Distributed energy storage stability adjusting method

Abstract

The invention discloses a distributed energy storage stability adjusting method, and relates to the technical field of stability adjustment, and the method comprises the following steps: obtaining distributed energy storage multi-source disturbance data, constructing a disturbance signal time-frequency characteristic matrix, and extracting a disturbance type index; collecting first data of the energy storage unit, evaluating the physical suitability based on the disturbance type index, and generating an adaptation list containing the respondable unit identifier; calling a pre-stored response template according to the adaptive list, and generating a control instruction set; issuing a control instruction to the target energy storage unit, executing a control action and synchronously acquiring second data; and performing correlation analysis on the second data and the disturbance type index, updating the disturbance-control effect mapping matrix, and optimizing a calling priority strategy of the response template through reinforcement learning. According to the method, the dynamic adaptation list is generated, so that each energy storage unit executes the optimal adjustment strategy in the physical boundary of the energy storage unit, and the response precision to different types of disturbances and the system stability recovery speed are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of stability regulation, and more specifically, to a distributed energy storage stability regulation method. Background Art

[0002] In modern distribution networks, with the widespread integration of renewable energy generation devices, nonlinear loads, and sensitive electronic equipment, power quality issues are becoming increasingly prominent. Typical manifestations include frequent disturbances such as voltage flicker, voltage sag, transient pulses, and harmonics. These disturbances not only affect the normal operation of end-user equipment but can also cause serious consequences such as malfunction, shortened lifespan, or data distortion. Consequently, an increasing number of power systems are deploying distributed energy storage devices to improve local power quality and system stability.

[0003] Most existing distributed energy storage systems rely on traditional voltage, current, or frequency deviation control and regulation, with response mechanisms primarily based on fixed threshold triggering and linear regulation strategies. While this approach achieves basic voltage stabilization and frequency regulation, it has significant limitations in complex disturbance scenarios. In particular, when the type of voltage disturbance is unclear, relying solely on numerical deviations can easily lead to over-regulation or delayed response. Misjudging the nature of the disturbance can even trigger secondary disturbances, reducing system stability.

[0004] For example, when a system experiences short transient pulses or high-frequency harmonic disturbances, traditional energy storage control systems are unable to distinguish the type and duration of the disturbance, often making regulation decisions that do not match the disturbance characteristics, resulting in energy waste or control failure. Further analysis revealed that current energy storage regulation mechanisms lack a closed-loop chain from "disturbance perception - type identification - strategy matching," making it impossible to achieve a categorized response based on the type of disturbance.

[0005] For example, the invention patent with announcement number CN202411878264.7 discloses a method and device for adaptively adjusting parameters of an analog light source controller. The method includes: collecting the input voltage signal, output current signal, ambient temperature signal and channel power signal of the analog light source controller, and inputting them into a neural network model for parameter mapping to obtain a dynamic response matrix of the light source parameters; performing feature dimensionality reduction and constructing a target optimization model; gridding the parameter space, and generating a steepest descent direction vector and a vertical search direction vector at the divided grid points to obtain a parameter optimization direction set; iteratively updating each control parameter to obtain a parameter iteration sequence set; performing a voltage stabilization output test to obtain multiple candidate control parameter combinations; performing a parallel verification test to determine the optimal control parameter combination, thereby effectively suppressing the crosstalk between channels of the analog light source controller and solving the mutual interference problem when multiple channels work in parallel.

[0006] For example, the invention patent with announcement number CN202411366785.4 discloses a model-adaptive predictive control method and system based on disturbance state compensation. This method, by introducing an adaptive disturbance identifier based on deep learning, can identify and compensate for system disturbances in real time, significantly improving the control accuracy and robustness of the system in the face of a variable disturbance environment. The invention uses deep learning technology to construct a disturbance identifier, making the disturbance identification process driven by data more intelligent, capable of more accurately capturing subtle changes in disturbances and improving the accuracy of disturbance identification. A performance evaluation mechanism has been established to allow the automatic selection of the optimal controller from a set of controllers to ensure optimal control of the system under disturbances. The accuracy and effectiveness of the control signal are further ensured through precise error evaluation, disturbance compensation, and correction of the prediction model. These technologies and methods not only optimize the control effect, but also extend the service life of the equipment and reduce maintenance costs.

[0007] The aforementioned technical solutions present at least the following technical issues: When the system experiences short transient pulses or high-frequency harmonic disturbances, traditional energy storage control systems are unable to distinguish the disturbance type and duration, often making regulation decisions that do not match the disturbance characteristics, resulting in energy waste or control failure. Further analysis revealed that the current energy storage regulation mechanism lacks a closed-loop chain from "disturbance perception - type identification - strategy matching", making it impossible to achieve a classified response based on the disturbance type.

[0008] In view of the above problems, the present invention proposes a solution. Summary of the Invention

[0009] To overcome the above-mentioned shortcomings of the prior art, an embodiment of the present invention provides a distributed energy storage stability regulation method. By constructing a time-frequency feature matrix of the disturbance signal to extract a disturbance type indicator, and combining physical state modeling and dynamic adaptive evaluation of the energy storage unit, a matching control instruction is generated. This solves the problem that when the system is subjected to short transient pulses or high-frequency harmonic disturbances, traditional energy storage control systems are unable to distinguish the disturbance type and duration, and often make regulation decisions that do not match the disturbance characteristics.

[0010] To achieve the above object, the present invention provides the following technical solutions: A distributed energy storage stability regulation method comprises the following steps: acquiring multi-source disturbance data of distributed energy storage, constructing a time-frequency characteristic matrix of the disturbance signal, and extracting a disturbance type index; collecting first data of an energy storage unit, evaluating physical adaptability based on the disturbance type index, and generating an adaptation list containing identifiers of responsive units; calling a pre-stored response template according to the adaptation list and generating a control instruction set; issuing a control instruction to a target energy storage unit, executing a control action, and synchronously collecting second data; correlating and analyzing the second data with the disturbance type index, updating a disturbance-control effect mapping matrix, and optimizing a calling priority strategy of the response template through reinforcement learning.

[0011] In a preferred embodiment, the method of acquiring multi-source disturbance data of distributed energy storage and constructing a time-frequency characteristic matrix of the disturbance signal is specifically as follows: collecting multi-source disturbance data and constructing a time series input matrix through clock calibration; performing short-time energy gain analysis on each channel data of the time series input matrix, generating a disturbance trigger criterion function, and locating the starting point of the dynamic slice; constructing a buffer analysis window with the dynamic slice starting point as the center, and performing trend item elimination and energy normalization processing on the disturbance signal in the window; performing modal decomposition on the normalized signal, extracting the intrinsic mode component, and generating a time-frequency characteristic matrix through Hilbert transform.

[0012] In a preferred embodiment, the extraction of disturbance type characteristic indicators includes: performing energy spectrum normalization and time domain boundary enhancement processing on the time-frequency feature matrix, extracting the energy center trajectory and local peak area, and constructing a disturbance local response feature set; extracting disturbance type characteristic indicators based on the disturbance local response feature set; inputting the extracted disturbance type characteristic indicators into a preset amplitude-frequency feature clustering algorithm, performing feature clustering analysis, and completing preliminary discrimination of the disturbance type based on the cluster center and the disturbance label space, and outputting a preliminary disturbance type label.

[0013] In a preferred embodiment, the first data of the energy storage unit is collected, the physical adaptability is evaluated based on the disturbance type index, and an adaptation list including the identification of the responsive unit is generated. Specifically, the dynamic response characteristic curve of each energy storage unit under different disturbance types is collected; features are extracted from the dynamic response characteristic curve, a multi-index evaluation matrix is ​​constructed, the entropy weight method is used to automatically assign indicator weights, and a dynamic fitness function is constructed in combination with the TOPSIS method; the response capability level of each energy storage unit under the current disturbance situation is calculated based on the fitness function, and a sorted energy storage adaptation list is generated.

[0014] In a preferred embodiment, the dynamic response characteristic curves of each energy storage unit under different disturbance types are collected as follows: Taking the disturbance type characteristic index as the input constraint condition, a digital twin model of the energy storage unit is constructed based on the first data, and a typical disturbance scenario is virtually simulated through the model predictive control method to output the dynamic response characteristic curve of each energy storage unit under different disturbance types.

[0015] In a preferred embodiment, the dynamic response characteristic curve is feature extracted, a multi-index evaluation matrix is ​​constructed, the entropy weight method is used to automatically assign indicator weights, and a dynamic fitness function is constructed in combination with the TOPSIS method. Specifically, the dynamic response characteristic curve is feature extracted to construct a response indicator set of the energy storage unit; a multi-index evaluation matrix is ​​constructed based on the response indicator set, and the information entropy value of each indicator in the evaluation matrix is ​​calculated based on the entropy weight method to generate an objective weight coefficient; the evaluation matrix is ​​weighted based on the objective weight coefficient, and the Euclidean distance of each energy storage unit to the positive ideal solution and the negative ideal solution is calculated using the TOPSIS method, and a fitness function is constructed based on relative closeness to output the comprehensive response score of each energy storage unit in the current disturbance scenario.

[0016] In a preferred embodiment, the method of calling a pre-stored response template according to the adaptation list to generate a control instruction set specifically comprises: screening the target energy storage unit group according to the fitness score sorting; matching the response template in the local control strategy library with the preliminary label of the disturbance type as the input condition; extracting the control parameter set corresponding to the disturbance type in the response template; and jointly encoding the control parameter set and the real-time system state quantity into a structured control instruction.

[0017] In a preferred embodiment, the control instructions are issued to the target energy storage unit, the control action is executed, and the second data is synchronously collected. Specifically, the control instructions are distributed according to the target energy storage unit number and sent to the corresponding energy storage interface layer through the scheduling controller to execute the corresponding control behavior; during the execution of the control action, key operating parameters in the disturbance response process are synchronously collected as disturbance response effect parameters; and the system recovery parameters of the energy storage unit itself are recorded in parallel, and the disturbance response effect parameters and the system recovery parameters are combined to form a disturbance response data set.

[0018] In a preferred embodiment, the association analysis of the second data and the disturbance type index is performed to update the disturbance-control effect mapping matrix, specifically: the disturbance response effect parameters, the energy storage unit stress parameters and the initial disturbance type index after the execution of the control instruction are structured and mapped to construct a causal association map; based on the causal association map and the causal Bayesian network algorithm, the direct impact path from the control instruction to the system recovery parameter is identified; based on the direct impact path, the control strategy-disturbance type mapping matrix is ​​annotated with path strength, the intervention contribution of each strategy is quantified, and the strategy is assigned a side cost weight according to the side effect path existing in the impact path; according to the path strength and the side cost weight, the priority and call weight of the control strategy in the disturbance type mapping space are dynamically adjusted to update the disturbance-control effect mapping matrix.

[0019] In a preferred embodiment, the calling priority strategy of the response template optimized by reinforcement learning is specifically as follows: the updated effect mapping matrix is ​​used as the state space input to construct a multidimensional state vector including the disturbance label, response effect, system recovery parameter and control path strength; based on the multidimensional state vector, a reward function including path strength gain and sub-cost penalty is set for benefit evaluation of the reinforcement learning model to form the optimization target of strategy selection; the constructed state vector and reward function are input into the time-aware reinforcement learning network, and the priority weights of different control strategies under various disturbance types are output through the training process; based on the strategy priority weights, the calling order of the control strategy in the disturbance response process is dynamically updated to realize the optimized scheduling of the control instruction calling priority under different disturbance scenarios.

[0020] The technical effects and advantages of the distributed energy storage stability adjustment method of the present invention are as follows: 1. This invention constructs a disturbance type index by extracting the time-frequency characteristics of disturbances, combines the multi-dimensional operating status data of distributed energy storage units, establishes a digital twin simulation and response characteristic evaluation mechanism, and generates a dynamic adaptation list; then, based on the ternary structure template of disturbance type and energy storage characteristics, it generates customized control instructions, enabling each energy storage unit to execute the optimal adjustment strategy within its physical boundaries, effectively improving the response accuracy to different types of disturbances and the speed of system stability recovery.

[0021] 2. The present invention introduces structured collection and causal analysis of disturbance response effects and system recovery parameters, uses causal Bayesian networks to identify the direct impact relationship of control paths, and labels the contribution and side costs of strategy intervention; on this basis, it constructs a reinforcement learning state space and reward function to form a self-learning and self-optimization mechanism of strategy call priority, realizes dynamic optimization of control strategies and closed-loop reinforcement regulation of disturbance response processes, and has sustainable adaptive control capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1The figure is a flow chart of a distributed energy storage stability adjustment method according to the present invention. DETAILED DESCRIPTION

[0023] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0024] Example 1, Figure 1 The present invention provides a distributed energy storage stability adjustment method, comprising the following steps: S1, obtains multi-source disturbance data of distributed energy storage, constructs the time-frequency characteristic matrix of the disturbance signal, and extracts the characteristic index of the disturbance type; The multi-source disturbance data includes voltage waveform, current change rate, frequency offset, harmonic distortion rate and power disturbance.

[0025] The method of obtaining multi-source disturbance data of distributed energy storage and constructing the time-frequency characteristic matrix of the disturbance signal is as follows: Collect multi-source disturbance data, perform clock alignment and multi-channel synchronization, and form a time series input matrix; Perform short-time energy gain analysis on each channel data of the time series input matrix to generate a disturbance trigger criterion function and identify the disturbance starting boundary as the starting point of dynamic slicing; A buffer analysis window is constructed with the dynamic slice starting point as the center, and trend term elimination and energy normalization processing are performed on the disturbance signal within the window to reduce the interference of DC offset and boundary effect on modal decomposition; The normalized disturbance signal is subjected to modal decomposition to extract the intrinsic modal components of the disturbance signal, and the time-frequency feature matrix of the disturbance signal is constructed through Hilbert transform.

[0026] The extraction of disturbance type characteristic indicators includes: The time-frequency feature matrix is ​​subjected to energy spectrum normalization and time domain boundary enhancement processing to extract the energy center trajectory and local peak area, and a disturbance local response feature set is constructed. The feature set is based on the wavelet decomposition results and reflects the excitation mode of the disturbance on multiple frequency channels. Based on the local response feature set of disturbance, characteristic indicators of disturbance type including frequency distribution width, number of main frequency points, energy concentration, local peak duration, and modal energy ratio are extracted as the physical semantic expression of disturbance; The extracted disturbance type characteristic indicators are input into the preset amplitude-frequency characteristic clustering algorithm, and clustering and classification are performed according to the frequency distribution width, energy concentration, and duration indicators of the characteristic points; Based on the clustering classification results and according to the preset disturbance type label space, preliminary disturbance type discrimination is performed, and preliminary disturbance type labels are output, including but not limited to transient pulses, voltage sags, harmonic disturbances and their combined disturbances.

[0027] The disturbance trigger criterion function is specifically:

[0028] in, is the perturbation energy gain ratio, is the disturbance signal sample value collected at the i-th moment, is the reference time of the current sliding window center, that is, the current detection point, is the half-width of one side of the sliding window.

[0029] The construction of the disturbance trigger criterion function identifies the disturbance starting boundary as the dynamic slicing starting point, specifically: If there is a signal channel whose disturbance energy gain ratio is greater than a preset ratio threshold, the signal channel is identified as a disturbance starting boundary and used as a dynamic slicing starting point.

[0030] S2, collecting first data of the energy storage unit, evaluating physical adaptability based on the disturbance type index, and generating an adaptability list including identifiers of responsive units; The first data includes state of charge, power margin, charge and discharge rate and current regulated load curve.

[0031] The first data of the current energy storage unit is obtained, and the physical adaptability of the energy storage response is evaluated based on the disturbance type index to generate an adaptation list, specifically: Using the disturbance type characteristic index as an input constraint, a digital twin model of the energy storage unit is constructed based on the first data. Without disturbing the actual equipment, a model predictive control method is used to perform virtual simulation of typical disturbance scenarios, and the dynamic response characteristic curve of each energy storage unit under different disturbance types is output; the typical disturbance scenarios include frequency drop, harmonic shock, and voltage sag; The dynamic response characteristic curve is extracted to construct a multi-index evaluation matrix. The entropy weight method is used to automatically assign index weights, and the dynamic fitness function is constructed in combination with the TOPSIS method. Based on the fitness function, the response capability level of each energy storage unit under the current disturbance situation is calculated, and a sorted energy storage adaptation list is generated.

[0032] The feature extraction of the dynamic response characteristic curve is performed, a multi-index evaluation matrix is ​​constructed, the entropy weight method is used to automatically assign the index weights, and the dynamic fitness function is constructed in combination with the TOPSIS method, specifically: Extract features from the dynamic response characteristic curves of each energy storage unit in the disturbance simulation, extract several response indices, and construct a response index set for the energy storage unit; the indices are used to characterize the transient and steady-state regulation capabilities of the energy storage unit under different disturbance types; A multi-index evaluation matrix is ​​constructed based on the response index set, and the information entropy value of each indicator in the evaluation matrix is ​​calculated based on the entropy weight method to generate an objective weight coefficient. The weight reflects the importance of each response indicator under the current disturbance type. The evaluation matrix is ​​weighted based on the objective weight coefficient, and the TOPSIS method is used to calculate the Euclidean distance of each energy storage unit to the positive ideal solution and the negative ideal solution respectively. A fitness function is constructed based on the relative closeness, and the comprehensive response score of each energy storage unit in the current disturbance scenario is output.

[0033] The information entropy value is specifically:

[0034]

[0035] in, is the information entropy of the j-th indicator, is the normalized index value, is a very small constant, is the total number of energy storage units, is a constant that normalizes information entropy.

[0036] The entropy weight is specifically:

[0037] The comprehensive response score is specifically:

[0038]

[0039] in, is the entropy weight of the j-th indicator, is the comprehensive response score, is the Euclidean distance from the i-th unit to the positive ideal solution, is the Euclidean distance from the i-th unit to the negative ideal solution, is the weighted normalized value of the i-th unit on index j, is the positive ideal value of the j-th index, is the negative ideal value of the j-th indicator.

[0040] It should be noted that a higher comprehensive response score indicates a better response of the energy storage unit to disturbances.

[0041] S3, calling the pre-stored response template according to the adaptation list to generate a control instruction set; The pre-stored response template is called according to the adaptation list to generate a control instruction set, specifically: According to the priority order of the energy storage units in the adaptation list, a target energy storage unit group with a fitness score greater than a preset threshold is screened. Combined with the preliminary label of the disturbance type as an input condition, a corresponding response template is retrieved from the local control strategy library. The response template is a preset ternary mapping structure of disturbance type, energy storage characteristics, and control behavior. Calling the matching response template to extract the control parameter set for the corresponding disturbance type, including the response trigger threshold, power regulation slope, frequency dead zone setting, and filter bandwidth setting. The control parameter set is dynamically adapted based on the energy storage unit type. The control parameter set is combined with the current system operating state to generate a structured control instruction. The control instruction includes the target unit number, adjustment direction, adjustment amplitude, and action duration, which are used to guide the specific control action of the selected energy storage unit during the disturbance response process. The current system operating state includes the grid-connected frequency, voltage amplitude, and load change rate.

[0042] The control instruction generation process is based on a template-driven generation mechanism to ensure that the response strategy is reconfigurable and timely, and the physical boundary conditions for instruction execution are set by the energy storage status information in the adaptation list.

[0043] S4, issuing a control instruction to the target energy storage unit, executing the control action and synchronously collecting the second data; The second data includes disturbance response effect and system recovery parameters The sending of control instructions to the target energy storage unit, executing the control action and synchronously collecting the second data is specifically as follows: Distribute structured control instructions according to the target energy storage unit number and send them to the corresponding energy storage interface layer through the dispatch controller to execute corresponding control actions, including but not limited to active output increase, reactive support injection, and filter control startup; During the execution of the control action, key operating parameters in the disturbance response process are synchronously collected, including the system frequency recovery curve, voltage recovery characteristics, total harmonic distortion rate, and node voltage fluctuation index as disturbance response effect parameters; The system recovery parameters of the energy storage unit itself are recorded in parallel, including frequency recovery slope, voltage recovery curve, harmonic content attenuation rate, and node voltage deviation recovery time, to measure the actual contribution of the control strategy to grid stability. The disturbance response effect parameters and system recovery parameters together constitute the disturbance response dataset, which serves as the input basis for subsequent disturbance-strategy-result mapping relationship optimization and strategy adaptive scheduling learning.

[0044] S5, associates and analyzes the second data with the disturbance type indicator, updates the disturbance-control effect mapping matrix, and optimizes the calling priority strategy of the response template through reinforcement learning.

[0045] The correlation analysis of the second data and the disturbance type indicator and updating of the disturbance-control effect mapping matrix are specifically as follows: The disturbance response effect parameters after executing the control instructions, the stress parameters of the energy storage unit and the initial disturbance type indicators are structured and mapped to construct a causal relationship map; Based on the causal association graph, the causal Bayesian network algorithm is used to identify the direct impact path of each control instruction on the system recovery parameters; Based on the identified direct impact paths, the original mapping matrix between control strategies and disturbance types is annotated with path strength, the intervention contribution of each strategy is quantified, and the strategy side cost weight is assigned according to the side effect paths in the impact path; According to the path strength and secondary cost weight, the priority and call weight of the control strategy in the disturbance type mapping space are dynamically adjusted, and the effect mapping matrix is ​​updated.

[0046] The call priority strategy of optimizing the response template through reinforcement learning is specifically as follows: The updated effect mapping matrix is ​​used as the state space input to construct a multidimensional state vector containing the disturbance label, response effect, system recovery parameters and control path strength; Based on the multi-dimensional state vector, a reward function including path strength gain and side cost penalty is set to evaluate the benefits of the reinforcement learning model and form the optimization goal of the strategy selection; The constructed state vector and reward function are input into a time-aware reinforcement learning network, and the priority weights of different control strategies under various disturbance types are output through the training process. Based on the strategy priority weights, the calling order of the control strategies in the disturbance response process is dynamically updated to achieve optimized scheduling of the control instruction calling priorities under different disturbance scenarios.

[0047] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0048] The above embodiments may be implemented in whole or in part through software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0049] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0050] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0051] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0052] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A distributed energy storage stability adjustment method, characterized in that: The following steps are involved: Obtain multi-source disturbance data of distributed energy storage, construct the time-frequency characteristic matrix of the disturbance signal, and extract the disturbance type index; collecting first data of the energy storage unit, evaluating physical adaptability based on a disturbance type indicator, and generating an adaptability list including identifiers of responsive units; Call the pre-stored response template according to the adaptation list to generate a control instruction set; Sending a control instruction to the target energy storage unit, executing the control action and synchronously collecting the second data; The second data and the disturbance type indicator are correlated and analyzed, the disturbance-control effect mapping matrix is ​​updated, and the calling priority strategy of the response template is optimized through reinforcement learning.

2. The distributed energy storage stability adjustment method according to claim 1, characterized in that: The method of obtaining multi-source disturbance data of distributed energy storage and constructing the time-frequency characteristic matrix of the disturbance signal is as follows: Collect multi-source disturbance data and construct a time series input matrix through clock calibration; Perform short-time energy gain analysis on each channel data of the time series input matrix to generate a disturbance trigger criterion function and locate the starting point of the dynamic slice; A buffer analysis window is constructed with the dynamic slice starting point as the center, and trend item elimination and energy normalization processing are performed on the disturbance signal within the window; The normalized signal is modally decomposed to extract the intrinsic mode components, and the time-frequency feature matrix is ​​generated through Hilbert transform.

3. The distributed energy storage stability adjustment method according to claim 2, characterized in that: The extraction of disturbance type characteristic indicators includes: The time-frequency feature matrix is ​​subjected to energy spectrum normalization and time domain boundary enhancement processing to extract the energy center trajectory and local peak area, and to construct a disturbance local response feature set; Extract disturbance type characteristic indicators based on disturbance local response feature set; The extracted disturbance type characteristic indicators are input into the preset amplitude-frequency feature clustering algorithm to perform feature clustering analysis, and the preliminary discrimination of disturbance type is completed based on the cluster center and the disturbance label space, and the preliminary disturbance type label is output.

4. The distributed energy storage stability adjustment method according to claim 3, characterized in that: The first data of the energy storage unit is collected, the physical adaptability is evaluated based on the disturbance type index, and an adaptability list including the identifier of the responsive unit is generated, specifically: Collect the dynamic response characteristic curves of each energy storage unit under different disturbance types; The dynamic response characteristic curve is extracted to construct a multi-index evaluation matrix. The entropy weight method is used to automatically assign index weights, and the dynamic fitness function is constructed in combination with the TOPSIS method. Based on the fitness function, the response capability level of each energy storage unit under the current disturbance situation is calculated, and a sorted energy storage adaptation list is generated.

5. The distributed energy storage stability adjustment method according to claim 4, characterized in that: The dynamic response characteristic curves of each energy storage unit under different disturbance types are collected as follows: Taking the disturbance type characteristic index as the input constraint condition, a digital twin model of the energy storage unit is constructed based on the first data, and a typical disturbance scenario is virtually simulated through the model predictive control method to output the dynamic response characteristic curve of each energy storage unit under different disturbance types.

6. The distributed energy storage stability adjustment method according to claim 5, characterized in that: The feature extraction of the dynamic response characteristic curve is performed, a multi-index evaluation matrix is ​​constructed, the entropy weight method is used to automatically assign the index weights, and the dynamic fitness function is constructed in combination with the TOPSIS method, specifically: Extracting features from the dynamic response characteristic curve to construct a response index set of the energy storage unit; A multi-index evaluation matrix is ​​constructed based on the response index set, and the information entropy value of each indicator in the evaluation matrix is ​​calculated based on the entropy weight method to generate an objective weight coefficient; The evaluation matrix is ​​weighted based on the objective weight coefficient, and the TOPSIS method is used to calculate the Euclidean distance of each energy storage unit to the positive ideal solution and the negative ideal solution respectively. A fitness function is constructed based on the relative closeness, and the comprehensive response score of each energy storage unit in the current disturbance scenario is output.

7. The distributed energy storage stability adjustment method according to claim 6, characterized in that: The pre-stored response template is called according to the adaptation list to generate a control instruction set, specifically: Filter the target energy storage unit group according to the fitness score ranking; Using the preliminary disturbance type label as input, the response template is matched in the local control strategy library; Extracting the control parameter set corresponding to the disturbance type in the response template; The control parameter set and the real-time system state are jointly encoded into structured control instructions.

8. The distributed energy storage stability adjustment method according to claim 7, characterized in that: The sending of control instructions to the target energy storage unit, executing the control action and synchronously collecting the second data is specifically as follows: Distribute the control instructions according to the target energy storage unit number and send them to the corresponding energy storage interface layer through the dispatch controller to execute the corresponding control behavior; During the execution of the control action, key operating parameters in the disturbance response process are synchronously collected as disturbance response effect parameters; The system recovery parameters of the energy storage unit itself are recorded in parallel, and the disturbance response effect parameters and the system recovery parameters together constitute the disturbance response data set.

9. The distributed energy storage stability adjustment method according to claim 8, characterized in that: The correlation analysis of the second data and the disturbance type indicator and updating of the disturbance-control effect mapping matrix are specifically as follows: The disturbance response effect parameters after executing the control instructions, the stress parameters of the energy storage unit and the initial disturbance type indicators are structured and mapped to construct a causal relationship map; Identify the direct impact path from control instructions to system recovery parameters based on causal association graphs and causal Bayesian network algorithms; Based on the direct impact path, the control strategy-disturbance type mapping matrix is ​​annotated with path strength, the intervention contribution of each strategy is quantified, and the strategy side cost weight is assigned according to the side effect path in the impact path; According to the path strength and secondary cost weight, the priority and call weight of the control strategy in the disturbance type mapping space are dynamically adjusted, and the disturbance-control effect mapping matrix is ​​updated.

10. The distributed energy storage stability adjustment method according to claim 9, characterized in that: The call priority strategy of optimizing the response template through reinforcement learning is specifically as follows: The updated effect mapping matrix is ​​used as the state space input to construct a multidimensional state vector containing the disturbance label, response effect, system recovery parameters and control path strength; Based on the multi-dimensional state vector, a reward function including path strength gain and side cost penalty is set to evaluate the benefits of the reinforcement learning model and form the optimization goal of the strategy selection; The constructed state vector and reward function are input into a time-aware reinforcement learning network, and the priority weights of different control strategies under various disturbance types are output through the training process. Based on the strategy priority weights, the calling order of the control strategies in the disturbance response process is dynamically updated to achieve optimized scheduling of the control instruction calling priorities under different disturbance scenarios.

Images