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

Abstract

The invention relates to a climbing power prediction model training method and device, equipment and a medium. Source domain training data and target domain training data of the reference wind power plant are obtained, the source domain training data comprise training data of various weather events, and the target domain training data comprise training data of climbing events; pre-training the original prediction model based on the source domain training data to obtain a pre-trained prediction model; and according to the target domain training data, performing fine tuning on the pre-trained prediction model to obtain a climbing power prediction model of the reference wind power plant. Due to the fact that feature generality and difference exist between a training sample under a climbing event and a training sample under conventional weather, the method adopts a knowledge migration means, pre-training is conducted through sufficient training data, then parameters of the model are finely adjusted through limited climbing events, the reliability of the climbing power prediction model can be guaranteed, and the prediction accuracy of the climbing power prediction model is improved. And effective adaptation to climbing event characteristics can be realized.

Description

Technical Field

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

[0002] In recent years, extreme weather conditions (such as ramp events) have occurred frequently. The output of wind farms has changed greatly in a short time, resulting in the ramp power of wind farms. It should be particularly noted that the ramp power of wind farms will cause serious imbalance in the active power of the power system, easily cause frequency crossing, and even lead to accidents such as load shedding or large-scale power outages.

[0003] Currently, the ramp power method mainly adopts a data-driven technical route, and a high-precision ramp power model depends on a large amount of training data. However, due to the low frequency of ramp events in wind farms, which belong to small sample events, and the low diversity of sample characteristics of ramp events, the conventional wind power prediction modeling method cannot accurately capture the wind power ramp characteristics under small sample events, and finally it is difficult to train a high-precision ramp power prediction model. Summary of the Invention

[0004] To solve the above technical problems, the present disclosure provides a method, device, equipment and medium for training a ramp power prediction model.

[0005] In a first aspect, the present disclosure provides a method for a ramp power prediction model, including:

[0006] Obtain source domain training data and target domain training data of a reference wind farm, where the source domain training data is training data including various weather events, and the target domain training data is training data including ramp events;

[0007] Pre-train an original prediction model based on the source domain training data to obtain a pre-trained prediction model;

[0008] Fine-tune the pre-trained prediction model according to the target domain training data to obtain the ramp power prediction model of the reference wind farm.

[0009] In a second aspect, the present disclosure provides a ramp power prediction model device, including:

[0010] A data acquisition module, configured to obtain source domain training data and target domain training data of a reference wind farm, where the source domain training data is training data including various weather events, and the target domain training data is training data including ramp events;

[0011] A pre-training module, configured to pre-train an original prediction model based on the source domain training data to obtain a pre-trained prediction model;

[0012] A model fine-tuning module, configured to fine-tune the pre-trained prediction model according to the target domain training data to obtain the ramp power prediction model of the reference wind farm.

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

[0014] One or more processors;

[0015] A storage device, configured to store one or more programs,

[0016] When the one or more programs are executed by the 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, on which a computer program is stored, and when the computer program is executed by a processor, the method provided in the first aspect is implemented.

[0018] The technical solutions provided by the embodiments of the present disclosure have the following advantages compared with the prior art:

[0019] A method, device, equipment and medium for training a ramp power prediction model according to an embodiment of the present disclosure. Source domain training data and target domain training data of a reference wind farm are obtained, wherein the source domain training data is training data including various weather events, and the target domain training data is training data including ramp events; the original prediction model is pre-trained based on the source domain training data to obtain a pre-trained prediction model; according to the target domain training data, the pre-trained prediction model is fine-tuned to obtain the ramp power prediction model of the reference wind farm. Since there are feature commonalities and differences between the training samples under ramp events and the training samples under normal weather, this method uses the means of knowledge transfer, first pre-trains with sufficient training data, and then fine-tunes the parameters of the model with limited ramp events, which can not only ensure the reliability of the ramp power prediction model, but also effectively adapt to the characteristics of ramp events, and finally improves the power prediction accuracy of small-sample ramp events. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The drawings herein are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present disclosure and used together with the specification to explain the principles of the present disclosure.

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0022] Figure 1 Schematic flowchart of a method for training a ramp power prediction model provided by an embodiment of the present disclosure;

[0023] Figure 2 Schematic flowchart of another method for training a ramp power prediction model provided by an embodiment of the present disclosure;

[0024] Figure 3 Schematic structural diagram of a device for training a ramp power prediction model provided by an embodiment of the present disclosure;

[0025] Figure 4 Schematic structural diagram of an electronic device provided by an embodiment of the present disclosure. Detailed implementation manners

[0026] In order to be able to more clearly understand the above objects, features, and advantages of the present disclosure, the solutions of the present disclosure will be further described below. It should be noted that, without conflict, the embodiments of the present disclosure and the features in the embodiments may be combined with each other.

[0027] Many specific details are set forth in the following description in order to fully understand the present disclosure, but the present disclosure may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only a part of the embodiments of the present disclosure, rather than all of the embodiments.

[0028] In order to ensure the reliability and applicability of the ramp power prediction model, an embodiment of the present disclosure provides a method, device, equipment, and medium for training a ramp power prediction model.

[0029] The following combines Figure 1 to illustrate the method for training a ramp power prediction model provided by an embodiment of the present disclosure. In the embodiment of the present disclosure, the method for training a ramp power prediction model may be executed by an electronic device or a server. Among them, the electronic device may include devices with communication functions such as tablet computers, desktop computers, and laptop computers, and may also include devices simulated by virtual machines or simulators. The server may include a server cluster and a cloud server.

[0030] Figure 1 Shows a schematic flowchart of a method for training a ramp power prediction model provided by an embodiment of the present disclosure.

[0031] As Figure 1 shown, the method for training a ramp power prediction model may include the following steps.

[0032] S110. Obtain the source domain training data and target domain training data of the reference wind farm. Among them, the source domain training data is the training data containing various weather events, and the target domain training data is the training data containing ramp events.

[0033] Among them, the reference wind farm is a wind farm used to collect training data to train the ramp power prediction model. Specifically, the reference wind farm, i.e., the wind power plant, can convert wind energy into mechanical energy and then convert the mechanical energy into electrical energy. Optionally, the composition of the reference wind farm includes but is not limited to anemometers, wind turbines, integrated circuits, booster stations, etc. The wind turbines receive electrical energy and finally convert it into electrical energy and output it through the output terminal of the booster station. The wind power of the wind farm is the total power output from the output terminal of the booster station.

[0034] Among them, the source domain training data refers to the training data with rich supervision information, specifically including the training data of various weather events in the reference wind farm.

[0035] Optionally, the source domain training data includes the weather event type and the first measured wind power. The weather event type includes two dimensions: time and features. The time dimension includes historical data of N moments, where the time interval is 15 minutes. The feature dimension includes M types of feature parameters. For ramp power prediction, mainly historical wind speed, historical wind direction, etc. In this way, the N data points of each feature parameter are arranged in chronological order to obtain a sequence, and then the M types of feature parameters are aligned according to the same time points to obtain a two-dimensional array. The first measured wind power is the true wind power of the reference wind farm at historical moments.

[0036] Among them, the target domain training data refers to the training data with few labels or no labels, specifically including the training data of ramp events in the reference wind farm.

[0037] Optionally, the target domain training data includes ramp events and the second measured wind power. The ramp events can also be in the form of a two-dimensional array. The second measured wind power is the true wind power of the reference wind farm at historical moments.

[0038] It can be understood that since the reference wind farm can only provide a small amount of target domain training data, it is difficult to train a ramp power prediction model for high-precision prediction based only on a small amount of target domain training data. Therefore, it is necessary to combine the source domain training data containing sufficient weather events for model training, so as to ensure the reliability of the ramp prediction model and effectively adapt to the characteristics of ramp events.

[0039] S120. Pre-train the original prediction model based on the source domain training data to obtain a pre-trained prediction model.

[0040] In this embodiment, first, methods such as zero initialization, random initialization, Xavier initialization, and He initialization are used to initialize the network parameters of the original prediction model. Then, the parameters of all layers of the original prediction model are updated E times using the weather event types and the first measured wind power in the source domain training data, so as to pre-train the original prediction model using the source domain training data containing sufficient weather events.

[0041] Among them, it is set that 1 ≤ E ≤ 1000, and the specific value of E is determined by the "early stopping method". The specific process of the "early stopping method" is as follows: the model parameters are updated using the training set sample data, and the accuracy is verified using the validation set sample data. When the validation set accuracy drops continuously for p times, the iteration process is terminated, and the wind power prediction model is no longer updated.

[0042] Optionally, the original prediction model includes, but is not limited to, complex neural networks with multi-layer structures such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU).

[0043] S130. According to the target domain training data, fine-tune the pre-trained prediction model to obtain the ramp power prediction model of the reference wind farm.

[0044] In order to enable the ramp power prediction model to effectively adapt to the ramp characteristics, after pre-training the original prediction model using the source domain training data, the ramp events and the second measured wind power in the target domain training data are continued to be used to update the network parameters of all layers or specific layers in the pre-trained prediction model G times, so as to obtain the ramp power prediction model of the reference wind farm.

[0045] Among them, it is set that 1 ≤ G ≤ 1000, and the specific value of G is determined by the "early stopping method". The specific process of the "early stopping method" can refer to the above description and will not be elaborated here. Optionally, the specific layer of the pre-trained prediction model can be selected according to the specific ramp event type. Generally, the output layer of the pre-trained prediction model is often selected.

[0046] In order to improve the fine-tuning efficiency of the model, first divide the ramp events in the target training data into multiple ramp types, and based on the multiple ramp types and the second measured wind power, fine-tune the network parameters of all layers or specific layers in the pre-trained prediction model to obtain the ramp power prediction model of the reference wind farm. Optionally, the ramp types include, but are not limited to, sharp up-ramp, up-ramp, gentle fluctuation ramp, down-ramp, and sharp down-ramp.

[0047] An embodiment of the present disclosure provides a method for training a ramp power prediction model, which obtains source domain training data and target domain training data of a reference wind farm. The source domain training data is training data including various weather events, and the target domain training data is training data including ramp events. Based on the source domain training data, the original prediction model is pre-trained to obtain a pre-trained prediction model. According to the target domain training data, the pre-trained prediction model is fine-tuned to obtain a ramp power prediction model of the reference wind farm. Since there are commonalities and differences in features between the training samples under ramp events and those under normal weather, this method uses the means of knowledge transfer. First, it performs pre-training with sufficient training data, and then fine-tunes the parameters of the model with limited ramp events, which can not only ensure the reliability of the ramp power prediction model but also effectively adapt to the characteristics of ramp events, ultimately improving the power prediction accuracy of small-sample ramp events.

[0048] Further, after determining the ramp power prediction model, the ramp power prediction model can be used to predict the ramp power of any target wind farm. Correspondingly, after executing S130, the method further includes: obtaining predicted ramp events that occur in the target wind farm within a future time period; using the ramp power prediction model to process the predicted ramp events to obtain the predicted power generation of the target wind farm when the predicted ramp events occur within the future time period.

[0049] The target wind farm is any power plant with similar power generation capabilities to the reference wind farm. Therefore, the ramp power prediction model of the reference wind farm can be migrated to the target wind farm for ramp power prediction.

[0050] The predicted ramp events that occur within the future time period can be any ramp type. It can be understood that the predicted ramp events that occur within the future time period will cause the active power imbalance of the power system within the future time period. Therefore, it is necessary to predict the predicted power generation of the target wind farm within the future time period based on the predicted ramp events, which is convenient for scheduling and operating the power system according to the predicted power generation and providing trading reference data for the power market.

[0051] In another embodiment of the present disclosure, the pre-training process and fine-tuning process of the ramp power prediction model are specifically explained.

[0052] Figure 2 The flowchart of another method for training a ramp power prediction model provided by an embodiment of the present disclosure is shown.

[0053] As Figure 2 shown, the method for training the ramp power prediction model may include the following steps.

[0054] S210. Obtain the source domain training data and target domain training data of the reference wind farm. Among them, the source domain training data is the training data containing various weather events, and the target domain training data is the training data containing ramp events.

[0055] Among them, S210 is similar to S110 and will not be elaborated here.

[0056] S220. Select multiple groups of first training data from the source domain training data. Among them, each group of first training data includes the weather event type and the first measured wind power.

[0057] Among them, the number of groups of the first training data is greater than or equal to 1, and can be specifically set manually.

[0058] S230. Use the original prediction model to predict the wind power for the weather event type to obtain the first predicted wind power.

[0059] Optionally, the original prediction model includes an information forgetting layer, an information selection layer, and an output layer; correspondingly, the specific implementation manner of S230 includes but is not limited to the following manner: S2301. Based on the information forgetting layer, process the weather event type at any current moment and the output information of the original prediction model at the previous moment to obtain the forgetting information of the original prediction model at the previous moment; S2302. Based on the information selection layer, process the weather event type at the current moment and the output information of the original prediction model at the previous moment to determine the selection information and candidate information of the original prediction model at the current moment; S2303. Based on the output layer, process the memory information of the original prediction model at the previous moment, the forgetting information of the original prediction model at the previous moment, the output information of the original prediction model at the previous moment, the selection information of the original prediction model at the current moment, and the candidate information of the original prediction model at the current moment to obtain the first predicted wind power.

[0060] Specifically, when the original prediction model is an LSTM, the information forgetting layer is the forgetting gate, the information selection layer is the input gate, and the output layer is the output gate.

[0061] For S2301, the forgetting information of the original prediction model at the previous moment can be determined by the following method:

[0062] f t =σ(W f ·(x t ,h t-1 )+b f )

[0063] Among them, f t is the forgetting information of the information forgetting layer at the current moment t for the previous moment t - 1, W f is the weight matrix of the information forgetting layer, xt is the weather event type at the current moment t, h t-1 is the output information of the original prediction model at the previous moment, b f is the bias matrix of the information forgetting layer, and σ(*) is the sigmoid activation function that can map σ(*) to the range [0, 1].

[0064] For S2302, the selection information and candidate information of the original prediction model at the current moment can be determined in the following way:

[0065] i t = σ(W i ·(x t , h t-1 ) + b i )

[0066]

[0067] where, i t is the selection information of the original prediction model at the current moment, W i is the weight matrix of the information selection layer, b i is the bias matrix of the information selection layer, is the candidate information of the original prediction model at the current moment, specifically the candidate cell information, W C is the weight matrix of the information selection layer, b C is the bias matrix of the information selection layer.

[0068] Among them, S2303 first includes the following steps: multiplying the memory information of the original prediction model at the previous moment by the forgetting information of the original prediction model at the previous moment to obtain the first product; multiplying the selection information of the original prediction model at the current moment by the candidate information of the original prediction model at the current moment to obtain the second product; adding the first product and the second product to obtain the new candidate information of the original prediction model at the current moment.

[0069] Optionally, the new candidate information of the original prediction model at the current moment can be determined in the following way:

[0070]

[0071] Furthermore, S2303 continues to include the following steps: determining the output layer information based on the weather event type at the current moment and the output information of the original prediction model at the previous moment; determining the output information of the original prediction model at the current moment based on the output layer information and the new candidate information, and using the output information of the original prediction model at the current moment as the first predicted wind power.

[0072] Optionally, the output layer information can be determined in the following way:

[0073] O t = σ(W O ·(X t , h t-1 ) + b O )

[0074] Among them, O t is the output layer information, W O is the weight matrix of the output layer, and b O is the bias matrix of the output layer.

[0075] Optionally, the first predicted wind power can be determined in the following manner:

[0076] h t = O t * tanh(C t )

[0077] Among them, tanh is the tanh activation function that maps the input to the range [-1, 1].

[0078] Thus, by using a prediction model including an information forgetting layer, an information selection layer, and an output layer, feature processing is sequentially performed on the weather event types to achieve sufficient pre-training between the original prediction models, ensuring the reliability of the pre-training process.

[0079] S240. Calculate the first prediction error of the original prediction model based on the first measured wind power and the first predicted wind power in multiple groups of first training data.

[0080] In this embodiment, error calculation methods such as the mean square error calculation method and the mean absolute error calculation method can be used to calculate the error between the first measured wind power and the first predicted wind power in multiple groups of first training data to obtain the first prediction error, that is, the first prediction error can be one of the mean square error and the mean absolute error.

[0081] S250. Iteratively update the parameters of the original prediction model according to the first prediction error to obtain a pre-trained prediction model.

[0082] Specifically, according to the first prediction error, the parameters of all layers of the original prediction model can be updated to adjust the parameters of the original prediction model to the optimal parameters, obtaining a pre-trained prediction model. Optionally, the parameter update algorithm of the original prediction model can adopt but is not limited to the SGD, AdaGrad, and Adam algorithms.

[0083] S260. Classify the ramp events in the target domain training data to obtain multiple ramp types.

[0084] In this embodiment, the specific implementation manner of S260 includes but is not limited to the following: obtaining the measured wind power at the first historical moment from the target domain training data; obtaining the measured wind power at the prediction moment after the first historical moment from the target domain training data; and determining multiple types of ramp events according to the measured wind power at the first historical moment, the measured wind power at the prediction moment, and the rated wind power of the wind farm.

[0085] Optionally, the specific determination manner of multiple types of ramp events includes but is not limited to the following:

[0086]

[0087] Wherein, ΔP is the type of ramp event, which may specifically include a sharp upward ramp, an upward ramp, a gentle fluctuation, a downward ramp, and a sharp downward ramp. A sharp upward ramp means that ΔP is greater than or equal to 50%; an upward ramp means that ΔP is greater than or equal to 20% and less than 50%; a gentle fluctuation means that ΔP is greater than -20% and less than 20%; a downward ramp means that ΔP is greater than -50% and less than or equal to -20%; a sharp downward ramp means that ΔP is less than or equal to -50%; is the measured wind power at the prediction moment after the first historical moment, is the measured wind power at the first historical moment, P rated is the rated wind power of the wind farm.

[0088] Thus, before fine-tuning the pre-trained prediction model, classifying the ramp events is convenient for subsequent accurately fine-tuning the pre-trained prediction model based on multiple types of ramp events to obtain a ramp power prediction model applicable to the characteristics of ramp events.

[0089] S270. Fine-tuning the pre-trained prediction model based on multiple types of ramp events and the second measured wind power in the target domain training data to obtain a ramp power prediction model for the reference wind farm.

[0090] In this embodiment, the specific implementation manner of S270 includes but is not limited to the following: selecting multiple groups of ramp event types and multiple groups of second measured wind power from the target domain training data; using the pre-trained prediction model to predict the wind power for multiple groups of ramp event types to obtain multiple groups of second predicted wind power; calculating the second prediction error of the pre-trained prediction model based on multiple groups of second measured wind power and multiple groups of second predicted wind power; and iteratively updating the parameters of the pre-trained prediction model according to the second prediction error to obtain a ramp power prediction model for the reference wind farm.

[0091] Wherein, the number of groups of ramp event types and the second measured wind power is greater than or equal to 1, and can be specifically set manually.

[0092] Among them, each group of second predicted wind power can be specifically obtained by processing through the information forgetting layer, information selection layer, and output layer of the pre-trained prediction model in sequence. The specific process is similar to the above description and will not be elaborated here.

[0093] Among them, the second prediction error of the pre-trained prediction model can be specifically calculated by means such as mean square error calculation method, mean absolute error calculation method, etc., that is, the second prediction error can be one of mean square error and mean absolute error.

[0094] Furthermore, after determining the second prediction error, the parameters of specific layers of the pre-trained prediction model can be updated according to the second prediction error, so that the parameters of the pre-trained prediction model are adjusted to optimal parameters, and a ramp power prediction model is obtained. Optionally, the parameter update algorithm of the pre-trained prediction model can adopt but is not limited to SGD, AdaGrad, and Adam algorithms.

[0095] Thus, it is possible to fully adjust the parameters of the pre-trained prediction model by using multiple groups of ramp types and multiple groups of second measured wind power, so as to obtain a ramp power prediction model that can perform high-precision power generation prediction when a ramp event occurs.

[0096] The embodiment of the present disclosure also provides a ramp power prediction model training device for implementing the above ramp power prediction model training method. The following will be described in conjunction with Figure 3 For illustration. In the embodiment of the present disclosure, the ramp power prediction model training device can be an electronic device or a server. Among them, the electronic device can include devices with communication functions such as tablet computers, desktop computers, and laptop computers, and can also include devices simulated by virtual machines or simulators. The server can include a server cluster and a cloud server.

[0097] Figure 3 FIG. shows a schematic structural diagram of a ramp power prediction model training device provided by an embodiment of the present disclosure.

[0098] As Figure 3 shown, the ramp power prediction model training device 300 can include:

[0099] A data acquisition module 310, configured to acquire source domain training data and target domain training data of a reference wind farm, where the source domain training data is training data including multiple weather events, and the target domain training data is training data including ramp events;

[0100] A pre-training module 320, configured to pre-train an original prediction model based on the source domain training data to obtain a pre-trained prediction model;

[0101] The model fine-tuning module 330 is used to fine-tune the pre-trained prediction model according to the target domain training data to obtain the ramp power prediction model of the reference wind farm.

[0102] A ramp power prediction model training device according to an embodiment of the present disclosure obtains source domain training data and target domain training data of a reference wind farm, where the source domain training data is training data including various weather events, and the target domain training data is training data including ramp events; pre-trains an original prediction model based on the source domain training data to obtain a pre-trained prediction model; and fine-tunes the pre-trained prediction model according to the target domain training data to obtain the ramp power prediction model of the reference wind farm. Since there are commonalities and differences in features between the training samples under ramp events and the training samples under normal weather, this method uses the means of knowledge transfer, first pre-trains with sufficient training data, and then fine-tunes the parameters of the model with limited ramp events, which can not only ensure the reliability of the ramp power prediction model, but also effectively adapt to the characteristics of ramp events, and finally improves the power prediction accuracy of small-sample ramp events.

[0103] In some embodiments of the present disclosure, the pre-training module 320 includes:

[0104] A selection unit for selecting multiple groups of first training data from the source domain training data, where each group of the first training data includes a weather event type and a first measured wind power;

[0105] A prediction unit for using the original prediction model to predict the wind power of the weather event type to obtain a first predicted wind power;

[0106] An error calculation unit for calculating a first prediction error of the original prediction model based on the first measured wind power and the first predicted wind power in multiple groups of the first training data;

[0107] An iterative update unit for iteratively updating the parameters of the original prediction model according to the first prediction error to obtain the pre-trained prediction model.

[0108] In some embodiments of the present disclosure, the original prediction model includes an information forgetting layer, an information selection layer, and an output layer; correspondingly, the first prediction unit is specifically configured to: based on the information forgetting layer, process the weather event type at any current moment and the output information of the original prediction model at the previous moment to obtain the forgetting information of the original prediction model at the previous moment;

[0109] Based on the information selection layer, process the weather event type at the current moment and the output information of the original prediction model at the previous moment to determine the selection information and candidate information of the original prediction model at the current moment;

[0110] Based on the output layer, process the memory information of the original prediction model at the previous moment, the forgetting information of the original prediction model at the previous moment, the output information of the original prediction model at the previous moment, the selection information of the original prediction model at the current moment, the candidate information of the original prediction model at the current moment, and the weather event type at the current moment to obtain the first predicted wind power.

[0111] In some embodiments of the present disclosure, the model fine-tuning module 330 includes:

[0112] A classification unit for classifying the ramp events in the target domain training data to obtain multiple ramp types;

[0113] A fine-tuning unit for fine-tuning the pre-trained prediction model based on the multiple ramp types in the target domain training data and the second measured wind power to obtain the ramp power prediction model of the reference wind farm.

[0114] In some embodiments of the present disclosure, the classification unit is specifically configured to: obtain the measured wind power at the first historical moment from the target domain training data;

[0115] Obtain the measured wind power at the prediction moment after the first historical moment from the target domain training data;

[0116] Determine the multiple ramp types according to the measured wind power at the first historical moment, the measured wind power at the prediction moment, and the rated wind power of the wind farm.

[0117] In some embodiments of the present disclosure, the fine-tuning unit is specifically configured to: select multiple groups of ramp types and multiple groups of second measured wind power from the target domain training data;

[0118] Use the pre-trained prediction model to predict the wind power for the multiple groups of ramp types to obtain multiple groups of second predicted wind power;

[0119] Based on the multiple groups of second measured wind power and the multiple groups of second predicted wind power, calculate the second prediction error of the pre-trained prediction model;

[0120] Iteratively update the parameters of the pre-trained prediction model according to the second prediction error to obtain the ramp power prediction model of the reference wind farm.

[0121] In some embodiments of the present disclosure, the device further includes:

[0122] A predicted ramping event acquisition module, configured to acquire predicted ramping events occurring in a target wind farm within a future time period;

[0123] A ramping power prediction module, configured to process the predicted ramping events by using the ramping power prediction model to obtain predicted power generation of the target wind farm when the predicted ramping events occur within the future time period.

[0124] It should be noted that, Figure 3 the ramping power prediction model training device 300 shown can execute Figures 1 - 2 each step in the method embodiments shown, and implement Figures 1 - 2 each process and effect in the method embodiments shown, which will not be elaborated herein.

[0125] Figure 4 FIG. shows a schematic structural diagram of an electronic device provided by an embodiment of the present disclosure.

[0126] As Figure 4 shown, the electronic device may include a processor 401 and a memory 402 storing computer program instructions.

[0127] Specifically, the processor 401 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0128] The memory 402 may include a mass storage 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 magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 402 may include removable or non-removable (or fixed) media. Where appropriate, the memory 402 may be internal or external to 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 (EEPROM), an electrically alterable ROM (EAROM), or a flash memory, or a combination of two or more of these.

[0129] The processor 401 reads and executes the computer program instructions stored in the memory 402 to perform the steps of the ramp power prediction model training method or the wind power prediction method provided by the embodiments of the present disclosure.

[0130] In one example, the electronic device may further include a transceiver 403 and a bus 404. Among them, as Figure 4 shown, the processor 401, the memory 402, and the transceiver 403 are connected through the bus 404 and complete communication with each other.

[0131] The bus 404 includes hardware, software, or both. By way of example and 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 bus or a combination of two or more of these. Where appropriate, the bus 404 may include one or more buses. Although the embodiments of the present application describe and illustrate specific buses, the present application contemplates any suitable bus or interconnect.

[0132] The following are embodiments of a computer-readable storage medium provided by the embodiments of the present disclosure. The computer-readable storage medium and the ramp power prediction model training method of the above embodiments belong to the same inventive concept. Details not described in detail in the embodiments of the computer-readable storage medium may refer to the embodiments of the ramp power prediction model training method.

[0133] This embodiment provides a storage medium containing computer-executable instructions. The computer-executable instructions, when executed by a computer processor, are used to execute a ramp power prediction model training method. The method is applied to an electronic device and includes:

[0134] Obtain source domain training data and target domain training data of a reference wind farm, where the source domain training data is training data containing multiple weather events, and the target domain training data is training data containing ramp events;

[0135] Pre-train an original prediction model based on the source domain training data to obtain a pre-trained prediction model;

[0136] Fine-tune the pre-trained prediction model according to the target domain training data to obtain the ramp power prediction model of the reference wind farm.

[0137] Certainly, the computer-executable instructions included in the storage medium provided by the embodiments of the present disclosure are not limited to the above method operations, and can also execute the related operations of the ramp power prediction model training method provided by any embodiment of the present disclosure.

[0138] From the above description of the embodiments, those skilled in the art can clearly understand that the present disclosure can be implemented by means of software and necessary general-purpose hardware. Of course, it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present disclosure, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as a floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disc of a computer, etc., including several instructions for causing a computer cloud platform (which can be a personal computer, a server, or a network cloud platform, etc.) to execute the ramp power prediction model training method provided by each embodiment of the present disclosure.

[0139] Note that the above is only the preferred embodiment of the present disclosure and the technical principles applied. Those skilled in the art will understand that the present disclosure is not limited to the specific embodiments here, and various obvious changes, re-adjustments and substitutions can be made by those skilled in the art without departing from the protection scope of the present disclosure. Therefore, although the present disclosure has been described in detail through the above embodiments, the present disclosure is not limited to the above embodiments. Without departing from the concept of the present disclosure, more other equivalent embodiments can be included, and the scope of the present disclosure is determined by the scope of the appended claims.

Claims

1. A climbing power prediction model training method, It is characterized in that include: Acquire source domain training data and target domain training data of a reference wind farm, wherein the source domain training data is training data containing multiple weather events, and the target domain training data is training data containing ramp events; Pre-training the original prediction model based on the source domain training data to obtain a pre-trained prediction model; The pre-trained prediction model is fine-tuned according to the target domain training data to obtain the ramp power prediction model of the reference wind farm.

2. The method according to claim 1, It is characterized in that The pre-training of the original prediction model based on the source domain training data to obtain a pre-trained prediction model includes: Selecting multiple groups of first training data from the source domain training data, wherein each group of the first training data includes a weather event type and a first measured wind power; Using the original prediction model, predicting wind power for the weather event type to obtain a first predicted wind power; Calculating a first prediction error of the original prediction model based on first measured wind power and first predicted wind power in multiple groups of the first training data; The parameters of the original prediction model are iteratively updated according to the first prediction error to obtain the pre-trained prediction model.

3. The method according to claim 2, It is characterized in that The original prediction model includes an information forgetting layer, an information selection layer and an output layer; accordingly, using the original prediction model to predict wind power for the weather event type to obtain a first predicted wind power includes: Based on the information forgetting layer, any weather event type at the current moment and the output information of the original prediction model at the previous moment are processed to obtain the forgotten information of the original prediction model at the previous moment; Based on the information selection layer, the weather event type at the current moment and the output information of the original prediction model at the previous moment are processed to determine the selection information and candidate information of the original prediction model at the current moment; Based on the output layer, the memory information of the original prediction model at the previous moment, the forgotten information of the original prediction model at the previous moment, the output information of the original prediction model at the previous moment, the selection information of the original prediction model at the current moment, the candidate information of the original prediction model at the current moment, and the weather event type at the current moment are processed to obtain the first predicted wind power.

4. The method according to claim 1, It is characterized in that The step of fine-tuning the pre-trained prediction model according to the target domain training data to obtain the ramp power prediction model of the reference wind farm includes: Classifying the hill climbing events in the target domain training data to obtain multiple hill climbing types; Based on the multiple climbing types and the second measured wind power in the target domain training data, the pre-trained prediction model is fine-tuned to obtain the climbing power prediction model of the reference wind farm.

5. The method according to claim 4, It is characterized in that The hill climbing events in the target domain training data are classified to obtain multiple hill climbing types, including: Acquire the measured wind power at a first historical moment from the target domain training data; Acquire the measured wind power at the time to be predicted after the first historical moment from the target domain training data; The multiple ramp types are determined according to the actually measured wind power at the first historical moment, the actually measured wind power at the moment to be predicted, and the rated wind power of the wind farm.

6. The method according to claim 4, It is characterized in that The method of fine-tuning the pre-trained prediction model based on the multiple climbing types in the target domain training data and the second measured wind power to obtain the climbing power prediction model of the reference wind farm includes: Selecting multiple groups of climbing types and multiple groups of second measured wind power from the target domain training data; Using the pre-trained prediction model, predicting wind power for the multiple groups of climbing types, to obtain multiple groups of second predicted wind power; Calculating a second prediction error of the pre-trained prediction model based on the multiple groups of second measured wind power and the multiple groups of second predicted wind power; The parameters of the pre-trained prediction model are iteratively updated according to the second prediction error to obtain the ramp power prediction model of the reference wind farm.

7. The method according to claim 1, It is characterized in that Also includes: Obtain predicted ramp events that will occur in the target wind farm in a future time period; The predicted ramp event is processed using the ramp power prediction model to obtain the predicted power generation of the target wind farm when the predicted ramp event occurs in the future time period.

8. A climbing power prediction model device, It is characterized in that include: A data acquisition module, used to acquire source domain training data and target domain training data of a reference wind farm, wherein the source domain training data is training data containing multiple weather events, and the target domain training data is training data containing ramp events; A pre-training module, used to pre-train the original prediction model based on the source domain training data to obtain a pre-trained prediction model; The model fine-tuning module is used to fine-tune the pre-trained prediction model according to the target domain training data to obtain the ramp power prediction model of the reference wind farm.

9. An electronic device, It is 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, It is 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