A wind power prediction method, device, equipment and medium
By extracting local and global features of wind power using a deep neural network model, the problem of error accumulation and difficulty in balancing feature relationships in wind power prediction in existing technologies is solved, thus achieving stable and accurate prediction of wind power.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HEBEI HANFENG POWER GENERATION CO LTD
- Filing Date
- 2026-02-14
- Publication Date
- 2026-06-02
AI Technical Summary
Existing wind power forecasting technologies cannot generate complete intraday power sequences, leading to the accumulation of forecasting errors. Furthermore, they struggle to balance local short-term dynamic characteristics with global dependencies over long time spans, resulting in insufficient stability and accuracy of forecasting results.
By employing a deep neural network model, the system acquires meteorological forecast time series of the target area, extracts local and global features using convolutional neural networks and Transformer encoders, and combines multi-head self-attention mechanism and gating generation layer for feature fusion, thereby achieving end-to-end modeling from meteorological series to wind power series.
It has achieved stable and accurate prediction of wind power, improved the temporal continuity and physical rationality of the prediction results, avoided error accumulation, and improved prediction accuracy and stability.
Smart Images
Figure CN122133105A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of new energy power generation prediction technology, and more specifically, to a wind power prediction method, device, equipment and medium. Background Technology
[0002] With the continuous growth of wind power installed capacity, the strong fluctuations and uncertainties in wind power output place higher demands on power system dispatching, safe and stable operation, and renewable energy absorption capacity. Wind power forecasting, as a key technology supporting the grid connection of a high proportion of renewable energy, directly affects grid operating efficiency, reserve capacity configuration, and dispatching economy. Therefore, there is an urgent need to develop advanced forecasting methods that can accurately depict the intraday variation patterns of wind power, maintain a consistent output structure, and provide a physically reasonable complete power sequence.
[0003] Existing wind power prediction technologies can be mainly divided into three categories: the first is based on statistics and traditional time series models (such as autoregressive models). These methods assume that power changes have linear or weakly nonlinear characteristics, making it difficult to cope with non-stationary processes under complex meteorological conditions. The second is machine learning methods (such as support vector regression, random forests, etc.). Although they introduce supervised learning mechanisms, they generally model the prediction task as a point-to-point regression problem, that is, using meteorological and power information within the current or limited historical window to predict the power value at a single future moment. The third is deep learning models (such as recurrent neural networks and their variants). Although they are widely used in time series modeling, they still mostly follow the time extrapolation paradigm, generating prediction results point by point through recursion or sliding windows, failing to fundamentally break through the limitations of single-point prediction.
[0004] The aforementioned existing technologies have two core drawbacks: First, the prediction target is limited to a single point output and cannot directly generate a complete intraday power sequence. This leads to the need to call the model multiple times in practical applications such as wind power dispatch and power balance, which can easily cause error accumulation and make it difficult to guarantee the temporal continuity and physical rationality of the prediction curve. Second, the model structure is difficult to take into account both local short-term dynamic characteristics and global dependencies over long periods. Traditional methods either rely too much on historical power autocorrelation and weaken meteorological driving information, or only capture local changes and ignore the overall evolution trend. Especially in ultra-short-term multi-step prediction scenarios, the stability, consistency and accuracy of the prediction results are difficult to meet the practical engineering requirements. Summary of the Invention
[0005] The purpose of this application is to provide a wind power prediction method, device, equipment and medium to solve the problem of insufficient prediction accuracy in the prior art, and to achieve stable and accurate prediction of wind power.
[0006] Firstly, a wind power prediction method is provided, which may include: Obtain the meteorological forecast time series of the target area within a future preset time period; wherein, the meteorological forecast time series includes multiple consecutive time points within the future preset time period and the corresponding meteorological forecast data; The meteorological forecast time series is input into a pre-trained wind power prediction model to obtain a wind power prediction sequence for the target area in a future preset time period; wherein, the wind power prediction model is trained using historical wind power meteorological sequences of the target area in multiple historical time periods; the historical wind power meteorological sequence in any historical time period includes multiple consecutive historical time points in the corresponding historical time period and the historical meteorological data and historical wind power of the corresponding historical time points.
[0007] In one optional implementation, the weather forecast data includes wind speed and wind direction.
[0008] In an optional implementation, the method for obtaining the meteorological forecast time series includes: Obtain raw weather forecast data for the target area within a preset time period; Based on the configured time interval, the original meteorological forecast data is interpolated or resampled to generate the meteorological forecast time series.
[0009] In an optional implementation, the wind power prediction model includes: The input layer is used to input the meteorological forecast time series of the target area within a preset future time period; A feature embedding layer is used to perform a linear transformation on the meteorological forecast time series to obtain embedded features; The first feature extraction layer is used to extract local temporal features from the embedded features to obtain a first feature vector; The second feature extraction layer is used to perform global feature extraction on the embedded features and the meteorological forecast time series to obtain a second feature vector. A feature fusion layer is used to fuse the first feature vector and the second feature vector to obtain fused features; A regression mapping layer is used to map the fused features into a wind power prediction sequence for the target area within a future preset time period; wherein, the wind power prediction sequence includes multiple consecutive time points within the future preset time period and corresponding wind power prediction values.
[0010] In an optional implementation, the first feature extraction layer includes: at least two parallel convolutional modules with different convolutional sizes and a first global average pooling layer; Each convolutional module is used to perform linear transformation, nonlinear activation and normalization on the embedded features using convolutional kernels of different sizes, extract local temporal features at different scales, and then concatenate and aggregate the local temporal features at different scales to obtain a local temporal feature sequence. The first global average pooling layer is used to aggregate the local temporal feature sequences in the time dimension to obtain the first feature vector.
[0011] In an optional implementation, the second feature extraction layer includes: a wind direction sector coding layer, a position coding layer, a multi-head self-attention layer, a feedforward neural network layer, and a second global average pooling layer; The wind direction sector coding layer is used to discretize the wind direction data in the meteorological forecast time series into multiple sectors and generate a corresponding sector embedding vector for each time step. The location encoding layer is used to add temporal location information to the embedded features to obtain an input sequence with location encoding; The multi-head self-attention layer is used to calculate the correlation between any two time steps based on wind direction sector perception according to the input sequence with position encoding and the sector embedding vector, and to weight and aggregate context information to obtain the attention output sequence. The feedforward neural network layer is used to perform position-by-position nonlinear transformation on the attention output sequence to obtain an enhanced temporal feature sequence; The second global average pooling layer is used to calculate the mean of the enhanced temporal feature sequence in the time dimension to obtain the second feature vector.
[0012] In an optional implementation, the feature fusion layer includes: a feature splicing layer, a gated generation layer, and a weighted fusion layer; The feature concatenation layer is used to concatenate the first feature vector and the second feature vector along the feature dimension to obtain a joint feature vector; A gated generation layer is used to perform a linear transformation on the joint feature vector and activate it with Sigmoid to obtain a gated coefficient vector. The weighted fusion layer is used to perform element-wise multiplication of the joint feature vector and the gate coefficient vector to obtain the fused features. Secondly, a wind power prediction device is provided, which may include: The acquisition unit is used to acquire the meteorological forecast time series of the target area within a future preset time period; wherein, the meteorological forecast time series includes multiple consecutive time points within the future preset time period and the corresponding meteorological forecast data; The prediction unit is used to input the meteorological prediction time series into a pre-trained wind power prediction model to obtain a wind power prediction sequence for the target area in a future preset time period; wherein, the wind power prediction model is trained using historical wind power meteorological sequences of the target area in multiple historical time periods; the historical wind power meteorological sequence in any historical time period includes multiple consecutive historical time points in the corresponding historical time period and the historical meteorological data and historical wind power of the corresponding historical time points.
[0013] Thirdly, an electronic device is provided, which includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements any of the steps described in the first aspect above.
[0014] Fourthly, a computer-readable storage medium is provided, wherein a computer program is stored therein, and when executed by a processor, the computer program implements the steps of any of the methods described in the first aspect above.
[0015] This application applies to ultra-short-term and short-term power prediction scenarios for wind farms. It utilizes a pre-trained wind power prediction model to model and analyze the meteorological forecast time series of a target area over a predetermined future time period, obtaining a wind power prediction sequence for the target area within that time period. The wind power prediction model employs a deep neural network to jointly learn the time dependencies and nonlinear characteristics of historical wind power data and historical meteorological forecast data, thereby achieving the prediction of wind power variation patterns within a certain future timeframe. This application realizes end-to-end modeling from meteorological time series to wind power prediction results. Attached Figure Description
[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 An architecture diagram of a wind power prediction system provided in this application embodiment; Figure 2 A flowchart illustrating a wind power prediction method provided in an embodiment of this application; Figure 3A flowchart illustrating another wind power prediction method provided in this application embodiment; Figure 4 This application provides a schematic diagram of training data screening as an embodiment. Figure 5 This is a schematic diagram of the structure of a wind power prediction device provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Unless otherwise defined, the technical or scientific terms used in this application should have the ordinary meaning understood by those skilled in the art. The words "first," "second," and similar terms used in this application do not indicate any order, quantity, or importance, but are only used to distinguish different components. The words "comprising" or "including," etc., mean that the element or object preceding the word covers the element or object listed after the word and its equivalents, but do not exclude other elements or objects. The words "connected," "coupled," or "connected," etc., are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up," "down," "left," "right," etc., are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0019] The wind power prediction method provided in this application embodiment can be applied to... Figure 1 In the system architecture shown, such as Figure 1 As shown, the system may include: a server and a data acquisition interface; wherein, the data acquisition interface is used to acquire raw meteorological forecast data of the target area within a preset time period in the future; The server is used to execute the wind power prediction method provided in the embodiments of this application. The server can be a physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.
[0020] The preferred embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit this application. Furthermore, the embodiments and features in the embodiments of this application can be combined with each other without conflict.
[0021] Figure 2 This is a flowchart illustrating a wind power prediction method provided in an embodiment of this application. Figure 2 As shown, the method may include: Step S210: Obtain the meteorological forecast time series of the target area within a preset time period in the future.
[0022] The meteorological forecast time series includes multiple consecutive time points within a future preset time period and corresponding meteorological forecast data; the time interval between any two adjacent time points is the same, and this time interval can be 15 minutes; the future preset time period can be 24 hours; the meteorological forecast data can include: wind speed, wind direction, air density, temperature, wind shear index, atmospheric stability, humidity, air pressure, and turbulence intensity.
[0023] In specific implementation, the method for obtaining meteorological forecast time series includes: acquiring numerical weather prediction data, real-time monitoring data from meteorological towers, and satellite-derived meteorological data for the target area within a preset future time period; aligning the acquired numerical weather prediction data, real-time monitoring data from meteorological towers, and satellite-derived meteorological data along the time axis and removing outliers to obtain the original meteorological forecast data for the target area within the preset future time period; inputting the original meteorological forecast data into a pre-trained error correction model to obtain corrected meteorological forecast data; wherein, the error correction model is trained using the deviation sequence between the original meteorological forecast data of historical time periods and the measured meteorological data of the corresponding time periods; interpolating or resampling the corrected meteorological forecast data according to the configured time interval to generate a resampled sequence; performing time-series stabilization processing on the resampled sequence using differencing, moving average, or wavelet denoising to obtain the meteorological forecast time series; wherein, the configured time interval can be 15 minutes.
[0024] Step S220: Input the meteorological forecast time series into the pre-trained wind power prediction model to obtain the wind power prediction sequence of the target area in the future preset time period.
[0025] The wind power prediction model is trained using historical wind power meteorological sequences of the target area over multiple historical time periods. The historical wind power meteorological sequence for any historical time period includes multiple consecutive historical time points within the corresponding historical time period, as well as historical meteorological data and historical wind power at the corresponding historical time points.
[0026] In specific implementation, such as Figure 3 As shown, the meteorological forecast time series is input into the pre-trained wind power prediction model, and CNN feature extraction, Transformer feature extraction, feature concatenation and linear mapping are performed in sequence to obtain the wind power prediction sequence.
[0027] In one embodiment of this application, the wind power prediction model may include: The input layer is used to input the meteorological forecast time series of the target area within a future preset time period; where the future preset time period can be a future day; the meteorological forecast time series is the meteorological forecast time series corresponding to the future forecast day, and its dimension is (96,F), where 96 represents the number of time points within the forecast day, and F represents the meteorological feature dimension. The feature embedding layer is used to map the meteorological forecast time series to a high-dimensional representation space of uniform dimension through linear transformation, thereby eliminating the dimensional differences between different physical quantities and enhancing the model's ability to express nonlinear relationships, thus obtaining embedded features; The first feature extraction layer uses a convolutional neural network (CNN) to extract local temporal features from the embedded features, resulting in a first feature vector. The CNN performs one-dimensional convolution operations along the time dimension on the embedded features, modeling the relationship between adjacent time points through local receptive fields, thereby capturing the local fluctuations in wind speed, meteorological conditions, and power changes over short time scales. The CNN employs multiple convolutions to progressively abstract local temporal patterns and compresses the time dimension through pooling operations, forming a compact feature vector representing local dynamic features. This CNN can effectively characterize the rapid changes and local trends in wind power over short time ranges. The second feature extraction layer employs a Transformer encoder with positional encoding to perform global feature extraction on the embedded features, resulting in a second feature vector. The Transformer encoder with positional encoding introduces positional encoding to explicitly represent the relative order of different time points within the intraday sequence, modeling the time series through multiple Transformer encoders. The Transformer's self-attention mechanism can establish correlations between any time points across the entire sequence, enabling the model to comprehensively consider the contextual information of the entire time series when modeling features at a specific time point. This effectively captures the global trend and long-term dependency structure of wind power changes with meteorological conditions. The time series features encoded by the Transformer are also compressed into a fixed-length global feature vector through pooling operations to characterize the overall structural features of intraday power changes. The feature fusion layer is used to fuse the first and second feature vectors using a gated fusion mechanism to obtain fused features. Specifically, the first and second feature vectors are concatenated along their feature dimensions using a gated fusion mechanism to obtain concatenated features. Gating coefficients are calculated using a linear mapping and a sigmoid activation function. By multiplying the gate coefficients element-wise with the concatenated features, adaptive weighting is applied to temporal features at different scales to dynamically balance the contributions of local temporal variation information and global temporal structure information during prediction. The gate coefficients range from 0 to 1, representing the relative importance of each feature dimension in the current sample. The gate coefficients provided in this embodiment are automatically learned through backpropagation during training, enabling the model to autonomously adjust the feature fusion strategy according to different meteorological conditions and power change patterns. The regression mapping layer is used to map the fused features into a wind power prediction sequence for the target region within a preset future time period using multiple linear regression. Specifically, it employs multiple linear regression (MLR) or its equivalent implementation, mapping the fused high-dimensional feature representation to the target power value through a fully connected layer. The regression mapping layer establishes a functional relationship between the temporal features extracted by the deep network and the wind power label, enabling end-to-end prediction modeling. Through this regression mapping layer, the model can transform the learned multi-scale temporal features into specific power prediction results, achieving a complete prediction process from meteorological sequence input to wind power sequence output. The wind power prediction sequence for the target region within a preset future time period is a meteorological prediction time series strictly time-aligned with the input meteorological sequence, with dimensions (96,1), representing the actual wind power value corresponding to each time point within the predicted day. Through this definition, the wind power prediction problem is transformed from a traditional point-to-point regression problem into a sequence-to-sequence mapping problem from meteorological sequence to power sequence, enabling the model to learn the intraday variation structure of wind power on an overall time scale.
[0028] In one embodiment of this application, the first feature extraction layer may include: at least two parallel convolutional modules with different convolutional sizes and a first global average pooling layer; each convolutional module is used to perform linear transformation, nonlinear activation, and normalization processing on the embedded features using convolutional kernels of different sizes to extract local temporal features at different scales, and then concatenate and aggregate the multi-scale features to obtain a local temporal feature sequence; wherein, each convolutional module includes a one-dimensional convolutional layer, a ReLU activation layer, and a batch normalization layer; the one-dimensional convolutional layer is used to perform linear transformation on the embedded features along the time dimension to obtain a convolutional time series; the ReLU activation layer is used to perform nonlinear mapping on the convolutional time series to obtain an activated time series; the batch normalization layer is used to normalize the activated time series in the channel dimension to obtain a normalized time series; the output of each batch normalization layer is used as the input of the one-dimensional convolutional layer of the next convolutional module; the first global average pooling layer is used to aggregate the local temporal feature sequence output by the last batch normalization layer in the time dimension to obtain a first feature vector.
[0029] In one embodiment of this application, the second feature extraction layer may include: The wind direction sector coding layer is used to discretize the wind direction data in the meteorological forecast time series into multiple sectors and generate a corresponding sector embedding vector for each time step. The positional encoding layer is used to add temporal and positional information to the embedded features, resulting in an input sequence with positional encoding. The multi-head self-attention layer is used to use historical wind power data of the target area as a mask constraint, calculate the correlation between any two time steps based on wind direction sector perception according to the input sequence with position encoding and the sector embedding vector, and weight and aggregate context information to obtain the attention output sequence. The feedforward neural network layer is used to perform position-by-position nonlinear transformation on the attention output sequence to obtain the enhanced temporal feature sequence. The second global average pooling layer is used to calculate the mean of the enhanced temporal feature sequence in the time dimension, and obtain the second feature vector.
[0030] In one embodiment of this application, the feature fusion layer may include: a feature splicing layer, a gated generation layer, and a weighted fusion layer; The feature concatenation layer is used to concatenate the first feature vector and the second feature vector along the feature dimension to obtain a joint feature vector; The gated generation layer is used to perform a linear transformation on the joint feature vector and activate it with Sigmoid to obtain the gated coefficient vector. The gated generation layer is used to dynamically adjust the gated coefficient vector according to the meteorological forecast time series of the target area in the future preset time period, so as to achieve adaptive matching of the fusion weights. The weighted fusion layer is used to multiply the joint feature vector and the gating coefficient vector element by element to obtain the fused features.
[0031] In one embodiment of this application, the method for obtaining historical wind power meteorological sequences may include: Historical meteorological data and historical wind power data for each historical time period are time-aligned and segmented. Specifically, historical wind power and historical meteorological data are resampled at a fixed time resolution (15 minutes) to form a daily time series sample containing 96 consecutive time points. To ensure that the time series model's assumption of continuous time structure holds, data integrity is constrained during sample construction: if there are missing, discontinuous, or time-aligned failures among the 96 time points of a certain day, the daily series is considered to not meet the continuity requirement and is not included in model training and prediction. Figure 4 As shown, the above sample selection mechanism ensures that the time series input to the model are consistent in length and temporal structure, thereby avoiding modeling bias caused by missing values disrupting the convolution window or attention calculation.
[0032] In another embodiment of this application, after obtaining the wind power prediction sequence for the target area within a preset future time period, the method may further include: Obtain the historical wind power prediction sequence for the target area in the previous time period based on the wind power prediction model for the current time period; Obtain the measured wind power sequence of the target area in the current time period; The difference between the historical wind power prediction value at each time point in the historical wind power prediction sequence corresponding to the current time period and the measured wind power value at the corresponding time point in the measured wind power sequence is calculated to obtain the historical prediction deviation time series. The historical predicted deviation time series is input into a pre-trained deviation prediction model to obtain the expected deviation sequence of the target area in the future within a preset time period; wherein, the deviation prediction model is a lightweight time series prediction model, which can be one of the following: ARIMA model, exponential smoothing model, or long short-term memory network; the deviation prediction model is trained using the deviation values of the historical wind power sequence and the corresponding measured wind power sequence of the target area. The wind power prediction sequence for the target area in the future preset time period is corrected by using the expected deviation sequence of the target area in the future preset time period, so as to obtain the target wind power prediction sequence for the target area in the future preset time period.
[0033] In another embodiment of this application, after obtaining the wind power prediction sequence of the target area within a preset time period in the future, the method may further include: The system verifies whether the wind power prediction sequence of the target area within a preset time period meets the preset wind turbine constraints. If not, the wind power prediction value at the corresponding time point is corrected. The wind turbine constraints are determined based on the physical characteristics of wind turbine operation. The wind turbine constraints may include: when the wind speed is lower than the cut-in wind speed, the wind power prediction value is 0; when the wind speed is higher than the cut-out wind speed, the wind power prediction value is 0; when the wind speed is within the configured rated wind speed range, the predicted power should not exceed the rated power of the wind turbine.
[0034] This application provides a wind power prediction method with a clear structure and closed-loop logic, which enables the wind power prediction model to make full use of local and global time series information in meteorological sequences while ensuring temporal continuity and physical consistency, thereby improving the accuracy and stability of wind power prediction.
[0035] This application discretizes the wind power prediction timeline into a fixed number of time points and uniformly sets the time interval between adjacent time points to 15 minutes, thereby constructing an intraday time series containing 96 time points. Under this setting, the wind power prediction problem is no longer regarded as an extrapolation of the power value at a single future moment, but is redefined as a sequence-to-sequence prediction problem that directly maps the complete power sequence within the next day from historical time series and corresponding meteorological forecast time series. By expanding the prediction target from a single point power value to a complete intraday power sequence, this application enables the wind power prediction model to learn the structural characteristics of wind power changes over time at a holistic level, thereby explicitly modeling the interrelationships between different time points. This modeling approach not only fully utilizes the influence of wind speed, wind direction, and power information from previous moments on subsequent power changes, but also comprehensively considers the continuity and consistency of meteorological forecast data in the time dimension during the prediction process, making the model's characterization of the intraday evolution of wind power more comprehensive. By adopting a sequence-to-sequence prediction modeling approach, the wind power prediction model is no longer limited to regression fitting at local time points. Instead, through joint learning of the complete time series, it improves the modeling ability for the overall trend of wind power change and local fluctuation characteristics, thereby achieving better performance in terms of the continuity, stability, and physical consistency of the prediction results.
[0036] This application can directly output a complete intraday power sequence in a single prediction process, avoiding the error accumulation problem caused by point-by-point recursive prediction. Furthermore, through joint modeling using CNN and Transformer, the model can simultaneously characterize both the local features and global structure of power changes, resulting in better performance in terms of temporal continuity and physical plausibility. The prediction results of this application are output in the form of a complete sequence, making this method highly practical in engineering applications such as wind power dispatching, grid power balance analysis, and renewable energy consumption assessment.
[0037] Corresponding to the above method, embodiments of this application also provide a wind power prediction device, such as... Figure 5 As shown, the device includes: The acquisition unit 510 is used to acquire the meteorological forecast time series of the target area within a future preset time period; wherein, the meteorological forecast time series includes multiple consecutive time points within the future preset time period and the corresponding meteorological forecast data; The prediction unit 520 is used to input the meteorological forecast time series into the pre-trained wind power prediction model to obtain the wind power prediction series of the target area in the future preset time period; wherein, the wind power prediction model is trained using the historical wind power meteorological series of the target area in multiple historical time periods; the historical wind power meteorological series in any historical time period includes multiple consecutive historical time points in the corresponding historical time period and the historical meteorological data and historical wind power of the corresponding historical time points.
[0038] The functions of each functional unit of the wind power prediction device provided in the above embodiments of this application can be realized through the above methods and steps. Therefore, the specific working process and beneficial effects of each unit in the wind power prediction device provided in the embodiments of this application will not be repeated here.
[0039] This application also provides an electronic device, such as... Figure 6 As shown, it includes a processor 610, a communication interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640.
[0040] Memory 630 is used to store computer programs; When the processor 610 executes the program stored in the memory 630, it performs the following steps: Obtain the meteorological forecast time series of the target area within a future preset time period; wherein, the meteorological forecast time series includes multiple consecutive time points within the future preset time period and the corresponding meteorological forecast data; The meteorological forecast time series is input into the pre-trained wind power prediction model to obtain the wind power prediction sequence of the target area in the future within a preset time period. The wind power prediction model is trained using the historical wind power meteorological sequence of the target area in multiple historical time periods. The historical wind power meteorological sequence in any historical time period includes multiple consecutive historical time points in the corresponding historical time period, as well as the historical meteorological data and historical wind power of the corresponding historical time points.
[0041] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.
[0042] The communication interface is used for communication between the aforementioned electronic devices and other devices.
[0043] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.
[0044] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0045] The implementation methods and beneficial effects of the various components of the electronic device in the above embodiments for solving the problem can be found in [reference needed]. Figure 2 The steps in the illustrated embodiments are used to implement the electronic device. Therefore, the specific working process and beneficial effects of the electronic device provided in this application will not be repeated here.
[0046] In another embodiment provided in this application, a computer-readable storage medium is also provided, which stores instructions that, when executed on a computer, cause the computer to perform any of the wind power prediction methods described in the above embodiments.
[0047] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to execute any of the wind power prediction methods described in the above embodiments.
[0048] Those skilled in the art will understand that the embodiments in this application can be provided as methods, systems, or computer program products. Therefore, the embodiments in this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the embodiments in this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0049] This application describes embodiments of methods, apparatus (systems), and computer program products according to embodiments of this application with reference to flowchart illustrations and / or block diagrams. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0050] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0051] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0052] Although preferred embodiments have been described in this application, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of this application.
[0053] Obviously, those skilled in the art can make various modifications and variations to the embodiments of this application without departing from the spirit and scope of the embodiments of this application. Therefore, if these modifications and variations to the embodiments of this application fall within the scope of this application and its equivalents, then these modifications and variations are also intended to be included in the embodiments of this application.
Claims
1. A method for predicting wind power output, characterized in that, The method includes: Obtain the meteorological forecast time series for the target area within a preset future time period; wherein, the meteorological forecast time series includes multiple consecutive time points within the preset future time period and the corresponding meteorological forecast data; The meteorological forecast time series is input into a pre-trained wind power prediction model to obtain a wind power prediction sequence for the target area in a future preset time period; wherein, the wind power prediction model is trained using historical wind power meteorological sequences of the target area in multiple historical time periods; the historical wind power meteorological sequence in any historical time period includes multiple consecutive historical time points in the corresponding historical time period and the historical meteorological data and historical wind power of the corresponding historical time points.
2. The method as described in claim 1, characterized in that, The meteorological forecast data includes wind speed and wind direction.
3. The method as described in claim 1, characterized in that, The method for obtaining the meteorological forecast time series includes: Obtain raw weather forecast data for the target area within a preset time period; Based on the configured time interval, the original meteorological forecast data is interpolated or resampled to generate the meteorological forecast time series.
4. The method as described in claim 2, characterized in that, The wind power prediction model includes: The input layer is used to input the meteorological forecast time series of the target area within a preset future time period; A feature embedding layer is used to perform a linear transformation on the meteorological forecast time series to obtain embedded features; The first feature extraction layer is used to extract local temporal features from the embedded features to obtain a first feature vector; The second feature extraction layer is used to perform global feature extraction on the embedded features and the meteorological forecast time series to obtain a second feature vector. A feature fusion layer is used to fuse the first feature vector and the second feature vector to obtain fused features; A regression mapping layer is used to map the fused features into a wind power prediction sequence for the target area within a future preset time period; wherein, the wind power prediction sequence includes multiple consecutive time points within the future preset time period and corresponding wind power prediction values.
5. The method as described in claim 4, characterized in that, The first feature extraction layer includes: at least two parallel convolutional modules with different convolutional sizes and a first global average pooling layer; Each convolutional module is used to perform linear transformation, nonlinear activation and normalization on the embedded features using convolutional kernels of different sizes, extract local temporal features at different scales, and then concatenate and aggregate the local temporal features at different scales to obtain a local temporal feature sequence. The first global average pooling layer is used to aggregate the local temporal feature sequences in the time dimension to obtain the first feature vector.
6. The method as described in claim 4, characterized in that, The second feature extraction layer includes: a wind direction sector encoding layer, a position encoding layer, a multi-head self-attention layer, a feedforward neural network layer, and a second global average pooling layer; The wind direction sector coding layer is used to discretize the wind direction data in the meteorological forecast time series into multiple sectors and generate a corresponding sector embedding vector for each time step. The location encoding layer is used to add temporal location information to the embedded features to obtain an input sequence with location encoding; The multi-head self-attention layer is used to calculate the correlation between any two time steps based on wind direction sector perception according to the input sequence with position encoding and the sector embedding vector, and to weight and aggregate context information to obtain the attention output sequence. The feedforward neural network layer is used to perform position-by-position nonlinear transformation on the attention output sequence to obtain an enhanced temporal feature sequence; The second global average pooling layer is used to calculate the mean of the enhanced temporal feature sequence in the time dimension to obtain the second feature vector.
7. The method as described in claim 4, characterized in that, The feature fusion layer includes: a feature splicing layer, a gated generation layer, and a weighted fusion layer; The feature concatenation layer is used to concatenate the first feature vector and the second feature vector along the feature dimension to obtain a joint feature vector; A gated generation layer is used to perform a linear transformation on the joint feature vector and activate it with Sigmoid to obtain a gated coefficient vector. The weighted fusion layer is used to multiply the joint feature vector and the gating coefficient vector element by element to obtain the fused features.
8. A wind power prediction device, characterized in that, The device includes: The acquisition unit is used to acquire the meteorological forecast time series of the target area within a future preset time period; wherein, the meteorological forecast time series includes multiple consecutive time points within the future preset time period and the corresponding meteorological forecast data; The prediction unit is used to input the meteorological prediction time series into a pre-trained wind power prediction model to obtain a wind power prediction sequence for the target area in a future preset time period; wherein, the wind power prediction model is trained using historical wind power meteorological sequences of the target area in multiple historical time periods; the historical wind power meteorological sequence in any historical time period includes multiple consecutive historical time points in the corresponding historical time period and the historical meteorological data and historical wind power of the corresponding historical time points.
9. An electronic device, characterized in that, The electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the method of any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method described in any one of claims 1-7.