基于全域扫描雷达的风电场集群调度与控制方法及系统

Abstract

本申请提供一种基于全域扫描雷达的风电场集群调度与控制方法及系统，涉及风电调度技术领域，其中，该方法包括：通过部署在风电场集群设定位置的全域扫描雷达站对目标区域进行扫描，获取三维风场数据；利用时空编码器处理三维风场数据，生成四维时空图谱；将四维时空图谱输入时空预测模型进行预测，输出预测功率数据；基于预测功率数据，结合电网调度指令，采用分层优化算法进行协同优化，生成协同调度指令；根据协同调度指令，控制风电场集群中各风电机组的发电功率和运行姿态。本申请提升了风电场集群的调度响应速度与控制精度。

Description

Technical Field

[0001] This application relates to the field of wind power dispatching technology, and in particular to a method and system for wind farm cluster dispatching and control based on full-domain scanning radar. Background Technology

[0002] With the rapid development of renewable energy, large-scale wind farm clusters have become an important part of the power system. The precise scheduling and control of wind farms is the key to ensuring the stable operation of the power grid and improving energy absorption capacity, and has broad application prospects.

[0003] In existing technologies, existing solutions employ scanning radar equipment deployed within wind farms to monitor wind resource conditions. For example, after acquiring wind farm data through lidar, the output power of the wind farm is predicted based on this data, and then a scheduling plan is formulated based on the prediction results. Such methods improve the ability to perceive changes in wind energy to a certain extent.

[0004] However, the existing solutions mentioned above do not adequately explore the inherent spatiotemporal evolution patterns of wind farm monitoring data when using it for dispatching decisions. This results in limited matching between the generated dispatching instructions and the actual dynamic changes of the wind farm. Furthermore, the coordination and foresight of dispatching strategies need to be strengthened when responding to complex grid regulation needs. Therefore, the existing technology suffers from technical problems such as insufficient accuracy and coordination of wind farm cluster dispatching instructions. Summary of the Invention

[0005] This application provides a method and system for scheduling and controlling wind farm clusters based on full-domain scanning radar, in order to solve the problems of slow scheduling response speed and low control accuracy of wind farm clusters in the prior art.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides a wind farm cluster scheduling and control method based on full-domain scanning radar, comprising:

[0007] Three-dimensional wind field data is obtained by scanning the target area with full-area scanning radar stations deployed at designated locations in the wind farm cluster.

[0008] The three-dimensional wind field data is processed using a spatiotemporal encoder to generate a four-dimensional spatiotemporal map;

[0009] The four-dimensional spatiotemporal map is input into the spatiotemporal prediction model for prediction, and the predicted power data is output.

[0010] Based on the predicted power data and combined with power grid dispatch instructions, a hierarchical optimization algorithm is used for collaborative optimization to generate collaborative dispatch instructions.

[0011] According to the coordinated scheduling instructions, the power generation and operating status of each wind turbine in the wind farm cluster are controlled.

[0012] Optionally, the step of generating collaborative dispatch instructions by combining the predicted power data with grid dispatch instructions using a hierarchical optimization algorithm includes:

[0013] Power change trend information and spatial distribution information are extracted from the predicted power data;

[0014] The power change trend information and the power grid dispatching instructions are input into the first-level optimization module of the hierarchical optimization algorithm. The first-level optimization module determines the total power target value of the wind farm cluster at each time point in the future period based on the power adjustment rate requirements and frequency adjustment requirements of the power grid and the power change trend information.

[0015] The total power target value and the spatial distribution information are input to the second-level optimization module of the hierarchical optimization algorithm. The second-level optimization module decomposes the total power target value at each time point into the power generation command of each wind turbine according to the current power output value, location information and spatial distribution information of each wind turbine.

[0016] The power generation command is input to the third-level optimization module of the hierarchical optimization algorithm. The third-level optimization module determines the attitude adjustment command corresponding to each power generation command based on the operating status of each wind turbine.

[0017] The power generation commands and attitude adjustment commands at each time point are combined to form a coordinated scheduling command.

[0018] Optionally, determining the total power target value of the wind farm cluster at each time point in the future period based on the power adjustment rate requirements and frequency regulation requirements of the power grid, combined with the power change trend information, includes:

[0019] Extract predicted power values ​​for multiple time points within the future period from the power change trend information;

[0020] Establish a first function and a second function, and combine the first function and the second function into an objective function, wherein the first function is used to minimize the difference between the total power target value and the predicted power value, and the second function is used to minimize the degree of change of the total power target value between adjacent time points;

[0021] The objective function is solved by using the power adjustment rate requirement and the frequency adjustment requirement as optimization constraints.

[0022] Based on the solution results, the total power target value at each time point is determined.

[0023] Secondly, this application provides a wind farm cluster scheduling and control system based on full-domain scanning radar, including:

[0024] The acquisition module is used to scan the target area by deploying a full-area scanning radar station at a set location in the wind farm cluster to acquire three-dimensional wind field data;

[0025] The processing module is used to process the three-dimensional wind field data using a spatiotemporal encoder to generate a four-dimensional spatiotemporal map.

[0026] The input module is used to input the four-dimensional spatiotemporal map into the spatiotemporal prediction model for prediction and output predicted power data.

[0027] The optimization module is used to perform collaborative optimization based on the predicted power data and in conjunction with power grid dispatch instructions, using a hierarchical optimization algorithm to generate collaborative dispatch instructions.

[0028] The control module is used to control the power generation and operating status of each wind turbine in the wind farm cluster according to the coordinated scheduling instructions.

[0029] Thirdly, this application provides an electronic device, comprising:

[0030] Memory, used to store computer programs;

[0031] A processor is configured to execute the computer program to implement the steps of the wind farm cluster scheduling and control method based on full-domain scanning radar as described in the first aspect above.

[0032] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the wind farm cluster scheduling and control method based on full-domain scanning radar as described in the first aspect above.

[0033] The technical solution provided in this application has the following beneficial effects:

[0034] This application achieves comprehensive and synchronous perception of wind resource conditions in a wind farm cluster area by acquiring three-dimensional wind field data through a full-domain scanning radar station. Then, a four-dimensional spatiotemporal map is generated using a spatiotemporal encoder to effectively represent the continuous evolution characteristics of the wind field in both time and space. Next, the four-dimensional spatiotemporal map is input into a prediction model for power prediction, resulting in more accurate power data reflecting future wind energy changes. Based on this predicted power data and fused with grid dispatch instructions, a hierarchical optimization algorithm is used for collaborative optimization, enabling the generated dispatch instructions to simultaneously meet grid regulation needs and wind farm dynamic characteristics. Finally, the wind turbines are controlled according to the generated collaborative dispatch instructions, achieving coordinated adjustment of power generation and operating attitude, improving the overall cluster's response consistency and dispatch response speed, and thus enhancing control accuracy.

[0035] Furthermore, this application also extracts trend and distribution information from the predicted power data, firstly determines the total power target of the cluster by combining it with the grid command, then decomposes the total target into power commands for each unit according to the wind turbine status and spatial distribution, then generates corresponding attitude adjustment commands according to the unit operating status, and finally integrates them into a complete coordinated scheduling command.

[0036] Furthermore, this process combines the macro-level dispatching needs of the power grid with the micro-level operating status of wind turbines through a hierarchical and progressive optimization approach. This ensures that the final generated instructions can accurately match the total power target while also taking into account the individual differences and spatial location effects of each wind turbine, thereby improving the rationality and executability of the dispatching instructions.

[0037] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 A flowchart illustrating a wind farm cluster scheduling and control method based on full-domain scanning radar, provided as an embodiment of this application;

[0040] Figure 2 A schematic diagram illustrating a specific implementation of a wind farm cluster scheduling and control method based on full-domain scanning radar, provided in an embodiment of this application;

[0041] Figure 3 This is a schematic diagram of the structure of a wind farm cluster scheduling and control system based on a full-domain scanning radar, provided in an embodiment of this application. Detailed Implementation

[0042] To address the issues of insufficient accuracy and coordination in wind farm cluster dispatching commands, existing solutions do not delve deeply enough into the spatiotemporal evolution patterns contained in radar wind farm data when utilizing it. Furthermore, the overall coordination between the real-time adjustment needs of the power grid and the individual operating status of wind turbines needs to be strengthened when formulating dispatching strategies. This limits the degree to which dispatching commands match the actual dynamics of the wind farm and the complex operating conditions of the power grid.

[0043] To address the aforementioned shortcomings, this application proposes a wind farm cluster scheduling and control method based on full-domain scanning radar. The core idea of ​​this method is as follows: First, three-dimensional wind field data is acquired via radar, and a four-dimensional map representing the continuous spatiotemporal changes of the wind field is constructed using encoding technology. Then, power prediction is performed based on this map to obtain more accurate information on future wind energy changes. Next, the prediction results are combined with grid dispatch commands, and dispatch commands are collaboratively generated through hierarchical optimization technology. Finally, the power output and operating attitude of each wind turbine within the cluster are uniformly controlled according to these commands. Therefore, this scheme, by deeply analyzing the spatiotemporal evolution characteristics of the wind farm and adopting a hierarchical collaborative optimization mechanism, enables the generated dispatch commands to more accurately match the dynamic changes of the wind farm, while effectively coordinating the macro-level demands of the grid and the micro-level states of the wind turbines, thus solving the problems of insufficient accuracy and coordination of dispatch commands in existing technologies.

[0044] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0045] The core of this application is to provide a method for scheduling and controlling wind farm clusters based on full-domain scanning radar. A flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes:

[0046] Step 101: Scan the target area using a full-area scanning radar station deployed at a designated location within the wind farm cluster to obtain three-dimensional wind field data.

[0047] In step 101, the full-domain scanning radar station is a monitoring device deployed at a selected high point within the wind farm cluster. The full-domain scanning radar station detects the movement of the wind field by emitting and receiving laser beams. The three-dimensional wind field data refers to the data set obtained through radar scanning that reflects the wind speed and wind direction information at different locations and heights within the target area. It includes the distribution status of the wind field in three spatial dimensions.

[0048] In this embodiment of the application, the target area to be monitored is continuously scanned by a full-area scanning radar station pre-installed at a key location in the wind farm cluster. The detection signal emitted by the radar station can cover the entire cluster range, thereby simultaneously collecting the original wind speed and wind direction information of multiple spatial points in the target area. After preliminary signal processing, the three-dimensional wind field data for subsequent analysis is finally obtained.

[0049] Step 102: Process the three-dimensional wind field data using a spatiotemporal encoder to generate a four-dimensional spatiotemporal map.

[0050] Among them, the four-dimensional spatiotemporal map is a structured data representation that includes the dynamic trend of wind field changes over time and the structural relationship of its spatial distribution.

[0051] The spatiotemporal encoder consists of a first network layer and a second network layer connected in series. The first network layer specifically adopts a bidirectional long short-term memory network structure to receive time series data points of individual spatial grid cells and output their encoded time feature vectors. The second network layer specifically adopts a graph attention network structure. This structure uses the time feature vectors of each cell output by the first network layer as node features and constructs edges based on the adjacency relationship of the spatial grid. It calculates the interaction weights between nodes and aggregates neighborhood information through a multi-attention head mechanism, thereby outputting a spatial structure information vector that integrates spatiotemporal context.

[0052] The training process of this encoder is as follows: historical three-dimensional wind field data and corresponding standard four-dimensional spatiotemporal maps are used as training samples. After obtaining the predicted map through the encoder forward propagation, the mean square error between the predicted map and the standard map is calculated as the loss function. The backpropagation algorithm combined with the gradient descent method is used to iteratively update all parameters in the two-layer network until the loss converges to the preset threshold.

[0053] It should be noted that the above structure is exemplary. This application does not impose specific limitations on the layer structure design, parameter design, etc. adopted in the internal structure of the spatiotemporal encoder. Corresponding settings can be made according to the actual situation.

[0054] In this embodiment, step 102 includes the following process:

[0055] Step 1021: Divide the three-dimensional wind field data into multiple spatial grid units.

[0056] In step 1021, a spatial grid cell refers to a basic analytical unit with a fixed coordinate range obtained by dividing the entire wind farm area according to preset rules.

[0057] In this embodiment of the application, the acquired three-dimensional wind field data is processed, and the entire target area is divided into several grids of regular size according to its geographical location coordinate information. Each grid covers a fixed geographical range, and each observation point in the wind field data belongs to a specific spatial grid unit.

[0058] Step 1022: In each spatial grid cell, identify multiple time-series data points associated with the spatial grid cell.

[0059] In step 1022, time series data points refer to a set of data records arranged in chronological order within the same spatial grid cell, with each data point recording the wind field state of the grid at a specific moment.

[0060] In this embodiment of the application, for each spatial grid cell obtained by division, all observation point data falling within the range of the cell are found and extracted from the three-dimensional wind field data, and then arranged according to the time order in which these observation points were collected, thereby constructing a series of data points arranged in chronological order for each spatial grid cell.

[0061] Step 1023: Analyze the correlation between multiple time series data points within the same spatial grid cell through the first network layer in the spatiotemporal encoder to extract trend information reflecting changes in wind field intensity.

[0062] In step 1023, trend information refers to data used to describe the overall change pattern of parameters such as wind speed and wind direction within a spatial grid cell over time.

[0063] In this embodiment, the time series data points corresponding to each spatial grid cell are input into the first network layer inside the spatiotemporal encoder. The network layer uses a long short-term memory network structure to analyze the sequential dependencies between these continuous data points, thereby capturing and outputting the evolution trend of wind field intensity within each grid cell over time, and generating trend information.

[0064] Step 1024: Analyze the correlation between trend information of adjacent spatial grid cells through the second network layer in the spatiotemporal encoder to extract spatial structure information reflecting changes in wind field morphology.

[0065] In step 1024, spatial structure information refers to data used to describe the mutual influence and correlation between wind field trends of different spatial grid units, reflecting the spatial morphology of the wind field.

[0066] In this embodiment of the application, the trend information of each spatial grid cell obtained in step 1023 is input into the second network layer inside the spatiotemporal encoder. The second network layer adopts a graph convolutional network structure, constructs connections based on the adjacency relationship of each grid cell in physical space, and analyzes the interaction between the trend information of these adjacent cells, thereby extracting spatial structure information that can reflect the propagation and change of the wind field in space.

[0067] Step 1025: Integrate the trend information of each spatial grid cell with the spatial structure information to generate a four-dimensional spatiotemporal map.

[0068] In this embodiment of the application, the trend information and spatial structure information corresponding to each spatial grid unit are combined, and these two types of information are jointly assigned to each spatial grid unit to obtain a complete data structure, namely the four-dimensional spatiotemporal map. Each node of the map represents a spatial grid unit, and each node carries the temporal trend data and spatial correlation data of the unit.

[0069] This application organizes wind field data into a gridded time series and extracts its temporal evolution trend and spatial correlation structure layer by layer, thereby constructing a map that can comprehensively depict the spatiotemporal dynamic changes of the wind field, providing a structured data foundation for subsequent accurate prediction.

[0070] Step 103: Input the four-dimensional spatiotemporal map into the spatiotemporal prediction model for prediction, and output the predicted power data.

[0071] The spatiotemporal prediction model adopts a sequential processing structure consisting of a feature extraction module, a temporal network module, a spatial network module, a fusion module, a power mapping module, and an output module. The feature extraction module is a convolutional neural network containing two convolutional layers and one max-pooling layer, which is used to transform each time slice of the input four-dimensional spatiotemporal map into a feature vector of length 256 to form a map sequence. The temporal network module is a bidirectional long short-term memory network containing three layers of long short-term memory units, each layer containing 128 hidden neurons, which is used to process the map sequence and output a 128-dimensional time trend vector for future time steps as the first prediction information.

[0072] The spatial network module is specifically a graph attention network containing two graph convolutional layers, each layer using 8 attention heads to analyze the spatial correlation of the graph sequence and output a 128-dimensional spatial structure vector as the second prediction information; the fusion module is specifically a fully connected neural network layer, which concatenates the first prediction information vector and the second prediction information vector at the same time step, and then performs nonlinear fusion through a fully connected layer with 128 neurons to output a 128-dimensional unified feature vector as the predicted wind field information;

[0073] The power mapping module is specifically a decoding network consisting of two fully connected layers. The first layer has 64 neurons, and the second layer outputs two parameters: wind speed and wind direction. These parameters are used to map the predicted wind field information into the state data of each wind turbine location. Then, by querying the power curve of each wind turbine, the state data is converted into predicted power values. The output module is responsible for summarizing the predicted power values ​​of each wind turbine into a cluster total power trend and spatial distribution matrix.

[0074] The training process of this model is as follows: historical full-domain scanning radar data and actual power data of wind farm clusters within the corresponding time period are used as training samples. The historical radar data is processed by a spatiotemporal encoder and used as input. The actual power data is used as labels. After the predicted power data is obtained by forward propagation, the loss between the predicted value and the actual value is calculated using the mean square error function. The network parameters of all modules of the model are iteratively updated using the backpropagation algorithm and the adaptive moment estimation algorithm optimizer until the loss function converges to a stable state on the validation set.

[0075] It should be noted that the above structure is exemplary. This application does not impose specific limitations on the layer structure design, parameter design, etc. adopted in the internal structure of the spatiotemporal prediction model. The corresponding settings can be made according to the actual situation.

[0076] Predicted power data is a collection of information containing the expected power generation of a wind farm cluster over a future period of time.

[0077] In this embodiment, step 103 includes the following process:

[0078] Step 1031: Input the four-dimensional spatiotemporal map into the spatiotemporal prediction model, and extract map features at multiple times from the four-dimensional spatiotemporal map through the feature extraction module of the spatiotemporal prediction model to form a map sequence.

[0079] In step 1031, the feature extraction module is a component of the spatiotemporal prediction model. Its function is to extract key information from the input data for subsequent prediction. The spectral feature refers to a set of values ​​extracted from the data at each time point in the four-dimensional spatiotemporal map that can represent the core state of the wind field at that moment. The spectral sequence is a set of spectral features arranged in chronological order.

[0080] In this embodiment, the complete four-dimensional spatiotemporal map generated in step 102 is first input into the spatiotemporal prediction model. The feature extraction module inside the model then starts working. This module is specifically composed of a convolutional neural network. The module scans the four-dimensional spatiotemporal map and, for each historical time point recorded in the map, extracts a concise, fixed-length feature vector from the complex spatiotemporal data corresponding to that time point through a series of convolution and pooling operations of the network. This vector is the map feature of that time point. After processing all historical time points, these map features are arranged in chronological order to form a map sequence describing the historical evolution of the wind field.

[0081] In practical applications, for example, for a four-dimensional spatiotemporal map containing data from 12 past time points, the feature extraction module will process the data from these 12 time points in sequence, outputting a feature vector of length 256 for each time point. Finally, these 12 vectors of length 256 are combined in chronological order to form a map sequence containing 12 elements.

[0082] Step 1032: Analyze the temporal variation pattern of the spectral sequence through the temporal network module of the spatiotemporal prediction model to generate the first prediction information.

[0083] In step 1032, the time series network module is a component of the spatiotemporal prediction model, used to analyze the patterns of data changes over time; the first prediction information refers to the prediction data generated by the time series network module that describes the overall trend of wind field changes at future time points.

[0084] In this embodiment of the application, the spectral sequence obtained in step 1031 is input into the temporal network module inside the spatiotemporal prediction model. This module is specifically composed of a long short-term memory network. This network receives the spectral feature sequence input in time steps. Its internal memory units can learn and remember the long-term and short-term dependencies in the sequence, thereby analyzing the inherent pattern of wind field state evolution over time. After the forward propagation calculation of the network, the temporal network module will output a series of vectors. These vectors together describe the expected evolution trend of the wind field at the next few future time points. This part of the output data is the first prediction information.

[0085] In practical applications, this Long Short-Term Memory (LSTM) network may contain three hidden layers, each with 128 neurons. After reading a graph sequence of length 12, the LSM network outputs trend prediction vectors corresponding to the next six time points, with each vector having a length of 128.

[0086] Step 1033: Analyze the spatial variation relationship of the map sequence through the spatial network of the spatiotemporal prediction model to generate second prediction information.

[0087] In step 1033, the spatial network is another component of the spatiotemporal prediction model, used to analyze the correlation of data in the spatial dimension; the second prediction information refers to the prediction data generated by the spatial network that describes the spatial distribution pattern of the wind field at future time points.

[0088] In this embodiment, steps 1033 and 1032 are processed in parallel. Specifically, the map sequence obtained in step 1031 is input into the spatial network inside the spatiotemporal prediction model. This network is specifically composed of a graph convolutional network. This network treats the map features at each time point in the map sequence as a graph structure, where each feature dimension corresponds to a node, and the connection relationship between nodes reflects the mutual influence of the wind field in space. The graph convolutional network updates the features of each node by aggregating the information of adjacent nodes, thereby capturing the propagation and correlation patterns of the wind field in the spatial dimension. After the network processing is completed, another series of vectors will be output. These vectors describe the expected spatial distribution structure of the wind field at future time points. This part of the output data is the second prediction information.

[0089] In practical applications, this graph convolutional network may define the adjacency relationship between nodes as a fixed matrix that reflects the typical geographical impact of wind farms. After processing a graph sequence of length 12, the graph convolutional network will also output spatial distribution prediction vectors corresponding to the next 6 time points.

[0090] Step 1034: Through the fusion module of the spatiotemporal prediction model, process the first prediction information and the second prediction information to obtain the predicted wind field information corresponding to multiple time points in the future period.

[0091] In step 1034, the fusion module is a component of the spatiotemporal prediction model, and its function is to combine prediction information from different sources; the predicted wind field information refers to a complete description of the future wind field state obtained by fusing temporal trends and spatial distribution.

[0092] In this embodiment, after obtaining the first prediction information from the temporal network module and the second prediction information from the spatial network, the fusion module inside the spatiotemporal prediction model starts working. Specifically, this module performs a weighted concatenation operation, that is, for each future prediction time point, the first prediction information vector and the second prediction information vector corresponding to that time point are concatenated into a longer combined vector in element-wise order. Then, a fully connected neural network layer is used to perform nonlinear transformation and information integration on this combined vector, and finally outputs a unified feature vector that contains both temporal trends and spatial structures. The set of unified feature vectors of all these future time points constitutes the complete predicted wind field information.

[0093] In practical applications, for a certain future time point, the first prediction information is a vector A with a length of 128, and the second prediction information is a vector B with a length of 128. The fusion module first concatenates vector A and vector B into a vector C with a length of 256. Then, vector C passes through a fully connected layer with 128 neurons, and finally outputs a vector D with a length of 128. This vector D is the predicted wind field information corresponding to that future time point.

[0094] Step 1035: Using the power mapping module of the spatiotemporal prediction model, extract the state data corresponding to each wind turbine from the predicted wind farm information based on the location of each wind turbine in the wind farm cluster.

[0095] In step 1035, the power mapping module is a component of the spatiotemporal prediction model. Its function is to map the macroscopic prediction information of the wind field to the microscopic operating state of the specific wind turbine. The state data refers to parameters that are directly related to the expected wind conditions at a future point in time for a specific wind turbine location. Typically, the state data includes wind speed and wind direction.

[0096] In this embodiment, the power mapping module receives the predicted wind field information generated in step 1034 and the pre-stored geographical coordinates of each wind turbine in the wind farm cluster. The module maintains a mapping relationship internally, which is established based on the spatial grid division of the wind farm area. This mapping relationship can locate the corresponding data unit in the predicted wind field information based on any geographical coordinate. For each wind turbine, the power mapping module queries the unified feature vector of the coordinate corresponding to each future time point in the predicted wind field information based on its geographical coordinate. Then, it decodes these feature vectors through a fully connected neural network layer and converts them into specific wind condition parameters of the turbine's location at each future time point as state data.

[0097] In practical applications, for a wind turbine located at coordinate X, the power mapping module finds the corresponding data index in the predicted wind field information based on coordinate X, extracts 6 feature vectors corresponding to the next 6 time points, and then uses a decoding network to convert these 6 vectors into 6 sets of state data, each set of data including a wind speed value and a wind direction angle value.

[0098] Step 1036: The power mapping module calculates the predicted power value of each wind turbine in the future scheduling cycle based on the status data and the power characteristics of each wind turbine.

[0099] In step 1036, power characteristics refer to the inherent properties that describe the relationship between the power generation of a wind turbine and the incoming wind speed, and are usually defined by a power curve; the future scheduling cycle is a complete time interval, while multiple future scheduling time steps refer to a series of consecutive time points or sub-intervals obtained by dividing this cycle into fixed time intervals.

[0100] In this embodiment, for each wind turbine, the power mapping module uses the state data of the turbine at various future time points calculated in step 1035, especially the wind speed value, and queries the preset power characteristic curve of the wind turbine. The power characteristic curve takes wind speed as input and theoretical power generation as output. The power mapping module substitutes the predicted wind speed value at each future time point into the curve to calculate the theoretical power generation value of the wind turbine at that time point. This process is repeated for all wind turbines and all future time points to obtain the predicted power value of each wind turbine at each time point in the future scheduling cycle.

[0101] In practical applications, the power characteristic curve of a wind turbine may show that when the wind speed is 10 meters per second, its theoretical output power is 1500 kilowatts. If step 1035 calculates that the predicted wind speed of the turbine at a certain future time point is 10 meters per second, the power mapping module can obtain the predicted power value of the turbine at that time point as 1500 kilowatts by looking up a table or by calculation.

[0102] Step 1037: Output predicted power data based on the predicted power values ​​of all wind turbine units through the output module of the spatiotemporal prediction model.

[0103] In this embodiment, the output module within the spatiotemporal prediction model summarizes and organizes the predicted power values ​​of all wind turbine units calculated in step 1036 at all future time points. This module adds the predicted power values ​​of all units at the same time point to obtain the total predicted power of the entire wind farm cluster at that time point. Then, the total predicted power at all time points is arranged in chronological order to form the power change trend of the cluster. At the same time, the predicted power values ​​of each wind turbine unit at each time point are retained to form the power spatial distribution of the cluster. Finally, this structured data is packaged and output as predicted power data for use in subsequent steps.

[0104] In practical applications, for a cluster of 100 wind turbines, predicting six future time points, the output module will generate a time series containing six total power values, as well as a 100-row by 6-column matrix. Each row of the matrix represents a wind turbine, each column represents a time point, and each element in the matrix is ​​the specific predicted power value.

[0105] In the above process, this application utilizes a prediction model that integrates feature extraction, temporal analysis, spatial correlation, information fusion, and power mapping to systematically transform the four-dimensional map describing the historical spatiotemporal state of the wind farm into quantitative prediction data for future power generation, providing accurate and reliable input basis for subsequent optimized scheduling.

[0106] Step 104: Based on the predicted power data and combined with the power grid dispatch instructions, a hierarchical optimization algorithm is used for collaborative optimization to generate collaborative dispatch instructions.

[0107] The coordinated scheduling instruction refers to a set of control commands that explicitly specifies the power generation value that each wind turbine in the wind farm cluster should achieve at different future time points, as well as the aerodynamic attitude that should be adjusted. The explanation of the hierarchical optimization algorithm can be found in relevant technical documents and will not be elaborated upon here.

[0108] In this embodiment, step 104 includes the following process, such as... Figure 2 As shown:

[0109] Step 1041: Extract power change trend information and spatial distribution information from the predicted power data.

[0110] In step 1041, the power change trend information refers to the data sequence describing the change of the total power generation of the entire wind farm cluster over time, and the spatial distribution information refers to the data set describing the spatial distribution of the power generation of each wind turbine unit within the wind farm cluster.

[0111] In this embodiment of the application, the predicted power data output in step 103 is first parsed. This data contains two core parts: one part is a sequence of total predicted power values ​​of the entire cluster at various future time points, arranged in chronological order. Extracting this sequence data yields the power change trend information. The other part is a data matrix, where the rows of the matrix correspond to each specific wind turbine, the columns correspond to various future time points, and each element in the matrix represents the predicted power value of the turbine at that time point. Directly extracting this matrix yields the spatial distribution information.

[0112] In practical applications, for example, if the predicted power data includes information for the next 6 time points, then the power change trend information is a list containing 6 power values, such as [10.5, 11.2, 9.8, 12.1, 10.9, 11.5] megawatts; the spatial distribution information can be a 100-row by 6-column matrix, representing the predicted power of 100 wind turbines at the next 6 time points.

[0113] Step 1042: Input the power change trend information and the power grid dispatching command into the first-level optimization module of the hierarchical optimization algorithm. The first-level optimization module determines the total power target value of the wind farm cluster at each time point in the future period based on the power adjustment rate requirements and frequency regulation requirements of the power grid and the power change trend information.

[0114] The first-layer optimization module is the component in the hierarchical optimization algorithm responsible for determining the overall power target of the cluster. The power adjustment rate requirement refers to the maximum power value that the power output of the wind farm cluster is allowed to increase or decrease per unit time by the power grid. The frequency regulation requirement refers to the power capacity range that the power grid requires the wind farm cluster to reserve for rapid increases or decreases in order to maintain frequency stability. The total power target value refers to the power value that the wind farm cluster should transmit to the power grid at various points in the future, as determined after optimization.

[0115] Step 1042 may specifically include the following steps:

[0116] A1: Extract the predicted power values ​​for multiple time points within the future period from the power change trend information.

[0117] In step A1, the predicted power value refers to the power value corresponding to a specific future time point, which is directly read from the power change trend information sequence.

[0118] In practical applications, the power change trend information sequence is [10.5, 11.2, 9.8, 12.1, 10.9, 11.5] MW, and the extracted predicted power value is these 6 values.

[0119] A2: Establish a first function and a second function, and combine the first function and the second function into an objective function, wherein the first function is used to minimize the difference between the total power target value and the predicted power value, and the second function is used to minimize the degree of change of the total power target value between adjacent time points.

[0120] In step A2, the first function is a mathematical expression used to quantify the difference between the total power target value and the corresponding predicted power value; the second function is another mathematical expression used to quantify the degree of fluctuation of the total power target value over time; the objective function is a comprehensive performance index that needs to be minimized, which is a combination of the first function and the second function in a certain proportion.

[0121] It should be noted that this embodiment does not specifically limit the specific expressions used for the first function, the second function, and the target function; these can be set according to the actual situation.

[0122] In this embodiment of the application, the total power target value is set to... The predicted power value at the corresponding time point is First, establish the first function. Its expression is , where the symbol This function represents the summation of data over all future time points t. Its meaning is to calculate the sum of squares of the differences between the total power target value and the predicted power value at all time points. Minimizing this function means making the total power plan as close as possible to the predicted natural trend. Next, a second function is established. Its expression is This function calculates the sum of squares of the differences between the total power target values ​​at all adjacent time points. Minimizing this function means making the changes in the total power plan as gradual as possible. Finally, the first and second functions are linearly combined to form the final objective function J, expressed as follows: ,in and These are pre-set target weights greater than zero, used to adjust the relative importance between the two objectives of "closeness to prediction" and "smooth change." For example, and Both are 0.5.

[0123] A3: Solve the objective function by taking the power adjustment rate requirement and the frequency adjustment requirement as optimization constraints.

[0124] In step A3, the optimization constraints refer to the mathematical inequalities that the total power target value must satisfy when solving the objective function.

[0125] In this embodiment, the power regulation rate requirement of the power grid is transformed into a mathematical constraint, such as requiring that the absolute value of the difference between the total power target values ​​at adjacent time points does not exceed a preset difference threshold R. The frequency regulation requirements are transformed into mathematical constraints, such as requiring that the total power target value cannot exceed a certain upper limit at any point in time. It also cannot be lower than a certain lower limit. ,Right now After setting these constraints, a quadratic programming algorithm is used to solve the objective function J established in step A2. This algorithm can find a sequence of total power target values ​​that minimizes the objective function J under the above linear inequality constraints. .

[0126] A4: Based on the solution results, determine the total power target value at each time point.

[0127] In this embodiment of the application, after the quadratic programming algorithm is solved, the output optimal solution sequence is the total power target value that satisfies all power grid constraints and comprehensively balances the tracking prediction and steady change objectives.

[0128] In practical applications, assuming the optimal total power target value sequence obtained by solving is [10.8, 11.0, 10.2, 11.5, 11.2, 11.5] MW, compared with the original predicted sequence [10.5, 11.2, 9.8, 12.1, 10.9, 11.5], the change of the new sequence is more gradual and does not exceed the set power change rate limit.

[0129] Step 1043: Input the total power target value and the spatial distribution information into the second-layer optimization module of the hierarchical optimization algorithm. The second-layer optimization module decomposes the total power target value at each time point into the power generation command of each wind turbine according to the current power output value, location information and spatial distribution information of each wind turbine.

[0130] In step 1043, the second-layer optimization module is a component in the hierarchical optimization algorithm responsible for allocating the total power target of the cluster to each wind turbine; the power generation instruction refers to the specific power generation value issued to each wind turbine, requiring it to reach a specific power generation value at a future point in time.

[0131] In this embodiment of the application, the second-layer optimization module receives the total power target value sequence determined in step 1042 and the spatial distribution information extracted in step 1041; for each future time point, the task of this module is to reasonably decompose the total power target value of that time point to each wind turbine.

[0132] The decomposition is based primarily on: the actual power output of each unit at the current moment to ensure the continuity of adjustment; the predicted power value of each unit at that time point in the spatial distribution information to respect the natural distribution of wind energy; and the geographical location information of each unit to consider the wake effect, i.e., the shading effect of upwind units on downwind units. This module generates specific power generation instructions for each wind turbine at all future time points by solving an optimization problem with the allocation result being closest to the predicted power distribution of each unit and the total allocation value being strictly equal to the total power target value.

[0133] In practical applications, at a certain point in time, the total power target value is 11.0 MW, and the spatial distribution information shows that the predicted total power of 100 wind turbine units is 11.2 MW. The second-level optimization module uses the predicted power value of each unit as a benchmark. Under the premise of ensuring that the sum of the commanded power of 100 wind turbine units is equal to 11.0 MW, it calculates the power value that each wind turbine unit should receive, based on the current output of each unit. For example, unit 1 is 0.12 MW, unit 2 is 0.15 MW, etc.

[0134] Step 1044: Input the power generation command into the third-layer optimization module of the hierarchical optimization algorithm. The third-layer optimization module determines the attitude adjustment command corresponding to each power generation command based on the operating status of each wind turbine.

[0135] In step 1044, the three-layer optimization module is a component of the hierarchical optimization algorithm responsible for matching the aerodynamic attitude of each wind turbine; the operating status refers to the current mechanical and aerodynamic parameters of the wind turbine, including the blade pitch angle, the nacelle yaw angle, and the generator speed; the attitude adjustment command refers to the control command issued to the wind turbine's pitch system and yaw system, requiring them to adjust the blade angle and nacelle orientation to a specific position.

[0136] In this embodiment of the application, the third-layer optimization module receives the power generation command generated in step 1043 and obtains the current operating status of each wind turbine in real time; for the power generation command of each wind turbine at each future time point, the module needs to calculate the optimal blade angle and nacelle orientation of the unit in order to achieve the power output.

[0137] The calculation process is based on the aerodynamic characteristic model and power curve of the wind turbine model. By querying or calculating, it finds the combination of pitch angle and yaw angle that can generate the target power at a given predicted wind speed, while making the mechanical load of the unit small or the power generation efficiency high. Finally, it generates a set of specific attitude adjustment commands for each wind turbine at each time point, which includes the target pitch angle and the target yaw angle.

[0138] In practical applications, for a generator set that receives a power generation command of 1.5 MW at a future point in time, the third-layer optimization module queries the power-pitch angle relationship curve of the generator set under the current predicted wind speed. It may calculate that the pitch angle that achieves 1.5 MW of power with high efficiency is 5 degrees. At the same time, based on the wind direction prediction, the optimal yaw angle is calculated to be positive 15 degrees.

[0139] Step 1045: Combine the power generation commands and attitude adjustment commands at each time point to form a coordinated scheduling command.

[0140] In this embodiment, the power generation instructions generated in step 1043 and the attitude adjustment instructions generated in step 1044 are integrated and arranged according to the unit number and time sequence. The power generation instructions specify how much electricity each wind turbine generates, and the attitude adjustment instructions specify how each wind turbine generates electricity. The two correspond one-to-one and together constitute a complete control plan for multiple future time points. This complete plan, which includes all units, all time points, and all control dimensions, is the final coordinated scheduling instruction. This coordinated scheduling instruction will serve as the control basis directly issued to the wind farm cluster execution layer.

[0141] This application, through the aforementioned hierarchical and progressive optimization process, first formulates a stable total power plan that meets the macro-constraints of the power grid, then fairly and reasonably decomposes it to each wind turbine, and finally matches the optimal aerodynamic attitude for each wind turbine, thereby generating a set of coordinated, highly executable, and safe and economical refined dispatch instructions.

[0142] Step 105: Control the power generation and operating status of each wind turbine in the wind farm cluster according to the coordinated scheduling instruction.

[0143] In this embodiment, step 105 includes the following process:

[0144] Step 1051: Generate power control signals and attitude control signals for each wind turbine according to the power control instructions and attitude control instructions in the coordinated scheduling instructions.

[0145] In step 1051, the power control signal is an electrical signal or digital command used to adjust the output power of the wind turbine generator, and the attitude control signal is an electrical signal or digital command used to adjust the blade angle and nacelle orientation of the wind turbine generator.

[0146] In this embodiment of the application, the collaborative scheduling instruction generated in step 104 is first parsed, and the power control instruction part and attitude control instruction part for each specific wind turbine are separated from it; then, the central controller at the site level generates a digital or analog power control signal that can drive the power setpoint of the wind turbine converter through the internal signal conversion unit according to the power control instruction corresponding to each wind turbine.

[0147] Meanwhile, based on the attitude control command corresponding to each wind turbine, another signal conversion unit generates digital or analog attitude control signals that can directly drive the pitch actuator and yaw drive motor of that wind turbine.

[0148] Step 1052: Send the power control signal and the attitude control signal to the control unit of the corresponding wind turbine, so that the control unit of each wind turbine can adjust the power output of the wind turbine, the pitch angle of the blades and the yaw angle of the nacelle in the wind turbine according to the received power control signal and attitude control signal.

[0149] In step 1052, the control unit refers to the local controller installed on each wind turbine, which is used to receive instructions from the superior unit and directly drive the actuators of the unit.

[0150] In this embodiment of the application, the central controller at the wind farm level sends the power control signal and attitude control signal generated in step 1051 for each specific wind turbine to the local control unit of the wind turbine in real time through the industrial communication network inside the wind farm.

[0151] After receiving the power control signal, the local control unit uses the signal as the target value and dynamically adjusts the electromagnetic torque of the generator through its internal generator torque control algorithm, so that the actual power output of the generator tracks and reaches the target value.

[0152] Meanwhile, after receiving the attitude control signal, the local control unit analyzes the target pitch angle and target yaw angle, and drives the pitch motor and yaw motor to operate through its internal pitch servo control loop and yaw servo control loop, respectively, so that the blade angle and nacelle orientation are adjusted to the positions required by the command.

[0153] In this embodiment, after step 105, the method further includes the following steps:

[0154] B1: Collect actual operating data of each wind turbine in the wind farm cluster.

[0155] In step B1, the actual operating data refers to the various physical quantities that the wind turbine generator receives in real time from sensors during the execution of coordinated dispatch instructions, reflecting its current operating status.

[0156] In this embodiment of the application, during and after the execution of the control command, the actual operating data of the unit is continuously collected by the data acquisition module built into the local controller of each wind turbine. The actual operating data mainly includes the actual output power value of the generator, the actual pitch angle value of the blades at the current position, the actual yaw angle value of the nacelle at the current position, and the actual wind speed value at the nacelle measured by the anemometer.

[0157] B2: Compare the actual operating data with the corresponding expected state data in the collaborative scheduling instruction to generate state deviation data.

[0158] In step B2, the expected state data refers to the target power value, target pitch angle, and target yaw angle that the wind turbine should achieve at a specific time point, which are preset in the coordinated scheduling instruction; the state deviation data is the difference between the actual measured value and the expected target value.

[0159] In this embodiment of the application, the site-level controller compares the actual operating data of each wind turbine collected in step B1 at the actual time with the expected state data preset for that turbine at that time in the coordinated scheduling instruction; specifically, it calculates the difference between the actual power and the target power, the difference between the actual pitch angle and the target pitch angle, and the difference between the actual yaw angle and the target yaw angle; these calculated difference data together constitute the state deviation data describing the deviation of the instruction execution.

[0160] B3: Input the state deviation data into the adaptive adjustment model, which calculates the parameter correction amount according to the reinforcement learning strategy, and adjusts the target weights of each function in the hierarchical optimization algorithm according to the parameter correction amount.

[0161] In step B3, the adaptive adjustment model can adopt a deep reinforcement learning structure based on the actor-critic framework. The actor network is a multilayer perceptron containing two fully connected hidden layers. The input layer receives a feature vector composed of a sequence of historical state deviation data. The first hidden layer has 64 neurons and uses the ReLU activation function, and the second hidden layer has 32 neurons and uses the ReLU activation function. The output layer outputs the discrete action probability distribution of the target weights α and β of the hierarchical optimization algorithm through the Softmax function. The critic network is a multilayer perceptron with a similar structure but a single neuron in the output layer, used to evaluate the long-term value of state-action pairs.

[0162] The training process of this model involves simulating the closed-loop scheduling and control of a wind farm cluster in a simulation environment. Historical state deviation data is used as the initial state. The actor network outputs actions, i.e., weight adjustment schemes, and applies them to a hierarchical optimization algorithm to generate new instructions. After simulation execution, new state deviations and reward signals are obtained. The reward function is designed as a negative weighted sum of squared deviations. These interaction data are stored in an experience replay buffer. During training, data is sampled from the buffer. First, the value estimate of the current state-action pair is calculated through the critic network, and the critic network parameters are updated through backpropagation using time difference error. Then, the actor network parameters are updated through backpropagation using the policy gradient method based on the advantage function provided by the critic network. This process is iterated repeatedly until the average reward of the model converges on the validation set.

[0163] It should be noted that the above structure is exemplary. This application does not impose specific limitations on the layer structure design, parameter design, etc. adopted in the internal structure of the adaptive adjustment model. The corresponding settings can be made according to the actual situation.

[0164] In this embodiment of the application, the historical and current state deviation data sequence generated in step B2 is input into a pre-trained adaptive adjustment model; the model encapsulates a reinforcement learning agent, which uses the state deviation data as the environment state and adjusts the target weights α and β of the two functions of the first layer objective function in the hierarchical optimization algorithm as optional actions;

[0165] Based on the magnitude and pattern of current and historical deviations, the model evaluates the long-term benefits of different weight adjustment actions through its internal policy network, thereby calculating and outputting a set of parameter corrections that make future deviations less expected. The station-level controller receives this set of corrections and uses it to update the objective function weights in the first-level optimization module of the hierarchical optimization algorithm. For example, the original weight α is adjusted from 0.7 to 0.8, and β is adjusted from 0.3 to 0.2, so that the optimization objective is more inclined to track predictions or to smooth changes.

[0166] B4: Input the actual operating data into the equipment health assessment model. The equipment health assessment model processes the actual operating data based on a deep belief network and analyzes and generates load information for each wind turbine.

[0167] In step B4, the equipment health assessment model can adopt a deep belief network structure, which consists of three stacked restricted Boltzmann machines. The first restricted Boltzmann machine is the visible layer-first hidden layer, which receives feature vectors composed of standardized actual operating data of the wind turbine. The first hidden layer contains 100 neurons. The second restricted Boltzmann machine is the first hidden layer-second hidden layer, which contains 50 neurons. The third restricted Boltzmann machine is the second hidden layer-third hidden layer, which contains 20 neurons. An additional fully connected output layer with 10 neurons is added at the top of the network, and a scalar between 0 and 1 is output using the Sigmoid activation function as a fatigue load accumulation index.

[0168] The training process of this model is divided into two stages: pre-training and fine-tuning. In the pre-training stage, a large amount of unlabeled historical running data is used to greedily train each restricted Boltzmann machine layer by layer through the contrastive divergence algorithm to learn the intrinsic distribution of the data. In the fine-tuning stage, labeled data is used. The labels are obtained by experts from theoretical fatigue damage values ​​calculated by high-fidelity simulation models or historical maintenance records. The pre-trained network is combined with the top output layer, and the mean square error between the predicted fatigue index and the true label is minimized through the backpropagation algorithm. Thus, all weights of the network are fine-tuned end-to-end until the loss converges.

[0169] It should be noted that the above structure is exemplary. This application does not impose specific limitations on the layer structure design, parameter design, etc. adopted in the internal structure of the equipment health assessment model. Corresponding settings can be made according to the actual situation.

[0170] Load information refers to data that quantitatively describes the stress level and fatigue accumulation degree of key mechanical components of wind turbine units.

[0171] In this embodiment, the actual operating data collected in step B1, including wind speed, power, and blade pitch angle, is simultaneously input into a pre-trained equipment health assessment model. The core of this model is a deep belief network, which, through its multi-layer nonlinear transformation, can automatically extract deep features strongly correlated with mechanical loads from complex operating data. After processing the data, the network outputs an assessment result of the dynamic load level borne by key parts such as the blade root and main shaft of each wind turbine under the current and recent operating conditions, as well as an index characterizing the overall fatigue damage accumulation. These assessment results together constitute the load information of the turbine.

[0172] B5: Adjust the optimization constraints of the hierarchical optimization algorithm based on the load information.

[0173] In step B5, optimization constraints refer to the restrictions that decision variables must satisfy during optimization calculations.

[0174] In this embodiment, the wind turbine load information generated in step B4 of the site-level controller analysis, particularly the fatigue accumulation index, is analyzed. If the fatigue accumulation index of one or more turbines shows a high level, indicating that these turbines have been subjected to a large load, then when generating scheduling instructions for the next cycle, the upper limit constraints on the output capacity of these turbines need to be tightened in the second and third optimization modules of the hierarchical optimization algorithm. For example, the maximum allowable output value can be temporarily reduced by a certain percentage, or a stricter load change rate constraint can be added. In this way, the actual health status of the turbines is fed back and incorporated into the constraints of the optimization decision.

[0175] B6: Substitute the adjusted target weights and adjusted optimization constraints into the hierarchical optimization algorithm to generate new collaborative scheduling instructions for the next scheduling cycle in a rolling time domain manner.

[0176] In this embodiment, after obtaining the adjusted objective function weights in step B3 and the adjusted optimization constraints in step B5, the station-level controller initiates the instruction generation process for the next scheduling cycle. This instruction generation process inputs the latest predicted power data, grid scheduling instructions, adjusted weights, and adjusted constraints into the hierarchical optimization algorithm. The algorithm operates in a rolling time domain manner, that is, it re-executes the complete optimization process of step 104 based on the latest information and state. Finally, it outputs a set of new collaborative scheduling instructions suitable for the next scheduling cycle, thereby achieving closed-loop, adaptive scheduling control.

[0177] This application introduces the collection, comparative analysis, and model evaluation of actual operating data after the execution of control, and uses the evaluation results to dynamically adjust the parameters and constraints of the optimization algorithm, enabling the entire dispatch control system to have the ability to learn and adapt. It can continuously improve the generation quality of subsequent dispatch instructions based on the actual execution effect and the health status of the unit, thereby improving the accuracy and safety of control.

[0178] Figure 3 A schematic diagram of a wind farm cluster scheduling and control system based on full-domain scanning radar is provided for an embodiment of this application, as shown below. Figure 3 As shown, the system includes:

[0179] The acquisition module 31 is used to scan the target area by deploying a full-area scanning radar station at a set location in the wind farm cluster to acquire three-dimensional wind field data.

[0180] The processing module 32 is used to process the three-dimensional wind field data using a spatiotemporal encoder to generate a four-dimensional spatiotemporal map.

[0181] The input module 33 is used to input the four-dimensional spatiotemporal map into the spatiotemporal prediction model for prediction and output predicted power data.

[0182] The optimization module 34 is used to perform collaborative optimization based on the predicted power data and in conjunction with the power grid dispatch instructions, using a hierarchical optimization algorithm to generate collaborative dispatch instructions.

[0183] The control module 35 is used to control the power generation and operating status of each wind turbine in the wind farm cluster according to the coordinated scheduling instructions.

[0184] The wind farm cluster scheduling and control system based on global scanning radar in this application embodiment is used to implement the aforementioned wind farm cluster scheduling and control method based on global scanning radar. Therefore, the specific implementation of the wind farm cluster scheduling and control system based on global scanning radar can be found in the embodiment section of the wind farm cluster scheduling and control method based on global scanning radar mentioned above. The specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0185] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the wind farm cluster scheduling and control method based on full-domain scanning radar described above.

[0186] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described wind farm cluster scheduling and control methods based on full-domain scanning radar.

[0187] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0188] The embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the embodiments of the wind farm cluster scheduling and control method based on global scanning radar.

[0189] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0190] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0191] The foregoing has provided a detailed description of a wind farm cluster scheduling and control method and system based on full-domain scanning radar, as provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A wind farm cluster scheduling and control method based on full-domain scanning radar, characterized in that, include: Three-dimensional wind field data is obtained by scanning the target area with full-area scanning radar stations deployed at designated locations in the wind farm cluster. The three-dimensional wind field data is processed using a spatiotemporal encoder to generate a four-dimensional spatiotemporal map; The four-dimensional spatiotemporal map is input into the spatiotemporal prediction model for prediction, and the predicted power data is output. Based on the predicted power data and combined with power grid dispatch instructions, a hierarchical optimization algorithm is used for collaborative optimization to generate collaborative dispatch instructions. According to the coordinated scheduling instructions, the power generation and operating status of each wind turbine in the wind farm cluster are controlled; The step of inputting the four-dimensional spatiotemporal map into the spatiotemporal prediction model for prediction and outputting predicted power data includes: The four-dimensional spatiotemporal map is input into the spatiotemporal prediction model. Through the feature extraction module of the spatiotemporal prediction model, map features at multiple times are extracted from the four-dimensional spatiotemporal map to form a map sequence. The temporal network module of the spatiotemporal prediction model is used to analyze the temporal variation pattern of the spectral sequence and generate the first prediction information. By analyzing the spatial variation relationship of the spectral sequence through the spatial network of the spatiotemporal prediction model, second prediction information is generated. The fusion module of the spatiotemporal prediction model processes the first prediction information and the second prediction information to obtain the predicted wind field information corresponding to multiple time points in the future period. The power mapping module of the spatiotemporal prediction model extracts the state data corresponding to each wind turbine from the predicted wind farm information based on the location of each wind turbine in the wind farm cluster. The power mapping module calculates the predicted power value of each wind turbine in the future scheduling cycle based on the status data and the power characteristics of each wind turbine. The output module of the spatiotemporal prediction model outputs predicted power data based on the predicted power values ​​of all wind turbine units. The process of using a spatiotemporal encoder to process the three-dimensional wind field data and generate a four-dimensional spatiotemporal map includes: The three-dimensional wind field data is divided into multiple spatial grid units; In each spatial grid cell, multiple time-series data points associated with the spatial grid cell are identified; The correlation between multiple time series data points within the same spatial grid cell is analyzed through the first network layer of the spatiotemporal encoder to extract trend information reflecting changes in wind field intensity. The second network layer in the spatiotemporal encoder is used to analyze the correlation between trend information of adjacent spatial grid cells in order to extract spatial structural information reflecting changes in wind field morphology. By integrating the trend information of each spatial grid unit with the spatial structure information, a four-dimensional spatiotemporal map is generated.

2. The method according to claim 1, characterized in that, Based on the predicted power data and combined with power grid dispatch instructions, a hierarchical optimization algorithm is used for collaborative optimization to generate collaborative dispatch instructions, including: Power change trend information and spatial distribution information are extracted from the predicted power data; The power change trend information and the power grid dispatching instructions are input into the first-level optimization module of the hierarchical optimization algorithm. The first-level optimization module determines the total power target value of the wind farm cluster at each time point in the future period based on the power adjustment rate requirements and frequency adjustment requirements of the power grid and the power change trend information. The total power target value and the spatial distribution information are input to the second-level optimization module of the hierarchical optimization algorithm. The second-level optimization module decomposes the total power target value at each time point into the power generation command of each wind turbine according to the current power output value, location information and spatial distribution information of each wind turbine. The power generation command is input to the third-level optimization module of the hierarchical optimization algorithm. The third-level optimization module determines the attitude adjustment command corresponding to each power generation command based on the operating status of each wind turbine. The power generation commands and attitude adjustment commands at each time point are combined to form a coordinated scheduling command.

3. The method according to claim 2, characterized in that, The process of determining the total power target value of the wind farm cluster at various points in time during the future period, based on the power adjustment rate requirements and frequency regulation requirements of the power grid, and in conjunction with the power change trend information, includes: Extract predicted power values ​​for multiple time points within the future period from the power change trend information; Establish a first function and a second function, and combine the first function and the second function into an objective function, wherein the first function is used to minimize the difference between the total power target value and the predicted power value, and the second function is used to minimize the degree of change of the total power target value between adjacent time points; The objective function is solved by using the power adjustment rate requirement and the frequency adjustment requirement as optimization constraints. Based on the solution results, the total power target value at each time point is determined.

4. The method according to claim 1, characterized in that, The step of controlling the power generation and operating status of each wind turbine in the wind farm cluster according to the coordinated scheduling instruction includes: Based on the power control command and attitude control command in the coordinated scheduling command, power control signals and attitude control signals corresponding to each wind turbine are generated respectively. The power control signal and the attitude control signal are sent to the control unit of the corresponding wind turbine, so that the control unit of each wind turbine can adjust the power output of the wind turbine, the pitch angle of the blades and the yaw angle of the nacelle in the wind turbine according to the received power control signal and attitude control signal.

5. The method according to claim 1, characterized in that, After controlling the power generation and operating status of each wind turbine in the wind farm cluster according to the coordinated scheduling instructions, the method further includes: Collect actual operating data of each wind turbine in the wind farm cluster; The actual operating data is compared with the expected state data corresponding to the cooperative scheduling instruction to generate state deviation data; The state deviation data is input into the adaptive adjustment model, which calculates the parameter correction amount according to the reinforcement learning strategy, and adjusts the target weights of each function in the hierarchical optimization algorithm according to the parameter correction amount. The actual operating data is input into the equipment health assessment model, which processes the actual operating data based on a deep belief network and analyzes and generates load information for each wind turbine. Based on the load information, adjust the optimization constraints of the hierarchical optimization algorithm; Substitute the adjusted target weights and adjusted optimization constraints into the hierarchical optimization algorithm to generate new collaborative scheduling instructions for the next scheduling cycle in a rolling time domain manner.

6. A wind farm cluster scheduling and control system based on full-domain scanning radar, characterized in that, include: The acquisition module is used to scan the target area by deploying a full-area scanning radar station at a set location in the wind farm cluster to acquire three-dimensional wind field data; The processing module is used to process the three-dimensional wind field data using a spatiotemporal encoder to generate a four-dimensional spatiotemporal map. The input module is used to input the four-dimensional spatiotemporal map into the spatiotemporal prediction model for prediction and output predicted power data. The optimization module is used to perform collaborative optimization based on the predicted power data and in conjunction with power grid dispatch instructions, using a hierarchical optimization algorithm to generate collaborative dispatch instructions. The control module is used to control the power generation and operating status of each wind turbine in the wind farm cluster according to the coordinated scheduling instructions. The step of inputting the four-dimensional spatiotemporal map into the spatiotemporal prediction model for prediction and outputting predicted power data includes: The four-dimensional spatiotemporal map is input into the spatiotemporal prediction model. Through the feature extraction module of the spatiotemporal prediction model, map features at multiple times are extracted from the four-dimensional spatiotemporal map to form a map sequence. The temporal network module of the spatiotemporal prediction model is used to analyze the temporal variation pattern of the spectral sequence and generate the first prediction information. By analyzing the spatial variation relationship of the spectral sequence through the spatial network of the spatiotemporal prediction model, second prediction information is generated. The fusion module of the spatiotemporal prediction model processes the first prediction information and the second prediction information to obtain the predicted wind field information corresponding to multiple time points in the future period. The power mapping module of the spatiotemporal prediction model extracts the state data corresponding to each wind turbine from the predicted wind farm information based on the location of each wind turbine in the wind farm cluster. The power mapping module calculates the predicted power value of each wind turbine in the future scheduling cycle based on the status data and the power characteristics of each wind turbine. The output module of the spatiotemporal prediction model outputs predicted power data based on the predicted power values ​​of all wind turbine units. The process of using a spatiotemporal encoder to process the three-dimensional wind field data and generate a four-dimensional spatiotemporal map includes: The three-dimensional wind field data is divided into multiple spatial grid units; In each spatial grid cell, multiple time-series data points associated with the spatial grid cell are identified; The correlation between multiple time series data points within the same spatial grid cell is analyzed through the first network layer of the spatiotemporal encoder to extract trend information reflecting changes in wind field intensity. The second network layer in the spatiotemporal encoder is used to analyze the correlation between trend information of adjacent spatial grid cells in order to extract spatial structural information reflecting changes in wind field morphology. By integrating the trend information of each spatial grid unit with the spatial structure information, a four-dimensional spatiotemporal map is generated.

7. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor is configured to execute the computer program to implement the steps of the wind farm cluster scheduling and control method based on full-domain scanning radar as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the wind farm cluster scheduling and control method based on full-domain scanning radar as described in any one of claims 1 to 5.

Images