A specific wind power scene generation method and device based on LSTM-ACGAN

Abstract

The application discloses a specific wind power scene generation method and device based on LSTM-ACGAN, the method comprises the following steps: based on historical load scenes and historical wind power scenes, solving an optimization problem to obtain the wind power abandonment rate of each wind power scene, classifying the wind power scenes according to the range to which the wind power abandonment rate belongs, obtaining the original wind power scene data set, then introducing a Bi-LSTM model into an ACGAN generator to construct an LSTM-ACGAN model, training the LSTM-ACGAN using the original wind power scene sample set, and obtaining the LSTM-ACGAN generator capable of generating different specific types of wind power scenes after the training is completed, and inputting Gaussian noise and a specified wind power scene label into the generator to enable the generator to generate a large number of specific wind power scenes similar to the specified type of real wind power scene. The application can provide assistance for subsequent analysis and decision of a power system.

Description

Technical Field

[0001] This invention relates to the field of new energy technology, and in particular to a method and apparatus for generating specific wind power scenarios based on LSTM-ACGAN. Background Technology

[0002] After large-scale integration of wind power into the power system, its intermittent and fluctuating nature often presents challenges in absorption. To account for the impact of wind power integration, many tasks in power system dispatch decisions require a large amount of wind power scenario data as a foundation. Therefore, generating wind power output scenarios is of great significance. Given increasingly stringent requirements for wind curtailment rates and the growing penetration rate of wind power, further considering the absorption difficulties of wind power scenarios during the generation of these scenarios, and conducting targeted generation and analysis, will help explore the impact of specific wind power scenarios on the power system and improve the power system's ability to absorb different wind power scenarios.

[0003] To generate wind power scenarios, Chinese Patent Publication No. CN107039981A uses Latin hypercube sampling to generate wind power scenarios, while simultaneously using scenario reduction techniques to obtain a small number of wind power scenarios with higher probability measures. Chinese Patent Publication No. CN117767387A, based on the probability distribution curve of wind power output prediction errors, uses the Monte Carlo method to uniformly sample prediction error values, and then calculates the actual output scenario of the wind farm using inverse transform sampling and the output prediction error formula. However, due to the complex characteristics of wind power scenarios, wind power scenarios obtained through traditional probability sampling or Monte Carlo simulation often differ significantly from real scenarios in certain features, making it difficult to ensure the similarity between the generated scenario and the real scenario. With continuous improvements in computing power and algorithms, researchers have begun to focus on using artificial intelligence methods for wind power scenario generation. Chinese Patent Publication No. CN111008455A proposes a wind power scenario generation method based on conditional generative adversarial networks (GANs), which improves the model's ability to capture the temporal features of wind power scenarios by replacing the convolutional layers in the generator and discriminator of the GAN with temporal enhancement modules. However, the time-series enhancement module used in this method is based on a traditional difference model, which makes it difficult to capture the complex and variable high-dimensional features of real-world wind power scenarios. Furthermore, existing methods often classify wind power output scenarios based on statistical features such as total output and rate of change, failing to reflect the integration difficulty of wind power scenarios and thus failing to meet the urgent needs of power systems to study wind power scenarios with varying integration difficulties. Summary of the Invention

[0004] The present invention aims to at least partially solve one of the technical problems in the related art.

[0005] To address this, the present invention proposes a method for generating specific wind power scenarios based on LSTM-ACGAN, which can generate specific wind power scenarios for different levels of grid integration difficulty.

[0006] Another objective of this invention is to propose a specific wind power scenario generation device based on LSTM-ACGAN.

[0007] To achieve the above objectives, this invention proposes a method for generating specific wind power scenarios based on LSTM-ACGAN, comprising:

[0008] Based on historical load data and historical wind power output data of the power system, historical load scenario datasets and historical wind power scenario datasets are obtained, and the historical load scenario datasets are clustered to obtain typical load scenario datasets.

[0009] Based on the typical loads in the typical load scenario dataset, the wind curtailment rate corresponding to each wind power scenario in the historical wind power scenario dataset is obtained by solving the optimization problem, and the wind power scenarios are classified according to the range of the wind curtailment rate to obtain the original wind power scenario dataset.

[0010] An LSTM-ACGAN model is constructed based on Bi-LSTM and a fully connected layer neural network; wherein, the LSTM-ACGAN model includes a generator, a discriminator, and a classifier;

[0011] The LSTM-ACGAN model was trained using the original wind power scene dataset, and the generator was used as the wind power scene generator after training.

[0012] Based on typical load scenarios and the types of wind power scenarios to be generated, wind power scenarios similar to real wind power scenarios are generated using the corresponding wind power scenario generator.

[0013] To achieve the above objectives, another aspect of the present invention proposes a specific wind power scenario generation device based on LSTM-ACGAN, comprising:

[0014] The data processing module is used to obtain historical load scenario datasets and historical wind power scenario datasets based on historical load data and historical wind power output data of the power system, and to cluster the historical load scenario datasets to obtain typical load scenario datasets.

[0015] The wind power scenario classification module is used to obtain the wind curtailment rate corresponding to each wind power scenario in the historical wind power scenario dataset by solving the optimization problem based on the typical load in the typical load scenario dataset, and classify the wind power scenarios according to the range of the wind curtailment rate to obtain the original wind power scenario dataset.

[0016] The model building module is used to build an LSTM-ACGAN model based on Bi-LSTM and fully connected layer neural networks; wherein, the LSTM-ACGAN model includes a generator, a discriminator, and a classifier;

[0017] The model training module is used to train the LSTM-ACGAN model using the original wind power scene dataset, and after training, the generator is used as a wind power scene generator.

[0018] The wind power scenario generation module is used to generate wind power scenarios similar to real wind power scenarios based on typical load scenarios and the type of wind power scenario to be generated, using the corresponding wind power scenario generator.

[0019] The method and apparatus for generating specific wind power scenarios based on LSTM-ACGAN in this invention can generate specific wind power scenarios with different absorption difficulties by training LSTM-ACGAN with classified wind power scenarios, thereby meeting the power system's analysis needs for wind power scenarios with different absorption difficulties.

[0020] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0021] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0022] Figure 1 This is a flowchart of a specific wind power scenario generation method based on LSTM-ACGAN according to an embodiment of the present invention;

[0023] Figure 2 This is a structural diagram of a specific wind power scenario generation device based on LSTM-ACGAN according to an embodiment of the present invention. Detailed Implementation

[0024] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0025] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0026] The following describes, with reference to the accompanying drawings, a method and apparatus for generating specific wind power scenarios based on LSTM-ACGAN, according to embodiments of the present invention.

[0027] Figure 1This is a flowchart of a specific wind power scenario generation method based on LSTM-ACGAN according to an embodiment of the present invention, such as... Figure 1 As shown, the method includes:

[0028] S1. Based on historical load data and historical wind power output data of the power system, a historical load scenario dataset and a historical wind power scenario dataset are obtained, and the historical load scenario dataset is clustered to obtain a typical load scenario dataset.

[0029] Specifically, based on historical load data and historical wind power output data of the power system, a historical wind power scenario dataset D is obtained through data processing. w0 and typical load scenario dataset D d This includes the following steps:

[0030] S11, for a time length of T w A one-dimensional historical wind power output sequence is divided into n parts according to the target scenario time length T. w There are historical wind power scenarios of length T, consisting of dimensions (n). w Historical wind power scenario dataset D (T) w0 ,in

[0031] S12, for a time length of T d A one-dimensional historical load sequence is divided into n parts according to the target scenario time length T. d There are historical load scenarios of length T, consisting of dimensions (n). d Historical load scenario dataset D of T) d0 ,in

[0032] S13, for D d0 n d Given a historical load scenario of length T, we divide it into M clusters using K-means clustering, denoted as Di, Di, and Di, respectively. d1 D d2 D dM Take the cluster centers P of these M clusters. d1 P d2 , ..., P dM Dataset D constituting a typical load scenario d ={P dm}, where m = 1, 2, ..., M. In one embodiment of the present invention, T is taken as... w =17520, T d =8760, T=144, M=3.

[0033] S2, based on the typical loads in the typical load scenario dataset, obtains the wind curtailment rate corresponding to each wind power scenario in the historical wind power scenario dataset by solving the optimization problem, and classifies the wind power scenarios according to the range of the wind curtailment rate to obtain the original wind power scenario dataset.

[0034] Specifically, for the typical loads in the aforementioned typical load scenario dataset, wind power scenarios are classified by solving an optimization problem to obtain the wind power scenario dataset D. w This includes the following steps:

[0035] S21, Construct the mathematical model for the optimization problem, the specific formula is as follows:

[0036] Objective function:

[0037]

[0038] Optimization problem constraints:

[0039] 0≤p w (t)≤P w (t)

[0040]

[0041] P gmin (i)X(i,t)≤P g (i,t)≤P gmax (i)X(i,t)

[0042] R d (i)≤P g (i,t+1)-P g (i,t)≤R u (i)

[0043]

[0044] X(i,t)-X(i,t-1)-Y(i,t)+Z(i,t)=0

[0045] -X(i,t)-X(i,t-1)+Y(i,t)≤0

[0046] X(i,t)+X(i,t-1)+Y(i,t)≤2

[0047] -X(i,t)-X(i,t-1)+Z(i,t)≤0

[0048] X(i,t)+X(i,t-1)+Z(i,t)≤2

[0049] Where t represents a point in time between 1 and T; P w P represents a specific wind power scenario.w (t) represents the wind power scenario P w The maximum transmittable power at time t, p w (t) represents the wind power scenario P w The actual power output at time t; N gen This indicates the number of thermal power units; i represents the unit number, i = 1, 2, ..., N gen ;P g (i,t) represents the active power output of the i-th thermal power unit at time t; P d P represents a typical load scenario. d (t) represents the load scenario P d The load at time t; X(i,t) represents the operating state of unit i at time t, which is 1 if it is running and 0 if it is shut down; Y(i,t) represents the startup process of unit i at time t, which is 1 if the unit is starting up and 0 otherwise; Z(i,t) represents the shutdown process of unit i at time t, which is 1 if the unit is shutting down and 0 otherwise; P re P is the system's spinning reserve factor; gmax (i) represents the maximum active power output of unit i, P gmin (i) represents the minimum active power output of unit i; R d (i) represents the maximum downhill speed of the i-th thermal power unit, R u (i) represents the maximum uphill speed of the i-th thermal power unit; T on and T off These represent the minimum start-up time and minimum shutdown time of the unit, respectively. In one embodiment of the invention, the power system under consideration is the IEEE 39-bus system, and the generator on bus 30 is converted into a wind farm, then N gen =9. Take P. re =5%, R d (i) = 0.05 × P gmax (i), R u (i) = 0.05 × P gmax (i), T on =T off =2h. P gmax (i) and P gmin (i) is given by the IEEE 39-node test system.

[0050] S22, for typical load scenario P dm , respectively for D w0 Each wind power scenario P wl By solving the optimization problem, we obtain the wind power scenario P at times 1 to T. wl Actual power output p wl (t), where l = 1, 2, ..., nw This indicates the sequence number of the wind power scenario. Then, P is calculated for each wind power scenario. wl The corresponding wind curtailment rate η l The calculation formula is as follows:

[0051]

[0052] S23, n is determined according to the range of wind curtailment rate. w The wind power scenarios are divided into K classes, and the dataset for each class of wind power scenarios is denoted as D. wm_1 D wm_2 D wm_K Together they constitute a typical load scenario P dm The corresponding original wind power scenario dataset D wm ={D wm_k}, where k = 1, 2, ... K. In one embodiment of the present invention, K = 4. The relationship between the wind curtailment rate range and the classification is defined as follows: a wind curtailment rate of 0 is defined as Class 1, a wind curtailment rate between (0, 25%) is defined as Class 2, a wind curtailment rate between (25%, 50%) is defined as Class 3, and a wind curtailment rate greater than 50% is defined as Class 4.

[0053] S24, for all typical load scenarios P d1 P d2 , ..., P dM Perform the operations in (S22) to (S23) respectively to obtain D. w1 D w2 D wM Together, they constitute the original wind power scenario dataset D. w ={D wm}, where m = 1, 2, ..., M.

[0054] S3, construct an LSTM-ACGAN model based on Bi-LSTM and a fully connected layer neural network; wherein, the LSTM-ACGAN model includes a generator, a discriminator, and a classifier.

[0055] Specifically, the LSTM-ACGAN model is constructed based on Bi-LSTM and fully connected neural networks, including the following steps:

[0056] S31, Construct generator M G The input is a Gaussian distribution with dimensions Z, p. z Given Gaussian noise z of (z) and type labels c = 1, 2, ..., K of the wind power scenario to be generated, the output is the generated wind power scenario P. wG The specific calculation formula is as follows:

[0057] P wG =M G(z,c)=FC(LSTM(z+Embedding(c)))

[0058] Where Embedding(·) represents the embedding layer, which maps the integer c to a vector with the same shape as z; LSTM(·) represents N L Bi-LSTM neural network with N layers; FC(·) represents N F The network consists of a fully connected layer, where each layer comprises a linear mapping (Linear(·)) and an activation function (ReLU(·)). In one embodiment of the invention, Z = 500, N... L =2, N F =4.

[0059] S32, Construct the discriminator M D Input y represents the actual wind power scenario P. w Or generate wind power scenario P wG The output is the judgment result S. S=1 represents the scene being judged as real, and S=0 represents the scene being judged as generated. The specific calculation formula is as follows:

[0060] S = M D (y) = Sigmoid(Linear(FC(y)))

[0061] The formula for calculating the Sigmoid function is as follows:

[0062] S33, Construct classifier M C Input y represents the actual wind power scenario P. w Or generate wind power scenario P wG The output is the classification result. The specific calculation formula is as follows:

[0063]

[0064] Where the function Softmax(·) is the normalized exponential function, and M C With M D FC(·) network function parameter sharing.

[0065] S34 defines the network's loss function, and the specific calculation formula is as follows:

[0066] L S =E[logP(S=1|P w )]+E[logP(S=0|P wG )]

[0067]

[0068] L G =-(LC +L S )

[0069] L D =-(L C -L S )

[0070] Where E[·] represents the mean, L S and L C These represent discrimination accuracy and classification correctness, respectively. The loss function L is used. G Update M G The parameters are used with the loss function L. D Update M D and M C The parameters.

[0071] S4 uses the original wind power scene dataset to train the LSTM-ACGAN model, and after training, the generator is used as the wind power scene generator.

[0072] Specifically, using the wind power scenario dataset D w Training LSTM-ACGAN to obtain a wind power scene generator includes the following steps:

[0073] S41, Set training parameters, including the number of training epochs Ep, batch size B, optimizer, and learning rate lr. In one embodiment of the present invention, Ep = 10000, B = 64, the optimizer is Adam, and lr = 0.0001.

[0074] S42, for the m-th typical load scenario P dm Using its corresponding original wind power scenario dataset D wm Train the corresponding LSTM-ACGAN model. After the model is trained, retrieve the generator M of the model. Gm As the m-th typical load scenario P dm The corresponding wind power scene generator.

[0075] S43, for all typical load scenarios P d1 P d2 , ..., P dM The operation steps in (S42) are performed respectively to obtain the wind power scenario generator corresponding to each typical load scenario, which together constitute the wind power scenario generator M for all typical loads. G ={M Gm}, where m = 1, 2, ..., M.

[0076] S5, based on typical load scenarios and the type of wind power scenario to be generated, uses the corresponding wind power scenario generator to generate wind power scenarios similar to real wind power scenarios.

[0077] Specifically, for specific typical load scenarios and specific wind power scenario types to be generated, a large number of specific generated wind power scenarios similar to specific real wind power scenarios are generated using the corresponding wind power scenario generator, including the following steps:

[0078] S51, determine the load scenario type m, the wind power scenario type c to be generated, and the number of wind power scenarios N to be generated. For the m-th typical load scenario, select its corresponding wind power scenario generator M. Gm .

[0079] S52, towards M Gm Given N Z-dimensional Gaussian noises z and type labels c, the generator will output N values ​​P for typical load scenarios. dm The wind power scenario of type c is generated to complete the generation of a specific wind power scenario. In one embodiment of the present invention, N = 10000.

[0080] According to embodiments of the present invention, the method for generating specific wind power scenarios based on LSTM-ACGAN uses a Generative Adversarial Network (ACGAN) with an auxiliary classifier as the basic model. Its deep neural network can accurately capture the characteristics of the power output probability distribution of wind power scenarios. By introducing a bidirectional LSTM network (Bi-LSTM) into the ACGAN generator, the generator can capture the complex and variable temporal characteristics of wind power scenarios, making the generated wind power scenarios similar to actual wind power scenarios in both power output distribution and temporal characteristics. This method processes the original wind power scenarios by solving an optimization problem, obtaining the wind curtailment rate corresponding to the wind power scenarios under different typical load scenarios. The absorption difficulty of the wind power scenarios is determined based on the range of the wind curtailment rate, and the scenarios are then classified. The LSTM-ACGAN is trained using these classified wind power scenarios, enabling the model to generate specific wind power scenarios with different absorption difficulties, thus meeting the power system's analytical needs for wind power scenarios with varying absorption difficulties.

[0081] To achieve the above embodiments, such as Figure 2 As shown, this embodiment also provides a specific wind power scenario generation device 10 based on LSTM-ACGAN, including:

[0082] The data processing module 100 is used to obtain historical load scenario datasets and historical wind power scenario datasets based on historical load data and historical wind power output data of the power system, and to cluster the historical load scenario datasets to obtain typical load scenario datasets.

[0083] The wind power scenario classification module 200 is used to obtain the wind curtailment rate corresponding to each wind power scenario in the historical wind power scenario dataset by solving the optimization problem based on the typical load in the typical load scenario dataset, and classify the wind power scenarios according to the range of the wind curtailment rate to obtain the original wind power scenario dataset.

[0084] The model building module 300 is used to build an LSTM-ACGAN model based on Bi-LSTM and fully connected layer neural networks; wherein, the LSTM-ACGAN model includes a generator, a discriminator and a classifier;

[0085] The model training module 400 is used to train the LSTM-ACGAN model using the original wind power scene dataset, and after training, the generator is used as a wind power scene generator.

[0086] The wind power scenario generation module 500 is used to generate wind power scenarios similar to real wind power scenarios based on typical load scenarios and the type of wind power scenario to be generated, using the corresponding wind power scenario generator.

[0087] The wind power scenario generation device based on LSTM-ACGAN according to embodiments of the present invention uses a generative adversarial network (ACGAN) with an auxiliary classifier as the basic model. Its deep neural network can accurately capture the characteristics of the power output probability distribution of wind power scenarios. By introducing a bidirectional LSTM network (Bi-LSTM) into the ACGAN generator, the generator can capture the complex and variable temporal characteristics of wind power scenarios, making the generated wind power scenarios similar to actual wind power scenarios in both power output distribution and temporal characteristics. The original wind power scenarios are processed using an optimization problem approach to obtain the wind curtailment rates corresponding to different typical load scenarios. The absorption difficulty of the wind power scenarios is determined based on the range of the wind curtailment rates, and the scenarios are classified. The LSTM-ACGAN is trained using the classified wind power scenarios, enabling the model to generate specific wind power scenarios with different absorption difficulties, meeting the power system's analysis needs for wind power scenarios with varying absorption difficulties.

[0088] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0089] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A method for generating specific wind power scenarios based on LSTM-ACGAN, characterized in that, include: S1. Based on historical load data and historical wind power output data of the power system, a historical load scenario dataset and a historical wind power scenario dataset are obtained, and the historical load scenario dataset is clustered to obtain a typical load scenario dataset. S2, based on the typical load in the typical load scenario dataset, the wind curtailment rate corresponding to each wind power scenario in the historical wind power scenario dataset is obtained by solving the optimization problem, and the wind power scenarios are classified according to the range of the wind curtailment rate to obtain the original wind power scenario dataset. S2 specifically includes: S21, Construct the mathematical model for the optimization problem, as shown in the following formula: Objective function: Optimization problem constraints: 0≤p w (t)≤P w (t) P gmin (i)X(i,t)≤P g (i,t)≤P gmax (i)X(i,t) R d (i)≤P g (i,t+1)-P g (i,t)≤R u (i) X(i,t)-X(i,t-1)-Y(i,t)+Z(i,t)=0 -X(i,t)-X(i,t-1)+Y(i,t)≤0 X(i,t)+X(i,t-1)+Y(i,t)≤2 -X(i,t)-X(i,t-1)+Z(i,t)≤0 X(i,t)+X(i,t-1)+Z(i,t)≤2 Where t represents a point in time between 1 and T; P w P represents a specific wind power scenario. w (t) represents the wind power scenario P w The maximum transmittable power at time t, p w (t) represents the wind power scenario P w The actual power output at time t; N gen This indicates the number of thermal power units; i represents the unit number, i = 1, 2, ..., N gen ;P g (i,t) represents the active power output of the i-th thermal power unit at time t; P d P represents a typical load scenario. d (t) represents the load scenario P d The load at time t; X(i,t) represents the operating state of unit i at time t, which is 1 if it is running and 0 if it is shut down; Y(i,t) represents the startup process of unit i at time t, which is 1 if the unit is starting up and 0 otherwise; Z(i,t) represents the shutdown process of unit i at time t, which is 1 if the unit is shutting down and 0 otherwise; P re P is the system's spinning reserve factor; gmax (i) represents the maximum active power output of unit i, P gmin (i) represents the minimum active power output of unit i; R d (i) represents the maximum downhill speed of the i-th thermal power unit, R u (i) represents the maximum uphill speed of the i-th thermal power unit; T on and T off These represent the minimum start-up time and minimum shutdown time of the unit, respectively. S22, for typical load scenario P dm , respectively for D w0 Each wind power scenario P wl By solving the optimization problem, we obtain the wind power scenario P at times 1 to T. wl Actual power output p wl (t), where l = 1, 2, ..., n w This indicates the sequence number of the wind power scenario; then, P is calculated for each wind power scenario. wl The corresponding wind curtailment rate η l The calculation formula is as follows: S23, n is determined according to the range of wind curtailment rate. w The wind power scenarios are divided into K classes, and the dataset for each class of wind power scenarios is denoted as D. wm_1 D wm_2 D wm_K Together they constitute a typical load scenario P dm The corresponding original wind power scenario dataset D wm ={D wm_k }, where k = 1, 2, ..., K; S24, for all typical load scenarios P d1 P d2 , ..., P dM Perform the operations in (S22) to (S23) respectively to obtain D. w1 D w2 D wM Together, they constitute the original wind power scenario dataset D. w ={D wm }, where m = 1, 2, ..., M; S3, Construct an LSTM-ACGAN model based on Bi-LSTM and a fully connected layer neural network; wherein, the LSTM-ACGAN model includes a generator, a discriminator, and a classifier; S4. The LSTM-ACGAN model is trained using the original wind power scene dataset, and the generator is used as the wind power scene generator after training. S4 specifically includes: S41, set the training parameters, including the number of training ep, batch size B, optimizer, and learning rate lr; S42, for the m-th typical load scenario P dm Using its corresponding original wind power scenario dataset D wm The corresponding LSTM-ACGAN model is trained. After the model is trained, the generator M of the model is retrieved. Gm As the m-th typical load scenario P dm The corresponding wind power scene generator; S43, for all typical load scenarios P d1 P d2 , ..., P dM The operation steps in (S42) are performed respectively to obtain the wind power scenario generator corresponding to each typical load scenario, which together constitute the wind power scenario generator M for all typical loads. G ={M Gm }, where m = 1, 2, ..., M; S5, based on typical load scenarios and the types of wind power scenarios to be generated, uses the corresponding wind power scenario generator to generate wind power scenarios similar to real wind power scenarios. S5 specifically includes: S51, determine the load scenario type m, the wind power scenario type c to be generated, and the number of wind power scenarios N to be generated. For the m-th typical load scenario, select the corresponding wind power scenario generator M. Gm ; S52, towards M Gm Given N Z-dimensional Gaussian noises z and type labels c, the generator will output N values ​​P for typical load scenarios. dm Generate wind power scenarios of type c, and complete the generation of specific wind power scenarios.

2. The method according to claim 1, characterized in that, Historical load scenario datasets and historical wind power scenario datasets were obtained based on historical power system load data and historical wind power output data. These historical load scenario datasets were then clustered to obtain typical load scenario datasets, including: S11, based on a time length of T w The one-dimensional historical wind power output sequence is divided into n parts according to the target scenario time length T. w There are historical wind power scenarios of length T, consisting of dimensions (n). w Historical wind power scenario dataset D (T) w0 ,in S12, based on a time length of T d A one-dimensional historical load sequence is divided into n parts according to the target scenario time length T. d There are historical load scenarios of length T, consisting of dimensions (n). d Historical load scenario dataset D of T) d0 ,in S13, for D d0 n d Given a historical load scenario of length T, we divide it into M clusters using K-means clustering, denoted as Di. d1 D d2 D dM Take the cluster centers P of M clusters. d1 P d2 , ..., P dM Dataset D constituting a typical load scenario d ={P dm }, where m = 1, 2, ..., M.

3. The method according to claim 1, characterized in that, The LSTM-ACGAN model is constructed based on Bi-LSTM and fully connected neural networks, including: S31, Construct generator M G The input is a Gaussian distribution with dimensions Z, p. z Given Gaussian noise z of (z) and type labels c = 1, 2, ..., K of the wind power scenario to be generated, the output is the generated wind power scenario P. wG The calculation formula is as follows: P wG =M G (z,c)=FC(LSTM(z+Embedding(c))) Where Embedding(·) represents the embedding layer, which maps the integer c to a vector with the same shape as z; LSTM(·) represents N L Bi-LSTM neural network with N layers; FC(·) represents N F A fully connected network with layers, each layer consisting of a linear mapping Linear(·) and an activation function ReLU(·); S32, Construct the discriminator M D Input y represents the actual wind power scenario P. w Or generate wind power scenario P wG The output is the judgment result S; S=1 represents the scene being judged as real, and S=0 represents the scene being judged as generated; the calculation formula is as follows: S=M D (y)=Sigmoid(Linear(FC(y))) The formula for calculating the function Sigmoid(·) is as follows: S33, Construct classifier M C Input y represents the actual wind power scenario P. w Or generate wind power scenario P wG The output is the classification result. The calculation formula is as follows: Wherein, the function Softmax(·) is the normalized exponential function, and M C With M D FC(·) network function parameter sharing; S34 defines the network's loss function, and the specific calculation formula is as follows: L S =E[logP(S=1|P w )]+E[logP(S=0|P wG )] L G =-(L C +L S ) L D =-(L C -L S ) Where E[·] represents the mean, L S and L C These represent discrimination accuracy and classification correctness, respectively; the loss function L is used. G Update M G The parameters are used with the loss function L. D Update M D and M C The parameters.

4. A device for generating specific wind power scenarios based on LSTM-ACGAN, characterized in that, The apparatus implements the method as described in claim 1, the apparatus comprising: The data processing module is used to obtain historical load scenario datasets and historical wind power scenario datasets based on historical load data and historical wind power output data of the power system, and to cluster the historical load scenario datasets to obtain typical load scenario datasets. The wind power scenario classification module is used to obtain the wind curtailment rate corresponding to each wind power scenario in the historical wind power scenario dataset by solving the optimization problem based on the typical load in the typical load scenario dataset, and classify the wind power scenarios according to the range of the wind curtailment rate to obtain the original wind power scenario dataset. The model building module is used to build an LSTM-ACGAN model based on Bi-LSTM and fully connected layer neural networks; wherein, the LSTM-ACGAN model includes a generator, a discriminator, and a classifier; The model training module is used to train the LSTM-ACGAN model using the original wind power scene dataset, and after training, the generator is used as a wind power scene generator. The wind power scenario generation module is used to generate wind power scenarios similar to real wind power scenarios based on typical load scenarios and the type of wind power scenario to be generated, using the corresponding wind power scenario generator.

Images