Energy storage power station grid-connected and off-grid adaptive control method based on load characteristic analysis

Abstract

The invention discloses an energy storage power station grid-connected and off-grid adaptive control method based on load characteristic analysis, and relates to the technical field of operation control, and the method comprises the steps: carrying out the collection of operation mode data, carrying out the preprocessing of the operation mode data, and obtaining a quasi-synchronization characteristic sequence; constructing a snapshot matrix based on the quasi-synchronization feature sequence, and calculating through a dynamic mode decomposition algorithm to obtain a load dynamic feature set; performing power distribution instruction calculation based on the load dynamic feature set to obtain a primary power distribution instruction, and performing secondary power distribution correction by combining the charge state of the storage battery and the charge state of the supercapacitor; optimizing the control parameters of the linear active-disturbance-rejection controller by adopting an improved spider-man algorithm, and calculating to obtain an optimal control parameter group of the linear active-disturbance-rejection controller; and calculating a current power balance state, generating a mode switching instruction in combination with the load dynamic feature set, and controlling the energy storage converter to execute grid-connected and off-grid adaptive switching based on the mode switching instruction.

Description

Technical Field

[0001] This invention relates to the field of operation control technology, and in particular to an adaptive control method for on-grid and off-grid power stations based on load characteristic analysis. Background Technology

[0002] In recent years, with the increasing penetration rate of renewable energy and the growing demand for flexible resource regulation in power systems, energy storage power stations, as key equipment supporting stable grid operation, improving power quality, and achieving peak shaving and valley filling, have seen their large-scale grid-connected operation become an important direction for the development of modern power systems. Energy storage power stations typically need to have both grid-connected and off-grid operating modes, and be able to switch smoothly and reliably between the two modes to adapt to various operating conditions such as grid faults, planned islanding, and microgrid operation. However, in actual operation, the dynamic characteristics of loads are complex and variable, especially the power surges and oscillations caused by impulsive loads and intermittent renewable energy access, posing severe challenges to the power distribution, voltage stability, and mode switching of energy storage systems.

[0003] Currently, Chinese invention patent application CN202110977189.X discloses a method for controlling the switching between grid-connected and off-grid operation of an energy storage converter. This method detects the operating state of the energy storage converter. When the converter is in grid-connected mode, it checks whether the three-phase grid voltage is less than a limit. If so, it uses the phase angle and amplitude of the three-phase grid voltage in grid-connected mode as the phase angle and amplitude of the three-phase grid voltage in off-grid mode, achieving a smooth transition from grid-connected to off-grid operation. When the converter is in off-grid mode, it checks whether the three-phase grid voltage is greater than a limit. If so, it controls the difference between the phase angle of the off-grid voltage and the phase angle of the three-phase grid voltage within a specified range, achieving a switch from off-grid to grid-connected operation. This method enables stable switching between grid-connected and off-grid modes, improving the stability and reliability of the energy storage converter during operation. However, the related technologies do not adequately consider the impact of load dynamic characteristics and power transient changes during the switching process, do not coordinate the resulting system power deficit, and lack the ability to adaptively adjust according to changes in load dynamic characteristics. Summary of the Invention

[0004] The technical problem addressed by this invention is that during the switching between grid-connected and off-grid operation of energy storage power stations and during load surges, when the power command of an energy storage unit exceeds its allowable operating range, existing technologies typically only perform individual limiting or protection disconnection on the over-limited energy storage unit, failing to address the resulting power deficit at the system level. This leads to a decrease in the overall power support capacity of the energy storage system, resulting in voltage fluctuations, power surges, and instability during grid-connected / off-grid switching. Furthermore, the lack of effective perception of the dynamic characteristics of the connected loads makes it difficult to promptly identify the oscillation modes induced by load disturbances and their impact on system stability, resulting in a degree of blindness and lag in grid-connected / off-grid switching and power regulation strategies. On the other hand, in existing technologies, power allocation, power limiting, and controller parameter tuning are independent of each other, lacking a unified and coordinated control mechanism. This makes it difficult to maintain the continuity and balance of the total system power while ensuring the safe operation of energy storage units, thus affecting the stability and reliability of grid-connected and off-grid operation of energy storage power stations.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: an adaptive control method for grid-connected and off-grid operation of energy storage power stations based on load characteristic analysis, comprising the following steps: Step S1: Collect the operating mode data, preprocess the operating mode data, and obtain the quasi-synchronous feature sequence; Step S2: Construct a snapshot matrix based on the quasi-synchronous feature sequence, and calculate the load dynamic feature set using the dynamic mode decomposition algorithm; Step S3: Calculate the power allocation command based on the load dynamic feature set to obtain the initial power allocation command, and perform secondary power allocation correction by combining the battery state of charge and the supercapacitor state of charge. Step S4: The improved spider monkey algorithm is used to optimize the control parameters of the linear active disturbance rejection controller and calculate the optimal control parameter set of the linear active disturbance rejection controller. Step S5: Calculate the current power balance state and combine it with the load dynamic feature set to generate a mode switching command. Based on the mode switching command, control the energy storage converter to execute and perform adaptive off-grid switching.

[0006] As a preferred embodiment of the adaptive control method for grid connection and disconnection of energy storage power stations based on load characteristic analysis described in this invention, step S1 specifically includes: Step S11: Collect operating mode data through sensors. The operating mode data includes the DC bus voltage signal of the energy storage system, the battery branch current, the supercapacitor branch current, the battery state of charge, and the supercapacitor state of charge. Step S12: The DC bus voltage signal is decomposed using a multi-scale adaptive wavelet algorithm to obtain the decomposition coefficients of each layer. The local variance of each layer decomposition coefficient within the sampling window is calculated and compared with a preset significance threshold. Decomposition coefficients with local variance less than the preset significance threshold are identified as noise components and cleared to zero. The remaining decomposition coefficients are used as feature components and inverse discrete wavelet transform is performed to obtain the voltage components after noise removal. Step S13: Extract the dynamic increment in the voltage component after noise removal as the voltage transient feature, and perform time reference alignment processing on the voltage transient feature with the battery branch current, supercapacitor branch current, battery state of charge and supercapacitor state of charge to obtain the quasi-synchronous feature sequence.

[0007] As a preferred embodiment of the adaptive control method for grid connection and disconnection of energy storage power stations based on load characteristic analysis described in this invention, step S13 specifically includes: Step S131: Calculate the mean of the voltage component after noise removal using a sliding window to obtain the DC fundamental component. Subtract the DC fundamental component from the voltage component after noise removal to obtain the voltage transient characteristic sequence. Step S132: Input the voltage transient characteristic sequence into the zero-phase filter, and perform forward filtering and reverse filtering to obtain the phase-compensated voltage transient characteristic sequence. Step S133: Using the timestamp sequence in the phase-compensated voltage transient characteristic sequence as the reference coordinate, the battery branch current, supercapacitor branch current, battery state of charge and supercapacitor state of charge are resampled by a linear interpolation algorithm and mapped to a unified sampling time. Step S134: The voltage transient characteristics, battery branch current, supercapacitor branch current, battery state of charge and supercapacitor state of charge mapped to the unified sampling time are merged into a matrix column to obtain a quasi-synchronous feature sequence.

[0008] As a preferred embodiment of the adaptive control method for grid connection and disconnection of energy storage power stations based on load characteristic analysis described in this invention, step S2 specifically includes: Step S21: Extract the sampling vector from the quasi-synchronous feature sequence and construct a snapshot matrix according to the sampling time step. and the corresponding offset snapshot matrix ; Step S22, for the snapshot matrix Singular value decomposition yields the left singular matrix, the singular value diagonal matrix, and the right singular matrix, calculated using the following formula: ; in, Describes a left singular matrix. Represents a singular value diagonal matrix. Represents a right singular matrix; Step S23, truncate the first part of the singular value diagonal matrix. The dimensionality of the first-order eigenvalues ​​and their corresponding eigenvectors is reduced to obtain a low-order linear mapping matrix, which is calculated using the following formula: ; in, Represents a low-order linear mapping matrix. This indicates the left singular matrix before the cut-off point. The conjugate transpose of the order eigenvectors. This represents the truncated right singular matrix. This represents the inverse of the truncated singular value diagonal matrix; Step S24: Perform eigenvalue decomposition on the low-order linear mapping matrix to obtain eigenvalues ​​and corresponding eigenvectors; Step S25: Calculate the modal frequencies and real-time damping ratios of each order based on the eigenvalues ​​to obtain the load dynamic characteristic set. The calculation formula is as follows: ; ; in, Indicates modal frequency, This indicates that the eigenvalues ​​are obtained after eigenvalue decomposition of a low-order linear mapping matrix. Indicates the sign of the imaginary part. The symbol for the natural logarithm. Indicates the sampling time step. Indicates the real-time damping ratio. Indicates the sign of taking the real part. Indicates the absolute value symbol; The load dynamic feature set includes the dominant oscillation mode, real-time damping ratio, and mode frequency under the current load disturbance.

[0009] As a preferred embodiment of the grid-connected adaptive control method for energy storage power stations based on load characteristic analysis described in this invention, step S24 involves eigenvalue decomposition of the low-order linear mapping matrix to obtain eigenvalues ​​and corresponding eigenvectors. The processing logic includes: Step S241: Perform eigenvalue decomposition on the low-order linear mapping matrix to calculate... Each discrete-time eigenvalue and its corresponding low-order eigenvector; Step S242: Obtain the left singular matrix obtained from the singular value decomposition in step S22. Multiply the left singular matrix with each low-order eigenvector to obtain the dynamic modal eigenvector. The calculation formula is as follows: ; in, Represents the dynamic modal feature vector. Represents a low-order eigenvector; Step S243: Establish a one-to-one correspondence between eigenvalues ​​and dynamic modal eigenvectors.

[0010] As a preferred embodiment of the adaptive control method for grid connection and disconnection of energy storage power stations based on load characteristic analysis described in this invention, step S3 specifically includes: Step S31: Extract the maximum modal frequency from the load dynamic feature set as the dominant modal frequency, and calculate the current total power demand. Step S32: Calculate the initial power allocation command based on the relationship between the dominant mode frequency and the preset frequency division threshold. Step S33: Substitute the state of charge of the battery and the state of charge of the supercapacitor into the membership function for mapping, and calculate the corresponding power correction increment. The calculation formula is as follows: ; ; in, This represents the power correction increment of the supercapacitor. This indicates the power correction increment of the battery. Indicates the first correction gain factor. This represents the second correction gain factor. This represents the target center value of the state of charge of the supercapacitor. This represents the target center value of the battery's state of charge. Indicates the state of charge of a supercapacitor. Indicates the state of charge of the battery; Step S34, Determine Does it exceed the preset safe operating threshold? ; If either side If the limit is exceeded, a secondary power allocation correction is performed to obtain the final reference power command, the calculation formula of which is: ; ; in, This indicates the amount of power overflow on the over-limit side. This indicates the current reference power command for the over-limit side energy storage unit. This indicates the maximum power of the over-limit side energy storage unit. This indicates the final reference power command after correction for energy storage units that have not exceeded their limits. This indicates the original reference power command before correction by the energy storage unit on the non-over-limit side; Step S35: The initial power command, the power correction increment, and the final reference power command are added together to obtain the execution command and sent to the underlying controller.

[0011] As a preferred embodiment of the adaptive control method for grid-connected and off-grid power storage stations based on load characteristic analysis described in this invention, in step S32, the initial power allocation command is calculated according to the magnitude relationship between the dominant mode frequency and the preset frequency division threshold. The processing logic includes: like The initial power command for the supercapacitor and the initial power command for the battery are calculated using the following formula: ; ; in, Indicates the dominant mode frequency. This indicates the preset frequency division threshold. This indicates the initial power command for the supercapacitor. This indicates the current total power demand. Represents the high-frequency power allocation factor. This indicates the initial power command for the battery; like Then let and will Set to zero.

[0012] As a preferred embodiment of the adaptive control method for grid connection and disconnection of energy storage power stations based on load characteristic analysis described in this invention, step S4 specifically includes: Step S41: Obtain the real-time damping ratio and substitute it into the preset damping deviation evaluation function to calculate the damping deviation value. Step S42: Using the controller bandwidth of the linear active disturbance rejection controller as the decision variable, the damping deviation value and voltage fluctuation deviation are weighted and calculated to construct the objective function; Step S43: Initialize the spider monkey population and calculate the objective function value corresponding to the position of each individual spider monkey within the preset search space, and identify the global optimum and local optimum positions; Step S44: A damping compensation perturbation factor is introduced to correct the individual spider monkey position, updating the individual spider monkey position. The calculation formula is as follows: ; in, This indicates the updated location of the individual spider monkey. This indicates the location of the individual spider monkey before the update. Indicates the local optimal position. Indicates the globally optimal position. This represents a uniformly distributed random number within the interval [0,1]. This represents a uniformly distributed random number within the interval [-1, 1]. The disturbance factor representing damping compensation; Step S45: Determine whether the preset iteration stop condition is met; If satisfied, the parameter value corresponding to the global optimal position is output, and the parameter value corresponding to the global optimal position is determined as the optimal control parameter set of the linear active disturbance rejection controller. The bandwidth and gain of the underlying controller are updated during mode switching. The control parameters include the control gain of the linear active disturbance rejection controller, the observer gain, and the corresponding controller bandwidth; If the conditions are not met, return to step S44 to continue iterative updates.

[0013] As a preferred embodiment of the adaptive control method for grid-connected and off-grid power stations based on load characteristic analysis described in this invention, in step S42, the controller bandwidth of the linear active disturbance rejection controller is used as the decision variable, and the damping deviation value and voltage fluctuation deviation are weighted and calculated to construct the objective function. Its processing logic includes: ; in, Represents the objective function value. This represents the first weighting coefficient. This represents the second weighting coefficient. Indicates the range from 0 to Integrate over the time period region. Indicates time period, This represents the absolute value of the deviation of the control quantity. Indicates the reference damping ratio. Indicates the real-time damping ratio. Represents the differential symbol.

[0014] As a preferred embodiment of the adaptive control method for grid connection and disconnection of energy storage power stations based on load characteristic analysis described in this invention, step S5 specifically includes: Step S51: Obtain the real-time damping ratio, dominant mode frequency, and voltage and current on the AC and DC sides of the energy storage converter from the load dynamic characteristic set, and calculate the current power balance state. Step S52: Input the real-time damping ratio, dominant mode frequency and power balance state into the preset logic discrimination matrix, calculate the matching degree between the current operating condition and the preset mode switching threshold, and generate a mode switching command based on the matching degree. Step S53: When the mode switching command is a grid-connected to off-grid command, control the circuit breaker at the grid connection point to open, extract the optimal control parameter group of the linear active disturbance rejection controller determined in step S45, update the bandwidth and gain of the underlying controller, adjust the bandwidth of the underlying controller to the optimal controller bandwidth, and switch to voltage source control mode. Step S54: When the mode switching command is an off-grid to grid-connected command, the output voltage of the energy storage converter is adjusted by the pre-synchronization control algorithm so that the phase and frequency of the output voltage are consistent with the external power grid. When the preset closing conditions are met, the circuit breaker at the grid connection point is closed, and the current source control mode is switched. Step S55: Output pulse width modulation signal to energy storage converter according to the switched control mode, and perform grid-connected adaptive switching action.

[0015] The beneficial effects of this invention are as follows: This application innovatively introduces a secondary power allocation correction. During the switching between grid-connected and off-grid operation of an energy storage power station and during load surges, when the initial power allocation result obtained based on load characteristic analysis shows that the power of an energy storage unit exceeds its limit, the initial power command is redistributed at the system level. This allows energy storage units with power regulation margins to compensate for the excess power demand, thereby avoiding the problem of insufficient total system power caused by only single-unit limiting processing in the prior art. This method fully utilizes the complementary characteristics of batteries and supercapacitors in terms of energy density and power density in a hybrid energy storage system, enabling different energy storage units to collaboratively undertake power support tasks according to the dynamic characteristics of the load. While ensuring the safe operation of energy storage units, it maintains the continuity of power output, effectively suppresses power surges and voltage fluctuations caused by load disturbances during grid-connected and off-grid switching and islanded operation, and improves the operational stability, power supply reliability, and adaptability to complex loads and weak grid environments of the energy storage power station under grid-connected and off-grid adaptive control conditions. Attached Figure Description

[0016] Figure 1 A flowchart illustrating the steps of an adaptive control method for on-grid and off-grid power stations based on load characteristic analysis, provided as an embodiment of the present invention; Figure 2 This is a flowchart illustrating the steps of the secondary power distribution correction method of the present invention. Detailed Implementation

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] Example, refer to Figure 1 This paper presents an adaptive control method for grid connection and disconnection of energy storage power stations based on load characteristic analysis, including the following steps: Step S1: Collect the operating mode data, preprocess the operating mode data, and obtain the quasi-synchronous feature sequence; Step S2: Construct a snapshot matrix based on the quasi-synchronous feature sequence, and calculate the load dynamic feature set using the dynamic mode decomposition algorithm; Step S3: Calculate the power allocation command based on the load dynamic feature set to obtain the initial power allocation command, and perform secondary power allocation correction by combining the battery state of charge and the supercapacitor state of charge. Step S4: The improved spider monkey algorithm is used to optimize the control parameters of the linear active disturbance rejection controller and calculate the optimal control parameter set of the linear active disturbance rejection controller. Step S5: Calculate the current power balance state and combine it with the load dynamic feature set to generate a mode switching command. Based on the mode switching command, control the energy storage converter to execute and perform adaptive off-grid switching.

[0019] In specific implementation, step S1 includes: Step S11: Collect operating mode data through sensors. The operating mode data includes the DC bus voltage signal of the energy storage system, the battery branch current, the supercapacitor branch current, the battery state of charge, and the supercapacitor state of charge. Step S12: The DC bus voltage signal is decomposed using a multi-scale adaptive wavelet algorithm to obtain the decomposition coefficients of each layer. The local variance of each layer decomposition coefficient within the sampling window is calculated and compared with a preset significance threshold. Decomposition coefficients with local variance less than the preset significance threshold are identified as noise components and cleared to zero. The remaining decomposition coefficients are used as feature components and inverse discrete wavelet transform is performed to obtain the voltage components after noise removal. Step S13: Extract the dynamic increment in the voltage component after noise removal as the voltage transient feature, and perform time reference alignment processing on the voltage transient feature with the battery branch current, supercapacitor branch current, battery state of charge and supercapacitor state of charge to obtain the quasi-synchronous feature sequence.

[0020] Specifically, the acquired DC bus voltage signal undergoes multi-scale adaptive wavelet decomposition to obtain decomposition coefficients representing signal components in different frequency bands. To effectively suppress noise, the local variance of each decomposition coefficient is calculated within a set time window, and this local variance is compared with a preset significance threshold. This threshold is a pre-set threshold based on the statistical characteristics of the system's steady-state operating noise, used to distinguish between real signal components and noise. The threshold can be determined based on historical noise analysis, signal-to-noise ratio estimation calibration, and adaptively adjusted according to operating conditions. By classifying coefficients with local variance below the threshold as noise components and clearing them to zero, and retaining coefficients above the threshold as effective components reflecting dynamic load changes, the denoised voltage components are reconstructed. This design allows the noise suppression process to adapt to changes in system operating conditions, effectively improving signal quality while preserving true transient characteristics.

[0021] The dynamic changes of the voltage components after noise reduction are extracted to form voltage transient characteristics. To achieve coordinated analysis of multi-source data, this scheme performs strict time synchronization processing on the voltage transient characteristics with the collected battery current, supercapacitor current, and their states of charge. By interpolating and resampling each signal to a unified time reference, a spatiotemporally consistent quasi-synchronous characteristic sequence is generated.

[0022] Compared to traditional methods using fixed threshold filtering or general wavelet denoising, this application employs an adaptively set saliency threshold for noise suppression. This enables more precise separation of true load characteristics from noise, avoiding feature loss or noise residue issues caused by fixed thresholds when operating conditions change. This significantly improves the accuracy and adaptability of voltage signal feature extraction. Simultaneously, by employing strict time synchronization based on voltage transient characteristics, it overcomes the data mismatch problem caused by inconsistent sampling delays in multi-source signals, ensuring the temporal consistency of quasi-synchronous feature sequences. These two improvements collectively provide a highly reliable and accurate data foundation for load dynamic characteristic analysis, power optimization allocation, and smooth grid connection / disconnection switching.

[0023] In specific implementation, step S13 includes: Step S131: Calculate the mean of the voltage component after noise removal using a sliding window to obtain the DC fundamental component. Subtract the DC fundamental component from the voltage component after noise removal to obtain the voltage transient characteristic sequence. Step S132: Input the voltage transient characteristic sequence into the zero-phase filter, and perform forward filtering and reverse filtering to obtain the phase-compensated voltage transient characteristic sequence. Step S133: Using the timestamp sequence in the phase-compensated voltage transient characteristic sequence as the reference coordinate, the battery branch current, supercapacitor branch current, battery state of charge and supercapacitor state of charge are resampled by a linear interpolation algorithm and mapped to a unified sampling time. Step S134: The voltage transient characteristics, battery branch current, supercapacitor branch current, battery state of charge and supercapacitor state of charge mapped to the unified sampling time are merged into a matrix column to obtain a quasi-synchronous feature sequence.

[0024] Specifically, for the noise-removed voltage component obtained in step S12, a sliding window of length L is used to calculate its local mean. This mean sequence is the DC fundamental component, mainly representing the steady-state change of the voltage. Subtracting this DC fundamental component from the noise-removed voltage component yields the voltage transient characteristic sequence. This operation is essentially a high-pass filtering process, with its cutoff frequency determined by both the sliding window length L and the sampling frequency. The selection of L requires a trade-off: a window that is too short may lead to some low-frequency fluctuations being misjudged as transient features, while a window that is too long will weaken the ability to capture slowly changing disturbances. L is determined based on the period of the lowest frequency oscillation mode of interest in the system, ensuring that the frequency bands related to load dynamics can be effectively separated. The extracted voltage transient characteristic sequence filters out DC offset and power frequency fundamental, focusing on the rapid voltage fluctuations caused by load switching, power surges, and system oscillations. It is the core signal for analyzing the dynamic behavior of the system.

[0025] The sliding window mean filtering or any causal filter in step S131 inevitably introduces phase lag, causing the extracted transient features to shift in time. To obtain a feature sequence without phase distortion, this invention uses a zero-phase filter. The voltage transient feature sequence is first forward-filtered, then the filtering result is time-reversed, followed by a reverse filtering, and finally the time axis is reversed again. This process uses a filter with a symmetrical impulse response, thereby realizing a purely real transfer function in the frequency domain and eliminating phase distortion. After this step, the phase-compensated voltage transient feature sequence has peak and trough time labels that strictly correspond to the occurrence times of the original physical events, providing a prerequisite for using it as a time reference for multi-signal synchronization.

[0026] After obtaining the phase-compensated voltage transient characteristic sequence, the precise and uniform timestamp sequence carried by the phase-compensated voltage transient characteristic sequence is used as the sole time reference axis for the entire system. For the battery branch current, supercapacitor branch current, battery state of charge, and supercapacitor state of charge that need to be synchronized, their original data points usually have time labels inconsistent with the reference axis. To achieve synchronization, a linear interpolation algorithm is used to resample each signal to be synchronized. The specific process is as follows: for each target moment on the reference time axis, the algorithm locates the two nearest neighbor data points that exactly surround the target moment in the original data points of the signal arranged in chronological order; based on the time interval formed by these two points and the corresponding physical quantity values, it is assumed that the signal changes linearly within this small interval, and the estimated value of the physical quantity at the target moment is calculated through a linear proportional relationship. By performing this operation sequentially on all time points on the reference axis, all data points of each original signal can be mapped to a unified time coordinate, thereby generating a new data sequence that is perfectly aligned with the voltage transient characteristic sequence in time.

[0027] After completing the time alignment mapping of all signals, the system enters the data integration phase, namely matrix column merging. This process integrates multiple synchronized one-dimensional data sequences into a structured two-dimensional data matrix. Specifically, at each identical sampling moment, the corresponding transient voltage characteristic values, battery branch current values, supercapacitor branch current values, battery state of charge values, and supercapacitor state of charge values ​​are combined into a feature row vector in a predefined order. This vector fully characterizes the multidimensional state of the system at that instant. Following the chronological order, the feature row vectors corresponding to all sampling moments are arranged from top to bottom, ultimately forming a quasi-synchronous feature sequence matrix. In this matrix, each row represents a unified sampling moment, and each column corresponds to a specific physical quantity dimension. This regular matrix form provides ready-to-use, high-quality standardized data input for dynamic mode decomposition and advanced matrix operation-based algorithms. It should be noted that column merging refers to horizontally integrating physical quantities of different dimensions such as voltage and current at the same moment, while the quasi-synchronous feature sequence matrix is ​​constructed by vertically arranging the merged row vectors from each moment in chronological order.

[0028] By refining the extraction logic of the quasi-synchronous feature sequence, the consistency and accuracy of multi-source monitoring data during transient processes are significantly improved. The combination of sliding window mean calculation and zero-phase filtering effectively filters out DC fundamental interference while eliminating the unavoidable phase lag phenomenon in traditional filtering stages. This allows the extracted voltage transient features to reflect the true physical state of the system during disturbances in real time and without distortion, solving the technical problem of spurious drift of feature signals on the time axis. Furthermore, the resampling and matrix column merging mechanism based on the phase-compensated voltage sequence overcomes the data mismatch problem caused by clock asynchrony between different communication links and sampling units in energy storage power stations, ensuring that key parameters such as voltage, current, and state of charge are mathematically described at the same sampling time. This seamless time base alignment not only provides a prerequisite for the accurate construction of the snapshot matrix in step S2 but also guarantees the reliability of load dynamic mode identification from the source, enabling the system to make accurate adaptive control decisions within milliseconds during grid-connected / off-grid switching based on high-precision quasi-synchronous data.

[0029] In specific implementation, step S2 includes: Step S21: Extract the sampling vector from the quasi-synchronous feature sequence and construct a snapshot matrix according to the sampling time step. and the corresponding offset snapshot matrix ; Step S22, for the snapshot matrix Singular value decomposition yields the left singular matrix, the singular value diagonal matrix, and the right singular matrix, calculated using the following formula: ; in, Describes a left singular matrix. Represents a singular value diagonal matrix. Represents a right singular matrix; Step S23, truncate the first part of the singular value diagonal matrix. The dimensionality of the first-order eigenvalues ​​and their corresponding eigenvectors is reduced to obtain a low-order linear mapping matrix, which is calculated using the following formula: ; in, Represents a low-order linear mapping matrix. This indicates the left singular matrix before the cut-off point. The conjugate transpose of the order eigenvectors. This represents the truncated right singular matrix. This represents the inverse of the truncated singular value diagonal matrix; Step S24: Perform eigenvalue decomposition on the low-order linear mapping matrix to obtain eigenvalues ​​and corresponding eigenvectors; Step S25: Calculate the modal frequencies and real-time damping ratios of each order based on the eigenvalues ​​to obtain the load dynamic characteristic set. The calculation formula is as follows: ; ; in, Indicates modal frequency, This indicates that the eigenvalues ​​are obtained after eigenvalue decomposition of a low-order linear mapping matrix. Indicates the sign of the imaginary part. The symbol for the natural logarithm. Indicates the sampling time step. Indicates the real-time damping ratio. Indicates the sign of taking the real part. Indicates the absolute value symbol; The load dynamic characteristic set includes the dominant oscillation mode, real-time damping ratio, and mode frequency under the current load disturbance.

[0030] Specifically, in step S2, the present invention performs deep extraction of load dynamic features based on the quasi-synchronous feature sequence generated in step S1. The quasi-synchronous feature sequence is a structured data matrix, where each row corresponds to the values ​​of multiple physical quantities at the same sampling time, including voltage transient features, two-branch currents, and two states of charge. Extracting the sampling vectors involves directly reading each row of data in this matrix, with each row vector representing a snapshot of the complete state of the system at a certain moment. Following the order of data acquisition, m consecutive such row vectors are arranged as column vectors to construct a high-dimensional snapshot matrix. Simultaneously, the snapshot matrix is ​​shifted in the positive direction of the time axis by one sampling time step, i.e., the m row vectors from the second moment to the (m+1)th moment are also arranged in columns to obtain the corresponding shifted snapshot matrix. These two matrices are strictly aligned in time, forming the basic data pair for dynamic mode decomposition.

[0031] After constructing the snapshot matrix, singular value decomposition (SVD) is performed on it. SVD is a powerful matrix factorization method that decomposes a complex matrix into the product of three specific matrices: a unitary matrix containing left singular vectors, a diagonal matrix whose diagonal elements are singular values, and the conjugate transpose of a unitary matrix containing right singular vectors. By analyzing the singular value diagonal matrix, the dominant information components (corresponding to large singular values) and noise components (corresponding to small singular values) can be identified. To achieve a balance between computational efficiency and accuracy, the system only extracts the top k largest singular values ​​and their corresponding left and right singular vectors, thus realizing a dimensionality reduction mapping from the high-dimensional original space to the low-dimensional dominant feature subspace. A low-order linear mapping matrix is ​​then calculated, which compactly describes the dominant evolution of the system state between adjacent sampling times.

[0032] After obtaining the low-order linear mapping matrix, eigenvalue decomposition is further performed. Eigenvalue decomposition aims to reveal the intrinsic dynamic characteristics of this linear mapping matrix. By solving its characteristic equation, a set of eigenvalues ​​and their corresponding eigenvectors are obtained. Each eigenvalue is typically a complex number, containing information about the growth rate and oscillation frequency of the corresponding dynamic mode, while the eigenvector describes the shape or spatial distribution of the mode in the original state space. Based on the obtained complex eigenvalues, the specific modal frequencies of each oscillation mode determined by the imaginary part of the eigenvalues ​​and their real-time damping ratios determined by the ratio of the real part of the eigenvalues ​​to the modulus can be calculated through mathematical transformation. The real-time damping ratio is a key indicator for measuring the oscillation decay rate of the system and directly reflects the relative stability of the system. All these calculated dominant oscillation modes, real-time damping ratios, and modal frequencies under the current load disturbance together constitute the load dynamic feature set describing the dynamic behavior of the system under the current load disturbance.

[0033] By introducing a dynamic mode decomposition algorithm based on truncated singular value decomposition, accurate perception of the load characteristics of the energy storage system is achieved. Through the construction of snapshot and offset snapshot matrices, the algorithm can directly extract the linear mapping relationship of the system from the time-domain sampled data, overcoming the limitation of traditional methods in describing the dynamic evolution of nonlinear loads. By truncating the first k-order eigenvalues ​​for spatial dimensionality reduction, the system can retain more than 90% of the energy characteristics while eliminating high-frequency random noise and redundant data. This not only significantly reduces the computational dimension of the low-order linear mapping matrix, ensuring the real-time performance of the control logic during millisecond-level grid-connected / off-grid switching, but also enables the system to identify the disturbance component that poses the greatest threat to stability by extracting the dominant oscillation mode. Through the mapping calculation from complex eigenvalues ​​to the real-time damping ratio, this invention can quantitatively evaluate the system stability margin under the current load disturbance. This deep mode characteristic perception allows step S4 to specifically optimize the parameters of the linear active disturbance rejection controller, ensuring that the energy storage converter has sufficient damping support during mode switching, fundamentally suppressing voltage oscillations and frequency drops, and greatly improving the reliability of microgrid operation.

[0034] In specific implementation, step S24 involves eigenvalue decomposition of the low-order linear mapping matrix to obtain eigenvalues ​​and corresponding eigenvectors. The processing logic includes: Step S241: Perform eigenvalue decomposition on the low-order linear mapping matrix to calculate... Each discrete-time eigenvalue and its corresponding low-order eigenvector; Step S242: Obtain the left singular matrix obtained from the singular value decomposition in step S22. Multiply the left singular matrix with each low-order eigenvector to obtain the dynamic modal eigenvector. The calculation formula is as follows: ; in, Represents the dynamic modal feature vector. Represents a low-order eigenvector; Step S243: Establish a one-to-one correspondence between eigenvalues ​​and dynamic modal eigenvectors.

[0035] Specifically, the one-to-one correspondence between eigenvalues ​​and dynamic modal eigenvectors is established based on the inherent index consistency during the computation process. After eigenvalue decomposition of the low-order linear mapping matrix, the system obtains a set of sequentially arranged eigenvalues ​​and their corresponding low-order eigenvectors, where each eigenvalue and low-order eigenvector has the same index, determined by the formula... Each low-order feature vector is mapped to a high-dimensional space to obtain a dynamic modal feature vector. This mapping process preserves the index number, which is automatically inherited and retained because the dynamic modal feature vector is calculated from the low-order feature vector with the index number. This index alignment mechanism based on the original calculation sequence allows each feature value to be uniquely and logically bound to its corresponding restored feature vector through a shared index number, without requiring an additional association table.

[0036] The index-consistent correspondence mechanism avoids the construction of a complex mapping database, directly utilizing the index sequence during the computation process to complete data binding. This significantly reduces the memory overhead and computational latency of the adaptive control algorithm, enabling the energy storage converter to complete feature reconstruction and retrieval with extremely high real-time performance, meeting the stringent requirements for millisecond-level control response during grid-connected / off-grid switching transients. Furthermore, through this established one-to-one correspondence, the system can not only quantify the severity of load oscillations through feature values, but also accurately locate the distribution weight of these oscillations on physical channels such as bus voltage and battery current through associated feature vectors. This achieves high-fidelity source tracing of load disturbance characteristics, providing deterministic data guidance for parameter optimization of specific sensitive control loops in step S4, fundamentally enhancing the operational stability of the energy storage power station under complex operating conditions.

[0037] In specific implementation, step S3 includes: Step S31: Extract the maximum modal frequency from the load dynamic feature set as the dominant modal frequency, and calculate the current total power demand. Step S32: Calculate the initial power allocation command based on the relationship between the dominant mode frequency and the preset frequency division threshold. Step S33: Substitute the state of charge of the battery and the state of charge of the supercapacitor into the membership function for mapping, and calculate the corresponding power correction increment. The calculation formula is as follows: ; ; in, This represents the power correction increment of the supercapacitor. This indicates the power correction increment of the battery. Indicates the first correction gain factor. This represents the second correction gain factor. This represents the target center value of the state of charge of the supercapacitor. This represents the target center value of the battery's state of charge. Indicates the state of charge of a supercapacitor. Indicates the state of charge of the battery; Step S34, Determine Does it exceed the preset safe operating threshold? ; If either side If the limit is exceeded, a secondary power allocation correction is performed to obtain the final reference power command, the calculation formula of which is: ; ; in, This indicates the amount of power overflow on the over-limit side. This indicates the current reference power command for the over-limit side energy storage unit. This indicates the maximum power of the over-limit side energy storage unit. This indicates the final reference power command after correction for energy storage units that have not exceeded their limits. This indicates the original reference power command before correction by the energy storage unit on the non-over-limit side; Step S35: The initial power command, the power correction increment, and the final reference power command are added together to obtain the execution command and sent to the underlying controller.

[0038] Specifically, this application innovatively introduces a secondary power distribution correction, fully utilizing the complementary characteristics of batteries and supercapacitors in terms of energy density and power density within the hybrid energy storage system to maximize the system's regulation potential. In existing technologies, when the commanded power of a storage unit exceeds its safety threshold, a passive approach of direct limiting is typically used. While this protects the individual device, it inevitably creates a gap in the system's total power output, potentially leading to voltage instability during critical on-grid and off-grid transients. Furthermore, it fails to fully utilize the global regulation potential of the hybrid energy storage system. To address this deficiency, this solution innovatively constructs a closed-loop power rebalancing circuit based on safety constraints. Its core principle is that when any storage unit's power command exceeds its limit, the system does not simply truncate the command. Instead, it precisely calculates the power overflow of the exceeded portion and uses this overflow as a dynamic compensation command, adding it in real-time to the command of the un-exceeded storage unit on the other side. This process achieves an active and smooth transfer of power responsibility within the system. Compared to existing technologies, this solution not only avoids system instability caused by over-limit protection of a single unit, but also provides more continuous and smooth power support during the transient process of grid-connected and off-grid switching. Through this multi-level coordinated correction, the system achieves 100% response to total power demand while ensuring that the energy storage units do not exceed their limits, fundamentally improving the microgrid's resilience and operational safety under complex load disturbances.

[0039] The preset safe operating threshold is the critical physical boundary ensuring the safe and stable operation of the hybrid energy storage system hardware. Essentially, it is a dynamic power limit determined based on the electrochemical characteristics, charge / discharge rate capabilities, and rated capacity of the battery and supercapacitor, as well as the converter's rated capacity. This threshold is not a fixed, rigid constant, but rather the maximum allowable power value obtained by the system in real-time from the battery management system (BMS) and the supercapacitor control unit. During execution, the system accumulates the initial command calculated in step S32 and the correction increment calculated in step S33 in real-time to generate the pre-execution power, and compares its absolute value online with the limit value in the threshold register. The significance of the threshold is to provide a logical trigger criterion for the final reference power command. Once it is determined that the pre-execution power of a certain energy storage unit touches or exceeds this physical boundary, the system immediately triggers the power overflow calculation logic. This setting method ensures that each energy storage unit can operate safely within its optimal physical envelope, and simultaneously, by using this threshold as an energy dispatch switch, a smooth transition from single-unit forced limiting protection to system-level power collaborative compensation is achieved.

[0040] In specific implementation, in step S32, the initial power allocation command is calculated based on the magnitude relationship between the dominant mode frequency and the preset frequency division threshold. The processing logic includes: like The initial power command for the supercapacitor and the initial power command for the battery are calculated using the following formula: ; ; in, Indicates the dominant mode frequency. This indicates the preset frequency division threshold. This indicates the initial power command for the supercapacitor. This indicates the current total power demand. Represents the high-frequency power allocation factor. This indicates the initial power command for the battery; like Then let and will Set to zero.

[0041] Specifically, the preset frequency division threshold is a key physical critical point used to define the frequency attributes of load fluctuations. Essentially, it is a spectral boundary determined by the response bandwidth of the supercapacitor and the economical operating frequency band of the battery in the energy storage system. This threshold is typically selected with reference to the highest cutoff frequency at which the battery can operate stably for a long period without excessive temperature rise; for example, it is set between 5Hz and 10Hz in typical engineering applications. During program execution, the system compares the dominant mode frequency with this preset frequency division threshold in real time. When the dominant mode frequency is in the high-frequency range of the preset frequency division threshold, the system activates a hybrid allocation mechanism, assigning rapidly changing transient power components to the supercapacitor through a high-frequency power allocation coefficient, utilizing its high power density to buffer grid impacts, while the remaining low-frequency stable components are assigned to the battery. Conversely, when the dominant mode frequency itself is in the low-frequency range of the preset frequency division threshold, the system determines that the current load fluctuation is in a slow-evolutionary state, requiring no intervention from the supercapacitor, thus allocating all power demand to the battery and setting the supercapacitor command to zero to avoid unnecessary charging and discharging losses. The initial purpose of setting this frequency division threshold is to maximize the utilization of the physical characteristics of energy storage resources, ensure that high-frequency kinetic energy is undertaken by power-type units and low-frequency potential energy is undertaken by energy-type units, thereby achieving accurate return of power commands in the frequency domain.

[0042] This application uses the dominant mode frequency as a dynamic trigger switch, solving the problem of energy storage resource misconfiguration caused by rigid frequency division strategies in traditional control. The adaptive frequency division logic, implemented through a preset frequency division threshold, effectively avoids the risks of overcharging and over-discharging batteries under high-frequency, drastic fluctuations. It also physically suppresses the risk of active material shedding and thermal runaway from the battery, significantly extending the lifespan of the core assets of the energy storage power station.

[0043] Compared to existing technologies that simply use fixed filters for frequency division, this solution offers the advantage of frequency sensitivity: when load characteristics shift due to grid connection / disconnection or the switching of high-power nonlinear loads, the system can accurately identify the dominant oscillation component and adjust the initial distribution ratio in real time. This precise power reallocation ensures that the supercapacitor only outputs power during the most critical transient phases, effectively preventing the blind waste of its energy. It also guarantees the smoothness of the battery's output power, thus providing high-quality, low-ripple power support for the microgrid and significantly enhancing the system's frequency stability margin during transient processes.

[0044] In specific implementation, step S4 includes: Step S41: Obtain the real-time damping ratio and substitute it into the preset damping deviation evaluation function to calculate the damping deviation value. Step S42: Using the controller bandwidth of the linear active disturbance rejection controller as the decision variable, the damping deviation value and voltage fluctuation deviation are weighted and calculated to construct the objective function; Step S43: Initialize the spider monkey population and calculate the objective function value corresponding to the position of each individual spider monkey within the preset search space, and identify the global optimum and local optimum positions; Step S44: A damping compensation perturbation factor is introduced to correct the individual spider monkey position, updating the individual spider monkey position. The calculation formula is as follows: ; in, This indicates the updated location of the individual spider monkey. This indicates the location of the individual spider monkey before the update. Indicates the local optimal position. Indicates the globally optimal position. This represents a uniformly distributed random number within the interval [0,1]. This represents a uniformly distributed random number within the interval [-1, 1]. The disturbance factor representing damping compensation; Step S45: Determine whether the preset iteration stop condition is met; If satisfied, the parameter value corresponding to the global optimal position is output, and the parameter value corresponding to the global optimal position is determined as the optimal control parameter set of the linear active disturbance rejection controller. The bandwidth and gain of the underlying controller are updated during mode switching. The control parameters include the control gain, observer gain, and corresponding controller bandwidth of the linear active disturbance rejection controller; If the conditions are not met, return to step S44 to continue iterative updates.

[0045] Specifically, the pre-defined damping deviation evaluation function is used to quantitatively characterize the degree of deviation between the current damping state of the system and the ideal stability requirement. The pre-defined damping deviation evaluation function is constructed as follows: ; in, This represents the preset damping deviation evaluation function value. Indicates the first weighting coefficient. Indicates the target value of the ideal damping ratio and , Indicates the real-time damping ratio. This represents the second weighting coefficient. Represents an exponential function. This represents the reference value for the damping attenuation coefficient; The pre-defined search space is the set of legal values ​​for the parameters of a linear active disturbance rejection controller (ADRC). Essentially, it is a multi-dimensional hypercube region composed of parameters such as controller bandwidth, observer bandwidth, and control gain. Based on the switching frequency, sampling period, and maximum allowable current change rate of the underlying converter, physical upper and lower bounds for each parameter are pre-defined. For example, the upper limit of the controller bandwidth search is constrained to within one-tenth of the sampling frequency to prevent high-frequency noise amplification. The significance of setting this search space is to define a safety margin for the intelligent algorithm, ensuring that the parameters represented by the individual spider monkey positions participating in the iteration are practically feasible in engineering, thereby avoiding the generation of extreme parameter sets that could lead to hardware damage or control failure during the optimization process.

[0046] During grid-connection and off-grid switching, energy storage systems are highly susceptible to underdamped oscillations due to the dynamic characteristics of the load. By constructing this deviation evaluation function, key input variables can be provided for the objective function. This ensures that the spider-monkey optimization algorithm, when searching for the bandwidth parameters of the linear active disturbance rejection controller, not only considers the absolute deviation of voltage fluctuations but also prioritizes ensuring sufficient damping margin based on the essence of system dynamic stability. This fundamentally suppresses parasitic oscillations caused by load mode switching.

[0047] This application uses the acquired real-time damping ratio as feedback input. An improved spider-monkey algorithm is employed to search for the most suitable linear active disturbance rejection controller bandwidth within a preset parameter space, tailored to the current load characteristics. Unlike traditional fixed-signal compensation schemes, this invention obtains a set of optimal control parameters through optimization, including controller gain, observer gain, and controller bandwidth. When the system detects a grid-connected / off-grid switching command, instead of superimposing a correction signal at the voltage reference, the register parameters of the underlying controller are directly rewritten online. This parameter update control strategy alters the frequency domain characteristics and stability margin of the control system itself, thereby fundamentally solving the problem of frequent voltage oscillations caused by sudden changes in system impedance during grid-connected / off-grid switching.

[0048] In specific implementation, in step S42, the controller bandwidth of the linear active disturbance rejection controller is used as the decision variable. The damping deviation value and voltage fluctuation deviation are weighted and calculated to construct the objective function. The processing logic includes: ; in, Represents the objective function value. This represents the first weighting coefficient. This represents the second weighting coefficient. Indicates the range from 0 to Integrate over the time period region. Indicates time period, This represents the absolute value of the deviation of the control quantity. Indicates the reference damping ratio. Indicates the real-time damping ratio. Represents the differential symbol.

[0049] In specific implementation, step S5 includes: Step S51: Obtain the real-time damping ratio, dominant mode frequency, and voltage and current on the AC and DC sides of the energy storage converter from the load dynamic characteristic set, and calculate the current power balance state. Step S52: Input the real-time damping ratio, dominant mode frequency and power balance state into the preset logic discrimination matrix, calculate the matching degree between the current operating condition and the preset mode switching threshold, and generate a mode switching command based on the matching degree. Step S53: When the mode switching command is a grid-connected to off-grid command, control the circuit breaker at the grid connection point to open, extract the optimal control parameter group of the linear active disturbance rejection controller determined in step S45, update the bandwidth and gain of the underlying controller, adjust the bandwidth of the underlying controller to the optimal controller bandwidth, and switch to voltage source control mode. Step S54: When the mode switching command is an off-grid to grid-connected command, the output voltage of the energy storage converter is adjusted by the pre-synchronization control algorithm so that the phase and frequency of the output voltage are consistent with the external power grid. When the preset closing conditions are met, the circuit breaker at the grid connection point is closed, and the current source control mode is switched. Step S55: Output pulse width modulation signal to energy storage converter according to the switched control mode, and perform grid-connected adaptive switching action.

[0050] Specifically, the current power balance calculation method is based on real-time energy accounting of the energy storage system according to the law of conservation of energy. Specifically, the system collects the battery output power on the DC side of the energy storage converter and the effective power output to the load on the AC side. The output power is obtained by multiplying the DC voltage and current. The current power balance calculation formula is as follows: ; in, This indicates the current power balance state. This indicates the battery output power on the DC side of the energy storage converter. This indicates the conversion efficiency of the energy storage converter. This indicates the effective power output from the AC side of the energy storage converter to the load; This indicator directly reflects the dynamic difference between the internal energy support capacity of the energy storage station and the external load demand, and is a prerequisite for determining whether the system has the physical capability to support and switch off-grid.

[0051] The pre-defined logical discrimination matrix is ​​the core of the intelligent decision-making for the entire adaptive switching strategy. It takes information from three key dimensions—real-time damping ratio, dominant mode frequency, and power balance state—as input, and assigns a series of pre-defined thresholds to each dimension. For example, the real-time damping ratio must be higher than the minimum safe threshold for maintaining stable off-grid operation; the dominant mode frequency should not be in a dangerous frequency band that could trigger resonance; and the power balance state must meet the minimum energy self-consistency requirements for off-grid operation. These criteria are combined into tightly packed logical rules, forming a multi-condition collaborative discrimination matrix. This upgrades the traditional single criterion of relying solely on the presence or absence of grid voltage to an intelligent decision-making process that comprehensively assesses the system's internal stability and external supply and demand relationship. It effectively prevents blind switching when the system's own damping is insufficient, there is a risk of oscillation, or power cannot be self-sufficient, thus fundamentally avoiding potential voltage collapse or equipment overload during the switching process and ensuring the absolute safety and reliability of the switching action.

[0052] Within the logical discrimination matrix, the matching degree between the current operating condition and the preset mode switching threshold is calculated. This matching degree is a quantitative value used to evaluate the degree of agreement between the real-time acquired operating condition data and the preset ideal switching conditions in the matrix. Its calculation is typically based on a fuzzy inference mechanism. For example, the system compares the real-time damping ratio with the fuzzy set of "sufficient damping," obtaining a matching degree score between 0 and 1 through a membership function. Similarly, a similar quantitative evaluation is performed on the dominant mode frequency and power balance state. Based on the importance of each criterion in the switching decision, different weights are assigned, and a weighted comprehensive calculation is performed to finally obtain a total matching degree score. When this total score exceeds the preset mode switching threshold, the switching conditions are deemed met, and a corresponding grid-connected to off-grid or off-grid to grid-connected instruction is generated. This method based on multi-dimensional matching degree calculation makes the switching decision no longer a simple Boolean judgment, but a continuous evaluation process that can delicately reflect the actual operating conditions of the system, thereby achieving optimal selection of switching timing and accurate generation of switching instructions.

[0053] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0054] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. A grid-connected adaptive control method for energy storage power stations based on load characteristic analysis, characterized in that, Includes the following steps: Step S1: Collect the operating mode data, preprocess the operating mode data, and obtain the quasi-synchronous feature sequence; Step S2: Construct a snapshot matrix based on the quasi-synchronous feature sequence, and calculate the load dynamic feature set using the dynamic mode decomposition algorithm; Step S3: Calculate the power allocation command based on the load dynamic feature set to obtain the initial power allocation command, and perform secondary power allocation correction by combining the battery state of charge and the supercapacitor state of charge. Step S4: The improved spider monkey algorithm is used to optimize the control parameters of the linear active disturbance rejection controller and calculate the optimal control parameter set of the linear active disturbance rejection controller. Step S5: Calculate the current power balance state and combine it with the load dynamic feature set to generate a mode switching command. Based on the mode switching command, control the energy storage converter to execute and perform adaptive off-grid switching.

2. The adaptive control method for grid connection and disconnection of an energy storage power station based on load characteristic analysis as described in claim 1, characterized in that, Step S1 specifically includes: Step S11: Collect operating mode data through sensors. The operating mode data includes the DC bus voltage signal of the energy storage system, the battery branch current, the supercapacitor branch current, the battery state of charge, and the supercapacitor state of charge. Step S12: The DC bus voltage signal is decomposed using a multi-scale adaptive wavelet algorithm to obtain the decomposition coefficients of each layer. The local variance of each layer decomposition coefficient within the sampling window is calculated and compared with a preset significance threshold. Decomposition coefficients with local variance less than the preset significance threshold are identified as noise components and cleared to zero. The remaining decomposition coefficients are used as feature components and inverse discrete wavelet transform is performed to obtain the voltage components after noise removal. Step S13: Extract the dynamic increment in the voltage component after noise removal as the voltage transient feature, and perform time reference alignment processing on the voltage transient feature with the battery branch current, supercapacitor branch current, battery state of charge and supercapacitor state of charge to obtain the quasi-synchronous feature sequence.

3. The adaptive control method for grid connection and disconnection of an energy storage power station based on load characteristic analysis as described in claim 2, characterized in that, Step S13 specifically includes: Step S131: Calculate the mean of the voltage component after noise removal using a sliding window to obtain the DC fundamental component. Subtract the DC fundamental component from the voltage component after noise removal to obtain the voltage transient characteristic sequence. Step S132: Input the voltage transient characteristic sequence into the zero-phase filter, and perform forward filtering and reverse filtering to obtain the phase-compensated voltage transient characteristic sequence. Step S133: Using the timestamp sequence in the phase-compensated voltage transient characteristic sequence as the reference coordinate, the battery branch current, supercapacitor branch current, battery state of charge and supercapacitor state of charge are resampled by a linear interpolation algorithm and mapped to a unified sampling time. Step S134: The voltage transient characteristics, battery branch current, supercapacitor branch current, battery state of charge and supercapacitor state of charge mapped to the unified sampling time are merged into a matrix column to obtain a quasi-synchronous feature sequence.

4. The adaptive control method for grid connection and disconnection of an energy storage power station based on load characteristic analysis as described in claim 1, characterized in that, Step S2 specifically includes: Step S21: Extract the sampling vector from the quasi-synchronous feature sequence and construct a snapshot matrix according to the sampling time step. and the corresponding offset snapshot matrix ; Step S22, for the snapshot matrix Singular value decomposition yields the left singular matrix, the singular value diagonal matrix, and the right singular matrix, calculated using the following formula: ； in, Describes a left singular matrix. Represents a singular value diagonal matrix. Represents a right singular matrix; Step S23, truncate the first part of the singular value diagonal matrix. The dimensionality of the first-order eigenvalues ​​and their corresponding eigenvectors is reduced to obtain a low-order linear mapping matrix, which is calculated using the following formula: ； in, Represents a low-order linear mapping matrix. This indicates the left singular matrix before the cut-off. The conjugate transpose of the order eigenvectors. This represents the truncated right singular matrix. This represents the inverse of the truncated singular value diagonal matrix; Step S24: Perform eigenvalue decomposition on the low-order linear mapping matrix to obtain eigenvalues ​​and corresponding eigenvectors; Step S25: Calculate the modal frequencies and real-time damping ratios of each order based on the eigenvalues ​​to obtain the load dynamic characteristic set. The calculation formula is as follows: ； ； in, Indicates modal frequency, This indicates that the eigenvalues ​​are obtained after eigenvalue decomposition of a low-order linear mapping matrix. Indicates the sign of the imaginary part. The symbol for the natural logarithm. Indicates the sampling time step. Indicates the real-time damping ratio. Indicates the sign of taking the real part. Indicates the absolute value symbol; The load dynamic feature set includes the dominant oscillation mode, real-time damping ratio, and mode frequency under the current load disturbance.

5. The adaptive control method for grid connection and disconnection of an energy storage power station based on load characteristic analysis as described in claim 4, characterized in that, In step S24, eigenvalue decomposition is performed on the low-order linear mapping matrix to obtain eigenvalues ​​and corresponding eigenvectors. The processing logic includes: Step S241: Perform eigenvalue decomposition on the low-order linear mapping matrix to calculate... Each discrete-time eigenvalue and its corresponding low-order eigenvector; Step S242: Obtain the left singular matrix obtained from the singular value decomposition in step S22. Multiply the left singular matrix with each low-order eigenvector to obtain the dynamic modal eigenvector. The calculation formula is as follows: ； in, Represents the dynamic modal feature vector. Represents a low-order eigenvector; Step S243: Establish a one-to-one correspondence between eigenvalues ​​and dynamic modal eigenvectors.

6. The adaptive control method for grid connection and disconnection of an energy storage power station based on load characteristic analysis as described in claim 1, characterized in that, Step S3 specifically includes: Step S31: Extract the maximum modal frequency from the load dynamic feature set as the dominant modal frequency, and calculate the current total power demand. Step S32: Calculate the initial power allocation command based on the relationship between the dominant mode frequency and the preset frequency division threshold. Step S33: Substitute the state of charge of the battery and the state of charge of the supercapacitor into the membership function for mapping, and calculate the corresponding power correction increment. The calculation formula is as follows: ； ； in, This represents the power correction increment of the supercapacitor. This indicates the power correction increment of the battery. Indicates the first correction gain factor. This represents the second correction gain factor. This represents the target center value of the state of charge of the supercapacitor. This represents the target center value of the battery's state of charge. Indicates the state of charge of a supercapacitor. Indicates the state of charge of the battery; Step S34, Determine Does it exceed the preset safe operating threshold? ; If either side If the limit is exceeded, a secondary power allocation correction is performed to obtain the final reference power command, the calculation formula of which is: ； ； in, This indicates the amount of power overflow on the over-limit side. This indicates the current reference power command for the over-limit side energy storage unit. This indicates the maximum power of the over-limit side energy storage unit. This indicates the final reference power command after correction for energy storage units that have not exceeded their limits. This indicates the original reference power command before correction by the energy storage unit on the non-over-limit side; Step S35: The initial power command, the power correction increment, and the final reference power command are added together to obtain the execution command and sent to the underlying controller.

7. The adaptive control method for grid connection and disconnection of an energy storage power station based on load characteristic analysis as described in claim 6, characterized in that, In step S32, the initial power allocation command is calculated based on the magnitude relationship between the dominant mode frequency and the preset frequency division threshold. The processing logic includes: like The initial power command for the supercapacitor and the initial power command for the battery are calculated using the following formula: ； ； in, Indicates the dominant mode frequency. This indicates the preset frequency division threshold. This indicates the initial power command for the supercapacitor. This indicates the current total power demand. Represents the high-frequency power allocation factor. This indicates the initial power command for the battery; like Then let and will Set to zero.

8. The adaptive control method for grid connection and disconnection of an energy storage power station based on load characteristic analysis as described in claim 1, characterized in that, Step S4 specifically includes: Step S41: Obtain the real-time damping ratio and substitute it into the preset damping deviation evaluation function to calculate the damping deviation value. Step S42: Using the controller bandwidth of the linear active disturbance rejection controller as the decision variable, the damping deviation value and voltage fluctuation deviation are weighted and calculated to construct the objective function; Step S43: Initialize the spider monkey population and calculate the objective function value corresponding to the position of each individual spider monkey within the preset search space, and identify the global optimum and local optimum positions; Step S44: A damping compensation perturbation factor is introduced to correct the individual spider monkey position, updating the individual spider monkey position. The calculation formula is as follows: ； in, This indicates the updated location of the individual spider monkey. This indicates the location of the individual spider monkey before the update. Indicates the local optimal position. Indicates the globally optimal position. This represents a uniformly distributed random number within the interval [0,1]. This represents a uniformly distributed random number within the interval [-1, 1]. The disturbance factor representing damping compensation; Step S45: Determine whether the preset iteration stop condition is met; If satisfied, the parameter value corresponding to the global optimal position is output, and the parameter value corresponding to the global optimal position is determined as the optimal control parameter set of the linear active disturbance rejection controller. The bandwidth and gain of the underlying controller are updated during mode switching. The control parameters include the control gain of the linear active disturbance rejection controller, the observer gain, and the corresponding controller bandwidth; If the conditions are not met, return to step S44 to continue iterative updates.

9. The adaptive control method for grid connection and disconnection of an energy storage power station based on load characteristic analysis as described in claim 8, characterized in that, In step S42, the controller bandwidth of the linear active disturbance rejection controller is used as the decision variable. The damping deviation value and voltage fluctuation deviation are weighted and calculated to construct the objective function. The processing logic includes: ； in, Represents the objective function value. This represents the first weighting coefficient. This represents the second weighting coefficient. Indicates the range from 0 to Integrate over the time period region. Indicates time period, This represents the absolute value of the deviation of the control quantity. Indicates the reference damping ratio. Indicates the real-time damping ratio. Represents the differential symbol.

10. The adaptive control method for grid connection and disconnection of an energy storage power station based on load characteristic analysis as described in claim 1, characterized in that, Step S5 specifically includes: Step S51: Obtain the real-time damping ratio, dominant mode frequency, and voltage and current on the AC and DC sides of the energy storage converter from the load dynamic characteristic set, and calculate the current power balance state. Step S52: Input the real-time damping ratio, dominant mode frequency and power balance state into the preset logic discrimination matrix, calculate the matching degree between the current operating condition and the preset mode switching threshold, and generate a mode switching command based on the matching degree. Step S53: When the mode switching command is a grid-connected to off-grid command, control the circuit breaker at the grid connection point to open, extract the optimal control parameter group of the linear active disturbance rejection controller determined in step S45, update the bandwidth and gain of the underlying controller, adjust the bandwidth of the underlying controller to the optimal controller bandwidth, and switch to voltage source control mode. Step S54: When the mode switching command is an off-grid to grid-connected command, the output voltage of the energy storage converter is adjusted by the pre-synchronization control algorithm so that the phase and frequency of the output voltage are consistent with the external power grid. When the preset closing conditions are met, the circuit breaker at the grid connection point is closed, and the current source control mode is switched. Step S55: Output pulse width modulation signal to energy storage converter according to the switched control mode, and perform grid-connected adaptive switching action.

Images