Mechanism-data dual-driven grid-formation type energy storage parameter adaptive regulation method

Abstract

The application relates to the technical field of network-constructing energy storage control, and discloses a mechanism-data double-driving network-constructing energy storage parameter adaptive regulation method, which collects original data of an energy storage side and a power grid side, carries out wavelet threshold denoising, interpolation completion and normalization pretreatment on the collected original data, and obtains a standardized time series data set; the standardized time series data set is input into a hybrid model based on a mechanism constraint-physical information fusion neural network, and an energy storage state parameter is output in real time; a FCM clustering algorithm is used to divide multiple working condition domains based on the energy storage state parameter, and a sparrow search algorithm is used to optimize initial values of control parameters of the working condition domains; a network-constructing energy storage digital twin is constructed, a LightGBM model and a buffer function are combined, the initial values of the control parameters are dynamically corrected, real-time optimal control parameters are obtained, and the real-time optimal control parameters are issued to a controller for regulation and control; and the application realizes dynamic, accurate and adaptive adjustment of network-constructing energy storage control parameters.

Description

Technical Field

[0001] This invention relates to the field of grid-type energy storage control technology, specifically to an adaptive control method for grid-type energy storage parameters based on mechanism-data dual-drive. Background Technology

[0002] In the context of power systems with a high proportion of new energy and high power electronics, grid-based energy storage has become a core device for ensuring grid stability due to its active voltage / frequency support and virtual inertia damping compensation capabilities. Currently, grid-based energy storage control mostly employs fixed-parameter virtual synchronous generator control or single-data-driven adaptive control, which has significant drawbacks: fixed-parameter strategies cannot adapt to complex operating conditions such as grid short-circuit ratio fluctuations, time-varying loads, and energy storage aging, easily leading to broadband oscillations, transient overcurrents, or instability risks; pure data-driven strategies rely on massive amounts of high-quality data, have poor generalization capabilities, lack physical constraints, and are prone to parameter drift and control inaccuracies under extreme conditions. Summary of the Invention

[0003] The purpose of this invention is to solve the above problems by designing an adaptive control method for grid-type energy storage parameters based on mechanism-data dual-drive.

[0004] This invention provides an adaptive control method for grid-type energy storage parameters based on a mechanism-data dual-drive approach. The method includes the following steps: Raw data from the energy storage side and the grid side are collected, and wavelet threshold denoising, interpolation completion, and normalization preprocessing are performed on the collected raw data to obtain a standardized time series dataset. The standardized time-series dataset is input into a hybrid model based on a mechanistic constraint-physical information fusion neural network, which outputs energy storage state parameters in real time. Based on the energy storage state parameters, the FCM clustering algorithm is used to divide multiple operating domains, and the initial values ​​of control parameters in each operating domain are optimized by the sparrow search algorithm. A grid-type energy storage digital twin is constructed. By combining the LightGBM model and buffer function, the initial values ​​of control parameters are dynamically corrected to obtain the real-time optimal control parameters, which are then sent to the controller for regulation.

[0005] Optionally, in the first implementation of the present invention, the raw data collected from the energy storage side and the grid side are subjected to wavelet threshold denoising, interpolation completion, and normalization preprocessing to obtain a standardized time-series dataset, including: Raw data from the energy storage side and the grid side are collected, and wavelet decomposition is performed on the raw data to obtain high-frequency and low-frequency components. Noise signals in the high-frequency components are suppressed by a preset threshold. Then, the processed high-frequency and low-frequency components are subjected to inverse wavelet transform to obtain clean data after denoising. Missing values ​​in the denoised clean data are filled in using a linear interpolation algorithm. The filled data is then normalized to obtain a standardized time series dataset.

[0006] Optionally, in the second implementation of the present invention, the step of using a linear interpolation algorithm to fill in missing values ​​in the denoised clean data includes: Traverse the clean data after noise reduction, mark the locations of missing information in the data sequence, and locate the two valid complete data samples before and after each missing point in the data sequence. A linear correlation is established based on two consecutive valid data points, and the numerical information corresponding to the missing positions is determined according to the interval ratio. Complete all missing points in the marked data to obtain the completed data.

[0007] Optionally, in a third implementation of the present invention, the hybrid model of the mechanistic constraint-physical information fusion neural network includes an input layer, an encoder module, a mechanistic constraint embedding module, a decoder module, and an output layer. The encoder module includes a position encoding layer, a multi-layer attention layer, and a fully connected layer. The mechanistic constraint embedding module is located between the encoder and the decoder. The decoder module includes a decoding layer and an adaptive weight adjustment layer.

[0008] Optionally, in the fourth implementation of the present invention, the step of inputting the standardized time-series dataset into a hybrid model based on a mechanistic constraint-physical information fusion neural network and outputting energy storage state parameters in real time includes: The input layer of a hybrid model based on a mechanistic constraint-physical information fusion neural network receives a standardized time-series dataset. The encoder module first adds temporal feature information to the standardized time series dataset through the position encoding layer, and then extracts features from the data through multiple attention layers. The extracted multi-dimensional temporal features are input into the fully connected layer for integration and dimensionality reduction to output a high-dimensional temporal feature vector. The mechanism constraint embedding module performs constraint screening on the high-dimensional time series feature vector. The decoder module receives the screened high-dimensional time series feature vector, first performs reverse decoding on the feature vector through the decoding layer, and then dynamically balances the data fitting accuracy and physical constraint requirements through the adaptive weight adjustment layer to obtain the decoded feature vector. The output layer converts the decoded feature vector into energy storage state parameters.

[0009] Optionally, in a fifth implementation of the present invention, the mechanism constraint embedding module performs constraint filtering on the high-dimensional temporal feature vector, including: The high-dimensional time series feature vector is decomposed dimension by dimension, and each decomposed dimension is compared with the corresponding physical constraints to determine whether the feature value of the current dimension is within the preset physical boundary range. Based on physical constraints, the feature dimensions that exceed the physical boundary are corrected, and the corrected dimensional features are re-integrated to form a filtered high-dimensional time-series feature vector.

[0010] Optionally, in the sixth implementation of the present invention, the step of dividing multiple operating domains based on energy storage state parameters using the FCM clustering algorithm and optimizing the initial values ​​of control parameters for each operating domain using the sparrow search algorithm includes: Define at least four clustering categories, including strong grid steady state, weak grid transient state, fault disturbance, and energy storage aging, and initialize the center vector of each clustering category; Calculate the fuzzy membership degree of each energy storage state parameter sample to each cluster center, iteratively update the cluster center and membership degree matrix, minimize the sum of squared distances between the sample and the corresponding cluster center, until the cluster center no longer changes significantly, and determine the operating condition category to which each sample belongs; For each of the divided operating conditions, the initial values ​​of the control parameters for each operating condition are obtained by optimization using the sparrow search algorithm.

[0011] Optionally, in the seventh implementation of the present invention, the step of optimizing the initial values ​​of the control parameters for each of the divided operating domains using a sparrow search algorithm includes: Initialize a sparrow population, where each individual in the population corresponds to a set of control parameters; The fitness value of each sparrow individual is calculated to screen out producers, followers and vigilants in the population. Producers are responsible for exploring new parameter regions in the search space, followers follow producers to conduct local searches, and vigilants are responsible for monitoring the search status of the population. When the search is found to be trapped in a local optimum, a position update is triggered to escape the local optimum region; Through multiple iterations until the preset number of iterations is reached, the initial values ​​of the control parameters for each operating condition domain are output.

[0012] Optionally, in the eighth implementation of the present invention, the construction of a grid-type energy storage digital twin, combined with the LightGBM model and buffer function, dynamically corrects the initial values ​​of control parameters to obtain real-time optimal control parameters, and sends the real-time optimal control parameters to the controller for regulation, including: Construct a grid-type energy storage digital twin, extract features from the digital twin simulation process, and form a high-dimensional feature set; The high-dimensional feature set is input into the LightGBM model, the model outputs the corresponding control parameter correction, and then the control parameter correction is processed using a buffer function. Features of real-time operating conditions and features from the digital twin simulation process are extracted and integrated to form a high-dimensional feature set.

[0013] Optionally, in the ninth implementation of the present invention, the construction of the grid-type energy storage digital twin involves extracting real-time operating condition features and features from the digital twin simulation process, and integrating them to form a high-dimensional feature set, including: By mapping the topology, operating parameters, and environmental conditions of the physical entity of the grid-type energy storage, a virtual simulation model corresponding to the physical entity is established. Real-time synchronization of physical entity operation data and mechanism model parameters to simulate the physical entity's operation response under different control parameters; Features of real-time operating conditions and features from the digital twin simulation process are extracted and integrated to form a high-dimensional feature set.

[0014] The technical solution provided by this invention involves collecting raw data from both the energy storage side and the grid side, performing wavelet threshold denoising, interpolation completion, and normalization preprocessing on the collected raw data to obtain a standardized time-series dataset. This standardized time-series dataset is then input into a hybrid model based on a mechanistic constraint-physical information fusion neural network, which outputs energy storage state parameters in real time. Based on the energy storage state parameters, an FCM clustering algorithm is used to divide the data into multiple operating domains, and a sparrow search algorithm is used to optimize the initial values ​​of control parameters in each operating domain. A network-type energy storage digital twin is constructed, and combined with the LightGBM model and buffer function, the initial values ​​of control parameters are dynamically corrected to obtain the real-time optimal control parameters. These real-time optimal control parameters are then sent to the controller for regulation. This invention achieves dynamic, precise, and adaptive adjustment of network-type energy storage control parameters, improving the transient stability, steady-state regulation accuracy, and broadband anti-interference capability of the power grid. Attached Figure Description

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0016] Figure 1 A schematic diagram of the first embodiment of the adaptive control method for grid-type energy storage parameters based on mechanism-data dual-drive provided by the present invention; Figure 2 A schematic diagram of the second embodiment of the adaptive control method for grid-type energy storage parameters based on mechanism-data dual-drive provided by the present invention; Figure 3 This is a schematic diagram of the structure of a mechanism-data dual-drive grid-type energy storage parameter adaptive control device provided in an embodiment of the present invention. Detailed Implementation

[0017] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0018] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 A schematic diagram of the first embodiment of the adaptive regulation method for grid-type energy storage parameters based on mechanism-data dual-drive provided by the present invention. The method specifically includes the following steps: Step 101: Collect raw data from the energy storage side and the grid side, and perform wavelet threshold denoising, interpolation completion, and normalization preprocessing on the collected raw data to obtain a standardized time series dataset; In this embodiment, full-dimensional raw operating data from both the energy storage side and the grid side are continuously and synchronously collected. The energy storage side includes parameters such as the active power, reactive power, port voltage, output current, operating temperature of the energy storage unit, state of charge, health status, and equivalent internal resistance. The grid side includes raw data such as the voltage amplitude at the point of common coupling, system operating frequency, regional load power consumption, line equivalent impedance, and grid-connected output power of new energy units. Then, a wavelet threshold denoising algorithm is used to perform data denoising processing. A wavelet basis function adapted to the characteristics of power time series data is pre-selected and a fixed number of decomposition layers is set. All the collected raw time series data are decomposed layer by layer by wavelet decomposition to separate low-frequency components that represent trend changes and high-frequency components that are mixed with random interference. A unified preset threshold is set according to the statistical characteristics of the data. The noise components mixed in the high-frequency components are screened and suppressed according to the threshold rules to retain effective high-frequency feature information. Then, the high-frequency components after threshold processing and the original low-frequency components are reconstructed by inverse wavelet transform to restore the denoised clean data after removing random noise interference. The clean, denoised time-series dataset is traversed, and each point and sequence is inspected sequentially to systematically identify data gaps, invalid data, and uncollected data points throughout the data chain. All identified missing locations are uniformly marked and recorded. Simultaneously, for each marked missing point, the sequence is searched and located precisely along the temporal order to pinpoint the immediately preceding and following valid data samples, determining the reference points required for interpolation calculations. Using these two located valid data samples as references, the data is then processed according to the time sequence. The continuous and gradual distribution characteristics of the ordered data establish a stable linear correlation between two valid data points. According to the proportion of missing points in the time interval between two valid samples, the corresponding change ratio is matched layer by layer. The specific numerical information corresponding to the missing point is determined according to the linear gradual change law. According to the judgment logic and numerical determination method of linear interpolation, batch calculation and numerical filling are performed on all previously marked missing points one by one. The data assignment operation of each missing position is completed one by one. After all points are processed, a complete dataset with no gaps, no omissions, and continuous temporal sequence is formed.

[0019] Normalization is performed on all the data samples after completion. The extreme value normalization method is uniformly adopted to map the running data with large differences in physical meaning, scale, and numerical range to the same numerical range, thereby eliminating the dimensional barriers and numerical span differences between different parameters and obtaining a standardized time series dataset with uniform format and aligned dimensions.

[0020] Step 102: Input the standardized time series dataset into the hybrid model based on the mechanistic constraint-physical information fusion neural network, and output the energy storage state parameters in real time; In this embodiment, the hybrid model of the mechanistic constraint-physical information fusion neural network includes an input layer, an encoder module, a mechanistic constraint embedding module, a decoder module, and an output layer. The encoder module includes a position encoding layer, a multi-layer attention layer, and a fully connected layer. The mechanistic constraint embedding module is located between the encoder and the decoder. The decoder module includes a decoding layer and an adaptive weight adjustment layer.

[0021] In this embodiment, the input layer of the hybrid model based on the mechanistic constraint-physical information fusion neural network specifically receives the standardized time-series dataset output by the preprocessing module. During the receiving process, the data undergoes preliminary verification, checking the data's dimension, temporal order, and numerical range to ensure that the input data conforms to the model's preset input specifications. Simultaneously, the received data is temporarily cached to prevent data loss, disorder, or temporal sequence errors during transmission. After the encoder module starts, it first uses its built-in position encoding layer to annotate the input standardized time-series dataset with temporal features. Since the standardized time-series data itself only contains numerical information and lacks temporal sequence association markers, the position encoding layer uses specific encoding logic to assign temporal features to each data point. The corresponding time location information is attached to clarify the chronological order and correlation of the data in the time series. Then, the time series data with location encoding is extracted layer by layer through a multi-layer stacked attention layer. Each attention layer will autonomously calculate the correlation weight between different data points, focusing on capturing the core data features that play a key role in the prediction of energy storage state parameters and filtering out irrelevant and redundant information. Finally, the multi-dimensional and scattered time series features extracted by the multi-layer attention layer are uniformly input into the fully connected layer. Through the neural operation of the fully connected layer, the multi-dimensional features are integrated and fused, and the feature dimensions are compressed and regularized. Finally, a high-dimensional time series feature vector with unified dimensions and concentrated features is output and passed to the mechanism constraint embedding module. After receiving the high-dimensional time-series feature vector output by the encoder module, the mechanism constraint embedding module immediately initiates the constraint screening process. First, it retrieves the preset physical constraints of the multi-scale coupling mechanism model for grid-type energy storage, including the relevant physical operating laws and boundary requirements of the bottom-layer electromagnetic transients, the middle-layer electromechanical transients, and the top-layer steady-state-aging coupling. These physical constraints are then transformed into judgment criteria that can be used for feature screening. Subsequently, the high-dimensional time-series feature vector is decomposed dimension by dimension, and the feature value of each dimension is compared one by one with the corresponding physical constraint judgment criteria to determine whether the feature value of that dimension conforms to the actual physical operating laws of the energy storage system and whether it is within the preset physical boundary range. Feature dimensions that exceed the physical boundary or do not meet the constraint requirements are specifically corrected and adjusted to a reasonable range. Feature dimensions that meet the constraint requirements are retained without changing their original values. After completing the constraint screening and correction for all dimensions, the features of each dimension are reintegrated to form a high-dimensional time-series feature vector that conforms to the physical mechanism constraints, and then transmitted to the decoder module. The decoder module receives the high-dimensional time-series feature vectors filtered by the mechanism constraint embedding module. First, it performs a reverse decoding operation on these feature vectors through its built-in decoding layer. Following the reverse logic of the encoder's feature extraction, the decoding layer analyzes the inherent meaning of the high-dimensional feature vectors layer by layer, gradually restoring the original data features and physical meaning corresponding to the feature vectors, thus transforming high-dimensional features into low-dimensional interpretable features. Next, an adaptive weight adjustment layer dynamically balances the data fitting accuracy with the physical constraint requirements. Based on the current feature decoding deviation, the weight allocation is adjusted in real time to ensure that the decoding result closely matches the true characteristics of the input data while remaining within the preset physical mechanism constraints, thus preventing the decoding result from deviating from the actual operating rules of the energy storage system. After decoding, a decoded feature vector with clear physical meaning is obtained and passed to the output layer. The output layer receives the decoded feature vector from the decoder module, performs final normalization and transformation on the feature vector, and transforms it into energy storage state parameters that can be directly used in subsequent optimization modules through built-in feature mapping logic. These parameters include key parameters such as energy storage equivalent internal resistance, virtual inertia reference value, damping coefficient reference value, and grid equivalent impedance. At the same time, the converted parameters are format-validated to ensure that the dimension and numerical range of the parameters meet the input requirements of subsequent modules, and finally completes the real-time output of energy storage state parameters.

[0022] Step 103: Based on the energy storage state parameters, the FCM clustering algorithm is used to divide the multiple operating domains, and the initial values ​​of the control parameters of each operating domain are optimized by the sparrow search algorithm. In this embodiment, based on the initialized cluster center vector, the fuzzy membership degree of each energy storage state parameter sample to the center vector of each cluster category is calculated one by one. The fuzzy membership degree is used to characterize the probability that the sample belongs to a certain cluster category. Through the preset membership degree calculation rules, the membership degree values ​​of each sample to all cluster categories are obtained and a membership degree matrix is ​​constructed. Then, according to the iteration rules of the FCM clustering algorithm, the center vector of each cluster category is updated according to the current membership degree matrix, and the membership degree matrix is ​​updated at the same time. During the iteration process, the goal is always to minimize the sum of squared distances between all samples and their corresponding cluster centers. The position of the center vector and the membership degree value are continuously adjusted and iterated continuously until the change in the cluster center vector is less than the preset threshold and no longer changes significantly. At this time, the iteration stops. Based on the maximum membership degree value of each sample, the working condition category to which the sample belongs is determined, and the division of the entire working condition domain is completed. For each operating range partitioned by the FCM clustering algorithm, a sparrow search algorithm is independently initiated to optimize the initial values ​​of control parameters. First, based on the operating characteristics and control requirements of the current operating range, a sparrow population is initialized, setting a reasonable population size, maximum number of iterations, and search range for control parameters. A set of control parameter combinations corresponding to each sparrow in the population is defined, with each individual representing a potential control parameter scheme. Then, according to the preset fitness evaluation criteria, combined with the relevant requirements of transient power angle stability, steady-state power balance, and broadband damping suppression, the fitness value of each sparrow is calculated. Producers and chasers in the population are selected based on their fitness values. Followers and watchdogs: Producers are responsible for exploring new parameter regions within a pre-defined search space to find better combinations of control parameters. Followers follow the producer's search trajectory to perform local fine-grained searches to further optimize parameter accuracy. Watchdogs monitor the population's search status in real time. When they find that the population search has fallen into a local optimum and the fitness value no longer increases, they promptly trigger their own position update to escape the local optimum. This process continues for multiple iterations until the pre-defined maximum number of iterations is reached, or the overall fitness value of the population tends to stabilize and no longer changes significantly. At this point, the optimal combination of control parameters with the best fitness value in this operating domain is output as the initial value of the control parameters for this operating domain.

[0023] Step 104: Construct a grid-type energy storage digital twin, combine the LightGBM model and buffer function to dynamically correct the initial values ​​of control parameters, obtain the real-time optimal control parameters, and send the real-time optimal control parameters to the controller for regulation.

[0024] In this embodiment, a digital twin that is completely mirrored with the physical entity of the grid-type energy storage is constructed to fully replicate the equipment topology, electrical characteristics, control logic and grid access environment. The physical side operating status and mechanism parameters are synchronized in real time. Multi-condition simulation is carried out in virtual space. During the entire simulation process, the state change characteristics, parameter response characteristics and operating condition correlation characteristics in the simulation evolution process are systematically captured. Various simulation characteristics are collected and organized, arranged in a regular manner according to the feature dimensions, and uniformly collected to form a high-dimensional feature set with complete dimensions and close correlation. The completed high-dimensional feature set is fully imported into the trained LightGBM model. Relying on the model's internal multi-decision tree ensemble and gradient boosting learning mechanism, the intrinsic correlation mapping relationship between high-dimensional features and control parameter deviations is automatically discovered. Based on the distribution state and working condition attributes of the input features, the model autonomously infers and outputs the required control parameter correction amount under the current working condition. Subsequently, a buffer function mechanism is introduced to smooth the transition of the control parameter correction amount output by the model, gradually releasing the correction magnitude according to the incremental adjustment logic, and constraining and buffering the instantaneously changing parameter correction amount to keep the parameter adjustment process continuous and smooth. Real-time operating features are extracted from on-site operational data from the grid and energy storage sides, covering actual on-site operating features such as voltage deviation, frequency dynamic change rate, energy storage state of charge, health degradation status, and real-time power fluctuation rate. At the same time, process features, response features, and state evolution features generated throughout the entire process of digital twin simulation are retrieved. The on-site real-time operating features and digital twin simulation features are fused across sources, aligned in dimensions, and normalized. Redundant duplicated information is removed, effective core features are retained, and integrated and spliced ​​to construct a standardized high-dimensional feature set.

[0025] Please see Figure 2 A schematic diagram of the second embodiment of the adaptive regulation method for grid-type energy storage parameters based on mechanism-data dual-drive provided by the present invention. The method includes: Step 201: Set at least four clustering categories, including strong grid steady state, weak grid transient state, fault disturbance, and energy storage aging, and initialize the center vector of each clustering category; In this embodiment, all energy storage state parameters are acquired and used as the core basis for operating condition classification. The FCM clustering algorithm is started, and four basic clustering categories are clearly defined based on the actual operation scenarios of grid-type energy storage and the characteristics of power grid environment changes. These categories correspond to four typical operating conditions: strong grid steady state, weak grid transient state, fault disturbance, and energy storage aging. At the same time, based on the general operating characteristics of each type of operating condition, energy storage state parameter samples that meet the characteristics of the corresponding operating condition are randomly selected as initial cluster centers. A corresponding center vector is initialized for each cluster category to ensure that the initial center vector can initially reflect the core characteristics of each type of operating condition.

[0026] Step 202: Calculate the fuzzy membership degree of each energy storage state parameter sample to each cluster center, iteratively update the cluster center and membership degree matrix, minimize the sum of squared distances between the sample and the corresponding cluster center, until the cluster center no longer changes significantly, and determine the operating condition category to which each sample belongs; Step 203: For each of the divided operating conditions, the initial values ​​of the control parameters for each operating condition are obtained by optimizing the sparrow search algorithm.

[0027] In this embodiment, the population initialization operation is completed according to the sparrow search algorithm. A reasonable search range for various control parameters of the grid-type energy storage is predefined, and a fixed number of individuals and a maximum number of iterations are set. Initially, evenly distributed sparrow individuals are randomly generated within the preset parameter search space. Each sparrow individual within the population independently corresponds to a complete set of energy storage control parameters, and all individuals together constitute a complete initial sparrow population. All sparrow individuals within the population are traversed, and according to preset operating condition adaptation and control performance evaluation criteria, the fitness value corresponding to each sparrow individual is calculated one by one. Based on the ranking rules of fitness values, the entire sparrow population is classified, and producer individuals, follower individuals, and vigilant individuals within the population are screened and distinguished. Producers, relying on their advantageous positions, conduct large-scale global exploration within the overall parameter search space, actively searching for untraversed new parameter regions. Followers follow the position trajectory of advantageous producers to conduct close-range, fine-grained local searches, performing fine-grained traversals around the neighborhood of high-quality parameters. Vigilant individuals monitor the entire population's search progress and positional change trends in real time, continuously perceiving the overall optimization dynamics of the population. During continuous monitoring of the population's search status, the monitor continuously identifies whether the population exhibits stagnation, such as clustering or a prolonged cessation of fitness improvement. If the population's optimization process is determined to be trapped in a local optimum, the monitor immediately triggers the individual sparrow position update mechanism. Following the algorithm's built-in position update rules, some individuals are repositioned, allowing the population to escape the current limited search area and enter a new parameter search range to continue the optimization process. The monitor iterates through the sparrow search algorithm's logic, performing population position updates, fitness recalculation, population role reclassification, and local optimum determination and exit procedures, completing iterative optimization operations round by round. This process is repeated until the algorithm reaches the pre-set maximum number of iterations. After terminating the iteration loop, the monitor selects the parameter combination corresponding to the sparrow with the best fitness from the final population and outputs it as the initial value of the control parameters for the current operating domain.

[0028] Figure 3This is a schematic diagram of the structure of a mechanism-data dual-drive grid-based adaptive energy storage parameter control device 300 provided in an embodiment of the present invention. This mechanism-data dual-drive grid-based adaptive energy storage parameter control device 300 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors) and a memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and storage media 330 can be temporary or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the diagram), each module may include a series of instruction operations on the mechanism-data dual-drive grid-based adaptive energy storage parameter control device 300. Furthermore, the processor 310 may be configured to communicate with the storage media 330 and execute a series of instruction operations in the storage media 330 on the mechanism-data dual-drive grid-based adaptive energy storage parameter control device 300 to implement the method provided in the above embodiment.

[0029] The mechanism-data dual-drive network-based adaptive regulation device 300 for energy storage parameters may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The mechanism-data dual-drive grid-type energy storage parameter adaptive control device structure shown does not constitute a limitation on the computer device provided by the present invention. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0030] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the various steps of the mechanism-data dual-drive grid-type energy storage parameter adaptive control method provided in the above embodiments.

[0031] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described equipment or apparatus / unit can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0032] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0033] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A mechanism-data dual-driven adaptive regulation method for grid-type energy storage parameters, characterized in that, The method includes the following steps: Raw data from the energy storage side and the grid side are collected, and wavelet threshold denoising, interpolation completion, and normalization preprocessing are performed on the collected raw data to obtain a standardized time series dataset. The standardized time-series dataset is input into a hybrid model based on a mechanistic constraint-physical information fusion neural network, which outputs energy storage state parameters in real time. Based on the energy storage state parameters, the FCM clustering algorithm is used to divide multiple operating domains, and the initial values ​​of control parameters in each operating domain are optimized by the sparrow search algorithm. A grid-type energy storage digital twin is constructed. By combining the LightGBM model and buffer function, the initial values ​​of control parameters are dynamically corrected to obtain the real-time optimal control parameters, which are then sent to the controller for regulation.

2. The adaptive control method for grid-type energy storage parameters based on mechanism-data dual-drive as described in claim 1, characterized in that, The raw data collected from the energy storage side and the grid side are subjected to wavelet threshold denoising, interpolation completion, and normalization preprocessing to obtain a standardized time-series dataset, including: Raw data from the energy storage side and the grid side are collected, and wavelet decomposition is performed on the raw data to obtain high-frequency and low-frequency components. Noise signals in the high-frequency components are suppressed by a preset threshold. Then, the processed high-frequency and low-frequency components are subjected to inverse wavelet transform to obtain clean data after denoising. Missing values ​​in the denoised clean data are filled in using a linear interpolation algorithm. The filled data is then normalized to obtain a standardized time series dataset.

3. The adaptive control method for grid-type energy storage parameters based on mechanism-data dual-drive as described in claim 2, characterized in that, The step of using a linear interpolation algorithm to fill in missing values ​​in the denoised clean data includes: Traverse the clean data after noise reduction, mark the locations of missing information in the data sequence, and locate the two valid complete data samples before and after each missing point in the data sequence. A linear correlation is established based on two consecutive valid data points, and the numerical information corresponding to the missing positions is determined according to the interval ratio. Complete all missing points in the marked data to obtain the completed data.

4. The adaptive control method for grid-type energy storage parameters based on mechanism-data dual-drive as described in claim 1, characterized in that, The hybrid model of the mechanistic constraint-physical information fusion neural network includes an input layer, an encoder module, a mechanistic constraint embedding module, a decoder module, and an output layer. The encoder module includes a position encoding layer, multiple attention layers, and a fully connected layer. The mechanistic constraint embedding module is located between the encoder and the decoder. The decoder module includes a decoding layer and an adaptive weight adjustment layer.

5. The adaptive regulation method for grid-type energy storage parameters based on mechanism-data dual-drive as described in claim 4, characterized in that, The process of inputting standardized time-series datasets into a hybrid model based on a mechanistic constraint-physical information fusion neural network to output energy storage state parameters in real time includes: The input layer of a hybrid model based on a mechanistic constraint-physical information fusion neural network receives a standardized time-series dataset. The encoder module first adds temporal feature information to the standardized time series dataset through the position encoding layer, and then extracts features from the data through multiple attention layers. The extracted multi-dimensional temporal features are input into the fully connected layer for integration and dimensionality reduction to output a high-dimensional temporal feature vector. The mechanism constraint embedding module performs constraint screening on the high-dimensional time series feature vector. The decoder module receives the screened high-dimensional time series feature vector, first performs reverse decoding on the feature vector through the decoding layer, and then dynamically balances the data fitting accuracy and physical constraint requirements through the adaptive weight adjustment layer to obtain the decoded feature vector. The output layer converts the decoded feature vector into energy storage state parameters.

6. The adaptive control method for grid-type energy storage parameters based on mechanism-data dual-drive as described in claim 5, characterized in that, The mechanism constraint embedding module performs constraint filtering on high-dimensional time-series feature vectors, including: The high-dimensional time series feature vector is decomposed dimension by dimension, and each decomposed dimension is compared with the corresponding physical constraints to determine whether the feature value of the current dimension is within the preset physical boundary range. Based on physical constraints, the feature dimensions that exceed the physical boundary are corrected, and the corrected dimensional features are re-integrated to form a filtered high-dimensional time-series feature vector.

7. The adaptive control method for grid-type energy storage parameters based on mechanism-data dual-drive as described in claim 1, characterized in that, The process involves dividing the energy storage state parameters into multiple operating domains using the FCM clustering algorithm and optimizing the initial values ​​of control parameters for each operating domain using the sparrow search algorithm, including: Define at least four clustering categories, including strong grid steady state, weak grid transient state, fault disturbance, and energy storage aging, and initialize the center vector of each clustering category; Calculate the fuzzy membership degree of each energy storage state parameter sample to each cluster center, iteratively update the cluster center and membership degree matrix, minimize the sum of squared distances between the sample and the corresponding cluster center, until the cluster center no longer changes significantly, and determine the operating condition category to which each sample belongs; For each of the divided operating conditions, the initial values ​​of the control parameters for each operating condition are obtained by optimization using the sparrow search algorithm.

8. The adaptive control method for grid-type energy storage parameters based on mechanism-data dual-drive as described in claim 7, characterized in that, For each of the divided operating conditions, the initial values ​​of the control parameters for each operating condition are obtained by optimization using a sparrow search algorithm, including: Initialize a sparrow population, where each individual in the population corresponds to a set of control parameters; The fitness value of each sparrow individual is calculated to screen out producers, followers and vigilants in the population. Producers are responsible for exploring new parameter regions in the search space, followers follow producers to conduct local searches, and vigilants are responsible for monitoring the search status of the population. When the search is found to be trapped in a local optimum, a position update is triggered to escape the local optimum region; Through multiple iterations until the preset number of iterations is reached, the initial values ​​of the control parameters for each operating condition domain are output.

9. The adaptive control method for grid-type energy storage parameters based on mechanism-data dual-drive as described in claim 1, characterized in that, The constructed grid-type energy storage digital twin, combined with the LightGBM model and buffer function, dynamically corrects the initial values ​​of control parameters to obtain real-time optimal control parameters, and then sends these real-time optimal control parameters to the controller for regulation, including: Construct a grid-type energy storage digital twin, extract features from the digital twin simulation process, and form a high-dimensional feature set; The high-dimensional feature set is input into the LightGBM model, the model outputs the corresponding control parameter correction, and then the control parameter correction is processed using a buffer function. Features of real-time operating conditions and features from the digital twin simulation process are extracted and integrated to form a high-dimensional feature set.

10. The adaptive control method for grid-type energy storage parameters based on mechanism-data dual-drive as described in claim 9, characterized in that, The construction of a grid-type energy storage digital twin involves extracting real-time operating condition features and features from the digital twin simulation process, integrating them to form a high-dimensional feature set, including: By mapping the topology, operating parameters, and environmental conditions of the physical entity of the grid-type energy storage, a virtual simulation model corresponding to the physical entity is established. Real-time synchronization of physical entity operation data and mechanism model parameters to simulate the physical entity's operation response under different control parameters; Features of real-time operating conditions and features from the digital twin simulation process are extracted and integrated to form a high-dimensional feature set.

Images