Wind power cluster power regulation system and method based on domestic SCADA

Abstract

The application discloses a kind of wind power cluster power regulation system and method based on domestic SCADA, including power grid dispatching layer, SCADA energy management layer, power scheduling execution layer, data acquisition layer and field device layer, power grid dispatching layer is connected with power grid dispatching center, receives the AGC active power instruction and AVC reactive power instruction issued by power grid, and real-time power generation information of wind power cluster is reported to power grid dispatching center, and carries out data interaction with SCADA energy management layer, SCADA energy management layer is deployed on domestic server cluster, power scheduling execution layer receives the wind turbine power set value issued by SCADA energy management layer, and executes power instruction to each wind turbine converter by closed-loop control, data acquisition layer is responsible for real-time acquisition operation data from field device layer and uploads to SCADA energy management layer and power scheduling execution layer, and field device layer is connected with data acquisition layer.The application reduces overall cost.

Description

Technical Field

[0001] This invention relates to the technical field of offshore wind power cluster control technology, and in particular to a wind power cluster power regulation system and method based on domestically produced SCADA. Background Technology

[0002] Currently, the power control of wind power clusters faces the following problems: 1. System Architecture Level: Currently, the power control systems of large-scale offshore wind power clusters in China mainly rely on imported SCADA / EMS platforms such as Siemens Spectrum Power, ABB MicroSCADA, and GE PowerOn. However, these systems have closed architectures, proprietary protocols, and unavailable source code, posing significant risks to data security and supply chain security. Moreover, most existing domestic alternatives adopt single-point replacements (such as only replacing server hardware or operating systems), thus lacking a complete domestic system from data acquisition, energy management, scheduling strategies to execution control to reduce the equipment cost of the system control layer. 2. From the perspective of scheduling strategy, most existing scheduling schemes adopt static scheduling or simple proportional allocation strategies. Static scheduling schemes have a slow response and cannot keep up with the rapid changes in sea wind speed. Simple proportional allocation does not take into account the differences in available power of each wind turbine, fatigue load balancing and grid constraints, resulting in some units being overloaded for a long time and the rest being underloaded, affecting the lifespan of the units and economic benefits. 3. Predictive capability: Most existing systems do not have the ability to predict wind speed and power together, or only use simple continuous prediction methods. However, in the marine environment, the turbulence intensity of wind speed is high and the direction changes rapidly. If a simple prediction method is used, the accuracy is low (the prediction error in 15 minutes usually reaches 15%-20% of the rated power), resulting in insufficient foresight of the dispatching scheme and inability to adjust the power allocation in advance to cope with the upcoming changes in wind conditions. 4. Control accuracy: Existing execution terminals mostly adopt simple proportional (P) control or open-loop stepped control, lacking closed-loop regulation capability. The steady-state deviation of power tracking is large (usually 2%-5% of rated power) and the response time is long (usually 5-10 seconds), which cannot meet the increasingly stringent requirements of power grid dispatch for fast and accurate power tracking (steady-state deviation ≤0.5% of rated power, response time ≤2 seconds). 5. Starting from the data acquisition layer (4), existing data acquisition solutions usually only support a single or a few communication protocols, which cannot flexibly adapt to multiple brands and models of wind turbine equipment; the data preprocessing capability is weak, lacking outlier filtering and timestamp alignment functions, resulting in poor quality of uploaded data, which directly affects the accuracy of the upper-layer scheduling strategy; 6. Security protection level: The existing system lacks a multi-layered security protection mechanism. When communication is interrupted, equipment fails, or power exceeds the limit, it cannot automatically handle the situation in a tiered manner, which poses a scheduling security risk. In summary, existing wind power cluster power control systems have significant shortcomings in terms of autonomous controllability, scheduling accuracy, predictive capability, control accuracy, data acquisition, and safety protection, and there is an urgent need to propose a comprehensive solution. Summary of the Invention

[0003] The purpose of this invention is to provide a wind power cluster power regulation system and method based on domestically produced SCADA, in order to solve the problems existing in the prior art.

[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a wind power cluster power regulation system based on domestic SCADA, comprising, from top to bottom, a power grid dispatching layer, a SCADA energy management layer, a power dispatching execution layer, a data acquisition layer, and a field equipment layer; The SCADA energy management layer is the core decision-making layer of the system, deployed on a domestically produced server cluster at an offshore booster station or onshore control center. This layer includes the following key components: a domestically produced SCADA master station server, an energy management system (EMS), a power prediction engine module (power prediction engine), a cluster power dynamic adjustment strategy engine module (strategy engine), a historical database that provides training and backtesting data for the prediction engine, and a human-machine interface. The domestically produced SCADA master station server, energy management system, power prediction engine module, cluster power dynamic adjustment strategy engine module, historical database, and human-machine interface are interconnected, and the SCADA master station server has its own real-time database, providing a unified data source for the power prediction engine, strategy engine, and human-machine interface. The power prediction engine module, based on historical wind speed-power data and real-time meteorological tower data, uses an LSTM neural network model to perform short-term power prediction for wind power clusters and outputs the predictions to the cluster power dynamic adjustment strategy engine module. The power prediction engine module includes a data preprocessing and feature engineering module, an LSTM inference module, a prediction result output and evaluation module, and a model training module, which are sequentially connected to these modules. The prediction result output and evaluation module is connected to the cluster power dynamic adjustment strategy engine module and the historical database, enabling the model training and update module to obtain training data from the historical database. After training, the updated model parameters are deployed to the LSTM inference module, and the prediction curve output by the prediction result output and evaluation module is sent to the cluster power dynamic adjustment strategy engine module as input for optimization. The solution results from the cluster power dynamic adjustment strategy engine module are distributed by the energy management system. The cluster power dynamic adjustment strategy engine module adopts a multi-objective optimization model and, based on the total power command, power prediction results, and current operating status of the energy management system, solves to generate the optimal power allocation scheme for each wind turbine. The cluster power dynamic adjustment strategy engine module includes a multi-source data access interface, a multi-objective optimization model construction module, an NSGA-II solution algorithm module, a TOPSIS scheme selection module, and a result output interface. The energy management system, power prediction engine module, and real-time database are respectively connected to the multi-objective optimization model construction module, NSGA-II solution algorithm module, and TOPSIS scheme selection module through the multi-source data access interface, and are connected to the energy management system's distribution channel and historical database through the result output interface. This enables the engine to solve for and generate the optimal power allocation scheme for each wind turbine based on the total power command, power prediction results, and current operating status of the energy management system. The grid dispatch layer is located at the top of the system and is connected to the grid dispatch center. It receives AGC (Automatic Generation Control) active power commands and AVC (Automatic Voltage Control) reactive power commands issued by the grid, reports the real-time generation information of the wind power cluster to the grid dispatch center, and performs bidirectional data interaction with the energy management system of the SCADA energy management layer through the IEC 104 protocol or dispatch data network to realize downlink transmission of AGC / AVC commands and uplink transmission of real-time power and unit status information of the wind power cluster. The core of the data acquisition layer is a data acquisition interface device, which internally includes a multi-protocol communication adaptation module interconnected via an internal bus, a data acquisition and preprocessing module, a domestically produced main control module, and a data upload and interface output module. The field equipment layer includes all field physical equipment in the wind power cluster, including converters, wind measurement tower sensors, grid metering devices (electricity meters, transformers, etc.) and substation protection devices for each wind turbine group. Each device has a standard communication interface and is connected to the data acquisition layer via Ethernet or RS485 serial port. The power scheduling execution layer is equipped with a power scheduling execution terminal, which is responsible for receiving the power setting values ​​of each wind turbine issued by the SCADA energy management layer, and accurately executing the power commands to each wind turbine converter in the field equipment layer through closed-loop control.

[0005] Preferably, the power scheduling execution terminal includes: Receive the power setting values ​​of each wind turbine sent by EMS, and send the control quantities to the instruction receiving and communication module of each wind turbine converter; A power allocation module used to perform local allocation according to a dynamic proportional algorithm when the EMS only issues a total power command (rather than the individual turbine values); The execution terminal PID control calculation module is used to perform closed-loop adjustment of the deviation between the actual output power of each fan and the set value. A status monitoring and protection module used to monitor communication status, power deviation and equipment alarms, and with automatic fault switching and safety interlock functions; The power control quantity output by the PID control calculation module (113) of the execution terminal is sent to the power setting port of the wind turbine converter through the downlink interface (4 Ethernet + 2 RS485) of the instruction receiving and communication module in the protocol format corresponding to each wind turbine.

[0006] Preferably, the data upload and interface output module in the data acquisition layer uploads the preprocessed standardized data frames to the communication front-end of the domestic SCADA master station server through a dual-path gigabit Ethernet uplink (main / backup). The communication protocol adopts IEC 61850 MMS or OPC UA, and the data is transmitted with SM4 national cryptographic encryption. At the same time, the data upload and interface output module in the data acquisition layer sends the actual power feedback data of the wind turbine to the power scheduling execution terminal through the internal bus bypass interface (101) for use as feedback by the PID module (this PID module is the PID for the entire scheduling command and the entire wind farm adjustment). The energy management system (EMS) sends the power setpoints of each wind turbine generated by the cluster power dynamic adjustment strategy engine module through gigabit Ethernet to the instruction receiving and communication module of the power scheduling execution terminal via IEC 61850 or Modbus TCP protocol.

[0007] As a preferred option, the cluster power dynamic adjustment strategy engine module adopts a multi-objective optimization model, simultaneously optimizing the following three objectives: Objective 1 (Power Tracking Accuracy): min Σ|Pi_actual - Pi_set|², to minimize the deviation between the actual total output power of the wind power cluster and the grid dispatch target power; Objective 2 (fatigue load balancing): min max(Li / Li_rated), to minimize the maximum value of fatigue load utilization of each wind turbine, and avoid long-term overload of a single unit; Objective 3 (Real-time response to grid constraints): Maximize the output of the wind power cluster while meeting the grid frequency deviation constraint Δf≤±0.2Hz and the voltage constraint ΔU≤±5%Un; In the above target optimization formula: Pi_actual represents the actual output power of the i-th wind turbine (unit: MW); Pi_set represents the power setpoint of the i-th wind turbine (unit: MW); Li represents the current cumulative fatigue load of the i-th wind turbine; Li_rated represents the rated fatigue load design value of the i-th wind turbine; Δf represents the grid frequency deviation (unit: Hz), that is, the difference between the actual frequency and the rated frequency of 50Hz; ΔU represents the grid voltage deviation, where Un is the rated voltage and ΔU is the percentage of the difference between the actual voltage and the rated voltage. The constraints include: the power setting value of each wind turbine does not exceed its current available power limit; the power change rate of the wind turbine does not exceed 10% / second of the rated power (to prevent sudden power changes); the total power of the wind power cluster does not exceed the grid connection point capacity limit. The solution algorithm adopts the improved NSGA-II (non-dominated sorting genetic algorithm), with a population size of 100, 200 iterations, adaptive crossover and mutation operators, and a solution cycle of 1 second. A complete optimization solution is completed in each scheduling cycle to generate the power setting value of each wind turbine for the next scheduling cycle.

[0008] Preferably, the power prediction engine module reads wind speed, wind direction, temperature, air pressure, and historical power output data of each wind turbine from historical / real-time databases for the past 24 hours. It also obtains forecast data from an external numerical weather prediction interface (the numerical weather prediction interface is the GRAPES system of the China Meteorological Administration, which stores long-term historical data). The module uses data preprocessing and feature engineering to imputate missing values, normalize, and construct time series features (such as moving average, rate of change, etc.) on the raw data. Then, it outputs a standardized multidimensional feature matrix and sends it to the LSTM inference module. The LSTM inference module receives the standardized feature matrix output by the feature engineering module as input and performs forward inference through a pre-trained three-layer LSTM neural network model (128 hidden units per layer) to output the wind power cluster power prediction curve for the next 15 minutes to 4 hours with a time resolution of 15 minutes. At the same time, the model training and update module uses historical wind speed-power data from the past 12 months in the historical database to periodically (e.g., once a month) retrain the LSTM inference module offline, updates the model weight parameters, and then deploys them to the LSTM inference module to adapt to seasonal changes in wind conditions.

[0009] This invention also discloses a method for dynamic power regulation of wind power clusters based on a domestically developed SCADA system, which includes the following steps: Step S1, Real-time Data Acquisition: The data acquisition interface device automatically identifies the communication protocol type of each field device through the multi-protocol communication adapter module, and collects the active power, reactive power, voltage, current and speed data of each wind turbine converter, the wind speed, wind direction, temperature and air pressure data of the wind tower sensor, and the grid connection point power and frequency data of the power grid metering device in real time at the configured period (default 100ms). Step S2, Data Preprocessing: The data acquisition and preprocessing module in the data acquisition interface device performs the following preprocessing on the raw data: (a) CRC check, which records the packet loss rate by discarding data frames that fail the check; (b) Sliding window median filtering: by obtaining 5 sampling points of the window size, outliers that deviate from the median by more than 3 times the standard deviation are removed; (c) Timestamp alignment: All data are appended with a unified UTC timestamp using a GPS / BeiDou dual-mode timing module, with an accuracy of ≤1ms; (d) Data format standardization: raw data in different protocol formats are uniformly converted into standardized data frames containing device ID, data type, timestamp, value and quality code. The preprocessed data is simultaneously uploaded to the SCADA energy management layer and power scheduling execution layer. The aforementioned "raw data" refers to the data collected during the real-time data acquisition phase in step S1, including the active power, reactive power, voltage, current, and speed of each wind turbine converter, the wind speed, wind direction, air temperature, and air pressure from the anemometer tower sensors, and the power and frequency data at the grid connection point of the power grid metering device. Step S2 performs preprocessing operations such as CRC verification, sliding window value filtering, GPS / BeiDou timestamp alignment, and data format standardization based on the raw data collected in step S1. Step S3, Power Prediction: The power prediction engine module of the SCADA energy management layer reads the wind speed and power time series of the past 24 hours and the current real-time wind measurement data from the historical database. Combined with numerical weather forecast information, it generates the wind power cluster power prediction curve for the next 15 minutes to 4 hours through a three-layer LSTM neural network model. The prediction engine runs once every 15 minutes and sends the prediction results to the strategy engine. Step S4, Multi-objective optimization solution: The cluster power dynamic adjustment strategy engine module receives the following inputs: grid AGC / AVC command (target power), power prediction results, current available power and fatigue load status of each wind turbine, and real-time frequency and voltage on the grid side. The strategy engine constructs a multi-objective optimization model (objectives: optimal power tracking accuracy, optimal fatigue load balance, and grid constraint satisfaction). Under the constraints (including the upper limit of power for each wind turbine, power change rate limit, and grid connection capacity limit), the improved NSGA-II algorithm is used to solve the Pareto optimal solution set (since Pareto comes from the Italian economist Pareto, the "Pareto optimal solution set" refers to a set of solutions that cannot improve a certain objective without harming other objectives, i.e., the optimal compromise solution set between objectives, also known as the "Pareto front". The TOPSIS method then selects the single optimal solution from it. The TOPSIS method (meaning the approximation of ideal solution sorting method) is used to select the optimal solution from the Pareto front, generate the power setpoint for each wind turbine, and the entire solution process is completed within 1 second. Step S5, Dynamic Power Allocation and Distribution: The power setting values ​​of each wind turbine generated by the strategy engine are distributed to the power scheduling execution terminal (11) via EMS. When communication is abnormal and only the total power command can be distributed, the power allocation module (112) of the execution terminal performs local allocation according to the dynamic ratio algorithm Pi_set = P_total × (Pi_avail / ΣPi_avail). When a wind turbine exits due to a fault, the power allocation module automatically excludes it and recalculates the setting values ​​of the remaining wind turbines. In step S5, the dynamic proportional allocation formula Pi_set = P_total × (Pi_avail / ∑Pi_avail) has the following meanings for each symbol: Pi_set: The power setting value (unit: MW) of the i-th wind turbine, that is, the target output power allocated to this wind turbine; P_total: The total power command (unit: MW) issued by the power grid dispatch center for the wind power cluster, which is the total output power target that the wind power cluster needs to achieve; Pi_avail: The current available power of the i-th wind turbine (unit: MW), that is, the maximum power that the wind turbine can output under the current wind conditions and equipment status; ∑Pi_avail: The sum of the available power of all operating wind turbines (unit: MW), that is, the total available output of the cluster at present; The physical meaning of this formula is: to dynamically allocate the total power command according to the proportion of the available power of each wind turbine to the total available power, so that the wind turbine with larger available power undertakes more power generation tasks, thereby achieving load balancing; Step S6, PID Closed-Loop Precise Control: The PID control calculation module of the execution terminal executes closed-loop control independently for each wind turbine with a control cycle of 50ms. The wind turbine power setpoint is the target value and the actual output power is the feedback value. The power deviation e(t) is calculated and the PID calculation control formula u(t) = Kp·e(t) + Ki·∫e(t)dt + Kd·de(t) / dt (Formula 1) is executed. The control quantity u(t) is sent to the corresponding wind turbine converter through the communication interface. The PID parameters support online self-tuning: when the system detects that the adjustment time or overshoot of the power step response does not meet the index, the Kp, Ki, and Kd parameters are automatically fine-tuned. In step S6, u(t) in the PID control formula represents the control quantity, that is, the power adjustment command (unit: MW) output by the PID controller at time t, which is sent to the wind turbine converter to adjust its output power. e(t) represents the power deviation, which is the difference between the setpoint power of the wind turbine at time t and the actual output power (e(t) = Pi_set - Pi_actual, unit: MW). Kp represents the proportional gain coefficient, which determines the proportional relationship between the control quantity and the current deviation. The larger Kp is, the faster the response but the more prone to oscillation. Ki represents the integral gain coefficient, which determines the relationship between the control quantity and the cumulative deviation, and is used to eliminate steady-state deviation. Kd represents the differential gain coefficient, which determines the relationship between the control quantity and the rate of change of the deviation, and is used to suppress oscillations and improve stability. ∫e(t)dt represents the integral of the power deviation over time, reflecting the historical cumulative deviation; de(t) / dt represents the derivative (rate of change) of the power deviation with respect to time, reflecting the trend of the deviation. Step S7: Multi-level security protection. The entire system implements security protection at the following three levels: (a) Data acquisition layer: When the edge computing node detects a serious fault in the wind turbine (overspeed, overtemperature, converter fault), it directly triggers local protection actions (power limit or emergency shutdown), with a response time of <50ms; (b) Power scheduling execution layer: When the status monitoring and protection module (114) detects that the packet loss rate of the communication link exceeds the threshold (5%), it automatically switches to the backup link. When it detects that the power deviation continues to exceed the limit (>3% of the rated power for 10 seconds), it triggers the amplitude limiting protection. When it detects a serious hardware failure, it triggers the main / backup terminal switching (switching time <500ms). (c) SCADA Energy Management Layer: When the Energy Management System (EMS) detects that the total power deviation of the wind power cluster exceeds the allowable range of the power grid dispatch, it automatically reduces the target power to the safety limit and reports the abnormal status to the power grid dispatch center; Step S8: Continuously execute steps S1 to S7 in a loop to form a complete closed-loop dynamic adjustment process.

[0010] As a preferred option, when the entire system is running normally, the end-to-end delay (end-to-end refers to the time from data acquisition to the control quantity being sent to the converter) from step S1 to step S6 is ≤200ms, which meets the requirements of power grid dispatch for rapid power response.

[0011] This invention provides a wind power cluster power regulation system and method based on domestically produced SCADA, which has the following advantages: ① Full-chain localization: From field data acquisition devices (domestic processors + domestic real-time operating systems), SCADA master station (domestic servers + domestic operating systems + independent software) to dispatch execution terminal (domestic industrial control processors), the entire wind power cluster power regulation system is fully localized, completely eliminating dependence on imported SCADA / EMS platforms and meeting the requirements for independent control of critical infrastructure; ② Improved dispatching accuracy: The multi-objective optimization dispatching strategy combined with PID closed-loop control reduces the steady-state deviation of wind power cluster power tracking from 2%-5% of rated power in the traditional scheme to ≤0.5% of rated power, and shortens the response time from 5-10 seconds to ≤2 seconds, significantly improving the grid dispatching response capability; ③ Proactive scheduling: The LSTM-based short-term power prediction engine enables the scheduling scheme to have a forward-looking capability of 15 minutes to 4 hours, with a 15-minute prediction error of ≤8% of rated power (reduced by about 50% compared to traditional continuous prediction methods), allowing the system to adjust power allocation in advance to cope with changes in wind conditions. ④ Fatigue load balancing: The multi-objective optimization strategy incorporates the fatigue load balancing of the wind turbine into the optimization objective, avoiding long-term overload operation of a single unit. It is expected to extend the service life of key components of the unit (gearbox, main bearing) by 15%-20%, and significantly reduce operation and maintenance costs. ⑤ Data quality assurance: The data acquisition interface device performs verification, filtering, timestamp alignment and format standardization locally, improving the data effectiveness rate from about 92% in the traditional solution to more than 99.5%, providing a high-quality data source for upper-level scheduling strategies; ⑥ High reliability: The system has a multi-level redundancy and protection mechanism, including hot standby switching of primary / backup terminals (<500ms), automatic switching of redundant communication links, local emergency protection of edge nodes (<50ms), and data interruption resumption, to ensure high reliability of the system in harsh marine environments. The system availability rate is ≥99.95%. Attached Figure Description

[0012] Figure 1 This is a connection diagram of a wind power cluster power regulation system based on domestic SCADA in Embodiment 1. Figure 2 This is a schematic diagram of the connection of the SCADA energy management layer in Embodiment 1; Figure 3 This is a connection diagram of the power prediction engine module in Embodiment 1; Figure 4 This is a connection diagram of the cluster power dynamic adjustment strategy engine module in Embodiment 1. Figure 5 This is a connection diagram of the data acquisition interface device in Embodiment 1; Figure 6 This is a schematic diagram of the connection of the power scheduling execution terminal in Embodiment 1.

[0013] The following are the labels in the diagram: Power Grid Dispatch Layer 1, SCADA Energy Management Layer 2, Domestic SCADA Master Server 201, Energy Management System 202, Power Prediction Engine Module 203, Cluster Power Dynamic Adjustment Strategy Engine Module 204, Historical Database 205, Human-Machine Interface 206, Real-Time Database 2011, Data Preprocessing and Feature Engineering Module 2031. LSTM Inference Module 2032, Prediction Result Output and Evaluation Module 2033, Model Training and Update Module 2034, Multi-Source Data Access Interface 2041, Multi-Objective Optimization Model Construction Module 2042, NSGA-II Solving Algorithm Module 2043, TOPSIS Scheme Selection Module 2044, Result Output Interface 2045, Power Scheduling Execution Layer 3, Data Acquisition Layer 4, Field Equipment Layer 5, Power Grid Dispatch Center 6, Multi-Protocol Communication Adaptation Module 7, Data Acquisition and Preprocessing Module 8, Domestic Main Control Module 9, Data Upload and Interface Output Module 10, Power Scheduling Execution Terminal 11, Command Receiving and Communication Module 111, Power Allocation Module 112, Execution Terminal PID Control Calculation Module 113, Status Monitoring and Protection Module 114, Internal Bus Bypass Interface 101. Detailed Implementation

[0014] 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, and 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.

[0015] Example 1 like Figures 1-6 The present invention provides an embodiment of a wind power cluster power regulation system based on domestic SCADA, which adopts a five-layer hierarchical architecture design, including, from top to bottom, a power grid dispatch layer 1, a SCADA energy management layer 2, a power dispatch execution layer 3, a data acquisition layer 4, and a field equipment layer 5. The SCADA energy management layer 2 is the core decision-making layer of the system, deployed on a domestically produced server cluster at an offshore booster station or onshore control center. This layer includes the following key components: a domestically produced SCADA master station server 201, an energy management system (EMS) 202, a power prediction engine module 203 (power prediction engine), a cluster power dynamic adjustment strategy engine module 204 (strategy engine), a historical database 205, and a human-machine interface 206. The domestically produced SCADA master station server 201, energy management system 202, power prediction engine module 203, cluster power dynamic adjustment strategy engine module 204, historical database 205 which provides training and backtesting data for the prediction engine, and human-machine interface 206 are interconnected. The SCADA master station server 201 has a built-in real-time database 2011 that provides a unified data source for the power prediction engine, strategy engine, and human-machine interface 206. The power prediction engine module 203 includes a data preprocessing and feature engineering module 2031, an LSTM inference module 2032, a prediction result output and evaluation module 2033, and a model training and update module 2034. The historical database 205 and the real-time database 2011 are sequentially connected to the data preprocessing and feature engineering module 2031, the LSTM inference module 2032, and the prediction result output and evaluation module 2033. The prediction result output and evaluation module 2033 outputs data that is connected to the cluster power dynamic adjustment strategy engine module 204 and the historical database 205, enabling the model training and update module 2034 to obtain training data from the historical database 205 and deploy the updated model parameters to the LSTM inference engine after training. Within the TM inference module 2032, the prediction results output and the prediction curve output by the evaluation module 2033 are sent to the cluster power dynamic adjustment strategy engine module 204 as input for optimization. The solution results of the cluster power dynamic adjustment strategy engine module 204 are distributed by the energy management system 202. In the above structure, the real-time database 2011 is a memory database built into the domestic SCADA master station server 201, which is specifically used to cache the real-time polling data of each wind turbine, with a refresh cycle of ≤1s. The historical database 205 is an independent time-series database that stores long-term historical data. Both provide data to the power prediction engine and the strategy engine, but have different functions. The described cluster power dynamic regulation strategy engine module 204 includes a multi-source data access interface 2041, a multi-objective optimization model construction module 2042, an NSGA-II solution algorithm module 2043, a TOPSIS solution optimization module 2044, and a result output interface 2045. The energy management system 202, the power prediction engine module 203 (since the power prediction engine module 203 is used as a data access source, it actually refers to the output interface of the prediction result output and evaluation module 2033 in its sub-module, that is, the prediction result output and evaluation module 2033 transmits the prediction curve to the multi-source data access interface 2041), and the real-time database 2011 are respectively connected to the multi-objective optimization model construction module 2042, the NSGA-II solution algorithm module 2043, and the TOPSIS solution optimization module 2044 in sequence through the multi-source data access interface 2041, and are connected to the distribution channel of the energy management system 202 and the historical database 205 through the result output interface 2045, and is responsible for solving and generating the optimal power distribution plan for each fan according to the total power instruction, power prediction result, and current operating state of the energy management system 202; The power grid dispatching layer 1 is connected to the power grid dispatching center 6 and the energy management system 202 to achieve the downlink transmission of AGC / AVC commands and the uplink transmission of the real-time power and unit status information of the wind power cluster; the power dispatching execution layer 3 is connected to the SCADA energy management layer 2 and the field device layer 5; the power grid dispatching layer 1 is located at the top layer of the system and is connected to the power grid dispatching center 6 to receive the AGC (AGC represents automatic generation control) active power command and AVC (AVC represents automatic voltage control) reactive power command issued by the power grid, and report the real-time power generation information of the wind power cluster to the power grid dispatching center 6, and perform two-way data interaction with the energy management system 202 of the SCADA energy management layer 2 through the IEC104 protocol or the dispatching data network to achieve the downlink transmission of AGC / AVC commands and the uplink transmission of the real-time power and unit status information of the wind power cluster; The core of the data acquisition layer 4 is a data acquisition interface device, and the data acquisition interface device internally includes a multi-protocol communication adaptation module 7, a data acquisition and preprocessing module 8, a domestic main control module 9, and a data upload and interface output module 10 interconnected through an internal bus; The field device layer 5 includes all the field physical devices in the wind power cluster, including the converters of each fan group, wind measurement tower sensors, power grid metering devices (including watt-hour meters, instrument transformers, etc.), and substation protection devices, and each device has a standard communication interface and is connected to the data acquisition layer 4 through Ethernet or RS485 serial port; The power dispatch execution layer 3 is internally deployed with power dispatch execution terminals 11, and is responsible for receiving the power set values of each fan sent by the SCADA energy management layer 2, and precisely executing the power commands to each fan converter in the field device layer 5 through closed-loop control.

[0016] The domestically produced SCADA master station server 201 uses domestically produced server hardware (Phytium / Kunpeng processor) and a domestically produced operating system (Kylin / Tongxin UOS), running independently developed SCADA master station software. The software of the domestically produced SCADA master station server 201 implements real-time database management, historical data archiving, alarm event management, access control, and communication front-end functions. The real-time database adopts a memory-mapped architecture, with a data refresh cycle of ≤100ms, supporting concurrent processing of no less than 500,000 measurement points. The Energy Management System (EMS) 202 runs on the domestically produced SCADA master station server 201, responsible for receiving AGC / AVC instructions from the power grid dispatch layer, combining the current real-time operating status of the wind power cluster and power prediction results, generating total power dispatch instructions, and sending them to the power dispatch execution layer. EMS calculates the optimal power dispatch scheme based on the target power required by the grid, the current available output of the wind power cluster, and grid-side operational constraints. The power prediction engine module 203 uses an LSTM (Long Short-Term Memory) neural network model to perform short-term power prediction for the wind power cluster based on historical wind speed-power data and real-time meteorological tower data (prediction time domain: 15 minutes to 4 hours, time resolution: 15 minutes). The prediction engine runs periodically (updated every 15 minutes), and the prediction results are input into the dynamic adjustment strategy engine to make the dispatch scheme forward-looking. Model input features include: wind speed sequence over the past 24 hours, wind... The system uses time series data, temperature and pressure data, historical power output data of each wind turbine, and numerical weather forecast data. The LSTM inference module has 3 layers, with 128 hidden units per layer. The training data uses historical data from the past 12 months, and the 15-minute prediction error target is ≤8% of rated power (significantly lower than the 15%-20% of traditional continuous prediction methods). The historical database 205 uses a domestic time series database (such as TDengine) to store all real-time acquired data, scheduling instructions, control variables, and operation logs. The data retention period is no less than 5 years, providing training data and backtesting verification data for the power prediction engine. The Human-Machine Interface (HMI) 206 is designed based on a domestically developed browser and a B / S architecture. Maintenance personnel can monitor the wind power cluster's operating status in real time, view scheduling plans, configure control parameters, query historical data, and generate operation reports through the web interface. The Power Scheduling Execution Layer 3 deploys power scheduling execution terminals, responsible for receiving power setpoints for each wind turbine from the SCADA energy management layer and accurately executing power commands to each wind turbine converter through closed-loop control. The execution layer of Power Scheduling Execution Layer 3 adopts a primary / backup dual-machine hot standby redundancy design. The primary and backup terminals maintain synchronization through a heartbeat detection mechanism (heartbeat cycle 200ms). When the primary terminal fails, the backup terminal automatically takes over within <500ms. The data acquisition layer 4 deploys a wind power data acquisition interface device, which is responsible for collecting various operating data from field equipment in real time and uploading them to the SCADA energy management layer and power dispatch execution layer; the data acquisition interface device is a domestic PLC module similar to a server, which includes four core modules: (1) Multi-protocol communication adapter module - supports Modbus TCP / RTU, IEC 61850, IEC 104, DNP3, OPC Automatic identification and dynamic adaptation of six protocols including UA; (2) Data acquisition and preprocessing module - collects data such as wind turbine power, wind speed, and grid parameters at configurable cycles, and completes CRC verification, sliding window mid-range filtering, GPS / BeiDou time synchronization and data format standardization locally; (3) Domestic hardware platform module - based on domestic ARM / Loongson processor and domestic real-time operating system, configured with redundant network ports and redundant power supplies; (4) Data upload and interface output module - built-in 512MB ring buffer supports breakpoint resume transmission, and data transmission adopts SM4 national cryptographic encryption; The data acquisition layer 4 also deploys edge computing nodes, which are responsible for completing real-time data quality assessment, intelligent diagnosis of abnormal values ​​and local emergency protection decisions at locations close to the site. When a serious fault (such as overspeed or overtemperature) reported by the wind turbine converter is detected, the edge computing node can directly trigger local protection actions without waiting for the SCADA energy management layer 2 instruction, with a response time of <50ms.

[0017] Preferably, the power scheduling execution terminal 11 includes: The instruction receiving and communication module 111 receives the power setting values ​​of each wind turbine sent by the EMS and sends the control quantities to each wind turbine converter. Power allocation module 112 is used to perform local allocation according to a dynamic proportional algorithm when the EMS only issues a total power command (rather than the individual wind turbine values); The execution terminal PID control calculation module 113 is used to perform closed-loop adjustment of the deviation between the actual output power of each fan and the set value; Status monitoring and protection module 114 is used to monitor communication status, power deviation and equipment alarms, and has automatic fault switching and safety interlock functions; Specifically, the power control quantity output by the execution terminal PID control calculation module 113 is sent to the power setting port of the wind turbine converter through the downlink interface of the instruction receiving and communication module 111 (the downlink interface includes 4 Ethernet channels + 2 RS485 channels) in the protocol format corresponding to each wind turbine. Furthermore, ① the instruction receiving and communication module 111 receives the power setting value sent by the dispatcher; ② the power distribution module 112 is connected to the instruction receiving and communication module 111; ③ the execution terminal PID control calculation module 113 receives the power setting value and outputs the control quantity; and ④ the status monitoring and protection module 114 is connected to all three modules, monitors the communication status, power deviation, and equipment alarms, and triggers master / slave switching or limiting protection in case of a fault.

[0018] Preferably, the data upload and interface output module 10 in the data acquisition layer 4 uploads the preprocessed standardized data frames to the communication front-end of the domestic SCADA master station server 201 via a dual-path gigabit Ethernet uplink (main / backup). The communication protocol adopts IEC 61850 MMS or OPC UA, and the data is transmitted using SM4 national cryptographic encryption. At the same time, the data upload and interface output module 10 in the data acquisition layer 4 also sends the actual power feedback data of the wind turbine to the power scheduling execution terminal 11 through the internal bus bypass interface 101, for use by the PID module (this PID module is the PID for the entire scheduling command and the entire wind farm adjustment) as feedback. The energy management system 202 sends the power setpoints of each wind turbine generated by the cluster power dynamic adjustment strategy engine module 204 via gigabit Ethernet to the instruction receiving and communication module 111 of the power scheduling execution terminal 11 via IEC 61850 or Modbus TCP protocol.

[0019] As a preferred option, the cluster power dynamic adjustment strategy engine module 204 adopts a multi-objective optimization model, simultaneously optimizing the following three objectives: Objective 1: Power point tracking accuracy: The corresponding optimization formula is min Σ|Pi_actual - Pi_set|², which minimizes the deviation between the actual total output power of the wind power cluster and the grid dispatch target power. Objective 2: Fatigue load balancing: The corresponding optimization formula is min max(Li / Li_rated), which minimizes the maximum value of fatigue load utilization of each wind turbine and avoids long-term overload of a single unit. Objective 3: Real-time response to grid constraints: Maximize the output of wind power clusters while meeting the grid frequency deviation constraint Δf≤±0.2Hz and voltage constraint ΔU≤±5%Un; In the above target optimization formula: Pi_actual represents the actual output power of the i-th wind turbine (unit: MW); Pi_set represents the power setpoint of the i-th wind turbine (unit: MW); Li represents the current cumulative fatigue load of the i-th wind turbine; Li_rated represents the rated fatigue load design value of the i-th wind turbine; Δf represents the grid frequency deviation (unit: Hz), that is, the difference between the actual frequency and the rated frequency of 50Hz; ΔU represents the grid voltage deviation, where Un is the rated voltage and ΔU is the percentage of the difference between the actual voltage and the rated voltage. The constraints include: the power setting value of each wind turbine does not exceed its current available power limit; the power change rate of the wind turbine does not exceed 10% / second of the rated power (to prevent sudden power changes); the total power of the wind power cluster does not exceed the grid connection point capacity limit. The solution algorithm adopts the improved NSGA-II (non-dominated sorting genetic algorithm), with a population size of 100, 200 iterations, adaptive crossover and mutation operators, and a solution cycle of 1 second. A complete optimization solution is completed in each scheduling cycle to generate the power setting value of each wind turbine for the next scheduling cycle.

[0020] Preferably, the power prediction engine module 203 reads the wind speed sequence, wind direction sequence, temperature and pressure, and historical power output data of each wind turbine from the historical database 205 for the past 24 hours, and obtains forecast data from an external numerical weather prediction interface (the numerical weather prediction interface is the GRAPES system of the China Meteorological Administration, which stores long-term historical data). It then uses the data preprocessing and feature engineering module 2031 to perform missing value imputation, normalization, and time series feature construction (such as moving average, rate of change, etc.) on the raw data. Finally, it outputs a standardized multidimensional feature matrix and sends it to the LSTM inference module 2032 for LSTM inference. The model inference module 2032 receives the standardized feature matrix output by the feature engineering module as input, performs forward inference through a pre-trained three-layer LSTM neural network model (128 hidden units per layer), and outputs the wind power cluster power prediction curve for the next 15 minutes to 4 hours with a time resolution of 15 minutes. At the same time, the model training and update module (2034) uses historical wind speed-power data from the past 12 months in the historical database 205 to periodically (e.g., once a month) retrain the LSTM inference module 2032 offline, updates the model weight parameters, and then deploys them to the LSTM inference module 2032 to adapt to seasonal changes in wind conditions.

[0021] This invention also discloses a method for dynamic power regulation of wind power clusters based on a domestically developed SCADA system, which includes the following steps: Step S1, Real-time Data Acquisition: The data acquisition interface device automatically identifies the communication protocol type of each field device through the multi-protocol communication adapter module 7, and collects the active power, reactive power, voltage, current and speed data of each wind turbine converter, the wind speed, wind direction, temperature and air pressure data of the wind tower sensor, and the grid connection point power and frequency data of the power grid metering device in real time at the configured period (default 100ms). Step S2, Data Preprocessing: The data acquisition and preprocessing module 8 in the data acquisition interface device performs the following preprocessing on the raw data: (a) CRC check, which records the packet loss rate by discarding data frames that fail the check; (b) Sliding window median filtering: by obtaining 5 sampling points of the window size, outliers that deviate from the median by more than 3 times the standard deviation are removed; (c) Timestamp alignment: All data are appended with a unified UTC timestamp using a GPS / BeiDou dual-mode timing module, with an accuracy of ≤1ms; (d) Data format standardization: raw data in different protocol formats are uniformly converted into standardized data frames containing device ID, data type, timestamp, value and quality code. The preprocessed data is simultaneously uploaded to the SCADA energy management layer and power scheduling execution layer 3. The aforementioned "raw data" refers to the data collected during the real-time data acquisition phase in step S1, including the active power, reactive power, voltage, current, and speed of each wind turbine converter, the wind speed, wind direction, air temperature, and air pressure from the anemometer tower sensors, and the power and frequency data at the grid connection point of the power grid metering device. Step S2 performs preprocessing operations such as CRC verification, sliding window value filtering, GPS / BeiDou timestamp alignment, and data format standardization based on the raw data collected in step S1. Step S3, Power Prediction: The power prediction engine module 203 of the SCADA energy management layer reads the wind speed and power time series of the past 24 hours and the current real-time wind measurement data from the historical database 205. Combined with numerical weather forecast information, it generates the wind power cluster power prediction curve for the next 15 minutes to 4 hours through a three-layer LSTM neural network model. The prediction engine runs once every 15 minutes and sends the prediction results to the strategy engine. Step S4, Multi-objective Optimization Solution: The cluster power dynamic adjustment strategy engine module 204 receives the following inputs: grid AGC / AVC command (target power), power prediction results, current available power and fatigue load status of each wind turbine, and real-time frequency and voltage on the grid side. The strategy engine constructs a multi-objective optimization model (objectives: optimal power tracking accuracy, optimal fatigue load balance, and satisfaction of grid constraints). Under the constraints (power upper limit of each wind turbine, power change rate limit, and grid connection capacity limit), the improved NSGA-II algorithm is used to solve the Pareto optimal solution set (since Pareto is derived from the Italian economist Pareto, the "Pareto optimal solution set" refers to a set of solutions that cannot improve a certain objective without harming other objectives, i.e., the optimal compromise solution set between objectives, also known as the "Pareto front"). The TOPSIS method then selects the single optimal solution from it, and selects the optimal solution from the Pareto front using the TOPSIS method (approximation ideal solution sorting method) to generate the power setpoint of each wind turbine. The entire solution process is completed within 1 second. Step S5, Dynamic Power Allocation and Distribution: The power setting values ​​of each wind turbine generated by the strategy engine are distributed to the power scheduling execution terminal 11 via EMS. When a communication failure results in only the total power command being distributed, the power allocation module 112 of the execution terminal performs local allocation according to the dynamic proportional algorithm Pi_set = P_total × (Pi_avail / ΣPi_avail). When a wind turbine exits due to a fault, the power allocation module 112 automatically excludes it and recalculates the setting values ​​of the remaining wind turbines. In step S5, the dynamic proportional allocation formula Pi_set = P_total × (Pi_avail / ∑Pi_avail) has the following meanings for each symbol: Pi_set: The power setting value (unit: MW) of the i-th wind turbine, that is, the target output power allocated to this wind turbine; P_total: The total power command (unit: MW) issued by the power grid dispatch center for the wind power cluster, which is the total output power target that the wind power cluster needs to achieve; Pi_avail: The current available power of the i-th wind turbine (unit: MW), that is, the maximum power that the wind turbine can output under the current wind conditions and equipment status; ∑Pi_avail: The sum of the available power of all operating wind turbines (unit: MW), which is the current total available output of the cluster.

[0022] The physical meaning of this formula is: to dynamically allocate the total power command according to the proportion of the available power of each wind turbine to the total available power, so that the wind turbine with larger available power undertakes more power generation tasks, thereby achieving load balancing; Step S6, PID Closed-Loop Precise Control: The PID control calculation module 113 of the execution terminal executes closed-loop control independently for each wind turbine with a control cycle of 50ms. The wind turbine power setpoint is the target value and the actual output power is the feedback value. The power deviation e(t) is calculated and the PID calculation control formula u(t) = Kp·e(t) + Ki·∫e(t)dt + Kd·de(t) / dt (Formula 1) is executed. The control quantity u(t) is sent to the corresponding wind turbine converter through the communication interface. The PID parameters support online self-tuning: when the system detects that the adjustment time or overshoot of the power step response does not meet the index, the Kp, Ki, and Kd parameters are automatically fine-tuned. In step S6, u(t) in the PID control formula represents the control quantity, that is, the power adjustment command (unit: MW) output by the PID controller at time t, which is sent to the wind turbine converter to adjust its output power. e(t) represents the power deviation, which is the difference between the setpoint power of the wind turbine at time t and the actual output power (e(t) = Pi_set - Pi_actual, unit: MW). Kp represents the proportional gain coefficient, which determines the proportional relationship between the control quantity and the current deviation. The larger Kp is, the faster the response but the more prone to oscillation. Ki represents the integral gain coefficient, which determines the relationship between the control quantity and the cumulative deviation, and is used to eliminate steady-state deviation. Kd represents the differential gain coefficient, which determines the relationship between the control quantity and the rate of change of the deviation, and is used to suppress oscillations and improve stability. ∫e(t)dt represents the integral of the power deviation over time, reflecting the historical cumulative deviation; de(t) / dt represents the derivative (rate of change) of the power deviation with respect to time, reflecting the trend of the deviation. Step S7: Multi-level security protection. The entire system implements security protection at the following three levels: (a) Data acquisition layer 4: When the edge computing node detects a serious fault in the wind turbine (overspeed, overtemperature, converter fault), it directly triggers local protection actions (power limit or emergency shutdown), with a response time of <50ms; (b) Power scheduling execution layer 3: When the status monitoring and protection module 114 detects that the packet loss rate of the communication link exceeds the threshold (5%), it automatically switches to the backup link. When it detects that the power deviation continues to exceed the limit (>3% of the rated power for 10 seconds), it triggers the amplitude limiting protection. When it detects a serious hardware failure, it triggers the main / backup terminal switching (switching time <500ms). (c) SCADA Energy Management Layer: When the energy management system 202 detects that the total power deviation of the wind power cluster exceeds the allowable range of the power grid dispatch, it automatically reduces the target power to the safety limit and reports the abnormal status to the power grid dispatch center 6; Step S8: Continuously execute steps S1 to S7 in a loop to form a complete closed-loop dynamic adjustment process.

[0023] As a preferred option, when the entire system is running normally, the end-to-end delay (end-to-end refers to the time from data acquisition to the control quantity being sent to the converter) from step S1 to step S6 is ≤200ms, which meets the requirements of power grid dispatch for rapid power response.

[0024] The details not elaborated in the above algorithm are all well-known technologies in this field and will not be repeated here.

[0025] System overall deployment example Taking a certain offshore wind power cluster as an example, the cluster includes 3 offshore wind farms, a total of 150 5MW wind turbine generators, and a total installed capacity of 750MW. The system is deployed according to a five-layer architecture as follows: On-site equipment layer 5: Each of the 150 wind turbines is equipped with one converter (80 of which use Modbus TCP protocol, 50 use IEC 61850 protocol, and 20 use DNP3 protocol), 3 wind measurement towers (each equipped with wind speed and direction sensors, temperature and pressure sensors), 3 sets of grid metering devices (installed at the grid connection points of each wind farm), and 1 offshore substation (equipped with protection devices). Data Acquisition Layer 4: Two data acquisition interface devices (one primary and one backup) are deployed in each wind farm, totaling six devices. Each device connects to 50 wind turbine converters, one meteorological tower, and one grid metering device in the wind farm via four RJ45 Gigabit Ethernet ports and four RS485 serial ports. The device's multi-protocol communication adaptation module automatically identifies the communication protocol type of each device upon initial power-up (identification time < 5 seconds / device), requiring no manual configuration. The device collects active power, reactive power, voltage, current, and speed of each wind turbine at a 100ms cycle, wind speed, wind direction, air temperature, and air pressure at the meteorological tower at a 1-second cycle, and power and frequency at the grid connection point at a 100ms cycle. After preprocessing, the collected data is uploaded to the SCADA master station via the primary communication port, while the backup communication port is in hot standby mode.

[0026] Edge computing nodes: One edge computing node is deployed in each wind farm, for a total of three nodes. The nodes receive the data stream from the data acquisition device of the wind farm and evaluate the data quality (packet loss rate, outlier ratio) in real time. When an overspeed fault (speed exceeds 110% of the rated value) or overtemperature fault (winding temperature exceeds 155℃) is detected in a wind turbine converter, a power limiting or emergency shutdown command is directly issued to the corresponding converter, with a response time of <50ms.

[0027] Power Dispatch Execution Layer 3: Two power dispatch execution terminals (one primary and one backup) are deployed at the offshore booster station. The primary and backup terminals maintain heartbeat synchronization via an independent Ethernet link (heartbeat cycle 200ms). The primary terminal receives the power setpoints of each wind turbine from the SCADA energy management layer 2, parses them through the instruction receiving and communication module 111, and sends them to the execution terminal PID control calculation module 113. The execution terminal PID control calculation module 113 maintains an independent control loop for each wind turbine and performs closed-loop control calculation with a control cycle of 50ms. The control quantity is sent to the converter through the downlink communication interface in the protocol format corresponding to each wind turbine. When the hardware watchdog of the primary terminal times out (timeout time is 3 seconds) or the heartbeat is interrupted, the backup terminal automatically takes over all control loops within <500ms.

[0028] SCADA Energy Management Layer 2: A domestically produced server cluster (using Phytium S2500 processor, 128GB memory, and Kylin V10 operating system) is deployed in the onshore control center to run SCADA master station software, EMS software, power prediction engine, and dynamic adjustment strategy engine. The domestically produced SCADA master station server 201 refreshes the real-time database at a 100ms cycle, supporting concurrent processing of more than 3,000 measurement points (150×20=3000). After receiving the grid AGC command, the EMS sends the target power, prediction results, and current status to the strategy engine. The NSGA-II algorithm of the strategy engine performs optimization and solution at a 1-second cycle, outputting power setpoint schemes for 150 wind turbines. Grid Dispatch Layer 1: Connects to the provincial grid dispatch center via the IEC 104 protocol, receives AGC / AVC commands, and reports the real-time power and unit status of the wind power cluster.

[0029] Example 2 illustrates the specific execution flow of the dynamic adjustment method. Taking the example of the power grid dispatch center 6 issuing an AGC command requiring the wind power cluster to adjust its output power from 500MW to 400MW in a stepwise manner, the specific execution flow of the method of the present invention is explained as follows: Steps S1-S2: The data acquisition interface device continuously collects real-time power data of 150 wind turbines and wind speed data of each meteorological tower. After preprocessing, the data is uploaded to the SCADA energy management layer 2. The current operating status of each wind turbine is as follows: 120 wind turbines are operating normally (total available power 620MW), and 30 wind turbines are operating with limited power due to low wind speed (total available power 80MW). The total available power of the cluster is 700MW, and the current actual output is 500MW. Step S3: The power prediction engine reads the wind speed time series of the most recent 24 hours and the current real-time wind speed data, and predicts that the wind speed will drop from the current 9.5 m / s to 8.8 m / s in the next 15 minutes. The corresponding available power of the cluster will drop from 700 MW to about 650 MW. This prediction result is sent to the strategy engine. Step S4: The strategy engine receives input—target power 400MW, current available power and fatigue load status of each wind turbine, predicted wind speed decline trend, and grid-side frequency 50.02Hz (normal range). The strategy engine constructs and solves a multi-objective optimization model: Under the premise of satisfying the upper limit of power and power change rate constraints of each wind turbine, with power tracking accuracy, fatigue load balancing, and grid constraint response as objectives, the NSGA-II algorithm is used to obtain the Pareto optimal solution set, and the TOPSIS method is used to select the comprehensive optimal solution. The final solution is: for 120 wind turbines... For the normally operating wind turbines, 400MW×(620 / 700)=354MW is allocated to the normal wind turbines according to the available power ratio, and the set value of each wind turbine is 354 / 120=2.95MW; for the 30 power-limited wind turbines, 400MW×(80 / 700)=46MW is allocated. The set value of each wind turbine is further subdivided according to its own available power ratio. At the same time, the strategy engine considers fatigue load balancing—appropriately reducing the power set value for units with a long cumulative operating time in recent times, and appropriately increasing the power set value for units with a short operating time, with the deviation range within ±5%; Step S5: The EMS sends the power setpoints for the aforementioned 150 wind turbines to the power dispatch execution terminal. The terminal's instruction receiving module parses the setpoints and distributes them to each PID control loop. Step S6: The terminal PID control calculation module 113 executes closed-loop control with a period of 50ms. Taking one of the wind turbines as an example: the set value is 2.95MW, the current actual output is 3.33MW (=500 / 150), the power deviation e(t)=2.95-3.33=-0.38MW, the PID controller outputs a negative control quantity, which is sent to the wind turbine converter through the Modbus TCP protocol. The converter gradually reduces the output power. After about 1.5 seconds (3 control cycles × 50ms × 10 iterations), the actual output power of the wind turbine stabilizes at around 2.95MW, and the steady-state deviation is <0.015MW (0.3% of rated power). Step S7: The status monitoring module continuously monitors the status of each control loop. It detects that the total power of the cluster smoothly decreases from 500MW to 400MW, and the power deviation of each wind turbine is within the allowable range, and the communication link is normal. If the communication of a wind turbine converter is interrupted during this process, the status monitoring module will automatically mark the wind turbine as offline after detecting 3 consecutive communication timeouts. The power allocation module will exclude it and reallocate the setpoints of the remaining wind turbines. The entire scheduling process has a full-link delay of <200ms, the adjustment time from the issuance of the AGC command to the wind power cluster power stabilizing at the target value is <2 seconds, and the steady-state deviation is <0.5% of the rated power.

[0030] Example 3 is for handling abnormal operating conditions. Scenario 1 – Primary Terminal Failure: When the primary terminal's hardware watchdog times out or heartbeat is interrupted, the backup terminal automatically takes over all control loops after detecting three consecutive lost heartbeats (600ms). During the takeover process, the backup terminal uses the last synchronized control state as the initial state, the PID controller switches seamlessly, and power fluctuations are less than 1% of rated power.

[0031] Scenario 2 – Communication Link Interruption: When the data acquisition device detects that the packet loss rate of the main communication port exceeds 5%, it automatically switches to the backup communication port with a switching delay of less than 50ms. During the switching process, the device writes the acquired data to a circular buffer. After the network recovers, it automatically resumes the transmission from the point of interruption, ensuring zero data loss.

[0032] Scenario 3 - Severe wind turbine failure: The edge computing node detects that a wind turbine converter reports an overspeed fault and issues an emergency shutdown command to the converter within <50ms. At the same time, it reports to the SCADA energy management layer 2. After receiving the fault information, the strategy engine automatically excludes the wind turbine and re-solves the optimal allocation scheme within the next 1-second solution cycle.

[0033] The main innovative points of this invention are as follows: (1) A five-layer hierarchical system scheme is adopted, which covers a complete domestic wind power cluster power control system architecture including grid dispatch layer 1, SCADA energy management layer 2, power dispatch execution layer 3, data acquisition layer 4, and field equipment layer 5. It integrates a multi-objective optimized cluster power dynamic adjustment strategy engine, and optimizes three objectives: power tracking accuracy, fatigue load balance, and grid constraint response. The improved NSGA-II algorithm is used to solve the problem and the TOPSIS method is used to select the best solution. (2) Simultaneously, a short-term prediction method based on LSTM wind speed-power coordination: using a three-layer LSTM neural network model, inputting multi-dimensional features (wind speed, wind direction, temperature, air pressure, historical power, weather forecast), predicting the power curve from 15 minutes to 4 hours, with a 15-minute prediction error ≤ 8% of rated power; (3) Full closed-loop dynamic adjustment method flow (using the seven-step method of steps S1-S7): a complete method flow from data acquisition and preprocessing, power prediction, multi-objective optimization solution, dynamic proportional allocation, PID closed-loop control to multi-level safety protection, with a full-link delay of <200ms and a steady-state deviation of ≤0.5% of rated power. (4) Three-level security protection mechanism: local protection of edge nodes in the data acquisition layer (<50ms), hot standby switching of primary / backup terminals in the execution layer (<500ms) and automatic switching of communication link redundancy, automatic adjustment of EMS power security limit in the management layer, and a scheme design for the coordinated operation of three-level protection. (5) Achieve full domestic technology stack: From processor (Phytium / Kunpeng / ARM / Loongson), operating system (Kylin / Tongxin UOS / RT-Thread), database (TDengine) to encryption algorithm (SM4 national cryptography), achieve full domestic production of the entire system; The present invention achieves the following technical effects: ① Full-chain localization: From field data acquisition devices (domestic processors + domestic real-time operating systems), SCADA master station (domestic servers + domestic operating systems + independent software) to dispatch execution terminal (domestic industrial control processors), the entire wind power cluster power regulation system is fully localized, completely eliminating dependence on imported SCADA / EMS platforms and meeting the requirements for independent control of critical infrastructure; ② Improved dispatch accuracy: The multi-objective optimization dispatch strategy combined with PID closed-loop control reduces the steady-state deviation of wind power cluster power tracking from 2%-5% of rated power in the traditional scheme to ≤0.5% of rated power, and shortens the response time from 5-10 seconds to ≤2 seconds, significantly improving the grid dispatch response capability.

[0034] ③ Proactive scheduling: The LSTM-based short-term power prediction engine enables the scheduling scheme to have a forward-looking capability of 15 minutes to 4 hours. The 15-minute prediction error is ≤8% of the rated power (which is about 50% lower than the traditional continuous prediction method), allowing the system to adjust power allocation in advance to cope with changes in wind conditions.

[0035] ④ Fatigue load balancing: The multi-objective optimization strategy incorporates the fatigue load balancing of the wind turbine into the optimization objective, avoiding long-term overload operation of a single unit. It is expected to extend the service life of key components of the unit (gearbox, main bearing) by 15%-20%, and significantly reduce operation and maintenance costs.

[0036] ⑤ Data quality assurance: The data acquisition interface device performs verification, filtering, timestamp alignment and format standardization locally, improving the data effectiveness rate from about 92% in traditional solutions to over 99.5%, providing a high-quality data source for upper-level scheduling strategies.

[0037] ⑥ High reliability: The system has a multi-level redundancy and protection mechanism, including hot standby switching of primary / backup terminals (<500ms), automatic switching of redundant communication links, local emergency protection of edge nodes (<50ms), and data interruption resumption, to ensure high reliability of the system in harsh marine environments. The system availability rate is ≥99.95%.

[0038] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A wind power cluster power regulation system based on domestically produced SCADA, characterized in that: The system includes, from top to bottom, the power grid dispatch layer (1), the SCADA energy management layer (2), the power dispatch execution layer (3), the data acquisition layer (4), and the field equipment layer (5). The SCADA energy management layer (2) is the core decision-making layer of the system, deployed on a domestic server cluster at an offshore booster station or onshore control center. This layer includes the following key components: domestic SCADA master station server (201), energy management system (202), power prediction engine module (203), cluster power dynamic adjustment strategy engine module (204), historical database (205) providing training and backtesting data for the prediction engine, and human-machine interface (206). The domestic SCADA master station server (201), energy management system (202), power prediction engine module (203), cluster power dynamic adjustment strategy engine module (204), historical database (205), and human-machine interface (206) are interconnected. The SCADA master station server (201) has its own real-time database (2011) to provide a unified data source for the power prediction engine, strategy engine, and human-machine interface (206). The power prediction engine module (203) uses an LSTM neural network model to perform short-term power prediction of the wind power cluster based on historical wind speed-power data and real-time wind tower data, and sends the prediction to the cluster power dynamic adjustment strategy engine module (204). The cluster power dynamic adjustment strategy engine module (204) uses a multi-objective optimization model and the total power command, power prediction results and current operating status of the energy management system (202) to solve and generate the optimal power allocation scheme for each wind turbine. The power grid dispatch layer (1) is connected to the power grid dispatch center (6) and the energy management system (202) to realize downlink transmission of AGC / AVC commands and uplink transmission of real-time power and unit status information of wind power clusters; the power dispatch execution layer (3) is connected to the SCADA energy management layer (2) and the field equipment layer (5); The field equipment layer (5) includes all field physical equipment in the wind power cluster, including converters, wind tower sensors, grid metering devices and booster station protection devices of each wind turbine group, and each device has a standard communication interface to connect to the data acquisition layer (4) via Ethernet or RS485 serial port.

2. The wind power cluster power regulation system based on domestically produced SCADA as described in claim 1, characterized in that: The power prediction engine module (203) includes a data preprocessing and feature engineering module (2031), an LSTM inference module (2032), a prediction result output and evaluation module (2033), and a model training module connected sequentially to the data preprocessing and feature engineering module (2031), the LSTM inference module (2032), and the prediction result output and evaluation module (2033). The prediction result output and evaluation module (2033) outputs and connects to the cluster power dynamic adjustment strategy engine module (204) and the historical database (205), enabling the model training and update module (2034) to obtain training data from the historical database (205). After training, the updated model parameters are deployed to the LSTM inference module (2032), and the prediction curve output by the prediction result output and evaluation module (2033) is sent to the cluster power dynamic adjustment strategy engine module (204) as input for optimization. The solution results of the cluster power dynamic adjustment strategy engine module (204) are distributed by the energy management system (202). The cluster power dynamic adjustment strategy engine module (204) includes a multi-source data access interface (2041), a multi-objective optimization model construction module (2042), an NSGA-II solution algorithm module (2043), a TOPSIS scheme selection module (2044), and a result output interface (2045). The energy management system (202), the power prediction engine module (203), and the real-time database (2011) are connected to the multi-objective optimization model construction module (2042), the NSGA-II solution algorithm module (2043), and the TOPSIS scheme selection module (2044) in sequence through the multi-source data access interface (2041), and are connected to the distribution channel of the energy management system (202) and the historical database (205) through the result output interface (2045). The engine is responsible for solving and generating the optimal power allocation scheme for each wind turbine based on the total power command, power prediction results, and current operating status of the energy management system (202). The power grid dispatch layer (1) is located at the top of the system and is connected to the power grid dispatch center (6). It receives the AGC active power command and AVC reactive power command issued by the power grid and reports the real-time power generation information of the wind power cluster to the power grid dispatch center (6). It also performs bidirectional data interaction with the energy management system (202) of the SCADA energy management layer (2) through the IEC 104 protocol or dispatch data network. The core of the data acquisition layer (4) is the data acquisition interface device. The data acquisition interface device contains a multi-protocol communication adapter module (7), a data acquisition and preprocessing module (8), a domestic main control module (9), and a data upload and interface output module (10) interconnected by an internal bus. The power dispatch execution layer (3) is equipped with a power dispatch execution terminal (11) and is responsible for receiving the power setting values ​​of each wind turbine issued by the SCADA energy management layer (2). It also executes the power command accurately to each wind turbine converter in the field equipment layer (5) through closed-loop control.

3. A wind power cluster power regulation system based on domestically produced SCADA as described in claim 2, characterized in that: The power scheduling execution terminal (11) includes: Receive the power setting values ​​of each wind turbine sent by EMS, and send the control quantity to the instruction receiving and communication module (111) of each wind turbine converter. Power allocation module (112) used to perform local allocation according to dynamic ratio algorithm when EMS only issues total power command. The execution terminal PID control calculation module (113) is used to perform closed-loop adjustment of the deviation between the actual output power of each wind turbine and the set value. Status monitoring and protection module (114) is used to monitor communication status, power deviation and equipment alarms, and has automatic fault switching and safety interlock functions. The power control quantity output by the execution terminal PID control calculation module (113) is sent to the power setting port of the wind turbine converter through the downlink interface of the instruction receiving and communication module (111) in the protocol format corresponding to each wind turbine.

4. A wind power cluster power regulation system based on domestically produced SCADA as described in claim 3, characterized in that: The data upload and interface output module (10) in the data acquisition layer (4) uploads the pre-processed standardized data frames to the communication front-end of the domestic SCADA master station server (201) through the main / backup dual gigabit Ethernet uplink. The communication protocol adopts IEC 61850 MMS or OPC UA, and the data is transmitted with SM4 national cryptographic encryption. At the same time, the data upload and interface output module (10) in the data acquisition layer (4) sends the actual power feedback data of the wind turbine to the power scheduling execution terminal (11) through the internal bus bypass interface (101) for the PID module to use as feedback. The energy management system (202) sends the power setpoint of each wind turbine generated by the cluster power dynamic adjustment strategy engine module (204) through gigabit Ethernet to the instruction receiving and communication module (111) of the power scheduling execution terminal (11) through IEC 61850 or Modbus TCP protocol.

5. A wind power cluster power regulation system based on domestically produced SCADA as described in claim 2, characterized in that: The cluster power dynamic adjustment strategy engine module (204) adopts a multi-objective optimization model, simultaneously optimizing the following three objectives: Objective 1: Power point tracking accuracy: min Σ|Pi_actual - Pi_set|², to minimize the deviation between the actual total output power of the wind power cluster and the grid dispatch target power; Objective 2: Fatigue load balancing: min max(Li / Li_rated), to minimize the maximum value of fatigue load utilization of each wind turbine, and avoid long-term overload of a single unit; Objective 3: Real-time response to grid constraints: Maximize the output of wind power clusters while satisfying the grid frequency deviation constraint Δf≤±0.2Hz and the voltage constraint ΔU≤±5%Un. In the above target optimization formula: Pi_actual represents the actual output power of the i-th wind turbine; Pi_set represents the power setting value of the i-th wind turbine; Li represents the current cumulative fatigue load value of the i-th wind turbine; Li_rated represents the rated fatigue load design value of the i-th wind turbine; Δf represents the grid frequency deviation, i.e., the difference between the actual frequency and the rated frequency of 50Hz; ΔU represents the grid voltage deviation, where Un is the rated voltage and ΔU is the percentage difference between the actual voltage and the rated voltage.

6. A method for dynamic power adjustment of wind power clusters based on a domestically developed SCADA architecture, as described in claim 2, 3, 4, or 5, characterized in that: The power prediction engine module (203) reads the wind speed sequence, wind direction sequence, temperature and pressure, and historical power output data of each wind turbine from the historical database (205) for the past 24 hours, and obtains forecast data from external numerical weather prediction. It uses the data preprocessing and feature engineering module (2031) to perform missing value imputation, normalization and time series feature construction on the original data, and then outputs a standardized multi-dimensional feature matrix and sends it to the LSTM inference module (2032). The LSTM inference module (2032) receives the standardized feature matrix output by the feature engineering module as input, performs forward inference through a pre-trained three-layer LSTM neural network model, and outputs the wind power cluster power prediction curve for the next 15 minutes to 4 hours with a time resolution of 15 minutes. At the same time, the model training and update module (2034) uses the historical wind speed-power data of the past 12 months in the historical database (205) to periodically retrain the LSTM inference module (2032) offline, updates the model weight parameters and deploys them to the LSTM inference module (2032) to adapt to seasonal changes in wind conditions.

7. A method for dynamic power regulation of wind power clusters based on a domestically developed SCADA system, characterized in that, Includes the following steps: Step S1, Real-time data acquisition: The data acquisition interface device automatically identifies the communication protocol type of each field device through the multi-protocol communication adapter module (7) and acquires the active power, reactive power, voltage, current and speed data of each wind turbine converter, the wind speed, wind direction, temperature and air pressure data of the wind tower sensor, and the grid connection point power and frequency data of the power grid metering device in real time according to the configuration period. Step S2, Data Preprocessing: The data acquisition and preprocessing module (8) in the data acquisition interface device performs the following preprocessing on the raw data: (a) CRC check, which records the packet loss rate by discarding data frames that fail the check; (b) Sliding window median filtering: by obtaining 5 sampling points of the window size, outliers that deviate from the median by more than 3 times the standard deviation are removed; (c) Timestamp alignment: All data are appended with a unified UTC timestamp using a GPS / BeiDou dual-mode timing module, with an accuracy of ≤1ms; (d) Data format standardization: By converting raw data of different protocol formats into standardized data frames containing device ID, data type, timestamp, value and quality code, the pre-processed data is simultaneously uploaded to the SCADA energy management layer and power scheduling execution layer (3). Step S3, Power Prediction: The power prediction engine module (203) of the SCADA energy management layer reads the wind speed and power time series of the past 24 hours and the current real-time wind measurement data from the historical database (205). Combined with numerical weather forecast information, it generates the wind power cluster power prediction curve for the next 15 minutes to 4 hours through a three-layer LSTM neural network model. The prediction engine runs once every 15 minutes and sends the prediction results to the cluster power dynamic adjustment strategy engine module (204). Step S4, Multi-objective optimization solution: The cluster power dynamic adjustment strategy engine module (204) receives the following inputs: grid AGC / AVC instructions, power prediction results, current available power and fatigue load status of each wind turbine, and real-time frequency and voltage on the grid side. The strategy engine constructs a multi-objective optimization model. Under the constraints, the improved NSGA-II algorithm is used to solve the Pareto optimal solution set, and the TOPSIS method is used to select the comprehensive optimal scheme from the Pareto front to generate the power set value of each wind turbine. The entire solution process is completed within 1 second. Step S5, Dynamic Power Allocation and Distribution: The power setting values ​​of each wind turbine generated by the cluster power dynamic adjustment strategy engine module (204) are distributed to the power scheduling execution terminal (11) via EMS. When communication is abnormal and only the total power command can be distributed, the power allocation module (112) of the execution terminal performs local allocation according to the dynamic ratio algorithm Pi_set = P_total × (Pi_avail / ΣPi_avail). When a wind turbine exits due to a fault, the power allocation module (112) automatically excludes it and recalculates the setting values ​​of the remaining wind turbines. In step S5, the dynamic proportional allocation formula Pi_set = P_total × (Pi_avail / ∑Pi_avail) has the following meanings for each symbol: Pi_set: The power setting value of the i-th fan, that is, the target output power allocated to this fan; P_total: The total power command issued by the power grid dispatch center for the wind power cluster, which is the total output power target that the wind power cluster needs to achieve; Pi_avail: The current available power of the i-th wind turbine, that is, the maximum power that the wind turbine can output under the current wind conditions and equipment status; ∑Pi_avail: The sum of the available power of all operating wind turbines, i.e., the total available output of the cluster at present; Step S6, PID Closed-Loop Precise Control: The PID control calculation module (113) of the execution terminal executes closed-loop control independently for each fan with a control cycle of 50ms. The fan power setpoint is the target value, and the actual output power is the feedback value. The power deviation e(t) is calculated and the PID calculation control formula is executed. u(t) = Kp·e(t) + Ki·∫e(t)dt + Kd·de(t) / dt (Formula 1); The control quantity u(t) is sent to the corresponding wind turbine converter through the communication interface; In Formula 1: u(t) represents the control quantity, i.e., the power adjustment command (unit: MW) output by the PID controller at time t, which is sent to the wind turbine converter to adjust its output power; e(t) represents the power deviation, i.e., the difference between the wind turbine power setpoint and the actual output power at time t (e(t) = Pi_set - Pi_actual); Kp represents the proportional gain coefficient, which determines the proportional relationship between the control quantity and the current deviation. The larger Kp is, the faster the response but the more prone to oscillation; Ki represents the integral gain coefficient, which determines the relationship between the control quantity and the cumulative deviation, and is used to eliminate steady-state deviation; Kd represents the derivative gain coefficient, which determines the relationship between the control quantity and the rate of change of the deviation, and is used to suppress oscillation and improve stability; ∫e(t)dt represents the integral of the power deviation with respect to time, reflecting the historical cumulative deviation; de(t) / dt represents the derivative of the power deviation with respect to time, reflecting the trend of deviation change; Step S7: Multi-level security protection. The entire system implements security protection at the following three levels: (a) Data acquisition layer (4): When the edge computing node detects a serious fault in the wind turbine, it directly triggers the local protection action with a response time of <50ms; (b) Power scheduling execution layer (3): When the status monitoring and protection module (114) detects that the packet loss rate of the communication link exceeds the threshold (5%), it automatically switches to the backup link; when it detects that the power deviation continues to exceed the limit, it triggers the amplitude limiting protection; when it detects a serious hardware failure, it triggers the main / backup terminal switching. (c) SCADA energy management layer: When the energy management system (202) detects that the total power deviation of the wind power cluster exceeds the allowable range of grid dispatch, it automatically reduces the target power to the safety limit and reports the abnormal status to the grid dispatch center (6); Step S8: Continuously execute steps S1 to S7 in a loop to form a complete closed-loop dynamic adjustment process.

8. The method for dynamic power adjustment of wind power clusters based on a domestic SCADA architecture according to claim 7, characterized in that: When the entire system is running normally, the end-to-end delay from step S1 to step S6 is ≤200ms, which meets the requirements of power grid dispatch for rapid power response.

Images