Energy storage power station distributed situation assessment method and system based on multi-agent model

Abstract

The application discloses a method and system for distributed situation assessment of energy storage power stations based on a multi-agent model, wherein a module-level battery management system of a battery module is used as a modeling layer agent, and an energy management system of an energy storage converter is used as an assessment layer agent; monitoring data of various working conditions of any modeling layer agent is filtered by using monitoring data of two adjacent modeling layer agents of the modeling layer agent to obtain fused monitoring data of the modeling layer agent in various working conditions; the modeling layer agent identifies characteristic parameters based on a battery module P2D model by using the fused monitoring data in various working conditions; the modeling layer agent constructs an operating state feature matrix of each battery module by using the characteristic parameters, time sequence features and frequency sequence features of real-time monitoring data; and the assessment layer agent performs situation assessment on the battery module. The application solves the problems of poor accuracy of battery energy storage power station operating situation assessment, insufficient model interpretability, high communication bandwidth and algorithmic power requirements and the like.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of new power system situation assessment, and in particular, relates to a distributed situation assessment method and system for energy storage power stations based on a multi-agent model. BACKGROUND

[0002] With the penetration of high proportions of renewable energy and the transformation of high power electronics in new power systems, the dynamic balance mechanism of power systems is facing profound changes. The strong volatility, anti-peaking characteristics and low inertia characteristics of renewable energy generation make it difficult to continue the traditional "source following load" mode, and it is urgent to realize the "source network load storage" collaborative interaction of large-scale flexible regulation resources. Battery energy storage power stations, with their millisecond response speed, multi-time scale energy throughput capacity and flexible configuration characteristics, have become a key technology carrier for solving new energy consumption bottlenecks and improving the frequency / voltage support capability of power grids, significantly improving the flexibility and economy of power systems. In addition, with the continuous decline in the cost of electrochemical energy storage and policy-driven, large-scale deployment of energy storage power stations has become an important path to achieve the "double carbon" goal.

[0003] In the prior art, for the operation and management of a battery energy storage system, real-time monitoring of its operating state is the basis for ensuring safe, efficient and reliable operation of the system. The main monitoring objects include batteries (single cells, battery packs and battery clusters, etc.), and the key monitoring data types include voltage, temperature, charging and discharging current, and remaining capacity, etc. The above monitoring data not only reflects the immediate working state of the energy storage system, but also provides an important basis for fault warning, performance evaluation and life cycle management. Through in-depth analysis of the monitoring data, intelligent management of the battery energy storage system can be achieved, the charging and discharging strategy can be optimized, the service life of the battery can be prolonged, and the economy and safety of the system can be improved. Therefore, building a perfect situation assessment system has important practical significance for promoting the application and development of battery energy storage technology.

[0004] Overall, the existing battery energy storage power station situation assessment technology based on monitoring data feature analysis has the following deficiencies: 1) the existing situation assessment methods are mostly full-data driven models, which lack physical mechanism support and have poor model interpretability; 2) the existing situation assessment methods are mostly based on centralized analysis mode, which is difficult to meet the communication bandwidth and calculation timeliness for large-scale battery energy storage systems. SUMMARY

[0005] To solve the problems in the prior art, the application provides a distributed situation assessment method and system for an energy storage power station based on a multi-agent model, a multi-level multi-agent model including a modeling layer, an assessment layer and a decision layer is established, the distributed situation awareness of the energy storage system is realized through collaborative analysis of the battery module and introduction of physical mechanism constraints in the data-driven model, which is different from the existing centralized energy storage operation situation awareness, and solves the problems of poor accuracy of the battery energy storage power station operation situation assessment, insufficient model interpretability, high communication bandwidth and computing power requirements, and provides a technical solution for intelligent operation and maintenance of the battery energy storage power station.

[0006] The application adopts the following technical scheme.

[0007] The application provides a distributed situation assessment method for an energy storage power station based on a multi-agent model, the multi-agent model includes a modeling layer, an assessment layer and a decision layer, and the method includes the following steps.

[0008] The module-level battery management system of the battery module is taken as the modeling layer agent, and the connection relationship of the battery module is mapped as the structural relationship of the modeling layer agent; the energy management system of the energy storage converter is taken as the assessment layer agent, and the connection relationship of the energy storage converter is mapped as the structural relationship of the assessment layer agent; the connection relationship of the battery module and the energy storage converter is mapped as the structural relationship of the modeling layer agent and the assessment layer agent; the decision layer agent receives the state information of each assessment layer agent;

[0009] For any modeling layer agent, the monitoring data of the modeling layer agent in various working conditions is filtered by using the monitoring data of the adjacent two modeling layer agents, and the fusion monitoring data of the modeling layer agent in various working conditions is obtained; each modeling layer agent uses the fusion monitoring data in various working conditions to identify the characteristic parameters representing the aging state of the battery module based on the P2D model of the battery module; each modeling layer agent uses the identified characteristic parameters, the time sequence features of the real-time monitoring data and the frequency sequence features of the real-time monitoring data to construct the operating state feature matrix of each battery module; the assessment layer agent obtains the operating state feature matrix of each battery module and performs situation assessment on the battery module.

[0010] The connection relationship of the battery module is mapped as the structural relationship of the modeling layer agent, which includes the following steps.

[0011] The modeling layer agents corresponding to the battery modules connected in series are located in the same modeling layer; the modeling layer agents corresponding to the battery modules connected in parallel are located in different modeling layers; the modeling layer agent with the smallest communication overhead in the assessment layer agent in each modeling layer is defined as the head-end agent;

[0012] In terms of communication, each modeling layer agent can only interact with adjacent agents except the head agent of the first end, the adjacent modeling layer agents store the monitoring data of the battery module under the same working condition, and the communication overhead of the adjacent modeling layer agents is minimum; the head agent of each modeling layer interacts with the adjacent modeling layer agents in the modeling layer, and interacts with the head agent and the evaluation layer agent in the adjacent modeling layer.

[0013] The monitoring data of each modeling layer agent is obtained from the corresponding battery management system, including the voltage of a group of series connected cells in the battery module, the cell current of all parallel points, and the temperature;

[0014] After the monitoring data of the modeling layer agent is normalized, the SOC of the battery module constitutes the monitoring sequence of the modeling layer agent.

[0015] The wavelet packet transform method and the sliding window interpolation method are used to unify the time axis and the sampling frequency of the monitoring sequence of the modeling layer agent in the same modeling layer, and a new monitoring data sequence is obtained;

[0016] The distance between the new monitoring data sequence of the two modeling layer agents adjacent to the modeling layer agent and the new monitoring data sequence of the modeling layer agent, and the weighted sum of the new monitoring data sequence of the modeling layer agent are used as the fused monitoring data of the modeling layer agent after sliding window filtering.

[0017] The window sequence obtained by windowing the monitoring data of the two modeling layer agents in the same modeling layer Wavelet packet decomposition is performed to obtain the window sequence The first layer decomposition under the first layer decomposition under the first layer decomposition under the first layer decomposition under the first layer decomposition under the first

[0018] layer decomposition under the first layer decomposition under the first layer decomposition under the first layer decomposition under the first layer decomposition under the first layer decomposition under the first layer decomposition under the first layer decomposition under the first layer decomposition under the first layer decomposition under the first layer decomposition under the first layer decomposition under the first layer decomposition under the first layer decomposition under the first

[0019] The correlation coefficient is obtained by using the correlation coefficient ; the cross-correlation coefficient is calculated by using the correlation coefficient In the first Sub-band residual under layer decomposition ; update sub-band coefficients using sub-band residual and cross-correlation coefficient ;

[0020] Based on the updated sub-band coefficients , the wavelet packet inverse transform method is used to obtain the intermediate window sequence with the same sampling frequency and time axis as the window sequence ; the window sequence and the intermediate window sequence are obtained by using the sliding window interpolation method to obtain the new monitoring data sequence with the same sampling frequency and time axis as the window sequence , as follows:

[0021]

[0022] In the formula, is the data of the intelligent agent in the new monitoring data sequence at time , , is the interpolation coefficient, and satisfies , , , is the interpolation calculated using the monitoring data at one time point before and after in the window sequence , is the interpolation calculated using the monitoring data at one time point before and after in the intermediate window sequence .

[0023] Fusion monitoring data after sliding window filtering , as follows:

[0024]

[0025] In the formula, is the fusion monitoring data after sliding window filtering, , , is the filtering parameter, and the value range is , and satisfies , is the distance between the new monitoring data sequence of the first battery module and the new monitoring data sequence of the first battery module, is the distance between the new monitoring data sequence New monitoring data sequence of battery module group and the first New monitoring data sequence of battery module group distance.

[0026] Based on the P2D model of the battery module, the to-be-recognized parameters for characterizing the aging state of the battery module include: solid-phase diffusion coefficient, effective reaction rate constant, liquid-phase diffusion coefficient, effective conductivity, current collector contact resistance, and active material volume fraction;

[0027] An offline parameter recognition model is established based on a long short-term memory network and a physical information neural network;

[0028] The first The parameters for characterizing the aging state of the battery module group are recognized, and the first The modeling layer agent The fusion monitoring data of each working condition of the adjacent modeling layer agent , The fusion monitoring data of each working condition of the adjacent modeling layer agent

[0029] In the first stage of recognition, different characteristic parameters are recognized according to the working conditions corresponding to the fusion monitoring data of the modeling layer agent , , The effective conductivity and the current collector contact resistance are recognized when the fusion monitoring data corresponds to the hybrid pulse power characteristic working condition; the solid-phase diffusion coefficient is recognized when the fusion monitoring data corresponds to the constant-current charge and discharge working condition; the active material volume fraction is recognized when the fusion monitoring data corresponds to the constant-voltage charge and discharge working condition; and the effective reaction rate constant and the liquid-phase diffusion coefficient are obtained when the fusion monitoring data corresponds to the dynamic stress test working condition;

[0030] In the second stage of recognition, the first stage of recognition results are optimized by using the fusion monitoring data of the normal working condition of the modeling layer agent , , And the typical data of the battery module aging process, to obtain the characteristic parameters for characterizing the aging state of the battery.

[0031] The real-time monitoring data of the first The battery module group is collected and discretely processed to obtain a discretized monitoring data sequence , , is the number of sampling points;

[0032] Constructing mechanism feature matrix As follows:

[0033]

[0034] wherein, is the normalized effective conductivity of the sampling point , is the normalized current collector contact resistance of the sampling point , is the normalized solid phase diffusion coefficient of the sampling point , is the normalized active material volume fraction of the sampling point , is the effective reaction rate constant of the sampling point , is the normalized liquid phase diffusion coefficient of the sampling point ;

[0035] Constructing timing feature matrix As follows:

[0036]

[0037] wherein, is the normalized dynamic time warping distance of the sampling point , is the normalized incremental capacity of the sampling point , is the normalized differential voltage of the sampling point , is the normalized distance metric of the sampling point , is the normalized Wasserstein distance of the sampling point , is the ratio of the output voltage of the P2D equivalent model to the actual voltage of the sampling point ;

[0038] Constructing frequency domain feature matrix As follows:

[0039]

[0040] wherein, , is the energy of the second and third modal of the sampling point , , is the kurtosis of the second and third modal of the sampling point , , is the distance of the second and third modal of the sampling point sample entropy of the second and third modalities of the first modality;

[0041] The mechanism feature matrix, the time sequence feature matrix and the frequency domain feature matrix are spliced to obtain a running state feature matrix.

[0042] Based on the connection relationship of each battery module, the evaluation layer agent splices the running state feature matrix of each battery module to obtain a series-parallel module feature matrix.

[0043] A variational Bayes feature evaluation network based on a coding and decoding structure is established; the network outputs a running state probability matrix of the evaluation layer agent according to the series-parallel module feature matrix, and the row vector of the running state probability matrix is a battery module running state evaluation result vector corresponding to each modeling layer agent, and the elements in the battery module running state evaluation result vector include the evaluation index values and confidence degrees of the SOC, SOH and RUL of the battery module corresponding to the modeling layer agent.

[0044] A distributed situation evaluation system of an energy storage power station based on a multi-agent model, the multi-agent model comprising a modeling layer, an evaluation layer and a decision layer; the system comprising:

[0045] The multi-agent model establishment module is used for taking the module-level battery management system of the battery module as the modeling layer agent, mapping the connection relationship of the battery module as the structural relationship of the modeling layer agent; taking the energy management system of the energy storage converter as the evaluation layer agent, mapping the connection relationship of the energy storage converter as the structural relationship of the evaluation layer agent; mapping the connection relationship of the battery module and the energy storage converter as the structural relationship of the modeling layer agent and the evaluation layer agent; the decision layer agent receives the state information of each evaluation layer agent;

[0046] The situation evaluation module is used for, for any modeling layer agent, filtering the monitoring data of various working conditions of the modeling layer agent by using the monitoring data of two adjacent modeling layer agents to obtain the fusion monitoring data of various working conditions of the modeling layer agent; each modeling layer agent identifies the characteristic parameters representing the aging state of the battery module based on the P2D model of the battery module by using the fusion monitoring data of various working conditions; each modeling layer agent constructs the running state feature matrix of each battery module by using the identified characteristic parameters, the time sequence features of real-time monitoring data and the frequency sequence features of real-time monitoring data; the evaluation layer agent obtains the running state feature matrix of each battery module and performs situation evaluation on the battery module.

[0047] The application also relates to a terminal comprising a processor and a storage medium; the storage medium is used for storing instructions; the processor is used for operating according to the instructions to execute the steps of the method.

[0048] The application is also a computer readable storage medium, which stores a computer program, and the program is executed by a processor to realize the steps of the method.

[0049] The application has the advantages that, compared with the prior art, at least, the method provided by the application uses a multi-level multi-agent model to realize distributed situation awareness of the energy storage system, reduces the communication bandwidth requirement, reduces the influence of single point failure on the situation awareness of the energy storage power station, and improves the confidence and interpretability of the situation assessment result.

[0050] The application divides the to-be-identified parameters of the battery module into two stages for prediction, identifies specific parameters under specific working conditions in the first stage, reduces the coupling relationship between the remaining to-be-optimized parameters and the parameters, optimizes the identified parameters again in the second stage based on general operating conditions, and improves the model robustness and parameter identification accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 A flowchart of the battery energy storage power station distributed situation assessment method based on the multi-agent model provided by the application is shown in the figure.

[0052] Figure 2 A multi-level multi-agent model topology diagram of the battery energy storage power station provided by the application is shown in the figure.

[0053] Figure 3 A flowchart of the two-stage battery aging parameter offline identification method and correction strategy of the application is shown in the figure.

[0054] Figure 4 A variational Bayes feature evaluation network model based on the encoding and decoding structure of the application is shown in the figure. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical scheme and advantages of the application clearer, the technical scheme of the application will be described clearly and completely in combination with the drawings in the embodiments of the application. The embodiments described in the application are only a part of the embodiments of the application, not all the embodiments. Based on the spirit of the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0056] The multi-agent model comprises a modeling layer, an evaluation layer and a decision layer; each layer is composed of multiple autonomous agents, each agent has the ability of perception, memory, planning, decision-making and action, and independently acts without external direct control according to the set target; meanwhile, the agents can interact and cooperate with each other to achieve common goals or tasks; the agents are distributed at different positions and communicate and cooperate through a network or other communication channels, and have distributed characteristics, therefore, the application provides a distributed situation evaluation method for a battery energy storage station based on a multi-agent model, as shown in Figure 1 The method comprises the following steps:

[0057] In step S10, a multi-layer (modeling layer, evaluation layer and decision layer) multi-agent model is constructed according to the topological relationship of the battery module in the battery energy storage station.

[0058] Specifically, the module-level battery management system of the battery module is taken as the modeling layer agent, and the connection relationship of the battery module is mapped as the structural relationship of the modeling layer agent; the energy management system of the energy storage converter is taken as the evaluation layer agent, and the connection relationship of the energy storage converter is mapped as the structural relationship of the evaluation layer agent; the connection relationship of the battery module and the energy storage converter is mapped as the structural relationship of the modeling layer agent and the evaluation layer agent; the decision layer agent receives the state information of each evaluation layer agent.

[0059] The application proposes to construct the modeling layer, the evaluation layer and the decision layer of the multi-agent model according to the topological relationship between the module-level battery management system of the battery module and the energy management system of the energy storage converter in the battery energy storage station, which is used to describe the basic rules of information interaction between the BMMS of the battery module and the EMS of the energy storage converter and the basic law of the dynamic response process in the battery energy storage station; the module-level battery management system (Battery module management system, BMMS) of the battery module is taken as a node to establish the modeling layer agent; the energy management system (Energy management system, EMS) of the energy storage converter is taken as a node to establish the evaluation layer agent; the decision layer agent is an exogenous system, the decision layer agent receives the state information of each evaluation layer agent, and the decision layer agent is embedded in the existing operation system of the battery energy storage station in the form of API;

[0060] In addition, the modeling layer agent, the evaluation layer agent and the decision layer agent realize data interaction based on an undirected graph, which comprises the following steps:

[0061] 1) For the battery module connected in series, each modeling layer agent can only realize data interaction with adjacent modeling layer agents; for the battery module connected in parallel, all modeling layer agents share data; the adjacent modeling layer agents store the monitoring data under the same working condition collected by the BMMS, and the communication overhead of the adjacent modeling layer agents is minimum;

[0062] 2) The evaluation layer agent realizes bidirectional data interaction with adjacent evaluation layer agents;

[0063] 3) All evaluation layer agents can unidirectionally send data to the decision layer agent.

[0064] The multi-level multi-agent model constructed by the application can be used for a battery energy storage power station connected in parallel on the DC side or connected in parallel on the AC side. Taking a battery energy storage power station connected in parallel on the DC side as an example, it is assumed that the battery energy storage power station comprises k groups of battery energy storage cabinets, each group of battery energy storage cabinets is connected in parallel to a DC bus through a bidirectional DC / DC energy storage converter, and then connected to an external power grid through a DC / AC inverter and a transformer; correspondingly, each DC / DC energy storage converter is provided with an energy management system (EMS) for collecting monitoring data such as voltage and current of the battery energy storage cabinet; in each battery energy storage cabinet, a cabinet-level battery management system (BCMS) is arranged for collecting monitoring data such as average current, voltage and temperature of each battery module in the energy storage cabinet, and calculating overall state of charge (SOC), state of health (SOH), state of power (SOP) and state of energy (SOE) of the battery energy storage cabinet; each battery energy storage cabinet comprises n groups of series-connected battery modules, each group of series-connected battery modules comprises m groups of series-connected battery modules, each battery module comprises p series-connected and q parallel-connected battery cells, each battery module is provided with a module-level battery management system (BMMS) for collecting and calculating monitoring information such as voltage U, current I, capacity , internal resistance and temperature of each battery module;

[0065] For the battery energy storage power station connected in parallel on the DC side, the communication topology between the agents is described by a directed graph, and a multi-level multi-agent model as shown in Figure 2 is constructed.

[0066] Firstly, taking the BMMS of the battery module as the node, the modeling layer agent is established , The modeling layer agents corresponding to the battery modules connected in series are located in the same modeling layer, Figure 2 , , The modeling layer agents corresponding to the battery modules connected in parallel are located in different modeling layers, Figure 2 , , , are located in the 1st modeling layer, …, the m+1th modeling layer, …, the nm-m+1th modeling layer respectively; the modeling layer agent with the minimum communication overhead with the evaluation layer agent in each modeling layer is defined as the head-end agent, and the modeling layer agent with the maximum communication overhead with the evaluation layer agent in each modeling layer is defined as the tail-end agent; in terms of communication, each modeling layer agent except the head-end agent can only interact data with adjacent agents , , and the tail-end agent only has one adjacent modeling layer agent that can interact data, while the head-end agent in each modeling layer can interact data with the head-end agent in the adjacent modeling layer and the evaluation layer agent in addition to the adjacent modeling layer agent in the modeling layer, and the data sharing of the battery modules connected in parallel is realized through the data interaction between the head-end agents of the modeling layers; in terms of function, the modeling layer agent model obtains the voltage of the inner series connection battery cell, the current of the parallel point battery cell and the temperature sensing monitoring data from the module-level battery management system of the corresponding battery module, and realizes the data-level multi-modal monitoring data fusion modeling based on the data cleaning and feature modeling method.

[0067] Then, taking the EMS of the energy storage converter as the node, the evaluation layer agent is established , In terms of communication, the evaluation layer agent can interact information with adjacent evaluation layer agents , , and in addition, all the evaluation layer agents can unidirectionally send information to the decision layer agent ; in terms of function, the evaluation layer agent is responsible for the feature matrix remodeling analysis of the modeling layer agent, and realizes the operation state probability evaluation of the battery module of the energy storage power station based on the operation state probability interval evaluation method on the basis of the feature-level fusion of the monitoring data.

[0068] Finally, the decision layer agent is an exogenous system, and this layer agent receives the information sent by the evaluation layer agents​ State information is embedded in the operation system of the existing battery energy storage power station in the form of API or the like.

[0069] In step S20, monitoring data is collected based on a system such as a BMS and is cleaned and preprocessed; the main object of step S20 is the modeling layer multi-agent;

[0070] Specifically, for any modeling layer agent, the monitoring data of the modeling layer agent under various operating conditions is filtered using the monitoring data of the two adjacent modeling layer agents, to obtain the fusion monitoring data of the modeling layer agent under various operating conditions.

[0071] Specifically, in the frequency domain, the monitoring data of the modeling layer agents in the same modeling layer is Unified time axis and sampling frequency are obtained to obtain the new monitoring data sequence of each modeling layer agent In the time domain, the new monitoring data sequence of any modeling layer agent is The new monitoring data sequence of the modeling layer agent is filtered using the new monitoring data sequence of the two adjacent modeling layer agents of the modeling layer agent 、 to obtain the fusion monitoring data sequence of the modeling layer agent .

[0072] Each battery module is provided with a BMMS, which is responsible for sampling and outputting monitoring data inside the module. In the prior art, the monitoring data inside the module is used to clean and preprocess the monitoring data of the battery module. Based on the construction of a multi-level multi-agent model, the communication and working framework of the multi-agent system are fully utilized to clean and preprocess the monitoring data of the modeling layer agent using the monitoring data of the adjacent modeling layer agent. The adjacent modeling layer agent stores the monitoring data under the same operating condition collected by the BMMS, and the communication overhead of the adjacent modeling layer agent is minimal. From the frequency domain and the time domain, a self-adaptive cleaning and preprocessing of the same type of operating condition data is realized

[0073] The monitoring data of the modeling layer agent is obtained from the module-level battery management system, including but not limited to: the voltage of a group of series-connected cells in the battery module under various operating conditions, the cell current at all parallel points, and the temperature; in the embodiment, the cells connected in series in the battery module are taken as a group, a sensor is provided for each group to collect the cell voltage of the group, and a sensor is provided at the parallel point of all groups of cells to collect the cell current and temperature at the parallel point. The sum of all series-connected cell voltages is the battery module voltage; the sum of all parallel cell currents is the battery module current; and the average of all parallel cell temperatures is the battery module temperature.​

[0074] These various operating conditions include, but are not limited to: normal operation, hybrid pulse power characteristic operation, constant current charge / discharge operation, constant voltage charge / discharge operation, and dynamic stress test operation. Specifically, normal operation refers to the battery's working state under typical application scenarios, such as stable charge / discharge, low-fluctuation loads, etc., with a smooth charge / discharge rate and no drastic fluctuations; hybrid pulse power characteristic operation assesses the battery's dynamic power capability and internal resistance characteristics through alternating short-duration high-power charge / discharge pulses, often based on pulsed loads, such as frequently alternating charge / discharge pulses (e.g., 10 seconds of discharge + 10 seconds of rest + 10 seconds of charging); constant current / constant voltage charge / discharge operation refers to the battery maintaining a constant current / voltage during charging / discharging, respectively; dynamic stress test operation simulates extreme or dynamic scenarios through complex and variable load / current curves, which can be equivalently simulated in practice through grid ancillary services such as real-time peak shaving and frequency regulation requirements.

[0075] Specifically, step S20 includes:

[0076] Step S201, for the modeling layer intelligent agent After normalization, the monitoring data is compared with the first The SOC of the battery module constitutes a monitoring sequence. , , , The first The normalized voltage of a specific group of series-connected cells in a battery module, the cell current at all parallel connection points, and the temperature. For the first SOC of battery module ;

[0077] Step S202: Using wavelet packet transform and sliding window interpolation methods, the monitoring sequences of the modeling layer agents in the same modeling layer are processed. A new monitoring data sequence is obtained by unifying the time axis and sampling frequency. ;

[0078] Using the same wavelet basis function, number of decomposition layers, and frequency band division method, window sequences are obtained by windowing the monitoring data of two agents in the same modeling layer. Perform wavelet packet decomposition to obtain the window sequence. In the The first layer under the decomposition coefficient of each sub-band , Or 2; In this embodiment, the db10 wavelet basis is selected, the number of decomposition layers is 6, and the number of frequency bands is 64;

[0079] Calculate window sequence The Energy of the body Calculate the window sequence The The and the first Correlation coefficient between subbands , , To calculate the covariance, , For the first The and the first The standard deviation of the energy of each element is used to obtain the cross-correlation coefficient using the correlation coefficient. Sub-band computation window sequence In the Subband residuals under layer decomposition Update subband coefficients using subband residuals and cross-correlation coefficients. ;

[0080] Based on the updated subband coefficients The wavelet packet inverse transform method is used to obtain the window sequence. Intermediate window sequences with the same sampling frequency and time axis ; for window sequences and intermediate window sequence The sliding window interpolation method is used to obtain the window sequence. New sequences with the same sampling frequency and time axis Thus making Become a unified window sequence and The time axis and sampling frequency of the new monitoring data sequence; wherein, the sliding window interpolation method is given by the following formula:

[0081]

[0082] In the formula, For modeling layer intelligent agents In the new monitoring data sequence, time Data, , Let be the interpolation coefficients, and satisfy . , , , To utilize window sequences Mid-moment Interpolation calculated from monitoring data at a previous and subsequent time point. To utilize intermediate window sequences Mid-time point Interpolation calculated from monitoring data at a previous and subsequent time point;

[0083] The function for calculating the interpolation includes but is not limited to a quadratic B-spline, a median filter, etc. , Through a plurality of experimental fitting optimizations in the embodiment, , .

[0084] The modeling layer agent corresponding to the battery module in series connection is the same layer modeling layer agent, and step S202 unifies the time axis and the sampling frequency of the monitoring data in the same layer modeling layer agent from the frequency domain, which is a mode of monitoring data fusion. The new monitoring data sequence of each layer modeling layer agent is determined by repeatedly executing step S202, so as to prepare for subsequent data cleaning.

[0085] In step S203, the distance between the new monitoring data sequence of the two modeling layer agents adjacent to the modeling layer agent and the new monitoring data sequence of the modeling layer agent is used to perform sliding window filtering on the new monitoring data sequence of the modeling layer agent. , , , .

[0086] The distance between the new monitoring data sequence of the two modeling layer agents adjacent to the modeling layer agent and the new monitoring data sequence of the modeling layer agent , , , characterizes the time sequence correlation of the monitoring data. By performing sliding window filtering on the new monitoring data sequence , abnormal data points such as sensing errors are preliminarily removed.

[0087] The distance between the new monitoring data sequence of the two modeling layer agents adjacent to the modeling layer agent and the new monitoring data sequence of the modeling layer agent , , characterizes the time sequence correlation of the monitoring data. By performing sliding window filtering on the new monitoring data sequence , abnormal data points such as sensing errors are preliminarily removed.​​​​​​​​​​​​​​​​​ a weighted sum of the modeling layer agent the fused monitoring data filtered by a sliding window as follows:

[0088]

[0089] wherein, the fused monitoring data filtered by a sliding window, , , filtering parameters, and the value range is , and satisfies , is the distance between the new monitoring data sequence of the first group of battery modules and the new monitoring data sequence of the first group of battery modules, is the distance between the new monitoring data sequence of the first group of battery modules and the new monitoring data sequence of the first group of battery modules; In particular, when the modeling layer agent has only one adjacent modeling layer agent or , the above formula is:

[0090]

[0091]

[0092]

[0093] The function for calculating the distance includes but is not limited to: a normalized dynamic time warping distance calculation method; the present application adopts a normalized dynamic time warping distance NDTW to represent the dynamic time warping distance between two curves, in the present application, NDTW represents the corresponding two groups of monitoring data, each point has been aligned by the time axis and the sampling rate in step S202, therefore, NDTW can be calculated once for each point to obtain a correlation value, and then the correlation value is multiplied by the adjacent monitoring data sequence, which is essentially a time sequence weighting process for each point, and generally = Each value is more obtained through engineering test, since the interpolation of the two groups of monitoring data will not be large, the corresponding NDTW value will also not be large, therefore, the values of and are generally small, and more retain the characteristics of the own agent monitoring data. Among them, , , Through multiple group experiments fitting optimization, in the embodiment,​​​​ , , .

[0094] On the basis of monitoring data of the modeling layer intelligent agent in each layer being processed in the frequency domain, step S203 gives full play to the advantages of the multi-level intelligent agent model from the time domain perspective, and the monitoring data of the adjacent battery module of the same type and with the minimum communication overhead is used to filter the monitoring data of the battery module, which is another mode of monitoring data fusion; the filtered new sequence of each modeling layer intelligent agent is determined by repeatedly executing step S203.

[0095] The data cleaning and feature modeling method provided by the application realizes data-level multi-modal monitoring data fusion modeling; in monitoring data cleaning and preprocessing, the monitoring data of multiple intelligent agents under the same type of working condition is complementarily modeled based on the information interaction advantages of the multi-agent model, the mechanism evolution law based on the same monitoring data has significant similarity priori knowledge, and the monitoring data interpolation accuracy and physical interpretability are improved; a new monitoring sequence is constructed according to the frequency domain cross-correlation coefficient, and the time domain correlation feature limitation problem caused by the inconsistent sampling period and time axis can be solved. Finally, the fusion monitoring data under normal operation condition, the fusion monitoring data under mixed pulse power characteristic condition, the fusion monitoring data under constant current charge and discharge condition, the fusion monitoring data under constant voltage charge and discharge condition, and the fusion monitoring data under dynamic stress test condition are obtained.

[0096] Step S30, constructing an equivalent mechanism model of the battery module, and identifying the mechanism model parameters based on the monitoring data; in addition, based on the edge learning framework, the model parameters are iteratively optimized offline; the main object of step S30 is the modeling layer multi-agent;

[0097] Specifically, the characteristic parameters representing the battery aging state are identified based on the P2D model of the battery module using the fusion monitoring data of the modeling layer intelligent agent under various working conditions.

[0098] Specifically, step S30 includes:

[0099] Step S301, based on the P2D model of the battery module, the to-be-identified parameters representing the battery aging state include: solid-phase diffusion coefficient , effective reaction rate constant , liquid-phase diffusion coefficient , effective conductivity , current collector contact resistance , and active material volume fraction .

[0100] For battery modules, a P2D model is used to describe the physical evolution mechanism of the battery charging and discharging process. The P2D model includes: partial differential equations and boundary conditions for solid-phase mass transfer, liquid-phase mass transfer, solid-phase charge conservation, liquid-phase charge conservation, reaction kinetics, and output voltage. These six partial differential equations can equivalently describe the battery's operating state, obtaining the changes in battery voltage and current under ideal operating conditions, which facilitates the subsequent calculation of battery SOC and other indicators.

[0101] 1) The partial differential equations for solid-state mass transfer and their boundary conditions are as follows:

[0102] (1a)

[0103] (1b)

[0104] In the formula, , The radial direction of the particles The optimized value of lithium concentration and the first The value of the next iteration. , The optimized value of the solid diffusion coefficient and the first The value of the next iteration. The local reaction current density is a nonlinear function of the Faraday constant. The first of the electrode particle radius The value of the next iteration. These are the spatial coordinates along the electrode thickness direction. For a moment.

[0105] 2) The partial differential equations for liquid-phase mass transfer and their boundary conditions are as follows:

[0106] (2a)

[0107] (2b)

[0108] In the formula, This refers to the lithium concentration in the electrolyte. The porosity of the electrolyte. To account for the effect of porosity on the liquid phase diffusion coefficient, The electrolyte diffusion coefficient is... For Bruggeman coefficients, This represents the lithium-ion transference number. The electrode specific surface area. The two ends of the electrode are characterized respectively, corresponding to the current collector interface and the diaphragm interface.

[0109] 3) The partial differential equations for solid-state charge conservation and their boundary conditions are as follows:

[0110] (3a)

[0111] (3b)

[0112] In the formula, For solid-state potential, To take into account the effective conductivity of the motor material by volume fraction, It is Faraday's constant. For output current, The cross-sectional area of ​​the electrode is... The thickness of the negative electrode. The thickness of the diaphragm. This represents the volume fraction of the active material.

[0113] 4) The partial differential equations for the conservation of charge in the liquid phase and their boundary conditions are as follows:

[0114] (4a)

[0115] (4b)

[0116] In the formula, The electrolyte potential, For effective liquid phase conductivity, Let be the ideal gas constant. For temperature.

[0117] 5) The reaction kinetic equation is as follows:

[0118] (5)

[0119] In the formula, For exchange current density, The anode transfer coefficient, The cathode transfer coefficient, This is an overpotential. To be related to surface concentration The relevant equilibrium potential, For interface film resistance, The effective reaction rate constant, This represents the maximum lithium concentration in the solid phase. The lithium concentration is at the solid surface.

[0120] 6) The output voltage equation is as follows:

[0121] (6)

[0122] In the formula, Vcellis the battery terminal voltage, Vcontis the potential difference between the positive and negative current collectors, solved from the solid phase conservation equation; Vionis the potential difference caused by ion migration in the electrolyte of the positive and negative regions, solved from the liquid phase charge conservation equation, Rctis the current collector contact resistance.

[0123] In the embodiment, the related parameters of the P2D model of the battery module are known constants when the battery is delivered to the operation, and the typical values of the related parameters are shown in Table 1:

[0124] Table 1 Typical values of related parameters of P2D model of battery module

[0125]

[0126] During the operation of the battery, the related parameters of the P2D model of the battery module in Table 1 are constantly degraded due to the material characteristics such as the thickening of the solid electrolyte interface film, leading to battery aging, so it is necessary to identify the parameters and use the new parameters obtained by identification to accurately describe the physical process after the battery is aged. The to-be-identified parameters, aging mechanism and parameter identification priority are shown in Table 2:

[0127] Table 2 To-be-identified parameters, aging mechanism and parameter identification priority

[0128]

[0129] Step S302, establishing a parameter offline identification model based on a long short-term memory network and a physical information neural network;

[0130] The application discloses a two-stage battery aging parameter offline identification method and correction strategy (DM-PINN-LSTM) based on combination of a long short-term memory network (LSTM) and a physics informed neural network (PINN), to identify six parameters in Table 2. The application introduces a PINN network on the basis of a typical LSTM, adds physical constraints in the super parameter training process, avoids super parameter overshoot, improves the explainability of the model, and can realize different network input dimensions. The input data is the monitoring data of the agent itself, the monitoring data of adjacent agents and typical battery attenuation monitoring data. The training process is different. The application is trained in two stages, and each time the training is performed on the significant influence parameters of the monitoring data under the working condition.

[0131] Step S303, identifying the parameters representing the aging state of the first group of battery modules, taking the modeling layer agent corresponding to the first group of battery modules fused monitoring data of each working condition, adjacent modeling layer agent 、 fused monitoring data of each working condition and typical data of battery module aging process, input of offline identification model of parameter, the offline identification model executes two-stage identification according to the input to obtain characteristic parameters representing the battery aging state, including:

[0132] 1) In the first stage identification, according to the working condition corresponding to the fused monitoring data of the modeling layer agent 、 、 , different characteristic parameters are identified, and when the fused monitoring data corresponds to the mixed pulse power characteristic working condition, the effective conductivity , the current collector contact resistance is identified; when the fused monitoring data corresponds to the constant current charge and discharge working condition, the solid phase diffusion coefficient is identified; when the fused monitoring data corresponds to the constant voltage charge and discharge working condition, the active material volume fraction is identified; when the fused monitoring data corresponds to the dynamic stress test working condition, the effective reaction rate constant and the liquid phase diffusion coefficient are obtained; different characteristic parameters show strong specificity in the corresponding working condition, so they are divided into two stages of identification, and the first stage identifies the characteristic parameters with strong specificity according to the working condition as the first stage identification result;

[0133] In the embodiment, as shown in Figure 3 , three parallel LSTM networks are used to extract the time sequence features of each input Input, and the time sequence features are formed into a feature matrix after being processed by a Transformer module and dimension splicing. After the feature matrix is processed by two fully connected layers Fullyconnector for feature dimension reduction, the identification result is obtained as the output Output, which is the identification parameters 1, 2 and 3 of the P2D model respectively; wherein each LSTM network includes k LSTM modules;

[0134] 2) In the second stage identification, the fused monitoring data of the normal working condition of the modeling layer agent 、 、 and the typical data of the battery module aging process are used to optimize the first stage identification result to obtain the characteristic parameters representing the battery aging state;

[0135] In the embodiment, the time sequence monitoring data of the modeling layer agent is input Input, and the initial value is optimized by the LSTM network and the fully connected layer to obtain the final value of the six to-be-identified parameters as the output Output.

[0136] Wherein, in two-stage identification, the loss function of the offline identification model of the parameters is as follows:

[0137]

[0138] In the formula, , are the data loss and the data loss weighting coefficient, respectively, , are the physical constraint loss and the physical constraint loss weighting coefficient, respectively, , are the regularization loss and the regularization loss weighting coefficient, respectively, , are the distance loss and the distance loss weighting coefficient, respectively;

[0139]

[0140] In the formula, is the parameter estimation value at time , is the parameter training value at time , is the parameter standard value at time , is the calculation of the two-norm distance; represents the kth iteration.

[0141]

[0142] In the formula, is the prediction value of the terminal voltage based on the parameter estimation value at time in the time period , is the true value of the terminal voltage in the time period ;

[0143] is used to constrain to be within the range of the interval with physical meaning;

[0144] characterizes the two-norm distance of the parameter training values between two adjacent modeling layer agents; the loss function introduces and the previous use of adjacent agent monitoring data to clean and pretreat the own agent cooperate, since the adjacent agents have similar operating conditions, the aging processes of the 6 parameters to be identified also have certain similarity, therefore, in order to avoid the risk of overfitting of the model, the two-norm distance of the identified parameters of the adjacent agents can be added in the parameter training process to optimize the parameter identification process of the own agent.

[0145] It is worth noting that the loss function forms of the two stages are the same, but the parameters of the two stages are different For example, in the first stage identification, the parameters are the effective conductivity , the current collector contact resistance ; in the second stage identification, the parameters are 6 characteristic parameters.

[0146] The DM-PINN-LSTM model proposed by the application predicts the to-be-identified parameters in two stages, identifies specific parameters in a specific working condition in the first stage, reduces the coupling relationship between the remaining to-be-optimized parameters and the parameters, and optimizes the identified parameters in the second stage based on the universal operating condition, thereby improving the robustness of the model and the identification accuracy of the parameters; the behaviors of adjacent agents under the same working condition and the typical evolution process are jointly modeled, and are adaptively fused by the Transformer module with weighting, so that the complementary information of different data sources is fully utilized, cross-single-body information complementation is realized, and the understanding of battery aging and heterogeneity of the model is enhanced.

[0147] In the embodiment, the modeling layer agent relies on the battery module BMMS operation, and the agent at this layer often has weak computing power and is difficult to train the DM-PINN-LSTM model. Therefore, the application constructs a model training strategy based on an edge learning framework for the modeling layer agent. First, the modeling layer agent and its adjacent agents , record the monitoring data under various working conditions such as hybrid pulse power characteristics, constant current / constant voltage charging and discharging, and dynamic stress testing at the same sampling frequency, and record the same length of conventional working condition monitoring data at the same time point, and transmit the above data to an edge computing server ; secondly, receive the data and after the data cleaning and preprocessing method of step S20 of the application, train the DM-PINN-LSTM model, and send the model hyperparameters to the modeling layer agent ; finally, the agent based on the new hyperparameters, verifies the effectiveness of the model based on the local conventional working condition monitoring data, iterates the new parameters when the accuracy meets the standard, and otherwise continues to use the old parameters.

[0148] Moreover, the battery has similar attenuation under similar working conditions, and the changes of the 6 parameters to be identified have similar evolution processes. From this perspective, the communication architecture between the battery modules established based on the multi-agent model realizes the use of similar evolution data of adjacent agents in the same energy storage station, the similar process degradation curve of the same type of battery to assist in identifying the aging parameters of the agent, thereby improving the parameter identification accuracy and the consistency of the physical characteristics of the batteries in the station.

[0149] Step S40, based on the monitoring data, the battery module is distributed data monitoring modeling, realizing data level parameter fusion, and constructing a feature matrix representing its running state; the main object of step S40 is the multi-agent of the modeling layer;

[0150] Specifically, each modeling layer agent uses the identified characteristic parameters, the time sequence characteristics of real-time monitoring data, and the frequency sequence characteristics of real-time monitoring data to construct the running state feature matrix of each battery module.

[0151] Specifically, step S40 includes:

[0152] Step S401, the real-time monitoring data of the first battery module is collected and discretized to obtain a discretized monitoring data sequence , , wherein n represents the number of sampling points.

[0153] Specifically, for the battery voltage, current, temperature, cumulative cycle number and time monitoring data of the first battery module to be modeled , represents any one time point, represents the time length, represents the span of the data to be analyzed on the time axis, and the coupling relationship between the monitoring quantity of the previous time period and the monitoring quantity of the current time period is considered to form the initial long-time-scale monitoring data . On this basis, the ratio of to the number of sampling points is the sampling interval, and after windowing by the Hanning window, a new discretized monitoring data sequence is obtained. In this embodiment, the number of sampling points is 40960, and the Hanning window overlap rate is 25%.

[0154] Step S402, based on the discretized monitoring sequence , the normalized control parameters , , 、 、 and is a column vector, and the sampling time is a row vector, and the mechanism feature matrix is constructed as follows:

[0155]

[0156] wherein, is the normalized effective conductivity of the sampling point is the normalized current collector contact resistance of the sampling point is the normalized solid-phase diffusion coefficient of the sampling point is the normalized active material volume fraction of the sampling point is the effective reaction rate constant of the sampling point is the normalized liquid-phase diffusion coefficient of the sampling point

[0157] Step S403, based on the discretized monitoring sequence , the normalized dynamic time warping distance NDTW, the normalized incremental capacity NIC, the normalized differential voltage NDV, the normalized distance measure NSBD, the normalized Wasserstein distance, the ratio of the output voltage of the P2D equivalent model to the actual voltage is a column vector, and the time axis is a row vector, and the time sequence feature matrix is constructed as follows:

[0158]

[0159] wherein, is the normalized dynamic time warping distance of the sampling point is the normalized incremental capacity of the sampling point is the normalized differential voltage of the sampling point is the normalized distance measure of the sampling point is the normalized Wasserstein distance of the sampling point is the ratio of the output voltage of the P2D equivalent model to the actual voltage of the sampling point

[0160] Step S404, based on the discretized monitoring sequence ​​​​​​​​​​​​​​, the second and third modal data are expanded in the frequency domain by variational modal decomposition, and the energy, kurtosis and sample entropy of the two modes are column vectors, and the time axis is a row vector, to construct a frequency domain feature matrix As follows:

[0161]

[0162] In the formula, , The energy of the second and third modes of the sampling point , The kurtosis of the second and third modes of the sampling point , The sample entropy of the second and third modes of the sampling point

[0163] In step S405, the mechanism feature matrix , the time sequence feature matrix , and the frequency domain feature matrix are spliced to obtain the final data level operation state feature matrix .

[0164] The multi-level operation state feature matrix modeling method combining time and frequency domains with mechanism evolution proposed in the application cooperatively models the current time operation state and the last time operation state, avoids time sequence fragmentation in the modeling process, enhances the time sequence stability of the operation features, cooperatively analyzes from multiple angles of mechanism evolution, time domain and frequency domain, enhances the interpretability of the features, and reduces the influence of single modal data disturbance on the modeling accuracy of the feature matrix.

[0165] In step S50, based on the feature matrix, the battery module is distributedly evaluated in operation state, feature level parameter fusion is realized, and the deterministic operation state evaluation is evolved into a probability distribution of the operation state interval; the main object of step S50 is the evaluation layer multi-agent;

[0166] Specifically, the evaluation layer agent obtains the operation state feature matrix of each battery module, and performs distributed situation evaluation on the battery module.

[0167] Based on the communication architecture of the multi-agent model, the modeling layer agent transmits the constructed operation state feature matrix of each battery module to the evaluation layer agent, and the evaluation layer agent is used for remodeling analysis of each operation state feature matrix, based on the operation state probability interval evaluation method proposed in the application, on the basis of feature level fusion of multi-modal monitoring data, the deterministic operation state evaluation is evolved into a probability distribution problem of the operation state interval, and the operation state probability evaluation of the battery module of the energy storage power station is realized. ​​​

[0168] Specifically, step S50 comprises:

[0169] Step S501, based on the connection relationship of each battery module, the layer intelligent agent evaluates the operating state feature matrix of each battery module to splice and obtain a series-parallel module feature matrix;

[0170] Specifically, first, the modeling layer intelligent agent The operating state feature matrix is spliced according to the series connection topology of the battery module in the row vector to obtain a series module feature matrix ; On this basis, the matrix is spliced according to the parallel connection topology in the channel to obtain a series-parallel module feature matrix , and as input network learning.

[0171] In the distributed operating state evaluation of the battery module, the multi-channel battery feature matrix is constructed according to the physical topology, realizing the series-parallel two-level state linkage analysis; On this basis, the confidence is introduced to convert the traditional deterministic state evaluation into a probability distribution modeling, further improving the robustness and interpretability of the model.

[0172] Step S502, a variational Bayes feature evaluation network based on a coding and decoding structure is established;

[0173] As shown in Figure 4 , the variational Bayes feature evaluation network based on the coding and decoding structure comprises: network input, coding process, decoding process, evaluation process and network output.

[0174] Figure 4 In the embodiment, the network input is the operating state feature matrix of the battery module in three dimensions of mechanism-time-frequency obtained based on step S40, which is spliced in the dimension and channel according to the series connection and parallel topology of the battery module to obtain a series-parallel module feature matrix as Input1;

[0175] Coding process: for the network input Input1, the output Bayes Conv N is obtained through N times of Bayes convolution module (Bayes Conv), and after splicing the first Bayes convolution output Bayes Conv1 through the residual module (ResidualConcate) and then performing Bayes convolution again, Bayes Conv 3 is obtained, and then the dimension is flattened through the Bayes linear layer (Bayes Liner) to obtain the coding layer output Bayes Liner 3; in this embodiment, N=8, each Bayes convolution module and Bayes linear module are the same in structure, and the input and output sizes are different from the matching data input and output dimensions.

[0176] Decoding process: For the encoding layer output Bayes Liner 3, after passing through P Bayesian linear layers, it is concatenated with the encoding layer output through the Bayes Liner Concate module to obtain the decoding layer output Bayes Liner Concate 1, thus completing feature decoding. In this embodiment, P=6.

[0177] Evaluation process: The output of the decoding layer Bayes Liner Concaten 1 is passed through Q Bayesian linear layers to obtain the network output Output. In this embodiment, Q=6.

[0178] Finally, the feature layer is evaluated. Specifically, the encoded and decoded features are extracted through a Q-order Bayesian linear module to obtain the evaluation layer agent. Operating state probability matrix as follows:

[0179]

[0180] In the formula, For modeling layer intelligent agents The corresponding battery module operating status assessment results;

[0181]

[0182] In the formula, , , These are the intelligent agents in the modeling layer. The corresponding evaluation index values ​​of SOC, SOH, and RUL for the battery module. , , These are the intelligent agents in the modeling layer. The confidence level of the corresponding battery module's SOC, SOH, and RUL evaluation index values.

[0183] Existing methods employ deterministic evaluation, directly yielding a specific numerical value. However, for black-box neural network models, providing probability distribution information on top of a deterministic value can improve the confidence level of the model results for maintenance personnel. The variational Bayesian feature evaluation network model based on an encoding / decoding structure disclosed in this invention offers advantages over existing methods in the following ways: relying on a multi-agent model, a multi-channel battery feature matrix is ​​constructed based on the physical topology, realizing two-level state linkage analysis of the module's serial and parallel connections; furthermore, by introducing confidence levels, traditional deterministic state evaluation is transformed into probabilistic distribution modeling, further improving the model's robustness and interpretability.

[0184] This invention also proposes a distributed situation assessment system for energy storage power stations based on a multi-agent model, comprising:

[0185] The multi-agent model establishing module is configured to: take the module-level battery management system of the battery module as a modeling layer agent, and map the connection relationship of the battery module to a structural relationship of the modeling layer agent; take the energy management system of the energy storage converter as an evaluation layer agent, and map the connection relationship of the energy storage converter to a structural relationship of the evaluation layer agent; map the connection relationship between the battery module and the energy storage converter to a structural relationship between the modeling layer agent and the evaluation layer agent; and the decision layer agent receives state information of each evaluation layer agent;

[0186] The situation assessment module is configured to: for any modeling layer agent, filter monitoring data of various operating conditions of the modeling layer agent by using monitoring data of two adjacent modeling layer agents, to obtain fused monitoring data of various operating conditions of the modeling layer agent; each modeling layer agent identifies a characteristic parameter representing an aging state of the battery module based on a P2D model of the battery module by using the fused monitoring data of various operating conditions; each modeling layer agent constructs an operating state feature matrix of each battery module by using the identified characteristic parameter, a time sequence feature of real-time monitoring data, and a frequency sequence feature of real-time monitoring data; and the evaluation layer agent acquires the operating state feature matrix of each battery module to perform situation assessment on the battery module.

[0187] The present disclosure can be a system, a method, and / or a computer program product. The computer program product can include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present disclosure.

[0188] The computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. The computer readable storage medium can be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium include the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device such as punch-cards or punched tape, a magnetically encoded device such as magnetic strip cards, an optically encoded device such as a compact disc (CD) or DVD, and / or any suitable combination of the foregoing. A computer readable storage medium, as used herein, is not to be construed as being transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission media (e.g., light pulses passing through a fiber-optic cable), or electrical signals transmitted through a wire.

[0189] Computer readable program instructions described herein can be downloaded to respective computing / processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and / or a wireless network. The network can comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and / or edge servers. A network adapter card or network interface in each computing / processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium within the respective computing / processing device.

[0190] Computer readable program instructions for carrying out operations of the present disclosure can be assembly instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, or either source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C++ or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer readable program instructions can execute entirely on the user's computing device, partly on the user's computing device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device or entirely on the remote computing device or server. In the latter scenario, the remote computing device can be connected to the user's computing device through any kind of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computing device, for example, through the Internet using an Internet Service Provider. In some embodiments, electronic circuitry including, for example, programmable logic circuitry, field-programmable gate arrays (FPGA), or programmable logic arrays (PLA) can execute the computer readable program instructions by utilizing state information of the computer readable program instructions to personalize the electronic circuitry, in order to perform aspects of the present disclosure.

[0191] Finally, it should be noted that the above-mentioned embodiments are merely intended for describing the technical solutions of the present application, but not for limiting it. Although the present application is described in detail with reference to the above embodiments, those skilled in the art should understand that the specific embodiments of the present application can be modified or replaced equivalently, and any modification or replacement without departing from the spirit and scope of the present application should be covered within the protection scope of the claims of the present application.

Claims

1. A method for distributed situation assessment of energy storage power stations based on multi-agent model, the multi-agent model comprising: The modeling layer, the evaluation layer and the decision layer, wherein the decision layer agent receives state information of each evaluation layer agent; and the method comprises the following steps of: The modeling layer agent is a module-level battery management system of a battery module, and the connection relationship of the battery module is mapped to the structural relationship of the modeling layer agent; the evaluation layer agent is an energy management system of an energy storage converter, and the connection relationship of the energy storage converter is mapped to the structural relationship of the evaluation layer agent; and the connection relationship of the battery module and the energy storage converter is mapped to the structural relationship of the modeling layer agent and the evaluation layer agent. For any modeling layer agent, the monitoring data of the modeling layer agent under various working conditions is filtered by using the monitoring data of two adjacent modeling layer agents, to obtain the fusion monitoring data of the modeling layer agent under various working conditions; the modeling layer agent identifies the characteristic parameters representing the aging state of the battery module by using the fusion monitoring data under various working conditions and based on a P2D model of the battery module; the modeling layer agent constructs an operating state feature matrix of each battery module by using the identified characteristic parameters, the time sequence features of the real-time monitoring data and the frequency sequence features of the real-time monitoring data; and the evaluation layer agent acquires the operating state feature matrix of each battery module to perform situation assessment on the battery module.

2. The method according to claim 1, wherein the connection relationship of the battery module is mapped to the structural relationship of the modeling layer agent, and the connection relationship of the battery module includes: The modeling layer agent corresponding to the battery module connected in series is located in the same modeling layer, the modeling layer agent corresponding to the battery module connected in parallel is located in different modeling layers, and the modeling layer agent with the minimum communication overhead to the evaluation layer agent in each modeling layer is defined as a head agent. In terms of communication, each modeling layer agent can only interact with adjacent agents except the head agent, the adjacent modeling layer agents store the monitoring data of the battery module under the same working condition, and the adjacent modeling layer agents have the minimum communication overhead; the head agent in each modeling layer interacts with the adjacent modeling layer agents in the modeling layer, and interacts with the head agents in the adjacent modeling layers and the evaluation layer agent.

3. The method according to claim 2, wherein the monitoring data of the modeling layer agent is acquired from the corresponding battery management system, and includes the voltage of a certain group of series-connected battery cells, the current of all parallel points and the temperature of the battery module. After the monitoring data of the modeling layer agent is normalized, the monitoring data and the SOC of the battery module constitute the monitoring sequence of the modeling layer agent.

4. The method according to claim 3, wherein the monitoring sequence of the modeling layer agent is filtered by using the monitoring data of the adjacent modeling layer agents.

5. The method according to claim 4, wherein the characteristic parameters representing the aging state of the battery module are identified by using the filtered monitoring sequence of the modeling layer agent.

6. The method according to claim 5, wherein the operating state feature matrix of each battery module is constructed by using the identified characteristic parameters, the time sequence features of the real-time monitoring data and the frequency sequence features of the real-time monitoring data. Adopt wavelet packet transform method and sliding window interpolation method, to the same modeling layer in the modeling layer agent monitoring sequence The time axis and sampling frequency are unified to obtain a new monitoring data sequence ; Utilizing and modeling layer intelligent agents Two adjacent modeling layer agents , New monitoring data sequence , and modeling layer intelligent agents New monitoring data sequence Distance between , For the modeling layer intelligent agent New monitoring data sequence Perform sliding window filtering; after filtering, obtain the intelligent agent of the modeling layer. Fusion monitoring data . ​ The window sequence obtained after windowing the monitoring data of two modeling layer agents in the same modeling layer Wavelet packet decomposition is performed to obtain the window sequence In the first layer decomposition, the coefficients of the first subband , or 2; Utilizing window sequence of the first subband of the window sequence of the first and the second subband , , for calculating covariance, , of the first and the second subband energy; obtaining cross-correlation coefficients using correlation coefficients ; computing window sequence per subband in the first subband residual under layer decomposition ; updating subband coefficients using subband residual and cross-correlation coefficients ; Based on the updated sub-band coefficients , the wavelet packet inverse transform method is adopted to obtain the window sequence with the same sampling frequency and time axis ; the window sequence and the intermediate window sequence are obtained by using the sliding window interpolation method to obtain the window sequence with the same sampling frequency and time axis , as follows: wherein is a new monitoring data sequence for the modeling layer agent at time , , is an interpolation coefficient and satisfies , , , is an interpolation calculated using monitoring data at a time point before and after time , is an interpolation calculated using monitoring data at a time point before and after time . ​ Fused monitoring data filtered with a sliding window As follows: In the formula, is a new window sequence after filtering, , , is a filtering parameter, and the value range is , and satisfies , is the distance between the new monitoring data sequence of the first group of battery modules and the new monitoring data sequence of the first group of battery modules, is the distance between the new monitoring data sequence of the first group of battery modules and the new monitoring data sequence of the first group of battery modules, is the distance between the new monitoring data sequence of the first group of battery modules and the new monitoring data sequence of the first group of battery modules, .

7. The multi-agent model-based distributed situation assessment method for energy storage power stations according to claim 1, characterized in that, the to-be-identified parameters for characterizing the aging state of the battery module based on the P2D model of the battery module include a solid-phase diffusion coefficient, an effective reaction rate constant, a liquid-phase diffusion coefficient, an effective conductivity, a current collector contact resistance, and an active material volume fraction; an offline identification model is established based on a long short-term memory network and a physical information neural network; The parameter of the battery module group is identified The parameter of the battery module group is identified The parameter of the battery module group is identified The parameter of the battery module group is identified The parameter of the battery module group is identified The parameter of the battery module group is identified 8. The multi-agent model-based distributed situation assessment method for energy storage power stations according to claim 7, characterized in that, In the first stage of identification, according to the working condition corresponding to the fusion monitoring data of the modeling layer agent , , , different characteristic parameters are identified. When the fusion monitoring data corresponds to the mixed pulse power characteristic working condition, the effective conductivity and the current collector contact resistance are identified. When the fusion monitoring data corresponds to the constant current charge and discharge working condition, the solid-phase diffusion coefficient is identified. When the fusion monitoring data corresponds to the constant voltage charge and discharge working condition, the active material volume fraction is identified. When the fusion monitoring data corresponds to the dynamic stress test working condition, the effective reaction rate constant and the liquid-phase diffusion coefficient are obtained. In the second stage identification, the modeling layer agent is used , , to optimize the first stage identification result by using the fusion monitoring data and typical data of the battery module aging process under normal working conditions, so as to obtain the characteristic parameters representing the battery aging state.

9. The multi-agent model-based distributed situation assessment method for energy storage power stations according to claim 1, characterized in that, Collection of the first Real-time monitoring data of the battery module is discretized to obtain a discretized monitoring data sequence. , , This represents the number of sampling points; Mechanism feature matrix As follows: In the formula, Sampling points The normalized effective conductivity, Sampling points Normalized current collector contact resistance, Sampling points The normalized solid-phase diffusion coefficient, Sampling points The normalized volume fraction of active material, Sampling points The effective reaction rate constant, Sampling points The normalized liquid-phase diffusion coefficient; Constructing a timing feature matrix As follows: In the formula, Sampling points Normalized dynamic time-warped distance, Sampling points Normalized incremental capacity, Sampling points The normalized differential voltage, Sampling points Normalized distance measure, Sampling points The normalized Wasserstein distance, Sampling points The ratio of the output voltage to the actual voltage of the P2D equivalent model; Constructing a frequency domain feature matrix As follows: wherein , is the energy of the second and third modalities at the sampling point , , is the kurtosis of the second and third modalities at the sampling point , , is the sample entropy of the second and third modalities at the sampling point , the operating state feature matrix is obtained by splicing the mechanism feature matrix, the time sequence feature matrix, and the frequency domain feature matrix.

10. The multi-agent model-based distributed situation assessment method for energy storage power stations according to claim 1, characterized in that, based on the connection relationship of each battery module, the operating state feature matrix of each battery module is spliced by the assessment layer agent to obtain a series-parallel module feature matrix; A codec structure-based variational Bayesian feature evaluation network is established; the network outputs an operating state probability matrix of the agent according to a series-parallel module feature matrix The row vector of the operating state probability matrix is the battery module operating state evaluation result vector corresponding to each modeling layer agent, and the elements in the battery module operating state evaluation result vector include: the evaluation index value and its confidence of the SOC, SOH and RUL of the battery module corresponding to the modeling layer agent ​ 11. A multi-agent model based energy storage power station distributed situation assessment system, the multi-agent model comprising: the modeling layer, the assessment layer, and the decision layer, and the state information of each assessment layer agent is received by the decision layer agent; The multi-agent model-based distributed situation assessment method for energy storage power stations according to any one of claims 1 to 10, characterized in that, comprises: a multi-agent model establishment module, which is used to take the module-level battery management system of the battery module as the modeling layer agent, and map the connection relationship of the battery module as the structural relationship of the modeling layer agent; take the energy management system of the energy storage converter as the assessment layer agent, and map the connection relationship of the energy storage converter as the structural relationship of the assessment layer agent; and map the connection relationship of the battery module and the energy storage converter as the structural relationship of the modeling layer agent and the assessment layer agent; a situation assessment module, which is used to, for any modeling layer agent, filter the monitoring data of various operating conditions of the modeling layer agent by using the monitoring data of the two adjacent modeling layer agents, to obtain the fusion monitoring data of various operating conditions of the modeling layer agent; each modeling layer agent uses the fusion monitoring data of various operating conditions to identify the characteristic parameters representing the aging state of the battery module based on the P2D model of the battery module; each modeling layer agent uses the identified characteristic parameters, the time sequence features of the real-time monitoring data, and the frequency sequence features of the real-time monitoring data to construct the operating state feature matrix of each battery module; and the assessment layer agent acquires the operating state feature matrix of each battery module to perform situation assessment on the battery module.

Images