Frequency modulation response intelligent monitoring method based on power station internal energy storage system

Abstract

This invention discloses an intelligent monitoring method for frequency regulation response based on an energy storage system within a power plant, relating to the field of power energy storage monitoring technology. The method involves collecting voltage time-series data, current time-series data, and temperature monitoring data from each energy storage unit in the power plant's energy storage system to form a raw operating data set. Fluctuation feature analysis is performed on the voltage time-series data, and dynamic response feature extraction is performed on the current time-series data. A fused electrical feature set is generated through cross-modal feature alignment. An improved Kalman filter algorithm is employed, and the filter gain matrix is ​​dynamically adjusted in conjunction with temperature monitoring data to estimate the real-time state of charge (SOC) of the energy storage units. Frequency regulation response anomalies are determined based on the comparison results between the SOC and safety thresholds, and a correction command is generated to feed back to the controller to complete online calibration of the frequency regulation parameters. This method closely aligns with the actual operating conditions of the energy storage units, accurately identifies frequency regulation response anomalies, and is suitable for the complex and variable frequency regulation operation scenarios of power plant energy storage systems.

Description

Technical Field

[0001] This invention belongs to the field of power energy storage monitoring technology, specifically a method for intelligent monitoring of frequency regulation response of energy storage systems within power plants. Background Technology

[0002] In current monitoring of frequency regulation response of power plant energy storage systems, a processing mode of independent analysis of single electrical parameters is generally adopted. Only basic analytical calculations are performed on the voltage or current time series data of the energy storage unit. Temperature monitoring data is mostly used only for unit overheating alarms and is not involved in the electrical characteristic analysis and state estimation process. Conventional monitoring methods rely solely on isolated data to complete the identification of the operating status of the energy storage unit and frequency regulation response analysis.

[0003] The processing mode of independently analyzing single electrical parameters cannot establish the intrinsic correlation between voltage and current time-series data, making it difficult to effectively integrate the two types of electrical characteristics, and limiting the dimensions of electrical characteristic representation. Traditional Kalman filtering algorithms use a fixed filter gain matrix for state-of-charge estimation, ignoring the impact of temperature changes on electrical operating characteristics during energy storage unit operation, resulting in discrepancies between the state estimation results and the actual operating conditions of the unit. Insufficient adaptability to operating conditions can lead to deviations in frequency regulation response anomaly detection, and subsequent frequency regulation parameter calibration cannot match the real-time operating status of the energy storage system, making it difficult to adapt to the complex and variable frequency regulation operation scenarios of power plant energy storage systems. Summary of the Invention

[0004] This invention aims to solve at least one of the technical problems existing in the prior art;

[0005] Therefore, this invention proposes an intelligent monitoring method for frequency regulation response based on an energy storage system within a power plant, comprising: Acquire the set of raw operating data of the energy storage system in the power station collected in real time by the monitoring, control and data acquisition system. The set of raw operating data includes voltage time-series data, current time-series data and temperature monitoring data of each energy storage unit. The voltage time series data is subjected to fluctuation feature parsing processing to obtain a voltage fluctuation feature sequence. The current time series data is subjected to dynamic response feature extraction processing to obtain a current response feature sequence. The voltage fluctuation feature sequence and the current response feature sequence are subjected to cross-modal feature alignment processing to generate a fused electrical feature set. An improved Kalman filter algorithm is invoked to perform state estimation processing on the fused electrical feature set to generate a real-time state of charge estimation sequence of the energy storage unit, wherein the improved Kalman filter algorithm dynamically adjusts the filter gain matrix based on the temperature monitoring data; Based on the comparison results between the real-time state of charge estimation value sequence and the preset state of charge safety threshold, the energy storage unit is subjected to frequency regulation response anomaly judgment processing to generate an abnormal unit marking sequence. Based on the abnormal unit marking sequence, a dynamic correction command for frequency regulation response parameters is generated, and the dynamic correction command is fed back to the energy storage system controller to trigger the online calibration operation of frequency regulation parameters.

[0006] Further, the step of performing fluctuation feature analysis processing on the voltage time series data to obtain a voltage fluctuation feature sequence specifically includes: The voltage time series data is subjected to sliding window segmentation to obtain multiple voltage window data segments of equal duration; For each voltage window data segment, perform detrending processing and subtract the linear fit value of that voltage window data segment to obtain the detrended voltage segment; Local extremum detection processing is performed on the detrended voltage segment to mark all local maxima and local minima in the detrended voltage segment; Calculate the amplitude difference between adjacent local maxima and local minima in each voltage window data segment to obtain the set of fluctuation amplitudes for that voltage window data segment; The maximum fluctuation amplitude in the set of fluctuation amplitudes is taken as the representative fluctuation amplitude of the voltage window data segment; The representative fluctuation amplitudes of all voltage window data segments are arranged in chronological order to generate the voltage fluctuation characteristic sequence.

[0007] Further, the step of performing dynamic response feature extraction processing on the current time-series data to obtain a current response feature sequence specifically includes: Perform step change point detection processing on the current time series data to identify all step start times and step end times when the rate of change of the current value exceeds a preset rate of change threshold per unit time. Starting from the start time of each step and ending at the corresponding end time of the step, multiple step response data segments are extracted from the current timing data. For each step response data segment, a response time calculation is performed to calculate the time required for the current value to reach 90% of the steady-state value from the start of the step response data segment. This time is used as the rise time characteristic value of the step response data segment. For each step response data segment, overshoot calculation is performed. The ratio of the difference between the maximum peak value of the current value and the steady-state value in the step response data segment to the steady-state value is calculated and used as the overshoot characteristic value of the step response data segment. The rise time characteristic value and overshoot characteristic value of each step response data segment are combined to form the response characteristic vector of that step response data segment; The response feature vectors of all step response data segments are arranged in chronological order according to the start time of the step, thus generating the current response feature sequence.

[0008] Further, the step of performing cross-modal feature alignment processing on the voltage fluctuation feature sequence and the current response feature sequence to generate a fused electrical feature set specifically includes: Obtain the first acquisition timestamp corresponding to each voltage fluctuation feature value in the voltage fluctuation feature sequence, and the second acquisition timestamp corresponding to each current response feature vector in the current response feature sequence; A unified timeline grid is constructed using the minimum timestamp between the first and second collection timestamps as the alignment start point and the maximum timestamp as the alignment end point. Each voltage fluctuation feature value in the voltage fluctuation feature sequence is mapped to the corresponding time grid point in the unified time axis grid according to its first acquisition timestamp to obtain the aligned voltage sequence; Each current response feature vector in the current response feature sequence is mapped to the corresponding time grid point in the unified time axis grid according to its second acquisition timestamp to obtain the aligned current sequence; For each time grid point in the unified time axis grid, the voltage fluctuation feature value in the aligned voltage sequence and the current response feature vector in the aligned current sequence at that time grid point are concatenated to generate the fused feature vector of that time grid point. Collect the fused feature vectors of all time grid points to form the fused electrical feature set.

[0009] Furthermore, the step of invoking the improved Kalman filter algorithm to perform state estimation processing on the fused electrical feature set and generating a real-time state-of-charge estimation sequence of the energy storage unit specifically includes: Extract the fused feature vector for each time grid point from the fused electrical feature set, use the voltage fluctuation feature value in the fused feature vector as the observation input of the improved Kalman filter algorithm, and use the current response feature vector in the fused feature vector as the control input of the improved Kalman filter algorithm; The state of charge estimate of the previous time grid point is used as the prior state estimate. The control input is fed into the state transition matrix of the improved Kalman filter algorithm to generate the predicted state value of the current time grid point. The observations are input into the observation matrix of the improved Kalman filter algorithm to generate the predicted observations for the current time grid point; Obtain the temperature value corresponding to the current time grid point from the temperature monitoring data, and dynamically adjust the temperature compensation coefficient in the filter gain matrix of the improved Kalman filter algorithm according to the temperature value. The residual between the predicted state value and the predicted observation value is weighted and corrected using the adjusted filter gain matrix to generate the estimated state of charge value of the current time grid point. The state of charge estimates of all time grid points are arranged in chronological order to generate the real-time state of charge estimate sequence.

[0010] Furthermore, the step of performing frequency regulation response anomaly determination processing on the energy storage unit and generating an anomaly unit marker sequence based on the comparison result of the real-time state of charge estimation value sequence and the preset state of charge safety threshold specifically includes: Obtain the state of charge estimate value corresponding to each time grid point in the real-time state of charge estimate value sequence, and a preset state of charge safety threshold range, wherein the state of charge safety threshold range includes a lower limit threshold and an upper limit threshold. For each time grid point, determine whether the estimated state of charge at that time grid point is less than the lower limit threshold of the state of charge or greater than the upper limit threshold of the state of charge; When the estimated state of charge is less than the lower limit threshold or greater than the upper limit threshold, the time grid point is marked as an abnormal state of charge time point, and the energy storage unit identifier corresponding to the time grid point is recorded. For each energy storage unit, count the number of abnormal state of charge time points that occur in the energy storage unit within a continuous time window, and determine whether the number exceeds a preset abnormality counting threshold. When the number exceeds the abnormal count threshold, the energy storage unit is marked as an abnormal unit, and the unit identifier and the time period of the abnormality are recorded. Arrange the unit identifiers and time periods of all abnormal units in chronological order of detection time to generate the abnormal unit tag sequence.

[0011] Furthermore, the length of the continuous time window and the value of the anomaly counting threshold are dynamically configured based on the historical fault statistics data of the energy storage system.

[0012] Further, the step of generating a dynamic correction command for frequency regulation response parameters based on the abnormal unit marking sequence, and feeding back the dynamic correction command to the energy storage system controller to trigger the online calibration operation of the frequency regulation parameters, specifically includes: Parse the abnormal unit marker sequence to extract the unit identifier and the time period of the abnormality for each abnormal unit; For each abnormal unit, the historical frequency regulation response parameter records of the abnormal unit during the abnormal period are retrieved from the monitoring, control and data acquisition system. The historical frequency regulation response parameter records include the charging and discharging power limit value, frequency regulation response dead zone threshold and ramp rate limit value of the abnormal unit. The type of target parameter to be adjusted is determined based on the type of abnormality of the abnormal unit. When the abnormality is over-discharge, the charging and discharging power limit value is used as the target parameter. When the abnormality is over-charge, the ramp rate limit value is used as the target parameter. The target parameter is adjusted in a stepwise manner, and the value of the target parameter is increased or decreased stepwise according to a preset adjustment step size to generate multiple candidate parameter adjustment values. Each candidate parameter adjustment value is input into the improved Kalman filter algorithm for simulation state estimation, and the predicted state of charge trajectory corresponding to each candidate parameter adjustment value is calculated. Candidate parameter adjustment values ​​that make the predicted state of charge trajectory completely within the safe threshold range of the state of charge are selected, and the candidate parameter adjustment value corresponding to the smallest adjustment magnitude is taken as the final parameter correction value. The final parameter correction value is encapsulated into a frequency modulation response parameter dynamic correction command, and the dynamic correction command is sent to the energy storage system controller.

[0013] Furthermore, the charge / discharge power limitation term in the state transition matrix applies a hard constraint to the boundary of the predicted state value of the improved Kalman filter algorithm based on the final parameter correction value.

[0014] Furthermore, the step of inputting each candidate parameter adjustment value into the improved Kalman filter algorithm for simulation state estimation and calculating the predicted state-of-charge trajectory corresponding to each candidate parameter adjustment value specifically includes: Obtain the historical fusion electrical characteristic set of the abnormal unit during the normal operation period before the time of the abnormality, and use it as the simulation baseline data; The voltage fluctuation characteristic value in the simulation reference data is used as the simulation observation input, and the current response characteristic vector in the simulation reference data is used as the simulation control input. The adjusted candidate parameter values ​​are used as constraint boundaries to modify the charge / discharge power limit term in the state transition matrix of the improved Kalman filter algorithm. Using the estimated state of charge of the abnormal unit at the beginning of the abnormality period as the initial state of the simulation, the simulation observation input and the simulation control input are input sequentially, and the modified Kalman filter algorithm is run. Record the simulated state of charge estimate output by the modified and improved Kalman filter algorithm at each simulation time step; By concatenating the simulated state of charge estimates from all simulation time steps in chronological order, the predicted state of charge trajectory corresponding to the candidate parameter adjustment value is generated.

[0015] Compared with the prior art, the beneficial effects of the present invention are: Fluctuation feature analysis is performed on voltage time-series data of energy storage units, and dynamic response feature extraction is implemented on current time-series data. Then, cross-modal feature alignment processing is carried out on the two types of feature sequences to complete the feature integration of multi-source electrical data. This breaks the inherent pattern of analyzing single electrical parameters separately, establishes a feature correlation channel between voltage and current time-series data, integrates electrical operation information from different dimensions, enriches the representation dimensions of the electrical characteristics of energy storage units, and restores the actual electrical operation variation patterns of the units.

[0016] An improved Kalman filter algorithm incorporates temperature monitoring data from the energy storage unit. The filter gain matrix is ​​dynamically adjusted based on the unit's real-time temperature conditions. State estimation is then performed on the fused electrical feature set based on the adjusted filter parameters. The filter parameters can change in real-time with the energy storage unit's temperature conditions, adapting to the unit's electrical operating characteristics under different temperature environments. The state estimation process closely follows the actual operating condition changes of the unit, the criteria for classifying abnormal units align with actual field operations, and the frequency response parameter correction can adapt to real-time fluctuations in the energy storage system's operating status. Attached Figure Description

[0017] Figure 1 This is a flowchart of the intelligent monitoring method for frequency regulation response based on an in-power station energy storage system as described in this invention; Figure 2 A flowchart for analyzing and processing wave characteristics; Figure 3 This is a flowchart for dynamic response feature extraction processing. Detailed Implementation

[0018] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] See Figure 1A method for intelligent monitoring of frequency regulation response based on an energy storage system within a power plant is proposed. This method acquires a set of raw operating data of the energy storage system within the power plant, collected in real time by a monitoring and control system and a data acquisition system. This raw operating data set includes voltage time-series data, current time-series data, and temperature monitoring data for each energy storage unit. Next, fluctuation feature analysis is performed on the voltage time-series data to obtain a voltage fluctuation feature sequence. Simultaneously, dynamic response feature extraction is performed on the current time-series data to obtain a current response feature sequence. The voltage fluctuation feature sequence and the current response feature sequence are then aligned across modes to generate a fused electrical feature set. An improved Kalman filter algorithm is used to perform state estimation on the fused electrical feature set to generate a real-time state-of-charge (SOC) estimation sequence for the energy storage units. The improved Kalman filter algorithm dynamically adjusts the filter gain matrix based on temperature monitoring data. Then, based on the comparison between the real-time SOC estimation sequence and a preset SOC safety threshold, frequency regulation response anomaly detection processing is performed on the energy storage units to generate an abnormal unit marker sequence. Based on the abnormal unit marker sequence, a dynamic correction command for frequency regulation response parameters is generated and fed back to the energy storage system controller to trigger online calibration of the frequency regulation parameters.

[0020] In one embodiment of the present invention, when performing fluctuation feature analysis processing on voltage time-series data to obtain a voltage fluctuation feature sequence, refer to... Figure 2 The voltage time series data is segmented into multiple voltage window data segments of equal duration by a sliding window process. Detrending processing is performed on each voltage window data segment, and the linear fitting value of the voltage window data segment is subtracted to obtain a detrended voltage segment. Local extremum detection processing is performed on the detrended voltage segment to mark all local maxima and local minima in the detrended voltage segment. Then, the amplitude difference between adjacent local maxima and local minima in each voltage window data segment is calculated to obtain the fluctuation amplitude set of the voltage window data segment. The maximum fluctuation amplitude in the fluctuation amplitude set is taken as the representative fluctuation amplitude of the voltage window data segment. The representative fluctuation amplitudes of all voltage window data segments are arranged in chronological order to generate a voltage fluctuation feature sequence.

[0021] In practical implementation, for a specific energy storage unit within a power plant's energy storage system, the monitoring, control, and data acquisition system continuously collects the voltage time-series data of that unit. This data records a sequence of voltage values ​​sampled every 100 milliseconds. To analyze the fluctuation characteristics in this voltage time-series data, during sliding window segmentation, the original voltage time-series data is divided into multiple voltage window data segments of equal duration, with a preset window width of 5 seconds and a sliding step size of 1 second. Each voltage window data segment contains 50 voltage sampling points. Detrending processing is then performed on each voltage window data segment. Specifically, the least-squares linear fitting line of all voltage sampling points within the segment relative to the time index is calculated, resulting in a sequence of linear fitting values. The original voltage sampling values ​​are subtracted one by one from the corresponding linear fitting values ​​to obtain the detrended voltage segment. This detrended voltage segment eliminates the overall rise and fall trend components in the voltage signal.

[0022] In some embodiments, local extremum detection processing is performed on the detrended voltage segment. The three-point comparison method is as follows: For the i-th data point in the detrended voltage segment, the amplitude of the data point is compared with the amplitudes of its adjacent left and right data points. If the amplitude of the data point is greater than the amplitudes of its left and right adjacent data points simultaneously, it is marked as a local maximum point; if it is less than the amplitudes of its left and right adjacent data points simultaneously, it is marked as a local minimum point. By traversing the entire detrended voltage segment, all local maximum and local minimum points can be marked. Optionally, within each voltage window data segment, adjacent combinations of local maximum and local minimum points are extracted sequentially according to time sequence. The amplitude difference between each group of adjacent maximum and minimum values ​​is calculated, and all calculated amplitude differences are collected into a fluctuation amplitude set. The largest fluctuation amplitude is found in the fluctuation amplitude set of a voltage window data segment, and this largest fluctuation amplitude is used as the representative fluctuation amplitude of the voltage window data segment. It can be understood that the formula for calculating the representative fluctuation amplitude is:

[0023] in: This represents the amplitude of the fluctuation in a voltage window data segment. This represents the voltage magnitude at the j-th local maximum point within the segment. This represents the voltage amplitude of the j-th local minimum point adjacent to the local maximum point. The absolute value is then taken, followed by the maximum value of the differences among all j-th corresponding points. The representative fluctuation amplitude is calculated sequentially for each voltage window data segment, and these representative fluctuation amplitudes are arranged according to the temporal order of the voltage window data segments to form a voltage fluctuation characteristic sequence.

[0024] In one embodiment of the present invention, see [reference] Figure 3When performing dynamic response feature extraction processing on current time series data to obtain the current response feature sequence, step change point detection processing is performed on the current time series data to identify all step start and step end times where the rate of change of the current value per unit time exceeds a preset rate of change threshold. Starting from each step start time and ending at the corresponding step end time, multiple step response data segments are extracted from the current time series data. Response time calculation processing is performed on each step response data segment to calculate the percentage change from the step start time to the current value reaching its steady-state value. The required time length is used as the rise time characteristic value of the step response data segment. At the same time, overshoot calculation is performed on each step response data segment. The ratio of the difference between the maximum peak value and the steady-state value of the current value in the step response data segment to the steady-state value is used as the overshoot characteristic value of the step response data segment. The rise time characteristic value and overshoot characteristic value of each step response data segment are combined to form the response characteristic vector of the step response data segment. The response characteristic vectors of all step response data segments are arranged in chronological order according to the step start time to generate the current response characteristic sequence.

[0025] When generating a fused electrical feature set by performing cross-modal feature alignment processing on the voltage fluctuation feature sequence and the current response feature sequence, the first acquisition timestamp corresponding to each voltage fluctuation feature value in the voltage fluctuation feature sequence and the second acquisition timestamp corresponding to each current response feature vector in the current response feature sequence are obtained. A unified time axis grid is constructed with the minimum timestamp between the first and second acquisition timestamps as the alignment starting point and the maximum timestamp as the alignment ending point. Then, each voltage fluctuation feature value in the voltage fluctuation feature sequence is mapped to the corresponding time grid point in the unified time axis grid according to its first acquisition timestamp to obtain the aligned voltage sequence. At the same time, each current response feature vector in the current response feature sequence is mapped to the corresponding time grid point in the unified time axis grid according to its second acquisition timestamp to obtain the aligned current sequence. For each time grid point in the unified time axis grid, the voltage fluctuation feature value in the aligned voltage sequence and the current response feature vector in the aligned current sequence are concatenated to generate the fused feature vector of that time grid point. The fused feature vectors of all time grid points are collected to form a fused electrical feature set.

[0026] In practical implementation, for the same energy storage unit within the power plant's energy storage system, the monitoring, control, and data acquisition system collects current time-series data in real time. This data records the current value at each sampling moment. When performing step change point detection processing on the current time-series data, the rate of change of the current value per unit time is obtained at a calculation frequency of once every 10 milliseconds. When the rate of change exceeds a preset threshold (e.g., a change of 50 amperes per millisecond), the moment when the rate of change exceeds the threshold is marked as the step start moment, and the moment when the rate of change falls back below the threshold is marked as the step end moment, thus identifying all step start and step end moments.

[0027] In some embodiments, starting from the start time of each step jump and ending at the corresponding end time of the step jump, current value segments are extracted from the original current time series data. Each current value segment is called a step response data segment. For each extracted step response data segment, during response time calculation, the current values ​​are iterated point by point from the step start time to find the moment when the current value first reaches 90% of the steady-state value (the final value to which the current value converges after the step jump). The time length between this moment and the step start time is calculated and used as the rise time characteristic value. Optionally, during overshoot calculation for each step response data segment, the current values ​​within the entire step response data segment are scanned to find the maximum peak current. Then, the steady-state current value at the end of the step response data segment is obtained, and the overshoot characteristic value is calculated according to the formula:

[0028] in: This represents the overshoot characteristic value of the step response data segment. This represents the maximum peak current value in the step response data segment. The steady-state current value represents the step response data segment, and the calculated overshoot characteristic value is a dimensionless ratio. The rise time characteristic values ​​and overshoot characteristic values ​​of the same step response data segment are sequentially combined into a two-dimensional vector, which is then used as the response characteristic vector for that step response data segment. The response characteristic vectors corresponding to each step response data segment are arranged sequentially along the time axis according to the chronological order of the step start times of all step response data segments, forming a current response characteristic sequence.

[0029] In practical implementation, when performing cross-modal feature alignment processing, the first acquisition timestamp corresponding to each voltage fluctuation feature value in the voltage fluctuation feature sequence and the second acquisition timestamp corresponding to each current response feature vector in the current response feature sequence are obtained respectively. The timestamp with the smallest value among all first and second acquisition timestamps is used as the alignment start point, and the timestamp with the largest value is used as the alignment end point. The time interval between the start and end points is divided into multiple time grids of equal length at fixed time intervals (e.g., 1 second). Each time grid corresponds to a center point of the time interval, and all time grids form a unified time axis grid. For the voltage fluctuation feature sequence, each voltage fluctuation feature value is traversed, and it is mapped to the time grid closest to that timestamp in the unified time axis grid according to the first acquisition timestamp. If multiple voltage fluctuation feature values ​​are mapped to the same time grid, the average value is taken to obtain the aligned voltage sequence. In some embodiments, for a current response feature sequence, each current response feature vector is traversed and mapped to the time grid point closest to the second acquisition timestamp in a unified time axis grid. If multiple current response feature vectors are mapped to the same time grid point, the average value of each component is taken to obtain an aligned current sequence.

[0030] Optionally, for each time grid point in the unified time axis grid, the voltage fluctuation characteristic value corresponding to that time grid point in the aligned voltage sequence is read, and simultaneously, the current response characteristic vector (containing two components: rise time characteristic value and overshoot characteristic value) corresponding to that time grid point in the aligned current sequence is read. The voltage fluctuation characteristic value is then concatenated as the first component before the two components of the current response characteristic vector to form a three-dimensional fused feature vector. It can be understood that the above concatenation operation is performed for each time grid point in the unified time axis grid, collecting the three-dimensional fused feature vectors generated by all time grid points, and arranging them according to the order of the time grid points to form a fused electrical feature set.

[0031] In one embodiment of the present invention, when the improved Kalman filter algorithm is invoked to perform state estimation processing on the fused electrical feature set to generate a real-time state of charge (SOC) estimation sequence of the energy storage unit, the fused feature vector of each time grid point is extracted from the fused electrical feature set. The voltage fluctuation feature value in the fused feature vector is used as the observation input of the improved Kalman filter algorithm, and the current response feature vector in the fused feature vector is used as the control input of the improved Kalman filter algorithm. The SOC estimation value of the previous time grid point is used as the prior state estimation value. The control input is fed to the state transition matrix of the improved Kalman filter algorithm to generate the predicted state value of the current time grid point. The observation input is fed to the observation matrix of the improved Kalman filter algorithm to generate the predicted observation value of the current time grid point. The temperature value corresponding to the current time grid point in the temperature monitoring data is obtained. The temperature compensation coefficient in the filter gain matrix of the improved Kalman filter algorithm is dynamically adjusted according to the temperature value. The adjusted filter gain matrix is ​​then used to perform weighted correction on the residual between the predicted state value and the predicted observation value to generate the SOC estimation value of the current time grid point. The SOC estimation values ​​of all time grid points are arranged in chronological order to generate a real-time SOC estimation sequence.

[0032] In practical implementation, for each time grid point in the fused electrical feature set, the corresponding fused feature vector is extracted from the fused electrical feature set. The fused feature vector contains voltage fluctuation feature values ​​and two components of the current response feature vector (rise time feature value and overshoot feature value). The voltage fluctuation feature value is used as the observation input of the improved Kalman filter algorithm, and the current response feature vector (two components) is used as the control input of the improved Kalman filter algorithm. For the first time grid point, the prior state estimate is taken as a preset initial value of the state of charge (e.g., 50%). For each subsequent current time grid point, the state of charge estimate output from the previous time grid point is used as the prior state estimate. This prior state estimate and the control input are respectively input into the state transition matrix of the improved Kalman filter algorithm. The state transition matrix contains the charging and discharging efficiency coefficient and rated capacity parameter of the energy storage unit. The state transition matrix, the prior state estimate, and the control input are linearly combined to generate the predicted state value of the current time grid point.

[0033] In some embodiments, the observations are input to the observation matrix of an improved Kalman filter algorithm, which describes the linear mapping between the state of charge and voltage fluctuation characteristics. The observation matrix is ​​multiplied by the predicted state values ​​to generate the predicted observations at the current time grid.

[0034] In practical implementation, the temperature value corresponding to the timestamp of the current time grid point is retrieved from the temperature monitoring data. Based on this temperature value, the temperature compensation coefficient in the filter gain matrix of the improved Kalman filter algorithm is dynamically adjusted. The formula for calculating the filter gain matrix is ​​as follows:

[0035] in: Indicates at time step The filter gain matrix at time t. This represents the prior estimation error covariance matrix. Represents the observation matrix. Represents the transpose of the observation matrix. This represents the observation noise covariance matrix at the reference temperature. This represents the rate of change of the temperature compensation coefficient. This represents the temperature value corresponding to the current time grid point. Indicates a reference temperature value. Optional, when... Higher than At that time, the exponent term Greater than 1, making It is amplified, thus reducing the filter gain matrix. The amplitude; when Below When the exponent term is less than 1, it makes Shrinking, thus enlarging The amplitude. Using the adjusted filter gain matrix. A weighted correction is applied to the residual between the predicted state value and the predicted observation value (observation input minus predicted observation value). The weighted correction is calculated as follows: the estimated state of charge at the current time grid point equals the predicted state value plus... Multiply by the residual.

[0036] It is understandable that the weighted correction yields a state-of-charge (POC) estimate that integrates dynamic information from voltage fluctuation characteristics and current response characteristic vectors, while also being corrected for real-time temperature readings. In some embodiments, the aforementioned POC estimation process is performed sequentially on each time grid point, and the POC estimates calculated for each time grid point are arranged chronologically and concatenated into a numerical sequence, which is the real-time POC estimate sequence. Optionally, for time grid points where temperature monitoring data has not yet been collected, the temperature value of the most recent time grid point is used as a substitute value to adjust the filter gain matrix. It is understood that the entire POC estimation process is recursively performed time-grid by time, with each time grid point relying only on the estimation result of the previous time grid point and the observation input, control input, and temperature value of the current time grid point, without needing to store all historical data.

[0037] In one embodiment of the present invention, when performing frequency regulation response anomaly judgment processing on energy storage units based on the comparison result of the real-time state of charge (SCC) estimation value sequence and the preset SCC safety threshold to generate an abnormal unit marker sequence, the SCC estimation value corresponding to each time grid point in the real-time SCC estimation value sequence and the preset SCC safety threshold range are obtained. The SCC safety threshold range includes a lower SCC threshold and an upper SCC threshold. For each time grid point, it is determined whether the SCC estimation value at that time grid point is less than the lower SCC threshold or greater than the upper SCC threshold. When the SCC estimation value is less than the lower SCC threshold or greater than the upper SCC threshold, [the following is considered:] The time grid point is marked as an abnormal state of charge time point, and the corresponding energy storage unit identifier is recorded. Then, for each energy storage unit, the number of abnormal state of charge time points occurring within the continuous time window is counted, and it is determined whether the number exceeds the preset abnormal count threshold. When the number exceeds the abnormal count threshold, the energy storage unit is marked as an abnormal unit, and the unit identifier and abnormal occurrence time period of the abnormal unit are recorded. The unit identifiers and abnormal occurrence time periods of all abnormal units are arranged in the order of detection time to generate an abnormal unit marking sequence. The length of the continuous time window and the value of the abnormal count threshold are dynamically configured based on the historical fault statistics data of the energy storage system.

[0038] In specific implementation, the state of charge (SOC) estimate for each time grid point is sequentially read from the real-time SOC estimate sequence. Simultaneously, a pre-configured SOC safety threshold range is obtained, which includes a lower SOC threshold (e.g., 20%) and an upper SOC threshold (e.g., 90%). A comparison operation is performed for each time grid point to determine whether the SOC estimate is less than the lower SOC threshold or greater than the upper SOC threshold. When the SOC estimate is less than the lower SOC threshold or greater than the upper SOC threshold, the time grid point is marked as an SOC anomaly time point, and the corresponding energy storage unit identifier (e.g., unit ID) is recorded. In some embodiments, for each energy storage unit, a sliding window with a continuous time window length (e.g., 300 seconds) is set. This sliding window moves along the time axis in fixed steps (e.g., 10 seconds). At each sliding window position, the total number of SOC anomaly time points occurring within the window for that energy storage unit is counted. Optionally, the expression for counting the number of anomalies is:

[0039] in: This represents the number of abnormal state of charge time points counted within a continuous time window. Indicates the start time of the continuous time window. Indicates the length of the continuous time window. This indicates an indicator function (it takes the value 1 when the condition inside the parentheses is true, and 0 when it is false). Indicates time Estimated state of charge, This represents the lower limit threshold for the charged state. This indicates the upper limit threshold for the state of charge.

[0040] The statistics obtained Compared with a pre-configured exception counting threshold, when When the anomaly count threshold is exceeded, the energy storage unit is marked as an abnormal unit, and the unit's identifier and the time period of the anomaly are recorded (the start time of the anomaly occurrence period is the left end point of the continuous time window, and the end time is the right end point of the continuous time window). It can be understood that the length of the continuous time window and the value of the anomaly count threshold are not fixed, but dynamically configured based on the historical fault statistics of the energy storage system. Specifically, the configuration method is as follows: data on units that have experienced frequency regulation response anomalies within the past year are extracted from the historical fault statistics database; the average frequency of occurrence of abnormal state of charge time points within the continuous time window before the anomaly event is calculated; the average frequency is multiplied by a safety factor to obtain the anomaly count threshold value; and the length of the continuous time window is set to the average duration of the historical anomaly event from start to end.

[0041] In some embodiments, the above-described sliding window statistical and comparison operations are performed on all energy storage units, and each marked abnormal unit records its unit identifier and the corresponding time period of the abnormality. Optionally, when marking abnormal units, if the same energy storage unit meets the requirements in multiple consecutive time windows... If the anomaly count threshold is exceeded, only the start time of the earliest anomaly occurrence time period and the end time of the latest anomaly occurrence time period are recorded and merged into a continuous anomaly occurrence time period. In essence, all detected abnormal units and their corresponding anomaly occurrence time periods are arranged in chronological order according to the detection time (i.e., the start time of the continuous time window), forming a sequence structure, which is the abnormal unit marking sequence.

[0042] In one embodiment of the present invention, when generating a dynamic correction command for frequency regulation response parameters based on the abnormal unit marking sequence and feeding the dynamic correction command back to the energy storage system controller to trigger the online calibration operation of the frequency regulation parameters, the abnormal unit marking sequence is parsed to extract the unit identifier and abnormal occurrence time period of each abnormal unit. For each abnormal unit, historical frequency regulation response parameter records of the abnormal unit within the abnormal occurrence time period are retrieved from the monitoring and control and data acquisition system. These historical frequency regulation response parameter records include the charging and discharging power limit value, frequency regulation response dead zone threshold, and ramp rate limit value of the abnormal unit. Then, the type of target parameter to be adjusted is determined according to the abnormal type of the abnormal unit. When the abnormal type is over-discharge, the charging and discharging power limit value is used as the target. In the case of an overcharge anomaly, the ramp rate limit is used as the target parameter. A step-by-step adjustment process is performed on the target parameter, increasing or decreasing the value of the target parameter step by step according to a preset adjustment step size to generate multiple candidate parameter adjustment values. Each candidate parameter adjustment value is then input into an improved Kalman filter algorithm for simulation state estimation. The predicted state of charge trajectory corresponding to each candidate parameter adjustment value is calculated. Candidate parameter adjustment values ​​that make the predicted state of charge trajectory completely within the safe state of charge threshold range are selected. The candidate parameter adjustment value with the smallest adjustment amplitude is used as the final parameter correction value. The final parameter correction value is encapsulated into a frequency modulation response parameter dynamic correction command and sent to the energy storage system controller.

[0043] The charging and discharging power limitation term in the state transition matrix applies a hard constraint to the boundary of the predicted state value of the improved Kalman filter algorithm based on the final parameter correction value. When inputting each candidate parameter adjustment value into the improved Kalman filter algorithm for simulation state estimation, the historical fused electrical feature set of the abnormal unit during its normal operating period before the abnormal occurrence time is obtained as simulation reference data. Voltage fluctuation feature values ​​from the simulation reference data are used as simulation observation inputs, and current response feature vectors from the simulation reference data are used as simulation control inputs. The charging and discharging power limitation term in the state transition matrix of the improved Kalman filter algorithm is then modified using the candidate parameter adjustment value as the constraint boundary. The state of charge estimate of the abnormal unit at the beginning of the abnormal occurrence time is used as the initial simulation state. The modified improved Kalman filter algorithm is run by sequentially inputting the simulation observation input and simulation control input, recording the simulated state of charge estimate output by the modified improved Kalman filter algorithm at each simulation time step, and connecting the simulated state of charge estimates output by all simulation time steps in chronological order to generate the predicted state of charge trajectory corresponding to the candidate parameter adjustment value.

[0044] In practice, when parsing the abnormal unit marker sequence, each entry in the sequence is read sequentially. Each entry contains the unit identifier of the abnormal unit and the corresponding time period of the abnormality. For each extracted abnormal unit, the historical frequency regulation response parameter records of the abnormal unit during the abnormality period are retrieved from the monitoring and control and data acquisition system. The historical frequency regulation response parameter records include the charging and discharging power limit value (in kilowatts), the frequency regulation response dead zone threshold (in Hertz), and the ramp rate limit value (in kilowatts per second). The type of target parameter to be adjusted is determined according to the abnormality type of the abnormal unit: when the abnormality type is the over-discharge abnormality type (i.e., the state of charge estimate is continuously lower than the state of charge upper limit threshold), the charging and discharging power limit value is used as the target parameter; when the abnormality type is the over-charge abnormality type (i.e., the state of charge estimate is continuously higher than the state of charge upper limit threshold), the ramp rate limit value is used as the target parameter.

[0045] In some embodiments, when performing a step-by-step adjustment on the target parameter, the current value of the target parameter is used as a reference, and the value of the target parameter is increased or decreased sequentially according to a preset adjustment step size (for example, the adjustment step size for the charge / discharge power limit is set to 5% of the current value, and the adjustment step size for the ramp rate limit is set to 10% of the current value). Each adjustment generates a candidate parameter adjustment value, until a preset number (e.g., ten) of candidate parameter adjustment values ​​are generated. Optionally, when the target parameter is a charge / discharge power limit, a sequential decreasing operation direction is used, and when the target parameter is a ramp rate limit, a sequential increasing operation direction is used.

[0046] In practice, each candidate parameter adjustment value is input into the improved Kalman filter algorithm for simulation state estimation, and the predicted state of charge trajectory corresponding to each candidate parameter adjustment value is calculated. During simulation state estimation, a set of historical integrated electrical characteristics of the abnormal unit during its normal operating period prior to the abnormality occurrence is obtained. This set of historical integrated electrical characteristics is used as the simulation baseline data, and the duration of the normal operating period is consistent with the duration of the abnormality occurrence period.

[0047] Voltage fluctuation characteristic values ​​from the simulation baseline data are used as simulation observation inputs, and current response characteristic vectors from the simulation baseline data are used as simulation control inputs. Candidate parameter adjustment values ​​are used as constraint boundaries, and the charging / discharging power limit term in the state transition matrix of the improved Kalman filter algorithm is modified. Specifically, the original charging / discharging power limit value in the state transition matrix is ​​replaced with the candidate parameter adjustment value, so that the boundary of the predicted state value is hard-constrained by the candidate parameter adjustment value during the state transition calculation. The estimated state of charge (SOC) of the abnormal unit at the beginning of the abnormal occurrence period is used as the initial state of the simulation. Simulation observation inputs and simulation control inputs are input sequentially according to time. The modified Kalman filter algorithm is run, and the simulated SOC estimated value output by the modified Kalman filter algorithm is recorded at each simulation time step. It can be understood that for a candidate parameter adjustment value, the simulated SOC estimated values ​​output from all simulation time steps are connected sequentially to form a continuous sequence, which is the predicted SOC trajectory corresponding to that candidate parameter adjustment value.

[0048] In some embodiments, candidate parameter adjustment values ​​are selected that ensure the predicted state of charge trajectory lies entirely within the safe state of charge threshold range, which includes a lower limit threshold and an upper limit threshold. For each candidate parameter adjustment value, each simulated state of charge estimate on the corresponding predicted state of charge trajectory is checked to determine if all are greater than the lower limit threshold and less than the upper limit threshold. If any simulated state of charge estimate exceeds this range, the candidate parameter adjustment value is excluded. From all candidate parameter adjustment values ​​that meet the conditions, the candidate parameter adjustment value with the smallest adjustment magnitude is selected as the final parameter correction value. The formula for calculating the adjustment magnitude is:

[0049] in: Indicates the adjustment range. This represents the numerical value of the candidate parameter adjustment. This represents the original value of the target parameter. Optionally, when no candidate parameter adjustment value can ensure that the predicted state of charge trajectory is completely within the safe threshold range of the state of charge, the candidate parameter adjustment value that minimizes the time the predicted state of charge trajectory exceeds the range is selected as the final parameter correction value. It can be understood that when the final parameter correction value is encapsulated as a dynamic correction instruction for the frequency regulation response parameters, the instruction format includes a unit identification field, a target parameter type field, and a final parameter correction value field. This dynamic correction instruction is sent to the energy storage system controller via industrial Ethernet. The charge / discharge power limitation term in the state transition matrix applies a hard constraint to the boundary of the predicted state value of the improved Kalman filter algorithm based on the final parameter correction value. Specifically, when the improved Kalman filter algorithm performs state transition calculations, if the predicted state value calculated according to the state transition matrix exceeds the boundary set by the charge / discharge power limitation term, the predicted state value is truncated to the boundary value.

[0050] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for intelligent monitoring of frequency regulation response of an energy storage system in a power plant, characterized in that, The method includes: Acquire the set of raw operating data of the energy storage system in the power station collected in real time by the monitoring, control and data acquisition system. The set of raw operating data includes voltage time-series data, current time-series data and temperature monitoring data of each energy storage unit. The voltage time series data is subjected to fluctuation feature parsing processing to obtain a voltage fluctuation feature sequence. The current time series data is subjected to dynamic response feature extraction processing to obtain a current response feature sequence. The voltage fluctuation feature sequence and the current response feature sequence are subjected to cross-modal feature alignment processing to generate a fused electrical feature set. An improved Kalman filter algorithm is invoked to perform state estimation processing on the fused electrical feature set to generate a real-time state of charge estimation sequence of the energy storage unit, wherein the improved Kalman filter algorithm dynamically adjusts the filter gain matrix based on the temperature monitoring data; Based on the comparison results between the real-time state of charge estimation value sequence and the preset state of charge safety threshold, the energy storage unit is subjected to frequency regulation response anomaly judgment processing to generate an abnormal unit marking sequence. Based on the abnormal unit marking sequence, a dynamic correction command for frequency regulation response parameters is generated, and the dynamic correction command is fed back to the energy storage system controller to trigger the online calibration operation of the frequency regulation parameters; The step of performing cross-modal feature alignment processing on the voltage fluctuation feature sequence and the current response feature sequence to generate a fused electrical feature set specifically includes: Obtain the first acquisition timestamp corresponding to each voltage fluctuation feature value in the voltage fluctuation feature sequence, and the second acquisition timestamp corresponding to each current response feature vector in the current response feature sequence; A unified timeline grid is constructed using the minimum timestamp between the first and second collection timestamps as the alignment start point and the maximum timestamp as the alignment end point. Each voltage fluctuation feature value in the voltage fluctuation feature sequence is mapped to the corresponding time grid point in the unified time axis grid according to its first acquisition timestamp to obtain the aligned voltage sequence; Each current response feature vector in the current response feature sequence is mapped to the corresponding time grid point in the unified time axis grid according to its second acquisition timestamp to obtain the aligned current sequence; For each time grid point in the unified time axis grid, the voltage fluctuation feature value in the aligned voltage sequence and the current response feature vector in the aligned current sequence at that time grid point are concatenated to generate the fused feature vector of that time grid point. Collect the fused feature vectors from all time grid points to form the fused electrical feature set; The step of using the improved Kalman filter algorithm to perform state estimation processing on the fused electrical feature set to generate a real-time state-of-charge estimation sequence for the energy storage unit specifically includes: Extract the fused feature vector for each time grid point from the fused electrical feature set, use the voltage fluctuation feature value in the fused feature vector as the observation input of the improved Kalman filter algorithm, and use the current response feature vector in the fused feature vector as the control input of the improved Kalman filter algorithm; The state of charge estimate of the previous time grid point is used as the prior state estimate. The control input is fed into the state transition matrix of the improved Kalman filter algorithm to generate the predicted state value of the current time grid point. The observations are input into the observation matrix of the improved Kalman filter algorithm to generate the predicted observations for the current time grid point; Obtain the temperature value corresponding to the current time grid point from the temperature monitoring data, and dynamically adjust the temperature compensation coefficient in the filter gain matrix of the improved Kalman filter algorithm according to the temperature value. The residual between the predicted state value and the predicted observation value is weighted and corrected using the adjusted filter gain matrix to generate the estimated state of charge value of the current time grid point. The state of charge estimates of all time grid points are arranged in chronological order to generate the real-time state of charge estimate sequence.

2. The method for intelligent monitoring of frequency modulation response based on the energy storage system in power station according to claim 1, characterized in that, The steps of performing fluctuation feature analysis on the voltage time series data to obtain the voltage fluctuation feature sequence specifically include: The voltage time series data is subjected to sliding window segmentation to obtain multiple voltage window data segments of equal duration; For each voltage window data segment, perform detrending processing and subtract the linear fit value of that voltage window data segment to obtain the detrended voltage segment; Local extremum detection processing is performed on the detrended voltage segment to mark all local maxima and local minima in the detrended voltage segment; Calculate the amplitude difference between adjacent local maxima and local minima in each voltage window data segment to obtain the set of fluctuation amplitudes for that voltage window data segment; The maximum fluctuation amplitude in the set of fluctuation amplitudes is taken as the representative fluctuation amplitude of the voltage window data segment; The representative fluctuation amplitudes of all voltage window data segments are arranged in chronological order to generate the voltage fluctuation characteristic sequence.

3. The method for intelligent monitoring of frequency modulation response based on the energy storage system in power station according to claim 1, characterized in that, The steps of performing dynamic response feature extraction processing on the current time series data to obtain the current response feature sequence specifically include: Perform step change point detection processing on the current time series data to identify all step start times and step end times when the rate of change of the current value exceeds a preset rate of change threshold per unit time. Starting from the start time of each step and ending at the corresponding end time of the step, multiple step response data segments are extracted from the current timing data. For each step response data segment, a response time calculation is performed to calculate the time required for the current value to reach 90% of the steady-state value from the start of the step response data segment. This time is used as the rise time characteristic value of the step response data segment. For each step response data segment, overshoot calculation is performed. The ratio of the difference between the maximum peak value of the current value and the steady-state value in the step response data segment to the steady-state value is calculated and used as the overshoot characteristic value of the step response data segment. The rise time characteristic value and overshoot characteristic value of each step response data segment are combined to form the response characteristic vector of that step response data segment; The response feature vectors of all step response data segments are arranged in chronological order according to the start time of the step, thus generating the current response feature sequence.

4. The method for intelligent monitoring of frequency modulation response based on the energy storage system in power station according to claim 3, characterized in that, The steps for determining frequency regulation response anomalies in the energy storage unit and generating an anomaly unit marker sequence based on the comparison between the real-time state of charge estimation sequence and the preset state of charge safety threshold specifically include: Obtain the state of charge estimate value corresponding to each time grid point in the real-time state of charge estimate value sequence, and a preset state of charge safety threshold range, wherein the state of charge safety threshold range includes a lower limit threshold and an upper limit threshold. For each time grid point, determine whether the estimated state of charge at that time grid point is less than the lower limit threshold of the state of charge or greater than the upper limit threshold of the state of charge; When the estimated state of charge is less than the lower limit threshold or greater than the upper limit threshold, the time grid point is marked as an abnormal state of charge time point, and the energy storage unit identifier corresponding to the time grid point is recorded. For each energy storage unit, count the number of abnormal state of charge time points that occur in the energy storage unit within a continuous time window, and determine whether the number exceeds a preset abnormality counting threshold. When the number exceeds the abnormal count threshold, the energy storage unit is marked as an abnormal unit, and the unit identifier and the time period of the abnormality are recorded. Arrange the unit identifiers and time periods of all abnormal units in chronological order of detection time to generate the abnormal unit tag sequence.

5. The method for intelligent monitoring of frequency modulation response of a power plant based on an in-plant energy storage system as claimed in claim 4, wherein, The length of the continuous time window and the value of the anomaly counting threshold are dynamically configured based on the historical fault statistics of the energy storage system.

6. The method for intelligent monitoring of frequency modulation response based on the energy storage system in power station according to claim 5, characterized in that, The steps of generating a dynamic correction command for frequency regulation response parameters based on the abnormal unit marking sequence, and feeding back the dynamic correction command to the energy storage system controller to trigger the online calibration operation of the frequency regulation parameters, specifically include: Parse the abnormal unit marker sequence to extract the unit identifier and the time period of the abnormality for each abnormal unit; For each abnormal unit, the historical frequency regulation response parameter records of the abnormal unit during the abnormal period are retrieved from the monitoring, control and data acquisition system. The historical frequency regulation response parameter records include the charging and discharging power limit value, frequency regulation response dead zone threshold and ramp rate limit value of the abnormal unit. The type of target parameter to be adjusted is determined based on the type of abnormality of the abnormal unit. When the abnormality is over-discharge, the charging and discharging power limit value is used as the target parameter. When the abnormality is over-charge, the ramp rate limit value is used as the target parameter. The target parameter is adjusted in a stepwise manner, and the value of the target parameter is increased or decreased stepwise according to a preset adjustment step size to generate multiple candidate parameter adjustment values. Each candidate parameter adjustment value is input into the improved Kalman filter algorithm for simulation state estimation, and the predicted state of charge trajectory corresponding to each candidate parameter adjustment value is calculated. Candidate parameter adjustment values ​​that make the predicted state of charge trajectory completely within the safe threshold range of the state of charge are selected, and the candidate parameter adjustment value corresponding to the smallest adjustment magnitude is taken as the final parameter correction value. The final parameter correction value is encapsulated into a frequency modulation response parameter dynamic correction command, and the dynamic correction command is sent to the energy storage system controller.

7. The intelligent monitoring method for frequency regulation response based on an energy storage system within a power plant according to claim 6, characterized in that, The charge / discharge power constraint term in the state transition matrix applies a hard constraint to the boundary of the predicted state value of the improved Kalman filter algorithm based on the final parameter correction value.

8. The method for intelligent monitoring of frequency modulation response based on the energy storage system in power station according to claim 7, characterized in that, The step of inputting each candidate parameter adjustment value into the improved Kalman filter algorithm for simulation state estimation and calculating the predicted state-of-charge trajectory corresponding to each candidate parameter adjustment value specifically includes: Obtain the historical fusion electrical characteristic set of the abnormal unit during the normal operation period before the time of the abnormality, and use it as the simulation baseline data; The voltage fluctuation characteristic value in the simulation reference data is used as the simulation observation input, and the current response characteristic vector in the simulation reference data is used as the simulation control input. The adjusted candidate parameter values ​​are used as constraint boundaries to modify the charge / discharge power limit term in the state transition matrix of the improved Kalman filter algorithm. Using the estimated state of charge of the abnormal unit at the beginning of the abnormality period as the initial state of the simulation, the simulation observation input and the simulation control input are input sequentially, and the modified Kalman filter algorithm is run. Record the simulated state of charge estimate output by the modified and improved Kalman filter algorithm at each simulation time step; By concatenating the simulated state of charge estimates from all simulation time steps in chronological order, the predicted state of charge trajectory corresponding to the candidate parameter adjustment value is generated.

Images