Wind power prediction method and system based on multi-branch depth separable convolution and mixer
By adopting a wind power prediction method based on multi-branch depthwise separable convolution and mixer, the problem of low data utilization efficiency of wind farms is solved, and efficient and accurate wind power prediction is achieved, thereby improving the operational efficiency and market competitiveness of wind farms.
Patent Information
- Application Number
- CN202411020102.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-29
- Publication Date
- 2024-11-12
AI Technical Summary
Existing technologies struggle to effectively utilize massive amounts of time-series data from wind farms, resulting in low production efficiency and poor corporate profits. Furthermore, traditional forecasting models are not ideal for handling multivariate time-series forecasts.
A wind power prediction method based on multi-branch depthwise separable convolution and mixer is adopted. By preprocessing the meteorological, equipment monitoring and basic data of wind farms, the trend term and periodic term are extracted by the depthwise separable convolution model, and the future wind power prediction is generated by combining fully connected layer and linear layer.
It significantly improved the accuracy of wind power time series forecasting, optimized wind power resource utilization, increased the grid connection rate and dispatch accuracy of wind farms, and enhanced dynamic adaptability and operational efficiency.
Smart Images

Figure CN118940897A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of wind power prediction, and in particular to a wind power prediction method and system based on multi-branch depth-separable convolution and mixer. Background Art
[0002] In recent years, with the rapid development of social economy and the continuous advancement of new energy technology, the power generation of new energy has achieved sustained growth, among which the business volume of wind farms in my country has also increased significantly. Every day, many wind farms are successfully connected to the grid, providing stable power supply for enterprises and individual users. In this process, wind farms have generated huge demand for accurate time series prediction data, which is essential for auxiliary decision-making and power dispatching. With the continuous increase in the power generation of wind farms, more stringent requirements are put forward for the accuracy and efficiency of time series prediction.
[0003] In the daily operation of wind farms, a large amount of time series data is generated, including but not limited to key parameters such as wind speed, blade speed, temperature, etc. These data contain rich information about the changes of various elements in business activities over time, which is of great significance for predicting the trend of future time series changes. However, with the rapid growth of data volume, the complexity and difficulty of data analysis tasks have also increased, making it difficult to effectively utilize these massive historical data. As a result, a large amount of data resources have been idle, which not only reduces the production efficiency of wind farms, but also has an adverse impact on the overall profitability of enterprises.
[0004] In view of the above problems, scientific modeling of time series data to achieve accurate prediction has become a key issue that needs to be solved urgently. Time series prediction technology can accurately predict future series values by deeply analyzing the potential laws in historical data, providing strong support for corporate decision-making and resource optimization.
[0005] Wind power time series prediction refers to predicting the future power generation of a wind farm based on the existing time series data of the wind farm, such as real-time wind speed, blade speed, temperature, weather forecast, etc. The main technical methods currently include the following two categories:
[0006] (1) Forecasting methods based on mathematical statistics usually assume that time series conform to a certain distribution or function form; for example, the Auto Regressive (AR) model assumes that there is a linear relationship between time series, and the Moving Average (MA) model uses recent trends to predict future values, which is suitable for non-periodic instant data. These two modeling methods have had a profound impact on subsequent time series forecasting technologies. For example, the Auto Regressive Moving Average (ARMA) model is another technology for analyzing random time series that absorbs the main advantages of the autoregressive model and the moving average model.
[0007] Although the prediction model of statistical time series can capture the linear relationship in time series very well, this method of modeling the predicted value as a linear function of the past observation value makes it difficult for the prediction method based on mathematical statistics to model the nonlinear pattern in practical problems, resulting in unsatisfactory application results on many practical data sets. In addition, although some prediction models based on mathematical statistics have made certain improvements, they still encounter difficulties in dealing with multivariate time series prediction problems, mainly because their ability to capture time series relationships is limited. They can only obtain information on a single sequence through a "single channel" method, which is not enough to reveal the intrinsic connection between multiple sequences.
[0008] (2) The prediction method based on traditional machine learning transforms the prediction problem into a parameter regression problem, which has a wide range of applications in real life. This type of method first extracts features from historical data and then inputs these features into a pre-designed machine learning model. Next, the model is trained by optimizing the loss function, and finally a model that can make predictions is obtained.
[0009] However, before training the model, manual feature engineering is required, which is not only labor-intensive but also difficult to generalize. Well-designed features often rely on expertise in a specific field and are difficult to extend to new tasks. When faced with large training data sets and high-dimensional time series, they are often stuck in local optimality because feature engineering design is too subjective and regression function design is simple, making it difficult to effectively model complex multivariate time series. Summary of the invention
[0010] The purpose of the present invention is to propose a wind power prediction method and system based on multi-branch deep separable convolution and mixer to improve the grid connection rate and scheduling accuracy of wind farms, and to have dynamic adaptability and reduce prediction delay.
[0011] According to a first aspect of an embodiment of the present disclosure, a wind power prediction method based on multi-branch depthwise separable convolution and mixer is provided, comprising the following steps:
[0012] Preprocess the wind farm meteorological data, equipment monitoring data, and wind turbine basic data, and sort the preprocessed data according to the timeline;
[0013] The sorted data is fed into a depth-wise separable convolutional model, and the trend term and period term are obtained through the average pooling decomposition layer;
[0014] The periodic term enters the depthwise separable convolutional layer to extract the association in the time dimension, and extracts the association in the channel dimension through point-by-point convolution, thereby mapping the vector of the periodic term into a hidden layer feature;
[0015] Through the fully connected layer, the hidden layer features are projected onto the periodic term output space to obtain the periodic term of the wind turbine power generation at each time point;
[0016] The trend item enters the linear layer, and the future trend item is directly generated according to the historical trend item to obtain the wind turbine power generation trend item at each time point;
[0017] The periodic term and the trend term of the wind turbine generator power generation at each time point are added together to obtain the wind turbine generator power generation at the corresponding time point.
[0018] The power generation cycle term and the trend term of the wind turbine generator at each time point are added together to obtain the power generation of the wind turbine generator at each time point.
[0019] According to a second aspect of an embodiment of the present disclosure, a wind power prediction system based on multi-branch depthwise separable convolution and mixer is provided, comprising:
[0020] The preprocessing module preprocesses the wind farm meteorological data, equipment monitoring data, and basic data of wind turbines, and sorts the preprocessed data according to the timeline;
[0021] The decomposition module feeds the sorted data into the depth-wise separable convolutional model and obtains the trend term and period term through the average pooling decomposition layer;
[0022] Hidden feature acquisition module, the periodic term enters the depth-separable convolution layer, extracts the association in the time dimension, extracts the association in the channel dimension through point-by-point convolution, and then maps the vector of the periodic term to the hidden layer feature;
[0023] The periodic term prediction module projects the hidden layer features onto the periodic term output space through the fully connected layer to obtain the periodic term of the wind turbine power generation at each time point;
[0024] A trend item prediction module, wherein the trend item enters the linear layer, directly generates future trend items based on historical trend items, and obtains the wind turbine power generation trend item at each time point;
[0025] The power generation prediction module adds the power generation cycle item and the trend item of the wind turbine generator at each time point to obtain the power generation of the wind turbine generator at the corresponding time point.
[0026] According to a third aspect of an embodiment of the present disclosure, an electronic device is provided, comprising a memory, a processor, and a computer program stored and running on the memory, wherein when the processor executes the program, a wind power prediction method based on multi-branch depthwise separable convolution and mixer is implemented.
[0027] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the wind power prediction method based on multi-branch depthwise separable convolution and mixer is implemented.
[0028] Compared with the prior art, the above technical solution adopted by the present invention has the following advantages: First, through deep learning, massive time series data is efficiently processed, which significantly improves the accuracy of wind power time series prediction, thereby improving the grid connection rate and scheduling accuracy of wind farms, which not only optimizes the utilization of wind power resources, but also directly improves the operating efficiency and revenue of enterprises. Secondly, the complexity of the prediction model is simplified through the deep separable convolution model, which solves the problem of low prediction efficiency caused by the excessive complexity of the traditional prediction model, enhances the dynamic adaptability of the time series method, and significantly reduces the prediction delay, so that wind farms can respond to market changes and scheduling needs more quickly, thereby improving the overall operating efficiency and market competitiveness of wind farms. By improving the accuracy and efficiency of predictions, the present invention has made positive contributions to the sustainable development of the wind power industry and the improvement of corporate benefits. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] The drawings in the specification, which constitute a part of the present application, are used to provide further understanding of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute improper limitations on the present application.
[0030] Figure 1 This is a principle block diagram of the wind power prediction method based on multi-branch deep separable convolution and mixer. Specific implementation methods
[0032] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.
[0033] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.
[0034] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.
[0035] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the methods and systems according to various embodiments of the present disclosure. It should be noted that each box in the flowchart or block diagram can represent a module, a program segment, or a part of a code, and the module, program segment, or a part of a code may include one or more executable instructions for implementing the logical functions specified in each embodiment. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, or they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the flowchart and / or block diagram, and the combination of boxes in the flowchart and / or block diagram can be implemented using a dedicated hardware-based system that performs a specified function or operation, or can be implemented using a combination of dedicated hardware and computer instructions.
[0036] Embodiment 1:
[0037] like Figure 1 As shown, this embodiment provides a wind power prediction method based on multi-branch depth-separable convolution and mixer, comprising the following steps:
[0038] Step 1: Preprocess the wind farm meteorological data, equipment monitoring data, and wind turbine basic data, and sort the preprocessed data according to the timeline;
[0039] Specifically, the meteorological data of wind farms, such as wind speed, temperature, humidity, and air pressure, are standardized and cleaned to remove abnormal data; the equipment monitoring data is normalized and its mapping space is converted into a constant space using functions such as sin and cos; and the basic data of wind turbines are supplemented and screened.
[0040] Step 2: Send the sorted data into the depth-wise separable convolutional model, and obtain the trend term and period term through the average pooling decomposition layer;
[0041] Specifically, the original power sequence in the data is average pooled through the average pooling decomposition layer, and the output after average pooling is used as the trend term; the original power sequence is differentiated from the trend term, and the output is used as the periodic term.
[0042] Step 3: The periodic term enters the depth-wise separable convolutional layer to extract the association in the time dimension, extract the association in the channel dimension through point-by-point convolution, and then map the vector of the periodic term to the hidden layer feature; specifically, it includes
[0043] Step 31: Initialize M convolution kernels (M is the feature dimension of the original power sequence);
[0044] Step 32: Perform a depth convolution on the periodic term. The M dimensions correspond to M convolution kernels respectively, and a one-to-one depth convolution is performed on the time dimension to extract the joint features on the time dimension.
[0045] Step 33: Perform point-by-point convolution on the joint features in the time dimension, extract the joint features in the channel dimension, and obtain the hidden layer features in both the time and channel dimensions.
[0046] Step 4: Project the hidden layer features onto the periodic term output space through the fully connected layer to obtain the periodic term of the wind turbine power generation at each time point;
[0047] Specifically, by limiting the output dimension of the fully connected layer, the hidden layer features are dimensionally mapped to obtain the power cycle term of the wind turbine generator at the future moment.
[0048] Step 5: The trend item enters the linear layer, and the future trend item is directly generated according to the historical trend item to obtain the wind turbine power generation trend item at each time point;
[0049] Specifically, the dimension and activation function of the fully connected layer are output, and the trend item is dimensionally mapped and rectified to obtain the wind turbine power generation trend item at the future moment.
[0050] Step 6: Add the cycle term and trend term of the wind turbine generator power generation at each time point to obtain the wind turbine generator power generation at the corresponding time point.
[0051] This method combines deep learning and time series analysis technology to achieve accurate prediction of wind farm power, which is helpful for the operation management of wind farms and the stable operation of power grids.
[0052] Embodiment 2:
[0053] This embodiment provides a wind power prediction system based on multi-branch depthwise separable convolution and mixer, including:
[0054] The preprocessing module preprocesses the wind farm meteorological data, equipment monitoring data, and basic data of wind turbines, and sorts the preprocessed data according to the timeline;
[0055] The decomposition module feeds the sorted data into the depth-wise separable convolutional model and obtains the trend term and period term through the average pooling decomposition layer;
[0056] Hidden feature acquisition module, the periodic term enters the depth-separable convolution layer, extracts the association in the time dimension, extracts the association in the channel dimension through point-by-point convolution, and then maps the vector of the periodic term to the hidden layer feature;
[0057] The periodic term prediction module projects the hidden layer features onto the periodic term output space through the fully connected layer to obtain the periodic term of the wind turbine power generation at each time point;
[0058] A trend item prediction module, wherein the trend item enters the linear layer, directly generates future trend items based on historical trend items, and obtains the wind turbine power generation trend item at each time point;
[0059] The power generation prediction module adds the power generation cycle item and the trend item of the wind turbine generator at each time point to obtain the power generation of the wind turbine generator at the corresponding time point.
[0060] Embodiment three:
[0061] An electronic device includes a memory, a processor, and a computer program stored and running on the memory, wherein the processor implements the above-mentioned wind power prediction method based on multi-branch depth-separable convolution and mixer when executing the program, including:
[0062] Preprocess the wind farm meteorological data, equipment monitoring data, and wind turbine basic data, and sort the preprocessed data according to the timeline;
[0063] The sorted data is fed into a depth-wise separable convolutional model, and the trend term and period term are obtained through the average pooling decomposition layer;
[0064] The periodic term enters the depthwise separable convolutional layer to extract the association in the time dimension, and extracts the association in the channel dimension through point-by-point convolution, thereby mapping the vector of the periodic term into a hidden layer feature;
[0065] Through the fully connected layer, the hidden layer features are projected onto the periodic term output space to obtain the periodic term of the wind turbine power generation at each time point;
[0066] The trend item enters the linear layer, and the future trend item is directly generated according to the historical trend item to obtain the wind turbine power generation trend item at each time point;
[0067] The periodic term and the trend term of the wind turbine generator power generation at each time point are added together to obtain the wind turbine generator power generation at the corresponding time point.
[0068] Embodiment 4:
[0069] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned wind power prediction method based on multi-branch depthwise separable convolution and mixer, comprising:
[0070] Preprocess the wind farm meteorological data, equipment monitoring data, and wind turbine basic data, and sort the preprocessed data according to the timeline;
[0071] The sorted data is fed into a depth-wise separable convolutional model, and the trend term and period term are obtained through the average pooling decomposition layer;
[0072] The periodic term enters the depthwise separable convolutional layer to extract the association in the time dimension, and extracts the association in the channel dimension through point-by-point convolution, thereby mapping the vector of the periodic term into a hidden layer feature;
[0073] Through the fully connected layer, the hidden layer features are projected onto the periodic term output space to obtain the periodic term of the wind turbine power generation at each time point;
[0074] The trend item enters the linear layer, and the future trend item is directly generated according to the historical trend item to obtain the wind turbine power generation trend item at each time point;
[0075] The periodic term and the trend term of the wind turbine generator power generation at each time point are added together to obtain the wind turbine generator power generation at the corresponding time point.
[0076] Those skilled in the art should understand that the modules or steps of the present disclosure can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present disclosure is not limited to any specific combination of hardware and software.
[0077] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.
[0078] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.
Claims
1. A wind power prediction method based on multi-branch deep separable convolution and mixer, characterized in that: The following steps are involved: Preprocess the wind farm meteorological data, equipment monitoring data, and wind turbine basic data, and sort the preprocessed data according to the timeline; The sorted data is fed into a depth-wise separable convolutional model, and the trend term and period term are obtained through the average pooling decomposition layer; The periodic term enters the depthwise separable convolutional layer to extract the association in the time dimension, and extracts the association in the channel dimension through point-by-point convolution, thereby mapping the vector of the periodic term into a hidden layer feature; Through the fully connected layer, the hidden layer features are projected onto the periodic term output space to obtain the periodic term of the wind turbine power generation at each time point; The trend item enters the linear layer, and the future trend item is directly generated according to the historical trend item to obtain the wind turbine power generation trend item at each time point; The wind turbine generator power generation cycle term and trend term at each time point are added together to obtain the wind turbine generator power generation at the corresponding time point.
2. According to claim 1, a wind power prediction method based on multi-branch depthwise separable convolution and mixer is characterized in that: Standardize and clean the meteorological data of wind farms to remove abnormal data; normalize the equipment monitoring data to transform its mapping space into a constant space; supplement and filter the basic data of wind turbines.
3. According to claim 1, a wind power prediction method based on multi-branch depthwise separable convolution and mixer is characterized in that: The average pooling decomposition layer performs an average pooling operation on the original power sequence in the data to obtain a trend term; the original power sequence is differentiated from the trend term to obtain a periodic term.
4. According to claim 3, a wind power prediction method based on multi-branch depthwise separable convolution and mixer is characterized in that: The periodic term enters the depthwise separable convolution layer, and the feature dimensions M of the original power sequence correspond to M convolution kernels, and a one-to-one depthwise convolution is performed to extract the joint features in the time dimension; The joint features in the time dimension are convolved point by point, the joint features in the channel dimension are extracted, and the hidden layer features in the dual dimensions of time and channel are obtained.
5. According to claim 1, a wind power prediction method based on multi-branch depthwise separable convolution and mixer is characterized in that: By limiting the output dimension of the fully connected layer, the hidden layer features are dimensionally mapped to obtain the wind turbine power generation cycle term at future moments.
6. According to claim 1, a wind power prediction method based on multi-branch depthwise separable convolution and mixer is characterized in that: The trend item is dimensionally mapped and rectified through the output dimension of the fully connected layer and the activation function to obtain the wind turbine power generation trend item at the future moment.
7. A wind power prediction system based on multi-branch deep separable convolution and mixer, characterized in that: include: The preprocessing module preprocesses the wind farm meteorological data, equipment monitoring data, and basic data of wind turbines, and sorts the preprocessed data according to the timeline; The decomposition module feeds the sorted data into the depth-wise separable convolutional model and obtains the trend term and period term through the average pooling decomposition layer; Hidden feature acquisition module, the periodic term enters the depth-separable convolution layer, extracts the association in the time dimension, extracts the association in the channel dimension through point-by-point convolution, and then maps the vector of the periodic term to the hidden layer feature; The periodic term prediction module projects the hidden layer features onto the periodic term output space through the fully connected layer to obtain the periodic term of the wind turbine power generation at each time point; A trend item prediction module, wherein the trend item enters the linear layer, directly generates future trend items based on historical trend items, and obtains the wind turbine power generation trend item at each time point; The power generation prediction module adds the power generation cycle item and the trend item of the wind turbine generator at each time point to obtain the power generation of the wind turbine generator at the corresponding time point.
8. An electronic device comprising a memory, a processor and a computer program stored and running on the memory, characterized in that: When the processor executes the program, the wind power prediction method based on multi-branch depth-separable convolution and mixer is implemented.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, a wind power prediction method based on multi-branch depth-separable convolution and mixer is implemented.
