A charging pile energy storage system response to wind and light fluctuation smoothing control method

Abstract

The present application relates to a kind of charging pile energy storage system to deal with the smoothing control method of wind and light fluctuation, to solve the problem of wind and light generation output prediction error and energy storage dynamic response matching.This method synchronously collects wind and light measured and predicted power, energy storage battery state of charge, voltage change rate and temperature gradient and other multidimensional physical quantities to generate standardized feature dataset.Use phase difference angle index to identify error evolution rhythm, drive the mode switching of three response strategies of conventional, inertia extension and inertia suppression.Inertia extension realizes buffer zone speed limit release hysteresis compensation, and under inertia suppression, weak harmonic current is injected by virtual impedance to absorb power deviation.Further based on state of charge change rate and electrochemical model, adaptive adjustment of buffer zone release time constant is realized, forming a closed-loop regulation of whole process.This scheme improves power smoothness and system safety, with error adaptive and autonomous control ability under local observation.

Description

Technical Field

[0001] This invention relates to the field of new energy storage system control technology, and in particular to a method for smoothing and controlling wind and solar power fluctuations in a charging pile energy storage system. Background Technology

[0002] Currently, charging pile energy storage systems generally adopt a PI control strategy based on predicted power and battery state, combined with filtering or feedforward compensation to regulate power fluctuations. A typical configuration relies on a cloud-based predictive model to generate a short-term power curve as a control benchmark. The BMS and converter collect battery SOC, voltage, and temperature data, adjusting the charging and discharging power according to the error amplitude. Some advanced methods introduce virtual inertia and virtual impedance to adaptively adjust the energy storage response and improve robustness.

[0003] However, existing technologies have three main shortcomings: First, they rely too heavily on prediction accuracy while neglecting the electrochemical inertial regulation potential of energy storage itself. When prediction errors are ahead or behind, they can easily lead to delayed or over-adjusted responses, causing safety hazards such as sudden changes in SOC, increased polarization, and excessive temperature rise. Second, PI regulation and fixed filter parameters are difficult to adapt to wide temperature ranges and low sampling rate conditions, lacking flexibility. Third, virtual inertial methods only address absolute errors and cannot handle the profound impact of phase relationship changes during error evolution on the energy storage response. Currently, there is no technology to construct a segmented lag compensation mechanism based on the phase difference angle of prediction errors. When predictions deviate significantly, relying solely on passive measures such as increasing the amplitude or shortening the response time can easily break through the safety envelope, affecting lifespan and grid safety.

[0004] To address the aforementioned shortcomings, this invention proposes an error-aware dynamic power redistribution method for energy storage. By analyzing the phase relationship between the predicted deviation and the measured power derivative, it divides the response into three intervals: synchronization, extension, and suppression. Conversely, it employs conventional adjustment, slow-release hysteresis compensation, and virtual impedance absorption strategies to adaptively match the error evolution. Furthermore, it introduces a SOC-phase coupling correction mechanism to anchor the electrochemical safety envelope. This technology transforms the response hysteresis from a passive constraint to an active control dimension, providing a highly robust power smoothing compensation path for wide-temperature-range energy storage stations. Summary of the Invention

[0005] This application provides a method for smoothing and controlling wind and solar fluctuations in a charging pile energy storage system, aiming to solve the problem of power compensation lag or over-adjustment caused by the mismatch between the evolution rhythm of wind and solar prediction errors and the timing of energy storage response.

[0006] This application provides a method for mitigating and controlling wind and solar power fluctuations in a charging pile energy storage system, specifically including:

[0007] Acquire measured and predicted power sequences of wind and solar power, and simultaneously collect current state of charge, terminal voltage change rate and temperature gradient data of energy storage batteries to generate a dual-channel state-aware dataset.

[0008] Based on the dual-channel state-aware dataset, the phase difference angle between the predicted deviation signal and the first derivative of the measured power within the sliding time window is calculated, and a phase difference angle index is generated.

[0009] The range of the phase difference angle index is determined. If the phase difference angle is between -30° and +30°, a conventional proportional-integral adjustment command is output. If it is greater than +30°, an inertial extension mode control flag is generated. If it is less than -30°, an inertial suppression mode control flag is generated, so as to generate a mode switching decision result containing three different response strategies.

[0010] When the inertial extension mode control flag is received, the original power command is split into a base command and an incremental command, and the incremental command is cached in a circular buffer. The release slope is set according to the terminal voltage drop rate to perform nonlinear asymptotic processing on the cached data to generate a hysteresis-compensated power command with electrochemical kinetic safety envelope constraints.

[0011] When the inertia suppression mode control flag is received, the current power regulation action is frozen and the virtual impedance simulation unit is activated. A weak harmonic current with the same frequency and opposite phase as the prediction deviation is injected into the grid connection point to generate equivalent power absorption by utilizing the inverter dead zone effect, so as to generate a reactive power support type error buffer current command.

[0012] Based on the execution process of the hysteresis compensation power command or reactive power support error buffer current command, the rate of change of state of charge within three cycles is continuously monitored and compared with the nominal slope threshold. If continuous deviation is detected, the buffer release time constant is dynamically adjusted to generate an adaptive correction factor that anchors the electrochemical characteristics of the battery.

[0013] The state transition conditions of the lightweight finite state machine are updated using the adaptive correction factor, and the closed-loop flow from the dual-channel state-aware dataset to the mode switching decision result is driven only by local observables to generate the final energy storage power allocation execution sequence with error adaptive correction capability.

[0014] The present application provides a method for mitigating and controlling wind and solar power fluctuations in a charging pile energy storage system, which has the following beneficial effects:

[0015] By constructing an error evolution feature extractor based on phase difference angle, this approach breaks through the traditional design paradigm that relies on the absolute value of the error or fixed filter parameters. It distinguishes between leading and lagging error modes by utilizing the phase relationship between the predicted deviation signal and the first derivative of the measured power. The system initiates corresponding compensation strategies before the error accumulates to a critical level, improving the system's response sensitivity and control pre-emptiveness in scenarios with severe fluctuations. Through a dual-mode collaborative mechanism of inertial extension mode and inertial suppression mode, the system combines the dynamic characteristics of energy storage electrochemicals with the inverter's programmable impedance capability, transforming response lag into a controllable and adjustable resource. In extension mode, the system employs instruction separation buffering and gradual release strategies to avoid polarization effects caused by sudden power increases. In suppression mode, the system injects in-phase harmonic currents with the same frequency through virtual impedance to achieve reactive power buffering using the dead-zone effect, improving power response smoothness and reducing battery cycle stress. By constructing a localized closed-loop response system through a state-of-charge-phase coupling correction mechanism and a finite state machine driven architecture, the system dynamically adjusts the release time constant in real time by monitoring the SOC change rate to anchor the electrochemical safety envelope. Mode switching is completed autonomously based solely on local observables, without the need for external communication or cloud collaboration. This solution transforms response lag from a passive constraint into an active control dimension, providing a highly robust localized error compensation path for low-cost, wide-temperature-range distributed energy storage systems. Attached Figure Description

[0016] Figure 1 This is the main flowchart of a method for smoothing and controlling wind and solar power fluctuations in a charging pile energy storage system.

[0017] Figure 2 This is a sub-flowchart of a method for mitigating wind and solar power fluctuations in a charging pile energy storage system.

[0018] Figure 3 This is another sub-flowchart of a method for smoothing and controlling wind and solar fluctuations in a charging pile energy storage system. Detailed Implementation

[0019] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0020] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0021] like Figure 1 As shown, this application provides a method for mitigating wind and solar power fluctuations in a charging pile energy storage system, specifically including:

[0022] S1: Acquire the measured power sequence and predicted power sequence of wind and solar power, and simultaneously collect the current state of charge, terminal voltage change rate and temperature gradient data of energy storage batteries to generate a dual-channel state-aware dataset containing time-series features of multi-dimensional physical quantities.

[0023] S2: Based on the dual-channel state-aware dataset, calculate the phase difference angle between the predicted deviation signal and the first derivative of the measured power within the sliding time window to generate a phase difference angle index that characterizes the rhythm of error evolution.

[0024] S3: Based on the numerical range of the phase difference angle index, the interval is determined. If the phase difference angle is between -30° and +30°, a conventional proportional-integral adjustment command is output. If it is greater than +30°, an inertial extension mode control flag is triggered. If it is less than -30°, an inertial suppression mode control flag is triggered to generate a mode switching decision result containing three different response strategies.

[0025] S4: When the inertial extension mode control flag is received, the original power command is split into a base command and an incremental command. The incremental command is cached in a circular buffer. The release slope is set according to the terminal voltage drop rate to perform nonlinear asymptotic processing on the cached data to generate a hysteresis-compensated power command with electrochemical kinetic safety envelope constraints.

[0026] S5: When the inertia suppression mode control flag is received, the current power regulation action is frozen and the virtual impedance simulation unit is activated. A weak harmonic current with the same frequency and opposite phase as the predicted deviation is injected into the grid connection point to generate equivalent power absorption by utilizing the inverter dead zone effect, so as to generate a reactive power support type error buffer current command.

[0027] S6: Based on the execution process of the hysteresis-compensated power command or the reactive power-supported error buffer current command, the state-of-charge change rate is continuously monitored within three cycles and compared with the nominal slope threshold. If a continuous deviation is detected, the buffer release time constant is dynamically adjusted to generate an adaptive correction factor that anchors the electrochemical characteristics of the battery.

[0028] S7: The state transition conditions of the lightweight finite state machine are updated using an adaptive correction factor. The closed-loop flow from the dual-channel state-aware dataset to the mode switching decision result is driven by local observables only, so as to generate the final energy storage power allocation execution sequence with error adaptive correction capability.

[0029] Step S1: Obtain the measured and predicted power sequences of wind and solar power, and simultaneously collect the current state of charge, terminal voltage change rate, and temperature gradient data of the energy storage battery to generate a dual-channel state-aware dataset containing multi-dimensional physical quantity time-series characteristics. Specifically, this includes:

[0030] S1.1: Acquire the raw sampling data from the wind and solar power generation side, and use the timestamp synchronization algorithm to perform millisecond-level time-series alignment processing on the measured power sequence and the predicted power sequence to generate a dual-channel raw power data stream with a unified time base.

[0031] Raw power sampling data streams are acquired from grid-connected acquisition terminals on the wind and solar power generation side. These streams contain both measured active power sequences and predicted power sequences transmitted from the cloud, each carrying independent high-precision timestamps. Time base alignment verification is performed on the measured and predicted power sequences to identify starting point deviations and sampling interval jitter on the time axis, establishing a unified global reference time coordinate system. A linear interpolation resampling algorithm is used to process non-equidistant sampling points, remapping the measured power sequences onto the standard time grid of the predicted power sequences, eliminating timing misalignments caused by communication delays or sensor response differences. A millisecond-level sliding synchronization window is constructed, calculating the cross-correlation coefficient between measured and predicted values ​​within the window. The time offset corresponding to the cross-correlation peak is used as a correction factor to perform microsecond-level phase shift compensation on the predicted sequence. Through these timing alignment and phase compensation processes, multi-source power data with time-domain deviations are transformed into a raw power dual-channel data stream with strict time synchronization, providing a highly consistent data foundation for the accurate extraction of subsequent error evolution characteristics.

[0032] For example, with a sampling frequency of 100Hz, the measured power sequence lags the predicted sequence by 50ms due to transmission delay. The system reads 100 sampling points from the previous second, calculates the cross-correlation function, and detects a peak offset of 5 sampling points. Using cubic spline interpolation, the measured sequence is shifted forward by 50ms, ensuring that the measured value at t=0.05s and the predicted value at t=0.05s are strictly coincident on the time axis. After processing, the time synchronization error of the dual-channel data is reduced to less than 1ms, significantly improving the accuracy of phase difference angle calculation.

[0033] S1.2: Based on the dual-channel raw data stream of power, the state of charge value, terminal voltage change rate gradient and temperature gradient distribution data of the energy storage battery are synchronously collected through the battery management system interface to generate a real-time monitoring set of multi-dimensional physical quantities containing electrochemical dynamic characteristics.

[0034] Receive the raw power dual-channel data stream with a unified time reference generated by S1.1 and use it as the time reference signal to trigger synchronous acquisition of battery status.

[0035] The system sends a high-frequency status query command to the battery management system via the controller area network bus interface, requesting the current state of charge (SOC) value, individual cell voltage sequence, and temperature sensor array data of the energy storage battery.

[0036] Differential operations are performed on the received discrete voltage sampling points to calculate the ratio of voltage change to time interval between adjacent sampling points, generating a terminal voltage change rate gradient index, which is used to characterize the dynamic intensity of electrochemical polarization reaction inside the battery.

[0037] Spatial interpolation algorithm is used to process temperature field data collected by distributed temperature sensors, construct a three-dimensional temperature distribution model inside the battery module, and extract the region with the highest temperature rise rate as temperature gradient distribution data to reflect the impact of heat accumulation on the battery's internal resistance.

[0038] The data on the state of charge, the rate of change of terminal voltage, and the temperature gradient are timestamped and aligned with the original power dual-channel data stream at the microsecond level to eliminate data misalignment caused by communication delay.

[0039] Through the above processing method, the results of the previous step are transformed into a real-time monitoring set of multi-dimensional physical quantities containing electrochemical dynamic characteristics, realizing the accurate spatiotemporal correlation between the physical state of the energy storage body and the power fluctuation of the grid, and providing a high-fidelity input basis for subsequent error evolution feature extraction.

[0040] For example, the battery management system sampling frequency is set to 100Hz, and the controller area network bus communication cycle is 10ms. At t=1.0s, the state of charge is read as 65.2%, and three consecutive voltage sampling points are collected, which are 3.20V, 3.18V, and 3.15V, respectively. The calculated terminal voltage change rate is -25V / s. The temperature sensor array detects that the temperature at the center of the module rises by 0.5℃ and at the edge by 0.2℃ within one second. The maximum temperature gradient is calculated to be 0.5℃ / s through interpolation. The above data is aligned and packaged with the measured wind and solar power of 100kW and the predicted value of 105kW at t=1.0s to form a complete state frame that includes power deviation and battery dynamic response. This significantly improves the accuracy of subsequent phase difference angle calculation and effectively avoids control lag caused by data asynchrony.

[0041] S1.3: Perform a sliding window truncation operation on the real-time monitoring set of multidimensional physical quantities, and use a linear interpolation compensation algorithm to fill in the data missing points caused by the difference in sampling frequency, so as to generate a power and battery status synchronization data frame that is perfectly matched in the time dimension.

[0042] The system receives a data stream containing a real-time monitoring set of multidimensional physical quantities, in which the sampling timestamps of wind and solar power data and battery status data exhibit non-uniform distribution characteristics.

[0043] Set a fixed-length sliding time window, and extract historical data segments of a preset duration based on the current control cycle to construct a local data processing interval.

[0044] Identify the timestamp intervals of each physical quantity sequence within a sliding time window to detect whether there are data gaps caused by communication delays or differences in sensor responses.

[0045] For the detected missing data location, locate the adjacent valid sampling points before and after it, and obtain the values ​​of the valid values ​​at the previous moment and the valid values ​​at the next moment, as well as the corresponding timestamps.

[0046] A linear interpolation compensation algorithm is applied to calculate the slope of the rate of change per unit time based on the time difference and numerical difference between adjacent valid points.

[0047] The calculated slope is multiplied by the time offset of the missing point relative to the previous point to obtain the numerical compensation increment.

[0048] The numerical compensation increment is accumulated to the previous valid value to generate the estimated physical quantity value at the missing moment, thus filling the data gap.

[0049] The above interpolation operation was performed on all sequences of measured power, predicted power, state of charge, rate of change of terminal voltage, and temperature gradient of wind and solar power to ensure that each sequence has continuous and corresponding numerical points on the same time axis.

[0050] By using linear interpolation compensation, the multidimensional physical quantity real-time monitoring set from the previous step is transformed into a power and battery status synchronization data frame that is perfectly matched in time dimension, achieving strict alignment of multi-source heterogeneous data in time sequence, and providing a data foundation without time delay deviation for subsequent phase difference angle calculation.

[0051] For example, the sliding time window length is set to 200ms, and the sampling reference frequency is 10Hz. When a missing data point is detected in the battery terminal voltage change rate sequence at t=1.05s, and the adjacent valid points are 0.5V / s at t=1.0s and 0.7V / s at t=1.1s, the slope is calculated as (0.7-0.5) / (1.1-1.0)=2.0V / s². The time offset of the missing point t=1.05s relative to t=1.0s is 0.05s. The compensation increment is calculated as 2.0*0.05=0.1V / s. The estimated value for t=1.05s is 0.5+0.1=0.6V / s. The same logic is performed on all missing points to generate a complete synchronous data frame, significantly improving the accuracy of subsequent phase calculations.

[0052] S1.4: Based on the power and battery state synchronization data frame, a normalized mapping function is used to transform the measured power sequence, predicted power sequence, state of charge value, terminal voltage change rate gradient and temperature gradient distribution data of different dimensions to the standard feature space, so as to generate dimensionless multidimensional physical quantity normalized feature vectors.

[0053] Receive the power and battery status synchronization data frame generated by S1.3, and extract the original time series data columns of five dimensions: measured power, predicted power, state of charge, terminal voltage change rate, and temperature gradient.

[0054] A Min-Max normalized mapping model is constructed. For the measured power and predicted power sequences, the global maximum and minimum values ​​within the sliding window are selected as scaling boundaries, and a linear transformation is performed to eliminate the weight interference of dimensional differences on subsequent phase calculations.

[0055] The Z-Score normalization algorithm is used to process the state of charge (SOC) values. Based on the mean and standard deviation obtained from historical operating cycles, the SOC data is converted into a standard normal distribution with a mean of zero and a variance of one, so as to highlight its relative deviation in the safety envelope constraint.

[0056] The hyperbolic tangent activation function is applied to the gradient of the voltage change rate at the opposite end for nonlinear compression mapping, which restricts the drastically fluctuating voltage differential signal to the interval [-1,1], suppresses extreme value overflow caused by sampling noise, and preserves the trend characteristics of the voltage dynamic response.

[0057] The temperature gradient distribution data is preprocessed using exponential smoothing filtering, and recent temperature changes are given higher weighting coefficients to generate a smooth temperature feature vector that reflects the thermal inertia hysteresis effect of the battery, ensuring that the thermal state index is consistent with the time scale of the electrochemical response rate.

[0058] The five-dimensional feature vectors processed by different normalization strategies are concatenated point by point according to the timestamp index to form a multi-dimensional physical quantity standardized feature matrix with a unified structure.

[0059] By using a unified mapping process in a multidimensional feature space, the heterogeneous synchronous data frames from the previous step are transformed into dimensionless multidimensional physical quantity standardized feature vectors, thereby achieving comparability and fusion of multi-source heterogeneous data in a unified feature space and providing numerical stability assurance for subsequent phase difference angle calculations.

[0060] For example, the sliding window length is set to 50 sampling points, the measured power range is 0-100kW, and the predicted power range is 0-100kW. For the measured power at time t, which is 85kW, the maximum window value is 95kW, and the minimum is 10kW. The normalized result is (85-10) / (95-10)=0.882. The mean SOC is 60%, the standard deviation is 5%, and the current SOC is 62%, with a normalized result of (62-60) / 5=0.4. The terminal voltage change rate is 0.5V / s, which becomes 0.462 after mapping with the tanh function. The temperature gradient, after exponential smoothing, is 0.2℃ / s. The final output contains a five-dimensional normalized feature vector [0.882, 0.850, 0.4, 0.462, 0.2], eliminating the dimensional conflict between kilowatt-level power and millivolt-level voltage, significantly improving the convergence speed of subsequent cross-correlation analysis.

[0061] S1.5: The dimensionless multidimensional physical quantity standardized feature vector is structurally encapsulated, and the power dual-channel data and battery electrochemical state data are integrated into a single data object using a data packaging protocol to generate a dual-channel state-aware dataset containing the time-series features of multidimensional physical quantities.

[0062] Dual channels refer to a power channel and a battery status channel. The power channel carries time-series data of measured and predicted power sequences of wind and solar power, while the battery status channel carries time-series data of state of charge, rate of change of terminal voltage, and temperature gradient. The data from the two channels are strictly synchronized on the time axis to form a unified dataset, which supports the joint analysis of power dynamics and battery dynamics by subsequent phase difference angle calculation.

[0063] like Figure 2 As shown, step S2: Based on the dual-channel state-aware dataset, calculate the phase difference angle between the predicted bias signal and the first derivative of the measured power within the sliding time window to generate a phase difference angle index characterizing the rhythm of error evolution. Specifically, this includes:

[0064] S2.1: Perform differential operations on the measured power sequence of wind and solar power in the dual-channel state-aware dataset to obtain the first derivative sequence of the measured power, which characterizes the actual rate of fluctuation.

[0065] The measured power sequence of wind and solar power was extracted from a dual-channel state-aware dataset. This sequence already had a unified time reference and had undergone normalization. Discrete difference operations were performed on the measured power sequence, and the power increment between adjacent sampling points was calculated using a forward difference operator to characterize the instantaneous change trend of power on a microscopic time scale. A moving average filter was introduced to smooth the difference results, filtering out high-frequency glitches caused by sensor noise or communication jitter, ensuring the physical authenticity of the derivative signal. The differential step size was determined according to the sampling period, and the smoothed power increment was divided by the time interval to obtain the measured power change rate value with clear physical dimensions. A first-order derivative sequence vector was constructed, with the length of the vector consistent with the original power sequence. Missing points at the beginning and end were filled using the boundary value replication method to ensure the integrity of the time series data. Through the above differential operations and filtering, the discrete power amplitude sequence was transformed into a continuous rate of change sequence, accurately capturing the dynamic characteristics of wind and solar fluctuations, and providing key input data reflecting the actual fluctuation rhythm for subsequent phase difference angle calculations.

[0066] For example, with a sampling period of 100ms, the measured power sequence is P=[10.5,10.8,11.2,11.1,10.9]kW. Forward differencing yields ΔP=[0.3,0.4,-0.1,-0.2]kW. After averaging with a sliding window size of 3, the smoothed difference value is [0.3,0.23,0.03,-0.17]kW. Dividing by the time interval of 0.1s, the first derivative sequence of the measured power is obtained as dP / dt=[3.0,2.3,0.3,-1.7]kW / s. This derivative sequence clearly reflects the dynamic process of the power first rising rapidly, then slowly decreasing, and then accelerating downwards, effectively eliminating the small fluctuation noise in the original data and significantly improving the anti-interference capability of subsequent phase analysis.

[0067] S2.2: Perform point-by-point subtraction operation based on the measured power sequence and the predicted power sequence of wind and solar power to generate a prediction deviation signal sequence that characterizes the instantaneous prediction accuracy.

[0068] Receive time-aligned measured power sequences and predicted power sequences of wind and solar power as raw data input for instantaneous prediction accuracy calculation.

[0069] Perform point-by-point subtraction between the measured power value and the predicted power value at the same timestamp to construct the original signal stream characterizing the power deviation.

[0070] The prediction bias signal at time t is calculated using the following formula:

[0071]

[0072] in, for Prediction deviation signal at time, for Predicted power value at time 10:00 for The measured power value at any given time.

[0073] By traversing all sampling points within the sliding time window, a set of prediction bias signals in the form of a continuous time series is generated.

[0074] Outlier detection is performed on the generated deviation signal sequence, and outlier noise caused by communication packet loss or sensor failure is removed using the three-standard-deviation criterion.

[0075] The deviation signal after noise removal is zero-mean processed to eliminate the baseline drift caused by the DC component in subsequent phase analysis.

[0076] Through the above processing method, the power dual-channel data from the previous step is transformed into a prediction deviation signal sequence that characterizes the instantaneous prediction accuracy, thereby achieving the data preprocessing goal of extracting the rhythmic features of error evolution.

[0077] For example, the sampling frequency is set to 100Hz, and the sliding time window length is 2 seconds, i.e., 200 sampling points. At a certain moment t=1.5s, the measured power P meas (1.5) is 45.2kW, predicted power P pred (1.5) is 48.5kW. Substituting into the formula, we get e(1.5) = 48.5 - 45.2 = 3.3kW. The system continuously collects 200 points, forming a deviation vector E = [e1, e2, ..., e200]. The deviation of the 50th point is detected to be 150kW, exceeding the threshold of the mean plus 3 times the standard deviation, and is identified as an outlier and replaced with the mean of the preceding and following neighborhoods. The sequence mean of 0.2kW is subtracted from the remaining 199 points to obtain a zero-mean deviation sequence. This sequence accurately reflects the instantaneous deviation of the predicted value from the measured value, providing high signal-to-noise ratio input data for subsequent calculation of the phase difference between the predicted and measured power first derivatives, significantly improving the accuracy of phase difference angle calculation.

[0078] S2.3: The predicted deviation signal sequence and the measured power first derivative sequence are synchronously truncated and aligned using a sliding time window of preset length to generate local signal segment pairs containing temporal correspondence.

[0079] The measured power first derivative sequence generated by S2.1 and the prediction deviation signal sequence generated by S2.2 are received as the input data source for sliding window processing.

[0080] The length parameter of the sliding time window is set, which is determined by a fixed number of sampling points based on the typical periodic characteristics of wind and solar power fluctuations, ensuring that the window covers a complete error evolution segment. The window length of the sliding time window is determined based on the main periodic characteristics of wind and solar power fluctuations, and its value should cover at least one complete fluctuation cycle to capture the complete phase information of error evolution; if the window length is too short, the phase calculation will be significantly affected by noise interference, while if the window length is too long, the response sensitivity to changes in the error rhythm will be reduced.

[0081] A time-series index traversal is performed on the measured power first derivative sequence. Starting from the current time point, a data subset of a preset length is extracted by backtracking, forming the measured derivative local window data. Simultaneously, the same time-series index positioning operation is performed on the prediction deviation signal sequence to extract the data subset that is strictly aligned with the measured derivative local window on the time axis, forming the prediction deviation local window data.

[0082] The dimensionality consistency of the two local window data is checked. If data gaps exist due to boundary effects, zero-padding or mirror extension strategies are used to fill them in, ensuring equal data lengths. The measured derivative local window data and the prediction bias local window data are structurally paired to construct local signal segment pairs containing temporal correspondences. A sliding window mechanism is used to synchronously truncate and align the prediction bias signal sequence and the measured power first derivative sequence, transforming the discrete sequence from the previous step into local signal segment pairs with spatiotemporal correlation. This achieves localized focusing of error evolution characteristics, providing a high signal-to-noise ratio input foundation for subsequent accurate calculation of the phase difference angle.

[0083] For example, the sliding time window length is set to 50 sampling points, corresponding to a time span of 500 ms. At the current time t, a subset D of data with an index range of [t-49, t] is extracted from the measured power first derivative sequence. real A subset E of data with the same index range is extracted from the prediction bias signal sequence. pred If t < 49, for D real and E pred The missing part of the front end is completed using mirror extension. D real With E pred Packed into local signal fragments Pair t This processing method eliminates global noise interference and significantly improves the local sensitivity of phase calculation.

[0084] S2.4: Calculate the time offset between local signal segments using cross-correlation analysis algorithms or orthogonal decomposition methods to generate the original phase difference value that characterizes the error lead or lag.

[0085] Receive the local signal segment pair generated by S2.3, which includes the prediction bias signal sequence and the measured power first derivative sequence.

[0086] The local signal segments are preprocessed by removing the mean. The arithmetic mean of the two sequences within the sliding time window is calculated, and the corresponding mean is subtracted from the original data to eliminate the interference of the DC component on the phase calculation, thereby generating the zero-mean deviation signal vector and the derivative signal vector.

[0087] Based on the zero-mean signal vector, a Hilbert transform kernel is constructed using the discrete orthogonal decomposition method. The measured power first derivative sequence is then analyzed to reconstruct the signal, extract its instantaneous phase information, and generate an orthogonal component matrix containing amplitude and phase characteristics.

[0088] The cross-correlation function between the predicted bias signal vector and the measured power first derivative vector is calculated using a cross-correlation analysis algorithm. All lag points within the sliding time window are traversed to find the lag index corresponding to the peak value of the cross-correlation function. This index represents the maximum similarity offset between the two signals in the time domain.

[0089] By combining the sampling frequency and hysteresis index, the time-domain offset is converted into a phase difference value in the angular domain, and the original phase difference is calculated using the following formula:

[0090]

[0091] in, This is the original phase difference value. The lag index corresponding to the cross-correlation peak. The sampling time interval, This represents the fundamental period corresponding to the sliding time window.

[0092] If the cross-correlation peak is lower than the preset confidence threshold, the instantaneous phase difference obtained by orthogonal decomposition is used as a backup calculation result. The phase angle between the orthogonal components of the two signals is calculated by the arctangent function to enhance the computational robustness under low signal-to-noise ratio conditions.

[0093] By combining the above-mentioned cross-correlation peak search with orthogonal phase calculation, the local signal segment of the previous step is transformed into the original phase difference value that characterizes the error leading or lagging characteristics. This achieves high-precision quantification of the evolution rhythm of the prediction error and provides a reliable timing basis for subsequent mode switching.

[0094] For example, the sliding time window length is set to 0.5 seconds, the sampling frequency to 100Hz, and the fundamental period T to 0.02 seconds. The 50 collected data points are processed to remove the mean. The cross-correlation function is calculated, and the peak is found to occur at the hysteresis index k=3. Substituting this into the formula, the phase difference is calculated, resulting in 3π radians. After mapping to the principal value interval, the original phase difference value is obtained. If the normalized value of the cross-correlation peak is 0.85, which is higher than the confidence threshold of 0.7, this result is directly adopted. This processing method effectively eliminates the influence of signal amplitude fluctuations on phase judgment and significantly improves the accuracy of phase difference calculation.

[0095] S2.5: Perform angle mapping normalization on the original phase difference values ​​to generate the final phase difference angle index used to characterize the rhythm of error evolution.

[0096] Step S3: Based on the numerical range of the phase difference angle index, determine the interval. If the phase difference angle is between -30° and +30°, output a standard proportional-integral (PI) control command. If it is greater than +30°, trigger the generation of an inertial extension mode control flag. If it is less than -30°, trigger the generation of an inertial suppression mode control flag, thus generating a mode switching decision result containing three different response strategies. Specifically, this includes:

[0097] S3.1: Perform numerical range boundary comparison processing on the phase difference angle index to obtain the interval assignment state quantity that characterizes the rhythm of error evolution. The input condition is the phase difference angle index calculated within the sliding time window. The technical processing method used is to set a synchronization threshold range of -30 degrees to +30 degrees and a lead-lag threshold of +30 degrees. The clear goal is to output the interval assignment state quantity that indicates the rhythm of error and the dynamic synchronization of the system, and the prediction of severe lag or excessive lead.

[0098] The phase difference angle index generated in the previous step is received. This index characterizes the temporal offset between the predicted bias signal and the first derivative of the measured power within the sliding time window.

[0099] The synchronization threshold interval is set to -30 degrees and +30 degrees, and an interval judgment logic matrix is ​​constructed to distinguish three typical states of error evolution rhythm. The phase difference angle index is compared with the positive 30-degree critical value. If the index is greater than positive 30 degrees, it is determined that the prediction is seriously lagging behind the actual fluctuation, and an inertial extension mode trigger flag is generated.

[0100] The ±30° threshold setting comprehensively considers three factors: wind and solar power fluctuation characteristics, energy storage electrochemical response time constant, and measurement noise tolerance. Measured wind and solar power data are affected by sensor noise and communication jitter, resulting in an inherent error tolerance of ±5° to ±10° in phase difference angle calculations. A sufficiently wide boundary needs to be set to avoid frequent mode switching. Under typical wind and solar power fluctuation cycles, the timing offset corresponding to a 30° phase difference is on the same order of magnitude as the battery's electrochemical response time constant. Exceeding this boundary indicates that the timing offset of the prediction deviation has exceeded the effective tracking range of the battery's inherent response capability. Without a dedicated compensation strategy, this will evolve into significant power compensation lag or overshoot. Therefore, ±30° is a compromise between control sensitivity and system robustness, and in engineering practice, it can be adaptively adjusted within the range of ±20° to ±40° depending on the system configuration.

[0101] The phase difference angle index is compared with the -30 degree critical value. If the index is less than -30 degrees, it is determined that the prediction is over-leading the actual fluctuation, and an inertial suppression mode trigger flag is generated. If the phase difference angle index is within the closed interval of -30 degrees to +30 degrees, it is determined that the error rhythm is basically synchronized with the system dynamics, and a conventional proportional-integral regulation enable signal is generated.

[0102] Through the above boundary comparison processing, continuous phase difference angle values ​​are transformed into discrete interval belonging state quantities, realizing accurate classification of prediction error leading, lagging and synchronous states, and providing clear logical input for the execution of subsequent differentiated control strategies.

[0103] For example, suppose the phase difference angle calculated by the current sliding time window is 45 degrees. This value is compared with a preset threshold of +30 degrees. Since 45 degrees is greater than 30 degrees, the system determines it to be in a state of severe prediction lag and outputs an inertial extension mode control flag. If the phase difference angle is -45 degrees, since -45 degrees is less than -30 degrees, the system determines it to be in a state of over-prediction and outputs an inertial suppression mode control flag. If the phase difference angle is 15 degrees, since it is within the range of -30 degrees to +30 degrees, the system determines it to be in a state of synchronization and outputs a conventional proportional-integral (PI) regulation command. This process ensures that the energy storage system can automatically match the corresponding power response strategy under different error evolution rhythms, significantly improving the adaptability of smoothing control.

[0104] S3.2: Perform finite state machine transition logic mapping processing based on the interval home state quantity to generate a mode switching decision result containing a regular proportional-integral adjustment command, an inertial extension mode control flag, or an inertial suppression mode control flag. The input condition is the interval home state quantity output from the previous step. The technical processing method used is to map the numerical interval to a specific control mode identifier using a predefined lightweight finite state machine state transition table. The explicit goal is to output a mode switching decision result with three different response strategy attributes.

[0105] The system receives the interval assignment state quantity output from the previous step. This state quantity identifies the phase lead or lag attribute of the current prediction error relative to the measured fluctuation. A lightweight finite state machine model is constructed, with three predefined core operating states: normal synchronization adjustment state, inertial extension compensation state, and inertial suppression buffer state. The state transition table is initialized, mapping the synchronization interval in the interval assignment state quantity to the normal synchronization adjustment state, the severely lagging interval to the inertial extension compensation state, and the excessively leading interval to the inertial suppression buffer state. The state matching logic is executed, reading the current interval assignment state quantity as the index key and retrieving the corresponding target control mode identifier in the state transition table. If the state quantity indicates a synchronization interval, a normal proportional-integral adjustment command flag is generated, maintaining the original control loop parameters unchanged. If the state quantity indicates a severely lagging interval, an inertial extension mode control flag is generated, triggering the subsequent power command splitting and buffering mechanism. If the state quantity indicates an excessively leading interval, an inertial suppression mode control flag is generated, activating the virtual impedance simulation unit to inject reverse harmonic current. The generated mode switching decision results are encoded and encapsulated to ensure that the control flags have clear binary semantics for rapid parsing by downstream modules. Through deterministic mapping processing using a finite state machine, the abstract phase difference angle interval is transformed into a specific control mode identifier, enabling the energy storage system to adaptively switch response strategies to the rhythm of prediction errors, significantly improving the timeliness and robustness of power smoothing control.

[0106] For example, the sliding time window length is set to 500ms and the sampling frequency to 1kHz. When the calculated phase difference angle is +45°, the interval home state is determined to be severely lagging. The finite state machine searches the state transition table, matches the inertial extension compensation state, and outputs the inertial extension mode control flag 0x02. The energy storage controller receives this flag, immediately suspends the integral term update of the conventional PI regulator, starts the ring buffer write pointer, and prepares to cache incremental power commands. When the phase difference angle is -45°, the interval home state is determined to be excessively leading, the state machine outputs the inertial suppression mode control flag 0x03, triggers the virtual impedance module enable signal, and sends a harmonic injection request to the grid-connected inverter. When the phase difference angle is +10°, the state machine outputs the conventional adjustment flag 0x01, and the system maintains the standard power tracking mode. This mechanism ensures that the energy storage system can automatically select the optimal control strategy at different error evolution stages, avoiding response lag or overshoot problems caused by a single control law, and significantly improving the smoothing accuracy of power fluctuations at the grid connection point.

[0107] S3.3: Perform local observability consistency verification on the mode switching decision results to generate a final mode switching decision result with error adaptive correction capability. The input condition is the mode switching decision result generated in the previous step. The technical processing method used is to perform cross-domain logic interlock verification by combining the terminal voltage change rate and temperature gradient data in the dual-channel state-aware dataset. The clear goal is to output a final mode switching decision result that excludes false trigger interference and relies only on local observables.

[0108] The system receives the mode switching decision result output from S3.2 and extracts the inertial extension or suppression mode identifier as the object to be verified. It synchronously reads the current battery terminal voltage change rate and cell temperature gradient distribution data from the dual-channel state-aware dataset. A cross-domain logic interlock verification matrix is ​​constructed, setting the terminal voltage change rate as the electrochemical response sensitivity criterion and the temperature gradient as the thermal stability constraint boundary. The ratio of the absolute value of the terminal voltage change rate to a preset safety threshold is calculated to generate a voltage response confidence factor. If severe prediction lag is detected and the terminal voltage drop rate exceeds the critical slope, it is determined to be in the battery polarization effect dominance region. At this time, the upper limit of the release slope of the inertial extension mode is forcibly corrected to prevent overcharging and over-discharging risks. If excessive prediction lead is detected and the temperature gradient shows a nonlinear surge trend, it is determined to be in the thermal runaway precursor region. At this time, the harmonic injection action of the inertial suppression mode is blocked, and power freeze protection is executed instead. By introducing a multi-dimensional physical quantity coupling verification mechanism, the mode switching decision triggered by a single phase difference angle is transformed into a final control command corrected by dual constraints of electrochemical and thermodynamics, thereby achieving a robust control effect that eliminates sensor noise interference and false trigger signals.

[0109] For example, when the phase difference angle is +45° and the inertial extension mode is triggered, the system readout voltage change rate is -0.5V / s, exceeding the warning threshold of -0.3V / s. The interlock logic determines that the battery is in a high polarization state and automatically extends the release time constant of the annular buffer from 100ms to 250ms, limiting the power increment injection rate. If the temperature gradient is 2°C / min at this time, which does not exceed the thermal limit of 5°C / min, the extension mode is maintained but the gain is reduced. The final output is a smoothed power command after voltage slope correction, effectively avoiding the risk of lithium plating inside the battery and significantly improving the operational safety of the energy storage system under extreme fluctuations.

[0110] like Figure 3 As shown, step S4: When the inertial extension mode control flag is received, the original power command is split into a base command and an incremental command, and the incremental command is cached in a circular buffer. The release slope is set according to the terminal voltage drop rate to perform nonlinear asymptotic processing on the cached data to generate a hysteresis-compensated power command with electrochemical kinetic safety envelope constraints. Specifically, this includes:

[0111] S4.1: Based on the received inertial extension mode control flag, perform decomposition operation on the original power command, separating the original power command into base value command and incremental command to generate an incremental command data stream to be processed.

[0112] Receive the inertial extension mode control flag and the original power command, and start the command decomposition logic.

[0113] The original power command refers to the initial power reference value calculated by the conventional proportional-integral regulator based on the prediction deviation signal without hysteresis compensation processing. In other words, it is the power output target value that the energy storage converter should have executed before entering the inertial extension mode.

[0114] Analyze the instantaneous amplitude and rate of change of the original power command, and extract the target power reference value at the current moment.

[0115] The target power reference value is low-pass filtered by a moving average filter to remove high-frequency fluctuation components and generate a base value instruction that represents the steady-state requirement.

[0116] Subtracting the base value command from the original power command yields the difference sequence representing the dynamic compensation requirement.

[0117] The dead zone threshold is checked on the difference sequence to remove minor disturbances below the noise threshold and generate a clean incremental instruction data stream to be processed.

[0118] Through the above decomposition operation, the original power command is separated into a base value command and an incremental command, realizing the dynamic and static decoupling of the power response and providing precise control input for subsequent hysteresis compensation based on electrochemical characteristics.

[0119] For example, the original power command is set to 100kW, and the moving average window length is 50ms. After low-pass filtering, the base command stabilizes at 98.5kW. The difference is calculated to obtain an initial increment of 1.5kW. The dead zone threshold is set to 0.1kW. Since 1.5kW is greater than the threshold, this increment is retained as the incremental command data stream to be processed. This process effectively isolates the steady-state load from the fluctuation compensation requirements, ensuring that the incremental command only contains the dynamic error component that needs to be processed by inertial extension, significantly improving the targeting and effectiveness of subsequent buffer control.

[0120] S4.2: Based on the generated incremental instruction data stream to be processed, the incremental instructions are temporarily stored using a circular buffer storage mechanism to generate a cached incremental instruction set containing time series characteristics.

[0121] The system receives the incremental instruction data stream generated in step S4.1, which contains a sequence of power compensation components that need to be delayed due to prediction lag. A circular buffer data structure based on the first-in, first-out (FIFO) principle is constructed, with the buffer depth matched to the sliding time window length to accommodate error evolution characteristic data for the entire cycle. The incremental instruction data stream is written to the write pointer positions of the circular buffer in timestamp order, achieving temporal temporary storage and queued management of power instructions. During the writing process, each incremental instruction unit is appended with a current terminal voltage change rate tag and a temperature gradient tag, forming a cached instruction tuple with electrochemical state markings. The data dwell time is calculated by the relative displacement of the read and write pointers to ensure that the cached instruction set strictly corresponds to the lag compensation time window required by the inertial extension mode. Integrity checks are performed on the instruction tuples in the buffer, eliminating null values ​​or abnormal jump points caused by communication jitter to maintain the temporal continuity of the cached incremental instruction set. By using a ring buffer storage mechanism, the instantaneous incremental instructions from the previous step are transformed into a cached incremental instruction set with time-series characteristics and associated electrochemical states, thereby achieving dynamic matching between the physical response rhythm of power instructions and the ion diffusion rate inside the battery.

[0122] For example, the circular buffer depth is set to 50 sampling points, corresponding to a 500ms time window at a sampling frequency of 100Hz. When a phase difference angle of +45 degrees is detected, the peak value of the incremental command decomposed by S4.1 is 20kW. The system decomposes this 20kW power increment into 50 400W differential instruction units and writes them sequentially into the buffer. Each unit is accompanied by the current terminal voltage change rate of -0.5V / s and a temperature of 35 degrees Celsius. The buffer adopts a circular overwrite strategy, returning to the 1st position after the write pointer reaches the 50th position, ensuring that the instruction history of the most recent 500ms is always retained. If data is missing at a certain moment, linear interpolation is used to fill in the missing data based on the adjacent valid units before and after, ensuring the continuity of the instruction stream. The final output cached incremental instruction set contains 50 power differential values ​​with status labels, providing an accurate time-aligned data basis for the slope mapping of S4.3, significantly reducing the risk of battery polarization caused by instruction mutations.

[0123] S4.3: Based on real-time acquired terminal voltage change rate data, a slope parameter is dynamically released through a slope mapping algorithm to generate a slope constraint factor characterizing the limit of the electrochemical response rate.

[0124] The cache incremental instruction set generated in step S4.2 and the real-time terminal voltage change rate data collected in step S1 are used as the input basis for calculating the dynamic release slope.

[0125] The absolute value of the real-time terminal voltage change rate data is extracted and the polarity is determined to identify whether the battery is currently in a state of enhanced charging polarization or reduced discharging polarization, and to determine the direction of electrochemical response constraint.

[0126] Based on the preset electrochemical kinetic safety envelope model, the nonlinear mapping function between the rate of change of terminal voltage and the maximum allowable rate of change of current is invoked to calculate the baseline slope limit value.

[0127] The dynamic release slope parameter is calculated using the following formula:

[0128]

[0129] in, To dynamically release the slope parameter, As the baseline slope limit value, Voltage sensitivity coefficient, The terminal voltage change rate.

[0130] The calculated dynamic release slope parameter is coupled with the read pointer rate of the annular buffer to generate a slope constraint factor characterizing the rate limitation of the electrochemical response.

[0131] By using a slope mapping algorithm, the terminal voltage change rate data from the previous step is transformed into a slope constraint factor technical indicator, thereby achieving the expected technical effect of adaptive matching between the energy storage power command release rate and the battery polarization state.

[0132] For example, a baseline slope limit value k is set. base The voltage sensitivity coefficient α is 0.8 s / V, and the measured voltage change rate dV / dt is -0.2 V / s. Substituting this into the formula, the dynamic release slope parameter k is calculated to be approximately 4.35 kW / s. This slope constraint factor is applied to the data release control of the ring buffer, enabling the power increment command to be smoothly injected into the grid at a rate of 4.35 kW / s. This effectively avoids the aggravation of electrode polarization caused by large current surges, significantly improving the electrochemical safety and response stability of the energy storage system during prediction error compensation.

[0133] S4.4: Based on the generated slope constraint factor, perform nonlinear progressive release processing on the cache incremental instruction set, and gradually transform the cache incremental instruction set into smooth power components according to the rate curve defined by the slope constraint factor to generate an electrochemically safe power compensation sequence.

[0134] Read the incremental instruction data stream temporarily stored in the circular buffer, and extract the power increment amplitude to be released at the current moment and the corresponding timestamp index.

[0135] The slope constraint factor generated in the previous step is invoked. This factor characterizes the upper limit of the electrochemical response rate corresponding to the rate of change of battery terminal voltage, and serves as the benchmark control parameter for nonlinear gradual release.

[0136] A smooth transition model based on the hyperbolic tangent function is constructed, and the time variable is mapped to the normalized gain interval to eliminate the impact of step command on the battery polarization state.

[0137] The smoothed power component of the current control cycle is calculated using the following formula:

[0138]

[0139] in, for The smooth power component output at all times, This represents the total incremental instruction magnitude in the cache. This is the moment when the inertial extension mode is activated. Based on the time constant, Based on the rate of change of terminal voltage The generated dynamic adjustment coefficient.

[0140] The battery terminal voltage change rate is monitored in real time. When the voltage drop rate exceeds the preset threshold, the denominator term in the formula is dynamically increased, thereby lengthening the rise time of power injection and suppressing the accumulation of concentration polarization effect.

[0141] The calculated smooth power component is subjected to amplitude limiting protection to ensure that its instantaneous rate of change does not exceed the maximum allowable charge / discharge slope set by the battery management system.

[0142] The processed smooth power components are encapsulated as a time series to generate an electrochemically safe power compensation sequence with an electrochemically kinetic safety envelope constraint.

[0143] It should be noted that the electrochemical kinetic safety envelope refers to the power command safety boundary jointly defined by the rate of change of battery terminal voltage, the rate of change of state of charge, and the temperature gradient. When the rate of change of power command exceeds this boundary, electrochemical safety events such as lithium plating, increased polarization, or thermal runaway may occur inside the battery. This application limits the release rate of the power increment command within this safety boundary by using a slope constraint factor mapped from the rate of change of terminal voltage and a nonlinear progressive release process.

[0144] By using hyperbolic tangent nonlinear mapping and terminal voltage change rate feedback mechanism, the cached step incremental instructions are transformed into a continuous and smooth power compensation sequence, realizing the flexible power response of the energy storage system under predictive lag scenarios, significantly reducing the internal polarization loss of the battery and improving the robustness of the smoothing control.

[0145] For example, setting the cache increment instruction P inc The power output is 50kW, and the base time constant τ is 2s. When the detected terminal voltage change rate dV / dt is -0.5V / s, the dynamic adjustment coefficient k is calculated to be 1.2. At time t-t0=1s, the tanh parameter is calculated as (1) / (21.2)=0.417, and tanh(0.417) is approximately 0.394. At this time, the output power P(1) is approximately 500.5*(1+0.394)=34.85kW. If the voltage drop rate increases to -1.0V / s, k increases to 1.5, and at the same time the tanh parameter decreases to 0.333, the output power decreases to 32.5kW, effectively slowing down the power injection rate and avoiding the risk of battery over-discharge.

[0146] S4.5: Based on the generated base value command and the electrochemically safe power compensation sequence, perform vector superposition operation to merge the base value command and the electrochemically safe power compensation sequence into a single control signal to generate a hysteresis compensation power command with electrochemical kinetic safety envelope constraints.

[0147] Step S5: When the inertia suppression mode control flag is received, the current power regulation action is frozen and the virtual impedance simulation unit is activated. A weak harmonic current with the same frequency and opposite phase as the predicted deviation is injected into the grid connection point to generate equivalent power absorption using the inverter dead-time effect, thereby generating a reactive power support type error buffer current command. Specifically, this includes:

[0148] S5.1: Perform state analysis on the received inertial suppression mode control flag, generate a power command freeze signal based on the inertial suppression mode control flag, lock the current active power reference value of the energy storage converter and suspend the output update of the conventional proportional-integral regulator.

[0149] The system receives the inertial suppression mode control flag and parses its status bit to confirm that the prediction deviation is in the lead range. It reads the current cycle's active power reference value from the energy storage converter and the accumulated value of the internal integral term of the conventional proportional-integral (PI) regulator. The current active power reference value is locked as a constant, blocking subsequent power command update channels from modifying the reference value. The integral term update logic of the conventional PI regulator is paused, and the integral term value is frozen to prevent control overshoot caused by error accumulation. A power command freeze signal containing the lock flag and the frozen value is generated as the enable trigger condition for the virtual impedance simulation unit. Through the above processing, the mode switching decision of the previous step is transformed into a specific controller state lock command, achieving instantaneous stability and silence of active power output in the prediction lead scenario, avoiding blind response that exacerbates grid fluctuations. For example, when the phase difference angle is -45 degrees, the system determines that the prediction is excessively leadful. The current active power reference value is read as 50kW, and the PI regulator integral term is 12.5. The system locks 50kW as the baseline and pauses the integral term update. The generated freeze signal is set to 1 and lasts for 20ms until the phase difference angle is reassessed in the next control cycle. This process effectively prevents invalid charging commands caused by falsely high predictions, significantly improving the smoothness of the grid connection point power.

[0150] S5.2: The enable logic of the virtual impedance simulation unit is triggered by the power command freeze signal. The prediction deviation signal generated in the previous step is used as the reference input. The virtual harmonic voltage reference quantity with the same frequency and opposite phase is calculated by the phase reversal algorithm.

[0151] It should be noted that the virtual impedance simulation unit refers to the virtual resistor-inductor-capacitor network model constructed by software algorithm in the energy storage converter control system. This model takes the frequency component of the predicted deviation signal as input and outputs a virtual harmonic voltage reference quantity that is in phase with the input at the same frequency. Thus, a controllable reactive power absorption impedance characteristic is equivalent to that at the grid connection point without the need to add additional physical impedance components.

[0152] Upon receiving the power command freeze signal, confirming that the active power reference value of the energy storage converter has been locked, activate the enable logic port of the virtual impedance simulation unit.

[0153] The predicted deviation signal sequence generated in the previous steps is used as the reference time-domain input data for virtual harmonic voltage calculation.

[0154] Perform a Fast Fourier Transform on the prediction deviation signal sequence to extract the dominant fluctuation frequency component and its corresponding phase angle information.

[0155] Based on the extracted phase angle information, a phase reversal algorithm model is constructed to increase the original phase angle by 180 degrees to generate an anti-phase reference phase.

[0156] By using the inverted reference phase and the dominant oscillation frequency, a virtual harmonic voltage reference quantity with the same frequency but opposite phase is synthesized.

[0157] Through the aforementioned phase reversal processing, the power fluctuation component caused by the prediction lead is transformed into an inverse voltage command that can be absorbed by the inverter dead-zone effect, thereby realizing the preconditions for reactive power-supported error buffering.

[0158] For example, when a dominant 50Hz component with a 30-degree phase lead is detected in the prediction deviation signal, the system extracts this frequency component and calculates an out-of-phase phase of 210 degrees. The amplitude coefficient A is set to 2% of the grid's rated voltage to generate a virtual harmonic voltage reference. This reference is then sent to the limiting module to ensure that the injected current does not exceed 5% of the converter's rated capacity, effectively utilizing dead-zone nonlinearity to offset the lead error and significantly improving the voltage stability at the grid connection point.

[0159] S5.3: Perform amplitude limiting and waveform shaping on the virtual harmonic voltage reference quantity, and generate a weak harmonic current injection command containing specific harmonics based on the inverter dead zone effect characteristic curve to match the grid impedance characteristics and form an equivalent power absorption channel.

[0160] The virtual harmonic voltage reference value with the same frequency and opposite phase generated in the previous step is received as the initial input data for waveform shaping and amplitude limiting processing.

[0161] Peak detection is performed on the virtual harmonic voltage reference quantity to extract the absolute value of the instantaneous voltage amplitude within the current control cycle and construct the amplitude monitoring sequence.

[0162] The inverter dead zone effect refers to the deviation between the inverter output voltage and the ideal pulse width modulation voltage during the switching delay protection period set to prevent direct short circuit between the upper and lower switches on the same bridge arm of the energy storage converter. This deviation characteristic generates nonlinear voltage loss as the load current direction changes. This application utilizes this voltage loss to form an equivalent absorption channel for the injected weak harmonic current.

[0163] The preset dead-time effect characteristic lookup table is invoked, and the maximum permissible harmonic voltage threshold is determined based on the dead-time parameters of the energy storage converter switching devices and the DC bus voltage level.

[0164] The absolute value of the instantaneous voltage amplitude is compared with the maximum harmonic voltage threshold. If it exceeds the threshold, a saturation truncation algorithm is used to limit the amplitude within the threshold range, generating a voltage reference signal with limited amplitude.

[0165] A low-pass filtering algorithm is applied to the voltage reference signal after amplitude limiting to filter out high-frequency switching noise interference, retain the fundamental and target subharmonic components, and form a smooth voltage waveform profile.

[0166] Based on the smooth voltage waveform profile and combined with the equivalent impedance model of the grid connection point, the theoretical injection current amplitude is calculated using Ohm's law.

[0167] Based on the frequency characteristics of the theoretical injected current amplitude and the predicted deviation signal, a weak harmonic current injection command containing specific subharmonics is synthesized to ensure that the current phase is strictly out of phase with the predicted deviation.

[0168] By using amplitude limiting and shaping and impedance matching, the virtual voltage reference value from the previous step is converted into a weak harmonic current injection command that conforms to the dead-zone nonlinearity of the inverter, thereby achieving the expected technical effect of using the dead-zone effect to form an equivalent power absorption channel to buffer the prediction lead error.

[0169] For example, with a dead time of 2μs and a DC bus voltage of 750V, the maximum harmonic voltage threshold is found to be 15V according to a table. The peak value of the virtual harmonic voltage reference is detected to be 18V, which is then limited to 15V after saturation cutoff. The equivalent impedance Z at the grid connection point... grid The impedance is 0.5Ω. The theoretical injected current amplitude is calculated to be I = 15 / 0.5 = 30A. A sinusoidal harmonic current command with a frequency of 50Hz, an amplitude of 30A, and the phase opposite to the prediction deviation is generated. This command drives the converter to generate equivalent reactive power absorption during the dead time, significantly suppressing the power overshoot caused by prediction lead and improving the voltage stability at the grid connection point.

[0170] S5.4: The pulse width modulation module of the energy storage converter is driven by the weak harmonic current injection command to perform the actual injection operation of reactive power support error buffer current at the grid connection point, so as to use the dead zone nonlinear effect to offset the power fluctuation component caused by the prediction lead.

[0171] It receives the weak harmonic current injection command generated in the previous step, and parses the fundamental frequency, harmonic order and target amplitude phase parameters contained therein, which are used as reference inputs for the pulse width modulation module.

[0172] Based on the real-time sampled value of the DC bus voltage of the energy storage converter, the maximum linear modulation ratio within the current switching cycle is calculated, and the dynamic upper limit boundary of the inverter output voltage is determined.

[0173] The weak harmonic current injection command is converted into a voltage reference vector in a two-phase stationary coordinate system and superimposed on the fundamental voltage reference quantity to generate a composite voltage reference signal containing high-frequency harmonic components.

[0174] A space vector pulse width modulation algorithm is used to determine the sector and calculate the action time of the composite voltage reference signal to generate an initial switching state sequence.

[0175] Dead time compensation logic is introduced to nonlinearly correct the pulse width in the initial switching state sequence based on the dead time characteristic curve of the inverter power device, thereby offsetting the voltage loss caused by the dead time effect.

[0176] The corrected switching state sequence is mapped to the drive pulse signal of the three-phase bridge arm and transmitted to the gate of the insulated gate bipolar transistor through the isolation drive circuit.

[0177] The drive energy storage converter outputs a weak harmonic current at the grid connection point that is in phase and at the same frequency as the prediction deviation. It utilizes the interaction between the grid impedance and the inverter output impedance to form an equivalent power absorption channel.

[0178] By using the above processing method, the current command in the previous step is converted into an actual physical current injection action. The dead-zone nonlinear effect is used to offset the power fluctuation component caused by the prediction lead, thereby realizing reactive power support error buffering.

[0179] S5.5: Real-time monitoring of the voltage and current timing data at the grid connection point after injection, extraction of characteristic parameters of the equivalent power absorption effect, and generation of the execution feedback state of the reactive power support type error buffer current command to complete the closed-loop control process in inertial suppression mode.

[0180] The instantaneous values ​​of the three-phase voltage and the instantaneous values ​​of the injected harmonic current at the grid connection point are collected, and the original electrical quantity timing data are obtained by using a high-speed analog-to-digital converter at a microsecond-level sampling frequency.

[0181] A synchronous rotating coordinate transformation is performed on the original electrical quantity timing data to map the voltage and current signals in the three-phase stationary coordinate system to the two-phase rotating coordinate system, separating the fundamental component and the high-frequency harmonic component.

[0182] Extract the voltage and current components of a specific subharmonic that have the same frequency as the predicted deviation signal, calculate their instantaneous power product in the complex domain, and construct an equivalent absorbed power observation vector.

[0183] A moving average filtering algorithm is used to denoise the instantaneous power product, eliminate switching noise interference, and generate a smooth power index that characterizes the actual energy consumption effect of the virtual impedance.

[0184] The deviation between the smoothing power index and the preset error buffer target power is compared, the residual energy value is calculated, and the equivalent power absorption generated by the current dead zone effect is evaluated to determine whether it is sufficient to offset the fluctuation impact caused by the forecast lead.

[0185] The gain coefficient of the virtual impedance simulation unit is adjusted based on the residual energy value. If the residual is greater than the threshold, the harmonic injection amplitude is increased; otherwise, it is maintained or finely adjusted, forming a closed-loop feedback adjustment mechanism.

[0186] Generate an execution feedback state word containing the current equivalent absorbed power state, residual energy value, and gain adjustment instructions, which is used to update the internal variables of the lightweight finite state machine.

[0187] By monitoring the electrical quantities at the grid connection point in real time and calculating the equivalent absorbed power residual, the harmonic injection action of the previous step is transformed into a quantifiable error compensation effect index, thereby achieving precise closed-loop control in inertial suppression mode.

[0188] For example, with a sampling frequency of 50kHz, for the 5th prediction bias harmonic components, the voltage harmonic amplitude of the d-axis in the rotating coordinate system is extracted to be 2.5V, and the current harmonic amplitude is 1.2A. The instantaneous power product is calculated and filtered through a 20ms sliding window, yielding an equivalent absorbed power of 3.0W. The preset target buffer power is 3.2W, and the residual energy is 0.2W. Since the residual is less than the 0.5W threshold, the compensation effect is deemed good. The current virtual impedance gain coefficient remains unchanged, and the output execution feedback status word indicates stable system operation, significantly improving the robustness of suppressing prediction lead errors.

[0189] Step S6: Based on the execution process of the hysteresis-compensated power command or the reactive power-supported error buffer current command, continuously monitor the rate of change of state of charge over three cycles and compare it with the nominal slope threshold. If continuous deviation is detected, dynamically adjust the buffer release time constant to generate an adaptive correction factor anchored to the electrochemical characteristics of the battery. Specifically, this includes:

[0190] S6.1: Synchronously acquire battery terminal voltage and current data during the execution of hysteresis compensation power command or reactive power support error buffer current command, and use the sliding window integral algorithm to calculate the actual state of charge change within three consecutive control cycles to generate a real-time state of charge change rate sequence containing time-series cumulative characteristics.

[0191] S6.2: Based on the real-time rate of change of state of charge sequence, the preset electrochemical kinetic safety envelope model is called to extract the nominal slope threshold under the current temperature gradient, and the real-time rate of change of state of charge sequence and the nominal slope threshold are compared point by point to generate a continuous deviation judgment flag that characterizes the degree of deviation.

[0192] The real-time charge state change rate sequence generated by S6.1 is received as the input data stream for determining the electrochemical kinetic safety boundary.

[0193] The preset electrochemical kinetic safety envelope model is invoked, which stores the maximum allowable SOC change slope threshold matrix corresponding to different temperature gradient ranges.

[0194] Based on the currently collected temperature gradient distribution data, multidimensional linear interpolation is performed in the threshold matrix to extract the nominal slope threshold under the current operating conditions.

[0195] The instantaneous deviation is calculated by performing a point-by-point subtraction operation between the value of each sampling point in the real-time rate of change of state of charge sequence and the extracted nominal slope threshold.

[0196] The absolute value of the instantaneous deviation is measured and compared with a preset tolerance threshold for small disturbances to eliminate false deviations caused by measurement noise.

[0197] If the absolute deviation exceeds the tolerance threshold, it is determined that there is a significant deviation in the cycle, and a high-level deviation judgment flag is generated.

[0198] If the absolute deviation does not exceed the tolerance threshold, the cycle is determined to be within the safe envelope, and a low-level normal flag is generated.

[0199] Perform a logical AND operation or sequence buffering on the decision flags of three consecutive control cycles to form a continuous deviation decision flag representing a continuous deviation state.

[0200] Through the above comparison and logical judgment processing, the time series data of the previous step is transformed into discrete state indicators that characterize the electrochemical stress risk of the battery, realizing real-time quantitative assessment of the dynamic response safety of the energy storage system.

[0201] S6.3: When the continuous deviation judgment flag indicates that there is a deviation for three consecutive cycles, the time constant correction coefficient is calculated by using a nonlinear mapping function based on the deviation magnitude of the real-time state of charge change rate sequence, so as to generate a dynamic adjustment gain parameter for quantifying the adjustment intensity.

[0202] Obtain the continuous deviation judgment flag and real-time state of charge change rate sequence output from step S6.2, and confirm that the trigger condition of deviation exists for three consecutive control cycles is met at the current moment.

[0203] Extract the absolute values ​​of the deviations between the real-time rate of change of state of charge and the nominal slope threshold within the most recent three control cycles, and construct the deviation magnitude vector.

[0204] Normalization is performed on the deviation magnitude vector to eliminate the influence of dimensional differences on subsequent nonlinear mapping and generate a standardized deviation factor.

[0205] The S-shaped sigmoid function is used as the core algorithm for nonlinear mapping. The standardized deviation factor is mapped to a preset gain range, and the time constant correction coefficient is calculated.

[0206] Based on the electrochemical polarization characteristics of the battery, the curve steepness parameter and deviation center offset of the sigmoid function are set to enhance the sensitivity to small deviations and balance the response weights of positive and negative deviations.

[0207] The calculated time constant correction coefficient is converted into a dynamically adjusted gain parameter, which characterizes the adjustment strength of the buffer release rate in the next cycle.

[0208] By using the S-shaped nonlinear mapping method, the continuous deviation judgment result of the previous step is transformed into a quantified time constant correction coefficient, thereby achieving the expected technical effect of adaptive gain adjustment for the degree of hysteresis in battery electrochemical response.

[0209] For example, when the SOC change rates for three consecutive cycles are detected to be 0.02% / s, 0.025% / s, and 0.03% / s, respectively, and the nominal slope threshold is 0.015% / s, the calculated deviation vector is [0.005, 0.01, 0.015]. After normalization to obtain the standardized deviation factor, the time constant correction coefficient is calculated using a nonlinear mapping with a sigmoid function. This coefficient is greater than 1, indicating that the buffer release rate needs to be appropriately accelerated to compensate for hysteresis, avoiding overshoot or insufficient response caused by fixed gain, and significantly improving the timeliness matching of power smoothing.

[0210] S6.4: The original initial release time constant of the annular buffer is updated by dynamically adjusting the gain parameter and reconstructing the buffer data release curve in combination with the battery polarization impedance characteristics, so as to generate an optimized buffer release time constant that adapts to the current electrochemical response state.

[0211] Obtain the dynamically adjusted gain parameters generated in the preceding steps and the initial release time constant currently stored in the ring buffer, as the reference input data for updating the time constant.

[0212] Read the estimated polarization impedance value reported in real time by the battery management system. This value characterizes the electrochemical response hysteresis characteristics under the current temperature gradient and state of charge.

[0213] A time constant coupling correction model is constructed to map the dynamically adjusted gain parameter, the initial release time constant, and the polarization impedance estimate to a unified computational space.

[0214] The optimized buffer release time constant is calculated using the following formula:

[0215]

[0216] in, To optimize the buffer release time constant, The initial release time constant, To dynamically adjust the gain parameters, This is a real-time polarization impedance estimate. This is the standard reference impedance value.

[0217] The calculated optimized time constant is subjected to physical boundary limiting to ensure that its value is within the range of the battery's minimum allowable response time and maximum safe delay time.

[0218] Based on the optimized time constant after limiting, the data dequeue rate curve of the annular buffer is reconstructed to match the release rhythm of power increment with the ion diffusion rate inside the battery.

[0219] By using the weighted update and curve reconstruction methods described above, the dynamic adjustment gain of the previous step is transformed into an optimized buffer release time constant that adapts to the current electrochemical state, thereby achieving adaptive synchronization between the energy storage power response and the battery physical characteristics and significantly reducing the risk of electrode polarization caused by response mismatch.

[0220] S6.5: Based on the coupling relationship between the optimized buffer release time constant and the nominal slope threshold, an error feedback correction logic is constructed and the final control variable is output to generate an adaptive correction factor that can anchor the battery electrochemical characteristics and has adaptive capabilities.

[0221] Obtain the optimized buffer release time constant and the nominal slope threshold, and construct error feedback correction logic to generate an adaptive correction factor.

[0222] Read the release time constant of the optimized buffer generated in S6.4 and the nominal slope threshold under the current temperature gradient extracted in S6.2 as the initial input variables for the coupled calculation.

[0223] The ratio of the optimized release time constant to the nominal slope threshold is calculated to quantify the deviation of the current electrochemical response rate from the ideal safety envelope boundary.

[0224] A nonlinear saturation function is used to map the values, suppressing parameter oscillations under extreme conditions and generating a normalized dynamic adjustment gain coefficient as an adaptive correction factor.

[0225] The adaptive correction factor is weighted and fused with the preset basic phase difference angle boundary value to reconstruct the state transition critical conditions of the finite state machine.

[0226] The output includes the updated state transition threshold and an adaptive correction factor that dynamically adjusts the gain coefficient, which is used to drive the next round of control loop.

[0227] By using the above-mentioned coupled calculation and nonlinear mapping processing, the time constant optimization result of the previous step is transformed into an adaptive correction factor with electrochemical characteristic anchoring capability, realizing real-time adaptive correction of the internal dynamic state of the battery by the energy storage power allocation strategy, and significantly improving the robustness and safety of the system under the fluctuation of prediction error.

[0228] For example, a nominal slope threshold S is set. nom The buffer release time constant τ is optimized to be 0.05% / s. opt The time interval is 2.5s. The calculated ratio is 50. After mapping with the nonlinear saturation function f, the dynamic adjustment gain coefficient K is 1.2. This coefficient is applied to the phase difference angle boundary, and the original +30° boundary is adjusted to +36°. In three consecutive monitoring cycles, the SOC change rate remained stable within the safety envelope, and overcharge protection was not triggered. The system response smoothness was significantly improved, effectively avoiding the phenomenon of increased battery polarization caused by prediction errors.

[0229] Step S7: Update the state transition conditions of the lightweight finite state machine using an adaptive correction factor, relying solely on local observables to drive the closed-loop flow from the dual-channel state-aware dataset to the mode-switching decision result in the next round, thereby generating the final energy storage power allocation execution sequence with adaptive error correction capabilities. Specifically, this includes:

[0230] S7.1: Analyze the adaptive correction factor generated in the previous round, extract the dynamically adjusted gain parameter and the optimized buffer release time constant contained therein, in order to generate a set of localized control variables for reconstructing the state transition logic.

[0231] S7.2: Based on the localized set of control variables, perform a nonlinear mapping update operation on the preset state transition threshold table of the lightweight finite state machine, and replace the original fixed phase difference angle boundary with a dynamic critical value that floats with the electrochemical response state to generate an updated state transition condition matrix with adaptive characteristics.

[0232] It receives dynamically adjusted gain parameters and optimized buffer release time constants from the set of localized control variables as input benchmarks for updating the state transition threshold table.

[0233] The pre-built nonlinear mapping function library is invoked, and the S-shaped sigmoid curve is selected as the basic mapping model for phase difference angle boundary drift, so as to smoothly handle the threshold mutation risk caused by drastic changes in electrochemical response.

[0234] The optimized buffer release time constant is normalized to an electrochemical inertia coefficient, which characterizes the degree of hysteresis in the execution of power commands due to the current internal ion diffusion resistance of the battery.

[0235] Based on the electrochemical inertia coefficient, the dynamic phase critical value offset is calculated using the following formula:

[0236]

[0237] in, This is the phase critical value offset. This is the maximum permissible offset constant. As a sensitivity adjustment factor, To optimize the buffer release time constant, This is the reference value for the nominal time constant.

[0238] The calculated dynamic phase critical value offset is superimposed on the original +30° and -30° fixed boundary values ​​to generate upper and lower critical values ​​that fluctuate with the electrochemical state.

[0239] By combining dynamically adjusted gain parameters, the update rates of the upper and lower critical values ​​are weighted and constrained to prevent high-frequency jitter of the state transition threshold caused by transient noise interference.

[0240] Write the updated upper and lower critical values ​​to the memory address of the state transition threshold table of the lightweight finite state machine, replacing the original static angle constants.

[0241] By using a nonlinear mapping update process, the adaptive correction factor from the previous step is transformed into a dynamic critical value matrix with electrochemical sensing capabilities, thereby achieving the expected technical effect of real-time adaptation of the state switching logic to the battery's physical characteristics.

[0242] For example, the maximum allowable offset amplitude constant K is set to 15°, the sensitivity adjustment factor α is 0.5, and the nominal time constant reference value τ0 is 200ms. When the optimized buffer release time constant τ is detected to increase to 300ms due to low temperature and high resistance characteristics, the exponential term parameter is calculated as -0.5×(300-200)=-50. At this time, e raised to the power of -50 approaches 0, and the denominator approaches 1, resulting in a dynamic phase critical value offset Δθ of approximately 15°. The system automatically raises the original upper limit critical value of +30° to +45° and automatically lowers the original lower limit critical value of -30° to -45°. This adjustment expands the trigger range of the conventional proportional-integral control mode, avoids frequent entry into the inertia extension mode when the battery response is slow, and significantly improves the control stability and robustness of the system under wide temperature range conditions.

[0243] S7.3: Using the updated state transition condition matrix, the dual-channel state-aware dataset at the current moment is matched and judged in real time. Based on the phase difference angle index in the local observable and the gradient of the terminal voltage change rate, the state jump logic is triggered to generate the next round mode switching decision result pointing to the conventional proportional-integral regulation, inertial extension mode or inertial suppression mode.

[0244] The dimensionless phase difference angle index and the terminal voltage change rate gradient data of the current control cycle are extracted from the dual-channel state-aware dataset and used as the original input variables for the state determination of the finite state machine.

[0245] The state transition condition matrix updated in step S7.2 is invoked to read the dynamic critical thresholds that fluctuate with the electrochemical response state, including the upper and lower limits of the dynamic synchronization interval and the boundary value of the inertial mode trigger.

[0246] The extracted phase difference angle index is compared with the upper and lower limits of the dynamic synchronization interval to determine whether the current error evolution rhythm is within the stable region where the predicted and actual fluctuations are basically synchronized.

[0247] If the phase difference angle index falls within the dynamic synchronization range and the gradient of the terminal voltage change rate does not exceed the preset safety limit, the system is determined to be in the steady-state tracking stage, and a conventional proportional-integral adjustment command identifier is generated.

[0248] If the phase difference angle index exceeds the upper limit of the dynamic synchronization range, it indicates that the predicted signal is seriously lagging behind the actual fluctuations. Further verification is needed to determine whether the voltage change rate gradient is in the downward trend range that allows for rapid power injection.

[0249] When the hysteresis determination condition is met and the voltage gradient meets the discharge safety constraints, the inertial extension mode control flag is generated, and the power incremental gradual release logic based on the ring buffer is started.

[0250] If the phase difference angle index is lower than the lower limit of the dynamic synchronization range, it indicates that the predicted signal is excessively ahead of the actual fluctuations, and whether the voltage change rate gradient at the synchronous monitoring end shows abnormal rising or oscillating characteristics.

[0251] When the advance judgment condition is met and the voltage gradient indicates that the battery is in a state of limited charging acceptance, the inertia suppression mode control flag is triggered, and the virtual impedance simulation unit is activated to perform reactive power support type error buffering.

[0252] Interlock verification is performed on the generated mode switching decision results to ensure the continuity of power commands at the moment of mode switching and to prevent current surges in the energy storage converter caused by sudden state changes.

[0253] By using the above-mentioned real-time matching and multi-physical quantity joint verification processing method based on dynamic threshold matrix, the adaptive correction factor updated in the previous step is transformed into a specific mode switching decision result, thereby achieving the expected technical effect of smooth mode transition and robust control of energy storage system under prediction error fluctuation.

[0254] For example, the sampling time of the current control cycle is set to t=100ms. The phase difference angle index is extracted from the dual-channel state-aware dataset as +45°, and the terminal voltage change rate gradient is -0.05V / s. The updated state transition condition matrix is ​​read. Due to the deviation of the SOC change rate in the previous cycle, the upper limit of the dynamic synchronization interval is adjusted from +30° to +35°, and the lower limit is adjusted from -30° to -35°. Comparing +45° with +35°, it is determined that the upper limit of the synchronization interval is exceeded. The terminal voltage change rate is verified to be -0.05V / s, confirming that it is in the voltage drop range of allowable discharge and has not reached the minimum voltage protection threshold. Based on this, the inertial extension mode control flag is generated. If the phase difference angle is -40°, which is lower than the dynamic lower limit of -35°, and the terminal voltage change rate is +0.02V / s, it indicates that the voltage is rising and the prediction is ahead, triggering the generation of the inertial suppression mode control flag. If the phase difference angle is +10°, falling between -35° and +35°, and the voltage gradient is stable, a conventional proportional-integral (PI) control command identifier is generated. This process ensures that mode switching strictly depends on the dynamic boundaries of local real-time observations, significantly improving the system's response accuracy to sudden changes in prediction errors.

[0255] S7.4: Based on the decision result of the next round of mode switching, call the corresponding power command generation algorithm module to convert the selected control mode into a specific active power reference value or reactive power support error buffer current command, so as to generate a single-cycle energy storage power allocation execution sequence with error adaptive correction capability.

[0256] Receive the next round of mode switching decision results, parse the currently active control mode identifier, and determine the type of power command generation algorithm module to be called.

[0257] If the control mode identifier indicates conventional proportional-integral control, read the deviation between the measured power sequence and the predicted power sequence from the dual-channel state-aware dataset.

[0258] The deviation value is input into the proportional-integral controller, and the dynamic adjustment gain parameter in the adaptive correction factor is used to calculate the reference value of the basic active power.

[0259] If the control mode identifier indicates inertial extension mode, fetch the cache increment instruction set from the circular buffer and read the optimized buffer release time constant.

[0260] Based on the slope constraint factor generated by the rate of change of terminal voltage, a nonlinear progressive release operation is performed on the buffer increment instruction to generate an electrochemically safe power compensation sequence.

[0261] The base value command and the power compensation sequence are vector-superimposed to synthesize an active power reference value with hysteresis compensation characteristics.

[0262] If the control mode identifier indicates inertial suppression mode, the frequency component of the predicted deviation signal is obtained, and the phase reversal algorithm is executed to generate a virtual harmonic voltage reference quantity with the same frequency and opposite phase.

[0263] Based on the inverter dead-zone effect characteristic curve, the virtual harmonic voltage reference is amplitude-limited and waveform-shaped to generate a weak harmonic current injection command.

[0264] The generated active power reference value or reactive power support error buffer current command is encapsulated into a single-cycle energy storage power allocation execution sequence.

[0265] By using pattern recognition and differentiated instruction synthesis processing, the mode switching decision result of the previous step is transformed into specific converter control instructions, enabling the energy storage system to adaptively and accurately respond to prediction errors.

[0266] S7.5: Perform timing splicing and smooth transition processing on the single-cycle energy storage power allocation sequence, embed it into the continuously running control flow and feed it back to the energy storage converter drive unit to complete the fully closed-loop autonomous control process from the dual-channel state-aware dataset to the final power output.

[0267] The system receives a single-cycle energy storage power allocation execution sequence with adaptive error correction capability and extracts the active power reference value or reactive power support current command component. It then performs a difference calculation between the extracted power command component and the final output command of the previous control cycle to obtain the power jump amplitude value for the current cycle. Based on the power jump amplitude value, it calls a preset ramp rate limiting function to calculate the maximum allowable slope that satisfies the IGBT switching frequency constraints of the energy storage converter. The system uses the maximum allowable slope to perform linear interpolation reconstruction on the power command for the current cycle, generating a smoothly transitioned intermediate power command sequence. Dead-time compensation verification is performed on the intermediate power command sequence, and minor interval disturbances that may cause bridge arm short circuits are eliminated based on the inverter hardware dead-time characteristic curve. The verified smoothed power command is encapsulated into a standardized communication frame and sent in real-time to the underlying drive unit of the energy storage converter via the SPI bus. After receiving the command, the drive unit updates the duty cycle register of the pulse width modulation module, completing the mapping from digital control signals to physical power output. By using timing splicing and smooth transition processing, discrete single-cycle commands are transformed into continuous and stable physical power output, eliminating the power step shock caused by mode switching and achieving stable operation of the fully closed-loop autonomous control process.

[0268] For example, the control cycle is set to 10ms, the output power of the previous cycle is 50kW, and the current single-cycle execution sequence calculation value is 55kW. The calculated power jump amplitude is 5kW. The preset maximum allowable change slope is 200kW / s, so the maximum allowable change within 10ms is 2kW. Since 5kW is greater than 2kW, linear interpolation is used to correct the target value of this cycle to 52kW. The 52kW command is checked for dead time, and after confirming that there are no abnormalities, it is packaged and issued. The converter drive unit updates the PWM duty cycle, and the actual output power smoothly rises to 52kW, avoiding the instantaneous step jump of 3kW, significantly reducing the harmonic content of grid voltage fluctuations, and improving the power quality at the grid connection point.

[0269] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0270] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The “multiple” mentioned in the embodiments of this application refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0271] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for mitigating and controlling wind and solar power fluctuations in a charging pile energy storage system, characterized in that, include: Acquire measured and predicted power sequences of wind and solar power, and simultaneously collect current state of charge, terminal voltage change rate and temperature gradient data of energy storage batteries to generate a dual-channel state-aware dataset. Based on the dual-channel state-aware dataset, the phase difference angle between the predicted deviation signal and the first derivative of the measured power within the sliding time window is calculated, and a phase difference angle index is generated. The range of the phase difference angle index is determined. If the phase difference angle is between -30° and +30°, a conventional proportional-integral adjustment command is output. If it is greater than +30°, an inertial extension mode control flag is generated. If it is less than -30°, an inertial suppression mode control flag is generated, so as to generate a mode switching decision result containing three different response strategies. When the inertial extension mode control flag is received, the original power command is split into a base command and an incremental command, and the incremental command is cached in a circular buffer. The release slope is set according to the terminal voltage drop rate to perform nonlinear asymptotic processing on the cached data to generate a hysteresis-compensated power command with electrochemical kinetic safety envelope constraints. When the inertia suppression mode control flag is received, the current power regulation action is frozen and the virtual impedance simulation unit is activated. A weak harmonic current with the same frequency and opposite phase as the predicted deviation is injected into the grid connection point to generate equivalent power absorption by utilizing the inverter dead zone effect, so as to generate a reactive power support type error buffer current command.

2. The method for mitigating wind and solar power fluctuations in a charging pile energy storage system according to claim 1, characterized in that, After generating the reactive power support type error buffer current command, it also includes: Based on the execution process of the hysteresis compensation power command or reactive power support error buffer current command, the rate of change of state of charge within three cycles is continuously monitored and compared with the nominal slope threshold. If continuous deviation is detected, the buffer release time constant is dynamically adjusted to generate an adaptive correction factor that anchors the electrochemical characteristics of the battery. The state transition conditions of the lightweight finite state machine are updated using the adaptive correction factor, and the closed-loop flow from the dual-channel state-aware dataset to the mode switching decision result is driven only by local observables to generate the final energy storage power allocation execution sequence with error adaptive correction capability.

3. The method for mitigating wind and solar power fluctuations in a charging pile energy storage system according to claim 1, characterized in that, The step of calculating the phase difference angle between the predicted deviation signal and the first derivative of the measured power within a sliding time window based on the dual-channel state-aware dataset, and generating a phase difference angle index, includes: Differential operations are performed on the measured power sequence of wind and solar power in the dual-channel state-aware dataset to obtain the first derivative sequence of the measured power, which characterizes the actual rate of fluctuation. A point-by-point subtraction operation is performed between the measured power sequence and the predicted power sequence to generate a prediction deviation signal sequence that characterizes the instantaneous prediction accuracy. The predicted deviation signal sequence and the measured power first derivative sequence are synchronously truncated and aligned using a sliding time window of preset length to generate local signal segment pairs containing temporal correspondence. Based on local signal segments, cross-correlation analysis algorithms or orthogonal decomposition methods are applied to calculate the time offset between the two to generate the original phase difference value that characterizes the error lead or lag characteristics. The original phase difference values ​​are normalized by angle mapping to generate the final phase difference angle index used to characterize the rhythm of error evolution.

4. The method for mitigating wind and solar power fluctuations in a charging pile energy storage system according to claim 1, characterized in that, The generation of hysteresis-compensated power commands with electrochemical kinetic safety envelope constraints includes: Based on the received inertial extension mode control flag, the original power command is decomposed into a base command and an incremental command to generate an incremental command data stream to be processed. Based on the generated incremental instruction data stream to be processed, the incremental instructions are temporarily stored using a circular buffer storage mechanism to generate a cached incremental instruction set containing time series features. Based on the real-time collected terminal voltage change rate data, the dynamic release slope parameter is calculated through the slope mapping algorithm to generate a slope constraint factor characterizing the limit of the electrochemical response rate. Based on the generated slope constraint factor, a nonlinear progressive release process is performed on the cache incremental instruction set, which is then gradually transformed into a smooth power component according to the rate curve defined by the slope constraint factor, in order to generate an electrochemically safe power compensation sequence. Based on the generated base value command and the electrochemically safe power compensation sequence, a vector superposition operation is performed to merge the base value command and the electrochemically safe power compensation sequence into a single control signal, so as to generate a hysteresis compensation power command with electrochemical kinetic safety envelope constraints.

5. A method for mitigating wind and solar power fluctuations in a charging pile energy storage system according to claim 1, characterized in that, The generated dual-channel state-aware dataset includes: A sliding window truncation operation is performed on the measured power sequence, the predicted power sequence, the state of charge, the rate of change of terminal voltage, and the temperature gradient data; a linear interpolation compensation algorithm is used to fill in the data missing points caused by the difference in sampling frequency; and the interpolated data set is used as the basic input of the dual-channel state-aware dataset.

6. The method for mitigating wind and solar power fluctuations in a charging pile energy storage system according to claim 1, characterized in that, Based on the numerical range boundary comparison processing of the phase difference angle index, the interval belonging state quantity is obtained by performing finite state machine transition logic mapping processing to generate the mode switching decision result containing the conventional proportional-integral adjustment command, the inertial extension mode control flag, or the inertial suppression mode control flag.

7. A method for mitigating wind and solar power fluctuations in a charging pile energy storage system according to claim 6, characterized in that, The mode switching decision results are subjected to local observability consistency verification to generate a final mode switching decision result with adaptive error correction capability.

8. A method for mitigating wind and solar power fluctuations in a charging pile energy storage system according to claim 1, characterized in that, Based on the power command freeze signal triggered by the current power adjustment action, the enable logic of the virtual impedance simulation unit is activated. Using the prediction deviation signal as a reference input, the virtual harmonic voltage reference quantity with the same frequency and opposite phase is calculated through the phase reversal algorithm.

9. A method for mitigating wind and solar power fluctuations in a charging pile energy storage system according to claim 8, characterized in that, After performing the current injection operation to the grid connection point based on the virtual harmonic voltage reference quantity with the same frequency and opposite phase, the voltage and current timing data after injection are monitored in real time, the characteristic parameters of the equivalent power absorption effect are extracted, and the execution feedback state of the reactive power support type error buffer current command is generated to complete the closed-loop control process in the inertial suppression mode.

10. A method for mitigating wind and solar power fluctuations in a charging pile energy storage system according to claim 1, characterized in that, All mode switching and power allocation rely on local observables, which include the measured power, the predicted power, the state of charge, the rate of change of terminal voltage, and the temperature gradient in the dual-channel state-aware dataset, without the need for real-time intervention from external prediction models.

Images