Cooperative peak regulation method and system based on wind power energy storage system

Abstract

The invention relates to the technical field of wind power energy storage, and discloses a collaborative peak regulation method and system based on a wind power energy storage system, and the method comprises the steps: collecting the detail data and SCADA data of a power generation fan, carrying out the time sequence alignment, constructing a prediction error model, calculating a standardized prediction error, constructing a hybrid prediction model, and dynamically adjusting the model parameters; a time-function double-layer decomposition strategy is constructed to calculate an energy storage attenuation value, a multi-objective optimization model is constructed in combination with the energy storage attenuation value, cross-day balance is carried out, and a charging and discharging strategy is optimized; constructing a three-channel reward function, dynamically adjusting the weight coefficient of the reward function, and automatically adjusting the weight according to the real-time state of the power grid to realize multi-target dynamic balance; the method comprises the following steps: constructing a hybrid hierarchical control execution framework, optimizing an edge layer, obtaining a wind power fluctuation rate, judging whether to trigger super capacitor pre-charging, performing model prediction control MPC optimization, realizing credible synchronization of energy storage SOC data in combination with an intelligent contract, and performing health state early warning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind power energy storage, in particular to a collaborative peak shaving method and system based on a wind power energy storage system. BACKGROUND

[0002] Wind power fluctuation has randomness, intermittency and seasonality, while the capacity, power and response speed of existing energy storage technology have inherent limitations; the energy storage system has charging and discharging efficiency loss, and the efficiency further decreases in extreme environments; the energy density of wind power itself is low, and after superimposing the energy storage loss, the overall energy utilization efficiency is weakened, especially in the high proportion of wind power access scenarios, the economy decreases obviously; the existing dispatching algorithm is mostly based on historical data or static model, and it is difficult to adapt to the dynamic fluctuation of wind power and the state change of energy storage in real time; In the prior art, existing patents often regard energy storage as a single ideal model, or only simply distinguish battery types, and lack deep modeling and optimization of the collaborative use of different energy storage technologies in response speed, cycle life, charging and discharging efficiency curve, self-discharge rate, aging cost and environmental adaptability; the collaborative strategy proposed in the prior art is too simple, and fails to fully utilize the complementary advantages of different types of energy storage at different time scales, and lacks fine consideration of the actual operation constraints of the energy storage system in the dynamic optimization model, ignoring the complex energy management and balance control problems in the energy storage system; Therefore, it is necessary to provide a collaborative peak shaving method and system based on a wind power energy storage system. SUMMARY

[0003] The purpose of the present application is to provide a collaborative peak shaving method and system based on a wind power energy storage system, in order to solve the above-mentioned prior art problems, the present application realizes the following technical scheme: In a first aspect, the collaborative peak shaving method based on a wind power energy storage system provided by the embodiments of the present application specifically comprises the following steps: Step one: collect the detailed data and SCADA data of the power generator and perform time sequence alignment, build a prediction error model to calculate the standardized prediction error, build a hybrid prediction model based on the standardized prediction error, introduce real-time error cloud feedback closed loop correction, and dynamically adjust the model parameters; Step two: build a time-function double-layer decomposition strategy based on the built hybrid prediction model to calculate the energy storage attenuation value, build a multi-objective optimization model combined with the energy storage attenuation value, perform cross-day balance through virtual power plant collaborative hydrogen storage, and optimize the charging and discharging strategy; Step three: build a three-channel reward function based on the multi-objective optimization model, dynamically adjust the weight coefficients of the reward function through a blockchain smart contract, automatically adjust the weight according to the real-time state of the power grid through the smart contract, and realize the dynamic balance of multiple objectives; Step 4: Construct a hybrid hierarchical control execution architecture based on the three-channel reward function, optimize the edge layer, obtain wind power volatility and determine whether to trigger supercapacitor pre-charging, perform model predictive control (MPC) optimization, combine smart contracts to achieve reliable synchronization of energy storage SOC data, and provide health status early warning.

[0004] Furthermore, the method for calculating the standardized prediction error is as follows: A prediction error model based on federated learning is constructed, with each wind farm acting as a local node. The model is trained using local data, and the result is obtained through the following formula: ; The standardized prediction error of the t-th sample is calculated. ,in, Let t be the actual value of the t-th sample. Let be the predicted value for the t-th sample. This is the actual average of the sample. Let be the total number of predicted values ​​for the t-th sample. The root mean square error of the prediction error for all samples; Standardized prediction error Follows a Gaussian mixture distribution: ; in, The weights of each Gaussian component are given. The mean of the Gaussian components. Let covariance matrix be the variance matrix. For modal number, For Gaussian components; During the federated learning optimization process, the local model parameters of each wind farm are encrypted by Paillier and then uploaded to the cloud for aggregation. The aggregated global model is then distributed to each local node to update the local prediction model. Furthermore, the method for constructing the hybrid prediction model is as follows: By using the obtained standardized prediction error, a hybrid prediction model is constructed, a real-time error cloud feedback closed-loop correction is introduced, and the model parameters are dynamically adjusted by integrating UAV inspection data. By combining the Attention mechanism with the LSTM network, the Attention mechanism is used to capture key time step information in the wind power output data, the LSTM network processes the temporal dependencies, and the prediction results are optimized by the Unscented Kalman Filter (UKF) to form an Attention-LSTM-UKF hybrid prediction model. A real-time error cloud model is established, incorporating the error data from each prediction. By analyzing the distribution characteristics and trends of the errors, a correction factor is generated and fed back to the hybrid prediction model to dynamically adjust the prediction parameters, forming a closed-loop correction mechanism. The detail data of the fan blade three-dimensional scanning collected by the unmanned aerial vehicle is converted into feature parameters, which are input into the hybrid prediction model, the weight and threshold of the model are dynamically adjusted, and the prediction result accurately reflects the actual operation state of the fan; Further, the method for calculating the energy storage attenuation value is: In the time dimension, a day is divided into multiple time periods of the same length, and the power grid load characteristics and wind power output in different time periods are decomposed; In the functional dimension, the functional positioning of different types of supercapacitors, lithium batteries and hydrogen energy storage is clear, supercapacitors are used to cope with short-term high-frequency fluctuations, lithium batteries are used for intraday peak shaving, and hydrogen energy storage is used for cross-day balance; The discharge depth and cycle life of different types of energy storage are obtained, and the formula is: ; The energy storage attenuation value is calculated , wherein, is the discharge depth of the i-th battery, is the cycle life of different types of batteries, which is updated in real time by the federal learning SOH model, is the charge and discharge power of the i-th battery at time t; Further, the method for constructing a multi-objective optimization model is: Obtain the power grid load characteristics and wind power output in different time periods for decomposition; According to the power grid load characteristics and wind power output in different time periods, a multi-objective optimization model is constructed in combination with the calculated energy storage attenuation value: ; , wherein, is the power grid purchase expenditure function, is the power purchased from the power grid at time t, is the energy storage loss weight coefficient, is the energy storage attenuation value of the wind energy storage at time t, is the wind power penalty coefficient, is the wind power at time t; Further, the method for constructing a three-channel reward function is: A deep reinforcement model DRL is designed, and the state space is set as: ; , wherein, is the state of charge of the energy storage at time t, is the corrected wind power output at time t, is the power grid load at time t, is the power grid frequency, is the standardized prediction error, is the newly added crack index, is the hydrogen storage capacity; The action space is set as: ; wherein, is the charge and discharge power of the super capacitor, is the charge and discharge power of the lithium battery, supporting hydrogen storage while participating in electrolysis and discharge, is the electrolysis power at time t, is the discharge power of the fuel cell at time t; The technical benefit channel considers the wind power consumption rate and the energy storage response speed technical index; the economic benefit channel covers the electricity purchase expenditure and the energy storage loss expenditure; the safety penalty channel punishes for the power grid frequency deviation, voltage fluctuation, and equipment failure risk; the values of the three channels form the reward function; Further, the method for dynamically adjusting the reward function weight coefficient is: The reward function weight coefficient is dynamically adjusted by the blockchain smart contract, through the formula: ; The reward function weight coefficient is calculated as wherein, is the immediate reward of the i-th channel, is the power grid safety utility, is the actual frequency of the power grid, is the voltage deviation, is the rated voltage, the weight is automatically adjusted according to the real-time state of the power grid through the smart contract, and the dynamic balance of multiple targets is realized; The dynamic reward function weight coefficient adjustment rule and the reward function calculation method are written into the blockchain smart contract; the smart contract automatically executes the reward function weight coefficient adjustment and reward calculation; Further, the method for performing model predictive control MPC optimization is: The edge layer is optimized, the CEP engine is deployed to process high-frequency data in real time, the wind power fluctuation rate is obtained within a preset time period, if an event of identifying that the wind power fluctuation rate exceeds a preset fluctuation rate threshold is identified, the super capacitor is pre-charged, and at the same time, the model predictive control MPC optimization is performed, and the optimization function is: ; wherein, is the prediction time domain, which is preset as 15 minutes, and the control time domain M is preset as 5 minutes, and are weight coefficients, is the actual power of the battery at time k, is the reference power at time k, The energy storage attenuation value of the wind power energy storage at the k moment is calculated as follows: The end layer control rule is that the super capacitor adopts adaptive PID control, and the super capacitor parameters are adjusted online, and the adjustment formula is as follows: The adjustment proportion coefficient is calculated as follows , wherein, is the initial proportion coefficient, is the main frequency component of the Fourier transform of the wind power change rate, is the wind power; If the communication is interrupted, the end layer controller makes autonomous decisions based on local data to ensure stable operation of the system; Further, the method for early warning of the health state is as follows: A local light node is deployed to store the SOC change chain in the last 24 hours, each block contains a hash, a timestamp, an SOC value and a battery management system BMS signature, and a distributed consensus algorithm is used to ensure that the data of each node is consistent; When the network is recovered, the real SOC is calculated by the formula , wherein, is the SOC value recorded by each node, is the corresponding signature; A three-dimensional digital twin model of the battery pack is constructed, the internal structure data obtained by X-ray tomography and the electrochemical impedance spectroscopy EIS data are fused, and the remaining useful life RUL is predicted: , wherein, is the initial capacity, is the current capacity, is the capacity attenuation amount of each cycle; if the RUL is greater than 300, the charge and discharge depth DoD is automatically adjusted to below 50%, thereby prolonging the battery life.

[0005] In the second aspect, the cooperative peak shaving system based on the wind power energy storage system provided by the embodiments of the present application specifically includes the following modules: The fusion prediction module collects the detailed data and SCADA data of the power generator and performs time sequence alignment, constructs a prediction error model to calculate the standardized prediction error, constructs a hybrid prediction model based on the standardized prediction error, introduces real-time error cloud feedback closed-loop correction, and dynamically adjusts the model parameters; The balance optimization module calculates the energy storage attenuation value based on the construction of the hybrid prediction model, constructs a multi-objective optimization model combined with the energy storage attenuation value, performs cross-day balance through virtual power plant cooperation with hydrogen storage, and optimizes the charging and discharging strategy; ​​Dynamic decision module: based on a multi-objective optimization model, a three-channel reward function is constructed, the weight coefficient of the reward function is dynamically adjusted through a blockchain smart contract, and the weight is automatically adjusted according to the real-time state of the power grid through the smart contract, so that the dynamic balance of multiple objectives is realized; Control warning module: based on the three-channel reward function, a hybrid hierarchical control execution architecture is constructed, the edge layer is optimized, the wind power fluctuation rate is obtained and whether the super capacitor pre-charging is triggered is judged, the model predictive control (MPC) optimization is carried out, the SOC data of the energy storage is realized in a trusted manner through the smart contract, and the health state warning is carried out.

[0006] The beneficial effects of the present application are: 1. Deploying unmanned aerial vehicles carrying laser radars and high-definition cameras to realize three-dimensional scanning of detailed data and break through the limitations of traditional SCADA data; each wind farm is trained as a local node to independently train a prediction error model, and the parameters are aggregated after being encrypted by Paillier, so that the data is not out of the domain and the generalization ability of the global model is improved; the dynamic time warping (DTW) algorithm is used to align the SCADA data and satellite cloud images, solving the problem of inconsistent timestamps of multiple source data; the battery health state is updated in real time, and the energy storage attenuation value is calculated in combination with the discharge depth and cycle life; a hydrogen storage dynamic balance equation is established to optimize the priority charging strategy; 2. Write the reward calculation rules and weight adjustment logic into the smart contract, and record all decision records on the chain to ensure fairness and auditability; formulate a hydrogen storage cross-day plan and virtual power plant collaborative scheduling, and use adaptive PID control to cope with communication interruption to ensure system stability; a three-dimensional digital twin model of the battery pack is constructed to predict the remaining life; distributed consensus algorithm is used to synchronize the energy storage SOC data, and the local light node stores the 24-hour change chain, and after the network is recovered, the real SOC is restored through the majority voting mechanism. BRIEF DESCRIPTION OF DRAWINGS

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creating any creative labor.

[0008] Figure 1 is a step flow chart of the collaborative peak shaving method based on the wind power energy storage system provided by the embodiment 1 of the present application; Figure 2 is a structural schematic diagram of the collaborative peak shaving system based on the wind power energy storage system provided by the embodiment 2 of the present application. DETAILED DESCRIPTION

[0009] In order to make the person skilled in the art better understand the technical scheme of the present application, the technical scheme in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, and not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by the person skilled in the art without creative labor should belong to the protection scope of the present application.

[0010] Embodiment 1 As Figure 1 shown, the method for coordinated peak shaving based on a wind power energy storage system provided by the embodiments of the present application specifically includes the following steps: Step one: collect the detail data and SCADA data of the power generation wind turbine and perform time sequence alignment, build a prediction error model to calculate the standardized prediction error, build a hybrid prediction model based on the standardized prediction error, introduce real-time error cloud feedback closed-loop correction, and dynamically adjust the model parameters; In a specific embodiment, a UAV is deployed to carry a laser radar and a high-definition camera, a preset scanning period is set to perform three-dimensional scanning on the wind turbine blades, the scanning frequency of the laser radar is preset to 10 Hz, the detail data of the blades is obtained, and the detail data includes but is not limited to: real-time morphology, cracks; The SCADA data of the power generation wind turbine is collected, the SCADA data includes but is not limited to: wind speed, power generation wind turbine speed, and power generation wind turbine output; a satellite cloud image updated once per preset update period is used to analyze the weather condition; The dynamic time warping algorithm (DTW) is used to align the data with different sampling frequencies; For the SCADA data of the power generation wind turbine and the satellite cloud image updated once per preset update period, the optimal time matching path between the two is calculated by the DTW algorithm, the accurate correspondence in time sequence is realized, and the consistency of the data in the time dimension is ensured; A prediction error model based on federated learning is built, each wind farm is used as a local node, the model is trained by using the local data, and the formula is: ; The standardized prediction error of the tth sample is calculated , wherein, is the actual value of the tth sample, is the prediction value of the tth sample, is the actual average value of the sample, is the total number of prediction values of the tth sample, is the root mean square error of all sample prediction errors; The standardized prediction error obeys a mixed Gaussian distribution: ; wherein, are weights of the Gaussian components, are means of the Gaussian components, is a covariance matrix, is a modal number, is a Gaussian component; In the federal learning optimization process, the local model parameters of each wind farm are uploaded to the cloud after Paillier encryption for aggregation, and the global model after aggregation is issued to each local node to update the local prediction model, effectively ensuring data privacy; The modal number K is dynamically adjusted according to the real-time weather type, K=2 for sunny days, K=3 for cloudy days, and K=4 for heavy rain, and the adjustment is automatically triggered by the meteorological radar echo intensity; For example, if the meteorological radar echo intensity is lower than the preset intensity threshold, it is determined as sunny, and K is automatically set to 2; By obtaining the standardized prediction error, a hybrid prediction model is constructed, real-time error cloud feedback is introduced for closed-loop correction, and unmanned aerial vehicle inspection data is fused to dynamically adjust model parameters to reduce thick-tailed errors; Specifically, the Attention mechanism is combined with the LSTM network, the Attention mechanism is used to capture the key time step information in the wind power output data, the LSTM network processes the time sequence dependency relationship, and then the prediction result is optimized through the unscented Kalman filter UKF, forming an Attention-LSTM-UKF hybrid prediction model; A real-time error cloud model is established, the error data of each prediction is included in the model, the distribution characteristics and change trend of the error are analyzed, a correction factor is generated and fed back to the hybrid prediction model, the prediction parameters are dynamically adjusted, and a closed-loop correction mechanism is formed; The detailed data of the fan blade three-dimensional scanning collected by the unmanned aerial vehicle is converted into feature parameters and input into the hybrid prediction model, the weights and thresholds of the model are dynamically adjusted, and the prediction result accurately reflects the actual running state of the fan; Step two: based on the prediction error model, a time-function double-layer decomposition strategy is constructed to calculate the energy storage attenuation value, a multi-objective optimization model is constructed combined with the energy storage attenuation value, and the virtual power plant cooperates with hydrogen storage for cross-day balance to optimize the charging and discharging strategy; In specific embodiments, a time-function double-layer decomposition strategy is constructed based on a prediction error model, cross-day balance is achieved through virtual power plant VPP cooperation with hydrogen storage, and the charging and discharging strategy is optimized combined with the energy storage life loss model; Specifically, in the time dimension, a day is divided into multiple time periods of the same length, and the power grid load characteristics and wind power output in different time periods are obtained for decomposition; In the functional dimension, the functional positioning of different types of energy storage such as supercapacitors, lithium batteries and hydrogen energy storage is clarified. Supercapacitors are used to cope with short-term high-frequency fluctuations, lithium batteries are used for intraday peak shaving, and hydrogen energy storage is used for cross-day balancing. The discharge depth and cycle life of different types of energy storage are obtained, and the formula is: ; The energy storage attenuation value is calculated , wherein, is the discharge depth of the i-th battery, is the cycle life of different types of batteries, which is updated in real time by federated learning SOH model, is the charge and discharge power of the i-th battery at time t; According to the load characteristics and wind power output of different time periods, a multi-objective optimization model is constructed combining the calculated energy storage attenuation value: ; , wherein, is the grid power purchase expenditure function, is the power purchased from the grid at time t, is the energy storage loss weight coefficient, is the energy storage attenuation value of wind energy storage at time t, is the wind power penalty coefficient, is the wind power at time t; The hydrogen energy storage is balanced across the day through the cross-day balancing equation of hydrogen energy storage: ; The hydrogen storage at time t is calculated , wherein, is the hydrogen storage at time t-1, is the electrolysis efficiency, is the electrolysis power at time t, is the fuel cell efficiency, is the discharge power of the fuel cell at time t, the electrolysis efficiency , and the fuel cell efficiency , is the battery temperature, and the electrolysis efficiency and fuel cell efficiency are dynamically adjusted according to the battery temperature; Based on the time-function double-layer decomposition strategy, the surrounding distributed photovoltaic and electric vehicle V2G resources are accessed, the predicted output of distributed photovoltaic and the charging demand of electric vehicles are predicted through federated learning, and the predicted output of distributed photovoltaic and the charging demand of electric vehicles are included in the optimization model. The hydrogen storage charge and discharge plan is optimized to realize multi-energy collaboration. Step three: based on the multi-objective optimization model, a three-channel reward function is constructed, and the weight coefficients of the reward function are dynamically adjusted through the smart contract of the blockchain. The weight is automatically adjusted according to the real-time state of the power grid through the smart contract, and the dynamic balance of multiple objectives is realized. Based on the multi-objective optimization model, a three-channel reward function is constructed, and the dynamic weight mechanism and the smart contract of the blockchain are combined to realize the dynamic balance of multiple objectives. Specifically, a deep reinforcement model DRL is designed, and the state space is set as: ; Among them, is the state of charge of the energy storage at time t, is the corrected wind power output at time t, is the power grid load at time t, is the grid frequency, is the standardized prediction error, is the newly added crack index, is the hydrogen reserve; The action space is set as: ; Among them, is the charge and discharge power of the super capacitor, is the charge and discharge power of the lithium battery, which supports hydrogen storage while participating in electrolysis and discharge, is the electrolysis power at time t, is the discharge power of the fuel cell at time t; The technical benefit channel considers the wind power consumption rate and the technical index of energy storage response speed; the economic benefit channel covers the purchase of electricity and the expenditure of energy storage loss; the safety penalty channel punishes the grid frequency deviation, voltage fluctuation and equipment failure risk; the values of the three channels are combined to form a reward function; Based on the obtained reward function of the three channels, the weight coefficients of the reward function are dynamically adjusted through the smart contract of the blockchain. The weight is automatically adjusted according to the real-time state of the power grid through the smart contract, and the dynamic balance of multiple objectives is realized. ; The reward function weight coefficients are calculated , wherein, is the immediate reward of the i-th channel, is the safety utility of the power grid, is the actual frequency of the power grid, is the voltage deviation, is the rated voltage, and the weight is automatically adjusted according to the real-time state of the power grid through the smart contract, and the dynamic balance of multiple objectives is realized. The dynamic reward function weight coefficient adjustment rule and reward function calculation method are written into the blockchain smart contract, ensuring the transparency, fairness and tamper resistance of the decision-making process; the smart contract automatically performs reward function weight coefficient adjustment and reward calculation, improving decision-making efficiency; Step four: based on the three-channel reward function, a hybrid hierarchical control execution architecture is constructed, the edge layer is optimized, the wind power fluctuation rate is obtained, and it is judged whether the super capacitor pre-charging is triggered, and the model predictive control (MPC) optimization is performed, through the distributed state consensus mechanism, combined with the smart contract, the energy storage SOC data is synchronized in a trusted manner, and the health state warning is performed; Based on the three-channel reward function, a hybrid hierarchical control execution architecture is constructed, through a three-level collaborative architecture of cloud-edge-end, the edge layer introduces complex event processing (CEP) technology to improve response speed, and the end layer adopts adaptive PID control to cope with communication interruption; Specifically, the cloud is responsible for global optimization decision-making, including: formulating hydrogen storage cross-day balance plan, coordinating virtual power plant scheduling; the edge layer is deployed in the local wind farm, and processes high-frequency data and real-time control; the end layer is the controller of various energy storage devices, which executes specific control instructions; The edge layer is optimized, the CEP engine is deployed to process high-frequency data in real time, the wind power fluctuation rate is obtained within a preset time period, if an event that the wind power fluctuation rate exceeds the preset fluctuation rate threshold is identified, the super capacitor pre-charging is triggered, and the model predictive control (MPC) optimization is performed, and the optimization function is: ; Wherein, is the prediction time domain, which is preset to 15 minutes, and the control time domain M is preset to 5 minutes, and are weight coefficients, is the actual power of the battery at time k, is the reference power at time k, is the energy storage attenuation value of wind power storage at time k; End layer control rule: super capacitor adopts adaptive PID control, super capacitor parameters are adjusted online, and the adjustment formula is: ; The adjustment proportion coefficient is calculated , wherein, is the initial proportion coefficient, is the main frequency component of the Fourier transform of the wind power change rate, is the wind power; If communication interruption occurs, the end layer controller makes autonomous decision based on local data to ensure stable operation of the system; Through the distributed state consensus mechanism of the blockchain, the SOC data of the energy storage is synchronously and reliably realized by combining with the smart contract, and the digital twin is introduced for health state early warning. Specifically, a local light node is deployed to store the SOC change chain in the last 24 hours, each block contains a hash, a timestamp, an SOC value and a battery management system (BMS) signature, and a distributed consensus algorithm is used to ensure the consistency of data of each node. When the network is recovered, the real SOC is calculated by the formula , wherein is the SOC value recorded by each node, is the corresponding signature, and a majority voting mechanism is used to prevent malicious node attacks; A three-dimensional digital twin model of the battery pack is constructed, the internal structure data obtained by X-ray tomography and electrochemical impedance spectroscopy (EIS) data are fused, and the remaining useful life (RUL) is predicted: . , wherein is the initial capacity, is the current capacity, is the capacity attenuation amount of each cycle; if the RUL is greater than 300, the charge and discharge depth (DoD) is automatically adjusted to below 50% to prolong the battery life; The blockchain nodes are regularly audited for security to prevent data leakage and tampering; the digital twin model monitors the equipment state in real time, and when an anomaly is found, an early warning is sent in time to remind the operation and maintenance personnel to handle it; For example, the unmanned aerial vehicle inspects once an hour, the laser radar scanning frequency is 10 Hz, and the satellite cloud image update cycle is 15 minutes; the federated learning aggregation cycle is adjusted according to the working conditions, once a day under normal working conditions and once an hour under extreme weather conditions, to ensure that the model can adapt to weather changes in time; the edge node updates the MPC prediction every 5 seconds to quickly respond to changes in wind power output and grid load; the cloud optimizes the hydrogen storage cross-day plan every day to realize long-term stable operation; the virtual power plant aggregates the surrounding resources to participate in the day-ahead market bidding, and optimizes the income distribution according to the bidding results; the super capacitor has a response time of less than 10 ms, which can quickly smooth short-term fluctuations; the lithium battery MPC control delay is less than 50 ms, which meets the daily peak shaving demand; the hydrogen storage system is preferentially charged during the low valley period of the power grid (23:00-7:00) and discharged during the peak period to improve energy utilization efficiency; the digital twin model is updated every hour, the RUL prediction error is less than 10%, and accurate basis is provided for equipment maintenance; the blockchain stores all SOC change records, supports 10-year data tracing, and is convenient for fault analysis and responsibility identification; IEC61850 protocol is used to realize cross-platform communication, DNP3.0 and Modbus protocol conversion are supported, and smooth communication between all parts of the system is ensured.

[0011] Example 2: AsFigure 2 As shown, the wind power energy storage system based coordinated peak shaving system provided by the embodiment of the present application specifically comprises the following modules: The fusion prediction module collects the detailed data and SCADA data of the power generator and performs time sequence alignment, constructs a prediction error model to calculate the standardized prediction error, constructs a hybrid prediction model based on the standardized prediction error, introduces real-time error cloud feedback closed-loop correction, and dynamically adjusts the model parameters. The balance optimization module calculates the energy storage attenuation value based on the constructed hybrid prediction model, constructs a multi-objective optimization model combined with the energy storage attenuation value, performs cross-day balance through virtual power plant coordination with hydrogen storage, and optimizes the charging and discharging strategy. The dynamic decision module constructs a three-channel reward function based on the multi-objective optimization model, dynamically adjusts the weight coefficient of the reward function through a blockchain smart contract, automatically adjusts the weight according to the real-time state of the power grid through the smart contract, and realizes dynamic balance of multiple objectives. The control and warning module constructs a hybrid hierarchical control execution architecture based on the three-channel reward function, optimizes the edge layer, obtains the wind power fluctuation rate and determines whether to trigger the super capacitor pre-charge, performs model predictive control (MPC) optimization, realizes reliable synchronization of energy storage SOC data combined with the smart contract, and performs health state warning.

[0012] The above describes one embodiment of the present application in detail, but the content described is only the preferred embodiment of the present application and cannot be considered as limiting the scope of the present application; the above formulas are all de-dimensioned to calculate the numerical values, the formulas are obtained by software simulation of a large amount of data to obtain a formula of the most recent real situation, the preset parameters in the formula are set by the person skilled in the art according to the actual situation and historical experience, and can be adjusted according to the actual situation; the above described is only the preferred embodiment of the present application and cannot be considered as limiting the scope of the present application; all equivalent changes and improvements made according to the scope of the present application should belong to the patent coverage range of the present application.

Claims

1. A collaborative peak-shaving method based on wind power energy storage systems, characterized in that, Includes the following steps: Collect detailed data of the wind turbine and SCADA data and align them in time series. Build a prediction error model to calculate the standardized prediction error. Based on the standardized prediction error, build a hybrid prediction model, introduce real-time error cloud feedback closed-loop correction, and dynamically adjust the model parameters. Based on the construction of a hybrid prediction model, a time-function dual-layer decomposition strategy is constructed to calculate the energy storage attenuation value. Combined with the energy storage attenuation value, a multi-objective optimization model is constructed. Through virtual power plant coordination with hydrogen storage, cross-day balancing is achieved, and the charging and discharging strategy is optimized. A three-channel reward function is constructed based on a multi-objective optimization model. The weight coefficients of the reward function are dynamically adjusted through a blockchain smart contract. The weights are automatically adjusted according to the real-time status of the power grid through the smart contract, thereby achieving a dynamic balance of multiple objectives. A hybrid hierarchical control execution architecture is constructed based on a three-channel reward function. The edge layer is optimized to obtain wind power volatility and determine whether to trigger supercapacitor pre-charging. Model predictive control (MPC) optimization is performed, and smart contracts are combined to achieve reliable synchronization of energy storage SOC data and provide health status early warning.

2. The collaborative peak-shaving method based on wind power energy storage system according to claim 1, characterized in that, The method for calculating the standardized prediction error is as follows: A prediction error model based on federated learning is constructed, with each wind farm acting as a local node. The model is trained using local data, and the result is obtained through the following formula: The standardized prediction error of the t-th sample is calculated. ,in, Let t be the actual value of the t-th sample. Let be the predicted value for the t-th sample. This is the actual average of the sample. Let be the total number of predicted values ​​for the t-th sample. The root mean square error of the prediction error for all samples; Standardized prediction error Follows a Gaussian mixture distribution: in, The weights of each Gaussian component are given. The mean of the Gaussian components. Let covariance matrix be the variance matrix. For modal number, For Gaussian components; During the federated learning optimization process, the local model parameters of each wind farm are encrypted by Paillier and uploaded to the cloud for aggregation. The aggregated global model is then distributed to each local node to update the local prediction model.

3. The collaborative peak-shaving method based on wind power energy storage system according to claim 1, characterized in that, The method for constructing the hybrid prediction model is as follows: By using the obtained standardized prediction error, a hybrid prediction model is constructed, a real-time error cloud feedback closed-loop correction is introduced, and the model parameters are dynamically adjusted by integrating UAV inspection data. By combining the Attention mechanism with the LSTM network, the Attention mechanism is used to capture key time step information in the wind power output data, the LSTM network processes the temporal dependencies, and the prediction results are optimized by the Unscented Kalman Filter (UKF) to form an Attention-LSTM-UKF hybrid prediction model. A real-time error cloud model is established, incorporating the error data from each prediction. By analyzing the distribution characteristics and trends of the errors, a correction factor is generated and fed back to the hybrid prediction model to dynamically adjust the prediction parameters, forming a closed-loop correction mechanism. The detailed data of the three-dimensional scan of the wind turbine blades collected by the UAV are transformed into feature parameters and input into the hybrid prediction model. The weights and thresholds of the model are dynamically adjusted so that the prediction results accurately reflect the actual operating status of the wind turbine.

4. The collaborative peak-shaving method based on wind power energy storage system according to claim 1, characterized in that, The method for calculating the energy storage attenuation value is as follows: In terms of time, a day is divided into multiple periods of equal length, and then decomposed according to the grid load characteristics and wind power output of different periods; In terms of functionality, the functional positioning of different types of energy storage, such as supercapacitors, lithium batteries, and hydrogen energy storage, is clearly defined. Supercapacitors are used to cope with short-term high-frequency fluctuations, lithium batteries are used for intraday peak shaving, and hydrogen energy storage is used for cross-day balancing. The depth of discharge and cycle life of different types of energy storage can be obtained using the formula: The energy storage attenuation value was calculated. ,in, Let be the depth of discharge of the i-th type of battery. The cycle life of different battery types is updated in real time using a federated learning SOH model. Let be the charging and discharging power of the i-th type of battery at time t.

5. The collaborative peak-shaving method based on wind power energy storage system according to claim 1, characterized in that, The method for constructing the multi-objective optimization model is as follows: The characteristics of the power grid load and wind power output at different time periods are obtained and decomposed. Based on the grid load characteristics and wind power output at different times, a multi-objective optimization model is constructed using the calculated energy storage attenuation values: in, For the power grid's electricity purchase expenditure function, Let t be the power purchased from the grid. This is the weighting coefficient for energy storage losses. Let be the energy storage degradation value of wind power energy storage at time t. This is the wind curtailment penalty coefficient. Let t be the wind curtailment power at time t.

6. The collaborative peak-shaving method based on wind power energy storage system according to claim 1, characterized in that, The method for constructing the three-channel reward function is as follows: Design a deep reinforcement learning (DRL) model with the state space set as follows: in, Let t be the state of charge of the stored energy. The wind power output is corrected at time t. Let be the grid load at time t. For the power grid frequency, To standardize the prediction error, For the newly added crack index, Hydrogen reserves; The motion space is set as follows: in, This refers to the charging and discharging power of the supercapacitor. The charging and discharging power of lithium batteries supports hydrogen storage while simultaneously participating in electrolysis and discharging. Let be the electrolysis power at time t. Let be the discharge power of the fuel cell at time t; The technology benefit channel considers technical indicators such as wind power absorption rate and energy storage response speed; the economic benefit channel covers electricity purchase expenditure and energy storage loss expenditure; the safety penalty channel penalizes grid frequency deviation, voltage fluctuation and equipment failure risk; and the values ​​of the three channels are combined to form a reward function.

7. The collaborative peak-shaving method based on wind power energy storage system according to claim 1, characterized in that, The method for dynamically adjusting the weight coefficients of the reward function is as follows: The reward function weight coefficients are dynamically adjusted via blockchain smart contracts, using the formula: ； Calculate the weight coefficients of the reward function ,in, The instant reward for the i-th channel. For the purpose of power grid security, The actual frequency of the power grid. For voltage deviation, Based on the rated voltage, the weights are automatically adjusted according to the real-time status of the power grid through smart contracts to achieve dynamic balance of multiple objectives. The rules for adjusting the weight coefficients of the dynamic reward function and the calculation method of the reward function are written into the blockchain smart contract; the smart contract automatically executes the adjustment of the weight coefficients of the reward function and the calculation of the reward.

8. The collaborative peak-shaving method based on wind power energy storage system according to claim 1, characterized in that, The method for performing model predictive control (MPC) optimization is as follows: The edge layer is optimized by deploying a CEP engine to process high-frequency data in real time. Wind power volatility is acquired within a preset time period. If an event is identified where wind power volatility exceeds a preset volatility threshold, supercapacitor pre-charging is triggered. Simultaneously, model predictive control (MPC) optimization is performed. The optimization function is as follows: in, To predict the time domain, and These are the weighting coefficients. Let k be the actual power of the battery at time k. The reference power at time k, Let k be the energy storage attenuation value of wind power energy storage at time k; End-level control rules: The supercapacitor adopts adaptive PID control, and the supercapacitor parameters are adjusted online. The adjustment ratio coefficient is calculated through the adjustment formula. ; If communication is interrupted, the endpoint controller makes autonomous decisions based on local data to ensure stable system operation.

9. The collaborative peak-shaving method based on wind power energy storage system according to claim 1, characterized in that, The method for issuing health status early warnings is as follows: Deploy local light nodes to store the SOC change chain of the most recent 24 hours. Each block contains a hash, timestamp, SOC value and battery management system (BMS) signature. A distributed consensus algorithm is used to ensure data consistency across nodes. When a network outage occurs and recovery is performed, the formula is used. The true SOC is calculated, where, The SOC value recorded for each node. For the corresponding signature; A three-dimensional digital twin model of the battery pack was constructed, and internal structural data obtained from X-ray tomography and electrochemical impedance spectroscopy (EIS) data were integrated to predict the remaining lifetime (RUL). in, For the initial capacity, For the current capacity, This represents the capacity decay per cycle; if RUL is greater than 300, the depth of charge / discharge (DoD) will be automatically adjusted to below 50% to extend battery life.

10. A collaborative peak-shaving system based on wind power energy storage, wherein the method is used to execute the peak-shaving method according to any one of claims 1-9, characterized in that, include: Fusion prediction module: Collect detailed data of wind turbine generators and SCADA data and perform time-series alignment, build a prediction error model to calculate standardized prediction error, build a hybrid prediction model based on standardized prediction error, introduce real-time error cloud feedback closed-loop correction, and dynamically adjust model parameters; Balance optimization module: Based on the construction of a hybrid prediction model, a time-function two-layer decomposition strategy is used to calculate the energy storage decay value. Combined with the energy storage decay value, a multi-objective optimization model is constructed. Through virtual power plant and hydrogen storage, cross-day balance is achieved, and the charging and discharging strategy is optimized. Dynamic decision-making module: Based on a multi-objective optimization model, a three-channel reward function is constructed. The weight coefficients of the reward function are dynamically adjusted through blockchain smart contracts. The weights are automatically adjusted according to the real-time status of the power grid through smart contracts, so as to achieve dynamic balance of multiple objectives. Control and early warning module: Based on the three-channel reward function, a hybrid hierarchical control execution architecture is constructed, the edge layer is optimized, the wind power volatility is obtained and it is determined whether the supercapacitor pre-charging is triggered, the model predictive control MPC is optimized, and the energy storage SOC data is reliably synchronized by smart contracts, and the health status is warned.

Images