An AI-based method and system for predicting wind power output.
By using an AI-based wind power prediction system and multimodal fusion technology to process wind speed, wind direction and historical data, the shortcomings of wind power prediction models in commercial applications have been addressed. This has enabled accurate wind power prediction, reduced wind curtailment, and improved the commercial application effect of wind power.
Patent Information
- Application Number
- CN202211486305.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2021-11-27
- Filing Date
- 2022-11-24
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2042-11-24
AI Technical Summary
Existing wind power prediction models lack mature solutions for commercial applications, leading to severe wind curtailment and failing to effectively improve prediction accuracy and stability.
A wind power prediction system based on AI data mining is adopted. The system processes wind speed and wind direction prediction values and historical data through the first and second branches respectively, extracts features using an attention mechanism model, and obtains wind power prediction values through multimodal fusion.
It enables accurate prediction of wind power output, reduces wind curtailment, enhances the competitiveness of wind power in the electricity market, and improves the stability of grid dispatch and resource utilization.
Smart Images

Figure CN115713159B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of wind power generation technology, in particular to a wind turbine wind power prediction method and system based on AI data mining. BACKGROUND
[0002] Wind energy is a renewable energy source with great potential for development. In recent years, with the popularization of wind power equipment manufacturing technology and the continuous expansion of wind turbine manufacturers, the price of wind turbines has gradually decreased. Wind power generation is currently one of the most mature and most commercially developed renewable energy generation technologies. Many countries have formulated power development plans to make wind power generation the main energy source for future development.
[0003] With the rapid growth of wind power installed capacity in China, the threat of wind power to the power grid is becoming more prominent. The rapid development, centralized access and long-distance transmission of large-scale wind power have put higher requirements on power grid dispatching. Because wind energy has intermittency and volatility, it affects efficient and safe grid connection, so there is currently a lot of wind curtailment behavior in China, which undoubtedly causes a lot of resources to be idle and wasted. Accurate prediction of wind turbine power not only solves the problem of wind power consumption, but also enhances the competitiveness of wind power in the power market and improves the on-grid price. Although researchers have proposed many wind power prediction models and have been successful in laboratory conditions and field tests, so far, in the wind power prediction scenario, there are still very few prediction systems that can achieve commercialization and industrial application, and there is no mature solution.
[0004] Therefore, further research on wind power prediction is still needed to improve the accuracy and stability of the model. SUMMARY
[0005] Based on the above problems, the present application provides a wind power prediction method and system based on AI (artificial intelligence) data mining. Based on the prediction system, reliable wind power prediction can be achieved. Based on the system, the method can effectively improve the prediction accuracy, reduce wind curtailment, and accurately predict future wind turbine power.
[0006] To achieve the above-mentioned purposes, the present application adopts the following technical solutions:
[0007] A wind power prediction system based on AI, characterized in that it comprises:
[0008] a first branch and a second branch,
[0009] The first branch obtains a prediction value feature based on a wind direction prediction value and a wind speed prediction value and based on a preset first prediction model, and obtains a prediction value attention feature through an attention mechanism model, and then fuses the prediction value attention feature and a historical data value feature to obtain a historical data value fusion feature;
[0010] The second branch obtains a historical data value feature based on historical data and based on a preset first prediction model, and obtains a historical data value attention feature through an attention mechanism model, and then fuses the historical data value attention feature and a prediction value feature to obtain a prediction value fusion feature:
[0011] The fifth data processing module performs splicing based on the historical data value fusion feature, the prediction value fusion feature and global feature channel dimension information to obtain a prediction value of the wind power.
[0012] Preferably, the first branch comprises:
[0013] a first data acquisition module for obtaining a wind direction prediction value and a wind speed prediction value,
[0014] a first data processing module that receives information fed back by the first data acquisition module and processes the information respectively to obtain respective features, fuses the respective features in terms of dimension information, and outputs the fused information to a second data processing module, and the second data processing module receives and responds to the fused information and processes the information to obtain a prediction value feature.
[0015] Preferably, the second branch comprises:
[0016] a second data acquisition module for obtaining a plurality of historical data of the same period as the first data acquisition module and transmitting the historical data to a third data processing module
[0017] The third data processing module receives and responds to the plurality of historical data and processes the historical data respectively to obtain respective features, fuses the respective features in terms of dimension information, and outputs the fused information to a fourth data processing module, and the fourth data processing module receives and responds to the fused information and processes the information to obtain a historical data value feature.
[0018] A prediction method of the prediction system, characterized in that the method comprises:
[0019] S1, obtaining a prediction value feature and a historical data value feature based on a weather forecast prediction value and a historical data value respectively,
[0020] S2, obtaining a prediction value attention feature and a historical data value attention feature based on an attention mechanism model,
[0021] S3, obtaining a prediction value fusion feature by fusing the historical data value attention feature and the prediction value feature, and
[0022] By fusing the attention features of predicted values with the features of historical data values, we obtain the fused features of historical data values.
[0023] S4. Based on the fusion features of historical data values, the fusion features of predicted values, and the dimensional information of global feature channels, the predicted value of wind power is obtained by splicing them together.
[0024] Beneficial effects
[0025] Compared to existing solutions, the prediction method of the wind power prediction system based on AI data mining in this application can effectively extract and fuse information from multi-channel signals, model the correlation between multi-channel weather forecast values and historical data values, automatically extract high-level semantic features that are strongly related to the target task, and achieve accurate wind turbine power prediction. Attached Figure Description
[0026] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0027] Figure 1 , Figure 2 This is a functional diagram of an AI-based wind power prediction system according to an embodiment of this application.
[0028] Figure 3 This is a flowchart illustrating the AI-based wind power prediction method according to an embodiment of this application. Detailed Implementation
[0029] To enable those skilled in the art to better understand the technical solutions proposed in this invention, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments in this specification, and not all of them. All other embodiments obtained by those skilled in the art based on one or more embodiments in this specification without creative effort should fall within the scope of protection of this invention. It should also be noted that in the apparatus, equipment, and methods of this application, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered equivalent solutions of this application.
[0030] Example:
[0031] This application proposes an AI-based method and system for predicting wind power output of wind turbines. The prediction system has a first branch and a second branch. The first branch obtains predicted value features based on predicted wind direction and wind speed values and a preset first prediction model. The second branch obtains historical data value features based on historical data matching weather forecasts and a second preset prediction model. The predicted value features and historical data value features are then input into a multimodal fusion model to obtain the predicted wind power output. The prediction system connects to wind turbines and performs wind power output prediction on the connected turbines. The method includes: obtaining predicted value features and historical data value features based on weather forecasts and historical data values respectively; obtaining predicted value attention features and historical data value attention features based on an attention mechanism model; obtaining predicted value fusion features by fusing historical data value attention features with predicted value features; obtaining historical data value fusion features by fusing predicted value attention features with historical data value features; and concatenating the historical data value fusion features, predicted value fusion features, and global feature channel dimension information to obtain the predicted wind power output. This method can effectively improve prediction accuracy, reduce wind curtailment, and accurately predict future wind turbine power.
[0032] The following section will describe in detail the wind power prediction method and system based on AI data mining proposed in this application, with reference to the figures.
[0033] The AI-based data mining-based wind power prediction system includes a first branch and a second branch. The first branch obtains predicted value features based on predicted wind direction and wind speed values and a preset prediction model. The second branch obtains historical data value features based on historical wind direction data, historical wind speed data, and historical wind power data, etc., and a preset prediction model.
[0034] The predicted value features and historical data value features are input into multimodal fusion to obtain the predicted value of wind power.
[0035] like Figure 1 The diagram shown is a block diagram of an AI (data mining)-based wind power prediction system according to an embodiment of this application.
[0036] The system includes: a first branch and a second branch.
[0037] The first branch road includes:
[0038] The first data acquisition module is used to obtain wind direction and wind speed prediction values.
[0039] The first data processing module receives and processes the information fed back from the first data acquisition module to obtain their respective features, fuses them with dimensional information, and outputs them to the second data processing module.
[0040] The second data processing module receives and responds to the fused information, and processes it through a local convolutional long short-term memory network model to obtain the predicted value features of the entire weather forecast prediction value branch.
[0041] Simultaneously, through the attention mechanism model, the attention features of the predicted values are obtained, and then the attention features of the predicted values are fused with the historical data value features to obtain the historical data value fusion features.
[0042] The second branch (i.e., the historical data value branch) includes:
[0043] The second data acquisition module obtains various historical data from the same historical period matched by the first data acquisition module and transmits it to the third data processing module.
[0044] The third data processing module receives and responds to the various historical data, processes them separately (e.g., through convolutional network model 2 with the same structure), obtains their respective (historical information) features, fuses them using dimensional information, and outputs the fused information to the fourth data processing module.
[0045] The fourth data processing module receives and responds to the fused (historical feature) information, processes it through a local convolutional long short-term memory network model to obtain historical data value features, and simultaneously obtains historical data value attention features through an attention mechanism model. Finally, by fusing the historical data value attention features with the predicted value features, the predicted value fused features are obtained.
[0046] The fifth data processing module concatenates historical data value fusion features, predicted value fusion features, and global feature channel dimension information to obtain the predicted wind power value. Specifically, the fifth data processing module concatenates the predicted value fusion features, historical data value fusion features, and global features through channel dimension information to obtain features that can fully express wind power prediction. These features are then processed through a local convolutional long short-term memory network and a fully connected layer to obtain the predicted wind power value. The global features are obtained by fusing the predicted value fusion features and historical data value fusion features through an additive model. The global features can represent the combined characteristics of weather forecast predictions and historical data values. The system in this implementation can predict the wind power generated by connected wind turbines. It can perform long-term predictions (predicting the following year), medium-term predictions (predicting the next few weeks or months), short-term predictions (predicting the next 2 or 3 days), and ultra-short-term predictions (predicting the next few minutes, commonly 15 minutes). Users only need to construct the data according to their needs and time scale. This solution proposes a multimodal fusion wind power prediction system. Through its designed modules, it effectively extracts and fuses information from multiple channels, models the correlation between multi-channel weather forecast values and historical data, and automatically extracts high-level semantic features strongly correlated with the target task, achieving accurate wind turbine power prediction. The system's prediction mechanism follows the wind turbine power curve; the power of the wind turbine increases with increasing wind speed. Therefore, wind power prediction is not only directly related to the wind speed and direction in weather forecasts but also has a certain correlation with various historical data.
[0047] like Figure 2 The diagram shown is a block diagram of an AI (data mining)-based wind power prediction system according to an embodiment of this application.
[0048] The system includes two sets of input signals: weather forecast values and historical data values. The weather forecast values include wind speed forecast values and wind direction forecast values; the historical data values include, but are not limited to, historical wind speed data, historical wind direction data, and historical wind power data.
[0049] In the weather forecast prediction branch, the wind direction prediction and wind speed prediction are operated on by the same convolutional network 1 to obtain their respective features. After being fused by dimensional information, the prediction features of the entire weather forecast prediction branch can be obtained by passing through a local convolutional long short-term memory network. At the same time, the attention features of the predictions are obtained through the attention mechanism.
[0050] In the historical data value branch, multiple historical data points are processed through a convolutional network with the same structure to obtain their respective features. These features are then fused using dimensional information, and further processed through a local convolutional long short-term memory network to obtain the historical data value features. Simultaneously, an attention mechanism is used to obtain signal attention features. By fusing the historical data value attention features with the predicted value features, the predicted value fusion feature is obtained. Similarly, by fusing the predicted value attention features with the historical data value features, the historical data value fusion feature is obtained.
[0051] The predicted value fusion feature and the historical data value fusion feature are fused together using an additive model to obtain a global feature, which is a comprehensive feature that can represent both weather forecast predictions and historical data values. Then, the predicted value fusion feature, the historical data value fusion feature, and the global feature are concatenated using channel dimension information to obtain a feature that can fully express wind power prediction. This feature is then passed through a local convolutional long short-term memory network and a fully connected layer to obtain the predicted wind power value. In this embodiment, convolutional network 1 and convolutional network 2 have the same model structure.
[0052] Next, combine Figure 3 To describe the prediction method for the above system:
[0053] The method includes:
[0054] S1. Obtain the characteristics of the predicted value and the characteristics of the historical data value based on the weather forecast and historical data values respectively.
[0055] S2, Attention-based models (e.g., features are passed through the sigmoid function) By obtaining the attention feature containing only spatial information, and then multiplying this attention feature by the same feature or other related features, we can obtain the attention feature for the predicted value and the attention feature for the historical data value.
[0056] S3. By fusing the attention features of historical data values with the features of predicted values, we obtain the fused features of predicted values; and by fusing the attention features of predicted values with the features of historical data values, we obtain the fused features of historical data values (in this step, the predicted feature values and historical data feature values are processed simultaneously).
[0057] S4. Based on the fusion features of historical data values, the fusion features of predicted values, and the dimensional information of global feature channels, the predicted wind power value is obtained by concatenating these features. In the initial training phase, the input data is constructed according to requirements. Through a convolutional network and the methods described above, features with strong expressive power for wind power prediction are obtained. Finally, mean squared error loss is applied. After forward propagation to obtain the outputs of each branch, the gradient is calculated according to the aforementioned loss function, and feedback training is performed, iterating until the model converges. In the testing phase, the input data is constructed according to requirements, and the output is the predicted wind power value.
[0058] Step S1 specifically includes:
[0059] The first data acquisition module obtains the predicted wind direction and wind speed values from the weather forecast and transmits them to the first data processing module. The first data processing module receives and processes them separately to obtain the features of each (predicted wind direction and wind speed values), and performs dimensional information fusion. The fused information is then transmitted to the second data processing module, which receives and processes the data through a local convolutional long short-term memory network model to obtain the predicted value features based on the weather forecast.
[0060] The second data acquisition module obtains multiple historical data from the same historical period as the first data acquisition module and transmits them to the third data processing module. The third data processing module receives these multiple historical data and processes them separately (e.g., through operation of a convolutional network model with the same structure) to obtain their respective (historical information) features and perform dimensional information fusion. The fused information is then output to the fourth data processing module. The fourth data processing module receives the fused (historical feature) information and processes it through a local convolutional long short-term memory network model to obtain historical data value features.
[0061] Step S4 specifically includes:
[0062] The fifth data processing module splices together historical data value fusion features, predicted value fusion features and global feature channel dimension information to obtain the first predicted feature of wind power. This first predicted feature is then processed by a local convolutional long short-term memory network and a fully connected layer to obtain the predicted value of wind power.
[0063] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.
[0064] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.
Claims
1. An AI-based wind power prediction system, characterized in that, include: First and second branch roads The first branch obtains predicted value features based on wind direction and wind speed prediction values and a preset first prediction model. It then obtains predicted value attention features through an attention mechanism model, and finally merges the predicted value attention features with historical data value features to obtain historical data value fusion features. The second branch obtains historical data value features based on historical data and a preset first prediction model. It then uses an attention mechanism model to obtain historical data value attention features. Finally, by fusing these historical data value attention features with the predicted value features, it obtains the predicted value fusion features. The fifth data processing module uses historical data value fusion features, predicted value fusion features, and global features to splice together the features expressing wind power prediction through channel dimension information. Then, it obtains the predicted value of wind power through a local convolutional long short-term memory network and a fully connected layer. The global features are obtained by fusing the predicted value fusion features and historical data value fusion features through an additive mode.
2. The AI-based wind power prediction system as described in claim 1, characterized in that, The first branch includes: The first data acquisition module is used to obtain wind direction and wind speed prediction values. The first data processing module receives information from the first data acquisition module, processes it separately to obtain individual features, fuses these features using dimensional information, and outputs it to the second data processing module. The second data processing module receives and responds to the fused information, and processes it to obtain the predicted value features.
3. The AI-based wind power prediction system as described in claim 1, characterized in that, The second branch includes: The second data acquisition module obtains various historical data from the same historical period matched by the first data acquisition module and transmits it to the third data processing module. The third data processing module receives and responds to the various historical data, processes them separately to obtain their respective features, fuses them with dimensional information, and outputs the fused information to the fourth data processing module. The fourth data processing module receives and responds to the fused information, processes it to obtain the historical data value features.
4. A prediction method for a prediction system according to any one of claims 1-3, characterized in that, The method includes the following steps: S1. Based on the predicted wind direction, predicted wind speed, and historical data, obtain the predicted value features and historical data features respectively; S2. Based on the attention mechanism model, obtain the attention features of the predicted values and the attention features of the historical data values; S3. By fusing the attention features of historical data values with the features of predicted values, the predicted value fusion feature is obtained, and By fusing the attention features of predicted values with the features of historical data values, we obtain the fused features of historical data values. S4. Based on the historical data value fusion feature, the predicted value fusion feature and the global feature channel dimension information, the predicted value of wind power is obtained by splicing them together. The global feature is obtained by adding the predicted value fusion feature and the historical data value fusion feature.
Citation Information
Patent Citations
Wind power ultra-short-term power prediction method using self-attention mechanism and bilinear fusion
CN112990553A
