Generated power prediction model training method and device, equipment and medium

Abstract

The invention relates to a generation power prediction model training method and device, equipment and a medium. First training data of a source wind power plant and second training data of a target wind power plant are obtained, and the first operation time of the source wind power plant is longer than the second operation time of the target wind power plant; determining a target learning rate for finely adjusting a pre-trained prediction model by using the first training data, the second training data and the pre-trained prediction model of the source wind power plant; and according to the target learning rate and the second training data, performing fine tuning on the pre-trained prediction model to obtain a generated power prediction model of the target wind power plant. Therefore, when facing the target wind power plant which can only provide a small amount of training data, the fine tuning method suitable for the target wind power plant can be determined according to the pre-trained prediction model, the small amount of training data of the target wind power plant and enough training data of the source wind power plant; and the reliability and applicability of the generated power prediction model obtained by training can be improved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of wind farm output power prediction, and in particular to a method, device, equipment and medium for training a power generation prediction model. Background Art

[0002] At present, small sample wind power prediction technology aims to obtain a high-precision power generation prediction model with only a small amount of training data. The common method is to first pre-train on a source wind farm with sufficient training data, transfer the learned prediction knowledge to a target wind farm with only a small amount of training data (such as a newly built or expanded wind farm) and perform fine-tuning to obtain a power generation prediction model.

[0003] The current fine-tuning method is to manually set the same learning rate to fine-tune the model parameters. The method is single and has poor universality, and its ability to improve the prediction effect is limited. Therefore, providing a fine-tuning method suitable for the target wind farm to improve the reliability and applicability of the training results of the power generation prediction model is a technical problem that needs to be solved urgently. Summary of the invention

[0004] In order to solve the above technical problems, the present disclosure provides a power generation prediction model training method, device, equipment and medium.

[0005] In a first aspect, the present disclosure provides a method for training a power generation prediction model, comprising:

[0006] Acquire first training data of a source wind farm and second training data of a target wind farm, wherein a first operating time of the source wind farm is greater than a second operating time of the target wind farm;

[0007] Determine a target learning rate for fine-tuning the pre-trained prediction model using the first training data, the second training data, and the pre-trained prediction model of the source wind farm;

[0008] The pre-trained prediction model is fine-tuned according to the target learning rate and the second training data to obtain a power generation prediction model of the target wind farm.

[0009] In a second aspect, the present disclosure provides a power generation prediction model training device, comprising:

[0010] A data acquisition module, configured to acquire first training data of a source wind farm and second training data of a target wind farm, wherein a first operating time of the source wind farm is greater than a second operating time of the target wind farm;

[0011] a learning rate determination module, configured to determine a target learning rate for fine-tuning the pre-trained prediction model by using the first training data, the second training data, and the pre-trained prediction model of the source wind farm;

[0012] The model fine-tuning module is used to fine-tune the pre-trained prediction model according to the target learning rate and the second training data to obtain the power generation prediction model of the target wind farm.

[0013] In a third aspect, an embodiment of the present disclosure further provides an electronic device, the device comprising:

[0014] one or more processors;

[0015] a storage device for storing one or more programs,

[0016] When one or more programs are executed by one or more processors, the one or more processors implement the method provided in the first aspect.

[0017] In a fourth aspect, an embodiment of the present disclosure further provides a computer-readable storage medium having a computer program stored thereon, which implements the method provided in the first aspect when the computer program is executed by a processor.

[0018] Compared with the prior art, the technical solution provided by the embodiments of the present disclosure has the following advantages:

[0019] The disclosed embodiments provide a method, device, equipment and medium for training a power generation prediction model, which obtains first training data of a source wind farm and second training data of a target wind farm, wherein the first operating time of the source wind farm is greater than the second operating time of the target wind farm; uses the first training data, the second training data and the pre-trained prediction model of the source wind farm to determine a target learning rate for fine-tuning the pre-trained prediction model; and fine-tunes the pre-trained prediction model according to the target learning rate and the second training data to obtain a power generation prediction model for the target wind farm. Thus, when faced with a target wind farm that can only provide a small amount of training data, a fine-tuning method suitable for the target wind farm can be determined based on the pre-trained prediction model, a small amount of training data of the target wind farm and sufficient training data of the source wind farm, which is conducive to improving the reliability and applicability of the trained power generation prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0022] Figure 1 A schematic diagram of a flow chart of a method for training a power generation prediction model provided by an embodiment of the present disclosure;

[0023] Figure 2 A flowchart of another power generation prediction model training method provided by an embodiment of the present disclosure;

[0024] Figure 3 A schematic diagram of the structure of a power generation prediction model training device provided by an embodiment of the present disclosure;

[0025] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0026] 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 in the embodiments can be combined with each other without conflict.

[0027] 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.

[0028] In recent years, artificial intelligence models have become the mainstream trend in the field of wind power prediction, which often require a large amount of training data for data-driven. Since the target wind farm (such as a newly built or expanded wind farm) has a short operating time and has not accumulated enough historical data, it is difficult to establish a high-precision prediction model, and the prediction accuracy is poor. It is necessary to pre-train on the source wind farm with sufficient training data, transfer the learned prediction knowledge to the target wind farm (such as a newly built or expanded wind farm) with only a small amount of training data, and perform fine-tuning to obtain a power generation prediction model.

[0029] It is understandable that it is important to set an appropriate learning rate for the fine-tuning method. If the learning rate is too large, the model may not converge or even oscillate; if the learning rate is too small, the convergence speed of the model will be reduced. If different target wind farms use the same learning rate to fine-tune the model parameters, the existing fine-tuning method cannot be applied to different target wind farms and the method is single, which cannot effectively improve the prediction effect and adaptability of the model.

[0030] In order to improve the reliability and applicability of a trained power generation prediction model, the embodiments of the present disclosure provide a power generation prediction model training method, device, equipment and medium.

[0031] Combine the following Figure 1 The power generation prediction model training method provided in the embodiment of the present disclosure is described. In the embodiment of the present disclosure, the power generation prediction model training method can be executed by an electronic device or a server. Among them, the electronic device may include a device with communication function such as a tablet computer, a desktop computer, a laptop computer, etc., and may also include a virtual machine or a device simulated by a simulator. The server may include a server cluster and a cloud server.

[0032] Figure 1 A flow chart of a method for training a power generation prediction model provided by an embodiment of the present disclosure is shown.

[0033] like Figure 1 As shown, the power generation prediction model training method may include the following steps.

[0034] S110 . Acquire first training data of a source wind farm and second training data of a target wind farm, wherein a first operating time of the source wind farm is greater than a second operating time of the target wind farm.

[0035] The source wind farm refers to a wind farm that has been in operation for a long time and has sufficient training data, while the target wind farm refers to a wind farm that has been newly built, expanded, or rebuilt and has only a small amount of training data.

[0036] It is understandable that, since the first operation time of the source wind farm is greater than the second operation time of the target wind farm, the source wind farm can provide sufficient training data, while the target wind farm can only provide a small amount of training data. Therefore, when facing a target wind farm that can only provide a small amount of training data, a small amount of training data of the target wind farm is obtained, and combined with the sufficient training data of the source wind farm, a power generation prediction model suitable for the target wind farm is trained.

[0037] The first training data includes first numerical weather forecast data and first reference wind power.

[0038] Among them, the first numerical weather forecast data can be understood as solving the fluid mechanics and thermodynamics equations describing the weather evolution process near the source wind farm through numerical calculation according to the actual atmospheric conditions under certain initial and boundary conditions, and obtaining meteorological forecast data reflecting the state of atmospheric movement in a certain period of time in the future. Optionally, the first numerical weather forecast data is the weather forecast data T hours after the historical moment, and the first numerical weather forecast data is divided into meteorological data at N moments. The first numerical weather forecast data includes but is not limited to data such as wind speed, wind direction, air pressure, temperature and humidity.

[0039] Specifically, each type of meteorological data in the first numerical weather forecast data can be arranged in chronological order to obtain a one-dimensional weather forecast sequence, and then the multiple meteorological data can be aligned at the same time point to obtain two-dimensional numerical weather forecast data, and the two-dimensional numerical weather forecast data can be used as the first numerical weather forecast data.

[0040] The first reference wind power refers to the actually measured power generation of the source wind farm.

[0041] Among them, the meta-learning rate generator can be understood as a neural network model, and the basic structure is a fully connected neural network. In order to control the learning rate output by the model between 0 and 1, the activation function of the last layer of neurons in the meta-learning rate generator is the Sigmoid function.

[0042] The second training data includes second numerical weather forecast data and second reference wind power, and the meaning of the second numerical weather forecast data is similar to that of the first numerical weather forecast data, and the meaning of the second reference wind power is similar to that of the first reference wind power.

[0043] S120 , using the first training data, the second training data, and the pre-trained prediction model of the source wind farm, determine a target learning rate for fine-tuning the pre-trained prediction model.

[0044] The method for determining the pre-trained prediction model includes: using the first training data to pre-train the original network to obtain the pre-trained prediction model.

[0045] Optionally, the trained prediction model includes but is not limited to pre-training a complex neural network with a multi-layer structure such as a long short-term memory network (LSTM) and a recurrent neural network (GRU). Specifically, firstly, the parameters of the original network are initialized by using zero initialization, random initialization, Xavier initialization, and He initialization, and then the parameters of the original network are pre-trained E times based on the first training data to obtain a pre-trained prediction model.

[0046] Among them, the pre-training process can usually choose conventional transfer learning pre-training methods and meta-learning MAML, Reptile pre-training methods and other methods. In addition, set 1≤E≤1000, the specific value of E is determined by the "early stopping method", which is to use the first training data to update the parameters of the original network, and use the validation set sample data to verify the accuracy. When the accuracy of the validation set drops p times in a row, the iterative process is terminated, and the parameters of the original network are no longer updated to obtain a pre-trained prediction model.

[0047] It is understandable that, since the target wind farm can only provide a small amount of second training data, the target learning rate applicable to the target wind farm can be obtained by combining the second training data, the pre-trained prediction model of the source wind farm, and the first training data, and then the power generation prediction model applicable to the target wind farm can be fine-tuned based on the target learning rate. Specifically, the learning rate generator can be determined using the first training data and the pre-trained prediction model of the source wind farm, and the target learning rate can be determined based on the learning rate generator and the second training data.

[0048] S130. Fine-tune the pre-trained prediction model according to the target learning rate and the second training data to obtain a power generation prediction model of the target wind farm.

[0049] In this embodiment, the pre-trained prediction model is iteratively fine-tuned using the second training data, and the step size in each iteration is determined by the target learning rate, so that the loss value of the pre-trained prediction model converges to a minimum value, thereby determining a power generation prediction model suitable for the target wind farm.

[0050] The disclosed embodiment provides a method for training a power generation prediction model, which obtains first training data of a source wind farm and second training data of a target wind farm, wherein the first operating time of the source wind farm is greater than the second operating time of the target wind farm; using the first training data, the second training data, and the pre-trained prediction model of the source wind farm, a target learning rate for fine-tuning the pre-trained prediction model is determined; and according to the target learning rate and the second training data, the pre-trained prediction model is fine-tuned to obtain a power generation prediction model of the target wind farm. Thus, when faced with a target wind farm that can only provide a small amount of training data, a fine-tuning method suitable for the target wind farm can be determined based on the pre-trained prediction model, a small amount of training data of the target wind farm, and sufficient training data of the source wind farm, which is conducive to improving the reliability and applicability of the trained power generation prediction model.

[0051] Further, after determining the power generation prediction model of the target wind farm, the power generation prediction model can be used to predict the power generation of the target wind farm. Accordingly, after executing S130, the method further includes: obtaining real-time numerical weather forecast data of the target wind farm; and processing the real-time numerical weather forecast data using the power generation prediction model to obtain the real-time predicted wind power of the target wind farm.

[0052] The real-time numerical weather forecast data includes but is not limited to wind speed, wind direction, air pressure, temperature, humidity, etc. Specifically, each meteorological data in the real-time numerical weather forecast data can be arranged in chronological order to obtain a one-dimensional weather forecast sequence, and then multiple meteorological data can be aligned at the same time point to obtain two-dimensional numerical weather forecast data, and the two-dimensional numerical weather forecast data can be used as the real-time numerical weather forecast data.

[0053] In this way, by using the fine-tuned power generation prediction model, a high-precision small-sample wind power prediction for the target wind farm is achieved, thereby ensuring the prediction accuracy of the power generation of the target wind farm.

[0054] In another embodiment of the present disclosure, a method for determining the target learning rate is explained in detail.

[0055] Figure 2 A flow chart of another power generation prediction model training method provided by an embodiment of the present disclosure is shown.

[0056] like Figure 2 As shown, the power generation prediction model training method may include the following steps.

[0057] S210: Acquire first training data of a source wind farm and second training data of a target wind farm, wherein a first operating time of the source wind farm is greater than a second operating time of the target wind farm.

[0058] Among them, S210 is similar to S110 and will not be described in detail here.

[0059] S220: Using the first training data and the pre-trained prediction model, train a meta-learning rate generator.

[0060] Among them, the first training data includes first numerical weather forecast data and a first reference wind power. Accordingly, S220 specifically includes the following steps: using a pre-trained prediction model to process the first numerical weather forecast data to obtain a first predicted wind power; based on the first predicted wind power and the first reference wind power, calculating a first loss value of the pre-trained prediction model; according to the first loss value, updating the parameters of the initial learner to obtain a meta-learning rate generator.

[0061] Specifically, first, the parameters of the initial learner are initialized using methods such as zero initialization, random initialization, Xavier initialization, and He initialization; then, the pre-trained prediction model, the first numerical weather forecast data, the first reference wind power, and the meta-learning algorithm are used to perform F rounds of parameter updates on the initial learner to obtain a meta-learning rate generator.

[0062] Among them, 1≤F≤1000 is set, and the specific value of F is determined by the "early stopping method". The principle of the "early stopping method" here is: use the training set sample data to update the meta-learning rate generator parameters; use the validation set sample data and the meta-learning rate generator after each round of update to update the wind power prediction source model parameters S times; use the test set sample data and the updated wind power prediction source model to verify the accuracy. When the test set accuracy decreases p times in a row, terminate the iterative process and no longer update the meta-learning generator.

[0063] Specifically, the pre-trained prediction model, the first numerical weather forecast data, the first reference wind power, and the meta-learning algorithm are used to perform F rounds of parameter updates on the initial learner to obtain a meta-learning rate generator, including the following processes: selecting N source wind farms, where N is an integer greater than or equal to 1 and needs to be set manually; calculating the loss value of each wind farm; calculating the total loss value of the N source wind farms; calculating the parameter update gradient of the meta-learning rate generator according to the total loss value, and updating the parameters of the meta-learning rate generator. SGD, AdaGrad, Adam and other algorithms can be selected here for parameter update until the parameters of the meta-learning rate generator meet the requirements of the "early stopping method" to obtain the meta-learning rate generator.

[0064] Specifically, the loss value of each wind farm is calculated, including the following process: copying the pre-trained prediction model as the wind power prediction sub-model exclusive to each source wind farm; sampling the first training data of the source wind farm S-1 times, and updating the parameters of the wind power prediction sub-model S-1 steps; sampling the source wind farm for the Sth time, inputting the updated wind power prediction sub-model, and obtaining the corresponding wind power prediction sequence; calculating the loss of each source wind farm according to the predicted value and actual value of the wind power sequence.

[0065] Specifically, the first training data of the source wind farm is sampled S-1 times, and the parameters of the wind power prediction sub-model are updated S-1 steps, including: randomly selecting the first training sample of group B from the first training sample of the source wind farm, where B is an integer greater than or equal to 1 and needs to be set manually; inputting the numerical weather forecast data of the first training sample of group B into the wind power prediction sub-model exclusive to each source wind farm to obtain the corresponding wind power prediction sequence of group B; calculating the loss value between the wind power prediction sequence of group B and the first reference wind power; calculating the weight gradient and bias gradient of each network layer in the wind power prediction sub-model exclusive to each source wind farm according to the obtained loss value; inputting the weight gradient and bias gradient of each network layer in the wind power prediction sub-model exclusive to each source wind farm into the meta-learning rate generator to obtain different learning rates corresponding to the weight and bias of each network layer; and updating the parameters of the wind power prediction sub-model exclusive to each source wind farm according to the different learning rates of the weight and bias of each network layer.

[0066] Optionally, the weight gradient and bias gradient can be determined as follows:

[0067]

[0068] Among them, g(θ) includes the weight gradient and bias gradient of each network layer in the wind power prediction sub-model exclusive to each source wind farm, L is the loss value between the wind power prediction sequence of group B and the first reference wind power, and θ is the parameter of the wind power prediction sub-model exclusive to each source wind farm.

[0069] Optionally, the parameters of the wind power prediction sub-model can be updated as follows:

[0070] θ t =θ t-1 -lrg t-1

[0071] Where t is the tth update, θ t is the parameter after the tth update, g t-1 is the parameter θ t-1 The gradient of lr is the gradient corresponding to the parameter θ t-1 The learning rate.

[0072] In this way, the meta-learning rate generator can be trained by utilizing sufficient training data and a meta-learning algorithm provided by the source wind farm, so that the learning rate generator can be utilized conveniently to accurately determine a target learning rate suitable for the target wind farm.

[0073] S230: Determine a target learning rate based on the second training data, the pre-trained prediction model, and the meta-learning rate generator.

[0074] Among them, the second training data includes second numerical weather forecast data and second reference wind power. Accordingly, S230 specifically includes the following steps: using a pre-trained prediction model to process the second numerical weather forecast data to obtain a second predicted wind power; based on the second predicted wind power and the second reference wind power, calculating the second loss value of the pre-trained prediction model; according to the second loss value and the meta-learning rate generator, calculating the target learning rate.

[0075] Among them, the target learning rate is calculated according to the second loss value and the meta-learning rate generator, including the following methods: according to the second loss value, the gradient data of each network layer in the pre-trained prediction model is calculated; using the meta-learning rate generator, the learning rate is calculated on the gradient data to obtain the target learning rate.

[0076] Specifically, first, the pre-trained prediction model is copied to obtain a wind power prediction sub-model exclusive to the target wind farm; then, the wind power prediction sub-model exclusive to the target wind farm is updated G times using the second training data to obtain a target learning rate.

[0077] Specifically, the second training data is used to update the wind power prediction sub-model exclusive to the target wind farm G times to obtain the target learning rate, including: randomly selecting C groups of training samples in the second training data, where C is an integer greater than or equal to 1 and needs to be set manually; inputting the second numerical weather forecast data of the C group of training samples into the wind power prediction sub-model exclusive to the target wind farm to obtain the corresponding C group of wind power prediction sequences; calculating the loss value between the C group of wind power prediction sequences and the second reference wind power; according to the loss value, calculating the weight gradient and bias gradient of each network layer in the wind power prediction sub-model exclusive to the target wind farm; inputting the weight gradient and bias gradient of each network layer in the wind power prediction sub-model into the meta-learning rate generator to obtain the target learning rate of the wind power prediction sub-model.

[0078] Optionally, the weight gradient and bias gradient can be determined as follows:

[0079]

[0080] Among them, g(θ) includes the weight gradient and bias gradient of each network layer in the wind power prediction sub-model exclusive to the target wind farm, L is the loss value between the wind power prediction sequence of group C and the second reference wind power, and θ is the parameter of the wind power prediction sub-model exclusive to the target wind farm.

[0081] Therefore, the meta-learning rate generator can be used to automatically generate the optimal learning rate for different parameters as the target learning rate, so as to efficiently update the parameters of the pre-trained prediction model.

[0082] S240: fine-tune the pre-trained prediction model according to the target learning rate and the second training data to obtain a power generation prediction model of the target wind farm.

[0083] Specifically, the target learning rate and the second training data are used to update the parameters of the pre-trained prediction model to achieve model fine-tuning and obtain the power generation prediction model of the target wind farm.

[0084] Optionally, the parameters of the pre-trained prediction model can be updated as follows:

[0085] θ t =θ t-1 -lrg t-1

[0086] Where t is the tth update, θ t is the parameter after the tth update, g t-1 is the parameter θ t-1 The gradient of lr is the gradient corresponding to the parameter θ t-1 The learning rate.

[0087] In the above manner, by using a target learning rate suitable for a target wind farm, the efficiency of the fine-tuning stage of small sample wind power prediction can be improved, thereby facilitating the realization of high-precision small sample wind power prediction.

[0088] The present disclosure also provides a power generation prediction model training device for implementing the above-mentioned power generation prediction model training method. Figure 3 In the disclosed embodiment, the power generation prediction model training device may be an electronic device or a server. The electronic device may include a tablet computer, a desktop computer, a laptop computer, or other device with a communication function, or may include a virtual machine or a device simulated by a simulator. The server may include a server cluster and a cloud server.

[0089] Figure 3 A schematic structural diagram of a power generation prediction model training device provided by an embodiment of the present disclosure is shown.

[0090] like Figure 3 As shown, the power generation prediction model training device 300 may include:

[0091] A data acquisition module 310 is used to acquire first training data of a source wind farm and second training data of a target wind farm, wherein the first operating time of the source wind farm is greater than the second operating time of the target wind farm;

[0092] A learning rate determination module 320, configured to determine a target learning rate for fine-tuning the pre-trained prediction model by using the first training data, the second training data, and the pre-trained prediction model of the source wind farm;

[0093] The model fine-tuning module 330 is used to fine-tune the pre-trained prediction model according to the target learning rate and the second training data to obtain the power generation prediction model of the target wind farm.

[0094] A power generation prediction model training device of an embodiment of the present disclosure obtains first training data of a source wind farm and second training data of a target wind farm, wherein the first operating time of the source wind farm is greater than the second operating time of the target wind farm; uses the first training data, the second training data, and a pre-trained prediction model of the source wind farm to determine a target learning rate for fine-tuning the pre-trained prediction model; and fine-tunes the pre-trained prediction model according to the target learning rate and the second training data to obtain a power generation prediction model of the target wind farm. Thus, when faced with a target wind farm that can only provide a small amount of training data, a fine-tuning method suitable for the target wind farm can be determined based on the pre-trained prediction model, a small amount of training data of the target wind farm, and sufficient training data of the source wind farm, which is conducive to improving the reliability and applicability of the trained power generation prediction model.

[0095] In some embodiments of the present disclosure, the learning rate determination module 320 includes:

[0096] A training unit, configured to train a meta-learning rate generator using the first training data and the pre-trained prediction model;

[0097] A determination unit is used to determine the target learning rate based on the second training data, the pre-trained prediction model and the meta learning rate generator.

[0098] In some embodiments of the present disclosure, the first training data includes first numerical weather forecast data and first reference wind power. Accordingly, the training unit is specifically used for:

[0099] Processing the first numerical weather forecast data using the pre-trained prediction model to obtain a first predicted wind power;

[0100] Calculating a first loss value of the pre-trained prediction model based on the first predicted wind power and the first reference wind power;

[0101] According to the first loss value, the parameters of the initial learner are updated to obtain the meta learning rate generator.

[0102] In some embodiments of the present disclosure, the second training data includes second numerical weather forecast data and second reference wind power. Accordingly, the determining unit is specifically configured to:

[0103] Processing the second numerical weather forecast data using the pre-trained prediction model to obtain a second predicted wind power;

[0104] Calculating a second loss value of the pre-trained prediction model based on the second predicted wind power and the second reference wind power;

[0105] The target learning rate is calculated according to the second loss value and the meta learning rate generator.

[0106] In some embodiments of the present disclosure, the determining unit is specifically used to: calculate the gradient data of each network layer in the pre-trained prediction model according to the second loss value;

[0107] The meta-learning rate generator is used to calculate the learning rate of the gradient data to obtain the target learning rate.

[0108] In some embodiments of the present disclosure, the device further includes:

[0109] The pre-training module is used to pre-train the original network using the first training data to obtain the pre-trained prediction model.

[0110] In some embodiments of the present disclosure, the device further includes:

[0111] A real-time data acquisition module, used to acquire real-time numerical weather forecast data of the target wind farm;

[0112] The power prediction module is used to process the real-time numerical weather forecast data using the power generation prediction model to obtain the real-time predicted wind power of the target wind farm.

[0113] It should be noted that Figure 3 The power generation prediction model training device 300 shown can execute Figures 1-2 The various steps in the method embodiment shown in the figure are implemented Figures 1-2 The various processes and effects in the method embodiment shown are not described in detail here.

[0114] Figure 4 A schematic structural diagram of an electronic device provided by an embodiment of the present disclosure is shown.

[0115] like Figure 4 As shown, the electronic device may include a processor 401 and a memory 402 storing computer program instructions.

[0116] Specifically, the processor 401 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0117] The memory 402 may include a large capacity memory for information or instructions. By way of example and not limitation, the memory 402 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a tape, or a universal serial bus (USB) drive or a combination of two or more of these. Where appropriate, the memory 402 may include a removable or non-removable (or fixed) medium. Where appropriate, the memory 402 may be inside or outside the integrated gateway device. In a particular embodiment, the memory 402 is a non-volatile solid-state memory. In a particular embodiment, the memory 402 includes a read-only memory (ROM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (Electrically Erasable Programmable ROM, EEPROM), an electrically rewritable ROM (EAROM) or a flash memory, or a combination of two or more of these.

[0118] The processor 401 reads and executes the computer program instructions stored in the memory 402 to perform the steps of the power generation prediction model training method or the wind power prediction method provided in the embodiments of the present disclosure.

[0119] In one example, the electronic device may further include a transceiver 403 and a bus 404. Figure 4 As shown, the processor 401, the memory 402 and the transceiver 403 are connected via a bus 404 and communicate with each other.

[0120] The bus 404 includes hardware, software, or both. For example, but not limitation, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industrial Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a Memory bus, a MicroChannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses or a combination of two or more of these. Where appropriate, the bus 404 may include one or more buses. Although embodiments of the present application describe and illustrate a particular bus, the present application contemplates any suitable bus or interconnect.

[0121] The following is an embodiment of a computer-readable storage medium provided in an embodiment of the present disclosure. The computer-readable storage medium and the power generation prediction model training method of the above-mentioned embodiments belong to the same inventive concept. For details not described in detail in the embodiment of the computer-readable storage medium, reference can be made to the embodiment of the above-mentioned power generation prediction model training method.

[0122] This embodiment provides a storage medium containing computer executable instructions. When the computer executable instructions are executed by a computer processor, they are used to execute a power generation prediction model training method. The method is applied to an electronic device. The method includes:

[0123] Acquire first training data of a source wind farm and second training data of a target wind farm, wherein a first operating time of the source wind farm is greater than a second operating time of the target wind farm;

[0124] Determine a target learning rate for fine-tuning the pre-trained prediction model using the first training data, the second training data, and the pre-trained prediction model of the source wind farm;

[0125] The pre-trained prediction model is fine-tuned according to the target learning rate and the second training data to obtain a power generation prediction model of the target wind farm.

[0126] Of course, the storage medium containing computer executable instructions provided in the embodiment of the present disclosure is not limited to the above method operations, and the computer executable instructions can also execute related operations of the power generation prediction model training method provided in any embodiment of the present disclosure.

[0127] Through the above description of the implementation method, the technical personnel in the relevant field can clearly understand that the present disclosure can be implemented with the help of software and necessary general hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present disclosure is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions to enable a computer cloud platform (which can be a personal computer, server, or network cloud platform, etc.) to execute the power generation prediction model training method provided by each embodiment of the present disclosure.

[0128] Note that the above are only preferred embodiments of the present disclosure and the technical principles used. Those skilled in the art will understand that the present disclosure is not limited to the specific embodiments herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present disclosure. Therefore, although the present disclosure is described in more detail through the above embodiments, the present disclosure is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present disclosure, and the scope of the present disclosure is determined by the scope of the appended claims.

Claims

1. A method for training a power generation prediction model, characterized in that: include: Acquire first training data of a source wind farm and second training data of a target wind farm, wherein a first operating time of the source wind farm is greater than a second operating time of the target wind farm; Determine a target learning rate for fine-tuning the pre-trained prediction model using the first training data, the second training data, and the pre-trained prediction model of the source wind farm; The pre-trained prediction model is fine-tuned according to the target learning rate and the second training data to obtain a power generation prediction model of the target wind farm.

2. The method according to claim 1, characterized in that The using the first training data, the second training data, and the pre-trained prediction model of the source wind farm to determine a target learning rate for fine-tuning the pre-trained prediction model includes: Using the first training data and the pre-trained prediction model, training a meta-learning rate generator; The target learning rate is determined based on the second training data, the pre-trained prediction model, and a meta learning rate generator.

3. The method according to claim 2, characterized in that The first training data includes first numerical weather forecast data and first reference wind power. Accordingly, the first training data and the pre-trained prediction model are used to train a meta-learning rate generator, including: Processing the first numerical weather forecast data using the pre-trained prediction model to obtain a first predicted wind power; Calculating a first loss value of the pre-trained prediction model based on the first predicted wind power and the first reference wind power; According to the first loss value, the parameters of the initial learner are updated to obtain the meta learning rate generator.

4. The method according to claim 2, characterized in that: The second training data includes second numerical weather forecast data and a second reference wind power. Accordingly, determining the target learning rate based on the second training data, the pre-trained prediction model and the meta-learning rate generator includes: Processing the second numerical weather forecast data using the pre-trained prediction model to obtain a second predicted wind power; Calculating a second loss value of the pre-trained prediction model based on the second predicted wind power and the second reference wind power; The target learning rate is calculated according to the second loss value and the meta learning rate generator.

5. The method according to claim 4, characterized in that The step of calculating the target learning rate according to the second loss value and the meta learning rate generator includes: Calculating the gradient data of each network layer in the pre-trained prediction model according to the second loss value; The meta-learning rate generator is used to calculate the learning rate of the gradient data to obtain the target learning rate.

6. The method according to claim 1, characterized in that Also includes: The original network is pre-trained using the first training data to obtain the pre-trained prediction model.

7. The method according to any one of claims 1 to 6, characterized in that: Also includes: Acquiring real-time numerical weather forecast data for the target wind farm; The real-time numerical weather forecast data is processed using the power generation prediction model to obtain the real-time predicted wind power of the target wind farm.

8. A power generation prediction model training device, characterized in that: include: A data acquisition module, configured to acquire first training data of a source wind farm and second training data of a target wind farm, wherein a first operating time of the source wind farm is greater than a second operating time of the target wind farm; a learning rate determination module, configured to determine a target learning rate for fine-tuning the pre-trained prediction model by using the first training data, the second training data, and the pre-trained prediction model of the source wind farm; The model fine-tuning module is used to fine-tune the pre-trained prediction model according to the target learning rate and the second training data to obtain the power generation prediction model of the target wind farm.

9. An electronic device, characterized in that: include: processor; A memory for storing executable instructions; The processor is used to read the executable instructions from the memory and execute the executable instructions to implement the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the processor implements the method according to any one of claims 1 to 7.

Images