Adaptive sliding mode control method for black start fault tolerance in grid-type wind-storage systems

Abstract

This invention discloses an adaptive sliding mode control method for black-start fault tolerance in a grid-type wind-storage system, belonging to the field of wind-storage system control technology. It solves the problem of existing methods failing to incorporate the actual state parameters of the grid-type wind-storage system for initial state estimation, leading to biased initial state estimation. The method includes preprocessing the wind-storage system state parameters, performing initial state estimation of the wind-storage system state parameters, and real-time estimation of unmeasurable states, external disturbances, and sensor faults in the wind-storage system. The adaptive sliding mode control model identifies faults in the real-time estimation set based on a dynamic event triggering mechanism. In this invention, by preprocessing the wind-storage system state parameters, multi-source heterogeneous actual state parameters can be integrated. Furthermore, based on the black-start state space model combined with the preprocessed multi-source heterogeneous state parameters, the initial state variables are accurately solved, providing a reliable benchmark for the subsequent adaptive sliding mode control strategy with high-precision initial state estimation.

Description

Technical Field

[0001] This invention belongs to the field of wind and energy storage system control technology, specifically relating to an adaptive sliding mode control method for black-start fault tolerance in grid-type wind and energy storage systems. Background Technology

[0002] As the core infrastructure for energy supply, the safe and stable operation of the power grid is directly related to the guarantee of power supply. When encountering severe faults (such as large-scale power outages, natural disasters, or system shutdowns caused by human sabotage), it is necessary to quickly restore the power grid's power supply capacity through black start technology. This means using the self-starting power sources within the system to gradually start generator units and rebuild the grid structure without relying on external power sources, ultimately achieving full grid restoration.

[0003] Traditional large power grids rely primarily on synchronous generator sets (such as hydro turbines and gas turbines) for black start, which provide stable voltage and frequency support to the grid through the synergistic effect of mechanical inertia, speed governors, and excitation systems. However, with the accelerated transformation of the energy structure, the installed capacity of new energy sources, represented by wind power, continues to rise. New energy units (especially direct-drive / semi-direct-drive wind turbines) are typically connected to the grid via power electronic converters, which lack the inertia and autonomous voltage / frequency regulation capabilities of traditional synchronous machines. In the initial stage of black start, new energy units experience high system equivalent impedance and large load fluctuations. New energy converters are prone to oscillations or even grid disconnection due to control parameter mismatch or external disturbances, resulting in poor adaptability to weak grids. To address the new demands for grid security under the high penetration rate of new energy, grid-forming (GFM) converter control technology has emerged. Unlike traditional grid-following converters (which only track grid voltage / frequency), grid-based control, by simulating the dynamic characteristics of synchronous generators (such as virtual inertia, droop control, and damping regulation), enables renewable energy units to actively provide voltage and frequency references. In renewable energy power plants, wind-storage integrated systems are a typical application of grid-based technology. They can serve as the "first power source" to independently establish the voltage and frequency reference of the local power grid, and gradually connect to other renewable energy units or loads, avoiding dependence on traditional synchronous generators.

[0004] Chinese patent CN119315649A discloses a black-start power control method for a grid-connected wind-storage system. The method includes: taking the power change of the grid-connected photovoltaic system during the black-start process as an uncertain disturbance to the grid-connected wind-storage system, and obtaining a black-start prediction model for the system; constructing a disturbance observer state equation based on the black-start prediction model; establishing a robust model predictive control optimization problem based on a preset objective function and constraints, using the disturbance observer state equation and the black-start prediction model; and solving the robust model predictive control optimization problem to obtain the black-start power control result of the grid-connected wind-storage system. However, existing methods, when determining the estimated value based on the disturbance observer state equation and the black-start prediction model, do not combine the actual state parameters of the grid-connected wind-storage system for initial state estimation. This makes the black-start prediction model unable to accurately reflect the nonlinear dynamic characteristics of the wind-storage system, leading to deviations in the initial state estimation. This causes subsequent control strategies to deviate from actual requirements, prolonging system recovery time or triggering voltage / frequency overshoot. To address these issues, we propose an adaptive sliding mode control method for black-start fault tolerance in grid-connected wind-storage systems. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing an adaptive sliding mode control method for black-start fault tolerance in grid-type wind-storage systems. This method solves the problem that existing methods, when determining estimated values ​​based on the disturbance observer state equation and the black-start prediction model, fail to incorporate the actual state parameters of the grid-type wind-storage system for initial state estimation. As a result, the black-start prediction model cannot accurately reflect the nonlinear dynamic characteristics of the wind-storage system, leading to deviations in the initial state estimation.

[0006] This invention is implemented as follows: an adaptive sliding mode control method for black-start fault tolerance in a grid-type wind-storage system, the method comprising: The system acquires the state parameters of the wind-storage system in real time, preprocesses the state parameters, and performs initial state estimation on the preprocessed state parameters based on the black-start state-space model. Based on the adaptive sliding mode observer, the system performs real-time estimation of the unmeasurable state, external disturbances, and sensor failures of the wind storage system based on the initial state estimation, and outputs a real-time estimation set. Historical power output data of the wind-storage system's generating equipment is captured as modeling sample data. The modeling sample data is used to train an adaptive sliding mode control model based on a dynamic event triggering mechanism, and a converged adaptive sliding mode control model is output. The adaptive sliding mode control model obtains a real-time estimation set, performs fault identification on the real-time estimation set based on a dynamic event triggering mechanism, and determines the adaptive sliding mode control strategy by combining the real-time estimation set.

[0007] Obtain the adaptive sliding mode control strategy, use the adaptive sliding mode control strategy as a constraint, and adjust the parameters of the adaptive sliding mode control strategy through the Lyapunov stability criterion to generate an adaptive sliding mode optimization strategy; The adaptive sliding mode optimization strategy is distributed to the local edge controller of the wind storage system, and the local edge controller realizes adaptive black start based on the adaptive sliding mode optimization strategy.

[0008] Preferably, the method for preprocessing the state parameters of the wind-storage system includes: Obtain the state parameters of the multi-source heterogeneous wind and storage system, perform outlier removal on the state parameters of the wind and storage system, and output the state parameters of the wind and storage system after outlier removal. After loading outlier-removed wind-storage system state parameters, using PTP 1588 time clock as a unified time reference, zero-order hold-linear extrapolation hybrid interpolation is performed on the wind-storage system state parameters to obtain timestamp-aligned wind-storage system state parameters; The system obtains the wind-storage system state parameters aligned with timestamps. The system state parameters are then low-pass filtered using a second-order low-pass filter to obtain an initial filtered signal. The initial filtered signal is then subjected to a discrete Fourier transform to obtain an initial frequency domain signal. The initial frequency domain signal is then filtered a second time using an inverse Fourier transform combined with a dynamic adjustment method for the filter coefficients to obtain a second-frequency signal after the second filtering. The secondary frequency signal after secondary filtering is obtained, and the dimensionless preprocessing of the secondary frequency signal is performed based on the maximum-minimum method. The preprocessed wind-storage system state parameters are then output.

[0009] Preferably, the method for secondary filtering of the initial frequency domain signal based on inverse Fourier transform combined with dynamic adjustment of filter coefficients includes: Load the initial frequency domain signal and the corresponding wind storage system state parameters, determine the preset noise reduction amount of the initial frequency domain signal based on the parameter type, and weight the initial frequency domain signal in combination with the preset noise reduction amount to obtain the frequency domain weighted value; Obtain the frequency domain weighting value of the initial frequency domain signal, and perform inverse Fourier transform on the frequency domain weighting value to obtain the weighted frequency domain signal; Identify the parameter type of the wind-storage system state parameters corresponding to the initial frequency domain signal, map the parameter change rate of the parameter type to the cutoff frequency of the low-pass filter, and adjust the initial cutoff frequency of the low-pass filter to obtain the cutoff adjustment frequency. The damping ratio coefficient of the low-pass filter is dynamically adjusted based on the parameter weight coefficients of the parameter type. The cutoff adjustment frequency and the adjusted damping ratio coefficient are obtained and updated synchronously to the low-pass filter. At the same time, the transfer function of the low-pass filter is updated. The low-pass filter performs secondary filtering on the weighted frequency domain signal based on the updated transfer function, and outputs a secondary frequency signal after secondary filtering.

[0010] Preferably, the black-start state-space model is represented as: in, This represents the initial state estimate. These represent the state transfer matrix, the Jacobian dynamic matrix between state variables, the additive fault input matrix, and the disturbance input matrix, respectively. Represents the parameter perturbation matrix. For the initial state variables, These are the control input vector, sensor fault vector, and external disturbance term, respectively. Indicates measurement noise. For observer gain, This represents the model's output vector.

[0011] Preferably, the method for initial state estimation of the preprocessed wind-storage system state parameters based on the black-start state-space model includes: Obtain the preprocessed state parameters of the wind-storage system and extract the steady-state mean of the state parameters of the wind-storage system; Load the black-start state-space model, use the steady-state mean of the wind-storage system state parameters as the input representation of the preprocessed wind-storage system state parameters, combine the black-start state-space model to solve for the initial values ​​of the state variables, and identify the steady-state signal points corresponding to the steady-state mean. A small-signal linear model characterizing the linear relationship between the steady-state mean and the steady-state signal points is constructed by using the linear relationship between the steady-state mean and the steady-state signal points. The input representation of the state parameters of the wind-storage system is calibrated based on a small-signal linear model, and the calibrated initial state variables are output. The calibrated initial state variables are matched and mapped to the black-start state space model. The black-start state space model is solved using the gradient descent method to obtain the optimal state estimate, which is then used as the initial state estimate.

[0012] Preferably, the method for real-time estimation of unmeasurable states, external disturbances, and sensor failures of the wind-storage system based on an adaptive sliding mode observer and initial state estimation includes: The initial state estimate is loaded, and the adaptive sliding mode observer dynamically updates the initial state estimate based on a preset adaptive law mechanism. Based on the updated initial state estimation, the uncertainty of the black-start state space model is corrected. The unmeasurable state is added to the state vector of the observer as an extended state variable. The unmeasurable state is estimated based on the adaptive sliding mode observer combined with the adaptive law mechanism to obtain the unmeasurable state estimate. External disturbance variables are used as unknown inputs to the wind-storage system. The external disturbance variables are separated by the disturbance observation channel of the adaptive sliding mode observer, and the estimated value of the external disturbance is output. Based on the adaptive sliding mode observer, the sensor deviation between the theoretical estimate and the actual measurement of the sensor is compared and it is determined whether the sensor deviation exceeds the preset deviation threshold. When the sensor deviation exceeds the preset deviation threshold, the corresponding sensor is determined to be faulty. Based on the residual characteristics, the fault type is distinguished and the fault type and sensor deviation are marked. The estimates of unmeasurable states, external disturbances, fault types, and sensing biases are integrated into a real-time estimation set.

[0013] Preferably, the adaptive sliding mode control model is based on an LSTM neural network architecture and includes an input layer and an output layer. A multilayer perceptron (MLP) layer is introduced between the LSTM neural network and the input layer. The MLP layer uses statistical feature analysis to identify faults in the real-time estimation set, detects and locates the fault identification results. An adaptive controller based on fault-tolerant control is set between the MLP layer and the LSTM neural network. The adaptive controller dynamically adjusts the sliding surface parameters and switching control law based on the fault-tolerant control and the fault identification results. A dynamic event triggering mechanism is introduced into the LSTM neural network. The output of the adaptive controller is sparsified based on the dynamic event triggering mechanism, and a sparse control strategy is output to reduce communication and computation load. The LSTM neural network balances the weights of the unmeasurable state estimates, external disturbance estimates, and sensor faults in the real-time estimation set based on the Floyd-Warshall algorithm. The sparse control strategy is dynamically adjusted based on the balanced weights of the unmeasurable state estimates, external disturbance estimates, and sensor faults, and an adaptive sliding mode control strategy is output.

[0014] Preferably, the adaptive sliding mode control model uses a dynamic event triggering mechanism to perform fault identification on the real-time estimation set, comprising: The real-time estimation set is acquired, and the multilayer perceptron (MLP) layer, combined with statistical feature analysis, is used to identify faults in the real-time estimation set, detect and locate the fault identification results. The adaptive controller dynamically adjusts the sliding surface parameters and switching control law based on fault-tolerant control and fault identification results. Based on the dynamic event triggering mechanism, the output of the adaptive controller is sparsified, and a sparsified control strategy is output. The Floyd-Warshall algorithm is used to balance the weights of the unmeasurable state estimates, external disturbance estimates, and sensor faults in the real-time estimation set. Based on the balanced weights of the unmeasurable state estimates, external disturbance estimates, and sensor faults, the sparse control strategy is dynamically adjusted to output an adaptive sliding mode control strategy.

[0015] Compared with the prior art, the embodiments of this application have the following main advantages: In this embodiment of the invention, by preprocessing the state parameters of the wind-storage system, multi-source heterogeneous actual state parameters can be integrated, thereby covering the full-dimensional dynamics of the wind-storage system. Furthermore, based on the black-start state space model combined with the preprocessed multi-source heterogeneous state parameters, the initial state variables are accurately solved, so that the high-precision initial state estimation provides a reliable benchmark for the subsequent adaptive sliding mode control strategy, avoiding control command mismatch caused by initial deviation.

[0016] In this embodiment of the invention, when preprocessing the state parameters of the wind-storage system, high-frequency noise is first removed through frequency domain analysis and dynamic filter parameter adjustment, while retaining key dynamic information of the system. The initial filtered signal is then subjected to discrete Fourier transform to obtain the initial frequency domain signal. A second filtering is then performed by inverse Fourier transform combined with dynamic adjustment of filter coefficients. This allows the invention to dynamically adjust the filter parameters according to the characteristics of different state parameters, ensuring that the most effective dynamic features can be extracted from all types of signals, providing high-quality input for subsequent initial state estimation and fault identification.

[0017] In this embodiment of the invention, targeted weighted noise reduction is performed on residual noise in specific frequency bands for different parameter types. This avoids the loss of effective signal caused by the one-size-fits-all approach of traditional fixed filtering, preserves the key dynamic characteristics of the system, and restores the frequency domain optimization results to the time domain signal. This provides a directly processable input for the dynamic parameter adjustment of the subsequent low-pass filter, ensuring the continuity of the filtering process. The cutoff frequency is adjusted according to the actual rate of change of the parameters, which not only preserves the effective dynamic information of rapidly changing parameters but also suppresses high-frequency noise of slowly changing parameters. This avoids the over-smoothing or under-filtering problems caused by a fixed cutoff frequency. Finally, the low-pass filter performs secondary filtering on the weighted frequency domain signal based on the updated transfer function, outputting a secondary frequency signal after secondary filtering. This results in a secondary frequency signal with both low noise and high dynamic fidelity, providing a reliable input for accurate analysis of the black start process.

[0018] In this embodiment of the invention, when performing initial state estimation of the preprocessed wind-storage system state parameters based on the black-start state-space model, the actual state parameters are closely associated with the black-start state-space model by matching the steady-state mean with the steady-state signal points of the model, thus avoiding the problem of the model deviating from reality. The small-signal linear model extracts the local dynamic characteristics of the system near the equilibrium point, thereby providing key dynamic information for the initial state estimation and improving the pertinence of the estimation. The gradient descent method finds the combination of state variables that minimizes the error between the model and the actual data through global search, ensuring that the initial state estimation result is optimal.

[0019] In this embodiment of the invention, when the adaptive sliding mode observer performs real-time estimation of the unmeasurable state, external disturbances, and sensor faults of the wind-storage system based on the initial state estimation, the real-time estimation set unifies the key information of the unmeasurable state, disturbances, and faults, providing comprehensive system state awareness for the adaptive sliding mode control strategy. The adaptive sliding mode observer incorporates unmeasurable states such as load impedance and line impedance into the observer, and dynamically adjusts the estimation accuracy through adaptive laws, solving the control mismatch problem caused by ignoring unmeasurable states in traditional methods. Furthermore, it separates external disturbances such as load mutations and short-circuit faults through the disturbance observation channel, providing a basis for disturbance compensation for the control strategy and improving the system's anti-interference capability. Finally, the unified real-time estimation set integrates the key data of unmeasurable states, disturbances, and faults, providing a comprehensive perception foundation for the dynamic optimization of the adaptive sliding mode control strategy.

[0020] In this embodiment of the invention, the adaptive sliding mode control model is based on an LSTM neural network architecture. A multilayer perceptron (MLP) layer is introduced between the LSTM neural network and the input layer, and statistical feature analysis is used to identify faults in the real-time estimation set. The MLP layer can extract and analyze features from multi-source data in the real-time estimation set, detect and locate multiple types of faults. Based on different fault types and system states, the adaptive controller can adjust the sliding surface parameters and switch control laws in real time, ensuring that the control strategy always matches the actual needs of the system, improving the accuracy and effectiveness of control. Furthermore, through the weight balancing of the Floyd-Warshall algorithm, the weights of each factor in the control strategy can be precisely adjusted according to the dynamic changes of factors in the real-time estimation set, making the control strategy more consistent with the actual needs of the system, further improving the accuracy and effectiveness of control. Simultaneously, the adaptive sliding mode control model possesses strong fault tolerance capabilities, enabling rapid fault location and corresponding fault-tolerant control measures after fault detection to compensate for the impact of the fault on the system. Moreover, by dynamically adjusting the control strategy, the system can automatically resume normal operation after fault elimination, exhibiting strong self-healing capabilities and ensuring the long-term stable operation of the system. Attached Figure Description

[0021] Figure 1 A schematic diagram of the implementation process of the adaptive sliding mode control method for black-start fault tolerance in a grid-type wind-storage system is shown.

[0022] Figure 2 A schematic diagram of the implementation process of the preprocessing method for the state parameters of the wind-storage system is shown.

[0023] Figure 3 A schematic diagram of the implementation process of a secondary filtering method for an initial frequency domain signal based on inverse Fourier transform combined with dynamic adjustment of filter coefficients is shown.

[0024] Figure 4A schematic diagram of the implementation process of the method for initial state estimation of preprocessed wind-storage system state parameters based on the black-start state-space model is shown.

[0025] Figure 5 A schematic diagram of the implementation process of a method for real-time estimation of unmeasurable states, external disturbances, and sensor failures in a wind-storage system based on an adaptive sliding mode observer and initial state estimation is shown.

[0026] Figure 6 The diagram illustrates the implementation process of an adaptive sliding mode control model for fault identification of real-time estimation sets based on a dynamic event triggering mechanism. Detailed Implementation

[0027] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0028] Existing methods, when determining estimates based on the disturbance observer state equation and the black-start prediction model, fail to incorporate the actual state parameters of the grid-type wind-storage system for initial state estimation. This results in the black-start prediction model failing to accurately reflect the nonlinear dynamic characteristics of the wind-storage system, leading to biased initial state estimates. To address this issue, we propose an adaptive sliding mode control method for black-start fault tolerance in grid-type wind-storage systems. In short, this method first acquires the wind-storage system state parameters in real time, preprocesses these parameters, and performs initial state estimation based on the preprocessed parameters using the black-start state-space model. Then, an adaptive sliding mode observer performs real-time estimation of unmeasurable states, external disturbances, and sensor faults based on the initial state estimate. Finally, an adaptive sliding mode control model based on a dynamic event triggering mechanism is trained using modeling sample data. The model uses a dynamic event-triggered mechanism to identify faults in the real-time estimation set and determines an adaptive sliding mode control strategy based on the real-time estimation set. Using the adaptive sliding mode control strategy as a constraint, the parameters of the adaptive sliding mode control strategy are adjusted using the Lyapunov stability criterion to generate an adaptive sliding mode optimization strategy. Finally, the adaptive sliding mode optimization strategy is distributed to the local edge controller of the wind-storage system. The local edge controller implements adaptive black start based on the adaptive sliding mode optimization strategy. In this embodiment of the invention, by preprocessing the state parameters of the wind-storage system, multi-source heterogeneous actual state parameters can be integrated, thereby covering the full-dimensional dynamics of the wind-storage system. Furthermore, based on the black-start state space model combined with the preprocessed multi-source heterogeneous state parameters, the initial state variables are accurately solved, providing a reliable benchmark for the subsequent adaptive sliding mode control strategy with high-precision initial state estimation, avoiding control command mismatch caused by initial deviations.

[0029] This invention provides an adaptive sliding mode control method for black-start fault tolerance in a grid-type wind-storage system. Figure 1 A schematic diagram of the implementation process of the adaptive sliding mode control method for black-start fault tolerance in a grid-type wind-storage system is shown. The adaptive sliding mode control method for black-start fault tolerance in a grid-type wind-storage system specifically includes: S10: Real-time acquisition of wind and storage system state parameters, preprocessing of wind and storage system state parameters, and initial state estimation of preprocessed wind and storage system state parameters based on black-start state space model. S20, based on the adaptive sliding mode observer, performs real-time estimation of the unmeasurable state, external disturbances and sensor failures of the wind storage system based on the initial state estimation, and outputs a real-time estimation set; S30: Capture historical power output data of the wind-storage system's generating equipment as modeling sample data, use the modeling sample data to train an adaptive sliding mode control model based on a dynamic event triggering mechanism, and output a converged adaptive sliding mode control model. S40, Obtain the real-time estimation set. The adaptive sliding mode control model uses a dynamic event triggering mechanism to identify faults in the real-time estimation set and combines the real-time estimation set to determine the adaptive sliding mode control strategy.

[0030] S50: Obtain the adaptive sliding mode control strategy. Using the adaptive sliding mode control strategy as a constraint, adjust the parameters of the adaptive sliding mode control strategy through the Lyapunov stability criterion to generate an adaptive sliding mode optimization strategy. When adjusting the parameters of the adaptive sliding mode control strategy using the Lyapunov stability criterion, a Lyapunov function containing a sliding surface can be selected. The sliding surface is constructed based on the system state variables and reflects the difference between the system state and the desired state. Then, the derivative of the Lyapunov function is calculated. The parameters in the adaptive sliding mode control strategy will affect the dynamic changes of the sliding surface, and thus affect the derivative of the Lyapunov function. The stability of the control strategy is verified by the Lyapunov stability criterion. If the derivative of the Lyapunov function satisfies the stability adjustment, no parameter adjustment is required for the adaptive sliding mode control strategy. If the derivative of the Lyapunov function does not satisfy the stability adjustment, the process can return to S40 to continue adjusting the parameters of the adaptive sliding mode control strategy.

[0031] S60 sends the adaptive sliding mode optimization strategy to the local edge controller of the wind storage system, and the local edge controller realizes adaptive black start based on the adaptive sliding mode optimization strategy.

[0032] In this embodiment of the invention, in order to ensure that the local edge controller can correctly parse and execute the optimization strategy, the optimization strategy can be encoded using data formats such as JSON and XML, and data can be transmitted using communication protocols such as Modbus and OPC UA.

[0033] In this embodiment of the invention, by preprocessing the state parameters of the wind-storage system, multi-source heterogeneous actual state parameters can be integrated, thereby covering the full-dimensional dynamics of the wind-storage system. Furthermore, based on the black-start state space model combined with the preprocessed multi-source heterogeneous state parameters, the initial state variables are accurately solved, so that the high-precision initial state estimation provides a reliable benchmark for the subsequent adaptive sliding mode control strategy, avoiding control command mismatch caused by initial deviation.

[0034] This invention provides a method for preprocessing state parameters of a wind-storage system. Figure 2 This diagram illustrates the implementation flow of a method for preprocessing state parameters of a wind-storage system. Specifically, this method includes: S101, acquire the state parameters of the multi-source heterogeneous wind and energy storage system. The wind and energy storage system includes a wind power generation subsystem, an energy storage subsystem, a power conversion subsystem, and a grid connection interface subsystem. The wind and energy storage system includes, but is not limited to, the following devices: wind turbine, shunt, AC converter, energy storage battery, supercapacitor, grid-connected inverter, and filter. The state parameters of the wind and energy storage system include, but are not limited to, wind turbine speed, generator output power, wind speed, wind direction, generator temperature, converter temperature, battery state of charge, battery current, battery temperature, converter input current, converter efficiency, grid connection voltage, grid connection current, grid connection frequency, and grid connection power. The original state parameters of the wind and energy storage system may be caused by sensor instantaneous failure, communication packet loss, or environmental noise, resulting in outliers, which are abnormal values ​​that deviate significantly from the normal data distribution. These outliers can directly contaminate subsequent analyses, causing initial state estimates to deviate from the true values ​​and even leading to misjudgments of faults. Therefore, in this embodiment, outlier removal processing is performed on the wind-storage system state parameters. Specifically, a Hampel filter is used to remove outliers where the difference between the original parameter and the median value within the sliding window is greater than 3.5. The original parameters are marked as outliers and removed. The outlier-removed wind storage system state parameters are output, making the data after removing outliers more accurate and reducing misjudgment and erroneous control caused by abnormal data. The Hampel filter has strong robustness to non-Gaussian distributed noise and can effectively distinguish between real dynamic changes and abnormal interference, avoiding the mistaken removal of normal data. S102, Load the wind-storage system state parameters after outlier removal, and use the PTP 1588 time clock as a unified time reference to perform zero-order hold-linear extrapolation hybrid interpolation on the wind-storage system state parameters to obtain timestamp-aligned wind-storage system state parameters; S103: Obtain the wind-storage system state parameters aligned with the timestamp; perform low-pass filtering on the wind-storage system state parameters based on a second-order low-pass filter to obtain an initial filtered signal; perform discrete Fourier transform on the initial filtered signal to obtain an initial frequency domain signal; perform secondary filtering on the initial frequency domain signal based on inverse Fourier transform combined with dynamic adjustment of filter coefficients to obtain a secondary frequency signal after secondary filtering. S104: Obtain the secondary frequency signal after secondary filtering, perform dimensionless preprocessing on the secondary frequency signal based on the maximum-minimum method, and output the preprocessed wind-storage system state parameters.

[0035] In this embodiment of the invention, when preprocessing the state parameters of the wind-storage system, high-frequency noise is first removed through frequency domain analysis and dynamic filter parameter adjustment, while retaining key dynamic information of the system. The initial filtered signal is then subjected to discrete Fourier transform to obtain the initial frequency domain signal. A second filtering is then performed by inverse Fourier transform combined with dynamic adjustment of filter coefficients. This allows the invention to dynamically adjust the filter parameters according to the characteristics of different state parameters, ensuring that the most effective dynamic features can be extracted from all types of signals, providing high-quality input for subsequent initial state estimation and fault identification.

[0036] Considering that while the initial frequency domain signal reduces overall high-frequency noise, the noise distribution characteristics differ for different parameter types. Traditional fixed noise reduction methods cannot accurately suppress residual noise in specific frequency bands, leading to over-smoothing of the effective signal or incomplete removal of residual noise. This invention provides a secondary filtering method for the initial frequency domain signal based on inverse Fourier transform combined with dynamic adjustment of filter coefficients. Figure 3 This diagram illustrates the implementation flow of a secondary filtering method for an initial frequency domain signal based on inverse Fourier transform combined with dynamic adjustment of filter coefficients. The method specifically includes: S1031, Load the initial frequency domain signal and the corresponding wind-storage system state parameters, determine the preset noise reduction amount of the initial frequency domain signal based on the parameter type, and weight the initial frequency domain signal in combination with the preset noise reduction amount to obtain the frequency domain weighted value. The preset noise reduction amount is determined based on the parameter type (voltage / current / frequency) (set through historical data statistics or prior knowledge, the noise reduction amount of the voltage signal in the 100-200Hz frequency band can be 30%), and the frequency band components of the initial frequency domain signal are weighted in combination with this noise reduction amount to obtain the frequency domain weighted value. The formula for frequency domain weighting is expressed as follows: in, This indicates the preset noise reduction amount corresponding to the frequency domain weighting value and parameter type. S1032: Obtain the frequency domain weighted value of the initial frequency domain signal, perform inverse Fourier transform on the frequency domain weighted value to obtain the weighted frequency domain signal, and perform inverse Fourier transform (IDFT) on the frequency domain weighted value to convert it back from the frequency domain to the time domain, obtaining the weighted frequency domain signal. This signal retains the effective components after frequency domain weighting and noise reduction, while removing residual noise in the target frequency band. S1033, identify the parameter type of the wind storage system state parameter corresponding to the initial frequency domain signal, map the parameter change rate of the parameter type to the cutoff frequency of the low-pass filter, adjust the initial cutoff frequency of the low-pass filter, and obtain the cutoff adjustment frequency. The cutoff adjustment frequency is expressed as follows: in, These are the cutoff adjustment frequency and the initial cutoff frequency, respectively. These are respectively adjusting the proportional coefficient and the rate of change of the parameter. These are the initial maximum cutoff frequency and the initial minimum cutoff frequency, respectively. Total number of parameter types; S1034, dynamically adjust the damping ratio coefficient of the low-pass filter based on the parameter weight coefficient of the parameter type, obtain the cutoff adjustment frequency and the adjusted damping ratio coefficient and update them synchronously to the low-pass filter, and update the transfer function of the low-pass filter at the same time. The adjusted damping ratio coefficient is expressed as follows: in, These are the adjusted damping ratio coefficient and the initial damping ratio coefficient, respectively. These are the scaling factor and the parameter weight factor for the parameter type, respectively. The updated transfer function of the low-pass filter is expressed as: S1035, the low-pass filter performs secondary filtering on the weighted frequency domain signal based on the updated transfer function, and outputs the secondary frequency signal after secondary filtering.

[0037] In this embodiment of the invention, targeted weighted noise reduction is performed on residual noise in specific frequency bands for different parameter types. This avoids the loss of effective signal caused by the one-size-fits-all approach of traditional fixed filtering, preserves the key dynamic characteristics of the system, and restores the frequency domain optimization results to the time domain signal. This provides a directly processable input for the dynamic parameter adjustment of the subsequent low-pass filter, ensuring the continuity of the filtering process. The cutoff frequency is adjusted according to the actual rate of change of the parameters, which not only preserves the effective dynamic information of rapidly changing parameters but also suppresses high-frequency noise of slowly changing parameters. This avoids the over-smoothing or under-filtering problems caused by a fixed cutoff frequency. Finally, the low-pass filter performs secondary filtering on the weighted frequency domain signal based on the updated transfer function, outputting a secondary frequency signal after secondary filtering. This results in a secondary frequency signal with both low noise and high dynamic fidelity, providing a reliable input for accurate analysis of the black start process.

[0038] This invention provides a method for initial state estimation of preprocessed wind-storage system state parameters based on a black-start state-space model. Figure 4The diagram illustrates the implementation flow of a method for initial state estimation of preprocessed wind-storage system state parameters based on a black-start state-space model. Specifically, this method includes: S201, Obtain the preprocessed wind-storage system state parameters, extract the steady-state mean of the wind-storage system state parameters, and filter out dynamic disturbances (such as transient fluctuations and measurement noise) through steady-state mean extraction, providing a more reliable reference benchmark for initial state estimation; S202, Load the black-start state-space model, use the steady-state mean of the wind-storage system state parameters as the input representation of the preprocessed wind-storage system state parameters, combine the black-start state-space model to solve the initial values ​​of the state variables, and identify the steady-state signal points corresponding to the steady-state mean. The black-start state-space model is represented as follows: in, This represents the initial state estimate. These represent the state transfer matrix, the Jacobian dynamic matrix between state variables, the additive fault input matrix, and the disturbance input matrix, respectively. Represents the parameter perturbation matrix. For the initial state variables, These are the control input vector, sensor fault vector, and external disturbance term, respectively. Indicates measurement noise. For observer gain, This represents the model's output vector.

[0039] S203. A small-signal linear model characterizing the linear relationship between the steady-state mean and the steady-state signal point is constructed through the linear relationship between the steady-state mean and the steady-state signal point. It should be noted that the black-start state-space model is usually a high-order nonlinear equation system (including state variable coupling and nonlinear function terms). Directly solving the initial state estimation based on the full model is highly complex and may lose key local dynamic information. The steady-state mean only reflects the global equilibrium state of the system and cannot describe the dynamic coupling relationship of the system near the equilibrium point. In this embodiment, a small-signal linear model characterizing the relationship between the steady-state mean and the steady-state signal point is constructed through the linear relationship between the steady-state mean and the steady-state signal point. Specifically, taking the steady-state signal point as the equilibrium point, the nonlinear state-space model is linearized (Taylor expansion and retention of first-order terms) to obtain a model describing the linear relationship between small offsets of state variables and small changes in input / disturbance. The small-signal linear model focuses on the local dynamic characteristics (inertial response, damping characteristics) of the system near the equilibrium point. The small-signal linear model extracts the key dynamic coupling relationship of the system near the steady state, avoiding the high complexity and redundant information of the full nonlinear model, and focusing on the local dynamic characteristics most important for the initial state estimation. S204, calibrates the input representation of the state parameters of the wind storage system based on the small-signal linear model, and outputs the calibrated initial state variables; In step S205, a matching mapping relationship is established between the calibrated initial state variables and the black-start state-space model. The black-start state-space model is solved using the gradient descent method to obtain the optimal state estimate. This optimal state estimate is used as the initial state estimate. The input representation of the wind-storage system state parameters is calibrated based on a small-signal linear model: through the analytical relationship of the small-signal linear model (a linear mapping between state variable offsets and input / disturbance), error terms in the original input representation are adjusted (e.g., correcting small deviations in the steady-state mean), resulting in the calibrated initial state variables. Essentially, the calibration process "projects" the observed steady-state mean onto a reasonable range predicted by the model using a linear model, thereby eliminating the static deviation between the model and the actual system. The gradient descent method, through a global search, finds the combination of state variables that minimizes the error between the model and the actual data, ensuring that the initial state estimate is optimal.

[0040] In this embodiment of the invention, when performing initial state estimation of the preprocessed wind-storage system state parameters based on the black-start state-space model, the actual state parameters are closely associated with the black-start state-space model by matching the steady-state mean with the steady-state signal points of the model, thus avoiding the problem of the model deviating from reality. The small-signal linear model extracts the local dynamic characteristics of the system near the equilibrium point, thereby providing key dynamic information for the initial state estimation and improving the pertinence of the estimation. The gradient descent method finds the combination of state variables that minimizes the error between the model and the actual data through global search, ensuring that the initial state estimation result is optimal.

[0041] This invention provides a method for real-time estimation of unmeasurable states, external disturbances, and sensor failures in wind-storage systems based on an adaptive sliding mode observer and initial state estimation. Figure 5 This diagram illustrates the implementation flow of a method for real-time estimation of unmeasurable states, external disturbances, and sensor faults in a wind-storage system based on an adaptive sliding mode observer and initial state estimation. The method specifically includes: S301, Initial State Estimation Loaded: The adaptive sliding mode observer dynamically updates the initial state estimate based on a preset adaptive law mechanism. While the initial state estimate provides a system baseline at the black start moment, the state parameters of the wind-storage system dynamically change over time during the black start process (load surges cause frequency shifts, and energy storage charging and discharging cause DC bus voltage fluctuations). Traditional methods use a fixed initial state, which cannot track these dynamic changes, causing subsequent estimates and control strategies to deviate from the actual system state. After loading the initial state estimate, the adaptive sliding mode observer dynamically updates the initial state estimate based on a preset adaptive law mechanism. This adaptive law dynamically adjusts the observer's internal parameters by monitoring the system state and observer output in real time, ensuring that the observer's state estimate always closely matches the actual system dynamics. For example, when a rapid frequency drop is detected, the adaptive law increases the frequency-related sliding surface weight, prioritizing the correction of the frequency estimate. The adaptive law mechanism is expressed as follows: in, This is an update of the initial state estimate based on the adaptive law mechanism. An extended sliding surface is constructed based on initial state estimation and matching correlation parameters. This represents the matching error between the initial state estimate and the matching associated parameters. For adaptive matching coefficients, For adaptive gain; S302, based on the updated initial state estimation to drive the correction of the uncertainty of the black-start state space model, the unmeasurable state is added as an extended state variable to the state vector of the observer, and the unmeasurable state is estimated based on the adaptive sliding mode observer combined with the adaptive law mechanism to obtain the unmeasurable state estimate. S303 takes external disturbance variables as unknown inputs to the wind storage system, and separates external disturbance variables by combining the disturbance observation channel of the adaptive sliding mode observer, outputting the external disturbance estimate. Through the dynamic separation of the disturbance observation channel, the external disturbance is decoupled from the system state, avoiding the direct interference of the disturbance on the state estimation and control strategy. S304, based on the adaptive sliding mode observer, compares the sensing deviation between the theoretical estimated value of the observation sensor and the actual measured value of the sensor, determines whether the sensing deviation exceeds the preset deviation threshold, and when the sensing deviation exceeds the preset deviation threshold, determines the corresponding sensor fault, and distinguishes the fault type based on the residual characteristics, and marks the fault type and sensing deviation. S305 integrates the estimates of unmeasurable states, external disturbances, fault types, and sensing deviations into a real-time estimation set. The unmeasurable states include, but are not limited to, load impedance and line equivalent impedance, while the external disturbances include, but are not limited to, load mutations and short-circuit fault currents.

[0042] In this embodiment of the invention, when the adaptive sliding mode observer performs real-time estimation of the unmeasurable state, external disturbances, and sensor faults of the wind-storage system based on the initial state estimation, the real-time estimation set unifies the key information of the unmeasurable state, disturbances, and faults, providing comprehensive system state awareness for the adaptive sliding mode control strategy. The adaptive sliding mode observer incorporates unmeasurable states such as load impedance and line impedance into the observer, and dynamically adjusts the estimation accuracy through adaptive laws, solving the control mismatch problem caused by ignoring unmeasurable states in traditional methods. Furthermore, it separates external disturbances such as load mutations and short-circuit faults through the disturbance observation channel, providing a basis for disturbance compensation for the control strategy and improving the system's anti-interference capability. Finally, the unified real-time estimation set integrates the key data of unmeasurable states, disturbances, and faults, providing a comprehensive perception foundation for the dynamic optimization of the adaptive sliding mode control strategy.

[0043] In this embodiment of the invention, the adaptive sliding mode control model is based on an LSTM neural network architecture and includes an input layer and an output layer. A multilayer perceptron (MLP) layer is introduced between the LSTM neural network and the input layer. The MLP layer uses statistical feature analysis to identify faults in the real-time estimation set, detects and locates the fault identification results, and detects and locates at least one of the following types of faults: sensor faults (voltage / current transformer drift or failure), actuator faults (power device open circuit or short circuit), communication faults (control signal transmission delay or interruption), and internal parameter mismatch faults (system equivalent impedance estimation deviation). A fault-tolerant mechanism is set between the MLP layer and the LSTM neural network. The adaptive controller dynamically adjusts the sliding surface parameters and switching control law based on fault-tolerant control and fault identification results. A dynamic event triggering mechanism is introduced into the LSTM neural network to sparsify the output of the adaptive controller, outputting a sparse control strategy to reduce communication and computation load. The LSTM neural network balances the weights of unmeasurable state estimates, external disturbance estimates, and sensor faults in the real-time estimation set based on the Floyd-Warshall algorithm. Based on the balanced weights of unmeasurable state estimates, external disturbance estimates, and sensor faults, the sparse control strategy is dynamically adjusted to output an adaptive sliding mode control strategy.

[0044] When training the adaptive sliding mode control model based on a dynamic event-triggered mechanism using modeling sample data, it is necessary to first collect system operation data under different operating conditions as modeling samples. Historical power output data of the generating equipment covers normal operating conditions and various preset fault states, thus ensuring the comprehensiveness and representativeness of the samples. The collected modeling sample data is preprocessed, using the same method as the preprocessing of the wind-storage system state parameters, making the preprocessed data easier for subsequent model training. The preprocessed modeling sample data is divided into training, validation, and test sets, with a ratio of 4:1:1. The training set is used for learning and optimizing model parameters, the validation set is used to adjust the model's hyperparameters, and the test set is used to evaluate the model's generalization ability. The adaptive sliding mode control model is trained using the training set. During training, the parameters in the multilayer perceptron (MLP) layer, adaptive controller, and LSTM neural network are continuously adjusted using the backpropagation algorithm, so that the model's output gradually approaches the true value. During training, the model is validated using the validation set, and the model's hyperparameters, learning rate, and number of iterations are adjusted based on the validation results to improve model performance. Once the model's performance on the validation set meets the preset requirements, the model is then evaluated using the test set. If the model's performance on the test set meets the requirements of the actual application, the model training is completed, and a trained adaptive sliding mode control model based on a dynamic event triggering mechanism is finally obtained.

[0045] This invention provides a method for fault identification of real-time estimation sets based on a dynamic event triggering mechanism using an adaptive sliding mode control model. Figure 6 This diagram illustrates the implementation flow of an adaptive sliding mode control model for fault identification based on a dynamic event-triggered mechanism using a real-time estimation set. The method specifically includes: S401: Obtain the real-time estimation set. The multilayer perceptron (MLP) layer, combined with statistical feature analysis, performs fault identification on the real-time estimation set, detects and locates the fault identification results. S402, the adaptive controller dynamically adjusts the sliding surface parameters and switching control law based on fault-tolerant control and fault identification results; S403, sparsify the output of the adaptive controller based on the dynamic event triggering mechanism, and output a sparsified sparse control strategy. S404 uses the Floyd-Warshall algorithm to balance the weights of unmeasurable state estimates, external disturbance estimates, and sensor faults in the real-time estimation set. Based on the balanced weights of unmeasurable state estimates, external disturbance estimates, and sensor faults, it dynamically adjusts the sparse control strategy and outputs an adaptive sliding mode control strategy.

[0046] In this embodiment of the invention, the adaptive sliding mode control model is based on an LSTM neural network architecture. A multilayer perceptron (MLP) layer is introduced between the LSTM neural network and the input layer, and statistical feature analysis is used to identify faults in the real-time estimation set. The MLP layer can extract and analyze features from multi-source data in the real-time estimation set, detect and locate multiple types of faults. Based on different fault types and system states, the adaptive controller can adjust the sliding surface parameters and switch control laws in real time, ensuring that the control strategy always matches the actual needs of the system, improving the accuracy and effectiveness of control. Furthermore, through the weight balancing of the Floyd-Warshall algorithm, the weights of each factor in the control strategy can be precisely adjusted according to the dynamic changes of factors in the real-time estimation set, making the control strategy more consistent with the actual needs of the system, further improving the accuracy and effectiveness of control. Simultaneously, the adaptive sliding mode control model possesses strong fault tolerance capabilities, enabling rapid fault location and corresponding fault-tolerant control measures after fault detection to compensate for the impact of the fault on the system. Moreover, by dynamically adjusting the control strategy, the system can automatically resume normal operation after fault elimination, exhibiting strong self-healing capabilities and ensuring the long-term stable operation of the system.

[0047] In summary, this invention provides an adaptive sliding mode control method for black-start fault tolerance in a grid-type wind-storage system. In the embodiments of this invention, by preprocessing the state parameters of the wind-storage system, multi-source heterogeneous actual state parameters can be integrated to cover the full-dimensional dynamics of the wind-storage system. Furthermore, based on the black-start state space model combined with the preprocessed multi-source heterogeneous state parameters, the initial state variables are accurately solved, so that the high-precision initial state estimation provides a reliable benchmark for the subsequent adaptive sliding mode control strategy, avoiding control command mismatch caused by initial deviation.

[0048] It should be noted that, for the sake of simplicity, the foregoing embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to the present invention. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0049] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on these embodiments, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art can still combine, add, delete, or otherwise adjust the features of the various embodiments of the present invention according to the circumstances without conflict or creative effort, thereby obtaining different technical solutions that do not fundamentally depart from the concept of the present invention. These technical solutions also fall within the scope of protection of the present invention.

Claims

1. An adaptive sliding mode control method for black-start fault tolerance in a grid-type wind-storage system, characterized in that, The method includes: The system acquires the state parameters of the wind-storage system in real time, preprocesses the state parameters, and performs initial state estimation on the preprocessed state parameters based on the black-start state-space model. Based on the adaptive sliding mode observer, the system performs real-time estimation of the unmeasurable state, external disturbances, and sensor failures of the wind storage system based on the initial state estimation, and outputs a real-time estimation set. Historical power output data of the wind-storage system's generating equipment is captured as modeling sample data. The modeling sample data is used to train an adaptive sliding mode control model based on a dynamic event triggering mechanism, and a converged adaptive sliding mode control model is output. The adaptive sliding mode control model obtains a real-time estimation set, performs fault identification on the real-time estimation set based on a dynamic event triggering mechanism, and determines the adaptive sliding mode control strategy by combining the real-time estimation set.

2. The adaptive sliding mode control method for black-start fault tolerance of a grid-type wind-storage system as described in claim 1, characterized in that: The method further includes: Obtain the adaptive sliding mode control strategy, use the adaptive sliding mode control strategy as a constraint, and adjust the parameters of the adaptive sliding mode control strategy through the Lyapunov stability criterion to generate an adaptive sliding mode optimization strategy; The adaptive sliding mode optimization strategy is distributed to the local edge controller of the wind storage system, and the local edge controller realizes adaptive black start based on the adaptive sliding mode optimization strategy.

3. The adaptive sliding mode control method for black-start fault tolerance of a grid-type wind-storage system as described in claim 1, characterized in that: The method for preprocessing the state parameters of the wind-storage system includes: Obtain the state parameters of the multi-source heterogeneous wind and storage system, perform outlier removal on the state parameters of the wind and storage system, and output the state parameters of the wind and storage system after outlier removal. After loading outlier-removed wind-storage system state parameters, using PTP 1588 time clock as a unified time reference, zero-order hold-linear extrapolation hybrid interpolation is performed on the wind-storage system state parameters to obtain timestamp-aligned wind-storage system state parameters; The system obtains the wind-storage system state parameters aligned with timestamps. The system state parameters are then low-pass filtered using a second-order low-pass filter to obtain an initial filtered signal. The initial filtered signal is then subjected to a discrete Fourier transform to obtain an initial frequency domain signal. The initial frequency domain signal is then filtered a second time using an inverse Fourier transform combined with a dynamic adjustment method for the filter coefficients to obtain a second-frequency signal after the second filtering. The secondary frequency signal after secondary filtering is obtained, and the dimensionless preprocessing of the secondary frequency signal is performed based on the maximum-minimum method. The preprocessed wind-storage system state parameters are then output.

4. The adaptive sliding mode control method for black-start fault tolerance of a grid-type wind-storage system as described in claim 3, characterized in that: The method for secondary filtering of the initial frequency domain signal based on inverse Fourier transform combined with dynamic adjustment of filter coefficients includes: Load the initial frequency domain signal and the corresponding wind storage system state parameters, determine the preset noise reduction amount of the initial frequency domain signal based on the parameter type, and weight the initial frequency domain signal in combination with the preset noise reduction amount to obtain the frequency domain weighted value; Obtain the frequency domain weighting value of the initial frequency domain signal, and perform inverse Fourier transform on the frequency domain weighting value to obtain the weighted frequency domain signal; Identify the parameter type of the wind-storage system state parameters corresponding to the initial frequency domain signal, map the parameter change rate of the parameter type to the cutoff frequency of the low-pass filter, and adjust the initial cutoff frequency of the low-pass filter to obtain the cutoff adjustment frequency. The damping ratio coefficient of the low-pass filter is dynamically adjusted based on the parameter weight coefficients of the parameter type. The cutoff adjustment frequency and the adjusted damping ratio coefficient are obtained and updated synchronously to the low-pass filter. At the same time, the transfer function of the low-pass filter is updated. The low-pass filter performs secondary filtering on the weighted frequency domain signal based on the updated transfer function, and outputs a secondary frequency signal after secondary filtering.

5. The adaptive sliding mode control method for black-start fault tolerance of a grid-type wind-storage system as described in claim 2, characterized in that: The black-start state-space model is represented as follows: in, This represents the initial state estimate. These represent the state transfer matrix, the Jacobian dynamic matrix between state variables, the additive fault input matrix, and the disturbance input matrix, respectively. Represents the parameter perturbation matrix. For the initial state variables, These are the control input vector, sensor fault vector, and external disturbance term, respectively. Indicates measurement noise. For observer gain, This represents the model's output vector.

6. The adaptive sliding mode control method for black-start fault tolerance of a grid-type wind-storage system as described in claim 5, characterized in that: The method for initial state estimation of preprocessed wind-storage system state parameters based on the black-start state-space model includes: Obtain the preprocessed state parameters of the wind-storage system and extract the steady-state mean of the state parameters of the wind-storage system; Load the black-start state-space model, use the steady-state mean of the wind-storage system state parameters as the input representation of the preprocessed wind-storage system state parameters, combine the black-start state-space model to solve for the initial values ​​of the state variables, and identify the steady-state signal points corresponding to the steady-state mean. A small-signal linear model characterizing the linear relationship between the steady-state mean and the steady-state signal points is constructed by using the linear relationship between the steady-state mean and the steady-state signal points. The input representation of the state parameters of the wind-storage system is calibrated based on a small-signal linear model, and the calibrated initial state variables are output. The calibrated initial state variables are matched and mapped to the black-start state space model. The black-start state space model is solved using the gradient descent method to obtain the optimal state estimate, which is then used as the initial state estimate.

7. The adaptive sliding mode control method for black-start fault tolerance of a grid-type wind-storage system as described in claim 6, characterized in that: The method for real-time estimation of unmeasurable states, external disturbances, and sensor failures in a wind-storage system based on an adaptive sliding mode observer and initial state estimation includes: The initial state estimate is loaded, and the adaptive sliding mode observer dynamically updates the initial state estimate based on a preset adaptive law mechanism. Based on the updated initial state estimation, the uncertainty of the black-start state space model is corrected. The unmeasurable state is added to the state vector of the observer as an extended state variable. The unmeasurable state is estimated based on the adaptive sliding mode observer combined with the adaptive law mechanism to obtain the unmeasurable state estimate. External disturbance variables are used as unknown inputs to the wind-storage system. The external disturbance variables are separated by the disturbance observation channel of the adaptive sliding mode observer, and the estimated value of the external disturbance is output. Based on the adaptive sliding mode observer, the sensor deviation between the theoretical estimate and the actual measurement of the sensor is compared and it is determined whether the sensor deviation exceeds the preset deviation threshold. When the sensor deviation exceeds the preset deviation threshold, the corresponding sensor is determined to be faulty. Based on the residual characteristics, the fault type is distinguished and the fault type and sensor deviation are marked. The estimates of unmeasurable states, external disturbances, fault types, and sensing biases are integrated into a real-time estimation set.

8. The adaptive sliding mode control method for black-start fault tolerance of a grid-type wind-storage system as described in claim 7, characterized in that: The adaptive sliding mode control model is based on an LSTM neural network and includes an input layer and an output layer. A multilayer perceptron (MLP) layer is introduced between the LSTM neural network and the input layer. The MLP layer combines statistical feature analysis to identify faults in the real-time estimation set, detect and locate the fault identification results. An adaptive controller based on fault-tolerant control is set between the MLP layer and the LSTM neural network. A dynamic event triggering mechanism is introduced into the LSTM neural network.

9. The adaptive sliding mode control method for black-start fault tolerance of a grid-type wind-storage system as described in claim 8, characterized in that: The adaptive sliding mode control model uses a dynamic event-triggered mechanism to perform fault identification on the real-time estimation set, including: The real-time estimation set is acquired, and the multilayer perceptron (MLP) layer, combined with statistical feature analysis, is used to identify faults in the real-time estimation set, detect and locate the fault identification results. The adaptive controller dynamically adjusts the sliding surface parameters and switching control law based on fault-tolerant control and fault identification results. The output of the adaptive controller is sparsified based on a dynamic event triggering mechanism, resulting in a sparse control strategy.

10. The adaptive sliding mode control method for black-start fault tolerance of a grid-type wind-storage system as described in claim 9, characterized in that: The adaptive sliding mode control model, based on a dynamic event triggering mechanism, includes a fault identification method for the real-time estimation set, further comprising: The Floyd-Warshall algorithm is used to balance the weights of the unmeasurable state estimates, external disturbance estimates, and sensor faults in the real-time estimation set. Based on the balanced weights of the unmeasurable state estimates, external disturbance estimates, and sensor faults, the sparse control strategy is dynamically adjusted to output an adaptive sliding mode control strategy.

Images