Assign each object to the nearest medoid and calculate the total cluster cost E using formula (11).

[0174] (11)

[0175] S36. For k from 2 to k - 1, calculate the newly added medoids using formula (12), and generate a new set of medoids and generate a new set of medoids .

[0176] (12)

[0177] S37. Repeat.

[0178] S38. Assign each object to the nearest medoid according to the nearest distance principle.

[0179] S39. Update the medoid set O.

[0180] S310. Calculate the total cluster cost E using formula (11).

[0181] S311. When the total cluster cost E no longer changes, end.

[0182] Step 4: Since the clean energy generation load data at different time scales is large in scale and long in time span, a temporal convolutional network model is constructed through causal convolution and residual modules to improve the evaluation efficiency of the power generation capacity.

[0183] The formula for causal convolution is:[[]]

[0184] (13)

[0185] Among them, X represents the expansion coefficient, K represents the size of the convolutional kernel, f(i) represents the i-th data in the convolutional kernel, and T - di means that the convolutional operation is only performed on the data from time T to time di.

[0186] As the depth of the TCN increases, its ability to mine complex correlation features between time series information is enhanced, but it also brings problems such as gradient explosion and gradient disappearance. To solve the degradation problem of the deep learning network, a residual module is introduced for error correction, and its definition is as follows:

[0187] (14)

[0188] Where x represents the input sequence of the residual module, R(x) represents the residual term, and Activation(·) represents the activation function.

[0189] Each residual module consists of two non-linear, dilated causal convolutional layers. After each dilated causal convolutional layer, a batch normalization layer is added to normalize the input of each layer network. After the normalization layer, the ReLU activation function is used to improve the model's ability to fit non-linear data, and Dropout is introduced to reduce the risk of model overfitting.

[0190] Step 5: Using the mean absolute percentage error and the root mean square error as the error evaluation metrics.

[0191] (15)

[0192] (16)

[0193] The smaller the values of the evaluation metrics in Formulas 15 and 16, the higher the evaluation accuracy.

[0194] Using the coefficient of determination to evaluate the effectiveness of the model. The larger the value, the more significant the fitting effect of the evaluation model on the data. The calculation formula is:

[0195] (17)

[0196] Where and represent the predicted value and the actual value of the clean energy power generation load within the t time period respectively, represents the average power generation load of clean energy, and N represents the number of samples.

[0197] Compared with the prior art, the present application has the following advantages:

[0198] 1. Collect the power generation load data of clean energy at different time scales, uniformly convert it into power output, process the numerical units of the data, and convert it into the same measurement metric, which is beneficial to subsequent data processing using the data and speeds up the progress of model training.

[0199] 2. Use the Wasserstein generative adversarial network-gradient penalty algorithm to generate a set of power generation load scenarios corresponding to the power generation load data, and simultaneously consider the power generation information of clean energy under multiple scenarios, thereby improving the evaluation accuracy of power generation capacity evaluation.

[0200] 3. Use the K-medoids algorithm to reduce scenarios, and select power generation load scenarios with typical characteristics from the scenario set, so as to obtain refined wind power and photovoltaic power generation load scenarios, improve the effectiveness of the scenario data input into the model, and further improve the training effect of the model.

[0201] 4. Since the clean energy power generation load data at different time scales is large in scale and long in time span, a temporal convolutional network model is constructed through causal convolution and residual modules to improve the evaluation efficiency of power generation capacity. At the same time, by calculating the mean absolute percentage error, root mean square error, and coefficient of determination, as the basis for whether the model continues to be trained, the evaluation accuracy of the trained model is ensured.

[0202] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are shown in sequence according to the arrows, these steps do not necessarily have to be executed in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least some of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages do not necessarily have to be executed at the same time, but can be executed at different times. The execution order of these steps or stages does not necessarily have to be sequential, but can be executed alternately or alternately with at least some of the steps or stages in other steps or other steps.

[0203] Based on the same inventive concept, an embodiment of the present application also provides a training device for a clean energy multi-time scale power generation capacity evaluation model for implementing the above-mentioned clean energy multi-time scale power generation capacity evaluation model training method. The implementation solution provided by this device to solve the problem is similar to the implementation solution described in the above method. Therefore, the specific limitations in one or more embodiments of the training device for the clean energy multi-time scale power generation capacity evaluation model provided below can refer to the limitations on the clean energy multi-time scale power generation capacity evaluation model training method in the above text, and will not be repeated here.

[0204] In an exemplary embodiment, as Figure 4 shown, a training device for a clean energy multi-time scale power generation capacity evaluation model is provided, including: a data acquisition module 401, a scenario construction module 402, a scenario reduction module 403, and a model training module 404, where:

[0205] The data acquisition module 401 is used to acquire sample power generation load data of clean energy at different time scales;

[0206] A scenario construction module 402, configured to input sample power generation load data into a pre-trained power generation load scenario generation model to obtain a set of sample power generation load scenarios; the set of sample power generation load scenarios includes multiple sample power generation load scenarios, and the sample power generation load scenarios are used to characterize the environmental state information of the sample power generation load data within any time scale;

[0207] A scenario reduction module 403, configured to use a clustering algorithm to reduce the scenarios of each sample power generation load scenario to obtain a set of target sample power generation load scenarios;

[0208] A model training module 404, configured to input the set of target sample power generation load scenarios into a pre-constructed clean energy multi-time scale power generation capacity evaluation model, and perform model training on the clean energy multi-time scale power generation capacity evaluation model to obtain a trained clean energy multi-time scale power generation capacity evaluation model.

[0209] In one embodiment, the power generation load scenario generation model includes a generator and a discriminator, and the clean energy multi-time scale power generation capacity evaluation model training device further includes a discriminator training module, configured to collect power generation load data of clean energy and input the power generation load data into the generator to generate corresponding predicted power generation load scenario data; the power generation load data includes power generation load scenario data; input the power generation load scenario data and the predicted power generation load scenario data into the discriminator to be trained to obtain first expected probability information and second expected probability information; the first expected probability information is used to characterize the scoring information of the discriminator to be trained for the power generation load scenario data, and the second expected probability information is used to characterize the scoring information of the discriminator to be trained for the predicted power generation load scenario data; based on the first expected probability information, the second expected probability information, and a gradient penalty term, obtain a discriminator loss value corresponding to the discriminator to be trained; update the discriminator parameters of the discriminator to be trained according to the discriminator loss value until the discriminator loss value reaches the minimum value or the number of updates of the discriminator parameters reaches a preset number threshold, and then determine that the discriminator training is completed.

[0210] In one of the embodiments, the clean energy multi-time scale power generation capacity evaluation model training device further includes a generator training module, configured to update the generator parameters of the generator according to the second expected probability information until the second expected probability information reaches the minimum value or the number of updates of the generator parameters reaches a preset number threshold, and then determine that the generator training is completed.

[0211] In an exemplary embodiment, the clean energy multi-time scale power generation capacity evaluation model includes a causal convolution module, a residual module, and an evaluation module. The model training module 404 is further configured to input the target sample power generation load scenario set into the causal convolution module, and obtain the first predicted power generation load data corresponding to the target sample power generation load scenario set through the causal convolution module; input the first predicted power generation load data into the residual module, and obtain the second predicted power generation load data through the residual module, and obtain the predicted power generation load data according to the first predicted power generation load data and the second predicted power generation load data; input the predicted power generation load data and the sample power generation load data into the evaluation module, and obtain the evaluation accuracy information corresponding to the clean energy multi-time scale power generation capacity evaluation model through the evaluation module; in the case that the evaluation accuracy information does not meet the preset conditions, update the model parameters of the clean energy multi-time scale power generation capacity evaluation model to train the clean energy multi-time scale power generation capacity evaluation model.

[0212] In one embodiment, the scenario reduction module 403 is further configured to obtain the distance information between each sample power generation load scenario, and obtain the set variance of the sample power generation load scenario set and the object variance of each sample power generation load scenario according to the distance information; the set variance is used to measure the similarity between each sample power generation load scenario, and the object variance is used to characterize the distribution of the sample power generation load scenario; determine the candidate scenario cluster corresponding to the sample power generation load scenario set according to the set variance, the object variance, and a preset stretching factor; based on the distance information between each sample power generation load scenario and the candidate scenario cluster, add each sample power generation load scenario into the corresponding candidate scenario cluster; update the candidate scenario cluster according to the added candidate scenario cluster, and obtain and record the total cluster cost corresponding to the updated candidate scenario cluster, and return to execute updating the candidate scenario cluster based on the distance information between each sample power generation load scenario and the candidate scenario cluster and obtaining the total cluster cost corresponding to it until the total cluster cost is the same as the previously recorded total cluster cost; construct the target sample power generation load scenario set based on the current candidate scenario cluster.

[0213] In one embodiment, the clean energy includes wind power energy and photovoltaic energy. The data acquisition module 401 is further configured to obtain the actual output power of the wind turbine, the rated capacity of the wind turbine, the wind speed at the hub height of the wind turbine, the rated wind speed, the cut-in wind speed, and the cut-out wind speed, and obtain the actual output power of the photovoltaic array, the rated capacity of the photovoltaic array, the sunshine information and temperature information on the surface of the photovoltaic array; obtain the first sample output power of the wind power energy according to the actual output power of the wind turbine, the rated capacity of the wind turbine, the wind speed at the hub height of the wind turbine, the rated wind speed, the cut-in wind speed, and the cut-out wind speed; obtain the second sample output power of the photovoltaic energy based on the actual output power of the photovoltaic array, the rated capacity of the photovoltaic array, the sunshine information and temperature information on the surface of the photovoltaic array; and obtain the sample power generation load data according to the first sample output power and the second sample output power.

[0214] Each module in the above clean energy multi-time scale power generation capacity evaluation model training device can be implemented in whole or in part by software, hardware, and their combination. Each of the above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory of the computer device in the form of software, so as to facilitate the processor to call and execute the operations corresponding to each of the above modules.

[0215] In an exemplary embodiment, a computer device is provided. The computer device can be a server, and its internal structure diagram can be as Figure 5 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store sample power generation load data, a set of sample power generation load scenarios, sample power generation load scenarios, and data of a set of target sample power generation load scenarios. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a method for training a clean energy multi-time scale power generation capacity evaluation model.

[0216] Those skilled in the art can understand, Figure 5The structure shown is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have a different component layout.

[0217] In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the method for training the clean energy multi-time scale power generation capacity evaluation model in the above embodiment is implemented.

[0218] In an embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method for training the clean energy multi-time scale power generation capacity evaluation model in the above embodiment is implemented.

[0219] In an embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the method for training the clean energy multi-time scale power generation capacity evaluation model in the above embodiment is implemented.

[0220] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0221] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logics based on quantum computing, artificial intelligence (AI) processors, etc., without limitation.

[0222] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in the present application.

[0223] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A clean energy multi-time scale power generation capacity assessment model training method, characterized in that: The method comprises: Collect sample power generation load data of clean energy at different time scales; Inputting the sample power generation load data into a pre-trained power generation load scenario generation model to obtain a sample power generation load scenario set; the sample power generation load scenario set includes multiple sample power generation load scenarios, and the sample power generation load scenarios are used to characterize the environmental state information of the sample power generation load data within any time scale; Using a clustering algorithm to reduce the sample power generation load scenarios to obtain a target sample power generation load scenario set; The target sample power generation load scenario set is input into a pre-built clean energy multi-time scale power generation capacity assessment model, and the clean energy multi-time scale power generation capacity assessment model is trained to obtain a trained clean energy multi-time scale power generation capacity assessment model.

2. The method according to claim 1, characterized in that The power generation load scenario generation model includes a generator and a discriminator, and the discriminator is trained by the following steps: Collecting clean energy power generation load data, and inputting the power generation load data into the generator to generate corresponding predicted power generation load scenario data; the power generation load data includes power generation load scenario data; Input the power generation load scenario data and the predicted power generation load scenario data into the discriminator to be trained to obtain first expected probability information and second expected probability information; the first expected probability information is used to characterize the scoring information of the discriminator to be trained on the power generation load scenario data, and the second expected probability information is used to characterize the scoring information of the discriminator to be trained on the predicted power generation load scenario data; Based on the first expected probability information, the second expected probability information and the gradient penalty term, obtaining a discriminator loss value corresponding to the discriminator to be trained; The discriminator parameters of the discriminator to be trained are updated according to the discriminator loss value until the discriminator loss value reaches a minimum value or the number of updates of the discriminator parameters reaches a preset number threshold, and then it is determined that the discriminator training is completed.

3. The method according to claim 2, characterized in that The generator is trained by the following steps: The generator parameters of the generator are updated according to the second expected probability information until the second expected probability information reaches a minimum value or the number of updates of the generator parameters reaches the preset number threshold, and then it is determined that the generator training is completed.

4. The method according to claim 1, characterized in that: The clean energy multi-time scale power generation capacity assessment model comprises a causal convolution module, a residual module and an assessment module; the target sample power generation load scenario set is input into a pre-built clean energy multi-time scale power generation capacity assessment model, and the clean energy multi-time scale power generation capacity assessment model is trained, including: Inputting the target sample power generation load scenario set into the causal convolution module, and obtaining first predicted power generation load data corresponding to the target sample power generation load scenario set through the causal convolution module; Inputting the first predicted power generation load data into the residual module, obtaining second predicted power generation load data through the residual module, and acquiring predicted power generation load data according to the first predicted power generation load data and the second predicted power generation load data; Inputting the predicted power generation load data and the sample power generation load data into the evaluation module, and obtaining the evaluation accuracy information corresponding to the clean energy multi-time scale power generation capacity evaluation model through the evaluation module; When the evaluation accuracy information does not meet the preset conditions, the model parameters of the clean energy multi-time scale power generation capacity evaluation model are updated to train the clean energy multi-time scale power generation capacity evaluation model.

5. The method according to claim 1, characterized in that The clustering algorithm is used to reduce the sample power generation load scenarios to obtain a target sample power generation load scenario set, including: Acquire the distance information between each of the sample power generation load scenarios, and obtain the set variance of the set of the sample power generation load scenarios and the object variance of each of the sample power generation load scenarios according to the distance information; the set variance is used to measure the similarity between each of the sample power generation load scenarios, and the object variance is used to characterize the distribution of the sample power generation load scenarios; Determining a candidate scenario cluster corresponding to the sample power generation load scenario set according to the set variance, the object variance and a preset stretch factor; Based on the distance information between each of the sample power generation load scenarios and the candidate scenario cluster, each of the sample power generation load scenarios is added to a corresponding candidate scenario cluster; Update the candidate scenario cluster according to the added candidate scenario cluster, obtain and record the total cluster cost corresponding to the updated candidate scenario cluster, return to execute based on the distance information between each of the sample power generation load scenarios and the candidate scenario cluster, update the candidate scenario cluster and obtain the total cluster cost corresponding thereto, until the total cluster cost is the same as the total cluster cost recorded last time; The target sample power generation load scenario set is obtained based on the current candidate scenario cluster construction.

6. The method according to claim 1, characterized in that The clean energy includes wind power energy and photovoltaic energy. The collected sample power generation load data of clean energy at different time scales includes: Obtain the actual output power of the wind turbine, the rated capacity of the wind turbine, the wind speed at the height of the wind turbine hub, the rated wind speed, the cut-in wind speed and the cut-out wind speed, as well as the actual output power of the photovoltaic array, the rated capacity of the photovoltaic array, the sunshine information and the temperature information on the surface of the photovoltaic array; Obtaining a first sample output power of the wind power energy according to the actual output power of the wind turbine, the rated capacity of the wind turbine, the wind speed at the height of the wind turbine hub, the rated wind speed, the cut-in wind speed and the cut-out wind speed; Based on the actual output power of the photovoltaic array, the rated capacity of the photovoltaic array, and the sunshine information and temperature information on the surface of the photovoltaic array, obtaining a second sample output power of the photovoltaic energy; The sample power generation load data is obtained according to the first sample output power and the second sample output power.

7. A clean energy multi-time scale power generation capacity assessment model training device, characterized in that: The device comprises: A data acquisition module, used to collect sample power generation load data of clean energy at different time scales; A scenario construction module, used for inputting the sample power generation load data into a pre-trained power generation load scenario generation model to obtain a sample power generation load scenario set; the sample power generation load scenario set includes a plurality of sample power generation load scenarios, and the sample power generation load scenarios are used to characterize the environmental state information of the sample power generation load data within any time scale; A scenario reduction module, used to reduce the scenarios of each of the sample power generation load scenarios by using a clustering algorithm to obtain a target sample power generation load scenario set; The model training module is used to input the target sample power generation load scenario set into a pre-built clean energy multi-time scale power generation capacity assessment model, perform model training on the clean energy multi-time scale power generation capacity assessment model, and obtain a trained clean energy multi-time scale power generation capacity assessment model.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images