Wind storage integrated intelligent regulation method and system

Abstract

This invention discloses an integrated intelligent control method and system for wind and energy storage. The method includes: real-time acquisition of operating data of each wind turbine in the wind farm, status data of the energy storage system, grid dispatch command data, and real-time electricity price data; using a multi-scale spatiotemporal fusion prediction model to generate a wind power prediction curve for a preset future time period; automatically identifying the current operating scenario and matching the optimal control mode based on the current wind power output characteristics, the grid dispatch command type in the grid dispatch command data, and the real-time electricity price signal in the real-time electricity price data; using the wind power prediction curve as the prediction input, employing a hierarchical distributed model predictive control algorithm to output the output commands of each wind turbine and the charging and discharging commands of the energy storage as control commands; using a multi-objective optimization pricing strategy to generate minute-level pricing schemes and automatically submitting them to the electricity spot market trading platform; and issuing control commands to each execution unit, monitoring the execution effect in real time, and performing closed-loop correction.

Description

Technical Field

[0001] This invention relates to the field of wind power generation and energy storage technology, and more specifically to an integrated intelligent control method and system for wind power and energy storage. Background Technology

[0002] With the accelerated transformation of the global energy structure, wind power, as an important component of clean and renewable energy, has seen its installed capacity grow rapidly. However, the inherent intermittent, volatile, and random characteristics of wind power generation pose serious challenges to grid dispatch and safe and stable operation after large-scale grid connection. When the wind power penetration rate exceeds a certain threshold, drastic fluctuations in wind power output may lead to problems such as grid frequency deviation, voltage flicker, and increased pressure on peak and frequency regulation, which in severe cases may even threaten the safe and stable operation of the grid.

[0003] Energy storage technology, as an effective means to mitigate fluctuations in renewable energy output and enhance grid absorption capacity, has become a mainstream trend in the industry when combined with wind power generation to form integrated wind-storage systems. Through rapid charge and discharge response, energy storage systems can effectively smooth wind power output fluctuations, provide frequency regulation and peak shaving ancillary services, participate in electricity market transactions, and significantly improve the grid-friendliness and economic benefits of wind farms.

[0004] Meanwhile, the national electricity market liberalization process continues to advance, with the electricity spot market having been launched for trial operation and gradually promoted in several provinces. Under the electricity spot market mechanism, new energy power plants need to actively participate in competitive bidding based on market price signals, which poses new requirements for the operation and management of wind and energy storage power plants: on the one hand, it is necessary to accurately predict wind power output and market prices and formulate the optimal bidding strategy; on the other hand, it is necessary to coordinate the operation of wind turbines and energy storage systems to ensure that the deviation between declared output and actual output is minimized, and to avoid economic losses caused by deviation assessments.

[0005] The existing wind and energy storage control systems mainly have the following problems: (1) The control strategy is too simple and difficult to adapt to the changing operating scenarios. The existing system usually adopts a control strategy with fixed parameters, which cannot be adaptively adjusted according to the wind power output characteristics, grid dispatching needs and market price signals, resulting in poor control effect.

[0006] (2) Single-machine control and field-level regulation are disconnected and lack coordinated optimization. Existing systems either adopt a single-machine independent control mode, which cannot achieve power coordination at the field level; or adopt a centralized control mode, which has the risk of single point of failure and slow response speed.

[0007] (3) The spot market trading function is missing or incomplete. Most existing systems do not integrate the electricity spot market trading function, and cannot realize the closed-loop linkage of forecasting-trading-control, which leads to power plants missing market opportunities or facing greater deviation assessment risks.

[0008] (4) Insufficient prediction accuracy affects regulation and trading decisions. Existing wind power prediction models usually use a single method and do not make full use of multi-source data and multi-scale features, making it difficult to meet the needs of refined regulation and trading in terms of prediction accuracy.

[0009] Therefore, there is an urgent need to develop an integrated intelligent control method and system for wind and energy storage that can achieve coordinated optimization and control of wind and energy storage, support electricity spot market transactions, and have multi-scenario adaptive capabilities. Summary of the Invention

[0010] The purpose of this invention is to propose an integrated intelligent control method and system for wind and energy storage to solve the problems existing in the prior art.

[0011] To achieve the above objectives, the first aspect of this invention proposes an integrated intelligent control method for wind and energy storage, comprising the following steps: S1. Real-time collection of operating data of each wind turbine in the wind farm, status data of the energy storage system, grid dispatch command data, and real-time electricity price data; S2. Based on historical operating data, numerical weather forecast data, and real-time wind measurement data in the wind turbine operation data, a multi-scale spatiotemporal fusion prediction model is used to generate wind power prediction curves for a future preset time period. S3. Based on the current wind power output characteristics obtained from the wind power prediction curve, the grid dispatch instruction type in the grid dispatch instruction data, and the real-time electricity price signal in the real-time electricity price data, automatically identify the current operating scenario and match the optimal control mode. S4. Using the wind power prediction curve as the prediction input, a hierarchical distributed model prediction control algorithm is adopted to realize local edge closed-loop control at the single-unit power distribution and storage level, and to realize the collaborative power distribution optimization of multiple wind turbines and energy storage systems at the field level. The output commands of each wind turbine and the charging and discharging commands of the energy storage system are output as control commands. S5. A price prediction model is trained based on wind power prediction curves and real-time electricity price data. A multi-objective optimization pricing strategy is adopted to generate minute-level pricing schemes, which are then automatically submitted to the electricity spot market trading platform. S6. Issue control commands to each execution unit, monitor the execution effect in real time, and perform closed-loop correction.

[0012] Furthermore, the multi-scale spatiotemporal fusion prediction model in step S2 adopts the following formula: ; in, For the future Predicted power at time; These are the predicted components based on numerical weather prediction. For prediction components based on the persistence model; For prediction components based on machine learning models; , , For dynamic weighting coefficients, satisfying It also adaptively adjusts based on the length of the forecast period and the weather type.

[0013] Furthermore, the method for adjusting the dynamic weighting coefficients is as follows: ; ; ; in, Use the Sigmoid activation function; This refers to the current wind speed; Wind direction; For the predicted duration; , This is the weight matrix. , The bias vector is used; the weight matrix and bias vector are obtained through the following training process: collecting historical operational data including wind speed. ,wind direction Predicted duration The actual power at the corresponding time point is used as the training sample set. The mean square error between the actual power and the predicted power is used as the loss function. The backpropagation algorithm is used to iteratively train the neural network containing the weight matrix and bias vector until the loss function converges to below a preset threshold, thus obtaining the trained neural network. , , , .

[0014] Furthermore, the hierarchical distributed model predictive control algorithm in step S4 includes: The local edge closed-loop control at the single-machine storage level adopts the following optimization model: The constraints are: in, For the optimization objective function at the single-unit energy storage level, represents the weighted sum of squares of grid-connected power tracking error and changes in energy storage operation; For the first Grid-connected power at any given time; For reference power curve; For the first The actual output of the wind turbine at any given time is obtained from the wind turbine operating data; For energy storage charging and discharging power; This represents the change in energy storage charging and discharging power between adjacent time points; The state of charge of the energy storage system is obtained through the state data of the energy storage system. Through constraints Directly affects the feasible region of the optimization problem, when near Time limit on energy storage discharge power, when near Time-limited energy storage charging power; , These are the weighting coefficients. The value range is determined based on the grid-connected power point tracking accuracy requirements. , The value is determined based on the requirements for the smoothness of energy storage charge and discharge, and the range is [value range missing]. , and By conducting adjustment experiments on different weight combinations in a simulation environment, the weight combination that minimizes the root mean square error of grid-connected power tracking and has the lowest energy storage operation frequency was selected. To predict the length of the time domain.

[0015] Furthermore, the field-level cooperative power allocation optimization adopts the following model: The constraints are: in, Let be the field-level optimization objective function, representing the weighted sum of squares of scheduling command tracking error, single-unit output deviation, and energy storage state of charge deviation; This represents the total number of wind turbines. The power specified in the power grid dispatch command is obtained from the power grid dispatch command data. For the first The output of the typhoon generator; For the first Reference output of the typhoon generator; For the first The state of charge of the energy storage system at any given time is obtained from the state data of the energy storage system. The target state of charge for energy storage; , , These are the weighting coefficients. The value range is determined based on the tracking accuracy requirements of the scheduling instructions. , The value range is determined based on the requirement for balanced output of a single unit. , The value range is determined based on the accuracy requirements for maintaining the state of charge of energy storage. , , , By conducting adjustment experiments on different weight combinations in a simulation environment, the weight combination that minimizes the root mean square error of power tracking across the entire field and achieves the most balanced power distribution among the wind turbines was selected.

[0016] Furthermore, the scene recognition in step S3 employs a multi-dimensional feature fusion method: in, This is the characteristic vector of wind power output, which includes wind speed fluctuation rate. Wind direction change rate turbulence intensity All of these are calculated from the wind turbine operating data; This is a power grid dispatch feature vector, containing AGC command type and frequency regulation requirements. Peak shaving demand All of these are extracted from the power grid dispatch command data; This is the electricity price feature vector, containing real-time electricity prices. Electricity price trend All of these are calculated from the real-time electricity price data; The feature weights are obtained through training with historical data; This is the feature matching function.

[0017] Furthermore, the feature matching function The specific expression is: in, For the first 3D eigenvalues For the scene Next The mean of the dimensional features, For the scene Next Standard deviation of dimensional features and It is obtained by statistical clustering of various scenarios in historical operation data.

[0018] Furthermore, the multi-objective optimization pricing strategy in step S5 adopts the following model: The constraints are: in, For the first The power of the application during the time period; For the first The predicted electricity price for a given period is output by the electricity price prediction model. The probability of clearing out; To contribute practically; This is the risk aversion coefficient, with a range of values. ; For the first The deviation penalty price for a given time period represents the penalty fee corresponding to a unit of deviation power. It is determined by the trading rules of the electricity spot market and is set as a preset multiple of the real-time electricity price. The allowable deviation range indicates the upper limit of the proportion by which the declared power exceeds the predicted wind power value, with a value range of [value missing]. , Indicates the first The maximum allowable power level for a given time period; The total available power generation of the wind-storage system during the predicted period is calculated by integrating the wind power prediction curve over the predicted period and summing the amount of electricity that the energy storage system can release.

[0019] Furthermore, the clearing probability is calculated using the following formula: in, This is a market sensitivity parameter, reflecting the degree to which market prices affect the probability of clearing. The marginal clearing price threshold is obtained through regression analysis of historical market data.

[0020] A second aspect of this invention proposes an integrated wind and energy storage intelligent control system, comprising: The data acquisition module is used to collect real-time operating data of each wind turbine in the wind farm, status data of the energy storage system, grid dispatch command data, and real-time electricity price data. The wind power prediction module is used to generate wind power prediction curves for a preset future time period based on historical operating data, numerical weather forecast data, and real-time wind measurement data from wind turbine operation data, using a multi-scale spatiotemporal fusion prediction model. The scene recognition module is used to automatically identify the current operating scene and match the optimal control mode based on the current wind power output characteristics obtained from the wind power prediction curve, the grid dispatch instruction type in the grid dispatch instruction data, and the real-time electricity price signal in the real-time electricity price data. The collaborative control module is used to take the wind power prediction curve as the prediction input, adopt a hierarchical distributed model predictive control algorithm, realize local edge closed-loop control at the single-unit power distribution and storage level, realize collaborative power distribution optimization of multiple wind turbines and energy storage systems at the field level, and output the output commands of each wind turbine and the charging and discharging commands of energy storage as control commands. The spot market trading module is used to train an electricity price prediction model based on wind power prediction curves and real-time electricity price data, generate minute-level bidding schemes using a multi-objective optimization bidding strategy, and automatically submit them to the electricity spot market trading platform. The execution feedback module is used to send control commands to each execution unit, monitor the execution effect in real time, and perform closed-loop correction.

[0021] Compared with the prior art, the present invention has the following advantages: (1) This invention adopts a multi-scale spatiotemporal fusion prediction model, which integrates the advantages of numerical weather prediction, persistence model and machine learning model. Through dynamic weight adaptive adjustment, it significantly improves the accuracy of wind power prediction, with an ultra-short-term prediction accuracy of over 85% and a short-term prediction accuracy of over 90%.

[0022] (2) The present invention adopts a hierarchical distributed model predictive control architecture, which realizes local edge closed-loop control at the single-machine power distribution level with a response time of less than 300ms; and realizes coordinated power distribution optimization at the field level with a response time of less than 500ms, effectively solving the coordination problem between single-machine control and field-level regulation.

[0023] (3) This invention innovatively integrates the spot market trading function with the risk storage regulation function. Through multi-objective optimization of the pricing strategy, it achieves the optimal balance between maximizing returns and minimizing risks, and the accuracy of spot pricing reaches more than 99%.

[0024] (4) The present invention adopts a scene recognition method based on multi-dimensional feature fusion, which can automatically identify the operating scene and match the optimal control mode according to the wind power output characteristics, grid dispatching requirements and market price signals, thus realizing multi-scene adaptive regulation.

[0025] (5) The system architecture of the present invention is clear and highly modular, easy to expand and maintain, and can be widely used in single-unit wind-storage power stations and field-level wind-storage joint control power stations. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a flowchart of the integrated wind and energy storage intelligent control method of the present invention. Detailed Implementation

[0028] The present application will now be described in more detail with reference to the accompanying drawings. The following embodiments are only used to illustrate the technical solutions of the present application more clearly and are not intended to limit the scope of protection of the present application.

[0029] This invention provides an integrated intelligent control method for wind and energy storage, such as... Figure 1 As shown, the integrated wind and energy storage intelligent control method includes the following steps: Step 1: Data Acquisition Steps: Real-time acquisition of operating data of each wind turbine in the wind farm, status data of the energy storage system, grid dispatch command data, and real-time electricity price data.

[0030] The system collects real-time operational data from each wind turbine within the wind farm, including active power, reactive power, wind speed, wind direction, engine speed, pitch angle, and nacelle temperature. It also collects status data from the energy storage system, including state of charge (SOC), charge / discharge power, battery temperature, and individual cell voltage. Furthermore, it collects grid dispatch command data, including AGC commands, AVC commands, and generation plan curves. Real-time electricity price data is also collected, including the current time-to-date price and day-ahead clearing price. The data acquisition cycle is configurable, supporting data acquisition as fast as 100ms.

[0031] Data acquisition employs a multi-protocol adaptation engine, supporting IEC 61850, GOOSE It supports mainstream communication protocols and enables seamless integration with various devices.

[0032] The four types of data collected are used in subsequent steps: wind turbine operation data is used for wind power prediction in step 2 and wind power output characteristics extraction in step 3; energy storage system status data is used for obtaining energy storage charge state and constructing constraints in step 4; grid dispatch command data is used for scenario identification in step 3 and dispatch command tracking in step 4; and real-time electricity price data is used for scenario identification in step 3 and electricity price prediction model training in step 5.

[0033] Step 2, Wind Power Prediction Step: Based on historical operating data, numerical weather forecast data, and real-time wind measurement data from wind turbine operation data, a multi-scale spatiotemporal fusion prediction model is used to generate wind power prediction curves for the future preset time period.

[0034] According to an embodiment of the present invention, the multi-scale spatiotemporal fusion prediction model includes three prediction branches: (1) NWP forecast branch: Based on numerical weather forecast data obtained from the operation of the mesoscale meteorological model WRF, including forecast values ​​of meteorological elements such as wind speed, wind direction, temperature, humidity, and air pressure, the forecast values ​​are converted into predicted power through power curves. ; (2) Persistent model prediction branch: Assuming that the wind speed remains constant or changes according to inertia in the short term, the predicted power is obtained by extrapolation. Persistent models are suitable for ultra-short-term forecasts (0-1 hour). (3) Machine learning prediction branch: Using LSTM neural network or XGBoost model, inputting historical power series, meteorological data, time features, etc., and outputting predicted power .

[0035] The outputs of the three prediction branches are dynamically weighted and fused: The dynamic weighting coefficients are adaptively adjusted based on the length of the forecast period and the weather type. Input: Current wind speed ,wind direction Predicted duration Weather type Output: Weighting coefficients , , Step 1: Feature Vector Construction: Step 2: Calculate using a neural network and : , in, , This is the weight matrix. , The bias vector is obtained through the following training process: collecting historical operational data including wind speed. ,wind direction Predicted duration The actual power at the corresponding time point is used as the training sample set. The mean square error between the actual power and the predicted power is used as the loss function. The backpropagation algorithm is used to iteratively train the neural network containing the weight matrix and bias vector until the loss function converges to below a preset threshold, thus obtaining the trained neural network. , , , .

[0036] Step 3: Normalization ensures the weights sum to 1: , , , Output , , .

[0037] The wind power prediction curve output in this step will be used in subsequent steps: in step 3, it is used to extract the current wind power output characteristics (including wind speed fluctuation rate, wind direction change rate, turbulence intensity, etc.); in step 4, it is used as the prediction input for the hierarchical distributed model predictive control algorithm; and in step 5, it is used as the power benchmark for the multi-objective optimization pricing strategy.

[0038] Step 3, Scene Recognition Step: Based on the current wind power output characteristics obtained from the wind power prediction curve, the grid dispatch instruction type in the grid dispatch instruction data, and the real-time electricity price signal in the real-time electricity price data, automatically identify the current operating scene and match the optimal control mode.

[0039] The current wind power output characteristics are extracted from the wind power prediction curve output in step 2, specifically including: wind speed fluctuation rate. (Calculated from the standard deviation of power changes at adjacent times in the wind power prediction curve), wind direction change rate (Calculated from the wind direction sequence in the wind turbine operating data), turbulence intensity (Obtained from wind speed sequence statistics in wind turbine operation data).

[0040] The power grid dispatch command type is obtained by parsing the power grid dispatch command data collected in step 1, and specifically includes AGC command type (such as frequency regulation up, frequency regulation down, frequency regulation exit) and frequency regulation requirement. and peak shaving demand .

[0041] The real-time electricity price signal is obtained from the real-time electricity price data collected in step 1, specifically including the real-time electricity price. and electricity price change trends (Calculated from the price difference between adjacent time periods).

[0042] Scene recognition employs a multi-dimensional feature fusion method: Input: Wind power output characteristics Power grid dispatch characteristics Electricity price characteristics Output: Scene Type Step 1: Feature Extraction (Wind speed fluctuation rate, wind direction change rate, turbulence intensity) (AGC type, frequency modulation requirement, peak shaving requirement) (Real-time electricity price, electricity price trend) Step 2: Feature Fusion Step 3: Scenario Probability Calculation For each scenario : in, The feature matching function has the following expression: in, For the first 3D eigenvalues For the scene Next The mean of the dimensional features, For the scene Next Standard deviation of dimensional features and It is obtained by statistical clustering of various scenarios in historical operation data. Step 4: Choose the scenario with the highest probability: Output .

[0043] Step 4, Coordinated Control Step: Using the wind power prediction curve as the prediction input, a hierarchical distributed model predictive control algorithm is adopted to achieve local edge closed-loop control at the single-unit power distribution and energy storage level, and to achieve coordinated power distribution optimization of multiple wind turbines and energy storage systems at the field level. The output commands of each wind turbine and the charging and discharging commands of the energy storage system are output as control commands.

[0044] The model predictive control algorithms at both the single-unit power supply and storage level and the field level in step 4 use the wind power prediction curve output in step 2 as the prediction input for wind turbine output in the future period, and are used to construct the state prediction equation in the optimization problem. The wind turbine output commands and energy storage charging and discharging commands output in step 4 determine the actual available output range of the wind-storage system. This output range serves as the power constraint condition for the bidding strategy in the subsequent step 5. The wind turbine output commands and energy storage charging and discharging commands output in step 4 are the control commands mentioned in the subsequent step 6. In step 6, these control commands are sent to each execution unit and the execution effect is monitored.

[0045] (1) Local edge closed-loop control at the single-unit distribution and storage level: Input: Fan power (Originally obtained from wind turbine operating data), reference power (Determined based on the control mode corresponding to the scenario identified in step S3), Energy storage status (Obtained from the status data of the energy storage system) Output: Energy storage charging and discharging commands Step 1: State Prediction: For arrive : (Wind power prediction curve from step S2). Step 2: Optimize the solution: in, To optimize the objective function, let represent the weighted sum of squares of the grid-connected power point tracking error and the change in energy storage operation; It represents the change in energy storage charging and discharging power between adjacent time points, used to suppress frequent energy storage operations.

[0046] Constraints: , , in, Through constraints Directly affecting the feasible region of the optimization problem: when near When the energy storage has limited discharge space, the optimization solver will automatically limit the energy storage discharge power; when near At that time, the available space for energy storage charging is limited, and the optimization solver will automatically limit the energy storage charging power.

[0047] This is the grid-connected power point tracking weighting coefficient, determined according to the grid-connected power point tracking accuracy requirements, with a value range of [value range missing]. ; This is the smoothing weighting coefficient for energy storage operations, determined based on the smoothness requirements of energy storage charging and discharging, with a value range of [value missing]. . and By conducting adjustment experiments on different weight combinations in a simulation environment, the weight combination that minimizes the root mean square error of grid-connected power tracking and has the lowest energy storage operation frequency was selected.

[0048] Step 3: Solution Method (Quadratic Programming): Constructing the Hessian Matrix and gradient vector : , Considering constraints, the interior point method or... Solution method: Step 4: Optimize scrolling, only perform step one: Output .

[0049] (2) Field-level cooperative power allocation optimization: Input: Power of each fan (Obtained from wind turbine operating data), dispatch instructions (Obtained from grid dispatch command data), Energy storage status (Obtained from the status data of the energy storage system) Output: Output commands for each fan Energy storage charging and discharging commands Step 1: Overall Game Performance Prediction: For arrive :for arrive : (Wind power prediction curve from step S2). , Step 2: Multi-objective optimization: in, The objective function for field-level optimization is represented by the weighted sum of squares of the scheduling command tracking error, the single-unit output deviation, and the energy storage state of charge deviation.

[0050] Constraints: , (for all) ), , The weighting coefficient for scheduling instruction tracking is determined based on the accuracy requirements of scheduling instruction tracking, and its value range is [value range missing]. ; This is the single-unit output balance weighting coefficient, determined based on the single-unit output balance requirements, with a value range of [value range missing]. ; The weighting coefficient for maintaining the state of charge (SOC) of energy storage is determined based on the accuracy requirements for maintaining SOC, and its value range is [value range missing]. . , , By conducting adjustment experiments on different weight combinations in a simulation environment, the weight combination that minimizes the root mean square error of power tracking across the entire field and achieves the most balanced power distribution among the wind turbines was selected.

[0051] Step 3: Decompose and coordinate the solution: Decompose the problem into Lagrange relaxation methods. The issue is about size.

[0052] For iteration arrive : Subproblems Optimization of wind turbine output: Subproblems Energy storage power optimization: Update the Lagrange multipliers: Convergence criterion: If If so, the iteration will exit.

[0053] Step 4: Scrolling optimization: (First instruction for each wind turbine) (First step instruction for energy storage) Output , .

[0054] Step 4 outputs the power output commands for each fan. and energy storage charging and discharging commands Together they constitute the control command, which will be used in step 5 to determine the actual available output range of the wind storage system, and will be issued to each execution unit for execution in step 6.

[0055] Step 5, Spot Market Trading Steps: Based on the wind power prediction curve and real-time electricity price data, a price prediction model is trained. A multi-objective optimization pricing strategy is adopted to generate minute-level pricing schemes, which are then automatically submitted to the electricity spot market trading platform.

[0056] The electricity price prediction model uses the real-time and historical electricity price data collected in step 1 as training samples. It is trained using an LSTM neural network or an ARIMA time series model and is used to predict electricity prices for future periods. The real-time electricity price signal used in step 3 As one of the real-time input features of the electricity price forecasting model, the model predicts future electricity prices by combining historical electricity price trends and market supply and demand information based on real-time electricity price signals.

[0057] Input: Wind power prediction (From the wind power forecast curve in step 2), electricity price forecast (From electricity price forecasting model), risk aversion coefficient Output: Reported power Step 1: Clearing probability prediction: For arrive : Step 2: Build an optimization model: in, For the first The time-period deviation penalty price represents the penalty fee (in yuan / MW) corresponding to a unit of deviation power (unit: MW), determined by the electricity spot market trading rules, and is taken as a preset multiple of the real-time electricity price, for example... .

[0058] Constraints: in, The allowable deviation range indicates the upper limit of the proportion by which the declared power exceeds the predicted wind power value, with a value range of [value missing]. . Indicates the first The maximum allowable power limit for a given time period is the allowable upward fluctuation range for the power declared after taking into account the uncertainty of forecasting. in, The total available power generation of the wind-storage system during the prediction period (unit: MWh) is calculated by integrating the wind power prediction curve output in step 2 over the prediction period and summing the amount of electricity that the energy storage system can release. The specific calculation formula is as follows: The first item is the amount of electricity that can be generated by wind power during the predicted period, and the second item is the amount of electricity that the energy storage system can currently release.

[0059] Step 3: Transform into a deterministic optimization problem: Use scenario-based methods to handle uncertainty. For scenarios... arrive : ( (for scene deviation) , Step 4: Solve the optimization problem: Use Mixed Integer Linear Programming (MILP) or Genetic Algorithm to solve it. Step 5: Power Requirement Adjustment: Considering energy storage regulation capabilities, appropriately increase the required power requirement. Output .

[0060] Step 6: Execution feedback step: Send the control commands (including the output commands of each wind turbine and the charging and discharging commands of energy storage) output in step 4 to each execution unit, monitor the execution effect in real time and perform closed-loop correction.

[0061] Input: Control command (From step 4), actual output Energy storage status Output: Correction instructions Step 1: Perform deviation calculation: Step 2: Deviation Judgment: If : Step 3: Energy Storage Compensation (Energy storage compensation deviation) Step 4: Energy storage capacity verification: If : ;like : Step 5: Issue correction instructions: otherwise: (The deviation is within the allowable range and no correction is needed) Output .

[0062] This invention also proposes an integrated wind and energy storage intelligent control system, comprising: The data acquisition module 201 is used to collect real-time operating data of each wind turbine in the wind farm, status data of the energy storage system, grid dispatch command data, and real-time electricity price data. The wind power prediction module 202 is used to generate wind power prediction curves for a preset future time period based on historical operating data, numerical weather forecast data, and real-time wind measurement data in the wind turbine operating data, using a multi-scale spatiotemporal fusion prediction model. The scene recognition module 203 is used to automatically identify the current operating scene and match the optimal control mode based on the current wind power output characteristics obtained from the wind power prediction curve, the grid dispatch instruction type in the grid dispatch instruction data, and the real-time electricity price signal in the real-time electricity price data. The collaborative control module 204 is used to take the wind power prediction curve as the prediction input, adopt a hierarchical distributed model prediction control algorithm, realize local edge closed-loop control at the single-unit power distribution and storage level, realize collaborative power distribution optimization of multiple wind turbines and energy storage system at the field level, and output the output command of each wind turbine and the charging and discharging command of energy storage as control commands. The spot market trading module 205 is used to train an electricity price prediction model based on wind power prediction curves and real-time electricity price data, generate minute-level bidding schemes using a multi-objective optimization bidding strategy, and automatically submit them to the electricity spot market trading platform. The execution feedback module 206 is used to send control commands to each execution unit, monitor the execution effect in real time, and perform closed-loop correction.

[0063] The data acquisition module 201 includes: The wind turbine data acquisition unit is used to collect operating data such as active power, reactive power, wind speed, wind direction, rotational speed, blade pitch angle, and nacelle temperature of each wind turbine. The energy storage data acquisition unit is used to collect status data such as SOC, charge and discharge power, battery temperature, and individual cell voltage of the energy storage system. The power grid data acquisition unit is used to collect power grid dispatch command data such as AGC commands, AVC commands, and power generation plan curves. The electricity price data acquisition unit is used to collect real-time electricity price data such as the current time period's real-time electricity price and day-ahead clearing electricity price; Protocol adapter unit, used to implement IEC 61850, GOOSE Parsing and conversion of multiple protocols.

[0064] The wind power prediction module 202 includes: The NWP forecasting unit is used to obtain numerical weather forecast data based on the WRF mesoscale meteorological model and convert it into predicted power. The persistent model prediction unit is used for ultra-short-term power prediction based on the assumption of wind speed persistence. A machine learning prediction unit is used to predict power using an LSTM neural network or an XGBoost model. The dynamic fusion unit is used to adaptively adjust the weights of each prediction branch according to the prediction period length and weather type, and output the fused prediction result.

[0065] The scene recognition module 203 includes: The feature extraction unit is used to extract multidimensional features from wind power output data, grid dispatch data, and electricity price data. The scene classification unit is used to identify the current running scene based on a multi-dimensional feature fusion method. The pattern matching unit is used to match the optimal control mode based on the identified scenario.

[0066] The coordinated control module 204 includes: The single-unit energy storage control unit is used to realize local edge closed-loop control at the single-unit energy storage level, and uses model predictive control algorithm to optimize the energy storage charging and discharging power; The field-level collaborative control unit is used to optimize the collaborative power distribution of multiple wind turbines and energy storage systems at the field level. The AGC / AVC response unit is used to respond to the grid's AGC / AVC commands and realize active / reactive power regulation.

[0067] The spot market trading module 205 includes: The electricity price forecasting unit is used to predict future electricity prices based on historical and real-time electricity price data. The clearing probability prediction unit is used to predict the clearing probability of the declared power. The pricing optimization unit is used to generate the optimal pricing scheme using a multi-objective optimization strategy. The transaction submission unit is used to automatically submit bidding proposals to the electricity spot market trading platform.

[0068] The execution feedback module 206 includes: The instruction issuing unit is used to issue control instructions to each execution unit; The execution monitoring unit is used to monitor the execution effect of each execution unit in real time. The closed-loop correction unit is used to perform closed-loop correction based on the execution deviation.

[0069] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software may depend on the specific application and design constraints of the technical solution. Those skilled in the art may use different methods for each specific application to implement the described functions, but such implementation should not be considered beyond the scope of the embodiments of this disclosure.

[0070] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly using hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, etc. or any other form of storage medium known in the art.

[0071] The methods and products (including but not limited to devices and equipment) disclosed in the embodiments herein can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units may only be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be electrical, mechanical, or other forms.

[0072] Unless otherwise defined, the technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used herein is for descriptive purposes only and is not intended to limit the invention. Terms such as “part” or “component” appearing herein can refer to a single part or a combination of multiple parts. Terms such as “installation” or “installation” appearing herein can refer to one component being directly attached to another component or one component being attached to another component via an intermediary. A feature described in one embodiment herein may be applied, alone or in combination with other features, to another embodiment, unless that feature is not applicable in that other embodiment or is otherwise stated.

[0073] The present invention has been described through the above embodiments. However, it should be understood that the above embodiments are for illustrative purposes only and are not intended to limit the present invention to the described embodiments. Furthermore, those skilled in the art will understand that the present invention is not limited to the above embodiments, and many variations and modifications can be made based on the teachings of the present invention, all of which fall within the scope of protection claimed by the present invention. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A wind-storage integrated intelligent control method, characterized in that, Includes the following steps: S1. Real-time collection of operating data of each wind turbine in the wind farm, status data of the energy storage system, grid dispatch command data, and real-time electricity price data; S2. Based on historical operating data, numerical weather forecast data, and real-time wind measurement data in the wind turbine operation data, a multi-scale spatiotemporal fusion prediction model is used to generate wind power prediction curves for a future preset time period. S3. Based on the current wind power output characteristics obtained from the wind power prediction curve, the grid dispatch instruction type in the grid dispatch instruction data, and the real-time electricity price signal in the real-time electricity price data, automatically identify the current operating scenario and match the optimal control mode. S4. Using the wind power prediction curve as the prediction input, a hierarchical distributed model prediction control algorithm is adopted to realize local edge closed-loop control at the single-unit power distribution and storage level, and to realize the collaborative power distribution optimization of multiple wind turbines and energy storage systems at the field level. The output commands of each wind turbine and the charging and discharging commands of the energy storage system are output as control commands. S5. A price prediction model is trained based on wind power prediction curves and real-time electricity price data. A multi-objective optimization pricing strategy is adopted to generate minute-level pricing schemes, which are then automatically submitted to the electricity spot market trading platform. S6. Issue control commands to each execution unit, monitor the execution effect in real time, and perform closed-loop correction.

2. The integrated wind and energy storage intelligent control method according to claim 1, characterized in that, The multi-scale spatiotemporal fusion prediction model in step S2 uses the following formula: ； in, For the future Predicted power at time; These are the predicted components based on numerical weather prediction. For prediction components based on the persistence model; For prediction components based on machine learning models; , , For dynamic weighting coefficients, satisfying It also adaptively adjusts based on the length of the forecast period and the weather type.

3. The integrated wind and energy storage intelligent control method according to claim 1, characterized in that, The method for adjusting the dynamic weighting coefficient is as follows: ； ； ； in, Use the Sigmoid activation function; This refers to the current wind speed; Wind direction; For the predicted duration; , This is the weight matrix. , The bias vector is used; the weight matrix and bias vector are obtained through the following training process: collecting historical operational data including wind speed. ,wind direction Predicted duration The actual power at the corresponding time point is used as the training sample set. The mean square error between the actual power and the predicted power is used as the loss function. The backpropagation algorithm is used to iteratively train the neural network containing the weight matrix and bias vector until the loss function converges to below a preset threshold, thus obtaining the trained neural network. , , , .

4. The integrated wind and energy storage intelligent control method according to claim 1, characterized in that, The hierarchical distributed model prediction control algorithm in step S4 includes: The local edge closed-loop control at the single-machine storage level adopts the following optimization model: ； The constraints are: ； ； ； in, For the optimization objective function at the single-unit energy storage level, represents the weighted sum of squares of grid-connected power tracking error and changes in energy storage operation; For the first Grid-connected power at any given time; For reference power curve; For the first The actual output of the wind turbine at any given time is obtained from the wind turbine operating data; For energy storage charging and discharging power; This represents the change in energy storage charging and discharging power between adjacent time points; The state of charge of the energy storage system is obtained through the state data of the energy storage system. Through constraints Directly affects the feasible region of the optimization problem, when near Time limit on energy storage discharge power, when near Time-limited energy storage charging power; , These are the weighting coefficients. The value range is determined based on the grid-connected power point tracking accuracy requirements. , The value is determined based on the requirements for the smoothness of energy storage charge and discharge, and the range is [value range missing]. , and By conducting adjustment experiments on different weight combinations in a simulation environment, the weight combination that minimizes the root mean square error of grid-connected power tracking and has the lowest energy storage operation frequency was selected. To predict the length of the time domain.

5. The integrated wind and energy storage intelligent control method according to claim 1, characterized in that, The field-level cooperative power allocation optimization adopts the following model: ； The constraints are: ； ； ； in, Let be the field-level optimization objective function, representing the weighted sum of squares of scheduling command tracking error, single-unit output deviation, and energy storage state of charge deviation; This represents the total number of wind turbines. The power specified in the power grid dispatch command is obtained from the power grid dispatch command data. For the first The output of the typhoon generator; For the first Reference output of the typhoon generator; For the first The state of charge of the energy storage system at any given time is obtained from the state data of the energy storage system. The target state of charge for energy storage; , , These are the weighting coefficients. The value range is determined based on the tracking accuracy requirements of the scheduling instructions. , The value range is determined based on the requirement for balanced output of a single unit. , The value range is determined based on the accuracy requirements for maintaining the state of charge of energy storage. , , , By conducting adjustment experiments on different weight combinations in a simulation environment, the weight combination that minimizes the root mean square error of power tracking across the entire field and achieves the most balanced power distribution among the wind turbines was selected.

6. The integrated wind and energy storage intelligent control method according to claim 1, characterized in that, The scene recognition in step S3 employs a multi-dimensional feature fusion method: ； ； in, This is the characteristic vector of wind power output, which includes wind speed fluctuation rate. Wind direction change rate turbulence intensity All of these are calculated from the wind turbine operating data; This is a power grid dispatch feature vector, containing AGC command type and frequency regulation requirements. Peak shaving demand All of these are extracted from the power grid dispatch command data; This is the electricity price feature vector, containing real-time electricity prices. Electricity price trend All of these are calculated from the real-time electricity price data; The feature weights are obtained through training with historical data; This is the feature matching function.

7. The integrated wind and energy storage intelligent control method according to claim 6, characterized in that, The feature matching function The specific expression is: in, For the first 3D eigenvalues For the scene Next The mean of the dimensional features, For the scene Next Standard deviation of dimensional features and It is obtained by statistical clustering of various scenarios in historical operation data.

8. The integrated wind and energy storage intelligent control method according to claim 1, characterized in that, The multi-objective optimization pricing strategy in step S5 adopts the following model: ； ； ； The constraints are: ； ； in, For the first The power of the application during the time period; For the first The predicted electricity price for a given period is output by the electricity price prediction model. The probability of clearing out; To contribute practically; This is the risk aversion coefficient, with a range of values. ; For the first The deviation penalty price for a given time period represents the penalty fee corresponding to a unit of deviation power. It is determined by the trading rules of the electricity spot market and is set as a preset multiple of the real-time electricity price. The allowable deviation range indicates the upper limit of the proportion by which the declared power exceeds the predicted wind power value, with a value range of [value missing]. , Indicates the first The maximum allowable power level for a given time period; The total available power generation of the wind-storage system during the predicted period is calculated by integrating the wind power prediction curve over the predicted period and summing the amount of electricity that the energy storage system can release.

9. The integrated wind and energy storage intelligent control method according to claim 8, characterized in that, The clearing probability is calculated using the following formula: ； in, This is a market sensitivity parameter, reflecting the degree to which market prices affect the probability of clearing. The marginal clearing price threshold is obtained through regression analysis of historical market data.

10. A wind-storage integrated intelligent control system, characterized in that, include: The data acquisition module is used to collect real-time operating data of each wind turbine in the wind farm, status data of the energy storage system, grid dispatch command data, and real-time electricity price data. The wind power prediction module is used to generate wind power prediction curves for a preset future time period based on historical operating data, numerical weather forecast data, and real-time wind measurement data from wind turbine operation data, using a multi-scale spatiotemporal fusion prediction model. The scene recognition module is used to automatically identify the current operating scene and match the optimal control mode based on the current wind power output characteristics obtained from the wind power prediction curve, the grid dispatch instruction type in the grid dispatch instruction data, and the real-time electricity price signal in the real-time electricity price data. The collaborative control module is used to take the wind power prediction curve as the prediction input, adopt a hierarchical distributed model predictive control algorithm, realize local edge closed-loop control at the single-unit power distribution and storage level, realize collaborative power distribution optimization of multiple wind turbines and energy storage systems at the field level, and output the output commands of each wind turbine and the charging and discharging commands of energy storage as control commands. The spot market trading module is used to train an electricity price prediction model based on wind power prediction curves and real-time electricity price data, generate minute-level bidding schemes using a multi-objective optimization bidding strategy, and automatically submit them to the electricity spot market trading platform. The execution feedback module is used to send control commands to each execution unit, monitor the execution effect in real time, and perform closed-loop correction.

Images