A method, device and medium for predicting wind power output under extreme weather scenarios

Abstract

The present invention relates to a method, device and medium for predicting wind power output in extreme weather scenarios, wherein the method comprises the following steps: obtaining a historical wind power output data set, wherein the wind power output data set comprises historical wind power generation and wind speed under different weather conditions; discretizing the wind speed and weather type labels of the wind power output data set to construct a first data set; using a data enhancement module constructed based on a conditional generative adversarial network (CGAN) to perform data enhancement on scarce samples in extreme weather scenarios in the first data set to obtain a second data set; obtaining real-time meteorological data, and using a prediction module constructed based on a regression generative adversarial network (RGAN) to predict wind power output online, wherein the prediction module uses wind speed and weather type data as input, outputs wind power output prediction results, and uses the second data set for training. Compared with the prior art, the present invention has the advantages of being able to cope with the high uncertainty and variability of wind power output under extreme weather conditions, and having high prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power output prediction, and in particular to a method, device and medium for predicting wind power output in extreme weather scenarios based on data enhancement. Background Art

[0002] As the most common green and renewable energy, wind power has been widely used and developed in countries around the world. However, wind power is also facing prominent problems while developing rapidly. Unlike traditional thermal power generation, wind power and photovoltaic power generation, as the main forms of renewable energy generation, are easily affected by external meteorological factors, and the power generation capacity is intermittent and volatile. Under normal meteorological conditions, the output of renewable energy can be predicted more accurately based on numerical weather forecasts, and the impact of weather fluctuations on the operation of the power grid is relatively small. However, in recent years, with the intensification of global warming, extreme weather has become more frequent, which has had a serious impact on the safe operation of renewable energy and even the power system. Therefore, there is a need to quantitatively analyze the operating data of wind farms under extreme weather conditions and evaluate their impact.

[0003] At present, the research on wind power prediction at home and abroad is becoming more and more extensive and in-depth. Among various prediction methods, the simplest and most direct one is the continuous method that takes the most recent wind speed or power observation value as the next prediction value. This method has a simple model, but its accuracy decreases rapidly as the time scale increases. Other commonly used advanced methods include statistical and intelligent methods such as Kalman filtering, time series, artificial neural networks, fuzzy logic, and support vector machines. These methods have their own applicability and limitations, but they have one common feature: they only need to predict a single wind speed or wind power time series, and establish a prediction model by statistically processing the time series data, and then predict the future wind speed or wind power. In the prediction process, only historical data and real-time data of the wind farm are required, which is easy to implement. However, these methods ignore the influence of physical factors such as geographical terrain data, wind direction, roughness, temperature, air pressure, and spatial position relationship between each wind turbine or wind farm in the wind farm area. Therefore, the prediction method system and model prediction accuracy need to be further improved. Wind power output is almost entirely determined by natural conditions and is difficult to control artificially. Although the large-scale power fluctuations of tens of millions of kilowatts of wind power bases have seasonal variations and are characterized by low power during the day and high power at night, and the uncertainty of wind power output is somewhat alleviated with the increase in cluster size, its volatility, intermittency and randomness are still much more prominent than the power load.

[0004] Due to the rarity of extreme weather, data on wind power output is extremely scarce, and the collection process faces many challenges. This lack of data has largely hindered in-depth research on wind power performance under extreme climate conditions. At the same time, the vast majority of current domestic and foreign literature focuses on analyzing the operating performance of wind farms under conventional meteorological conditions, and based on this, studies the operation and maintenance data of wind farms under conventional weather conditions. Therefore, there is currently a lack of a reliable method for analyzing the operating conditions of wind farms under extreme climate conditions. Summary of the invention

[0005] The purpose of the present invention is to provide a method, device and medium for predicting wind power output in extreme weather scenarios based on data enhancement, so as to achieve accurate wind power output prediction when extreme weather sample data is scarce.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] A method for predicting wind power output in extreme weather scenarios based on data enhancement includes the following steps:

[0008] Acquire a historical wind power output data set, wherein the wind power output data set includes historical wind power generation and wind speed under different weather conditions;

[0009] Discretize the wind speed and weather type labels of the wind power output data set to construct the first data set;

[0010] The scarce samples in the extreme weather scenarios in the first dataset are enhanced using a data enhancement module based on a conditional generation adversarial network (CGAN) to obtain a second dataset.

[0011] Real-time meteorological data is obtained, and a prediction module based on a regression generation adversarial network (RGAN) is used to predict wind power output online. The prediction module takes wind speed and weather type data as input, outputs wind power output prediction results, and is trained using a second data set.

[0012] The label discretization process is specifically as follows:

[0013] The wind speed data in the wind power output data set is divided into multiple intervals according to the intensity, and marked with different conditional values ​​to form the historical wind speed sequence V t ;

[0014] According to the four weather types of normal weather, hurricane, frost and strong wind and wave, the weather is marked as different condition values ​​to form the historical weather type sequence W t;

[0015] The historical wind speed sequence V t and the historical weather type sequence W t Merge it into the wind power output data set to obtain the first data set.

[0016] The input data of the data enhancement module are random noise and required condition values, where the condition values ​​are wind speed sequence and weather type sequence. The data enhancement module generates and outputs wind power generation under different extreme weather conditions according to the given condition values ​​to form an enhanced wind power output data set, which is merged with the first data set to obtain a second data set.

[0017] The data enhancement module consists of a generator G c and the discriminator D C It consists of two parts, among which the generator G c The goal is to input the wind speed sequence V L and the weather type sequence W L and random noise z to generate enhanced data G with the same time series distribution as the real wind power output data x c (z), discriminator D C The goal is to distinguish the augmented data generated by the generator from the real data, and the training of the data augmentation module is achieved by maximizing the loss function of the generator and minimizing the loss function of the discriminator.

[0018] The input data of the prediction module are random noise and required condition values, wherein the condition values ​​are historical wind speed sequences and historical weather type sequences. The prediction module performs continuous numerical prediction according to the given condition values ​​and outputs wind power output prediction values.

[0019] The prediction model includes a generator G R and the discriminator D R , where the generator G R The goal is to input the historical wind speed sequence V t and the historical weather type sequence W t Generate a predicted value G that is as close to the actual wind power output as possible R (x), the input label is passed through G R Converted into predicted wind power output Right now:

[0020]

[0021] Where z is random noise, θ g represents the parameters of the generator network;

[0022] Discriminator D R The goal is to evaluate the predicted values ​​produced by the generator and the actual wind power output value yt The error between R Accepts a pair of inputs or (V t ,W t ;y t ), and outputs a scalar value or (V t ,W t ,y t θ d ), which represents the probability that the predicted value matches the true value:

[0023]

[0024] Among them, θ d represents the parameters of the generator network;

[0025] The prediction module is trained by maximizing the loss function of the generator and minimizing the loss function of the discriminator.

[0026] The generator G R Based on the LSTM network, the entire LSTM network predicts and outputs the future wind power output sequence by learning the input historical wind speed sequence and historical weather type sequence;

[0027] The temporal sequence of the input data is retained, and the linear transformation is performed through the fully connected layer and gradually input into the LSTM network. The input of each time step is recursively processed through the update rule of the LSTM network, and the prediction output of the corresponding time step is generated. The output of the generator is converted through one or more fully connected layers to adapt to specific prediction tasks, as shown below:

[0028]

[0029] Among them, W y and b y are the weights and biases of the output layer, f is the linear transformation function of the fully connected layer, is the activation function;

[0030] Generator G R Generate a series of predicted values It is used to simulate the time series of real wind power output data.

[0031] The data enhancement module uses the autocorrelation function and FID to evaluate the training results of the network, and the prediction module uses the root mean square error to evaluate the training results of the network.

[0032] An electronic device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the method described above when executing the program.

[0033] A storage medium stores a program, which implements the method described above when executed.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] The deep learning-based extreme weather wind power output prediction method provided by the present invention combines the long-term dependency capture capability of LSTM and the generation capability of GAN to maintain a good match with the actual data in the nonlinear characteristics of wind power data, and is specially designed to cope with the high uncertainty and variability of wind power output under extreme weather conditions. The present invention simulates extreme weather data through a data enhancement module, solves the problem of scarcity of wind power data under extreme weather, and can continuously improve the prediction capability through iterative evolution, effectively improve wind power management and utilization efficiency, and make accurate and real-time predictions of wind power output under extreme weather. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a flow chart of the method of the present invention;

[0037] Figure 2 A diagram illustrating the process of label discretization processing according to an embodiment of the present invention;

[0038] Figure 3 A specific structural diagram of a CGAN generator network according to an embodiment of the present invention;

[0039] Figure 4 A specific structural diagram of a CGAN discriminator network according to an embodiment of the present invention;

[0040] Figure 5 A specific structural diagram of an RGAN generator network according to an embodiment of the present invention;

[0041] Figure 6 This is a specific structural diagram of the RGAN discriminator network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0043] Example 1

[0044] This embodiment provides a method for predicting wind power output in extreme weather scenarios based on data enhancement. Figure 1 As shown, the following steps are included:

[0045] S101, obtain historical wind power output data set Ω Untreated,The wind power output data set at least includes the historical wind power generation and wind speed data under different weather conditions.

[0046] S102, discretizing the wind speed and weather type labels of the wind power output data set to construct a first data set Ω Labeled .

[0047] like Figure 2 As shown in the figure, the label discretization process is as follows:

[0048] According to the standards of the World Meteorological Organization, the wind speed data in the wind power output dataset is divided into four intervals according to the intensity, and marked as conditional values ​​1 / 2 / 3 / 4 respectively, that is, wind speed 0-10 m / s is marked as 1, wind speed 10-15 m / s is marked as 2, wind speed 15-20 m / s is marked as 3, and wind speed above 20 m / s is marked as 4, thus forming the historical wind speed sequence V t ;

[0049] According to the four weather types of normal weather, hurricane, frost and strong wind and wave, the weather is marked as conditional values ​​of 1 / 2 / 3 / 4, that is, normal weather is marked as 1, hurricane weather is marked as 2, frost weather is marked as 3, and strong wind and wave weather is marked as 4, thus forming the historical weather type sequence W t ;

[0050] The historical wind speed sequence V t and the historical weather type sequence W t Merged into the wind power output data set, and constructed together with the historical wind power output sequence as the first data set Ω Labeled .

[0051] S103, using a data enhancement module constructed based on a conditional generative adversarial network (CGAN) to perform data enhancement on scarce samples in extreme weather scenarios in the first data set to obtain a second data set.

[0052] Generative adversarial networks (GANs) are an emerging branch of data-driven methods in recent years and have been widely used in the field of power systems. Through the adversarial training process between the generator and the discriminator, GANs have a strong ability to learn the inherent distribution of data sets. CGAN is a variant of GAN. The introduction of conditional quantities enables the GAN model to generate data under specified labels.

[0053] The input data of the data enhancement module is the random noise z and the required condition value, and the condition value is the wind speed sequence V L and the weather type sequence W L The data enhancement module generates and outputs wind power generation under different extreme weather conditions according to the given condition values, forming an enhanced wind power output data set Ω generated, and the first data set Ω Labeled After merging, we get the second data set Ω RGAN .

[0054] The data augmentation module consists of the generator G c and the discriminator D C It consists of two parts, and its structures are as follows Figure 3 and Figure 4 As shown, the generator G c The goal is to input the wind speed sequence V L and the weather type sequence W L and random noise z to generate enhanced data G with the same time series distribution as the real wind power output data x c (z), discriminator D C The goal is to distinguish between the augmented data generated by the generator and the real data.

[0055] The training of the CGAN model is a substitution and adversarial process: G c Trying to generate something that can deceive D C Sample G c (z), D C Learn to distinguish G c (z) and x by assigning a larger probability to x and a smaller probability to G c (z). This process is described as a minimax game.

[0056]

[0057] Where P r and P d is the probability distribution of training data and generated data, E is the expectation operator,

[0058] The training of the data augmentation module is achieved by maximizing the loss function of the generator and minimizing the loss function of the discriminator.

[0059] S104, obtaining real-time meteorological dataΩ Real-time , using a prediction module built based on a regression generative adversarial network RGAN to predict wind power output online, the prediction module takes wind speed and weather type data as input, outputs wind power output prediction results, and uses the second data set for training.

[0060] RGAN is also a variant of GAN. By combining the GAN framework with the goal of regression analysis, it can be used for continuous numerical prediction tasks.

[0061] The input data of the prediction module are random noise z and the required condition value, where the condition value is the historical wind speed sequence V t and the historical weather type sequence Wt , the prediction module performs continuous numerical prediction according to the given condition value and outputs the wind power output prediction value. The prediction process of the prediction module is expressed by the formula:

[0062]

[0063] Among them, z is random noise, and f(·) represents the mapping relationship of RGAN.

[0064] The prediction model consists of a generator G R and the discriminator D R , whose structures are as follows Figure 5 and Figure 6 As shown, the generator G R The goal is to input the historical wind speed sequence V t and the historical weather type sequence W t Generate a predicted value G that is as close to the actual wind power output as possible R (x), the input label is passed through G R Converted into predicted wind power output Right now:

[0065]

[0066] Where z is random noise, θ g Represents the parameters of the generator network.

[0067] Discriminator D R The goal is to evaluate the predicted values ​​produced by the generator and the actual wind power output value y t The error between R Accepts a pair of inputs or (V t ,W t ;y t ), and outputs a scalar value or (V t ,W t ,y t θ d ), which represents the probability that the predicted value matches the true value:

[0068]

[0069] Among them, θ d represents the parameters of the generator network;

[0070] Similar to CGAN, RGAN training also involves minimizing the generator loss function and maximizing the discriminator loss function, but the formula is slightly different: G R Trying to generate something that can deceive D R The predicted value of D R Learn to distinguish and t , by assigning a larger probability to y t , assigning a smaller probability to It can be expressed as the following min-max problem:

[0071]

[0072] Among them, p data (V t ,W t ;y t ) is the data distribution of real wind power output and its related labels, It's G R Based on the input (z, V t ,W t ) Generate predicted values The conditional distribution of .

[0073] In this way, G R Learning produces increasingly accurate wind power output forecasts, while D R The learning method is to better distinguish the difference between the generated predictions and the real data, and the ultimate goal is to make the generated predictions unable to be detected by D R To distinguish and achieve prediction accuracy.

[0074] The operation of wind power at a specific time point will be affected by the conditions in the past period of time, such as wind speed, temperature, weather conditions, etc. At the same time, wind power output shows daily and seasonal periodicity, which reflects the periodic changes in wind speed and weather conditions in nature. Therefore, wind power data has a strong time series.

[0075] In order to capture the long-term dependency between historical wind power output and historical wind speed and historical weather type, this embodiment uses LSTM with full connection layer, activation function layer, etc. to construct the generator network G of RGAN R , to effectively capture the historical wind power output sequence Y t With the historical wind speed series V t and the historical weather type sequence W t long-term dependencies in .

[0076] An LSTM unit consists of four main parts: forget gate t 、Input gate i t , output gate t , and the cell state C tThese parts together determine how the LSTM unit updates and transfers information. For each time step t, given the input (z, V t ,W t ), the input is merged into a vector X through a fully connected layer linear transformation t , as the input of the LSTM layer, that is:

[0077] X t =f(z,V t ,W t )

[0078] Input X to LSTM t and the hidden state h at the previous time step t-1 and cell status C t-1 , the update of LSTM unit can be expressed by the following formula:

[0079] ①The forget gate determines which information will be discarded:

[0080] f t =σ(W f ·[h t-1 ,X t ]+b f )

[0081] ②The input gate determines what new information will be stored in the cell state:

[0082] i t =σ(W i ·[h t-1 ,X t ]+b i )

[0083]

[0084] ③Unit status update: old status C t-1 Update to new state C t , which reflects the information forgotten by the forget gate and added by the input gate:

[0085]

[0086] ④Output gate and hidden state update: Finally, determine the output value o t , and update the hidden state h t :

[0087] o t =σ(W o ·[h t-1 ,X t ]+b o )

[0088] ht =o t *tanh(C t )

[0089] The entire LSTM network predicts the output sequence (future wind power output) by learning the input sequence (such as historical wind speed and weather conditions). The temporal nature of the input data is retained and linearly transformed through the fully connected layer and gradually input into the LSTM network. The input of each time step is recursively processed through the update rule of the LSTM network, and the prediction output of the corresponding time step is generated. The output of the generator can be converted through one or more fully connected layers to adapt to specific prediction tasks, as shown below:

[0090]

[0091] Among them, W y and b y are the weights and biases of the output layer, f is the linear transformation function of the fully connected layer, is the activation function.

[0092] Generator G R Generate a series of predicted values This is used to simulate the time series of real wind power output data. In this way, the generator G built based on LSTM R It can effectively capture the time series characteristics in wind power data and generate accurate future wind power output forecasts.

[0093] When making online predictions, real-time data is continuously collected from multiple data sources (such as weather stations, sensor networks, etc.). Real-time , such as wind speed V and weather type W. The collected real-time data is processed using the label discretization preprocessing method described above, and the pre-trained RGAN model is used to perform instant prediction. The model generates prediction values ​​based on the real-time input data, which represent estimates of future wind power output. The execution of online prediction relies on efficient data processing and fast model inference capabilities to ensure the accuracy and timeliness of the prediction.

[0094] Example 2

[0095] This embodiment provides the following specific implementation process based on the method described in Embodiment 1:

[0096] S1. Get data

[0097] The data set Ω used in this example UntreatedProvided by an Irish energy group, it contains wind power generation data in Ireland and Northern Ireland for about six and a half years from 2014 to 2020. The data set covers key indicators such as wind power generation and wind speed, where the wind power generation data comes from SCADA system readings that have not been quality controlled. The load data in the data set is in megawatts (MW), and the data is updated every 15 minutes. In order to construct the training data set, this embodiment specifically screens data under extreme weather conditions and appropriately integrates data under normal weather conditions to ensure the diversity and representativeness of the training set.

[0098] S2. Preprocessing Data

[0099] According to the label discretization processing method mentioned in Example 1 of the present invention, the data set Ω Untreated Preprocessing is performed to divide the wind speed into four intervals according to the intensity, marked as conditional values ​​1 / 2 / 3 / 4, and the weather is marked as conditional values ​​1 / 2 / 3 / 4 according to the type, forming the processed data set Ω Labeled , to suit the data format of machine learning models.

[0100] S3. Training CGAN

[0101] Using a dataset with discrete numerical labels Ω Labeled The training of the generative data augmentation model CGAN was completed, and the autocorrelation function (ACF) and Fréchet Inception Distance (FID) were used to measure the performance of the CGAN model in generating wind power data under extreme weather conditions.

[0102] The present invention uses autocorrelation function (ACF) and Fréchet Inception Distance (FID) as validity criteria for the generative model CGAN.

[0103] ACF reflects the degree of correlation between the values ​​at any two moments in the wind power output data sequence. It is a quantitative representation of the internal dependency of the data sequence. By comparing the ACF of the generated data and the real data, it is possible to evaluate whether the model can capture the inherent laws of the time series. If the ACFs of the two are similar, it means that CGAN performs better in time dependency. FID measures the distribution difference between the generated output data and the real output data in the feature space. A lower FID value indicates that the generated data is closer to the real data in statistical characteristics, and therefore the model is considered to have better performance.

[0104] S4. Enhanced Dataset

[0105] The trained and verified CGAN is used to generate wind power output data Ω under extreme weather conditions with wind intensity and weather type as conditional parameters. Generated , to solve the problem of insufficient sample size due to the rarity of extreme weather events, and to convert Ω Generated With Ω Labeled Merge into enhanced dataset Ω RGAN .

[0106] S5. Training RGAN

[0107] Use the enhanced training set Ω RGAN The training of the prediction model RGAN is completed, and the root mean square error (RMSE) is used as an evaluation indicator to measure the performance of the RGAN model in the task of wind power output prediction under extreme weather conditions.

[0108] The present invention uses the root mean square error (RMSE) as the validity criterion of the prediction model RGAN, which is achieved by calculating the square root of the mean of the square of the difference between the predicted value and the true value under the same label. The formula is as follows:

[0109]

[0110] Where N is the total number of samples, y i is the actual wind power output value of the i-th sample, is the corresponding generator based on the input (V t ,W t )Predicted wind power output value.

[0111] Use the trained RGAN generator G R Each sample in the test set is predicted to obtain the predicted value, and for each sample in the test set, the mean square error between the predicted value and the actual value is calculated. A lower RMSE value indicates that the model has a higher prediction accuracy and a smaller error. Therefore, one of the goals of optimizing the model is to minimize the RMSE value. Using RMSE as an evaluation indicator can effectively measure the performance of the model in the task of predicting wind power output under extreme weather conditions.

[0112] S6. Data Prediction

[0113] After completing the validation of the model, RGAN completed the training. At this point, RGAN can obtain very accurate wind power output data prediction values, achieving the expected purpose.

[0114] S7. Online wind power forecasting

[0115] Get real-time meteorological data as the conditional value, use the trained RGAN model, and use the formula Predict real-time wind power output data.

[0116] Example 3

[0117] This embodiment provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the program, the method described in the above embodiment 1 or 2 is implemented.

[0118] The electronic device of the present invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The CPU, ROM and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0119] Multiple components in the device are connected to the I / O interface, including: input units, such as keyboards, mice, etc.; output units, such as various types of displays, speakers, etc.; storage units, such as disks, optical disks, etc.; and communication units, such as network cards, modems, wireless communication transceivers, etc. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunication networks.

[0120] The processing unit performs the various methods and processes described above, such as methods S101 to S104. For example, in some embodiments, methods S101 to S104 may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via a ROM and / or a communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more steps of methods S101 to S104 described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute methods S101 to S104 in any other appropriate manner (e.g., by means of firmware).

[0121] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0122] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0123] Example 4

[0124] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method described in the above embodiment 1 or 2 is implemented.

[0125] In the context of the present invention, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0126] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning, or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for predicting wind power output in extreme weather scenarios based on data enhancement, characterized in that: The following steps are involved: Acquire a historical wind power output data set, wherein the wind power output data set includes historical wind power generation and wind speed under different weather conditions; Discretize the wind speed and weather type labels of the wind power output data set to construct the first data set; The data enhancement module based on the conditional generative adversarial network (CGAN) is used to enhance the scarce samples in the extreme weather scenarios in the first data set to obtain the second data set. Real-time meteorological data is obtained as a conditional value, and a prediction module based on a regression generative adversarial network (RGAN) is used to predict wind power output online. The prediction module takes wind speed and weather type data as input, outputs wind power output prediction results, and is trained using a second data set; The label discretization process is specifically as follows: The wind speed data in the wind power output data set is divided into multiple intervals according to the intensity, and marked with different conditional values ​​to form the historical wind speed sequence V t ; According to the four weather types of normal weather, hurricane, frost and strong wind and wave, the weather is marked as different condition values ​​to form the historical weather type sequence W t ; The historical wind speed sequence V t and the historical weather type sequence W t Merge into the wind power output data set to obtain a first data set; The input data of the prediction module are random noise and required condition values, wherein the condition values ​​are historical wind speed sequences and historical weather type sequences. The prediction module performs continuous numerical prediction based on the given condition values ​​and outputs wind power output prediction values. The prediction module includes a generator G R and the discriminator D R , where the generator G R The goal is to input the historical wind speed sequence V t and the historical weather type sequence W t Generate a predicted value G that is as close to the actual wind power output as possible R (x), the input label is passed through G R Converted into predicted wind power output Right now: Where z is random noise, θ g represents the parameters of the generator network; Discriminator D R The goal is to evaluate the predicted values ​​produced by the generator and the actual wind power output value y t The error between R Accepts a pair of inputs or (V t ,W t ;y t ), and outputs a scalar value or (V t ,W t ,y t θ d ), which represents the probability that the predicted value matches the true value: Among them, θ d represents the parameters of the generator network; The prediction module is trained by maximizing the loss function of the generator and minimizing the loss function of the discriminator.

2. The method for predicting wind power output in extreme weather scenarios based on data enhancement according to claim 1 is characterized in that: The input data of the data enhancement module are random noise and required condition values, where the condition values ​​are wind speed sequence and weather type sequence. The data enhancement module generates and outputs wind power generation under different extreme weather conditions according to the given condition values ​​to form an enhanced wind power output data set, which is merged with the first data set to obtain a second data set.

3. The method for predicting wind power output in extreme weather scenarios based on data enhancement according to claim 2 is characterized in that: The data enhancement module consists of a generator G c and the discriminator D C It consists of two parts, among which the generator G c The goal is to input the wind speed sequence V L and the weather type sequence W L and random noise z to generate enhanced data G with the same time series distribution as the real wind power output data x c (z), discriminator D C The goal is to distinguish the augmented data generated by the generator from the real data, and the training of the data augmentation module is achieved by maximizing the loss function of the generator and minimizing the loss function of the discriminator.

4. The method for predicting wind power output in extreme weather scenarios based on data enhancement according to claim 1 is characterized in that: The generator G R Based on the LSTM network, the entire LSTM network predicts and outputs the future wind power output sequence by learning the input historical wind speed sequence and historical weather type sequence; The temporal sequence of the input data is retained, and the linear transformation is performed through the fully connected layer and gradually input into the LSTM network. The input of each time step is recursively processed through the update rule of the LSTM network, and the prediction output of the corresponding time step is generated. The output of the generator is converted through one or more fully connected layers to adapt to specific prediction tasks, as shown below: Among them, W y and b y are the weights and biases of the output layer, f is the linear transformation function of the fully connected layer, is the activation function; Generator G R Generate a series of predicted values It is used to simulate the time series of real wind power output data.

5. The method for predicting wind power output in extreme weather scenarios based on data enhancement according to claim 1 is characterized in that: The data enhancement module uses the autocorrelation function and FID to evaluate the training results of the network, and the prediction module uses the root mean square error to evaluate the training results of the network.

6. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the program, the method according to any one of claims 1 to 5 is implemented.

7. A storage medium having a program stored thereon, characterized in that: When the program is executed, the method according to any one of claims 1 to 5 is implemented.

Images