A method and device for processing wind speed power data of a wind farm

Abstract

The present application provides a data processing method and device for wind speed and power data of a wind farm, which belongs to the field of electrical digital data processing technology and is applied to the field of renewable energy power generation. The method includes: obtaining historical real sample data of wind speed and power of a target wind farm; inputting the data into a generative adversarial network model offline for pre-training until the generative adversarial network model converges to obtain a pre-trained model of the generative adversarial network; adjusting the pre-trained model of the generative adversarial network to obtain a target model of the generative adversarial network; generating random noise data that satisfies the same distribution as the noise sample training data set; inputting the random noise data into the target model of the adversarial network to obtain a wind speed and power data sample that meets the real data distribution of the target wind farm. The above method can efficiently expand the wind speed and power data used for training, thereby improving the prediction accuracy of the wind power and power generation model of the target wind farm.

Description

Technical Field

[0001] The present application belongs to the field of electrical digital data processing technology, is applied to the field of renewable energy power generation, and particularly relates to a data processing method and device for wind speed power data of a wind farm. Background Art

[0002] With the continuous development of wind power generation, more and more wind farms are put into operation and connected to the grid. During the operation process, more and more wind farms optimize the operation scheduling and power generation income of wind farms by predicting wind power and power generation. Machine learning is currently the main way to predict wind power and power generation, but machine learning needs to rely on training data, that is, a large amount of basic wind speed-power data is required during the training process. However, the commissioning time of new wind farms is relatively short, and the wind speed-power data generated is relatively small, which is bound to cause the prediction model to easily overfit on small sample training data sets.

[0003] There is currently no effective solution for how to expand the wind speed and power data used for training to improve the prediction accuracy of wind power and generation models. Summary of the invention

[0004] The purpose of the present application is to provide a data processing method and device for wind speed and power data of a wind farm, which can efficiently expand the wind speed and power data used for training, thereby improving the prediction accuracy of the wind power and power generation model of the target wind farm.

[0005] The present application provides a method and device for processing wind speed and power data of a wind farm, which is implemented as follows:

[0006] A method for processing wind speed power data of a wind farm, the method comprising:

[0007] Obtain historical real sample data of wind speed and power of the target wind farm;

[0008] Preprocessing the historical real sample data of the wind speed power to obtain preprocessed real sample data of the wind speed power, and randomly generating noise data;

[0009] Construct a generative adversarial network model to capture the distribution of real sample data of wind speed and power;

[0010] The real sample data after preprocessing is used to construct a real sample training data set, and the noise sample training data set is constructed using the randomly generated noise data;

[0011] Inputting the real sample training data set and the noise sample training data set into a generative adversarial network model offline for pre-training until the generative adversarial network model converges, thereby obtaining a pre-trained model of the generative adversarial network;

[0012] Inputting the wind speed power data generated by the real-time operation of the target wind farm into the pre-trained model of the generative adversarial network in real time to adjust the pre-trained model of the generative adversarial network to obtain a target model of the generative adversarial network;

[0013] Generate random noise data that satisfies the same distribution as the noise sample training data set;

[0014] The random noise data is input into the target model of the adversarial network to obtain wind speed power data samples that conform to the real data distribution of the target wind farm.

[0015] In one embodiment, the historical real sample data includes: a timestamp field, a wind speed field, and a power field;

[0016] Accordingly, preprocessing the historical real sample data of wind speed power includes:

[0017] Perform default value processing, outlier processing and normalization processing on wind speed field and power field;

[0018] The timestamp field, the processed wind speed field and the power field are combined into pre-processed wind speed and power real sample data.

[0019] In one embodiment, a generative adversarial network model for capturing the distribution of real sample data of wind speed and power is constructed, including:

[0020] A generative adversarial network model containing a generator and a discriminator is constructed, wherein the generator is used to learn the distribution of real historical sample data of wind speed and power to generate synthetic data close to the real data distribution, and the discriminator is used to learn the difference between the distribution of real historical sample data of wind speed and power and the synthetic data to identify the synthetic data.

[0021] The generator receives random noise data satisfying multimodal distribution and outputs synthetic data according to the random noise data;

[0022] The discriminator receives historical real sample data and synthetic data, and outputs a first data group and a second data group, wherein the first data group includes: a probability for characterizing that the data is historical real sample data and a parameter value of the discriminator, and the second data group includes: a probability for characterizing that the data is synthetic data and a parameter value of the discriminator.

[0023] In one embodiment, the generator includes: a sequence input module, a sequence parsing module and a sequence generation module, wherein the sequence input module is used to input historical real sample data in the form of a one-dimensional noise vector with a timestamp, and encode and mark the position sequence information of each field in the sequence and encode and mark the position sequence information of each field in the input data, the sequence parsing module is used to perform sequence normalization, sequence subspace information extraction, sequence normalization, and full-connection linear transformation on the encoded and marked data sequence to extract time series features, and the sequence generation module is used to input the extracted time series features into a two-dimensional convolutional layer in the sequence generation module and output to obtain synthetic data, wherein the fully connected linear transformation in the sequence parsing module is a two-layer neural network, and the two-layer neural network is used to first perform linear transformation, then use a Relu function for nonlinear transformation, and then perform linear transformation;

[0024] The discriminator includes: a mixed sequence input module, a mixed sequence parsing module and a binary classification recognition module, wherein the mixed sequence input module is used to receive historical real sample data in the form of a one-dimensional noise vector with a timestamp or synthetic data output by a generator, and encode and mark the position order information of each field in the sequence and encode and mark the position order information of each field in the input data; the mixed sequence parsing module is used to perform mixed sequence normalization, mixed sequence subspace information extraction, mixed sequence normalization, and full-connected linear transformation on the encoded and marked data sequence to extract time series features; the binary classification recognition module is used to perform binary classification recognition on the time series features, wherein the binary classification recognition module uses a sigmoid function to map the sequence value to the [0, 1] interval, and divides the output probability value into two categories: real data and synthetic data.

[0025] In one embodiment, the real sample training data set and the noise sample training data set are input into the generative adversarial network model offline for pre-training until the generative adversarial network model converges, to obtain a pre-trained model of the generative adversarial network, including:

[0026] Control the generator and discriminator to compete with each other and alternately update the weight parameters of the deep neural network;

[0027] The generative adversarial network model is pre-trained according to the following objective function:

[0028]

[0029] Where D represents the discriminator, G represents the generator, z represents noise data, x represents real sample training data, D(x,γ) represents the probability that the discriminator distinguishes the data in the real sample training data set as real data, D(G(z,θ),γ) represents the probability that the discriminator distinguishes the synthetic data as real data, θ represents the parameter of the generator, and γ represents the parameter of the discriminator; when the input data is the real sample training data set, the discriminator maximizes log(D(x,γ)) and log(1-D(G(z,θ),γ)), and controls the output of the discriminator D(G(z,θ),γ) to be close to 1; when the input data is synthetic data, the generator minimizes log(1-D(G(z,θ),γ)), and controls the output of the discriminator D(G(z,θ),γ) to be close to 0.

[0030] In one embodiment, the real sample training data set and the noise sample training data set are input into the generative adversarial network model offline for pre-training until the generative adversarial network model converges, to obtain a pre-trained model of the generative adversarial network, including:

[0031] Collecting a predetermined number of real sample data from the real sample training data set;

[0032] Collecting a predetermined number of noise sample data from the noise sample training data set;

[0033] The stochastic gradient algorithm is used to update the weight parameters of the discriminator by increasing or decreasing the gradient based on a predetermined number of real sample data and a predetermined number of noise sample data:

[0034]

[0035] Wherein, m represents the predetermined number, D represents the discriminator, G represents the generator, z represents the noise data, x represents the real sample training data, D(x,γ) represents the probability that the discriminator distinguishes the data in the real sample training data set as the real data, D(G(z,θ),γ) represents the probability that the discriminator distinguishes the synthetic data as the real data, θ represents the parameters of the generator, and γ represents the parameters of the discriminator.

[0036] A data processing device for wind speed power data of a wind farm, comprising:

[0037] An acquisition module is used to acquire historical real sample data of wind speed and power of a target wind farm;

[0038] A preprocessing module is used to preprocess the historical real sample data of the wind speed power to obtain the preprocessed real sample data of the wind speed power and randomly generate noise data;

[0039] The first building module is used to build a generative adversarial network model for capturing the distribution of real sample data of wind speed and power;

[0040] The second construction module is used to construct a real sample training data set using the preprocessed wind speed power real sample data, and to construct a noise sample training data set using the randomly generated noise data;

[0041] An input module, used to input the real sample training data set and the noise sample training data set offline into the generative adversarial network model for pre-training until the generative adversarial network model converges, thereby obtaining a pre-trained model of the generative adversarial network;

[0042] An adjustment module, used for inputting the wind speed power data generated by the real-time operation of the target wind farm into the pre-trained model of the generative adversarial network in real time, so as to adjust the pre-trained model of the generative adversarial network and obtain a target model of the generative adversarial network;

[0043] A first generating module, used to generate random noise data that satisfies the same distribution as the noise sample training data set;

[0044] The second generation module is used to input the random noise data into the target model of the adversarial network to obtain wind speed power data samples that conform to the real data distribution of the target wind farm.

[0045] In one embodiment, the first building block comprises:

[0046] A construction unit is used to construct a generative adversarial network model including a generator and a discriminator, wherein the generator is used to learn the distribution of historical real sample data of wind speed and power to generate synthetic data close to the distribution of real data, and the discriminator is used to learn the difference between the distribution of historical real sample data of wind speed and power and the distribution of synthetic data to identify the synthetic data;

[0047] A first output unit, configured to receive random noise data satisfying a multi-modal distribution through a generator, and output synthetic data according to the random noise data;

[0048] The second output unit is used to receive historical real sample data and synthetic data through the discriminator, and output a first data group and a second data group, wherein the first data group includes: a probability for characterizing that the data is historical real sample data and a parameter value of the discriminator, and the second data group includes: a probability for characterizing that the data is synthetic data and a parameter value of the discriminator.

[0049] An electronic device comprises a processor and a memory for storing instructions executable by the processor, wherein the steps of the above method are implemented when the processor executes the instructions.

[0050] A computer-readable storage medium stores a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0051] The present application provides a data processing method for wind speed and power data of a wind farm, which obtains historical real sample data of wind speed and power of a target wind farm; then, preprocesses the historical real sample data of wind speed and power to obtain preprocessed real sample data of wind speed and power, and randomly generates noise data; then, constructs a generative adversarial network model for capturing the distribution of the real sample data of wind speed and power; uses the preprocessed real sample data of wind speed and power to construct a real sample training data set, and uses the randomly generated noise data to construct a noise sample training data set; inputs the real sample training data set and the noise sample training data set offline into a generative adversarial network model for pre-training until the generative adversarial network model converges, thereby obtaining a pre-trained model of the generative adversarial network; then, inputs the wind speed and power data generated by the real-time operation of the target wind farm into the pre-trained model of the generative adversarial network in real time, so as to adjust the pre-trained model of the generative adversarial network, thereby obtaining a target model of the generative adversarial network; generates random noise data that satisfies the same distribution as the noise sample training data set; inputs the random noise data into the target model of the adversarial network, thereby obtaining a wind speed and power data sample that conforms to the real data distribution of the target wind farm. The above method can efficiently expand the wind speed and power data used for training, thereby improving the prediction accuracy of the wind power and power generation model of the target wind farm. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0053] Figure 1 It is a method flow chart of an embodiment of a method for processing wind speed power data of a wind farm provided by the present application;

[0054] Figure 2 It is a model structure diagram of the generative adversarial network provided by this application;

[0055] Figure 3 It is a network structure diagram of the generator G provided by this application;

[0056] Figure 4 is a network structure diagram of the discriminator D provided in this application;

[0057] Figure 5It is a hardware structure block diagram of an electronic device for a data processing method of wind speed power data of a wind farm provided by the present application;

[0058] Figure 6 It is a schematic diagram of the module structure of an embodiment of a data processing device for wind speed power data of a wind farm provided by the present application. DETAILED DESCRIPTION

[0059] In order to enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.

[0060] It should be noted that the user-related information and data involved in the embodiments of this specification are all information and data authorized by the user or fully authorized by relevant parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users or relevant parties to choose to authorize or refuse.

[0061] It should also be noted that in the embodiments of this specification, certain software, components, models and other existing solutions in the industry may be mentioned, which should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of this application, but it does not mean that the applicant has or will necessarily use the solution.

[0062] Taking into account the objective differences between the surrounding wind farms and the newly built wind farms in terms of environment, wind turbine brand, wind turbine model, etc., the prediction effect after data fusion is not good. In addition, the data with low similarity in the surrounding wind farms will be eliminated during the data fusion process, which destroys the time series characteristics of the data, which will also cause the prediction accuracy to be low. Furthermore, the newly built wind farm has a short operation time and the generated wind speed and power data is relatively small, which will inevitably lead to the prediction model being prone to overfitting problems on small sample training data sets. Based on this, in this example, a data enhancement method is provided to utilize the existing small sample data of wind speed and power in the newly built wind farm to generate wind speed and power sample data that is more in line with the spatiotemporal characteristics of the newly built wind farm, so as to achieve the purpose of effectively expanding the training data set. Specifically, a data enhancement method for insufficient historical sample data of wind speed and power in a newly built wind farm is provided, which may include the following steps:

[0063] Figure 1It is a method flow chart of an embodiment of the data processing method of wind speed power data of a wind farm provided by the present application. Although the present application provides method operation steps or device structures as shown in the following embodiments or drawings, more or fewer operation steps or module units may be included in the method or device based on routine or without creative labor. In the steps or structures where there is no necessary causal relationship logically, the execution order of these steps or the module structure of the device is not limited to the execution order or module structure described in the embodiments of the present application and shown in the drawings. When the method or module structure described is applied to an actual device or terminal product, it can be connected according to the method or module structure shown in the embodiments or drawings for sequential execution or parallel execution (for example, a parallel processor or a multi-threaded processing environment, or even a distributed processing environment).

[0064] Specifically, Figure 1 As shown, the data processing method of the wind speed power data of the wind farm may include the following steps:

[0065] Step 101: Acquire historical real sample data of wind speed and power of a target wind farm;

[0066] Step 102: preprocessing the historical real sample data of wind speed power to obtain preprocessed real sample data of wind speed power, and randomly generating noise data;

[0067] Among them, the above-mentioned historical real sample data may include: a timestamp field, a wind speed field and a power field; accordingly, preprocessing the historical real sample data of wind speed and power may include: performing default value processing, outlier processing and normalization processing on the wind speed field and the power field; and combining the timestamp field, the processed wind speed field and the power field into preprocessed wind speed power real sample data.

[0068] That is, the above wind speed power real sample data may include: data of timestamp field, wind speed field and power field. Preprocessing of wind speed-power real sample data may include: default value processing of wind speed field and power field, abnormal value processing and normalization processing, obtaining processed wind speed field S, power field P and timestamp field T, and finally obtaining the structure of processed wind speed-power real sample data which can be expressed as Y=(S,P,T).

[0069] Step 103: constructing a generative adversarial network model for capturing the distribution of real sample data of wind speed and power;

[0070] Specifically, a generative adversarial network model containing a generator and a discriminator can be constructed, wherein the generator is used to learn the distribution of historical real sample data of wind speed and power to generate synthetic data close to the real data distribution, and the discriminator is used to learn the difference between the distribution of historical real sample data of wind speed and power and the synthetic data to identify the synthetic data; the generator receives random noise data that satisfies a multi-peak distribution, and outputs synthetic data based on the random noise data; the discriminator receives historical real sample data and synthetic data, and outputs a first data group and a second data group, wherein the first data group includes: a probability for characterizing that the data is historical real sample data and a parameter value of the discriminator, and the second data group includes: a probability for characterizing that the data is synthetic data and a parameter value of the discriminator.

[0071] That is, a multi-peak distribution is first adopted to increase the diversity of the generator input noise z, thereby alleviating the mode collapse phenomenon that may occur during the training process.

[0072] The generator may include: a sequence input module, a sequence parsing module and a sequence generation module, wherein the sequence input module is used to input historical real sample data in the form of a one-dimensional noise vector with a timestamp, and encode and mark the position sequence information of each field in the sequence and encode and mark the position sequence information of each field in the input data, the sequence parsing module is used to perform sequence normalization, sequence subspace information extraction, sequence normalization, and full-connected linear transformation on the encoded and marked data sequence to extract time series features, the sequence generation module is used to input the extracted time series features into the two-dimensional convolutional layer in the sequence generation module and output to obtain synthetic data, wherein the fully connected linear transformation in the sequence parsing module is a two-layer neural network, the two-layer neural network is used to first perform linear transformation, then use the Relu function for nonlinear transformation, and then perform linear transformation Transformation; the above-mentioned discriminator may include: a mixed sequence input module, a mixed sequence parsing module and a binary classification recognition module, wherein the mixed sequence input module is used to receive historical real sample data in the form of a one-dimensional noise vector with a timestamp or the synthetic data output by the generator, and encode and mark the position order information of each field in the sequence and encode and mark the position order information of each field in the input data, the mixed sequence parsing module is used to perform mixed sequence normalization, mixed sequence subspace information extraction, mixed sequence normalization, and full-connected linear transformation on the encoded and marked data sequence to extract time series features, and the binary classification recognition module is used to perform binary classification recognition on the time series features, wherein the binary classification recognition module uses a sigmoid function to map the sequence value to the [0, 1] interval, and divides the output probability value into two categories: real data and synthetic data.

[0073] Specifically, the generator G can receive random noise data z that satisfies the multi-peak distribution B, and output synthetic data G(z,θ), where θ represents all parameters in the generator G. The one-dimensional noise vector with a timestamp is input through the sequence input module and the position order information of each field in the sequence is encoded and marked. Then, the sequence is repeated n times (n>1) in sequence through the sequence normalization, sequence subspace information extraction, sequence normalization, and fully connected linear transformation in the sequence parsing module to extract the timing features. Finally, the extracted timing features are input into the two-dimensional convolutional layer in the sequence generation module and output to obtain the generated data. Among them, the double normalization in the sequence parsing module in the generator G can be the dimensional normalization of each neuron in a single sample, the sequence subspace information extraction object in the sequence parsing module can be a one-dimensional noise vector that has encoded the mark position order information in the sequence input module, and the fully connected linear transformation in the sequence parsing module can be a two-layer neural network. The two-layer neural network can first perform a linear transformation, then use the Relu function for nonlinear transformation, and finally perform a linear transformation again; the two-dimensional convolution layer in the sequence generation module in the generator G can use a two-dimensional convolution calculation with a 1*1 convolution kernel and 1 channel number.

[0074] The above discriminator D can receive real sample data x and synthetic data G(z,θ), and output D(x,γ) or D(G(z,θ),γ), where the values ​​of D(x,γ) and D(G(z,θ),γ) respectively represent the probability that the discriminator D discriminates x and G(z,θ) as real data, and γ represents all parameters in the discriminator D. The generated data or real historical sample data output by the generator G is input through the mixed sequence input module and the position order information of each field in the sequence is encoded and marked. Then, the mixed sequence normalization, mixed sequence subspace information extraction, mixed sequence normalization, and fully connected linear transformation in the mixed sequence parsing module are repeated m times (m>1) to extract the time series features, and finally the time series features are input into the binary classification recognition module for discrimination. The two normalizations in the mixed sequence parsing module in the above discriminator D are used to normalize the dimensions of each neuron in a single sample, and the mixed sequence subspace information extraction object in the mixed sequence parsing module is used for the generated / real sample sequence with the encoded position order information in the mixed sequence input module. The fully connected linear transformation in the hybrid sequence parsing module can be a two-layer neural network, which is used to perform linear transformation first, then use the Relu function for nonlinear transformation, and finally perform linear transformation again. Among them, the binary classification recognition module uses the sigmoid function to map the sequence value to the [0, 1] interval, and divides the output probability value into two categories: real data and synthetic data.

[0075] Step 104: construct a real sample training data set using the preprocessed wind speed power real sample data, and construct a noise sample training data set using the randomly generated noise data;

[0076] For example, we can construct a real sample training dataset X={x n} N , where x n =(S n ,P n ,T n ), n represents the nth row of data (1≤n≤N). Construct a noise sample training data set Z={z m} M , where Z m =(S m ,P m ,T m ), m represents the mth row of data (1≤m≤M).

[0077] Step 105: Inputting the real sample training data set and the noise sample training data set into the generative adversarial network model offline for pre-training until the generative adversarial network model converges, thereby obtaining a pre-trained model of the generative adversarial network;

[0078] Step 106: inputting the wind speed power data generated by the real-time operation of the target wind farm into the pre-trained model of the generative adversarial network in real time, so as to adjust the pre-trained model of the generative adversarial network and obtain a target model of the generative adversarial network;

[0079] Step 107: Generate random noise data that satisfies the same distribution as the noise sample training data set;

[0080] Step 108: Input the random noise data into the target model of the adversarial network to obtain wind speed power data samples that conform to the real data distribution of the target wind farm.

[0081] During implementation, the real sample training data set and the noise sample training data set are input offline into the generative adversarial network model for pre-training until the generative adversarial network model converges to obtain a pre-trained model of the generative adversarial network, which may include:

[0082] S1: Control the generator and discriminator to compete with each other and alternately update the weight parameters of the deep neural network;

[0083] S2: Pre-train the generative adversarial network model according to the following objective function:

[0084]

[0085] Where D represents the discriminator, G represents the generator, z represents noise data, x represents real sample training data, D(x,γ) represents the probability that the discriminator identifies the data in the real sample training data set as real data, D(G(z,θ),γ) represents the probability that the discriminator identifies the synthetic data as real data, θ represents the parameters of the generator, and γ represents the parameters of the discriminator. When the input data is the real sample training data set, the discriminator maximizes log(D(x,γ)) and log(1-D(G(z,θ),γ)), and controls the output of the discriminator D(G(z,θ),γ) to be close to 1. When the input data is synthetic data, the generator minimizes log(1-D(G(z,θ),γ)), and controls the output of the discriminator D(G(z,θ),γ) to be close to 0.

[0086] That is, during the training process, the generator G and the discriminator D can be trained separately in batches. In each batch, the generator G is first fixed and the discriminator D is trained, and then the discriminator D is fixed and the generator G is trained. In order to alleviate the mode collapse phenomenon that may occur during training, a multi-peak distribution can be used to increase the diversity of the generator input noise z, and then an L2 regularization term is added to the deep neural network of the generator and the discriminator to ensure the stability of the training.

[0087] During implementation, the real sample training data set and the noise sample training data set are input offline into the generative adversarial network model for pre-training until the generative adversarial network model converges to obtain a pre-trained model of the generative adversarial network, which may also include:

[0088] S1: Collecting a predetermined number of real sample data from the real sample training data set;

[0089] S2: Collecting a predetermined number of noise sample data from the noise sample training data set;

[0090] S3: Using the stochastic gradient algorithm, the weight parameters of the discriminator are updated by increasing or decreasing the gradient based on a predetermined number of real sample data and a predetermined number of noise sample data:

[0091]

[0092] Wherein, m represents the predetermined number, D represents the discriminator, G represents the generator, z represents the noise data, x represents the real sample training data, D(x,γ) represents the probability that the discriminator distinguishes the data in the real sample training data set as the real data, D(G(z,θ),γ) represents the probability that the discriminator distinguishes the synthetic data as the real data, θ represents the parameters of the generator, and γ represents the parameters of the discriminator.

[0093] The above method is described below in conjunction with a specific embodiment. However, it should be noted that this specific embodiment is only for better illustrating the present application and does not constitute an improper limitation on the present application.

[0094] In this example, a data enhancement method is provided to utilize the existing small sample data of wind speed-power of the newly built wind farm to generate wind speed-power sample data that is more consistent with the spatiotemporal characteristics of the newly built wind farm, so as to effectively expand the training data set. Figure 2 As shown, real data x is obtained, noise z is randomly generated, and the noise z is passed through the generator G to obtain synthetic data G(z). The real data z and the synthetic data G(z) are judged as true / false by the discriminator D. Specifically, a data enhancement method for insufficient historical sample data of wind speed-power of a newly built wind farm is provided, which may include the following steps:

[0095] S1: Obtain the existing wind speed-power real sample data of the target new wind farm and preprocess it;

[0096] The above wind speed-power real sample data may include: data of timestamp field, wind speed field and power field. Preprocessing of wind speed-power real sample data may include: default value processing, abnormal value processing and normalization processing of wind speed field and power field, and obtaining processed wind speed field S, power field P and timestamp field T. The structure of the processed wind speed-power real sample data can be expressed as Y=(S,P,T).

[0097] S2: Construct a generative adversarial network model that can capture the distribution of wind speed-power real sample data. Specifically, it can include the following steps:

[0098] S21: Construct a generative adversarial network model M containing a generator G and a discriminator D, wherein both the generator G and the discriminator D adopt a deep neural network structure, wherein the generator G is used to learn the distribution of real sample data of wind speed-power to generate synthetic data close to the real data distribution, and the discriminator D is used to learn the difference between the distribution of real wind speed-power data and synthetic data to identify synthetic data.

[0099] S22: Generator G receives random noise data z that satisfies multimodal distribution B, and outputs synthetic data G(z,θ), where θ represents all parameters in generator G, such as Figure 3As shown, the network structure of the generator G can be composed of a sequence input module, a sequence parsing module and a sequence generation module. Specifically, a one-dimensional noise vector with a timestamp is input through the sequence input module and the position order information of each field in the sequence is encoded and marked. Then, the sequence normalization, sequence subspace information extraction, sequence normalization, and fully connected linear transformation in the sequence parsing module are repeated n times (n>1) to extract the time series features. Finally, the extracted time series features are input into the two-dimensional convolutional layer in the sequence generation module and output to obtain the generated data.

[0100] The double normalization in the sequence parsing module in the above generator G can be the dimensional normalization of each neuron in a single sample, the sequence subspace information extraction object in the sequence parsing module can be a one-dimensional noise vector that has encoded the mark position order information in the sequence input module, and the fully connected linear transformation in the sequence parsing module can be a two-layer neural network. The two-layer neural network can first perform a linear transformation, then use the Relu function for nonlinear transformation, and finally perform a linear transformation again; the two-dimensional convolution layer in the sequence generation module in the generator G can use a two-dimensional convolution calculation with a 1*1 convolution kernel and 1 channel number.

[0101] S23: The discriminator D receives the real sample data x and the synthetic data G(z,θ), and outputs D(x,γ) and D(G(z,θ),γ), where the values ​​of D(x,γ) and D(G(z,θ),γ) represent the probabilities that the discriminator D discriminates x and G(z,θ) as real data, respectively.

[0102] γ represents all parameters in the discriminator D. Figure 4 As shown, the network structure of the discriminator D can be composed of a mixed sequence input module, a mixed sequence parsing module and a binary classification recognition module, wherein the generated data output by the generator G or the real historical sample data is input through the mixed sequence input module and the position order information of each field in the sequence is encoded and marked, and then, the mixed sequence normalization, mixed sequence subspace information extraction, mixed sequence normalization, and fully connected linear transformation in the mixed sequence parsing module are repeated m times (m>1) to extract the time series features, and finally the time series features are input into the binary classification recognition module for discrimination.

[0103] The double normalization in the mixed sequence parsing module in the above discriminator D is used to normalize the dimensions of each neuron in a single sample, and the mixed sequence subspace information extraction object in the mixed sequence parsing module is used to generate / real sample sequences that have encoded the mark position order information in the mixed sequence input module. The fully connected linear transformation in the mixed sequence parsing module can be a two-layer neural network, which is used to perform linear transformation first, then use the Relu function for nonlinear transformation, and finally perform linear transformation again. Among them, the binary classification recognition module uses the sigmoid function to map the sequence value to the [0, 1] interval, and divides the output probability value into two categories: real data and synthetic data.

[0104] S3: Use the preprocessed wind speed-power real sample data and randomly generated noise data to construct a real sample training data set and a noise sample training data set respectively;

[0105] Specifically, the following steps may be included:

[0106] S31: Construct a real sample training dataset X={x n} N , where x n =(S n ,P n ,T n ), n represents the nth row of data (1≤n≤N).

[0107] S32: Construct a noise sample training data set Z={z m} M , where Z m =(S m ,P m ,T m ), m represents the mth row of data (1≤m≤M).

[0108] S4: Input the training dataset offline into the generative adversarial network model for pre-training until the model converges;

[0109] Specifically, the generator G and the discriminator D can be made to compete with each other and update the weight parameters of the deep neural network alternately. D should maximize log(D(x,γ)) and log(1-D(G(z,θ),γ)), that is, when the input data is real data x, the output D(x,γ) of the discriminator is close to 1, and when the input data is synthetic data G(z,θ), the output D(G(z,θ),γ) of the discriminator is as close to 0 as possible. Preferably, G should minimize log(1-D(G(z,θ),γ)), that is, the synthetic data G(z,θ) generated by G can deceive D as much as possible, so that the output D(G(z,θ),γ) of the discriminator D is close to 1.

[0110] The objective function of the above training process can be expressed as:

[0111]

[0112] During the training process, the generator G and the discriminator D can be trained separately in batches. In each batch, the generator G is first fixed and the discriminator D is trained, and then the discriminator D is fixed and the generator G is trained.

[0113] In order to alleviate the mode collapse phenomenon that may occur during training, a multi-peak distribution can be used to increase the diversity of the generator input noise z. Then, an L2 regularization term can be added to the deep neural networks of the generator and the discriminator to ensure the stability of the training.

[0114] S5: Input the wind speed-power data generated by the operation of the target new wind farm into the pre-trained model in real time for continuous fine-tuning;

[0115] Specifically, the wind speed and power flow data generated by the newly built wind farm can be collected in real time, and the data can be processed through the above-mentioned steps S1 and S31, and the processed data can be input into the generator G of the generative adversarial network model M in real time as real sample training data to continuously fine-tune the model.

[0116] S6: Generate random noise data that has the same distribution as the noise sample training data set and input it into the generative adversarial network model to obtain generated data samples that conform to the distribution of real data. q} Q , where R q =(S q ,P q ,T q ), q represents the qth row of data (1≤q≤Q).

[0117] In implementation, the generator G and the discriminator D may be constructed as follows, for example, including:

[0118] Input: learning rate η, batch size batch, number of discriminator iterations k for each generator iteration; Output: parameters θ and γ of the generator network and discriminator network after convergence.

[0119] Specifically, m noise sample data are collected from the noise sample training data set, and m real sample data are collected from the real sample training data set. Then,

[0120] Use the stochastic gradient algorithm and update the discriminator weight parameters by increasing the gradient:

[0121]

[0122] Collect m noise sample data from the noise sample training data set, use the stochastic gradient algorithm and update the discriminator weight parameters by reducing the gradient:

[0123]

[0124] In the above example, a generator and discriminator based on a deep neural network are constructed, and a parallel training strategy of existing offline and incremental real-time is adopted to improve the accuracy and generalization ability of the generated data. In view of the existing data fusion method using different wind farms, the spatial characteristics vary greatly and the time series characteristics of wind speed-power data are ignored, which cannot reflect the actual operation status of the newly built wind farms. In this example, the spatiotemporal characteristics are mined based on the small sample historical actual operation data of the newly built wind farm, so that the generated data is more in line with the distribution law of the actual operation data. Furthermore, the existing method is to unify the data format by using a multi-source heterogeneous data integration method, rather than mining and analyzing the semantic features of the data. In this example, a deep neural network is used to automatically extract spatiotemporal semantic features. The generated data is not a simple integration in format, but a data coupling based on the understanding of the contextual semantics of the actual operation data.

[0125] The method embodiments provided in the above embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on an electronic device as an example, Figure 5 This is a hardware structure block diagram of an electronic device for a data processing method of wind speed power data of a wind farm provided by the present application. Figure 5 As shown, the electronic device 10 may include one or more (only one is shown in the figure) processors 02 (the processor 02 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 04 for storing data, and a transmission module 06 for communication functions. It can be understood by those skilled in the art that Figure 5 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 5 More or fewer components as shown, or with Figure 5 Different configurations are shown.

[0126] The memory 04 can be used to store software programs and modules of application software, such as program instructions / modules corresponding to the data processing method of wind speed power data of the wind farm in the embodiment of the present application. The processor 02 executes various functional applications and data processing by running the software programs and modules stored in the memory 04, that is, the data processing method of wind speed power data of the wind farm of the above-mentioned application is realized. The memory 04 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 04 may further include a memory remotely arranged relative to the processor 02, and these remote memories may be connected to the electronic device 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0127] The transmission module 06 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the electronic device 10. In one example, the transmission module 06 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission module 06 can be a radio frequency (Radio Frequency, RF) module, which is used to communicate with the Internet wirelessly.

[0128] At the software level, the data processing device for the wind speed power data of the wind farm can be as follows: Figure 6 As shown, including:

[0129] An acquisition module 601 is used to acquire historical real sample data of wind speed and power of a target wind farm;

[0130] A preprocessing module 602 is used to preprocess the historical real sample data of wind speed power to obtain preprocessed real sample data of wind speed power and randomly generate noise data;

[0131] A first construction module 603 is used to construct a generative adversarial network model for capturing the distribution of real sample data of wind speed and power;

[0132] The second construction module 604 is used to construct a real sample training data set using the preprocessed wind speed power real sample data, and to construct a noise sample training data set using the randomly generated noise data;

[0133] An input module 605 is used to input the real sample training data set and the noise sample training data set offline into a generative adversarial network model for pre-training until the generative adversarial network model converges, thereby obtaining a pre-trained model of the generative adversarial network;

[0134] An adjustment module 606 is used to input the wind speed power data generated by the real-time operation of the target wind farm into the pre-trained model of the generative adversarial network in real time, so as to adjust the pre-trained model of the generative adversarial network and obtain a target model of the generative adversarial network;

[0135] A first generating module 607, configured to generate random noise data having the same distribution as the noise sample training data set;

[0136] The second generating module 608 is used to input the random noise data into the target model of the adversarial network to obtain wind speed power data samples that conform to the real data distribution of the target wind farm.

[0137] In one embodiment, the historical real sample data may include: a timestamp field, a wind speed field and a power field; accordingly, preprocessing the historical real sample data of wind speed and power includes: performing default value processing, outlier processing and normalization processing on the wind speed field and the power field; and combining the timestamp field, the processed wind speed field and the power field into preprocessed wind speed power real sample data.

[0138] In one embodiment, the first construction module 603 may include: a construction unit, used to construct a generative adversarial network model containing a generator and a discriminator, wherein the generator is used to learn the distribution of historical real sample data of wind speed and power to generate synthetic data close to the real data distribution, and the discriminator is used to learn the difference between the distribution of historical real sample data of wind speed and power and the synthetic data to identify the synthetic data; a first output unit, used to receive random noise data that satisfies a multi-peak distribution through the generator, and output synthetic data based on the random noise data; a second output unit, used to receive historical real sample data and synthetic data through the discriminator, and output a first data group and a second data group, wherein the first data group includes: a probability for characterizing that the data is historical real sample data and a parameter value of the discriminator, and the second data group includes: a probability for characterizing that the data is synthetic data and a parameter value of the discriminator.

[0139] In one embodiment, the generator may include: a sequence input module, a sequence parsing module and a sequence generation module, wherein the sequence input module is used to input historical real sample data in the form of a one-dimensional noise vector with a timestamp, and encode and mark the position sequence information of each field in the sequence and encode and mark the position sequence information of each field in the input data, the sequence parsing module is used to perform sequence normalization, sequence subspace information extraction, sequence normalization, and full-connected linear transformation on the encoded and marked data sequence to extract time series features, the sequence generation module is used to input the extracted time series features into the two-dimensional convolutional layer in the sequence generation module and output to obtain synthetic data, wherein the fully connected linear transformation in the sequence parsing module is a two-layer neural network, and the two-layer neural network is used to first perform linear transformation, then use the Relu function for nonlinear transformation, and then perform linear transformation;

[0140] The above-mentioned discriminator may include: a mixed sequence input module, a mixed sequence parsing module and a binary classification recognition module, wherein the mixed sequence input module is used to receive historical real sample data in the form of a one-dimensional noise vector with a timestamp or synthetic data output by the generator, and encode and mark the position order information of each field in the sequence and encode and mark the position order information of each field in the input data, the mixed sequence parsing module is used to perform mixed sequence normalization, mixed sequence subspace information extraction, mixed sequence normalization, and fully connected linear transformation on the encoded and marked data sequence to extract time series features, and the binary classification recognition module is used to perform binary classification recognition on the time series features, wherein the binary classification recognition module uses a sigmoid function to map the sequence value to the [0, 1] interval, and divides the output probability value into two categories: real data and synthetic data.

[0141] In one embodiment, the input module 605 can control the generator and the discriminator to confront each other and alternately update the weight parameters of the deep neural network; the generative adversarial network model is pre-trained according to the following objective function:

[0142]

[0143] Where D represents the discriminator, G represents the generator, z represents noise data, x represents real sample training data, D(x,γ) represents the probability that the discriminator distinguishes the data in the real sample training data set as real data, D(G(z,θ),γ) represents the probability that the discriminator distinguishes the synthetic data as real data, θ represents the parameter of the generator, and γ represents the parameter of the discriminator; when the input data is the real sample training data set, the discriminator maximizes log(D(x,γ)) and log(1-D(G(z,θ),γ)), and controls the output of the discriminator D(G(z,θ),γ) to be close to 1; when the input data is synthetic data, the generator minimizes log(1-D(G(z,θ),γ)), and controls the output of the discriminator D(G(z,θ),γ) to be close to 0.

[0144] In one embodiment, the input module 605 may collect a predetermined number of real sample data from the real sample training data set; collect a predetermined number of noise sample data from the noise sample training data set;

[0145] The stochastic gradient algorithm is used to update the weight parameters of the discriminator by increasing or decreasing the gradient based on a predetermined number of real sample data and a predetermined number of noise sample data:

[0146]

[0147] Wherein, m represents the predetermined number, D represents the discriminator, G represents the generator, z represents the noise data, x represents the real sample training data, D(x,γ) represents the probability that the discriminator distinguishes the data in the real sample training data set as the real data, D(G(z,θ),γ) represents the probability that the discriminator distinguishes the synthetic data as the real data, θ represents the parameters of the generator, and γ represents the parameters of the discriminator.

[0148] The embodiment of the present application also provides a specific implementation of an electronic device capable of implementing all the steps in the data processing method for wind speed and power data of a wind farm in the above embodiment. The electronic device specifically includes the following contents: a processor, a memory, a communication interface and a bus; wherein the processor, the memory and the communication interface communicate with each other through the bus; the processor is used to call a computer program in the memory, and when the processor executes the computer program, all the steps in the data processing method for wind speed and power data of a wind farm in the above embodiment are implemented. For example, when the processor executes the computer program, the following steps are implemented:

[0149] Step 1: Obtain historical real sample data of wind speed and power of the target wind farm;

[0150] Step 2: preprocessing the historical real sample data of the wind speed power to obtain the preprocessed real sample data of the wind speed power, and randomly generating noise data;

[0151] Step 3: Build a generative adversarial network model to capture the distribution of real sample data of wind speed and power;

[0152] Step 4: Use the preprocessed wind speed and power real sample data to construct a real sample training data set, and use the randomly generated noise data to construct a noise sample training data set;

[0153] Step 5: Input the real sample training data set and the noise sample training data set into the generative adversarial network model offline for pre-training until the generative adversarial network model converges, thereby obtaining a pre-trained model of the generative adversarial network;

[0154] Step 6: Inputting the wind speed power data generated by the real-time operation of the target wind farm into the pre-trained model of the generative adversarial network in real time to adjust the pre-trained model of the generative adversarial network to obtain a target model of the generative adversarial network;

[0155] Step 7: Generate random noise data that satisfies the same distribution as the noise sample training data set;

[0156] Step 8: Input the random noise data into the target model of the adversarial network to obtain wind speed power data samples that conform to the real data distribution of the target wind farm.

[0157] The embodiments of the present application also provide a computer-readable storage medium capable of implementing all the steps in the data processing method for wind speed and power data of a wind farm in the above embodiments. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, all the steps in the data processing method for wind speed and power data of a wind farm in the above embodiments are implemented. For example, when the processor executes the computer program, the following steps are implemented:

[0158] Step 1: Obtain historical real sample data of wind speed and power of the target wind farm;

[0159] Step 2: preprocessing the historical real sample data of the wind speed power to obtain the preprocessed real sample data of the wind speed power, and randomly generating noise data;

[0160] Step 3: Build a generative adversarial network model to capture the distribution of real sample data of wind speed and power;

[0161] Step 4: Use the preprocessed wind speed and power real sample data to construct a real sample training data set, and use the randomly generated noise data to construct a noise sample training data set;

[0162] Step 5: Input the real sample training data set and the noise sample training data set into the generative adversarial network model offline for pre-training until the generative adversarial network model converges, thereby obtaining a pre-trained model of the generative adversarial network;

[0163] Step 6: Inputting the wind speed power data generated by the real-time operation of the target wind farm into the pre-trained model of the generative adversarial network in real time to adjust the pre-trained model of the generative adversarial network to obtain a target model of the generative adversarial network;

[0164] Step 7: Generate random noise data that satisfies the same distribution as the noise sample training data set;

[0165] Step 8: Input the random noise data into the target model of the adversarial network to obtain wind speed power data samples that conform to the real data distribution of the target wind farm.

[0166] From the above description, it can be seen that the embodiment of the present application obtains the historical real sample data of the wind speed and power of the target wind farm; then, the historical real sample data of the wind speed and power is preprocessed to obtain the preprocessed real sample data of the wind speed and power, and randomly generates noise data; then a generative adversarial network model for capturing the distribution of the real sample data of the wind speed and power is constructed; the real sample training data set is constructed using the preprocessed real sample data of the wind speed and power, and the noise sample training data set is constructed using the randomly generated noise data; the real sample training data set and the noise sample training data set are input offline into the generative adversarial network model for pre-training until the generative adversarial network model converges to obtain a pre-trained model of the generative adversarial network; then the wind speed and power data generated by the real-time operation of the target wind farm is input in real time into the pre-trained model of the generative adversarial network to adjust the pre-trained model of the generative adversarial network to obtain a target model of the generative adversarial network; random noise data that satisfies the same distribution as the noise sample training data set is generated; the random noise data is input into the target model of the adversarial network to obtain a wind speed and power data sample that conforms to the real data distribution of the target wind farm. The above method can efficiently expand the wind speed and power data used for training, thereby improving the prediction accuracy of the wind power and power generation model of the target wind farm.

[0167] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the hardware + program embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0168] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0169] Although the present application provides method operation steps as described in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-creative labor. The order of steps listed in the embodiments is only one way of executing the order of many steps and does not represent the only execution order. When the actual device or client product is executed, it can be executed in the order of the method shown in the embodiments or the drawings or in parallel (for example, in a parallel processor or multi-threaded processing environment).

[0170] Although the embodiments of this specification provide method operation steps as described in the embodiments or flow charts, more or less operation steps may be included based on conventional or non-creative means. The order of steps listed in the embodiments is only one way of executing the order of many steps and does not represent the only execution order. When the device or terminal product in practice is executed, it can be executed in sequence or in parallel according to the method shown in the embodiments or the drawings (for example, a parallel processor or a multi-threaded processing environment, or even a distributed data processing environment). The term "include", "include" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, product or equipment including a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such process, method, product or equipment. In the absence of more restrictions, it is not excluded that there are other identical or equivalent elements in the process, method, product or equipment including the elements.

[0171] For the convenience of description, the above devices are described in various modules according to their functions. Of course, when implementing the embodiments of this specification, the functions of each module can be implemented in the same or more software and / or hardware, or the module implementing the same function can be implemented by a combination of multiple sub-modules or sub-units. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0172] Those skilled in the art also know that, in addition to implementing the controller in a purely computer-readable program code, the controller can be made to implement the same function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered as a hardware component, and the devices for implementing various functions included therein can also be considered as structures within the hardware component. Or even, the devices for implementing various functions can be considered as both software modules for implementing the method and structures within the hardware component.

[0173] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0174] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, the embodiments of this specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the embodiments of this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0175] The present specification embodiments may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present specification embodiments may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0176] Each embodiment in this specification is described in a progressive manner, and the same and similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. In the description of this specification, the description of the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the embodiment of this specification. In this specification, the schematic representation of the above terms does not necessarily target the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, in the absence of contradiction, a person skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0177] The above is only an example of the embodiment of the present specification and is not intended to limit the embodiment of the present specification. For those skilled in the art, the embodiment of the present specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the embodiment of the present specification shall be included in the scope of the claims of the embodiment of the present specification.

Claims

1. A method for processing wind speed power data of a wind farm, characterized in that: The method comprises: Obtain historical real sample data of wind speed and power of the target wind farm; Preprocessing the historical real sample data of the wind speed power to obtain preprocessed real sample data of the wind speed power, and randomly generating noise data; Construct a generative adversarial network model to capture the distribution of real sample data of wind speed and power; The real sample data after preprocessing is used to construct a real sample training data set, and the noise sample training data set is constructed using the randomly generated noise data; Inputting the real sample training data set and the noise sample training data set into a generative adversarial network model offline for pre-training until the generative adversarial network model converges, thereby obtaining a pre-trained model of the generative adversarial network; Inputting the wind speed power data generated by the real-time operation of the target wind farm into the pre-trained model of the generative adversarial network in real time to adjust the pre-trained model of the generative adversarial network to obtain a target model of the generative adversarial network; Generate random noise data that satisfies the same distribution as the noise sample training data set; Inputting the random noise data into the target model of the adversarial network to obtain wind speed power data samples that conform to the real data distribution of the target wind farm; Among them, a generative adversarial network model is constructed to capture the distribution of real sample data of wind speed and power, including: A generative adversarial network model containing a generator and a discriminator is constructed, wherein the generator is used to learn the distribution of historical real sample data of wind speed and power to generate synthetic data close to the distribution of real data, and the discriminator is used to learn the difference between the distribution of historical real sample data of wind speed and power and the distribution of synthetic data to identify synthetic data; The generator receives random noise data satisfying multimodal distribution and outputs synthetic data according to the random noise data; Receiving historical real sample data and synthetic data through a discriminator, and outputting a first data group and a second data group, wherein the first data group includes: a probability for characterizing that the data is historical real sample data and a parameter value of the discriminator, and the second data group includes: a probability for characterizing that the data is synthetic data and a parameter value of the discriminator; The generator includes: a sequence input module, a sequence parsing module and a sequence generation module, wherein the sequence input module is used to input historical real sample data in the form of a one-dimensional noise vector with a timestamp, and encode and mark the position sequence information of each field in the sequence and encode and mark the position sequence information of each field in the input data, the sequence parsing module is used to perform sequence normalization, sequence subspace information extraction, sequence normalization, and full-connection linear transformation on the encoded and marked data sequence to extract time series features, and the sequence generation module is used to input the extracted time series features into the two-dimensional convolutional layer in the sequence generation module and output to obtain synthetic data, wherein the fully connected linear transformation in the sequence parsing module is a two-layer neural network, and the two-layer neural network is used to first perform linear transformation, then use the Relu function for nonlinear transformation, and then perform linear transformation; The discriminator includes: a mixed sequence input module, a mixed sequence parsing module and a binary classification recognition module, wherein the mixed sequence input module is used to receive historical real sample data in the form of a one-dimensional noise vector with a timestamp or synthetic data output by a generator, and encode and mark the position order information of each field in the sequence and encode and mark the position order information of each field in the input data; the mixed sequence parsing module is used to perform mixed sequence normalization, mixed sequence subspace information extraction, mixed sequence normalization, and full-connected linear transformation on the encoded and marked data sequence to extract time series features; the binary classification recognition module is used to perform binary classification recognition on the time series features, wherein the binary classification recognition module uses a sigmoid function to map the sequence value to the [0, 1] interval, and divides the output probability value into two categories: real data and synthetic data.

2. The method according to claim 1, characterized in that The historical real sample data includes: a timestamp field, a wind speed field and a power field; Accordingly, preprocessing the historical real sample data of wind speed power includes: Perform default value processing, outlier processing and normalization processing on wind speed field and power field; The timestamp field, the processed wind speed field and the power field are combined into pre-processed wind speed and power real sample data.

3. The method according to claim 1, characterized in that Inputting the real sample training data set and the noise sample training data set into the generative adversarial network model offline for pre-training until the generative adversarial network model converges, thereby obtaining a pre-trained model of the generative adversarial network, including: Control the generator and discriminator to compete with each other and alternately update the weight parameters of the deep neural network; The generative adversarial network model is pre-trained according to the following objective function: ； Where D represents the discriminator, G represents the generator, z represents noise data, x represents real sample training data, D(x,γ) represents the probability that the discriminator distinguishes the data in the real sample training data set as real data, D(G(z,θ),γ) represents the probability that the discriminator distinguishes the synthetic data as real data, θ represents the parameter of the generator, and γ represents the parameter of the discriminator; when the input data is the real sample training data set, the discriminator maximizes log(D(x,γ)) and log(1-D(G(z,θ),γ)), and controls the output of the discriminator D(G(z,θ),γ) to be close to 1; when the input data is synthetic data, the generator minimizes log(1-D(G(z,θ),γ)), and controls the output of the discriminator D(G(z,θ),γ) to be close to 0.

4. The method according to claim 1, characterized in that: Inputting the real sample training data set and the noise sample training data set into the generative adversarial network model offline for pre-training until the generative adversarial network model converges, thereby obtaining a pre-trained model of the generative adversarial network, including: Collecting a predetermined number of real sample data from the real sample training data set; Collecting a predetermined number of noise sample data from the noise sample training data set; The stochastic gradient algorithm is used to update the weight parameters of the discriminator by increasing or decreasing the gradient based on a predetermined number of real sample data and a predetermined number of noise sample data: ； Wherein, m represents the predetermined number, D represents the discriminator, G represents the generator, z represents the noise data, x represents the real sample training data, D(x,γ) represents the probability that the discriminator distinguishes the data in the real sample training data set as the real data, D(G(z,θ),γ) represents the probability that the discriminator distinguishes the synthetic data as the real data, θ represents the parameters of the generator, and γ represents the parameters of the discriminator.

5. A data processing device for wind speed power data of a wind farm, characterized in that: include: An acquisition module is used to acquire historical real sample data of wind speed and power of a target wind farm; A preprocessing module is used to preprocess the historical real sample data of the wind speed power to obtain the preprocessed real sample data of the wind speed power and randomly generate noise data; The first building module is used to build a generative adversarial network model for capturing the distribution of real sample data of wind speed and power; The second construction module is used to construct a real sample training data set using the preprocessed wind speed power real sample data, and to construct a noise sample training data set using the randomly generated noise data; An input module, used to input the real sample training data set and the noise sample training data set offline into the generative adversarial network model for pre-training until the generative adversarial network model converges, thereby obtaining a pre-trained model of the generative adversarial network; An adjustment module, used for inputting the wind speed power data generated by the real-time operation of the target wind farm into the pre-trained model of the generative adversarial network in real time, so as to adjust the pre-trained model of the generative adversarial network and obtain a target model of the generative adversarial network; A first generating module, used to generate random noise data that satisfies the same distribution as the noise sample training data set; A second generating module is used to input the random noise data into the target model of the adversarial network to obtain a wind speed power data sample that conforms to the real data distribution of the target wind farm; Among them, a generative adversarial network model is constructed to capture the distribution of real sample data of wind speed and power, including: A generative adversarial network model containing a generator and a discriminator is constructed, wherein the generator is used to learn the distribution of historical real sample data of wind speed and power to generate synthetic data close to the distribution of real data, and the discriminator is used to learn the difference between the distribution of historical real sample data of wind speed and power and the synthetic data to identify the synthetic data; The generator receives random noise data satisfying multimodal distribution and outputs synthetic data according to the random noise data; Receiving historical real sample data and synthetic data through a discriminator, and outputting a first data group and a second data group, wherein the first data group includes: a probability for characterizing that the data is historical real sample data and a parameter value of the discriminator, and the second data group includes: a probability for characterizing that the data is synthetic data and a parameter value of the discriminator; The generator includes: a sequence input module, a sequence parsing module and a sequence generation module, wherein the sequence input module is used to input historical real sample data in the form of a one-dimensional noise vector with a timestamp, and encode and mark the position sequence information of each field in the sequence and encode and mark the position sequence information of each field in the input data, the sequence parsing module is used to perform sequence normalization, sequence subspace information extraction, sequence normalization, and full-connection linear transformation on the encoded and marked data sequence to extract time series features, and the sequence generation module is used to input the extracted time series features into the two-dimensional convolutional layer in the sequence generation module and output to obtain synthetic data, wherein the fully connected linear transformation in the sequence parsing module is a two-layer neural network, and the two-layer neural network is used to first perform linear transformation, then use the Relu function for nonlinear transformation, and then perform linear transformation; The discriminator includes: a mixed sequence input module, a mixed sequence parsing module and a binary classification recognition module, wherein the mixed sequence input module is used to receive historical real sample data in the form of a one-dimensional noise vector with a timestamp or synthetic data output by a generator, and encode and mark the position order information of each field in the sequence and encode and mark the position order information of each field in the input data; the mixed sequence parsing module is used to perform mixed sequence normalization, mixed sequence subspace information extraction, mixed sequence normalization, and full-connected linear transformation on the encoded and marked data sequence to extract time series features; the binary classification recognition module is used to perform binary classification recognition on the time series features, wherein the binary classification recognition module uses a sigmoid function to map the sequence value to the [0, 1] interval, and divides the output probability value into two categories: real data and synthetic data.

6. The device according to claim 5, characterized in that The first building block comprises: A construction unit is used to construct a generative adversarial network model including a generator and a discriminator, wherein the generator is used to learn the distribution of historical real sample data of wind speed and power to generate synthetic data close to the distribution of real data, and the discriminator is used to learn the difference between the distribution of historical real sample data of wind speed and power and the distribution of synthetic data to identify the synthetic data; A first output unit, configured to receive random noise data satisfying a multi-modal distribution through a generator, and output synthetic data according to the random noise data; The second output unit is used to receive historical real sample data and synthetic data through the discriminator, and output a first data group and a second data group, wherein the first data group includes: a probability for characterizing that the data is historical real sample data and a parameter value of the discriminator, and the second data group includes: a probability for characterizing that the data is synthetic data and a parameter value of the discriminator.

7. An electronic device comprising a processor and a memory for storing instructions executable by the processor, characterized in that: When the processor executes the instructions, the steps of the method according to any one of claims 1 to 4 are implemented.

8. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images