Wind turbine power generation prediction model training method, device and equipment

Abstract

The present disclosure proposes a method, device and equipment for training a wind turbine generator power generation prediction model. The method includes: obtaining multiple training data, wherein each training data includes a first matrix composed of multiple unit operation data at multiple time points every day, a second matrix composed of multiple wind tower data, and a third matrix composed of multiple weather forecast data; the first matrix, the second matrix and the third matrix of each training data are respectively used as feature maps and fused to obtain training feature maps; the training feature maps are respectively input into an initial network model to output multiple predicted powers at multiple time points; a preset loss function is used to calculate a target loss value based on the actual power and the predicted power at the multiple time points; and the initial network model is optimized with reference to the target loss value until convergence to obtain a power generation prediction model, so that the power generation prediction model can be used to accurately predict the power generation of the wind turbine.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to a method, device, and equipment for training a wind turbine generator system power generation prediction model. Background Art

[0002] Renewable energy generation technologies include wind power, solar power, ocean power, geothermal power, and biomass power. These technologies all utilize equipment to convert renewable energy into electricity, thereby generating electricity. With the continuous advancement of technology and reduction in costs, renewable energy generation has become an increasingly important component of global energy supply. As the proportion of renewable energy generation continues to increase, ensuring the stable operation of the power grid requires higher accuracy in power forecasting. Summary of the Invention

[0003] The present disclosure proposes a method, device and equipment for training a wind turbine generator power generation prediction model, aiming to solve one of the technical problems in the related art at least to a certain extent.

[0004] An embodiment of the first aspect of the present disclosure proposes a method for training a wind turbine power generation prediction model, comprising: obtaining multiple training data, wherein each training data includes a first matrix composed of multiple unit operation data at multiple time points every day, a second matrix composed of multiple wind tower data, and a third matrix composed of multiple weather forecast data; using the first matrix, the second matrix, and the third matrix of each training data as feature maps and fusing them to obtain training feature maps; and inputting the training feature maps into an initial network model to output multiple predicted powers at multiple time points; using a preset loss function to calculate a target loss value based on the actual power and predicted power at multiple time points, and optimizing the initial network model with reference to the target loss value until convergence to obtain a power generation prediction model.

[0005] A second aspect embodiment of the present disclosure proposes a method for predicting the generated power of a wind turbine, the method comprising: obtaining data to be predicted, wherein the data to be predicted includes a first matrix composed of multiple unit operation data at multiple time points on a target date, a second matrix composed of multiple wind tower data, and a third matrix composed of multiple weather forecast data; using the first matrix, the second matrix, and the third matrix as feature maps and performing fusion processing to obtain a prediction feature map; and inputting the prediction feature map into a generated power prediction model trained by the training method of the first aspect embodiment described above to output the generated power at multiple time points on the target date.

[0006] An embodiment of the third aspect of the present disclosure proposes a wind turbine power generation prediction model training device, including: a first acquisition module, used to acquire multiple training data, wherein each training data includes a first matrix composed of multiple unit operation data at multiple time points every day, a second matrix composed of multiple wind tower data, and a third matrix composed of multiple weather forecast data; a first fusion module, used to use the first matrix, the second matrix and the third matrix of each training data as feature maps and perform fusion processing to obtain a training feature map; and an input module, used to input the training feature maps into the initial network model respectively to output multiple predicted powers at multiple time points; a training module, used to calculate a target loss value based on the actual power and predicted power at multiple time points using a preset loss function, and optimize the initial network model with reference to the target loss value until convergence to obtain a power generation prediction model.

[0007] An embodiment of the fourth aspect of the present disclosure proposes a wind turbine power generation prediction device, including: a second acquisition module, used to obtain data to be predicted, wherein the data to be predicted includes a first matrix composed of multiple unit operation data at multiple time points on a target date, a second matrix composed of multiple wind tower data, and a third matrix composed of multiple meteorological forecast data; a second fusion module, used to use the first matrix, the second matrix, and the third matrix as feature maps and perform fusion processing to obtain a prediction feature map; and a prediction module, used to input the prediction feature map into a power generation prediction model trained by the above-mentioned training device to output the power generation at multiple time points on the target date.

[0008] The fifth aspect embodiment of the present disclosure proposes an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the wind turbine power generation prediction model training method of the embodiment of the present disclosure, or execute the wind turbine power generation prediction method.

[0009] The sixth aspect embodiment of the present disclosure proposes a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the wind turbine power generation prediction model training method disclosed in the embodiment of the present disclosure, or to execute the wind turbine power generation prediction method.

[0010] In this embodiment, a plurality of training data are obtained, wherein each training data includes a first matrix composed of a plurality of unit operation data at a plurality of time points every day, a second matrix composed of a plurality of wind tower data, and a third matrix composed of a plurality of weather forecast data, and the first matrix, the second matrix and the third matrix of each training data are respectively used as feature maps and fused to obtain training feature maps, and the training feature maps are respectively input into the initial network model to output a plurality of predicted powers at a plurality of time points, and a preset loss function is used to calculate a target loss value based on the actual power and the predicted power at a plurality of time points, and the initial network model is optimized with reference to the target loss value until convergence to obtain a power generation prediction model, and the power generation prediction model can be trained by integrating a plurality of data, so that the power generation prediction model can be used to accurately predict the power generation of the wind turbine.

[0011] Additional aspects and advantages of the present disclosure will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The above and / or additional aspects and advantages of the present disclosure will become apparent and readily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0013] Figure 1 This is a flow chart of a method for training a wind turbine generator power generation prediction model according to an embodiment of the present disclosure;

[0014] Figure 2 This is a schematic diagram of the overall process of wind turbine power generation prediction model training and prediction according to an embodiment of the present disclosure;

[0015] Figure 3 is a schematic diagram of a matrix dimension mapping process provided according to an embodiment of the present disclosure;

[0016] Figure 4 is a schematic diagram of a feature map fusion process provided according to an embodiment of the present disclosure;

[0017] Figure 5 is a flow chart of a method for training a wind turbine generator power generation prediction model according to another embodiment of the present disclosure;

[0018] Figure 6 is a schematic diagram of a wind turbine generator power generation prediction model training device provided according to another embodiment of the present disclosure;

[0019] Figure 7 is a schematic diagram of a wind turbine generator system power generation prediction device according to another embodiment of the present disclosure;

[0020] Figure 8A block diagram of an exemplary electronic device suitable for implementing embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0021] In order to more clearly understand the above-mentioned objectives, features and advantages of the present disclosure, the scheme of the present disclosure will be further described below. It should be noted that the embodiments of the present disclosure and the features therein can be combined with each other in the absence of conflict.

[0022] In the following description, many specific details are set forth to facilitate a full understanding of the present disclosure, but the present disclosure may also be implemented in other ways different from those described herein; it is obvious that the embodiments in the specification are only part of the embodiments of the present disclosure, rather than all of the embodiments.

[0023] It should be noted that the executor of the wind turbine power generation prediction model training method of this embodiment can be a wind turbine power generation prediction model training device, which can be implemented by software and / or hardware. The device can be configured in an electronic device, and the electronic device can include but is not limited to a terminal, a server, etc.

[0024] Figure 1 FIG. 1 is a flow chart of a method for training a wind turbine generator power generation prediction model according to an embodiment of the present disclosure, such as Figure 1 As shown, the method includes:

[0025] S101: Acquire multiple pieces of training data.

[0026] Specifically, Figure 2 FIG. 1 is a schematic diagram of the overall process of training and predicting the power generation prediction model of a wind turbine according to an embodiment of the present disclosure. Figure 2 As shown, the process includes data acquisition, data preprocessing, model building, model training and prediction stages in sequence. In the data acquisition stage, this embodiment first obtains multiple training data.

[0027] In the embodiment of the present disclosure, each piece of training data is the historical unit operation data of the wind turbine unit at multiple time points every day, wind tower data, numerical weather prediction data (NWP), and any other possible data, without limitation.

[0028] The unit operation data at each time point includes multiple (N) types, such as wind speed, wind direction, power, speed, pitch angle, ambient temperature, ambient humidity and other dimensions, which are expressed as follows:

[0029]

[0030] The wind tower data also includes multiple types (M types), such as wind speed, wind direction, temperature, humidity, etc. at different heights, which can be expressed as:

[0031]

[0032] Weather forecast data also includes multiple types (K types), such as wind speed and direction data at different heights (10 meters, 30 meters, 50 meters, 70 meters, 90 meters, etc.), ambient temperature, ambient humidity, atmospheric pressure data, etc., which can be expressed as:

[0033]

[0034] The multiple time points (t) represent multiple moments in a day. For example, in this embodiment, a sliding time window is used to determine a set of unit operating data, wind tower data, and weather forecast data every 15 minutes. Therefore, each day can be divided into 96 time points, i.e., t = 96. Furthermore, in this embodiment, each time point in each piece of training data has a corresponding generated power, which is referred to as the actual power.

[0035] In this embodiment, the various unit operation data at t time points every day can be constructed into a t*N dimensional matrix, which is called the first matrix; similarly, in this embodiment, the various wind tower data at t time points can be constructed into a t*M dimensional matrix, which is called the second matrix; similarly, the various weather forecast data at t time points can be constructed into a t*K dimensional matrix, which is called the third matrix. That is to say, each training data in this embodiment includes the first matrix, the second matrix and the third matrix.

[0036] In practical applications, this embodiment can preprocess the acquired training data so that the training data can be used for model training.

[0037] In some embodiments, there may be missing data sources in the various unit operation data, wind tower data, and weather forecast data at multiple time points every day (for example, the wind speed data in the unit operation data at a certain time point is missing). In this case, this embodiment can fill in the missing values ​​of the data source (wind speed), wherein this embodiment uses, for example, the average value of the data source in each day for filling.

[0038] In other embodiments, each data source has a corresponding value range. In this embodiment, training data that is not within the value range can be eliminated. For example, the normal range of the unit speed variable is [0,1850], and training data that exceeds the normal range can be eliminated.

[0039] In some other embodiments, normalization processing may be performed on different data sources in the first matrix, the second matrix, and the third matrix. The calculation formula is as follows:

[0040]

[0041] in, represents the normalized data, x represents the original data in the matrix, mean(x) represents the mean of the data source, and std(x) represents the variance of the data source.

[0042] S102: The first matrix, the second matrix, and the third matrix of each piece of training data are respectively used as feature maps and fused to obtain a training feature map.

[0043] Furthermore, in the model building process, this embodiment can use the first matrix, the second matrix and the third matrix of each training data as feature maps and perform feature fusion processing. The feature map obtained after fusion is called a training feature map.

[0044] In some embodiments, since the dimensions of the first matrix, the second matrix, and the third matrix may be different, it is necessary to regularize the dimensions of different data sources in order to better perform feature fusion. Specifically, the model building process includes a self-attention (SA) module, a CNN feature fusion module, and a state prediction module. The SA module of this embodiment can first map the first matrix, the second matrix, and the third matrix to matrices of the same dimension.

[0045] For example, Figure 3 is a schematic diagram of a matrix dimension mapping process according to an embodiment of the present disclosure, such as Figure 3 As shown, this embodiment may adopt a self-attention mechanism, for example, to map the t*N-dimensional first matrix, the t*M-dimensional second matrix, and the t*K-dimensional third matrix into matrices of the same dimension (for example, t*D dimension). The calculation formula is as follows:

[0046]

[0047] Among them, Q, K, and V are matrices obtained by linearly transforming the input x. The linear transformation process is expressed as: Q = Wx, where W is a trainable weight matrix.

[0048] Furthermore, the matrices with the same dimensions after SA module mapping are used as feature maps of different channels, and multiple (ie, 3) feature maps are fused to obtain training feature maps. Specifically, Figure 4 is a schematic diagram of the feature map fusion process provided according to an embodiment of the present disclosure, such as Figure 4As shown, the CNN feature fusion module extracts features from the three feature maps using three learnable 1×1 convolution kernels, resulting in feature map A (corresponding to the first matrix), feature map B (corresponding to the second matrix), and feature map C (corresponding to the third matrix). Feature maps A and B are then matrix-multiplied and normalized using a softmax operation to produce the attention matrix. The attention matrix is ​​matrix-multiplied with feature map C, and then subjected to a learnable 1×1 convolution kernel for feature extraction. The final result is feature map D, which serves as the training feature map.

[0049] S103: Input the training feature maps into the initial network model respectively to output multiple predicted powers at multiple time points.

[0050] Specifically, the state prediction module of this embodiment pre-sets an initial network model, which includes, for example, gated recurrent units (Gated Recurrent Units) and deep neural networks (DNNs) connected in sequence. A GRU network is used to learn historical correlation information at different time steps, providing more feature information for predicting future states. The results of the GRU network are directly used as input to the DNN network. The number of output nodes of the DNN network in this embodiment is 96, that is, it can achieve power prediction every 15 minutes for the next day.

[0051] In this embodiment, during the training process, the training feature map can be input into the initial network model, and the initial network model outputs the predicted power of multiple time points (96).

[0052] S104: Calculate a target loss value based on the actual power and predicted power at multiple time points using a preset loss function, and optimize the initial network model with reference to the target loss value until convergence, to obtain a power generation prediction model.

[0053] Furthermore, the model training module can use a preset loss function to calculate a target loss value based on the actual power and predicted power at multiple time points, and optimize the parameters in the initial network model (for example, the dimension of the GRU network output space, learning rate, hyperparameters, etc.) with reference to the target loss value. Optimization methods such as random search, grid search, and Bayesian optimization can be used. Here, Bayesian optimization is used until the model converges to obtain a power generation prediction model. The loss function can be any possible loss function, without limitation.

[0054] In some embodiments, as the proportion of renewable energy generation continues to increase, in order to ensure the stable operation of the power grid, higher requirements are placed on the accuracy of power forecasting. As power forecasting technology continues to develop, the assessment of power forecast accuracy on the grid side is also evolving. Taking into account the changes in grid supply and demand in different periods, three periods of grid operation are defined:

[0055] (1) The new energy vehicle launch period is from 10:00 to 16:00;

[0056] (2) Peak electricity consumption hours are 6:00 to 9:00 and 17:00 to 22:00;

[0057] (3) Other time periods are 0:00 to 6:00, 9:00 to 10:00, 16:00 to 17:00, and 22:00 to 24:00;

[0058] There are also adjustments to the assessment strategies for different time periods. The assessment scores for different time periods are as follows:

[0059]

[0060] As can be seen, during periods of high renewable energy generation, if the predicted power is low and below the threshold, the assessment will be doubled; during peak renewable energy consumption periods, if the predicted power is high and above the threshold, the assessment will be doubled. Therefore, to further improve the accuracy of wind power forecasts and reduce the economic impact of assessments, it is necessary to optimize the loss function during model training. In view of this, this embodiment designs a piecewise loss function during model training, taking into account the power generation characteristics of different time periods. Bayesian optimization is used to optimize the key hyperparameters of the entire network. After obtaining the optimal hyperparameters, the model is trained using the full data set.

[0061] Specifically, this embodiment divides multiple time points in each training data into a new energy peak period (10:00 to 16:00), a peak electricity consumption period (6:00 to 9:00, 17:00 to 22:00), and other periods (0:00 to 6:00, 9:00 to 10:00, 16:00 to 17:00, 22:00 to 24:00), where the number of time points in the new energy peak period can be represented by N, the number of time points in the peak electricity consumption period can be represented by M, and the number of time points in other periods can be represented by L, and N+M+L=96.

[0062] Furthermore, the loss value is calculated based on the actual power and predicted power at N time points during the period of peak new energy generation in the training data. This loss value is called the first loss value, which can be represented by Loss1; similarly, the loss value is calculated based on the actual power and predicted power at M time points during the peak electricity consumption period. This loss value is called the second loss value, which can be represented by Loss2; similarly, the loss value is calculated based on the actual power and predicted power at L time points in other time periods. This loss value is called the third loss value, which can be represented by Loss3.

[0063] Furthermore, the first loss value Loss1, the second loss value Loss2, and the third loss value Loss3 are accumulated to obtain the total loss, that is, the target loss value.

[0064] In some embodiments, for periods of high renewable energy production, it is desirable to have the predicted value as high as possible above the actual value. To this end, this embodiment may set a hyperparameter interval for these periods, which can be represented as α∈[0.55,0.9]. In calculating the first loss value, this embodiment may determine a parameter from the hyperparameter interval [0.55,0.9] corresponding to the period of high renewable energy production. This parameter, referred to as the first hyperparameter α, is used to calculate the first loss value. Furthermore, based on the actual power and predicted power at N time points within the period of high renewable energy production and the first hyperparameter α, the first loss value Loss1 is calculated.

[0065] For example, the first loss value Loss1 can be calculated according to the following loss function:

[0066]

[0067] Among them, Loss1 represents the first loss value, N represents the number of time points in the new energy generation period, and yi represents the actual power at time point i in the new energy generation period. represents the prediction power at time point i, α represents the first hyperparameter, and α∈[0.55,0.9]. Greater than or equal to y i Time point utilization Calculate the loss; and Less than y i Time point utilization Calculate the losses.

[0068] In other embodiments, for peak periods of renewable energy consumption, it is desirable that the predicted value be slightly lower than the actual value. In view of this, this embodiment can set a hyperparameter interval for peak periods, which can be expressed as β∈[0.3,0.7]. In the process of calculating the second loss value, this embodiment can determine a parameter from the hyperparameter interval [0.3,0.7] corresponding to the peak period, which is referred to as the second hyperparameter β. Furthermore, the first loss value Loss2 is calculated based on the actual power and predicted power at multiple time points during the peak period L and the second hyperparameter β.

[0069] The second loss value can be calculated, for example, according to the following loss function:

[0070]

[0071] Among them, Loss2 represents the second loss value, L represents the number of time points during the peak power consumption period, and y i represents the actual power at time i during the peak electricity period, represents the prediction power at time point i, β represents the second hyperparameter, and β∈[0.3,0.7]. Greater than or equal to y i Time point utilization Calculate the loss; and Less than y i Time point utilization Calculate the losses.

[0072] The third loss value Loss3 in other periods is calculated as follows:

[0073]

[0074] Where M represents the number of time points in other stages.

[0075] The overall loss function is set as follows:

[0076]

[0077] For the model training and hyperparameter part of this embodiment, optimization methods such as random search, grid search, and Bayesian optimization can be used. Here, the Bayesian optimization method is used for long-term parameter optimization. The hyperparameters that need to be optimized include: the dimension after dimension regularization in the SA module, the dimension of the GRU network output space in the state prediction module, α and β in the function shown, and the learning rate. After parameter optimization, the optimal parameters are obtained, and the optimal model is retrained using the optimal parameters and the full amount of data.

[0078] In this embodiment, a plurality of training data are obtained, wherein each training data includes a first matrix composed of a plurality of unit operation data at a plurality of time points every day, a second matrix composed of a plurality of wind tower data, and a third matrix composed of a plurality of weather forecast data, and the first matrix, the second matrix and the third matrix of each training data are respectively used as feature maps and fused to obtain a training feature map, and the training feature maps are respectively input into the initial network model to output a plurality of predicted powers at a plurality of time points, and a preset loss function is used to calculate a target loss value based on the actual power and the predicted power at a plurality of time points, and the initial network model is optimized with reference to the target loss value until convergence to obtain a power generation prediction model, and the power generation prediction model can be trained by integrating a plurality of data, thereby improving the accuracy of the power generation prediction model.

[0079] Figure 5 FIG. 1 is a flow chart of a method for predicting wind turbine power generation according to another embodiment of the present disclosure, such as Figure 5 As shown, the method includes:

[0080] S501: Obtain data to be predicted.

[0081] Among them, the data to be predicted includes a first matrix composed of multiple unit operation data at multiple time points (96) on the target date, a second matrix composed of multiple wind tower data, and a third matrix composed of multiple weather forecast data. The specific contents of the matrices can be found in the above embodiment and will not be repeated here.

[0082] S502: The first matrix, the second matrix, and the third matrix are used as feature maps and fused to obtain a predicted feature map.

[0083] In this embodiment, the same fusion method as in the training phase can be used to take the first matrix, the second matrix, and the third matrix as feature maps and perform fusion processing to obtain a predicted feature map, which will not be described in detail here.

[0084] S503: Inputting the prediction feature graph into the power generation prediction model trained by the training method of the above embodiment to output the power generation at multiple time points on the target date.

[0085] Furthermore, the power generation prediction model obtained by the training method of the above embodiment is used to predict the prediction feature graph to obtain the power generation at 96 time points on the target date.

[0086] In this embodiment, the data to be predicted is obtained, and the first matrix, the second matrix, and the third matrix are used as feature maps and fused to obtain a prediction feature map, and the prediction feature map is input into the power generation prediction model trained by the training method of the above embodiment to output the power generation at multiple time points on the target date, so that the power generation prediction model can be used to accurately predict the power generation of the wind turbine.

[0087] In order to implement the above embodiment, the present disclosure also proposes a wind turbine generator power generation prediction model training device.

[0088] Figure 6 2 is a schematic diagram of a wind turbine generator set power generation prediction model training device provided according to another embodiment of the present disclosure.

[0089] like Figure 6 As shown, the wind turbine generator power generation prediction model training device 60 includes:

[0090] A first acquisition module 601 is configured to acquire a plurality of training data, wherein each training data piece includes a first matrix consisting of various unit operation data at various time points each day, a second matrix consisting of various wind tower data, and a third matrix consisting of various weather forecast data;

[0091] A first fusion module 602 is configured to fuse the first matrix, the second matrix, and the third matrix of each piece of training data as feature maps to obtain a training feature map; and

[0092] An input module 603 is used to input the training feature maps into the initial network model respectively to output multiple predicted powers at multiple time points;

[0093] The training module 604 is used to calculate the target loss value based on the actual power and predicted power at multiple time points using a preset loss function, and optimize the initial network model with reference to the target loss value until convergence to obtain a power generation prediction model.

[0094] In some embodiments, the training module 604 is specifically used to: divide multiple time points into a period of peak new energy generation, a period of peak electricity consumption, and other periods; calculate a first loss value based on the actual power and predicted power at multiple time points in the period of peak new energy generation, and calculate a second loss value based on the actual power and predicted power at multiple time points in the period of peak electricity consumption, and calculate a third loss value based on the actual power and predicted power at multiple time points in other periods; and accumulate the first loss value, the second loss value, and the third loss value to obtain a target loss value.

[0095] In some embodiments, the training module 604 is specifically used to: determine a first hyperparameter from a hyperparameter interval corresponding to a new energy peak generation period; and calculate a first loss value based on the actual power and predicted power at multiple time points within the new energy peak generation period and the first hyperparameter.

[0096] In some embodiments, the training module 604 is specifically used to: determine a second hyperparameter from a hyperparameter interval corresponding to the peak power consumption period; and calculate a second loss value based on the actual power and predicted power at multiple time points within the peak power consumption period and the second hyperparameter.

[0097] In some embodiments, the first loss value is calculated according to the following loss function:

[0098]

[0099] Among them, Loss1 represents the first loss value, N represents the number of time points during the new energy generation period, and y i It represents the actual power at time i during the new energy generation period. represents the prediction power at time point i, α represents the first hyperparameter, and α∈[0.55,0.9].

[0100] In some embodiments, the second loss value is calculated according to the following loss function:

[0101]

[0102] Among them, Loss2 represents the second loss value, L represents the number of time points during the peak power consumption period, and y i represents the actual power at time i during the peak electricity period, represents the prediction power at time point i, β represents the first hyperparameter, and β∈[0.3,0.7].

[0103] In some embodiments, the fusion module 602 is specifically used to: map the first matrix, the second matrix, and the third matrix into matrices of the same dimension; and use the mapped matrices of the same dimension as feature maps and perform fusion processing to obtain a training feature map.

[0104] Figure 7 Schematic diagram of a wind turbine generator system power generation prediction device according to another embodiment of the present disclosure.

[0105] like Figure 7 As shown, the wind turbine generator power generation prediction device 70 includes:

[0106] The second acquisition module 701 is configured to acquire data to be predicted, wherein the data to be predicted includes a first matrix consisting of various unit operation data at multiple time points on a target date, a second matrix consisting of various wind tower data, and a third matrix consisting of various weather forecast data;

[0107] A second fusion module 702 is configured to fuse the first matrix, the second matrix, and the third matrix as feature maps to obtain a predicted feature map; and

[0108] The prediction module 703 is used to input the prediction feature graph into the power generation prediction model trained by the above-mentioned training device to output the power generation at multiple time points on the target date.

[0109] In this embodiment, a plurality of training data are obtained, wherein each training data includes a first matrix composed of a plurality of unit operation data at a plurality of time points every day, a second matrix composed of a plurality of wind tower data, and a third matrix composed of a plurality of weather forecast data, and the first matrix, the second matrix and the third matrix of each training data are respectively used as feature maps and fused to obtain training feature maps, and the training feature maps are respectively input into the initial network model to output a plurality of predicted powers at a plurality of time points, and a preset loss function is used to calculate a target loss value based on the actual power and the predicted power at a plurality of time points, and the initial network model is optimized with reference to the target loss value until convergence to obtain a power generation prediction model, and the power generation prediction model can be trained by integrating a plurality of data, so that the power generation prediction model can be used to accurately predict the power generation of the wind turbine.

[0110] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0111] In order to implement the above embodiments, the present disclosure further proposes a computer program product. When an instruction processor in the computer program product is executed, the wind turbine power generation prediction model training method proposed in the above embodiments of the present disclosure is executed.

[0112] Figure 8 A block diagram of an exemplary electronic device suitable for implementing embodiments of the present disclosure is shown. Figure 8 The electronic device 12 shown is only an example and should not limit the functionality and scope of use of the embodiments of the present disclosure.

[0113] like Figure 8 As shown, electronic device 12 is implemented as a general-purpose computing device. Components of electronic device 12 may include, but are not limited to, one or more processors or processing units 16, system memory 28, and a bus 18 that connects various system components (including system memory 28 and processing unit 16).

[0114] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor, or a local bus using any of a variety of bus architectures. Examples of such architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnection (PCI) bus.

[0115] The electronic device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the electronic device 12, including volatile and non-volatile media, removable and non-removable media.

[0116] The memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The electronic device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be configured to read and write non-removable, non-volatile magnetic media ( Figure 8 Not shown, often called a "hard drive").

[0117] although Figure 8 Although not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a Compact Disc Read Only Memory (hereinafter referred to as: CD-ROM), a Digital Video Disc Read Only Memory (hereinafter referred to as: DVD-ROM), or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The memory 28 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the various embodiments of the present disclosure.

[0118] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in memory 28. Such program modules 42 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data, each of which, or some combination thereof, may include an implementation of a network environment. Program modules 42 generally implement the functions and / or methods of the embodiments described herein.

[0119] The electronic device 12 can also communicate with one or more external devices 14 (e.g., a keyboard, pointing device, display 24, etc.), one or more devices that enable a user to interact with the electronic device 12, and / or any device that enables the electronic device 12 to communicate with one or more other computing devices (e.g., a network card, a modem, etc.). This communication can occur via an input / output (I / O) interface 22. Furthermore, the electronic device 12 can communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network such as the Internet) via a network adapter 20. As shown, the network adapter 20 communicates with other modules of the electronic device 12 via the bus 18. It should be understood that, although not shown, other hardware and / or software modules can be used in conjunction with the electronic device 12, including but not limited to microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0120] The processing unit 16 executes various functional applications by running programs stored in the system memory 28 , such as implementing the wind turbine power generation prediction model training method mentioned in the above embodiment.

[0121] Other embodiments of the present disclosure will readily occur to those skilled in the art after considering the specification and practicing the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the present disclosure being indicated by the following claims.

[0122] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes can be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

[0123] It should be noted that, in the description of this disclosure, the terms "first", "second", etc. are used for descriptive purposes only and should not be understood as indicating or implying relative importance. In addition, in the description of this disclosure, unless otherwise specified, the meaning of "plurality" is two or more.

[0124] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present disclosure includes additional implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present disclosure belong.

[0125] It should be understood that various parts of the present disclosure can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0126] Those skilled in the art will understand that all or part of the steps in the method of the above embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0127] In addition, the functional units in the various embodiments of the present disclosure may be integrated into a single processing module, or each unit may exist physically separately, or two or more units may be integrated into a single module. The aforementioned integrated modules may be implemented in the form of hardware or in the form of software functional modules. If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they may also be stored in a computer-readable storage medium.

[0128] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0129] Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present disclosure. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0130] Although the embodiments of the present disclosure have been shown and described above, it is understood that the above embodiments are illustrative and are not to be construed as limitations on the present disclosure. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present disclosure.

Claims

1. A wind turbine power generation prediction model training method, characterized in that: The method comprises: Acquire multiple pieces of training data, wherein each piece of training data includes a first matrix consisting of multiple unit operation data at multiple time points every day, a second matrix consisting of multiple wind tower data, and a third matrix consisting of multiple weather forecast data; The first matrix, the second matrix, and the third matrix of each piece of training data are respectively used as feature maps and fused to obtain a training feature map; and Inputting the training feature graphs into the initial network model respectively to output a plurality of predicted powers at a plurality of time points; Calculating a target loss value based on the actual power and the predicted power at multiple time points using a preset loss function, and optimizing the initial network model with reference to the target loss value until convergence, thereby obtaining a power generation prediction model; The calculating the loss value based on the actual power and the predicted power at multiple time points using a preset loss function includes: Dividing the multiple time points into new energy peak generation period, electricity peak period, and other periods; Calculating a first loss value based on the actual power and the predicted power at multiple time points during the renewable energy peak generation period, calculating a second loss value based on the actual power and the predicted power at multiple time points during the peak power consumption period, and calculating a third loss value based on the actual power and the predicted power at multiple time points during the other time periods; and The first loss value, the second loss value, and the third loss value are accumulated to obtain the target loss value.

2. The method according to claim 1, wherein The calculating of the first loss value based on the actual power and the predicted power at multiple time points during the new energy generation period includes: Determining a first hyperparameter from the hyperparameter interval corresponding to the new energy peak generation period; and The first loss value is calculated based on the actual power and the predicted power at multiple time points within the new energy generation period and the first hyperparameter.

3. The method according to claim 1, wherein The calculating the second loss value based on the actual power and the predicted power at multiple time points during the peak power consumption period includes: Determining a second hyperparameter from the hyperparameter interval corresponding to the peak power consumption period; and The second loss value is calculated based on the actual power and the predicted power at multiple time points during the peak power consumption period and the second hyperparameter.

4. The method according to claim 2, wherein in, The first loss value is calculated according to the following loss function: in, represents the first loss value, Indicates the number of time points during the new energy generation period, It represents the actual power at time i during the new energy generation period. represents the predicted power at time point i, represents the first hyperparameter, and .

5. The method according to claim 3, wherein in, The second loss value is calculated according to the following loss function: in, represents the second loss value, Indicates the number of time points during peak electricity consumption period, represents the actual power at time i during the peak electricity period, represents the predicted power at time point i, represents the second hyperparameter, and .

6. The method according to claim 1, wherein The first matrix, the second matrix, and the third matrix of each piece of training data are respectively used as feature maps and fused to obtain a training feature map, including: Mapping the first matrix, the second matrix, and the third matrix into matrices of the same dimension; and The matrix of the same dimension after mapping is used as the feature map and fused to obtain the training feature map.

7. A method for predicting the power generation of a wind turbine, characterized in that: The method comprises: Acquiring data to be predicted, wherein the data to be predicted includes a first matrix consisting of various unit operation data at multiple time points on a target date, a second matrix consisting of various wind tower data, and a third matrix consisting of various weather forecast data; The first matrix, the second matrix, and the third matrix are used as feature maps and are fused to obtain a predicted feature map; and The prediction feature graph is input into a power generation prediction model trained by the method according to any one of claims 1 to 6 above, so as to output the power generation at multiple time points on the target date.

8. A wind turbine generator power generation prediction model training device, characterized in that: include: A first acquisition module is configured to acquire a plurality of training data, wherein each training data includes a first matrix consisting of a plurality of unit operation data at a plurality of time points each day, a second matrix consisting of a plurality of wind tower data, and a third matrix consisting of a plurality of weather forecast data; a first fusion module, configured to use the first matrix, the second matrix, and the third matrix of each piece of training data as feature maps and perform fusion processing to obtain a training feature map; and An input module, configured to input the training feature maps into an initial network model respectively to output a plurality of predicted powers at a plurality of time points; A training module is used to calculate a target loss value based on the actual power and the predicted power at multiple time points using a preset loss function, and optimize the initial network model with reference to the target loss value until convergence to obtain a power generation prediction model. The calculation of the loss value based on the actual power and the predicted power at multiple time points using the preset loss function includes: dividing the multiple time points into a new energy peak generation period, a peak power consumption period, and other periods; calculating a first loss value based on the actual power and the predicted power at multiple time points within the new energy peak generation period, and calculating a second loss value based on the actual power and the predicted power at multiple time points within the peak power consumption period, and calculating a third loss value based on the actual power and the predicted power at multiple time points within the other periods; and accumulating the first loss value, the second loss value, and the third loss value to obtain the target loss value.

9. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 7.

Images