Energy storage system risk management method and system

Abstract

The embodiment of the present application provides a kind of energy storage system risk management method and system, it is related to risk analysis technical field.The method comprises the following steps: in response to the user question query input by user, structured execution plan information is obtained by parsing;If the target function includes risk analysis function, the risk assessment result of each evaluation instance corresponding to the energy storage system is determined based on the risk analysis on at least one evaluation instance obtained by mapping the evaluation data range, the energy storage system includes multi-level risk indicators, and the risk assessment result includes: the risk indicators of the next lower level of the existing risk, the risk propagation path containing the risk indicators of the next lower level of the existing risk and the lowest level of the abnormal risk indicators associated with each risk indicator on the risk propagation path;Based on the output data corresponding to all called target functions, the summary report corresponding to the user question query is generated and output, and the summary report at least includes: the related information of risk assessment result.

Description

Technical Field

[0001] This invention relates to the field of energy storage risk management technology, specifically to a method and system for energy storage system risk management. Background Technology

[0002] Against the backdrop of "dual carbon" goals and the construction of new power systems, electrochemical energy storage systems (EESS) have become core equipment supporting new energy grid connection, peak shaving and valley filling, and frequency and voltage regulation due to their advantages such as fast response, flexible configuration, and high energy density. Their large-scale deployment continues to accelerate. However, the system consists of multiple levels of equipment, including cells, battery packs, clusters, cabinets, battery management systems (BMS), process control systems (PCS), thermal management, and fire protection. Its complex operating conditions and strong multi-physics coupling make it prone to problems such as increased cell inconsistency, insulation degradation, localized overheating, internal short circuits, and the spread of thermal runaway risks during long-term operation. Traditional operation and maintenance rely on BMS basic alarms, manual inspections, and expert judgment, which suffers from significant problems such as large data volumes but a lack of closed-loop governance, inaccurate risk identification, difficulty in tracing cross-system propagation, non-standard measures, delayed response, high costs, and difficulties in large-scale promotion.

[0003] Currently, research and engineering applications related to the operation and maintenance of electrochemical energy storage systems, both domestically and internationally, still primarily focus on two technical systems: data-driven and model-driven. The mainstream technical solutions at present mainly rely on building mechanistic models of energy storage systems or introducing intelligent algorithms such as artificial intelligence and machine learning to mine the operational data characteristics of energy storage devices, complete the analysis and judgment of system operating conditions, and thus achieve dynamic optimization of energy storage operation strategies, accurate identification of equipment faults, and anomaly alarms. These technical approaches effectively improve the problems of high reliance on manual labor and delayed fault response in traditional energy storage operation and maintenance, significantly improving the automation level and fault handling efficiency of energy storage system operation and maintenance.

[0004] However, from the perspective of practical application, existing operation and maintenance technologies still have significant shortcomings. Firstly, most of these methods rely on fixed mechanism models or single intelligent algorithm systems for analysis, imposing stringent requirements on the completeness, accuracy, and model compatibility of the original operating data, resulting in poor versatility and adaptability. Secondly, the technical outputs are mostly various operating indicators and fault alarm signals, failing to clearly and completely explain the derivation logic of data analysis and fault judgment, thus lacking interpretability. At the same time, the monitoring and judgment dimensions of existing operation and maintenance technologies are relatively singular, mostly focusing on the identification and detection of obvious equipment faults and short-term sudden anomalies. They cannot effectively capture the gradual accumulation of operating condition differences, potential safety hazards, and systemic performance degradation issues during the long-term operation of energy storage systems, making it difficult to achieve comprehensive and long-term status assessment. Furthermore, the current technical system fails to effectively combine industry standards and specifications with the practical experience of front-line operation and maintenance experts, failing to provide on-site operation and maintenance personnel with professional, interactive, and reusable intelligent question-and-answer decision support, thus failing to meet the actual needs of refined and intelligent energy storage operation and maintenance.

[0005] In existing technologies, in engineering scenarios where electrochemical energy storage systems are deployed on a large scale and safety risks are becoming increasingly prominent, traditional risk assessments only output risk scores and levels, which cannot be automatically transformed into implementable, traceable, and standard-compliant operation and maintenance management measures. They rely heavily on manual interpretation by experts, resulting in low efficiency, high cost, and slow response. Furthermore, traditional expert systems have rigid rules, poor generalization, and are prone to reasoning illusions, making it difficult to cover complex risks that are multi-dimensional and co-propagated across subsystems. They cannot provide operation and maintenance personnel with integrated intelligent operation and maintenance support that combines professionalism, compliance, and interpretability. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for risk management of energy storage systems. This method analyzes user queries, decomposes tasks, and performs collaborative reasoning. It integrates risk data, standard provisions, and mechanistic knowledge to generate an executable, verifiable, and traceable energy storage risk governance report. This reduces reliance on manual intervention, improves the efficiency of anomaly tracing and governance, avoids the omission of key risks, and provides safe, compliant, and expert-level decision support for energy storage safety operation and maintenance personnel.

[0007] To achieve the above objectives, the present invention provides a risk management method for energy storage systems, comprising: In response to a user's query, structured execution plan information is parsed and obtained. The execution plan information includes: the target function to be invoked, the execution graph between the target functions, and the evaluation data range. If the target function includes a risk analysis function, then based on the risk analysis performed on at least one assessment instance obtained by mapping the assessment data range, the risk assessment result of the energy storage system corresponding to each assessment instance is determined. The energy storage system includes multi-level risk indicators, and the risk assessment result includes: the second-lowest level risk indicator with risk, the risk propagation path containing the second-lowest level risk indicator with risk, and the lowest level abnormal risk indicator associated with each of the risk indicators on the risk propagation path. Based on the output data corresponding to all the invoked target functions, a summary report corresponding to the user problem query is generated and output. The summary report includes at least the relevant information of the risk assessment results.

[0008] The present invention also provides a risk management system for energy storage systems, comprising: a leading intelligent agent, a risk analysis intelligent agent, and a summarizing intelligent agent; The dominant intelligent agent is used to respond to user-inputted questions and queries, and parse the structured execution plan information. The execution plan information includes: the intelligent agent corresponding to the target function to be called, the execution graph between the intelligent agents corresponding to the target function, and the evaluation data range. The intelligent agent corresponding to the target function to be called includes the risk analysis intelligent agent. The risk analysis agent is used to determine the risk assessment result of the energy storage system corresponding to each assessment instance based on risk analysis of at least one assessment instance obtained by mapping the assessment data range. The energy storage system includes multi-level risk indicators. The risk assessment result includes: the second-lowest level risk indicator with risk, the risk propagation path containing the second-lowest level risk indicator with risk, and the lowest level abnormal risk indicator associated with each of the risk indicators on the risk propagation path. The summary agent uses the output data of all the agents corresponding to the invoked target functions to generate and output a summary report corresponding to the user question query. The summary report includes at least the relevant information of the risk assessment results.

[0009] The present invention also provides an energy storage system risk management device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the energy storage system risk management method as described above.

[0010] The present invention also provides a computer-readable storage medium, which is a non-volatile or non-transient storage medium, on which a computer program is stored, and the computer program is executed by a processor to perform the energy storage system risk management method described above.

[0011] In one embodiment, determining the risk assessment result of the energy storage system corresponding to each assessment instance based on risk analysis performed on at least one assessment instance mapped from the assessment data range includes: For each evaluation instance: The initial risk scores of the risk indicators at each level are obtained, and the initial risk scores of the risk indicators at non-lowest levels are corrected based on the coupling relationship network between the risk indicators at the same level, so as to obtain the corrected risk scores of each risk indicator. Based on the corrected risk scores and weights of each risk indicator in the non-lowest level, at least one first risk indicator is selected from the risk indicators in the second-lowest level, and based on the coupling relationship network between the risk indicators in the second-lowest level, a second risk indicator that can be reached along a directed path or that can be reached by the first risk indicator along a directed path is selected. Based on the first risk indicator, the second risk indicator, and the risk indicators connected by each of the directed paths, at least one risk propagation path is obtained. Each risk propagation path includes at least one of the first risk indicators, and no two risk propagation paths overlap. For each risk indicator on the aforementioned risk propagation path, the associated abnormal risk indicator is selected based on the initial risk score of the lower-level risk indicator associated with the risk indicator.

[0012] In one embodiment, extending and connecting all directed paths that connect to the same risk indicator yields at least one risk propagation path, including: For each risk indicator connected by the directed path, the difference between the initial risk score and the corrected risk score of the risk indicator is calculated as the coupling-induced score of the risk indicator, and based on the coupling-induced score of the risk indicator, the most relevant target directed path is selected from all the directed paths pointing to the risk indicator. Extend and connect all the directed paths that connect the same risk indicator in the target directed paths to obtain at least one risk propagation path.

[0013] In one embodiment, based on the coupled-induced scoring of the risk indicator, selecting the most relevant target directed path from all directed paths pointing to the risk indicator includes: Based on the coupling-induced score of the risk indicator, the degradation level of the risk indicator caused by risk coupling is determined, and the degradation level includes: high risk level, medium risk level and low risk level. If the degradation level corresponding to the risk indicator is a high-risk level, then all directed paths pointing to the risk indicator will be regarded as the most relevant target directed paths. If the degradation level corresponding to the risk indicator is medium risk level, then select a portion of the directed paths from all the directed paths pointing to the risk indicator as the most relevant target directed paths; If the degradation level corresponding to the risk indicator is low risk, then the most relevant target directed path will not be selected from all directed paths pointing to the risk indicator.

[0014] In one embodiment, based on the modified risk scores and weights of each risk indicator in the non-lowest level, at least one first risk indicator is selected from the risk indicators in the next lower level, including: The risk indicators are traversed from top to bottom and then to the next lowest level. For each risk indicator encountered: Based on the corrected risk scores of each subordinate risk indicator of the aforementioned risk indicator, the risk severity level of each subordinate risk indicator is determined, and based on the risk severity level of each subordinate risk indicator, each subordinate risk indicator that can be retained is selected. Based on the corrected risk scores and weights of each of the lower-level risk indicators of the risk indicators, the first risk impact factor of the lower-level risk indicators is calculated, and based on the first risk impact factor of each of the lower-level risk indicators, each of the lower-level risk indicators that can be retained is selected. All the risk indicators that can be retained in the second-lowest level are used as the first risk indicator.

[0015] In one embodiment, selecting the abnormal risk indicator associated with the risk indicator based on the initial risk score of the lower-level risk indicator associated with the risk indicator includes: Based on the initial risk scores and weights of each subordinate risk indicator associated with the aforementioned risk indicator, a second risk impact factor for each subordinate risk indicator is calculated. The lower-level risk indicator with the largest second risk impact factor is selected as the abnormal risk indicator.

[0016] In one embodiment, if the target function includes a standard guidance function and the execution graph includes directed edges from the risk analysis function to the standard guidance function, then based on the user question query and the risk assessment result, target clause data is retrieved from a pre-built standard knowledge graph, and the target clause data package is transformed into evidence-driven information including explicit clause references. The summary report also includes: operation and maintenance suggestions and strategies corresponding to the evidence-driven information.

[0017] In one embodiment, if the target function includes a mechanism explanation function and the execution graph includes directed edges from the risk analysis function to the mechanism explanation function, then target retrieval data with labeled evidence sources can be retrieved from the pre-built energy storage knowledge base based on the user question query and the risk assessment result. The summary report also includes risk analysis information corresponding to the target retrieved data. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the energy storage system risk management method in the first embodiment of the present invention; Figure 2 This is a schematic diagram of the energy storage system risk management system corresponding to the energy storage system risk management method in the first embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the process of risk location performed by the risk analysis agent in the first embodiment of the present invention; Figure 4 This is a schematic diagram of the risk factor tree and risk coupling relationship network corresponding to the multi-level risk index system of the energy storage system in the first embodiment of the present invention. Figure 5 This is a schematic diagram of the risk positioning effect in the first embodiment of the present invention; Figure 6 A schematic diagram of the risk propagation path in the first embodiment of the present invention; Figure 7 A schematic diagram of the response of the energy storage system risk management system in the first embodiment of the present invention to two types of user query questions: knowledge query and risk positioning query. Figure 8 The first embodiment of the present invention is a schematic diagram of the energy storage risk management and operation and maintenance report output by the summary intelligent agent. Detailed Implementation

[0019] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings to provide a clearer understanding of the purpose, features, and advantages of the present invention. It should be understood that the embodiments shown in the drawings are not intended to limit the scope of the present invention, but are merely illustrative of the essential spirit of the technical solution of the present invention.

[0020] In the following description, certain specific details are set forth for the purpose of illustrating various disclosed embodiments in order to provide a thorough understanding of the various disclosed embodiments. However, those skilled in the art will recognize that embodiments may be practiced without one or more of these specific details. In other instances, well-known apparatuses, structures, and techniques associated with this application may not have been shown or described in detail to avoid unnecessarily obscuring the description of the embodiments.

[0021] Unless the context requires otherwise, throughout the specification and claims, the word “comprising” and its variations, such as “including” and “having”, shall be understood to have an open, inclusive meaning, that is, to be interpreted as “including, but not limited to”.

[0022] Throughout this specification, references to "an embodiment" or "an embodiment" indicate that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Therefore, the appearance of "in an embodiment" or "an embodiment" in various places throughout the specification does not necessarily refer to the same embodiment. Furthermore, a particular feature, structure, or characteristic may be combined in any manner in one or more embodiments.

[0023] The singular forms “a” and “the” used in this specification and the appended claims include plural references unless otherwise expressly stated herein. It should be noted that the term “or” is generally used to include the meaning of “or / and” unless otherwise expressly stated herein.

[0024] In the following description, in order to clearly demonstrate the structure and working method of the present invention, a number of directional terms will be used. However, terms such as "front", "back", "left", "right", "outside", "inside", "outward", "inward", "up", and "down" should be understood as convenient terms and not as limiting terms.

[0025] With the continuous increase in data scale and scenario complexity, industrial multimodal data association and annotation are gradually showing the characteristics of cross-modal, multi-stage, and multi-strategy collaborative processing. How to automate the analysis of discrete event text data and environmental monitoring time series data under a unified framework, and achieve accurate and reliable association and annotation through the synergistic effect of multiple analysis methods, has become a technical problem that urgently needs to be solved in the field of intelligent industrial data processing.

[0026] The first embodiment of this invention relates to a risk management method for energy storage systems, applied to an energy storage system risk management device. This device can be, for example, an electronic device, a server, or a common computer device such as a desktop computer or laptop. The energy storage system risk management device can generate corresponding reports based on this method, performing risk analysis on the energy storage system. A systematic risk indicator system covering eight categories and dozens of detailed factors is constructed for multiple risk indicators involved in the actual operation of the electrochemical energy storage system. Based on this systematic risk indicator system, a multi-level risk indicator system covering all key aspects can be constructed, forming a risk factor tree. For example, the risk indicators of the energy storage system are divided into four levels, from top to bottom: subsystem (risk surface), risk factor, secondary risk factor, and risk item. Each level includes at least one risk indicator, and the risk indicators in the higher level are associated with one or more risk indicators in the lower level (referred to as lower-level risk indicators).

[0027] At the subsystem level, there are eight core subsystems: cell subsystem, battery pack subsystem, battery cluster subsystem, energy storage cabinet / station subsystem, thermal management subsystem, fire protection subsystem, power electronic devices and BMS subsystem (BMS, Battery Management System); the above eight core subsystems constitute the first-level risk classification covering the main subsystems of the energy storage system.

[0028] For each subsystem, based on the physical attributes and impact paths that generate risks, four risk factors are identified: electrical characteristics, thermal characteristics, mechanical characteristics, and environmental characteristics, covering the most important risk dimensions in energy storage systems. These four factors comprehensively reflect the potential sources of problems in energy storage systems and, together with the aforementioned subsystem classification, constitute a two-dimensional risk factor matrix of "subsystem × risk factor".

[0029] Each risk factor within a subsystem can include multiple sub-risk factors. For example, the electrical characteristics of a cell subsystem can include sub-risk characteristics such as performance failure, internal short circuit, and abnormal use. Each sub-risk factor can also include multiple risk items. For instance, the sub-risk factor "performance failure" can further include two risk factors: low coulombic efficiency and excessive self-discharge rate. Other risk factors, sub-factors, and risk items can be set as needed and will not be detailed here.

[0030] After constructing the systematic risk indicator system, a coupling relationship network is introduced to characterize the mutual influence between different risk indicators. The coupling relationship network is constructed using a directed graph structure to represent the relationships between secondary risk factors, i.e., the relationships between risk indicators at the next lower level. Nodes in the coupling relationship network represent risk indicators (i.e., secondary risk factors) at the next lower level, and edges represent the influence relationships between these risk indicators (i.e., secondary risk factors). Of course, the coupling relationship network can also represent the relationships between risk indicators at other levels and can cross levels; for example, a risk factor may influence a secondary risk factor, or a risk item may influence risk items under another risk factor. The following explanation uses the coupling relationship network to represent the relationships between risk indicators at the next lower level as an example.

[0031] like Figure 1 The diagram shown is a detailed flowchart of the energy storage system risk management method in this embodiment. The energy storage system risk management method in this embodiment is implemented by a corresponding energy storage system risk management system. This system is configured with multiple agents performing different functions, including at least: a leading agent, an analysis agent (risk), and a summarizing agent (sum); furthermore, it may also include: a standard guidance agent (std) and / or a mechanism interpretation agent (mech); for example, as shown... Figure 2 As shown, the energy storage system risk management system includes: a leading agent, an analytical agent (risk), a summarizing agent (sum), a standard guiding agent (std), and a mechanism explaining agent (mech).

[0032] Step 101: In response to the user's input query, the structured execution plan information is parsed and obtained. The execution plan information includes: the target function to be called, the execution graph between the target functions, and the evaluation data range.

[0033] Specifically, the leading agent, acting as the entry point and coordinator of the risk management system, receives user-input queries and transforms them into structured execution plans for other agents. It analyzes user query requests, clarifies task planning, determines which agents to activate and how they collaborate, and obtains structured execution plan information, including the target functions to be invoked, the execution graph between target functions, and the scope of evaluation data.

[0034] For example, the leading agent responds to the user's query χ by outputting structured execution plan information. for: ; in, This represents the set of agents corresponding to the target functions to be invoked as determined by the user's query χ. Since agent sum always executes the corresponding function last, it is not included. ; The execution graph representing the target functions, that is, the calling order and dependency relationship between the agents to be called and between the agents to be called and the summative agent, is a directed acyclic graph. The nodes are the agents to be called or the summative agent, and the directed edges represent the output of one agent to another. This represents the range of evaluation data obtained by the leading intelligent agent through mapping the user's question query χ. This range of evaluation data is the dataset range of the power plant and cycle to be evaluated and analyzed. ; in, This indicates the name of the power station to be queried, obtained by the leading agent from parsing the user's query χ. The data period to be queried is obtained by parsing the user question query χ of the dominant intelligent agent. Therefore, the operating data of the power station in different time periods can be queried based on the user question query χ to obtain the evaluation data range. This evaluation data range is used to select evaluation instances and form a subset of evaluation instances.

[0035] In determining execution plan information Then, based on the execution graph The indicated dependencies construct the inputs for the invoked agents: For example, when the risk analysis agent risk is activated and invoked, risk∈ The input to the risk analysis agent (risk) is: the leading agent based on the assessment data range. The selected evaluation instances constitute a subset of evaluation instances. It can be defined as: ; in, Indicates based on the range of assessment data The selected subset of assessment instances may include one or more power plants and assessment instances for a specific period. Each assessment instance is the operational data of a power plant within a set time period. If there are multiple assessment instances, risk analysis will be performed separately for each assessment instance. Indicates that power station n is in a periodic period Evaluation examples below; This represents the operational dataset of the energy storage system.

[0036] For example, when the standard guiding agent std is activated and invoked, std∈ The input to the standard-guided agent std includes: the leading agent parsing standard-related content from the user's query; furthermore, if the graph is executed... If an edge points from risk to std, then the input to the standard guiding agent std also includes the risk assessment results output by the risk analysis agent risk.

[0037] For example, the mechanism explains that when the agent mech is activated, mech∈ The input to the mechanism explanation agent `mech` includes: the mechanism-related content obtained by the leading agent from parsing the user's query; furthermore, if the graph is executed... If the edge contains a path from risk to mech, then the input to the mechanism explanation agent mech also includes the risk assessment result output by the risk analysis agent risk.

[0038] It is evident that the main function of the dominant agent is to understand user intent, schedule other agents according to the task requirements corresponding to the user's question query, transform open user question queries into compact and explicit task coordination, and establish a traceable route from user intent to expert analysis and final report generation.

[0039] For example, a user's question query χ is: "Conduct a risk assessment of Power Plant 2 between March and April 2024. Based on relevant standards, propose maintenance recommendations and use relevant knowledge to explain the causes of the identified risks."

[0040] The set of intelligent agents to be invoked ; Execution diagram Includes: edges with risk pointing to std and edges with risk pointing to mech.

[0041] Evaluation of a subset of instances This includes: an assessment example of Power Plant 2 in March 2024 and an assessment example of Power Plant 2 in April 2024.

[0042] Step 102: If the target function includes a risk analysis function, then based on the risk analysis performed on at least one assessment instance obtained by mapping the assessment data range, determine the risk assessment result of the energy storage system corresponding to each assessment instance. The energy storage system includes multi-level risk indicators, and the risk assessment result includes: the second-lowest level risk indicator with risk, the risk propagation path containing the second-lowest level risk indicator with risk, and the lowest level abnormal risk indicator associated with each of the risk indicators on the risk propagation path.

[0043] Specifically, for the risk analysis agent, when invoked, it performs risk localization for each input assessment instance, obtaining corresponding risk assessment results. These results include at least: the second-lowest level risk indicator where risk exists, the risk propagation path containing the second-lowest level risk indicator, and the lowest-level abnormal risk indicator associated with each of the risk indicators along the propagation path. This allows identification of the most relevant deterioration branch in the hierarchical indexing system of the energy storage system. For each assessment instance, please refer to... Figure 3 The specific risk assessment process is as follows: (1) For the current assessment example, the multi-level risk indicator system obtained by dividing the energy storage system can be determined based on the energy storage risk assessment system. The assessment example includes: multi-level risk indicators, risk coupling relationship network and risk score correction. Among them, the division of multi-level risk indicators can refer to the above content and will not be detailed here; the risk coupling relationship network is an acyclic directed graph, which is used at least to sort out the correlation and influence relationship between risk indicators in the second-lowest level, that is, the correlation and influence relationship between sub-risk factors in the sub-risk factor layer.

[0044] First, calculate the initial risk scores for each risk indicator at the lowest level, which includes the following steps: The lowest level of risk indicators are divided into quantitative and qualitative items; For quantitative items, an initial risk score is obtained using a cumulative risk point model based on alarm level frequency; for qualitative items, an initial risk score is obtained using a scoring model with three dimensions: severity, frequency of occurrence, and exposure.

[0045] Based on the initial risk scores of each risk indicator at the lowest level, and based on the weights of the risk indicators at each level, the initial risk scores of each level of risk indicators are calculated respectively. Subsequently, the coupling relationship network between the risk indicators at the same level is used to correct the initial risk scores of the risk indicators at non-lowest levels, resulting in corrected risk scores for each risk indicator. The coupling relationship network represents the potential impact of the influence relationship between risk indicators at each level on the risk of the energy storage system. The corrected risk score is obtained by subtracting downwards based on the initial risk score and the strength of the coupling relationship between the indicators indicated by the coupling relationship network.

[0046] From the above, we can obtain the initial index scores of risk indicators at each level, the revised risk scores of risk indicators at non-lowest levels, and the coupling relationship network. For relevant content about the above process, please refer to the content disclosed in the existing application with application number CN121073208A. All content in that patent can be cited in this application.

[0047] For example, the risk factor tree and risk coupling relationship network corresponding to the multi-level risk indicator system of an energy storage system, such as... Figure 4 As shown.

[0048] The multi-level risk indicator system for energy storage systems also includes: the weights of risk indicators at each level; specifically, the weights of risk factors under each subsystem, the weights of each sub-risk factor under each risk factor, and the weights of each risk item under each sub-risk factor. Except for the lowest-level risk indicator (risk item), the sum of the weights of all its subordinate risk indicators is 1. The calculation process for the weights of each risk indicator can be found in the application disclosed in application number CN121073208A, and will not be detailed here.

[0049] (2) Risk positioning in the hierarchical risk index system. The goal of risk positioning is to identify the key lower-level risk sources that drive the degradation of upper-level nodes in the selected assessment cases.

[0050] Based on the modified risk scores and weights of each risk indicator in the second-lowest level, at least one first risk indicator is selected from the risk indicators in the second-lowest level. Specifically: the risk indicators are traversed from top to bottom to the second-lowest level. For each risk indicator traversed: based on the modified risk scores of each lower-level risk indicator of the risk indicator, the risk severity level of each lower-level risk indicator is determined, and based on the risk severity level of each lower-level risk indicator, each lower-level risk indicator that can be retained is selected; based on the modified risk scores and weights of each lower-level risk indicator of the risk indicator, the first risk impact factor of the lower-level risk indicator is calculated, and based on the first risk impact factor of each lower-level risk indicator, each lower-level risk indicator that can be retained is selected; all the retainable risk indicators in the second-lowest level are used as the first risk indicators.

[0051] In detail, for each risk indicator in the risk factor tree Calculate the risk severity level of risk indicator i respectively. and the first risk factor ; ; ; in, This represents the adjusted risk score for risk indicator i. This represents the weight of risk indicator i, which is its local structural weight relative to its parent node (the higher-level risk indicator). It represents the set of all risk indicators in the risk factor tree (i.e., a multi-level indicator system).

[0052] First risk impact factor It is used to measure the contribution of risk indicator i to the degree of correction of the parent node score.

[0053] The risk identification process is executed recursively from top to bottom from the highest level (i.e., the root node r) of the multi-level risk indicators (risk factor tree); Each risk indicator that is not at the lowest level, i.e., any parent node ,set up For the set of sub-index nodes of its lower-level risk indicators, retain a subset of them. This is used to check for risks at a lower level.

[0054] Retained subset It is determined by the following two parallel rules: Rule 1: For each risk indicator that is not at the lowest level, the highest preset risk severity level shall be applied. To select which sub-risk indicators to retain for this risk indicator; The highest risk level ; If the highest severity level : Retain all child nodes (i.e., lower-level risk indicators) with a risk severity level of 3 from all lower-level risk indicators of this risk indicator. If the highest severity level : Retain the two child nodes (i.e., lower-level risk indicators) among all the lower-level risk indicators of this risk indicator that have a risk verification level of 2 and a corrected risk score of the reciprocal. If the highest severity level If the risk indicator is not specified, then all its subordinate risk indicators will not be retained.

[0055] Rule 2: For each risk indicator that is not at the lowest level, select which lower-level risk indicators to retain based on the first risk impact factor of that risk indicator. Specifically, from all the lower-level risk indicators of that risk indicator, retain the two child nodes (i.e., lower-level risk indicators) whose first risk impact factor is greater than 0 and whose first risk impact factor is the largest.

[0056] The risk indicators selected by Rule 1 and Rule 2 above are independent of each other, and the final retained subset This is the union of the sets of risk indicators selected based on the above two rules.

[0057] In this embodiment, the process recursively starts from the root node of the risk factor tree and proceeds to the next lowest level, i.e., the sub-risk factor layer, to obtain all the risk indicators (i.e., sub-risk factor nodes) that should be retained in the next lowest level. These risk indicators are the first risk indicators, and all the first risk indicators will constitute the local risk seed set. As the starting point for subsequent causal tracking phases; among which, It is the set of all risk indicators in the next lower level, that is, the set of sub-risk factor nodes.

[0058] (3) Construction of the causal perception risk tracing chain; After risk positioning, causal tracing is performed at the next lower level (sub-risk factor layer). The goal is to construct an energy storage risk propagation path supported by the modified risk score value based on the coupling relationship network. For risk indicators that do not have a modified risk score, the risk indicator is not included in the tracing stage.

[0059] For the current evaluation instance, its coupling network It can be represented as: ; in, It is the set of all risk indicators in the next lower level, that is, the set of sub-risk factor nodes; It is the set of directed edges in a coupled network; This refers to the set of coupling correction coefficients corresponding to the coupling relationship network. These coefficients characterize the correlation strength between two risk indicators in the secondary level. For example, the coupling correction coefficients between risk indicator x and risk indicator y in the secondary level... The expression is: ; in, This represents the initial risk score for the risk indicator y. This represents the initial risk score for risk indicator x.

[0060] Subsequently, based on the coupling relationship network between the risk indicators in the next lower level, a second risk indicator that can be reached along a directed path or that can be reached by the first risk indicator along a directed path is selected.

[0061] From local risk seed set Starting from the coupling relationship network Extract the reachable risk indicator, denoted as the second risk indicator: the second risk indicator can be: the first risk indicator can be obtained along the coupling relationship network. The path leads to the risk indicator, where the first risk indicator points to the second risk indicator; or it can be along the coupling network. The path reaches the first risk indicator, and at this point, the second risk indicator points to the first risk indicator.

[0062] The first risk indicator and the second risk indicator mentioned above constitute the target node set. The directed paths from the first risk indicator to the second risk indicator and the directed paths from the second risk indicator to the first risk indicator constitute a reachable edge set. .

[0063] For example, evaluate a subset of instances This includes assessment examples of Power Plant 2 in March 2024 and April 2024; risk identification can be performed on these two examples to obtain the following results: Figure 5 The diagram shown illustrates the effectiveness of risk positioning.

[0064] Based on the first risk indicator, the second risk indicator, and the risk indicators connected by each of the directed paths, at least one risk propagation path is obtained. Each risk propagation path includes at least one of the first risk indicators, and no two risk propagation paths overlap. For each of the risk indicators connected by the directed path, the difference between the initial risk score and the corrected risk score of the risk indicator is calculated as the coupling-induced score of the risk indicator. Subsequently, based on the coupled-induced scoring of the risk indicator, the most relevant target directed path is selected from all directed paths pointing to the risk indicator, specifically: Based on the coupling-induced score of the risk indicator, the degradation level of the risk indicator caused by risk coupling is determined, and the degradation level includes: high risk level, medium risk level and low risk level. If the degradation level corresponding to the risk indicator is a high-risk level, then all directed paths pointing to the risk indicator will be regarded as the most relevant target directed paths. If the degradation level corresponding to the risk indicator is medium risk level, then select a portion of the directed paths from all the directed paths pointing to the risk indicator as the most relevant target directed paths; If the degradation level corresponding to the risk indicator is low risk, then the most relevant target directed path will not be selected from all directed paths pointing to the risk indicator.

[0065] For example, for the target node set The risk indicator v in the middle, its coupling induces a score The expression is: ;in, This represents the initial risk score of the risk indicator v. This represents the adjusted risk score for the risk indicator v.

[0066] Coupling-induced scoring based on risk indicator v Determine the level of degradation of risk indicator v due to risk coupling. The expression is: ; It can be seen that coupling-induced scoring The larger the value, the stronger the coupling correction and score reduction capability. Among them, level 3 is the high-risk level, level 2 is the medium-risk level, and level 1 is the low-risk level.

[0067] Then from the reachable edge set The target directed path is selected from the most relevant propagation edges. The specific selection method is as follows: The highest risk degradation level is defined as: ; Traverse the target node set All risk indicators in the dataset, for each risk indicator: If the risk indicator is at a high risk level, then all reachable edges pointing to the risk indicator are retained; If the risk indicator is of medium risk level, then retain the two reachable edges that have the largest coupling-induced score and point to the risk indicator. If the risk indicator is at a low risk level, then reachable edges pointing to the risk indicator will not be retained.

[0068] From the above, all directed paths to the target constitute the set of retained edges. .

[0069] Additionally, in reachable edge sets If the set is empty, a directed path to the target cannot be selected.

[0070] Then, extend and connect all the directed paths that connect the same risk indicator in the directed paths of the target to obtain at least one risk propagation path; Based on the target node set With reachable edge set A maximal directed path is constructed based on the target directed path, where at least one node in each maximal directed path is a first risk indicator, and the maximal directed path cannot extend to either end of the target directed path; the set of maximal directed paths constitutes the risk propagation path. .

[0071] For example, in the lower-level coupling network Risk transmission paths identified in By tracing the risk propagation path, the root causes of the risk can be identified, and potential downstream risks can be found. Please refer to [reference needed]. Figure 6 The dashed box represents the traced risk propagation path. .

[0072] For each risk indicator on the aforementioned risk propagation path, the associated abnormal risk indicator is selected based on the initial risk score of the lower-level risk indicator associated with the risk indicator.

[0073] For example, regarding risk propagation paths Each risk indicator s on the chart is a tracking sub-risk factor s. Its associated set of lower-level risk indicators (risk items) nodes is: This is the lowest level of risk indicator, which only includes the initial risk score.

[0074] Based on the initial risk scores and weights of each subordinate risk indicator associated with the aforementioned risk indicator, a second risk impact factor is calculated for each subordinate risk indicator; the subordinate risk indicator with the largest second risk impact factor is selected as the abnormal risk indicator.

[0075] For example, the formula for calculating the second influencing factor of the subordinate risk indicator j of risk indicator s is: ; in, This represents the second influencing factor of the lower-level risk indicator j. This represents the initial risk score of the lower-level risk indicator j. This represents the weight of the lower-level risk indicator j, that is, its local structural weight relative to the parent node s.

[0076] Then, the most representative abnormal risk indicator (abnormal risk item) among the lower-level risk indicators j of risk indicator s is: ; Based on the above, regarding the risk transmission path For each risk indicator, the subordinate risk indicator with the largest impact of the second risk factor can be taken as its most representative abnormal risk indicator (abnormal risk item) as observable evidence; thus, the risk propagation path... Representative anomaly risk indicators for all risk indicators can be summarized into the following set of trackable anomaly items. : ; in, Indicates the risk transmission path The set of nodes for all risk indicators.

[0077] Based on the above process, two closely related results can be output: the risk propagation path of risk indicators in the second-lowest level. Risk transmission path Representative item-level evidence corresponding to each of the above risk indicators .

[0078] (4) Structured prompts and evidence-driven reasoning in the large language model: After risk identification and risk tracing, instead of directly inputting a large number of original monitoring records into the large language model, the risk assessment results are organized into meta-data packages to guide the large language model to reason based on clear logical hierarchies and causal evidence, so as to improve the interpretability and robustness of the energy storage operation and maintenance framework; specifically: The metadata generated by the risk analysis agent consists of three parts: Hierarchical risk factor tree: During the positioning phase, a multi-level risk factor tree is generated with the power plant-level assessment level as the root, recording the risk indicators retained in each level of subsystem, risk factor, and sub-risk factor; each risk indicator includes the initial risk score, the revised risk score, the coupling-induced score, the risk severity level, the first risk impact factor, the second risk impact factor, and the reason for positioning retention.

[0079] Risk indicators at the next lower level, i.e., the causal tracing chain between sub-risk factors: the final propagation path on the coupling graph generated during the tracking phase. This describes how risk factors spread through score deterioration.

[0080] Supporting evidence at the lowest level of risk items: Each traceable sub-risk factor is accompanied by a representative anomalous risk item, allowing people's judgments on high-level risks to be clearly anchored to observable evidence.

[0081] Based on the aforementioned metadata, the large language model will act as a structured reasoning module rather than an unconstrained text generator, identifying the most critical risk degradation branches, explaining the physical meaning of the risk tracing chain, summarizing the supporting evidence for anomalies at the risk item level, and determining whether the observed energy storage risk anomalies point to a common root cause or originate from multiple relatively independent risk sources.

[0082] For example, for a single selected assessment instance, the structured evidence package output by the risk analysis agent is: ; in, For evidence-driven diagnostic results generated by the large language model for power plant n-cycle μ, the local risk seed set within power plant n-cycle μ is: Risk propagation path at the sub-risk factor level , Set of all sub-risk factor nodes in the path Supporting evidence for risk items .

[0083] Once multiple power plants and periodic instances are selected, the total output of the risk analysis agent is the set of corresponding risk assessment results. : ; in, It can be sent directly to the summarizing agent, or forwarded as risk-triggered evidence to the standard guidance agent or the mechanism explanation agent.

[0084] Step 103: If the target function includes a standard guidance function and the execution graph includes directed edges from the risk analysis function to the standard guidance function, then based on the user question query and the risk assessment result, target clause data is retrieved from the pre-built standard knowledge graph, and the target clause data package is transformed into evidence-driven information including explicit clause references.

[0085] Specifically, for the standard guidance agent corresponding to the standard guidance function, the leading agent, upon determining that the standard guidance agent needs to be invoked, i.e. The standard guidance agent's input includes at least the user's question query χ. The user's question query χ is parsed using a large language model to extract hints through standard queries. It is then determined whether the user's question query χ contains standard-related requirements. If it does, key phrases are extracted from the user's question query χ to construct the first query set for risk-triggered standard queries. The output will be displayed; if the user's question query χ does not contain standard related requirements, the first query set of the risk trigger standard query will be an empty set.

[0086] In some embodiments, if the execution diagram is... If an edge points from risk to std, then the input to the standard guidance agent std also includes: the risk assessment result output by the risk analysis agent risk. Specifically: the risk assessment results Supporting evidence for medium-risk items The second query set is transformed into a risk triggering standard query. Otherwise, the second query set is empty.

[0087] Based on the above, the effective standard query set for guiding intelligent agents is... .

[0088] The standard-guided agent first constructs a standard knowledge graph, then performs graph-guided retrieval on the graph, and finally uses the retrieved clause data as the context for interpretation by the large language model, thereby reducing the evidence space and improving the traceability and reliability of the final output. The specific workflow includes: 1. Construction of standard knowledge graph: Organize existing standards, clauses, and reference relationships related to energy storage systems into a graph structure to form a standard knowledge graph; 2. Graph-guided standard retrieval: Based on the elements of the effective standard query set, the entity of the most relevant clause (target clause) is identified in the standard knowledge graph according to semantic similarity. Then, along the entity relationship in the standard knowledge graph, the entity of the target clause is extended with context to obtain the extended associated entity (associated clause). Thus, the target clause data can be constructed from the target clause, associated clause, and connection relationship. 3. Large Language Model (LLM) Interpretation: The large language model in the standard guidance agent transforms the target clause data into evidence-driven information with explicit clause references. This evidence-driven information can be presented in report form and forwarded to the summarizing agent.

[0089] Step 104: If the target function includes a mechanism explanation function and the execution graph includes directed edges from the risk analysis function to the mechanism explanation function, then based on the user question query and the risk assessment result, target retrieval data with labeled evidence sources is retrieved from the pre-built energy storage knowledge base.

[0090] Specifically, for the mechanism explanation agent corresponding to the mechanism explanation function, the dominant agent, when determining that the mechanism explanation agent needs to be invoked, i.e. The input to the mechanism explanation agent includes at least the user's question query χ. The user's question query χ is parsed using a large language model to extract hints through mechanism queries. It is then determined whether the user's question query χ contains mechanism-related analysis requirements. If it does, key phrases are extracted from the user's question query χ to construct a third query set for risk-triggered mechanism queries. The third query set of the risk triggering mechanism query is output; if the user's question query χ does not include the mechanism-related analysis requirements, the output third query set of the risk triggering mechanism query is an empty set.

[0091] In some embodiments, if the execution diagram is... If the edge contains a path from risk to mech, then the input to the mechanism explanation agent mech also includes the risk assessment result output by the risk analysis agent risk. Specifically: the risk assessment results Supporting evidence for medium-risk items The fourth query set is transformed into a risk triggering mechanism query. Otherwise, the fourth query set is empty.

[0092] Based on the above, the effective mechanism query set for explaining intelligent agents is explained. .

[0093] The mechanism explanation agent has a pre-set independent electrochemical energy storage knowledge base, which mainly covers the failure mechanism and risk assessment knowledge of electrochemical energy storage systems. This energy storage knowledge base does not directly store the original document data, but extracts the original document data into semantically complete knowledge units for storage, so as to improve retrieval accuracy and reduce irrelevant context. Mechanistic explanation of intelligent agents will effectively query the mechanism query set. Each element in the database is matched with a knowledge index in the energy storage knowledge base. For each element, the similarity between the element and the knowledge unit in the energy storage knowledge base can be calculated by cosine similarity. The query result corresponding to the maximum cosine similarity is taken as the best index knowledge fragment of the element. The best index knowledge fragments of all elements form a retrieval package.

[0094] The large language model in the mechanism explanation agent then interprets the retrieval package and generates statements of the best index knowledge fragments for each element. These statements are labeled with the sources of evidence, including: [RAG] statements, which are mainly based on retrieval mechanism knowledge; [LLM] statements, which are mainly generated by model inference; and [RAG][LLM] statements, which are jointly synthesized from retrieval knowledge and model inference.

[0095] The target retrieval data output by the mechanism-explaining agent includes: a retrieval package formed by the best index knowledge fragments of all elements, and statements corresponding to each element for labeling the source of evidence; the target retrieval data is sent to the summarizing agent.

[0096] For user queries of two types—knowledge query and risk location query—in the energy storage system risk management system of this embodiment, for knowledge query requests, the system performs semantic matching and content retrieval in the expert knowledge base based on the user query, integrating expert knowledge with the inference capabilities of the large model to output a knowledge-based answer that is highly relevant to the question and logically consistent. For risk location queries, this framework first decomposes the user query into sub-questions by the risk location agent, breaking it down into two sub-tasks: risk assessment and analysis, and knowledge query answering. Then, it completes the risk assessment and analysis and knowledge reasoning respectively, and merges and summarizes the two types of results to form the final operation and maintenance response.

[0097] Please refer to Figure 7 This refers to the response from the intelligent agent in the energy storage system risk management system when the risk is identified as: a short circuit within the battery cell; and when a user queries: "Based on existing knowledge, what are the applications of grid-scale battery energy storage systems?", the intelligent agent in the energy storage system risk management system provides the response.

[0098] Step 105: Based on the output data corresponding to all the invoked target functions, generate and output a summary report corresponding to the user problem query. The summary report includes at least: relevant information of the risk assessment results; further, the summary report also includes: operation and maintenance opinions and strategies corresponding to the evidence-driven information, and / or risk analysis information corresponding to the target retrieval data.

[0099] Specifically, the summary agent receives the output data of all invoked target functions and transforms this output data into a coherent, user-oriented report; the specific process is as follows: In summary, the intelligent agent follows the execution plan information Π(χ) generated for the user's query χ, and collects the execution graph. The output of the activated agent is called; for example The set of outputs of the invoked agents under the current execution plan, based on the set of activated agents. , It can contain the output set of the risk analysis agent. Standard guidance for the output set of intelligent agents Mechanistic explanation of the intelligent agent's output set .

[0100] The final output of user question query χ for: ; Fuse(·) is the report generation operator that summarizes the agent, and Fuse(·) is evidence-driven synthesis performed by the large language model.

[0101] In this embodiment, in addition to including the output generated by the called agent, the summarizing agent further refines the multi-source information into a concise summary for operations and maintenance personnel, presenting key findings, major risks, key evidence, and recommended points of attention in a compact form to help users quickly grasp the core information before reading the full report; the final output should be concise, traceable, and practical for operations and maintenance personnel.

[0102] For example, the query χ for a user question is: Conduct a risk assessment of Power Station 2 between March and April 2024. Based on relevant standards, propose maintenance recommendations and use relevant knowledge to explain the causes of the identified risks. For a summary report (summary report) of the energy storage risk management and operation and maintenance provided by the intelligent agent, please refer to... Figure 8 .

[0103] As can be seen from the above, this embodiment, under the conditions of large-scale operational energy storage power stations and massive operational data, achieves a fully automated, interpretable, and traceable closed-loop support for electrochemical energy storage from risk assessment to risk governance based on evidence-driven and multi-agent collaborative technologies. Specifically, it includes: constructing a hierarchical risk index and coupled correction assessment system; quantitatively analyzing long-term field data to form a risk score and propagation path dataset that supports precise location; simultaneously, constructing a standard knowledge graph based on national standards in the energy storage field and performing knowledge distillation on authoritative literature to form a reliable knowledge base supported by both standard basis and mechanistic knowledge; and using a multi-agent collaborative mechanism to perform intent parsing, task decomposition, and collaborative reasoning on user queries, generating executable, verifiable, and traceable energy storage risk governance reports by integrating risk data, standard provisions, and mechanistic knowledge. This reduces reliance on manual labor, improves the efficiency of anomaly tracing and governance, avoids the omission of key risks, and provides safe, compliant, and expert-level decision support for energy storage safety operation and maintenance personnel.

[0104] A risk localization and path tracing method based on hierarchical risk indexing and coupled propagation analysis; a standard knowledge graph construction, chapter-level retrieval, and cross-standard association reasoning method for electrochemical energy storage; a failure mechanism knowledge base organization and evidence anchoring generation method based on domain literature distillation; and a multi-agent collaborative scheduling, task decomposition, evidence fusion, and structured operation and maintenance report generation mechanism; by combining physical monitoring data, risk assessment results, standard knowledge, and mechanism knowledge to form a closed-loop decision-making process, it achieves intelligent operation and maintenance across the entire chain from data to risk, from risk to measures, and from measures to execution, thereby improving the efficiency, compliance, and reliability of electrochemical energy storage system safety governance and avoiding the omission of cross-system coupled risks and root causes.

[0105] A second embodiment of the present invention relates to a risk management system for an energy storage system, comprising: a leading agent, an analyzing agent (risk), and a summarizing agent (sum); further, it may also include: a standard guiding agent (std) and / or a mechanism interpreting agent (mech); exemplarily, such as... Figure 2 As shown, the energy storage system risk management system includes: a leading agent, an analytical agent (risk), a summarizing agent (sum), a standard guiding agent (std), and a mechanism explaining agent (mech).

[0106] The dominant intelligent agent is used to respond to user-inputted questions and queries, and parse the structured execution plan information. The execution plan information includes: the intelligent agent corresponding to the target function to be called, the execution graph between the intelligent agents corresponding to the target function, and the evaluation data range. The intelligent agent corresponding to the target function to be called includes the risk analysis intelligent agent. The risk analysis agent is used to determine the risk assessment result of the energy storage system corresponding to each assessment instance based on risk analysis of at least one assessment instance obtained by mapping the assessment data range. The energy storage system includes multi-level risk indicators. The risk assessment result includes: the second-lowest level risk indicator with risk, the risk propagation path containing the second-lowest level risk indicator with risk, and the lowest level abnormal risk indicator associated with each of the risk indicators on the risk propagation path. The summary agent uses the output data of all the agents corresponding to the invoked target functions to generate and output a summary report corresponding to the user question query. The summary report includes at least the relevant information of the risk assessment results.

[0107] For example, the system further includes: a standard guidance agent; the agent corresponding to the target function to be invoked also includes the standard guidance agent, and the execution graph includes directed edges from the risk analysis function to the standard guidance agent; The standard guidance agent is used to retrieve target clause data from a pre-built standard knowledge graph based on the user's question query and the risk assessment results, and to transform the target clause data package into evidence-driven information including explicit clause references.

[0108] The summary report generated by the summarizing agent also includes: the operation and maintenance opinion strategy corresponding to the evidence-driven information.

[0109] For example, the system further includes: a mechanism explanation agent; the agent corresponding to the target function to be invoked also includes the mechanism explanation agent, and the execution graph includes directed edges from the risk analysis agent to the mechanism explanation agent; The mechanism explanation agent is used to retrieve target retrieval data with labeled evidence sources from a pre-built energy storage knowledge base based on the user question query and the risk assessment results. The summary report generated by the summarizing agent also includes risk analysis information corresponding to the target retrieved data.

[0110] This embodiment utilizes a hierarchical risk localization, coupled propagation and tracing, standard knowledge graph, mechanism knowledge base, and multi-agent collaborative governance architecture to transform raw collected physical data such as voltage, temperature, insulation, and alarms into structured risk information that is localizable, quantifiable, and traceable. Simultaneously, it performs knowledge distillation and graph organization of national standards, industry norms, and failure mechanisms to form an authoritative knowledge system that directly supports compliant operation and maintenance and physical mechanism explanation. Furthermore, through multi-agent intent parsing, task decomposition, division of labor execution, and result fusion for operation and maintenance queries, the system can automatically complete the entire closed-loop process of risk localization, path tracing, standard matching, mechanism explanation, and report generation, outputting authoritative governance solutions with standard clauses, physical evidence, and execution steps without human intervention. This mechanism not only significantly improves the accuracy and coverage of risk identification and completely breaks down the critical gap between risk assessment and implementation of operation and maintenance, but also ensures that every operation and maintenance suggestion has standards to follow, mechanisms to explain, and data to support it, significantly reducing reliance on experts and manual workload, while supporting unified and standardized governance across sites, cycles, and subsystems. This intelligent operation and maintenance capability, which is explainable, traceable, executable, and closed-loop, can help on-site personnel quickly identify the root cause, block the coupling propagation, and implement compliant rectification, effectively reducing major safety hazards such as thermal runaway, improving operation and maintenance efficiency and governance level, and ultimately forming outstanding advantages in ensuring system security, reducing missed judgments and misjudgments, improving compliance, and achieving large-scale intelligent operation and maintenance.

[0111] Since the first embodiment corresponds to this embodiment, this embodiment can be implemented in conjunction with the first embodiment. The relevant technical details mentioned in the first embodiment remain valid in this embodiment, and the technical effects achievable in the first embodiment can also be achieved in this embodiment. To reduce repetition, they will not be repeated here. Correspondingly, the relevant technical details mentioned in this embodiment can also be applied to the first embodiment.

[0112] The third embodiment of the present invention relates to a risk management device for an energy storage system. The device may be, for example, an electronic device, a server, or a common computer device, such as a desktop computer or a laptop computer.

[0113] The energy storage system risk management device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the energy storage system risk management method in the first embodiment.

[0114] It should be noted that the memory may include random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. Similarly, the processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0115] Since the first embodiment corresponds to this embodiment, this embodiment can be implemented in conjunction with the first embodiment. The relevant technical details mentioned in the first embodiment remain valid in this embodiment, and the technical effects achievable in the first embodiment can also be achieved in this embodiment. To reduce repetition, they will not be repeated here. Correspondingly, the relevant technical details mentioned in this embodiment can also be applied to the first embodiment.

[0116] The fourth embodiment of the present invention relates to a computer-readable storage medium, which is a non-volatile storage medium or a non-transient storage medium, and stores a computer program thereon. The computer program is executed by a processor to perform the energy storage system risk management method in the first embodiment.

[0117] The preferred embodiments of the present invention have been described in detail above, but it should be understood that, if necessary, aspects of the embodiments can be modified to utilize aspects, features, and concepts from various patents, applications, and publications to provide other embodiments.

[0118] In light of the detailed description above, these and other changes can be made to the embodiments. Generally, the terminology used in the claims should not be considered limited to the specific embodiments disclosed in the specification and claims, but should be understood to include all possible embodiments together with the full scope of equivalents enjoyed by these claims.

Claims

1. A risk management method for energy storage systems, characterized in that, include: In response to a user's query, structured execution plan information is parsed and obtained. The execution plan information includes: the target function to be invoked, the execution graph between the target functions, and the evaluation data range. If the target function includes a risk analysis function, then based on the risk analysis performed on at least one assessment instance obtained by mapping the assessment data range, the risk assessment result of the energy storage system corresponding to each assessment instance is determined. The energy storage system includes multi-level risk indicators, and the risk assessment result includes: the second-lowest level risk indicator with risk, the risk propagation path containing the second-lowest level risk indicator with risk, and the lowest level abnormal risk indicator associated with each of the risk indicators on the risk propagation path. Based on the output data corresponding to all the invoked target functions, a summary report corresponding to the user problem query is generated and output. The summary report includes at least the relevant information of the risk assessment results.

2. The energy storage system risk management method according to claim 1, characterized in that, Based on risk analysis performed on at least one assessment instance obtained by mapping the assessment data range, the risk assessment result of the energy storage system corresponding to each assessment instance is determined, including: For each evaluation instance: The initial risk scores of the risk indicators at each level are obtained, and the initial risk scores of the risk indicators at non-lowest levels are corrected based on the coupling relationship network between the risk indicators at the same level, so as to obtain the corrected risk scores of each risk indicator. Based on the corrected risk scores and weights of each risk indicator in the non-lowest level, at least one first risk indicator is selected from the risk indicators in the second-lowest level, and based on the coupling relationship network between the risk indicators in the second-lowest level, a second risk indicator that can be reached along a directed path or that can be reached by the first risk indicator along a directed path is selected. Based on the first risk indicator, the second risk indicator, and the risk indicators connected by each of the directed paths, at least one risk propagation path is obtained. Each risk propagation path includes at least one of the first risk indicators, and no two risk propagation paths overlap. For each risk indicator on the aforementioned risk propagation path, the associated abnormal risk indicator is selected based on the initial risk score of the lower-level risk indicator associated with the risk indicator.

3. The energy storage system risk management method according to claim 2, characterized in that, Extend and connect all directed paths that connect to the same risk indicator to obtain at least one risk propagation path, including: For each risk indicator connected by the directed path, the difference between the initial risk score and the corrected risk score of the risk indicator is calculated as the coupling-induced score of the risk indicator, and based on the coupling-induced score of the risk indicator, the most relevant target directed path is selected from all the directed paths pointing to the risk indicator. Extend and connect all the directed paths that connect the same risk indicator in the target directed paths to obtain at least one risk propagation path.

4. The energy storage system risk management method according to claim 3, characterized in that, Based on the coupled-induced scoring of the risk indicator, the most relevant target directed path is selected from all directed paths pointing to the risk indicator, including: Based on the coupling-induced score of the risk indicator, the degradation level of the risk indicator caused by risk coupling is determined, and the degradation level includes: high risk level, medium risk level and low risk level. If the degradation level corresponding to the risk indicator is a high-risk level, then all directed paths pointing to the risk indicator will be regarded as the most relevant target directed paths. If the degradation level corresponding to the risk indicator is medium risk level, then select a portion of the directed paths from all the directed paths pointing to the risk indicator as the most relevant target directed paths; If the degradation level corresponding to the risk indicator is low risk, then the most relevant target directed path will not be selected from all directed paths pointing to the risk indicator.

5. The energy storage system risk management method according to claim 2, characterized in that, Based on the modified risk scores and weights of the risk indicators in the non-lowest level, at least one first risk indicator is selected from the risk indicators in the second-lowest level, including: The risk indicators are traversed from top to bottom and then to the next lowest level. For each risk indicator encountered: Based on the corrected risk scores of each subordinate risk indicator of the aforementioned risk indicator, the risk severity level of each subordinate risk indicator is determined, and based on the risk severity level of each subordinate risk indicator, each subordinate risk indicator that can be retained is selected. Based on the corrected risk scores and weights of each of the lower-level risk indicators of the risk indicators, the first risk impact factor of the lower-level risk indicators is calculated, and based on the first risk impact factor of each of the lower-level risk indicators, each of the lower-level risk indicators that can be retained is selected. All the risk indicators that can be retained in the second-lowest level are used as the first risk indicator.

6. The energy storage system risk management method according to claim 2, characterized in that, Based on the initial risk score of the lower-level risk indicators associated with the aforementioned risk indicator, the abnormal risk indicators associated with the aforementioned risk indicator are selected, including: Based on the initial risk scores and weights of each subordinate risk indicator associated with the aforementioned risk indicator, a second risk impact factor for each subordinate risk indicator is calculated. The lower-level risk indicator with the largest second risk impact factor is selected as the abnormal risk indicator.

7. The energy storage system risk management method according to claim 1, characterized in that, If the target function includes a standard guidance function and the execution graph includes directed edges from the risk analysis function to the standard guidance function, then based on the user question query and the risk assessment result, the target clause data is retrieved from the pre-built standard knowledge graph, and the target clause data package is transformed into evidence-driven information including explicit clause references. The summary report also includes: operation and maintenance suggestions and strategies corresponding to the evidence-driven information.

8. The energy storage system risk management method according to claim 1, characterized in that, If the target function includes a mechanism explanation function and the execution graph includes directed edges from the risk analysis function to the mechanism explanation function, then based on the user question query and the risk assessment result, target retrieval data with labeled evidence sources can be retrieved from the pre-built energy storage knowledge base. The summary report also includes risk analysis information corresponding to the target retrieved data.

9. A risk management system for energy storage systems, characterized in that, include: The leading agent, the risk analysis agent, and the summarizing agent; The dominant intelligent agent is used to respond to user-inputted questions and queries, and parse the structured execution plan information. The execution plan information includes: the intelligent agent corresponding to the target function to be called, the execution graph between the intelligent agents corresponding to the target function, and the evaluation data range. The intelligent agent corresponding to the target function to be called includes the risk analysis intelligent agent. The risk analysis agent is used to determine the risk assessment result of the energy storage system corresponding to each assessment instance based on risk analysis of at least one assessment instance obtained by mapping the assessment data range. The energy storage system includes multi-level risk indicators. The risk assessment result includes: the second-lowest level risk indicator with risk, the risk propagation path containing the second-lowest level risk indicator with risk, and the lowest level abnormal risk indicator associated with each of the risk indicators on the risk propagation path. The summary agent uses the output data of all the agents corresponding to the invoked target functions to generate and output a summary report corresponding to the user question query. The summary report includes at least the relevant information of the risk assessment results.

10. The energy storage system risk management system according to claim 9, characterized in that, The system further includes: a standard guidance agent; the agent corresponding to the target function to be invoked also includes the standard guidance agent, and the execution graph includes directed edges from the risk analysis function to the standard guidance agent; The standard guidance agent is used to retrieve target clause data from a pre-built standard knowledge graph based on the user question query and the risk assessment result, and to transform the target clause data package into evidence-driven information including explicit clause references. The summary report generated by the summarizing agent also includes: the operation and maintenance opinion strategy corresponding to the evidence-driven information.

11. The energy storage system risk management system according to claim 9, characterized in that, The system further includes: a mechanism explanation agent; the agent corresponding to the target function to be invoked also includes the mechanism explanation agent, and the execution graph includes directed edges from the risk analysis agent to the mechanism explanation agent; The mechanism explanation agent is used to retrieve target retrieval data with labeled evidence sources from a pre-built energy storage knowledge base based on the user question query and the risk assessment results. The summary report generated by the summarizing agent also includes risk analysis information corresponding to the target retrieved data.

Images