A wind farm active power-wind speed inversion method and system
Patent Information
- Application Number
- CN202611054654.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-16
- Publication Date
- 2026-08-18
AI Technical Summary
[0005]综上，现有技术在时序关联挖掘、空间特征融合、反演精度与效率平衡等方面仍存在短板
[0019]The active power-wind speed inversion method and system for wind farms disclosed in this application, based on the PatchTST architecture, divides the time-series data of total active power and measured wind speed at the center point of the wind farm into multiple continuous subsequence segments. These segments are then aggregated by a Transformer encoder to obtain a correlation feature vector of nonlinear coupling between active power and wind speed. The correlation feature vector is concatenated with the current active power, mapped by a linear regression head, and subjected to physical constraints using the equivalent power curve of the wind farm to obtain the wind speed inversion sequence at the center point. The wind speed inversion sequence at the center point is then concatenated with static spatial features and fed into a selective state-space model based on the Mamba architecture to output a time-varying correlation feature vector. Finally, a linear regression head is designed for each spatial grid to obtain the regional wind speed inversion time series, and physical constraints are applied using the single-unit power curve to form a wind speed inversion link from the center point to the region. This achieves accurate inversion of wind speed at the center point and in the region of the wind farm, improving the refinement and reliability of wind speed sensing.
Smart Images
Figure CN122600340A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of new energy power generation and power system optimization operation technology, and particularly relates to a method and system for wind farm active power-wind speed inversion. Background Technology
[0002] Wind power generation is currently the mainstream form of clean energy, and the safe operation of wind farms and the efficient dispatch of the power system are highly dependent on the accurate sensing of wind speed. As the core driving parameter for wind power output, the monitoring accuracy of wind speed directly affects power generation prediction, unit control, and grid dispatch decisions. Currently, wind speed acquisition mainly relies on physical equipment such as anemometer towers and lidar, but this method suffers from problems such as high deployment costs, insufficient regional coverage, and susceptibility to environmental interference. Single-point data cannot reflect the spatially heterogeneous wind speed distribution within a wind farm affected by terrain and wake, making refined wind speed sensing a pain point in the industry.
[0003] There is a strong physical correlation between active power and wind speed, providing a theoretical basis for "wind speed inversion from active power." Compared to adding physical wind measurement equipment, wind speed inversion based on active power data has advantages such as low cost, wide coverage, and traceability, making it a research hotspot in recent years. Existing methods can be divided into physical mechanism methods and data-driven methods. Physical mechanism methods establish mapping models based on aerodynamics and wind turbine power curves, but the modeling is complex, parameters are difficult to obtain accurately, and accuracy is limited. Data-driven methods use machine learning models (such as RNN, LSTM, Transformer, etc.) to learn the time-series dependency relationship between power and wind speed, making them more suitable for complex nonlinear scenarios.
[0004] However, traditional sequence models (RNN, LSTM) suffer from gradient vanishing or insufficient memory capacity when processing long sequences. PatchTST reduces computational complexity and improves the ability to capture long-range dependencies by dividing the time series into "patch" subsequences, providing an effective tool for mining high-dimensional correlations between active power and wind speed. However, PatchTST itself cannot directly achieve regional wind speed inversion. On the other hand, the computational complexity of the Transformer model increases quadratically with the sequence length, while the recently proposed Mamba model achieves efficient selective state-space modeling of long sequences with linear complexity, accurately capturing dynamic temporal patterns and is suitable for constructing dynamic correlations between wind speed at the center point and wind speeds in various regions.
[0005] In summary, existing technologies still have shortcomings in temporal correlation mining, spatial feature fusion, and balancing inversion accuracy and efficiency. Therefore, combining the feature extraction advantages of PatchTST with the efficient sequence modeling capabilities of Mamba to construct an integrated method that takes into account both central point wind speed inversion and regional wind speed inversion is of great value for overcoming technical bottlenecks and improving the accuracy of wind speed sensing. Summary of the Invention
[0006] To address the shortcomings of existing wind farm wind speed sensing methods, such as insufficient physical equipment coverage, data gaps, and deficiencies in temporal correlation capture, spatial feature fusion, and the balance between inversion accuracy and efficiency, this invention aims to provide a wind farm active power-wind speed inversion method and system. By integrating the PatchTST architecture and the Mamba architecture, it achieves accurate inversion of wind speed at the center point and in the surrounding area of the wind farm, improving the precision and reliability of wind speed sensing, and providing support for the safe operation of wind farms, power output prediction, and efficient power system dispatch.
[0007] In a first aspect, the present invention provides a method for inverting active power and wind speed in a wind farm, comprising:
[0008] The system acquires the time-series data of total active power of wind farms and the time-series data of measured wind speed at the center point within a historical time window. Based on the PatchTST architecture, the time-series data of total active power of wind farms and the time-series data of measured wind speed at the center point are divided into multiple continuous subsequence segments. Each continuous subsequence segment is mapped into a segment-level feature vector by a Transformer encoder. All segment-level feature vectors are aggregated into a correlation feature vector of nonlinear coupling between active power and wind speed.
[0009] The associated feature vector is concatenated with the total active power value of the wind farm collected in real time at the current moment to form the inversion input feature vector. The center point wind speed inversion value is obtained by mapping through a linear regression head. The center point wind speed inversion value is then subjected to physical constraints using the wind farm equivalent power curve, and the time-series center point wind speed inversion sequence is output.
[0010] The wind speed inversion sequence at the center point is used as a dynamic driving signal and concatenated with the static spatial feature vectors of each spatial grid in the wind farm to form an input sequence. The input sequence is then fed into a selective state-space model based on the Mamba architecture, and the time-varying correlation feature vector that evolves over time is output.
[0011] The time-varying correlation feature vector is input into a linear regression head designed independently for each spatial grid, and the wind speed inversion time series is mapped to obtain the wind speed inversion time series. The measured power and the corresponding single-unit power curve of each spatial grid in the wind farm are used to physically constrain and correct the wind speed inversion time series, forming an inversion link from the center point wind speed to the regional wind speed.
[0012] Secondly, the present invention provides a wind farm active power-wind speed inversion system, comprising:
[0013] The acquisition module is configured to acquire time-series data of total active power of wind farms and time-series data of measured wind speed at the center point within a historical time window. Based on the PatchTST architecture, the time-series data of total active power of wind farms and the time-series data of measured wind speed at the center point are divided into multiple continuous subsequence segments. Each continuous subsequence segment is mapped into a segment-level feature vector by a Transformer encoder. All segment-level feature vectors are aggregated into a correlation feature vector of nonlinear coupling between active power and wind speed.
[0014] The mapping module is configured to concatenate the associated feature vector with the total active power value of the wind farm collected in real time at the current moment to form an inversion input feature vector, obtain the center point wind speed inversion value through linear regression head mapping, and apply physical constraints to the center point wind speed inversion value using the wind farm equivalent power curve, and output the center point wind speed inversion sequence in time series form.
[0015] The output module is configured to use the center point wind speed inversion sequence as a dynamic driving signal, concatenate it with the static spatial feature vectors of each spatial grid in the wind farm to form an input sequence, and send the input sequence into a selective state-space model based on the Mamba architecture to output a time-varying correlation feature vector that evolves over time.
[0016] The correction module is configured to input the time-varying correlation feature vector into a linear regression head designed independently for each spatial grid, map it to obtain the wind speed inversion time series, and use the measured power of each spatial grid in the wind farm and the corresponding single-unit power curve to physically constrain and correct the wind speed inversion time series, forming an inversion link from the center point wind speed to the regional wind speed.
[0017] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the wind farm active power-wind speed inversion method according to any embodiment of the present invention.
[0018] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the steps of the wind farm active power-wind speed inversion method according to any embodiment of the present invention.
[0019] The active power-wind speed inversion method and system for wind farms disclosed in this application, based on the PatchTST architecture, divides the time-series data of total active power and measured wind speed at the center point of the wind farm into multiple continuous subsequence segments. These segments are then aggregated by a Transformer encoder to obtain a correlation feature vector of nonlinear coupling between active power and wind speed. The correlation feature vector is concatenated with the current active power, mapped by a linear regression head, and subjected to physical constraints using the equivalent power curve of the wind farm to obtain the wind speed inversion sequence at the center point. The wind speed inversion sequence at the center point is then concatenated with static spatial features and fed into a selective state-space model based on the Mamba architecture to output a time-varying correlation feature vector. Finally, a linear regression head is designed for each spatial grid to obtain the regional wind speed inversion time series, and physical constraints are applied using the single-unit power curve to form a wind speed inversion link from the center point to the region. This achieves accurate inversion of wind speed at the center point and in the region of the wind farm, improving the refinement and reliability of wind speed sensing. Attached Figure Description
[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0021] Figure 1 A flowchart of a wind farm active power-wind speed inversion method provided in an embodiment of the present invention;
[0022] Figure 2 This is a structural block diagram of a wind farm active power-wind speed inversion system provided in an embodiment of the present invention;
[0023] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation
[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0025] Please see Figure 1 The diagram shows a flowchart of a wind farm active power-wind speed inversion method according to this application.
[0026] like Figure 1As shown, the active power-wind speed inversion method for wind farms specifically includes the following steps:
[0027] Step S101: Obtain the time series data of total active power of the wind farm and the time series data of measured wind speed at the center point within the historical time window. Based on the PatchTST architecture, the time series data of total active power of the wind farm and the time series data of measured wind speed at the center point are divided into multiple continuous subsequence segments. Each continuous subsequence segment is mapped into a segment-level feature vector by a Transformer encoder. All segment-level feature vectors are aggregated into a correlation feature vector of nonlinear coupling between active power and wind speed.
[0028] In this step, a PatchTST model is constructed, which includes a segmentation module, a Transformer encoder, an average pooling layer, and a linear reconstruction head. The PatchTST model is trained by minimizing the correlation reconstruction loss, where the expression for the correlation reconstruction loss is:
[0029] ,
[0030] In the formula, For the associated reconstruction loss, The total time step length of the subsequence segment. The active power adaptive weighting coefficient is dynamically generated based on the data state at time t. The measured value of the total active power of the wind farm at time t. Let t be the reconfigured value of the total active power of the wind farm. The wind speed adaptive weighting coefficient is dynamically generated based on the data state at time t. Let t be the measured wind speed at the center point of the wind farm. Let t be the reconstructed wind speed value at the center point of the wind farm. The reconstructed wind speed value at the center point of the wind farm at time t-1. The measured wind speed at the center point of the wind farm at time t-1 is... The loss function is Log-Cosh. The regularization coefficient is used.
[0031] Using time t as the end of the window, select a length of Construct a power time series vector within a time window. = and the time series vector of wind speed at the center point , For a moment Measured total active power of the wind farm For a moment Measured total active power of the wind farm For a moment Measured wind speed at the center point of the wind farm. For a moment Measured wind speed at the center point of the wind farm. The transpose symbol is used, and the power time series vector is concatenated with the center point wind speed time series vector to form a two-dimensional time series matrix. , It is the set of real numbers;
[0032] The two-dimensional time series matrix Input the PatchTST model and perform the following operations in sequence:
[0033] Using the segmentation module, according to the length of... The size of the continuous subsequence segments will affect the two-dimensional time series matrix. Divided into 3 consecutive subsequence segments, of which Let be the sliding step size, and let the m-th continuous subsequence segment be represented as: ;
[0034] The Transformer encoder maps each segment to a segment-level hidden feature vector. = ,in, For Transformer encoder functions, For Transformer encoder parameters, For the m-th continuous subsequence segment, , To hide the feature dimension, It is the set of real numbers;
[0035] The average pooling layer is used to perform average pooling on all segment-level feature vectors to obtain the associated feature vector of the active power-wind speed nonlinear coupling. = , This represents the total number of continuous subsequence segments.
[0036] Step S102: The associated feature vector is concatenated with the total active power value of the wind farm collected in real time at the current moment to form an inversion input feature vector. The center point wind speed inversion value is obtained by mapping through a linear regression head. The center point wind speed inversion value is then subjected to physical constraints using the wind farm equivalent power curve, and a time-series center point wind speed inversion sequence is output.
[0037] In this step, a center point wind speed inversion model is constructed. This model includes a feature stitching layer and a linear regression head. The center point wind speed inversion model is trained by minimizing the center point wind speed inversion loss. The expression is:
[0038] ,
[0039] In the formula, The total time step length of the subsequence segment. The measured value of the total active power of the wind farm at time t. , These are the adaptive weighting coefficients for the wind speed inversion term at time step t and the adaptive weighting coefficients for the power physical constraint term at time step t, respectively. Let t be the measured wind speed at the center point of the wind farm. The wind speed inversion value at the center point output by the linear regression head. For scale parameters, The penalty coefficient is... This is the center point power inversion value output by the linear regression head. The first difference of the true wind speed at time step t. To retrieve the first difference of wind speed at time step t, This is a distance function used to measure measured wind speed. With inversion wind speed The differences between them;
[0040] Obtain the measured value of the total active power of the wind farm at the current moment, as well as the associated feature vector;
[0041] Input the associated feature vector and the measured value of the total active power of the wind farm at the current moment into the center point wind speed inversion model, and perform the following operations in sequence:
[0042] The inverted input feature vector is constructed through the feature concatenation layer. , To hide the feature dimension, Let be the set of real numbers. It is the transpose symbol. For associated feature vectors, For time t, the input feature vector or state variable;
[0043] The linear regression head maps the inverted input feature vector to the center point wind speed inversion value. ,in, The weight vector of the center point wind speed inversion model. This is the bias term for the center point wind speed inversion model;
[0044] Output the center point wind speed inversion value The wind speed inversion values at each center point were sorted according to time sequence to obtain a time-series center point wind speed inversion sequence. .
[0045] Step S103: The center point wind speed inversion sequence is used as a dynamic driving signal and concatenated with the static spatial feature vectors of each spatial grid in the wind farm to form an input sequence. The input sequence is then fed into a selective state-space model based on the Mamba architecture to output a time-varying correlation feature vector that evolves over time.
[0046] In this step, a selective state space model based on the Mamba architecture is constructed. The selective state space model includes a feature splicing layer and a Mamba selective state space layer.
[0047] Obtain the center point wind speed inversion sequence And the static spatial feature vector of the i-th turbine location or grid within the wind farm. ,in, Includes coordinates, altitude, height, and terrain roughness. For state space dimension;
[0048] The center point wind speed inversion sequence and the static spatial feature vector are input into the selective state-space model, and the following operations are performed sequentially:
[0049] The input feature vector is constructed at time t through the feature concatenation layer. ;
[0050] The hidden state is updated and associated features are output via the Mamba selective state space layer in the following manner:
[0051] ,
[0052] ,
[0053] In the formula, Let be the hidden state vector at time t. Let be the hidden state vector at time t-1. To output the associated feature vector, This is a matrix dynamically generated from the input by Mamba's selectivity parameters;
[0054] Output the associated feature vector , recorded as , serving as the time-varying correlation feature vector between the center point and the wind speed in the i-th region, where, For channel dimensions.
[0055] Step S104: Input the time-varying correlation feature vector into the linear regression head designed independently for each spatial grid, map to obtain the wind speed inversion time series, and use the measured power of each spatial grid in the wind farm and the corresponding single-unit power curve to physically constrain and correct the wind speed inversion time series, forming an inversion link from the center point wind speed to the regional wind speed.
[0056] In this step, a regional wind speed inversion model is constructed. This model includes a linear regression head and a power curve constraint layer designed independently for each emplacement or grid. The regional wind speed inversion model minimizes the regional wind speed inversion loss. The training yielded, among other things, regional wind speed inversion losses. The expression is:
[0057] ,
[0058] In the formula, The total number of spatial grids, Let be the actual wind speed value at the i-th position. The region wind speed inversion value output by the linear regression head. Let be the measured value of the single active power of the i-th unit. To utilize the power curve function of the i-th unit The obtained single-unit power estimate, This is the single-unit power error weighting coefficient;
[0059] Furthermore, the regional wind speed inversion model, the PatchTST model, and the center point wind speed inversion model minimize the end-to-end joint loss. Joint training was conducted, including:
[0060] ,
[0061] In the formula, These are the weighting coefficients for the associated reconstruction loss, the weighting coefficients for the center point wind speed inversion loss, and the weighting coefficients for the regional wind speed inversion loss, respectively.
[0062] Obtain the time-varying correlation feature vector and the measured value of the single active power of the i-th unit. ;
[0063] The time-varying correlation feature vector is input into the regional wind speed inversion model, and the regional wind speed inversion value is obtained by mapping through the linear regression head of the i-th camera position. ,in, Let be the regression mapping weight matrix for the i-th region. Let be the regression mapping bias vector for the i-th region;
[0064] Output the regional wind speed inversion values of each spatial grid, and sort the regional wind speed inversion values of each spatial grid in the same spatial grid according to time sequence to obtain the wind speed inversion time sequence of each spatial grid in the wind farm area, forming an inversion link from the center point wind speed to the regional wind speed.
[0065] In summary, the method of this application, based on the PatchTST architecture, divides the time-series data of total active power of wind farms and the time-series data of measured wind speed at the center point within a historical time window into multiple continuous subsequence segments. These segments are then aggregated by a Transformer encoder to obtain a correlation feature vector of nonlinear coupling between active power and wind speed. The correlation feature vector is concatenated with the current active power, mapped by a linear regression head, and physical constraints are applied using the equivalent power curve of the wind farm to obtain the wind speed inversion sequence at the center point. The wind speed inversion sequence at the center point is then concatenated with static spatial features and fed into a selective state-space model based on the Mamba architecture to output a time-varying correlation feature vector. Finally, a linear regression head is designed for each spatial grid to obtain the regional wind speed inversion time series, and physical constraints are applied using the single-unit power curve to form a wind speed inversion link from the center point to the region. This method achieves accurate inversion of wind speed at the center point and in the region of the wind farm, improving the refinement and reliability of wind speed sensing.
[0066] Please see Figure 2 The diagram shows a structural block diagram of a wind farm active power-wind speed inversion system according to this application.
[0067] like Figure 2 As shown, the active power-wind speed inversion system 200 for wind farms includes an acquisition module 210, a mapping module 220, an output module 230, and a correction module 240.
[0068] The acquisition module 210 is configured to acquire time-series data of total active power and measured wind speed at the center point of the wind farm within a historical time window. Based on the PatchTST architecture, the time-series data of total active power and measured wind speed at the center point are divided into multiple continuous subsequence segments. Each continuous subsequence segment is mapped into a segment-level feature vector by a Transformer encoder. All segment-level feature vectors are aggregated into a correlation feature vector that is nonlinearly coupled with active power and wind speed. The mapping module 220 is configured to concatenate the correlation feature vector with the real-time acquired total active power value of the wind farm to form an inversion input feature vector. The inversion value of the wind speed at the center point is obtained by mapping with a linear regression head, and the wind farm equivalent power curve is used to map the... Physical constraints are applied to the center point wind speed inversion values, and a time-series center point wind speed inversion sequence is output. Output module 230 is configured to use the center point wind speed inversion sequence as a dynamic driving signal, concatenate it with the static spatial feature vectors of each spatial grid in the wind farm to form an input sequence, and send the input sequence into a selective state-space model based on the Mamba architecture to output a time-varying correlation feature vector that evolves over time. Correction module 240 is configured to input the time-varying correlation feature vector into a linear regression head designed independently for each spatial grid, map it to obtain the wind speed inversion time series, and use the measured power and corresponding single-unit power curves of each spatial grid in the wind farm to physically constrain and correct the wind speed inversion time series, forming an inversion link from center point wind speed to regional wind speed.
[0069] It should be understood that Figure 2 The modules and references described in the document Figure 1 The steps described in the text correspond to those in the method described above. Therefore, the operations, features, and corresponding technical effects described above also apply to the method described in the text. Figure 2 The various modules in the document will not be described in detail here.
[0070] In other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the wind farm active power-wind speed inversion method in any of the above method embodiments.
[0071] In one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, which are configured as follows:
[0072] The system acquires the time-series data of total active power of wind farms and the time-series data of measured wind speed at the center point within a historical time window. Based on the PatchTST architecture, the time-series data of total active power of wind farms and the time-series data of measured wind speed at the center point are divided into multiple continuous subsequence segments. Each continuous subsequence segment is mapped into a segment-level feature vector by a Transformer encoder. All segment-level feature vectors are aggregated into a correlation feature vector of nonlinear coupling between active power and wind speed.
[0073] The associated feature vector is concatenated with the total active power value of the wind farm collected in real time at the current moment to form the inversion input feature vector. The center point wind speed inversion value is obtained by mapping through a linear regression head. The center point wind speed inversion value is then subjected to physical constraints using the wind farm equivalent power curve, and the time-series center point wind speed inversion sequence is output.
[0074] The wind speed inversion sequence at the center point is used as a dynamic driving signal and concatenated with the static spatial feature vectors of each spatial grid in the wind farm to form an input sequence. The input sequence is then fed into a selective state-space model based on the Mamba architecture, and the time-varying correlation feature vector that evolves over time is output.
[0075] The time-varying correlation feature vector is input into a linear regression head designed independently for each spatial grid, and the wind speed inversion time series is mapped to obtain the wind speed inversion time series. The measured power and the corresponding single-unit power curve of each spatial grid in the wind farm are used to physically constrain and correct the wind speed inversion time series, forming an inversion link from the center point wind speed to the regional wind speed.
[0076] Computer-readable storage media may include a stored program area and a stored data area, wherein the stored program area may store an operating system and an application program required for at least one function; the stored data area may store data created based on the use of the wind farm active power-wind speed inversion system, etc. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely disposed relative to a processor, which can be connected to the wind farm active power-wind speed inversion system via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.
[0077] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention, such as... Figure 3 As shown, the device includes a processor 310 and a memory 320. The electronic device may also include an input device 330 and an output device 340. The processor 310, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 3 Taking a bus connection as an example, the memory 320 is the computer-readable storage medium described above. The processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 320, thereby implementing the wind farm active power-wind speed inversion method described in the above embodiment. The input device 330 can receive input digital or character information and generate key signal inputs related to user settings and function control of the wind farm active power-wind speed inversion system. The output device 340 may include a display screen or other display device.
[0078] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.
[0079] In one implementation, the above-described electronic device is applied to a wind farm active power-wind speed inversion system for a client application, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:
[0080] The system acquires the time-series data of total active power of wind farms and the time-series data of measured wind speed at the center point within a historical time window. Based on the PatchTST architecture, the time-series data of total active power of wind farms and the time-series data of measured wind speed at the center point are divided into multiple continuous subsequence segments. Each continuous subsequence segment is mapped into a segment-level feature vector by a Transformer encoder. All segment-level feature vectors are aggregated into a correlation feature vector of nonlinear coupling between active power and wind speed.
[0081] The associated feature vector is concatenated with the total active power value of the wind farm collected in real time at the current moment to form the inversion input feature vector. The center point wind speed inversion value is obtained by mapping through a linear regression head. The center point wind speed inversion value is then subjected to physical constraints using the wind farm equivalent power curve, and the time-series center point wind speed inversion sequence is output.
[0082] The wind speed inversion sequence at the center point is used as a dynamic driving signal and concatenated with the static spatial feature vectors of each spatial grid in the wind farm to form an input sequence. The input sequence is then fed into a selective state-space model based on the Mamba architecture, and the time-varying correlation feature vector that evolves over time is output.
[0083] The time-varying correlation feature vector is input into a linear regression head designed independently for each spatial grid, and the wind speed inversion time series is mapped to obtain the wind speed inversion time series. The measured power and the corresponding single-unit power curve of each spatial grid in the wind farm are used to physically constrain and correct the wind speed inversion time series, forming an inversion link from the center point wind speed to the regional wind speed.
[0084] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.
[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for inverting active power-wind speed in a wind farm, characterized in that, include: The system acquires the time-series data of total active power of wind farms and the time-series data of measured wind speed at the center point within a historical time window. Based on the PatchTST architecture, the time-series data of total active power of wind farms and the time-series data of measured wind speed at the center point are divided into multiple continuous subsequence segments. Each continuous subsequence segment is mapped into a segment-level feature vector by a Transformer encoder. All segment-level feature vectors are aggregated into a correlation feature vector of nonlinear coupling between active power and wind speed. The associated feature vector is concatenated with the total active power value of the wind farm collected in real time at the current moment to form the inversion input feature vector. The center point wind speed inversion value is obtained by mapping through a linear regression head. The center point wind speed inversion value is then subjected to physical constraints using the wind farm equivalent power curve, and the time-series center point wind speed inversion sequence is output. The wind speed inversion sequence at the center point is used as a dynamic driving signal and concatenated with the static spatial feature vectors of each spatial grid in the wind farm to form an input sequence. The input sequence is then fed into a selective state-space model based on the Mamba architecture, and the time-varying correlation feature vector that evolves over time is output. The time-varying correlation feature vector is input into a linear regression head designed independently for each spatial grid, and the wind speed inversion time series is mapped to obtain the wind speed inversion time series. The measured power and the corresponding single-unit power curve of each spatial grid in the wind farm are used to physically constrain and correct the wind speed inversion time series, forming an inversion link from the center point wind speed to the regional wind speed.
2. The active power-wind speed inversion method for wind farms according to claim 1, characterized in that, The PatchTST architecture divides the active power time-series data and the center point wind speed time-series data into multiple continuous subsequence segments. Each continuous subsequence segment is mapped to a segment-level feature vector by a Transformer encoder. All segment-level feature vectors are aggregated into a correlation feature vector for nonlinear coupling of active power and wind speed, including: A PatchTST model is constructed, comprising a segmentation module, a Transformer encoder, an average pooling layer, and a linear reconstruction head. The PatchTST model is trained by minimizing the correlation reconstruction loss, where the expression for the correlation reconstruction loss is: ， In the formula, For the associated reconstruction loss, The total time step length of the subsequence segment. The active power adaptive weighting coefficient is dynamically generated based on the data state at time t. The measured value of the total active power of the wind farm at time t. Let t be the reconfigured value of the total active power of the wind farm. The wind speed adaptive weighting coefficient is dynamically generated based on the data state at time t. Let t be the measured wind speed at the center point of the wind farm. Let t be the reconstructed wind speed value at the center point of the wind farm. The reconstructed wind speed value at the center point of the wind farm at time t-1. The measured wind speed at the center point of the wind farm at time t-1 is... The loss function is Log-Cosh. The regularization coefficient is used. Using time t as the end of the window, select a length of Construct a power time series vector within a time window. = and the time series vector of wind speed at the center point , For a moment Measured total active power of the wind farm For a moment Measured total active power of the wind farm For a moment Measured wind speed at the center point of the wind farm. For a moment Measured wind speed at the center point of the wind farm. The transpose symbol is used, and the power time series vector is concatenated with the center point wind speed time series vector to form a two-dimensional time series matrix. , It is the set of real numbers; The two-dimensional time series matrix Input the PatchTST model and perform the following operations in sequence: Using the segmentation module, according to the length of... The size of the continuous subsequence segments will affect the two-dimensional time series matrix. Divided into 3 consecutive subsequence segments, of which Let be the sliding step size, and let the m-th continuous subsequence segment be represented as: ; The Transformer encoder maps each segment to a segment-level hidden feature vector. = ,in, For Transformer encoder functions, For Transformer encoder parameters, For the m-th continuous subsequence segment, , To hide the feature dimension, It is the set of real numbers; The average pooling layer is used to perform average pooling on all segment-level feature vectors to obtain the associated feature vector of the active power-wind speed nonlinear coupling. = , This represents the total number of continuous subsequence segments.
3. The active power-wind speed inversion method for wind farms according to claim 2, characterized in that, The process involves concatenating the associated feature vector with the real-time collected total active power value of the wind farm to form an inversion input feature vector. This vector is then mapped using a linear regression head to obtain the center point wind speed inversion value. Physical constraints are then applied to the center point wind speed inversion value using the wind farm's equivalent power curve. The output time-series center point wind speed inversion sequence includes: A center-point wind speed inversion model is constructed, comprising a feature stitching layer, a linear regression head, and a power curve constraint layer. This model is trained by minimizing the center-point wind speed inversion loss. The expression is: , In the formula, The total time step length of the subsequence segment. The measured value of the total active power of the wind farm at time t. , These are the adaptive weighting coefficients for the wind speed inversion term at time step t and the adaptive weighting coefficients for the power physical constraint term at time step t, respectively. Let t be the measured wind speed at the center point of the wind farm. The wind speed inversion value at the center point output by the linear regression head. For scale parameters, The penalty coefficient is... This is the center point power inversion value output by the linear regression head. The first difference of the true wind speed at time step t. To retrieve the first difference of wind speed at time step t, This is a distance function used to measure measured wind speed. With inversion wind speed The differences between them; Obtain the measured value of the total active power of the wind farm at the current moment, as well as the associated feature vector; Input the associated feature vector and the measured value of the total active power of the wind farm at the current moment into the center point wind speed inversion model, and perform the following operations in sequence: The inverted input feature vector is constructed through the feature concatenation layer. , To hide the feature dimension, Let be the set of real numbers. It is the transpose symbol. For associated feature vectors, For time t, the input feature vector or state variable; The linear regression head maps the inverted input feature vector to the center point wind speed inversion value. ,in, The weight vector of the center point wind speed inversion model. This is the bias term for the center point wind speed inversion model; Output the center point wind speed inversion value The wind speed inversion values at each center point were sorted according to time sequence to obtain a time-series center point wind speed inversion sequence. .
4. The active power-wind speed inversion method for a wind farm according to claim 3, characterized in that, The process of using the center point wind speed inversion sequence as a dynamic driving signal and concatenating it with the static spatial feature vectors of each spatial grid in the wind farm to form an input sequence, and then feeding the input sequence into a selective state-space model based on the Mamba architecture, outputting a time-varying correlation feature vector that evolves over time, includes: A selective state space model based on the Mamba architecture is constructed, which includes a feature splicing layer and a Mamba selective state space layer. Obtain the center point wind speed inversion sequence And the static spatial feature vector of the i-th turbine location or grid within the wind farm. ,in, Includes coordinates, altitude, height, and terrain roughness. For state space dimension; The center point wind speed inversion sequence and the static spatial feature vector are input into the selective state-space model, and the following operations are performed sequentially: The input feature vector is constructed at time t through the feature concatenation layer. ; The hidden state is updated and associated features are output via the Mamba selective state space layer in the following manner: ， ， In the formula, Let be the hidden state vector at time t. Let be the hidden state vector at time t-1. To output the associated feature vector, This is a matrix dynamically generated from the input by Mamba's selectivity parameters; Output the associated feature vector , recorded as , serving as the time-varying correlation feature vector between the center point and the wind speed in the i-th region, where, For channel dimensions.
5. The active power-wind speed inversion method for a wind farm according to claim 4, characterized in that, The process of inputting the time-varying correlation feature vector into a linear regression head designed independently for each spatial grid, mapping it to obtain the wind speed inversion time series, and using the measured power and corresponding single-unit power curves of each spatial grid in the wind farm to physically constrain and correct the wind speed inversion time series, forming an inversion link from the center point wind speed to the regional wind speed includes: A regional wind speed inversion model is constructed, comprising a linear regression head and a power curve constraint layer designed independently for each machine site or grid. The regional wind speed inversion model minimizes the regional wind speed inversion loss. The training yielded, among other things, regional wind speed inversion losses. The expression is: , In the formula, The total number of spatial grids, Let be the actual wind speed value at the i-th position. The region wind speed inversion value output by the linear regression head. Let be the measured value of the single active power of the i-th unit. To utilize the power curve function of the i-th unit The obtained single-unit power estimate, This is the single-unit power error weighting coefficient; Furthermore, the regional wind speed inversion model, the PatchTST model, and the center point wind speed inversion model minimize the end-to-end joint loss. Joint training was conducted, including: ， In the formula, These are the weighting coefficients for the associated reconstruction loss, the weighting coefficients for the center point wind speed inversion loss, and the weighting coefficients for the regional wind speed inversion loss, respectively. Obtain the time-varying correlation feature vector and the measured value of the single active power of the i-th unit. ; The time-varying correlation feature vector is input into the regional wind speed inversion model, and the regional wind speed inversion value is obtained by mapping through the linear regression head of the i-th camera position. ,in, Let be the regression mapping weight matrix for the i-th region. Let be the regression mapping bias vector for the i-th region; Output the regional wind speed inversion values of each spatial grid, and sort the regional wind speed inversion values of each spatial grid in the same spatial grid according to time sequence to obtain the wind speed inversion time sequence of each spatial grid in the wind farm area, forming an inversion link from the center point wind speed to the regional wind speed.
6. A wind farm active power-wind speed inversion system, characterized in that, include: The acquisition module is configured to acquire time-series data of total active power of wind farms and time-series data of measured wind speed at the center point within a historical time window. Based on the PatchTST architecture, the time-series data of total active power of wind farms and the time-series data of measured wind speed at the center point are divided into multiple continuous subsequence segments. Each continuous subsequence segment is mapped into a segment-level feature vector by a Transformer encoder. All segment-level feature vectors are aggregated into a correlation feature vector of nonlinear coupling between active power and wind speed. The mapping module is configured to concatenate the associated feature vector with the total active power value of the wind farm collected in real time at the current moment to form an inversion input feature vector, obtain the center point wind speed inversion value through linear regression head mapping, and apply physical constraints to the center point wind speed inversion value using the wind farm equivalent power curve, and output the center point wind speed inversion sequence in time series form. The output module is configured to use the center point wind speed inversion sequence as a dynamic driving signal, concatenate it with the static spatial feature vectors of each spatial grid in the wind farm to form an input sequence, and send the input sequence into a selective state-space model based on the Mamba architecture to output a time-varying correlation feature vector that evolves over time. The correction module is configured to input the time-varying correlation feature vector into a linear regression head designed independently for each spatial grid, map it to obtain the wind speed inversion time series, and use the measured power of each spatial grid in the wind farm and the corresponding single-unit power curve to physically constrain and correct the wind speed inversion time series, forming an inversion link from the center point wind speed to the regional wind speed.
7. An electronic device, characterized in that, include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 5.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method described in any one of claims 1 to 5.