Method, device and equipment for generating stability control strategy of wind power integration system based on digital twinning and medium

Abstract

The invention discloses a method, a device and equipment for generating a stability control strategy of a wind power grid-connected system based on digital twinning and a medium. The method comprises the following steps: acquiring historical operation data and wind power operation data of the wind power grid-connected system; constructing a digital twinborn model and acquiring real-time measurement data according to the historical operation data in combination with a topological structure of the wind power grid-connected system; calling a preset filtering algorithm to perform state estimation optimization according to the real-time measurement data, and determining an optimal system state; extracting feature state vectors from the optimal system state and the wind power operation data; and predicting the feature state vector according to a built-in strategy through a preset target reinforcement learning model, generating a stability control strategy instruction, and issuing the stability control strategy instruction to a corresponding module in the wind power grid-connected system, thereby effectively improving the adaptability of wind power grid-connected control.

Description

Technical Field

[0001] The present invention relates to the technical field of stabilization control strategy generation, and in particular to a method, device, equipment and medium for generating a stabilization control strategy for a wind power grid-connected system based on digital twins. Background Art

[0002] With the global energy transition and the advancement of the "dual carbon" goals, wind energy, as a key renewable energy source, has been widely integrated into power systems. In recent years, as wind power's share of the energy mix has continued to rise, its role has gradually shifted from supplementary to primary. In some regions, a new paradigm has emerged where wind power output dominates power system operations.

[0003] In scenarios where a high proportion of wind power is connected to the grid, the operating characteristics of the power system differ significantly from traditional models. Wind power output, influenced by natural factors such as wind speed and meteorological conditions, exhibits significant intermittency, volatility, and unpredictability, placing unprecedented pressure on system power balance and frequency regulation. Particularly in areas with concentrated wind power access, the support capacity of conventional power sources decreases, and the system inertia level decreases, which can easily lead to various problems such as frequency, voltage, and transient stability. To address this, wind power systems are typically regulated through stability control strategies. However, traditional stability control strategies rely on offline modeling and expert experience rules, making them difficult to cope with rapid changes in wind power output, multi-scenario operating conditions, and high-dimensional uncertainty. They suffer from response lag, poor control rigidity, and poor adaptability.

[0004] Therefore, existing wind power grid-connected control systems are mostly based on static optimization models, lacking high-fidelity perception and prediction capabilities of the system's dynamic evolution, making it difficult to dynamically adjust and adaptively update strategies during operation. Furthermore, some existing research has proposed using artificial intelligence methods to assist in decision-making, but these methods lack integration with power system mechanism models, posing the risk of "black box" manipulation and lacking credibility and interpretability, which in turn reduces the adaptability of wind power grid-connected control. Summary of the Invention

[0005] The present invention provides a method, device, equipment, and medium for generating a stabilization control strategy for a wind power grid-connected system based on digital twins. This addresses the problem that existing wind power grid-connected control systems are mostly based on static optimization models, lacking high-fidelity perception and prediction capabilities of the system's dynamic evolution, and making it difficult to dynamically adjust and adaptively update strategies during operation. Furthermore, some existing studies have proposed using artificial intelligence methods to assist in decision-making, but these methods lack integration with power system mechanism models, pose a "black box" risk, and lack credibility and interpretability, leading to technical issues such as reduced adaptability of wind power grid-connected control.

[0006] A first aspect of the present invention provides a method for generating a stabilization control strategy for a wind power grid-connected system based on digital twins, characterized in that the method comprises:

[0007] Acquiring historical operating data and wind power operating data of the wind power grid-connected system;

[0008] Constructing a digital twin model according to the historical operating data and the topological structure of the wind power grid-connected system, and obtaining real-time measurement data within the digital twin model;

[0009] Calling a preset filtering algorithm to perform state estimation optimization according to the real-time measurement data to determine the optimal system state;

[0010] extracting a characteristic state vector from the optimal system state and the wind power operation data;

[0011] The characteristic state vector is predicted according to the built-in strategy through a preset target reinforcement learning model, and a stabilization control strategy instruction is generated and sent to the corresponding module in the wind power grid-connected system.

[0012] Optionally, constructing a digital twin model and acquiring real-time measurement data according to the historical operation data in combination with the topology of the wind power grid-connected system includes:

[0013] Determining a node power balance relationship according to the topology of the wind power grid-connected system;

[0014] Constructing a physical part model according to the node power balance relationship;

[0015] Based on the error correction function and the historical operating data, a preset machine learning model is trained to obtain a data-driven model;

[0016] Associating the physical part model with the data-driven model to construct a digital twin model;

[0017] The data acquisition unit is called to obtain real-time measurement data from the digital twin model.

[0018] Optionally, calling a preset filtering algorithm to perform state estimation optimization according to the real-time measurement data to determine the optimal system state includes:

[0019] Substituting the real-time measurement data into a state estimation optimization function, and calling a preset filtering algorithm to solve the state estimation optimization function to determine the optimal system state;

[0020] The state estimation optimization function is:

[0021]

[0022] in, is the real-time measurement data at time t, is the system state at time t-1, is the stabilization strategy instruction at time t-1, is the modeling function of the physical part model, Convert the real state of the system into real-time measurement data The nonlinear transfer function of the same form, is the optimal system state, is the weight coefficient.

[0023] Optionally, the method further includes:

[0024] When the generation cycle of the stabilization control strategy instruction is equal to the preset optimization cycle, calling the rolling optimizer to modify the stabilization control strategy instruction, generating a new stabilization control strategy instruction and issuing it to the corresponding module in the wind power grid connection system;

[0025] The optimization goal of the rolling optimizer is:

[0026]

[0027] in, for The optimal system state at the moment In execution Stable control strategy instructions at all times After the system steady-state deviation loss, for -1 moment stabilization strategy instruction, To predict the rolling window length, It is the penalty coefficient for regulating the action change rate.

[0028] Optionally, the method further includes:

[0029] Call the preset wind speed prediction model to predict the system's predicted operating status in the future;

[0030] Calculating the deviation probability between the predicted operating state of the system and the preset safety domain;

[0031] If the deviation probability is greater than a preset correction threshold, a backup strategy instruction in a preset candidate strategy pool is called and sent to a corresponding module in the wind power grid connection system.

[0032] Optionally, the method further includes:

[0033] Obtain the initial reinforcement learning model and initialize the corresponding policy network;

[0034] Generate training strategy instructions based on the predicted operation state and the real-time operation state output by the digital twin model through the initial reinforcement learning model, combined with the strategy network, and output the training strategy instructions to the digital twin model to update the predicted operation state;

[0035] Calculating the current reward value of the initial reinforcement learning model according to a preset reward function until a preset number of cycles is reached;

[0036] With the goal of maximizing the weighted cumulative value of the current reward value, calling the proximal policy optimization algorithm to calculate the policy gradient of the policy network and optimize the model parameters of the initial reinforcement learning model;

[0037] When the current reward value change is less than a preset convergence threshold, the initial reinforcement learning model at the current moment is determined as the target reinforcement learning model.

[0038] Optionally, the method further includes:

[0039] After the corresponding module in the wind power grid-connected system executes the stabilization control strategy instruction, control effect data is collected;

[0040] Calculating a reward error according to the control effect data and the expected effect data of the target reinforcement learning model;

[0041] According to the operation prediction state and the reward error, the parameters of the digital twin model and the target reinforcement learning model are updated, and the step of obtaining real-time measurement data in the digital twin model is jumped to execute.

[0042] A second aspect of the present invention provides a device for generating a stabilization control strategy for a wind power grid-connected system based on digital twins, the device comprising:

[0043] A data acquisition module, used to acquire historical operation data and wind power operation data of the wind power grid-connected system;

[0044] A model building module, configured to build a digital twin model and obtain real-time measurement data based on the historical operating data and the topology of the wind power grid-connected system;

[0045] A state optimization module is used to call a preset filtering algorithm to perform state estimation optimization according to the real-time measurement data to determine the optimal system state;

[0046] a vector construction module, configured to extract a characteristic state vector from the optimal system state and the wind power operation data;

[0047] The instruction prediction module is used to predict the characteristic state vector according to the built-in strategy through a preset target reinforcement learning model, generate a stabilization control strategy instruction and send it to the corresponding module in the wind power grid-connected system.

[0048] The third aspect of the present invention provides an electronic device, characterized in that it includes a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the digital twin-based wind power grid-connected system stabilization strategy method as described in any one of the first aspects of the present invention.

[0049] The fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed, it implements the wind power grid-connected system stabilization strategy method based on digital twin as described in any one of the first aspects of the present invention.

[0050] It can be seen from the above technical solutions that the present invention has the following advantages:

[0051] The present invention obtains historical operating data and wind power operating data of the wind power grid-connected system; constructs a digital twin model and obtains real-time measurement data according to the historical operating data combined with the topological structure of the wind power grid-connected system; calls a preset filtering algorithm to perform state estimation optimization according to the real-time measurement data to determine the optimal system state; extracts characteristic state vectors from the optimal system state and wind power operating data; predicts the characteristic state vectors according to the built-in strategy through a preset target reinforcement learning model, generates stabilization control strategy instructions and sends them to the corresponding module in the wind power grid-connected system, thereby interacting based on the digital twin model and the reinforcement learning model to achieve accurate generation of stabilization control strategy instructions, thereby effectively improving the adaptability of wind power grid-connected control. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0053] Figure 1 A flowchart of the steps of a method for generating a stabilization control strategy for a wind power grid-connected system based on digital twins provided in an embodiment of the present invention;

[0054] Figure 2 A structural block diagram of a wind power grid-connected system stabilization control strategy generation device based on digital twins provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0055] Embodiments of the present invention provide a method, apparatus, device, and medium for generating a stabilization control strategy for a wind power grid-connected system based on digital twins. These methods are designed to address the problem that existing wind power grid-connected control systems are mostly based on static optimization models, lack high-fidelity perception and prediction capabilities of the system's dynamic evolution, and are difficult to dynamically adjust and adaptively update during operation. Furthermore, some existing studies have proposed using artificial intelligence methods to assist decision-making, but these methods lack integration with power system mechanism models, pose a "black box" risk, and lack credibility and interpretability, leading to technical issues such as reduced adaptability of wind power grid-connected control.

[0056] In order to make the purpose, features, and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0057] See also Figure 1 , Figure 1 A flowchart of the steps of a method for generating a wind power grid-connected system stabilization control strategy based on digital twins provided in an embodiment of the present invention.

[0058] The present invention provides a method for generating a wind power grid-connected system stabilization control strategy based on digital twins, the method comprising:

[0059] Step 101, obtaining historical operation data and wind power operation data of the wind power grid-connected system;

[0060] The historical operation data includes the wind power grid-connected system at multiple historical moments, such as the power system status and control instructions at time t-1.

[0061] Wind power operation data refers to various operating variables of the environment in which wind power generation units are located within the wind power grid-connected system, including but not limited to wind speed, wind direction, scheduling constraints and predicted load.

[0062] In the embodiments of the present application, a controller acquires historical operating data and wind power operation data of the wind power grid-connected system as the data basis for subsequent system stabilization. Specifically, intelligent sensors can be installed at each wind turbine generator set and key nodes of the transmission line. These sensors can collect and upload historical operating data such as unit speed, blade angle, output voltage, and current in real time, with a collection frequency of once per second. At the same time, meteorological monitoring buoys are set up around the wind farm to continuously collect wind power operation data such as wind speed, wind direction, and air pressure. The data is updated every 10 minutes, and the data for the past three years is stored in a central database.

[0063] It should be noted that this method can be applied to devices with communication and data processing functions, such as controllers and servers.

[0064] Step 102: construct a digital twin model based on historical operating data and the topology of the wind power grid-connected system, and obtain real-time measurement data within the digital twin model;

[0065] In this embodiment, the nodes and connection lines of the wind power grid-connected system are abstracted and converted into the topology structure of the wind power grid-connected system. According to this topological structure and combined with the power flow calculation equation, the physical modeling part of the digital twin model is carried out. , then build a data-driven model based on the machine learning model, and link the two models to build a digital twin model.

[0066] After building the digital twin model, real-time measurement data is obtained from it through the data acquisition unit for subsequent estimation of the optimal system state.

[0067] In one example of the present application, step 102 may include the following sub-steps:

[0068] Determine the node power balance relationship according to the topology of the wind power grid-connected system;

[0069] Construct the physical part model based on the node power balance relationship;

[0070] Based on the error correction function and historical operating data, the preset machine learning model is trained to obtain a data-driven model;

[0071] Associate the physical part model with the data-driven model to build a digital twin model;

[0072] Call the data acquisition unit to obtain real-time measurement data from the digital twin model.

[0073] In this embodiment, after obtaining the topological structure of the wind power grid-connected system, it is represented in the form of graph theory as follows: , where N is a node set such as busbars and equipment, and E is an edge set such as lines and connection relationships. Analyze this topology and describe the node power balance relationship, namely:

[0074]

[0075]

[0076] in, is the node active power, is the node reactive power, is the voltage amplitude of the i-th node, is the voltage amplitude of the jth node, is the admittance between the i-th and j-th nodes, is the phase angle of the i-th node, is the phase angle of the j-th node.

[0077] In this embodiment, the physical part model is constructed based on the power system's power flow equation, and the parameters in the preset machine learning model are adjusted based on the error correction function and historical operation data. Training is performed to correct the prediction error of the physical model and improve modeling accuracy. The two models are linked to generate a digital twin model, which is encapsulated in the digital twin platform to achieve consistent mapping and behavior prediction of the virtual and real systems. The model is expressed as:

[0078]

[0079] in, is the predicted value of the power system state variable, is a dynamic modeling function based on the physical laws of the power system, is the device parameter set of the physical part model, For the measurement data set, is the error correction function based on the machine learning model, with parameters Obtained by training with historical operation data, is the power system state variable at time t-1, Represents the stabilization control strategy instruction at time t-1.

[0080] After the digital twin model is built, real-time measurement data is acquired from the model by invoking a data acquisition unit (DAU). This DAU can include, but is not limited to, a PMU (Phasor Measurement Unit) and a SCADA (Supervisory Control and Data Acquisition System). The PMU utilizes clock synchronization technologies like GPS to achieve high-precision, synchronized measurement of voltage and current phasors in the power system, enabling dynamic monitoring of system operating status. The DASC is an automated system within the power system used to collect real-time data, monitor equipment status, and implement remote control. It is primarily used for steady-state operation monitoring and control.

[0081] Step 103: Calling a preset filtering algorithm to perform state estimation optimization according to real-time measurement data to determine the optimal system state;

[0082] In this embodiment, after acquiring real-time measurement data, such as bus voltage amplitude and phasor, current phasor, active power, reactive power, system frequency, wind speed, wind power output forecast, and real-time load data, the real-time measurement data undergoes preprocessing, including denoising and normalization, to ensure data accuracy and consistency. Algorithms such as Kalman filtering (KF), extended Kalman filtering (EKF), or particle filtering (PF) are then used to determine the optimal system state based on the real-time measurement data and the predicted values ​​from the digital twin model.

[0083] In an example of the present application, step 103 may include the following sub-steps:

[0084] Substitute the real-time measurement data into the state estimation optimization function, and call the preset filtering algorithm to solve the state estimation optimization function to determine the optimal system state;

[0085] The state estimation optimization function is:

[0086]

[0087] in, is the real-time measurement data at time t, is the system state at time t-1, is the stabilization strategy instruction at time t-1, is the modeling function of the physical part model, Convert the real state of the system into real-time measurement data The nonlinear transfer function of the same form, is the optimal system state, is the weight coefficient. To measure the function, the updated state is used for online calibration of the twin model to improve its simulation accuracy and dynamic adaptability.

[0088] In this embodiment, the Kalman filter is applicable to nonlinear systems, using Taylor expansion linearization to iteratively update the state prediction value and covariance matrix. It can also represent the state probability distribution through a set of randomly sampled particles, making it suitable for non-Gaussian and nonlinear scenarios and more robust.

[0089] Step 104: extracting a characteristic state vector from the optimal system state and wind power operation data;

[0090] In this embodiment, feature vectors are extracted from historical and real-time data. , including wind speed, wind direction, scheduling constraints, load forecast, active power, reactive power, voltage, current, etc., to construct the characteristic state vector.

[0091] In one example of this application, training of the target reinforcement learning model can be achieved by the following steps:

[0092] Obtain the initial reinforcement learning model and initialize the corresponding policy network;

[0093] Generate training strategy instructions based on the predicted operation state and the real-time operation state output by the digital twin model through the initial reinforcement learning model, combined with the strategy network, and output the training strategy instructions to the digital twin model to update the predicted operation state;

[0094] Calculating the current reward value of the initial reinforcement learning model according to a preset reward function until a preset number of cycles is reached;

[0095] With the goal of maximizing the weighted cumulative value of the current reward value, calling the proximal policy optimization algorithm to calculate the policy gradient of the policy network and optimize the model parameters of the initial reinforcement learning model;

[0096] When the current reward value change is less than a preset convergence threshold, the initial reinforcement learning model at the current moment is determined as the target reinforcement learning model.

[0097] In this embodiment, based on the reinforcement learning algorithm, the state space S, action space A, and reward function are defined as , use the proximal policy optimization (PPO) algorithm for policy training, the goal is to maximize the expected total reward:

[0098]

[0099] Among them, the strategy Approximation through deep neural networks, is the discount rate parameter of reinforcement learning, and R is the system stability reward function.

[0100] The typical reward function is designed as follows:

[0101]

[0102] in, is the system frequency, is the system rated frequency, is the voltage amplitude, is the rated voltage amplitude, is the voltage offset threshold, To cut off the load power, is the frequency stability weight coefficient, is the voltage stability weight coefficient, It is the load power supply reliability weight coefficient.

[0103] After training is completed, the strategy is deployed to the power dispatching control center in the form of a control interface module and linked with the digital twin platform.

[0104] Step 105 : predicting the characteristic state vector according to the built-in strategy through the preset target reinforcement learning model, generating a stabilization control strategy instruction and sending it to the corresponding module in the wind power grid-connected system.

[0105] In this embodiment, the target reinforcement learning model can be a multi-layer perceptron MLP, whose input includes system operation characteristics such as wind speed, wind direction, scheduling constraints, load forecast, voltage, current, etc., and the feature vector is mapped into the control action of the stabilization strategy instruction through the policy network, that is, the hidden layer performs linear transformation, ReLU activation and Dropout in sequence to extract high-dimensional features; the output layer generates a 5-dimensional continuous value action through linear transformation, converts the above action value into an engineering executable stabilization strategy instruction and sends it to the corresponding module to respond to the rapid fluctuation of wind power output.

[0106] In one example of the present application, the method further includes the following steps S11:

[0107] S11. When the generation cycle of the stabilization control strategy instruction is equal to the preset optimization cycle, the rolling optimizer is called to modify the stabilization control strategy instruction, generate a new stabilization control strategy instruction and send it to the corresponding module in the wind power grid connection system;

[0108] The optimization goals of the rolling optimizer are:

[0109]

[0110] in, for The optimal system state at the moment In execution Stable control strategy instructions at all times After the system steady-state deviation loss, for -1 moment stabilization strategy instruction, To predict the rolling window length, It is the penalty coefficient for regulating the action change rate.

[0111] In this embodiment, model prediction and online rolling optimization are combined to Local optimization is performed within the system to cope with sudden changes in wind speed or system disturbances.

[0112] In one example of the present application, the method further includes the following steps S21-S23:

[0113] S21. Calling a preset wind speed prediction model to predict the system operating status in the future;

[0114] S22. Calculate the probability of deviation between the system's predicted operating state and the preset safety domain;

[0115] S23. If the deviation probability is greater than a preset correction threshold, a backup strategy instruction in a preset candidate strategy pool is called and sent to a corresponding module in the wind power grid connection system.

[0116] In this embodiment, the wind speed prediction model based on the Bayesian network or LSTM model is called to predict the system operating status in the future. By evaluating the probability of deviation between the system's predicted operating state and the preset safety domain If it is greater than the preset correction threshold, the backup strategy instruction in the preset candidate strategy pool is called and sent to the corresponding module in the wind power grid connection system:

[0117]

[0118] is the power system state security domain, if , then the policy adjustment mechanism is triggered. Not in the security domain When , it means that a point in the predicted trajectory deviates from the safe region. (correction threshold), the control strategy correction mechanism is triggered, and the candidate strategy pool is called or the backup control strategy is activated for intervention.

[0119] In one example of the present application, the method further includes the following steps S31-S33:

[0120] S31. After the corresponding module in the wind power grid-connected system executes the stabilization control strategy instruction, control effect data is collected;

[0121] S32. Calculate the reward error according to the control effect data and the expected effect data of the target reinforcement learning model;

[0122] S33. Update the parameters of the digital twin model and the target reinforcement learning model according to the operation prediction state and the reward error, and jump to the step of obtaining real-time measurement data in the digital twin model.

[0123] In this embodiment, a closed-loop training structure is formed by feeding back the actual control effect (such as frequency recovery time, system stability margin, load loss, etc.) into the reinforcement learning environment. and control error , adaptively update the model parameters and control strategy, the strategy parameters and twin model parameters Joint updates are performed according to the following rules:

[0124]

[0125]

[0126] in, is the reward error, is the model prediction error, , is the learning rate.

[0127] In another example of this application, to more intuitively demonstrate the applicability and effectiveness of the embodiments of the present invention in a high-proportion wind power grid-connected environment, field deployment and testing were conducted in a typical wind power access area with a wind power penetration rate exceeding 40%. This area includes three wind farms (with a total installed capacity of 800MW), one load center (with a peak load of approximately 1GW), six main transformers, 15 buses, and multiple dynamic reactive power compensation devices (SVG, SVC, STATCOM, etc.), forming a complex large-scale power subsystem.

[0128] The method of the present invention uses the dispatching master station system as the deployment platform and combines the following software and hardware architecture for system integration:

[0129] Data source integration: Access the dispatch center SCADA system, the power grid company's PMU data platform, and the wind farm EMS system to uniformly receive real-time data streams with a sampling period of 100ms. This includes the following signal items:

[0130] 1) Wind speed, wind power active / reactive output, predicted power ,

[0131] 2) Bus voltage, frequency, and power angle ,Phasor Measurement ,

[0132] 3) Load forecast data, flow boundary data, fault recording data

[0133] Digital twin system deployment: A twin simulation platform is deployed using a containerized microservices architecture, integrating a power grid simulation engine (such as a combination of DSS / OpenDSS or PSCAD), a state estimation algorithm (UKF / EKF), an LSTM wind speed prediction model, and a system stability assessment module. The twin modeling frequency is 10 Hz, and latency is controlled within 200 ms.

[0134] Control strategy operation: The trained reinforcement learning strategy \pi_\theta and rolling optimizer are deployed on the real-time control server, and the interface communicates with the dispatch master station control command port via the IEC60870-5-104 protocol. The control commands issued by the strategy include:

[0135] 1) Reactive power regulation instructions for each wind farm:

[0136] 2) Load shedding / restoration control signal

[0137] 3) Dynamic reactive power compensation device adjustment command (SVG voltage set point)

[0138] 4) Flexible DC interconnection power setting command (if applicable)

[0139] The control strategy deployment process is as follows:

[0140] 1) In the initialization phase, data such as wind speed, load, and disconnection status are collected to build an initial digital twin model;

[0141] The real-time twin system receives SCADA data and provides status updates and predictions every 100ms;

[0142] 2) The control strategy evaluates the current state every 1s and generate the optimal control action , call the interface module to issue commands;

[0143] 3) Run the rolling optimizer every 10 seconds to make short-term adjustments to the strategy;

[0144] 4) If the LSTM wind speed prediction model predicts that there will be or Risk of crossing the limit (i.e. ), then the robustness priority strategy in the redundant strategy pool is called to execute instead;

[0145] 5) All control behaviors, twin model errors and stability indicators are stored in the database for later analysis.

[0146] The performance evaluation results are as follows:

[0147] During a seven-day test run during a typical spring period with drastic wind speed fluctuations, the following key performance indicators were collected:

[0148] The maximum fluctuation of system frequency is controlled within In line with GB / T15945 standard;

[0149] Voltage stability recovery time (defined as the time when the fluctuation is restored to % range) was reduced by 34.6%, from 4.2s to 2.75s;

[0150] During three sudden wind speed drops (>20%), the wind farm experienced no grid disconnection, and the stabilization strategy successfully triggered dynamic reactive power regulation and load-side support mechanisms.

[0151] The average response delay of the control strategy is 480ms, much lower than the 2.5s of the traditional manual scheduling-automatic device hybrid system;

[0152] Twin model prediction error ( norm) is controlled at around 1.8%, meeting the system-level simulation accuracy requirements.

[0153] Actual deployment benefit evaluation:

[0154] Increase wind power utilization by approximately 3.2%, reducing wind power curtailment by 120 million kWh annually;

[0155] Reduce frequency regulation standby capacity by approximately 8%, optimizing system scheduling resource allocation;

[0156] It has the ability to flexibly adapt to new wind farms and energy storage power stations in the future and has strong scalability.

[0157] It can be seen that the intelligent stabilization control method proposed in the present invention shows excellent control performance and engineering application value in actual high-proportion wind power systems, can significantly enhance the system's adaptability to fluctuations in renewable energy output, and provide effective technical support for the future construction of adaptive power grids.

[0158] In an embodiment of the present application, historical operating data and wind power operating data of the wind power grid-connected system are obtained; a digital twin model is constructed and real-time measurement data is obtained according to the historical operating data combined with the topological structure of the wind power grid-connected system; a preset filtering algorithm is called to perform state estimation optimization according to the real-time measurement data to determine the optimal system state; a characteristic state vector is extracted from the optimal system state and the wind power operation data; the characteristic state vector is predicted according to the built-in strategy through a preset target reinforcement learning model, and a stabilization control strategy instruction is generated and issued to the corresponding module in the wind power grid-connected system, so as to interact based on the digital twin model and the reinforcement learning model to realize the accurate generation of the stabilization control strategy instruction, thereby effectively improving the adaptability of the wind power grid-connected control.

[0159] Please refer to Figure 2 , Figure 2 A structural block diagram of a wind power grid-connected system stabilization control strategy generation device based on digital twins in an embodiment of the present application is shown.

[0160] An embodiment of the present invention provides a device for generating a stabilization control strategy for a wind power grid-connected system based on digital twins, the device comprising:

[0161] The data acquisition module 201 is used to acquire historical operation data and wind power operation data of the wind power grid-connected system;

[0162] The model building module 202 is used to build a digital twin model according to historical operation data and the topology of the wind power grid-connected system, and obtain real-time measurement data in the digital twin model;

[0163] The state optimization module 203 is used to call a preset filtering algorithm to perform state estimation optimization according to real-time measurement data to determine the optimal system state;

[0164] A vector construction module 204 is used to extract a characteristic state vector from the optimal system state and wind power operation data;

[0165] The instruction prediction module 205 is used to predict the characteristic state vector according to the built-in strategy through the preset target reinforcement learning model, generate the stabilization control strategy instruction and send it to the corresponding module in the wind power grid-connected system.

[0166] Optionally, the model building module 202 is specifically configured to:

[0167] Determine the node power balance relationship according to the topology of the wind power grid-connected system;

[0168] Construct the physical part model based on the node power balance relationship;

[0169] Based on the error correction function and historical operating data, the preset machine learning model is trained to obtain a data-driven model;

[0170] Associate the physical part model with the data-driven model to build a digital twin model;

[0171] Call the data acquisition unit to obtain real-time measurement data from the digital twin model.

[0172] Optionally, the state optimization module 203 is specifically configured to:

[0173] Substitute the real-time measurement data into the state estimation optimization function, and call the preset filtering algorithm to solve the state estimation optimization function to determine the optimal system state;

[0174] The state estimation optimization function is:

[0175]

[0176] in, is the real-time measurement data at time t, is the system state at time t-1, is the stabilization strategy instruction at time t-1, is the modeling function of the physical part model, Convert the real state of the system into real-time measurement data The nonlinear transfer function of the same form, is the optimal system state, is the weight coefficient.

[0177] Optionally, the device further includes a first optimization module, specifically configured to:

[0178] When the generation cycle of the stabilization control strategy instruction is equal to the preset optimization cycle, the rolling optimizer is called to modify the stabilization control strategy instruction, generate a new stabilization control strategy instruction and send it to the corresponding module in the wind power grid connection system;

[0179] The optimization goals of the rolling optimizer are:

[0180]

[0181] in, for The optimal system state at the moment In execution Stable control strategy instructions at all times After the system steady-state deviation loss, for -1 moment stabilization strategy instruction, To predict the rolling window length, It is the penalty coefficient for regulating the action change rate.

[0182] Optionally, the device further includes a second optimization module, specifically configured to:

[0183] Call the preset wind speed prediction model to predict the system's predicted operating status in the future;

[0184] Calculate the probability of deviation between the system's predicted operating state and the preset safety domain;

[0185] If the deviation probability is greater than a preset correction threshold, the backup strategy instruction in the preset candidate strategy pool is called and sent to the corresponding module in the wind power grid connection system.

[0186] Optionally, the device further includes a target reinforcement learning model training module, specifically configured to:

[0187] Obtain the initial reinforcement learning model and initialize the corresponding policy network;

[0188] The initial reinforcement learning model generates training policy instructions based on the predicted and real-time operating status output by the digital twin model, combined with the policy network, and outputs them to the digital twin model to update the predicted operating status.

[0189] Calculate the current reward value of the initial reinforcement learning model according to the preset reward function until the preset number of cycles is reached;

[0190] With the goal of maximizing the weighted cumulative value of the current reward value, the proximal policy optimization algorithm is called to calculate the policy gradient of the policy network and optimize the model parameters of the initial reinforcement learning model;

[0191] When the current reward value change is less than the preset convergence threshold, the initial reinforcement learning model at the current moment is determined as the target reinforcement learning model.

[0192] Optionally, the device further includes a third optimization module, specifically configured to:

[0193] After the corresponding module in the wind power grid-connected system executes the stabilization control strategy instruction, the control effect data is collected;

[0194] Calculate the reward error based on the control effect data and the expected effect data of the target reinforcement learning model;

[0195] According to the running prediction state and reward error, the parameters of the digital twin model and the target reinforcement learning model are updated, and the execution jumps to the step of obtaining real-time measurement data in the digital twin model.

[0196] An embodiment of the present invention provides an electronic device, characterized in that it includes a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the wind power grid-connected system stabilization control strategy method based on digital twins as described in any embodiment of the present invention.

[0197] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed, a wind power grid-connected system stabilization control strategy method based on digital twins as described in any embodiment of the present invention is implemented.

[0198] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and modules can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0199] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is merely a logical function division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0200] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected to achieve the purpose of the present embodiment according to actual needs.

[0201] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing module, or each module may exist physically separately, or two or more modules may be integrated into a single module. The aforementioned integrated modules may be implemented in the form of hardware or software functional modules.

[0202] If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing an electronic device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0203] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating a stabilization control strategy for a wind power grid-connected system based on digital twins, characterized in that: The method comprises: Acquiring historical operating data and wind power operating data of the wind power grid-connected system; Constructing a digital twin model according to the historical operating data and the topological structure of the wind power grid-connected system, and obtaining real-time measurement data within the digital twin model; Calling a preset filtering algorithm to perform state estimation optimization according to the real-time measurement data to determine the optimal system state; extracting a characteristic state vector from the optimal system state and the wind power operation data; The characteristic state vector is predicted according to the built-in strategy through a preset target reinforcement learning model, and a stabilization control strategy instruction is generated and sent to the corresponding module in the wind power grid-connected system.

2. The method according to claim 1, characterized in that The process of constructing a digital twin model and acquiring real-time measurement data according to the historical operation data and the topology of the wind power grid-connected system includes: Determining a node power balance relationship according to the topology of the wind power grid-connected system; Constructing a physical part model according to the node power balance relationship; Based on the error correction function and the historical operating data, a preset machine learning model is trained to obtain a data-driven model; Associating the physical part model with the data-driven model to construct a digital twin model; The data acquisition unit is called to obtain real-time measurement data from the digital twin model.

3. The method according to claim 1, characterized in that The calling of a preset filtering algorithm to perform state estimation optimization according to the real-time measurement data to determine the optimal system state includes: Substituting the real-time measurement data into a state estimation optimization function, and calling a preset filtering algorithm to solve the state estimation optimization function to determine the optimal system state; The state estimation optimization function is: in, is the real-time measurement data at time t, is the system state at time t-1, is the stabilization strategy instruction at time t-1, is the modeling function of the physical part model, Convert the real state of the system into real-time measurement data The nonlinear transfer function of the same form, is the optimal system state, is the weight coefficient.

4. The method according to claim 1, wherein The method further comprises: When the generation cycle of the stabilization control strategy instruction is equal to the preset optimization cycle, calling the rolling optimizer to modify the stabilization control strategy instruction, generating a new stabilization control strategy instruction and issuing it to the corresponding module in the wind power grid connection system; The optimization goal of the rolling optimizer is: in, for The optimal system state at the moment In execution Stable control strategy instructions at all times After the system steady-state deviation loss, for -1 moment stabilization strategy instruction, To predict the rolling window length, It is the penalty coefficient for regulating the action change rate.

5. The method according to claim 1, characterized in that The method further comprises: Call the preset wind speed prediction model to predict the system's predicted operating status in the future; Calculating the deviation probability between the predicted operating state of the system and the preset safety domain; If the deviation probability is greater than a preset correction threshold, a backup strategy instruction in a preset candidate strategy pool is called and sent to a corresponding module in the wind power grid connection system.

6. The method according to claim 1, characterized in that The method further comprises: Obtain the initial reinforcement learning model and initialize the corresponding policy network; Generate training strategy instructions based on the predicted operation state and the real-time operation state output by the digital twin model through the initial reinforcement learning model, combined with the strategy network, and output the training strategy instructions to the digital twin model to update the predicted operation state; Calculating the current reward value of the initial reinforcement learning model according to a preset reward function until a preset number of cycles is reached; With the goal of maximizing the weighted cumulative value of the current reward value, calling the proximal policy optimization algorithm to calculate the policy gradient of the policy network and optimize the model parameters of the initial reinforcement learning model; When the current reward value change is less than a preset convergence threshold, the initial reinforcement learning model at the current moment is determined as the target reinforcement learning model.

7. The method according to claim 1, characterized in that The method further comprises: After the corresponding module in the wind power grid-connected system executes the stabilization control strategy instruction, control effect data is collected; Calculating a reward error according to the control effect data and the expected effect data of the target reinforcement learning model; According to the operation prediction state and the reward error, the parameters of the digital twin model and the target reinforcement learning model are updated, and the step of obtaining real-time measurement data in the digital twin model is jumped to execute.

8. A wind power grid-connected system stabilization control strategy generation device based on digital twin, characterized in that: The device comprises: A data acquisition module, used to acquire historical operation data and wind power operation data of the wind power grid-connected system; A model building module, configured to build a digital twin model and obtain real-time measurement data based on the historical operating data and the topology of the wind power grid-connected system; A state optimization module is used to call a preset filtering algorithm to perform state estimation optimization according to the real-time measurement data to determine the optimal system state; a vector construction module, configured to extract a characteristic state vector from the optimal system state and the wind power operation data; The instruction prediction module is used to predict the characteristic state vector according to the built-in strategy through a preset target reinforcement learning model, generate a stabilization control strategy instruction and send it to the corresponding module in the wind power grid-connected system.

9. An electronic device, characterized in that: It includes a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the wind power grid-connected system stabilization control strategy method based on digital twins as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the wind power grid-connected system stabilization control strategy method based on digital twin is implemented according to any one of claims 1 to 7.

Images