Power equipment whole life cycle management system based on digital twinning technology

Abstract

The application relates to the technical field of power asset management, and discloses a power equipment whole life cycle management system based on digital twin technology, which comprises a causal knowledge graph construction unit, a strategy rule learning unit, a decision option value quantification unit, a combined decision optimization unit and a maintenance decision voucher management unit. The method comprises the following steps: constructing a knowledge graph comprising the causal relationship among equipment, environment and management activities; performing counterfactual deduction based on the graph to dynamically learn strategy rules; quantifying the value of potential maintenance decisions by using a real option model; performing combined optimization under the constraints of resources such as budget and spare parts to generate an optimal decision combination; and finally generating standardized maintenance decision vouchers for the decisions in the combination. The application can quantify decision flexibility and environmental uncertainty into specific economic values, realize future-oriented and globally optimal resource allocation, and solidify the decision process and results into traceable and manageable digital assets.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power asset management, in particular to a power equipment whole life cycle management system based on digital twin technology. BACKGROUND

[0002] As the core asset of the power grid, the health management and cost control of the whole life cycle of power equipment are crucial to the safe and stable operation of the power system. With the development of digital technology, digital twin, big data analysis and other technologies have been gradually applied to the management of power equipment, aiming to improve the scientificity of operation and maintenance decisions through data-driven methods.

[0003] However, the existing technology still has some limitations in practice. The analysis of massive operation data in the existing technology mostly stays at the level of state monitoring and correlation analysis, and fails to deeply mine the deep causal relationship between equipment state, external environment and operation and maintenance activities, resulting in insufficient basis for decision-making. Therefore, the development of maintenance strategies often relies on fixed threshold rules or personal experience, and lacks a closed-loop learning mechanism that can dynamically evolve and self-optimize according to actual results.

[0004] In terms of economic evaluation of decision-making, traditional cost-benefit analysis methods are difficult to quantify the value of future uncertainty and management decision flexibility, such as the option value of delayed maintenance and phased investment, which makes the real economic benefits of decision-making underestimated. In addition, the decision-making process is often decentralized and isolated, lacking a mechanism for systematic combination optimization under global resource (such as budget, spare parts, manpower) constraints, making it difficult to maximize overall benefits, and the formation process and value basis of decision-making also lack standardized recording and management means, which is not conducive to the accumulation and reuse of knowledge.

[0005] Therefore, the present application proposes a power equipment whole life cycle management system based on digital twin technology to solve the deficiencies of the prior art. SUMMARY

[0006] In view of the deficiencies of the prior art, the present application provides a power equipment whole life cycle management system based on digital twin technology, which solves the problem that in the management of power equipment, the decision-making process lacks the mining of deep causal logic, the quantification of the value of future uncertainty, and the optimization ability of global resources, resulting in the economic value of maintenance decisions cannot be maximized and the process is not transparent.

[0007] To achieve the above purpose, the present application realizes the following technical scheme: a power equipment whole life cycle management system based on digital twin technology, the system comprises:

[0008] The causal knowledge graph construction unit is configured to obtain design data, operation data, maintenance data and external environment data of the power equipment to form a full life cycle dataset, and construct a causal knowledge graph containing causal relationships between equipment entities, environmental factors and management activities based on the full life cycle dataset.

[0009] The strategy rule learning unit is configured to perform counterfactual reasoning on historical maintenance events based on the causal knowledge graph to determine utility gains of alternative decisions, generate new strategy rules, and inject the new strategy rules into the causal knowledge graph for dynamic updating.

[0010] The decision option value quantification unit is configured to identify potential maintenance decision options based on the dynamically updated causal knowledge graph, and calculate a real option value of each potential maintenance decision option using a real option model.

[0011] The combined decision optimization unit is configured to receive the potential maintenance decision options and the real option values, and perform combined optimization under preset resource constraints to generate an optimal decision combination that maximizes the total real option value.

[0012] The maintenance decision voucher management unit is configured to generate a standardized maintenance decision voucher containing task obligations, opportunity values and traceability information for selected maintenance decision options based on the optimal decision combination.

[0013] Preferably, the causal knowledge graph construction unit is specifically configured to:

[0014] The constraint-based causal discovery algorithm is used to learn and determine the preliminary causal structure between data variables by performing a series of conditional independence tests on the operation data in the full life cycle dataset.

[0015] The verified equipment failure analysis patterns and domain expert knowledge are converted into formalized causal rules, and the preliminary causal structure is verified, corrected or completed using the formalized causal rules.

[0016] Preferably, the strategy rule learning unit is specifically configured to:

[0017] The latent outcomes framework is used to simulate and calculate actual outcomes caused by the actual decisions taken in the historical maintenance events, and counterfactual outcomes that may occur if alternative decisions are taken.

[0018] The utility gain is determined by calculating the difference between the utility of the counterfactual outcomes and the utility of the actual outcomes, and new strategy rules are refined and generated based on the utility gain.

[0019] Preferably, the decision option value quantification unit is specifically configured to:

[0020] modeling each potential maintenance decision option as a real option;

[0021] and by taking the predicted avoidable loss present value as the underlying asset present value, the maintenance execution cost as the execution price, and combining the effective decision time window, the risk-free rate and the quantified uncertainty, into a preset real option pricing model to determine the real option value;

[0022] wherein the expected present value of future economic benefits and the quantified uncertainty are both calculated according to the dynamically updated causal knowledge graph.

[0023] Preferably, the calculation of the quantified uncertainty is specifically to take the dynamically updated causal knowledge graph as input, and to obtain by comprehensively quantifying at least two of the following sources of uncertainty:

[0024] the uncertainty of the deterioration rate of the equipment health state determined by the physical characteristics and historical operation data of the equipment itself;

[0025] the uncertainty of the external economic environment determined by the electricity market price and the future load forecast.

[0026] Preferably, the combined decision optimization unit is specifically used for:

[0027] constructing multiple potential maintenance decision options into a decision portfolio;

[0028] defining the budget, spare parts inventory and professional human resources as resource constraints of the decision portfolio;

[0029] using an integer programming algorithm to solve the optimal decision combination that maximizes the total real option value of the decision portfolio under the condition of satisfying the resource constraints.

[0030] Preferably, the maintenance decision voucher includes:

[0031] an obligation module for describing the specific content and technical quality standards of the maintenance task according to the diagnosis results of the causal knowledge graph;

[0032] an opportunity value module for recording the real option value associated with the maintenance task calculated by the decision option value quantification unit;

[0033] a traceability hash module for storing a hash value calculated according to an analysis snapshot used to generate the maintenance decision voucher, to establish an unalterable link between the maintenance decision voucher and the analysis snapshot.

[0034] Preferably, the maintenance decision voucher management unit is further used for:

[0035] Based on the maintenance decision voucher, a double-layer transaction platform including an internal market layer and an external market layer is constructed and managed;

[0036] The internal market layer provides a voucher value query and an internal resource transfer interface to support resource replacement between different departments within an enterprise based on the value of the maintenance decision voucher;

[0037] The external market layer provides a voucher listing, bidding and contract generation function to support market procurement of maintenance tasks between pre-authenticated third-party service providers.

[0038] Preferably, the system further comprises:

[0039] A fault diagnosis unit is configured to perform Bayesian reverse probability reasoning along the causal relationship from the monitored abnormal state node in the causal knowledge graph to calculate and output the probability ranking of the root cause node causing the abnormal state.

[0040] The application also provides a power equipment full life cycle management method based on digital twin technology, which comprises the following steps:

[0041] S1, obtaining design data, operation data, maintenance data and external environment data of the power equipment to form a full life cycle data set, and constructing a causal knowledge graph containing the causal relationship between equipment entities, environmental factors and management activities based on the full life cycle data set;

[0042] S2, based on the causal knowledge graph, the utility gain of the alternative decision is determined by counterfactual reasoning on the historical maintenance event, a new strategy rule is generated, and the new strategy rule is injected into the causal knowledge graph for dynamic updating;

[0043] S3, based on the dynamically updated causal knowledge graph, identify potential maintenance decision options, and calculate the real option value of each potential maintenance decision option using a real option model;

[0044] S4, receiving the potential maintenance decision options and real option values, and performing combinatorial optimization under the preset resource constraint condition to generate an optimal decision combination that maximizes the total real option value;

[0045] S5, according to the optimal decision combination, a standardized maintenance decision voucher containing task obligations, opportunity value and traceability information is generated for the selected maintenance decision option.

[0046] The application provides a power equipment full life cycle management system based on digital twin technology. The application has the following advantages:

[0047] 1. The invention discloses a causal knowledge graph that reveals deep causal relationships, and uses counterfactual reasoning to dynamically update strategy rules, surpassing state monitoring based on surface correlation, enabling more accurate fault tracing and forward-looking decision-making, and improving the depth and accuracy of life cycle state assessment and decision-making of power equipment.

[0048] 2. By introducing a real option model to quantify the value of potential maintenance decisions, the invention scientifically transforms technical and market uncertainties into decision-making option values, enabling dynamic evaluation of decision-making economic benefits. This provides a scientific basis for choosing the best investment opportunity in an uncertain environment, avoiding economic losses caused by premature or late decision-making.

[0049] 3. The invention constructs all potential maintenance decision options as a decision-making portfolio and optimizes the solution with the goal of maximizing total real option value, enabling the output of optimal decision combinations at the system level under budget, manpower and other actual resource constraints. This realizes the transition from handling isolated events to global resource optimization, improving the overall use of limited resources.

[0050] 4. By generating optimized maintenance decisions into standardized maintenance decision vouchers, the invention transforms intangible management behavior into tangible, traceable and transferable digital assets. This provides a technical foundation for subsequent value discovery and resource trading of maintenance tasks in internal or external markets, promoting more efficient allocation of operation and maintenance resources within the entire industry chain. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 The functional module architecture diagram of the power equipment life cycle management system based on digital twin technology of the invention;

[0052] Figure 2 The working flowchart of the causal knowledge graph construction unit of the invention;

[0053] Figure 3 The principle diagram of counterfactual reasoning of the strategy rule learning unit of the invention;

[0054] Figure 4 The principle diagram of value calculation of the decision option value quantization unit of the invention;

[0055] Figure 5 The data structure diagram of the standardized maintenance decision voucher of the invention;

[0056] Figure 6 The flowchart of the power equipment life cycle management method based on digital twin technology of the invention.

[0057] Wherein, 10, a causal knowledge graph construction unit; 20, a strategy rule learning unit; 30, a decision option value quantification unit; 40, a combined decision optimization unit; 50, a maintenance decision voucher management unit; 60, a fault diagnosis unit. DETAILED DESCRIPTION

[0058] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the specification of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0059] Referring to the drawings Figure 1 , Figure 1 is a functional module architecture diagram of a power equipment full life cycle management system based on digital twin technology according to an embodiment of the present application. The system provided by the present application can include: a causal knowledge graph construction unit 10, a strategy rule learning unit 20, a decision option value quantification unit 30, a combined decision optimization unit 40, and a maintenance decision voucher management unit 50.

[0060] The causal knowledge graph construction unit 10 is used to obtain a full life cycle data set of the power equipment, the data set including design data, operation data, maintenance data and external environment data, and to construct a causal knowledge graph containing the causal relationship between equipment entities, environmental factors and management activities based on the data set.

[0061] The strategy rule learning unit 20 is connected with the data output end of the causal knowledge graph construction unit 10. The strategy rule learning unit 20 receives the causal knowledge graph and performs counterfactual reasoning on historical maintenance events based on the causal knowledge graph to generate new strategy rules. The strategy rule learning unit 20 simultaneously outputs the new strategy rules generated by it to the causal knowledge graph construction unit 10 for dynamic updating of the causal knowledge graph.

[0062] The decision option value quantification unit 30 is connected with the output end of the strategy rule learning unit 20. The decision option value quantification unit 30 receives the dynamically updated causal knowledge graph, identifies potential maintenance decision options based on the dynamically updated causal knowledge graph, and calculates the real option value of each potential maintenance decision option using a real option model.

[0063] The combined decision optimization unit 40 is connected with the output end of the decision option value quantification unit 30. The combined decision optimization unit 40 receives the potential maintenance decision options and the corresponding real option values, and performs combined optimization under the preset resource constraint conditions to generate an optimal decision combination.

[0064] The maintenance decision credential management unit 50 is connected to the output end of the combined decision optimization unit 40. The maintenance decision credential management unit 50 receives the optimal decision combination and generates standardized maintenance decision credentials for the selected maintenance decision options in the combination.

[0065] In an optional embodiment, the system can further include a fault diagnosis unit 60 connected to the output end of the causal knowledge graph construction unit 10, for performing reverse probability reasoning in the causal knowledge graph to calculate and output the probability ranking of the root cause nodes.

[0066] Referring to the accompanying drawings Figure 6 , Figure 6 is a flowchart of a power equipment life cycle management method based on digital twin technology according to an embodiment of the present application. The method is applied to the aforementioned system and includes the following steps:

[0067] Step S1, obtain the full life cycle data set of the power equipment, and construct a causal knowledge graph based on the data set. This step is performed by the causal knowledge graph construction unit 10.

[0068] Step S2, based on the causal knowledge graph generated in step S1, perform counterfactual reasoning on historical maintenance events, generate new strategy rules, and inject the rules into the causal knowledge graph for dynamic updating. This step is performed by the strategy rule learning unit 20.

[0069] Step S3, based on the dynamically updated causal knowledge graph in step S2, identify potential maintenance decision options and calculate the real option value of each potential maintenance decision option. This step is performed by the decision option value quantification unit 30.

[0070] Step S4, receive the potential maintenance decision options and their real option values generated in step S3, perform combination optimization under the preset resource constraints, and generate an optimal decision combination. This step is performed by the combined decision optimization unit 40.

[0071] Step S5, according to the optimal decision combination generated in step S4, generate standardized maintenance decision credentials for the selected maintenance decision options. This step is performed by the maintenance decision credential management unit 50.

[0072] The internal implementation of each core unit in the system will be described in detail below.

[0073] Referring to the accompanying drawings Figure 2 , Figure 2This is a schematic diagram of the workflow of a causal knowledge graph construction unit 10 according to an embodiment of the present invention. The causal knowledge graph construction unit 10 functions to provide a knowledge foundation containing deep causal logic for subsequent analysis, learning, and decision-making. The first operation performed by this unit 10 is to acquire and integrate data to form a full lifecycle dataset. This dataset includes, but is not limited to: design data of power equipment, such as material specifications, rated voltage, insulation class, and design drawings; operational data, such as time-series monitoring data of current, voltage, temperature, vibration, dissolved gases in oil, and partial discharge signals; maintenance data, such as historical maintenance records, component replacement dates, and inspection reports; and external environmental data, such as ambient temperature, humidity, altitude, and air pressure at the equipment operating location.

[0074] After acquiring the full lifecycle dataset, the causal knowledge graph construction unit 10 constructs the graph using a combination of data-driven and knowledge-driven approaches.

[0075] In the data-driven approach, the causal knowledge graph construction unit 10 employs a constraint-based causal discovery algorithm. The core of this type of algorithm lies in performing a series of conditional independence tests on the operational data within the full lifecycle dataset. Conditional independence tests are used to determine whether a direct statistical dependency exists between two variables given one or more other variables. For example, testing variables... and variables In a given set of variables Independence under certain conditions can be expressed as determining whether the following equation holds:

[0076] ;

[0077] In the formula, Represents probability; Indicates in variable and variable set When variables occur simultaneously The conditional probability of occurrence; Indicates in the set of variables Variables at occurrence The conditional probability of occurrence. If the equation holds, then the variable is determined. and In a given Under the condition of conditional independence, by systematically performing such tests on each variable pair in the whole life cycle dataset, the causal knowledge graph construction unit 10 can learn and determine the causal structure between data variables, forming a preliminary causal relationship network.

[0078] In one specific embodiment, the constraint-based causal discovery algorithm can employ the PC algorithm. The execution flow of this algorithm is as follows: first, a complete undirected graph is constructed, which contains all data variables as nodes, i.e. it is assumed that there is an edge between any two variables. Second, a zero-order conditional independence test is performed, i.e. for any two variables in the graph , an unconditional independence test is performed, and if , the edge between them is removed. Then, a high-order conditional independence test is iteratively performed, e.g. in the first-order test, for the variable pair , all its adjacent nodes are traversed, and a conditional independence test is performed with as the condition, and if and are conditionally independent given , the edge between them is removed. This iteration process continues until all order conditional independence tests are completed. Finally, the causal directions on the remaining edges are determined using known causal rules (e.g. temporal precedence) and V-structure (Collider) identification rules, and a preliminary causal relationship network is output. For the conditional independence test in the test process, when dealing with discrete data, G-test or Chi-square test can be used; when dealing with continuous data, a test method based on partial correlation can be used.

[0079] In the knowledge-driven path, the causal knowledge graph construction unit 10 formalizes the verified expert knowledge and equipment failure analysis patterns in the field. These knowledge comes from equipment operation manuals, failure mode and effects analysis (FMEA) reports, and the diagnosis experience of senior operation and maintenance engineers. The formalization process converts unstructured natural language descriptions, such as "damp insulation causes the dielectric loss tangent value to increase", into structured causal rules. This conversion process can be represented in the form of triples (damp insulation, causes, dielectric loss tangent value increases) or IF-THEN rules.

[0080] Finally, the causal knowledge graph construction unit 10 fuses the preliminary causal structure generated by the data-driven path with the formalized causal rules generated by the knowledge-driven path. In this process, the formalized causal rules are injected as prior knowledge to verify, correct, or complete the causal relationships learned by data-driven learning, thereby constructing the final causal knowledge graph containing the causal relationships between equipment entities, environmental factors, and management activities. This knowledge graph is stored in the form of a graph structure, where nodes represent equipment entities, state variables, or events, and edges represent the causal relationships between them.

[0081] Referring to the accompanying drawings Figure 3 , Figure 3Fig. 1 is a schematic diagram showing the principle of counterfactual reasoning by the strategy rule learning unit 20 according to an embodiment of the present application. The strategy rule learning unit 20 is responsible for enabling the system to evolve and optimize its decision logic by deep analysis of historical events. The input to the unit 20 is the causal knowledge graph constructed and dynamically maintained by the causal knowledge graph construction unit 10.

[0082] The working process of the strategy rule learning unit 20 adopts a potential outcomes framework. First, the unit 20 selects a completed maintenance event from the historical maintenance data for analysis. The historical event contains the actual decision taken under specific equipment state and environmental conditions, as well as the actual outcome resulting from the decision.

[0083] Subsequently, the strategy rule learning unit 20 performs counterfactual reasoning based on the causal knowledge graph. Specifically, the strategy rule learning unit 20 replaces the actual decision taken in the historical event with one or more alternative decisions while keeping all the preconditions other than the decision variable unchanged. Using the causal relationships defined in the causal knowledge graph, the system simulates and calculates how the equipment state and related economic indicators would evolve if these alternative decisions were taken, thereby obtaining a series of possible counterfactual outcomes. The essence of this reasoning process is to perform intervention calculus on the established causal model to predict the consequences triggered by different decision behaviors.

[0084] After obtaining the actual outcome and the counterfactual outcome, the strategy rule learning unit 20 needs to quantitatively compare the pros and cons of the two outcomes, which is achieved by calculating the counterfactual utility gain. First, define a utility function that maps a specific event outcome to a scalar value to represent the overall benefit of the outcome. Based on this utility function, the following counterfactual utility gain calculation model can be used:

[0085] ;

[0086] In the formula: is the counterfactual utility gain, representing the amount of benefit change that can be brought about by taking alternative decisions relative to the actual decision;

[0087] is the utility function that combines various indicators of an event outcome into a single utility value. In a specific embodiment, this function can be defined as the weighted sum of multiple key performance indicators, and its specific form is:

[0088] ;

[0089] wherein, represents a specific event outcome; is the outcome corresponding total maintenance cost; for result corresponding equipment available duration; for result corresponding equipment remaining life; are preset weight coefficients for total maintenance cost, equipment available duration and equipment remaining life, respectively.

[0090] is counterfactual result, which is the synthetic result of simulating the alternative decision taken in the historical event;

[0091] is actual result, which is the synthetic result actually occurred and recorded in the historical event.

[0092] When the calculated counterfactual utility gain is greater than a preset threshold, it indicates that the analyzed alternative decision has higher application value in similar scenarios. Based on this judgment, the strategy rule learning unit 20 refines and generates a new strategy rule. The form of the rule can be: IF [specific equipment state and environmental conditions] THEN [execute alternative decision]. Finally, the new strategy rule is injected back into the causal knowledge graph construction unit 10 for updating or enhancing the relationship between management activities and results in the causal knowledge graph.

[0093] Referring to the accompanying drawings Figure 4 Figure 4 is a schematic diagram of the principle of value calculation by the decision option value quantification unit 30 according to an embodiment of the present application. The function of the decision option value quantification unit 30 is to quantify the flexibility and uncertainty in management decisions, thereby providing an economic value scale for decision-making beyond traditional cost-benefit analysis.

[0094] The decision option value quantification unit 30 first identifies potential maintenance decision options based on the dynamically updated causal knowledge graph. This identification process is achieved by querying nodes representing equipment health status in the causal knowledge graph and filtering out those nodes that have shown a deterioration trend but have not yet reached the preset intervention threshold. For each filtered state node, the system matches one or more maintenance activities that can prevent or delay its deterioration path. The combination of each potential problem to be solved and the corresponding maintenance activity constitutes a potential maintenance decision option.

[0095] ​​​After identifying the potential maintenance decision options, the decision option value quantification unit 30 models each option as a real option, specifically, as an European call option. The logical basis of this modeling is that holding a maintenance decision option is equivalent to having the right, but not the obligation, to acquire an asset (avoiding future larger losses due to equipment failure) at a certain cost (the cost of performing the maintenance) within a certain time window in the future.

[0096] In order to calculate the value of this real option, several core input parameters of the model need to be determined. The calculation of these parameters is based on the dynamically updated causal knowledge graph as data and logic.

[0097] First, by performing a forward reasoning on the causal knowledge graph, the expected present value of the possible failure modes, failure probabilities and the resulting economic losses (including repair costs, downtime losses, penalty fines, etc.) of the equipment in the future without executing the maintenance decision option is simulated, which is defined as the underlying asset present value of the real option model .

[0098] Second, the direct costs required to execute the maintenance decision option, including labor, spare parts and planned downtime costs, are defined as the execution price of the real option model .

[0099] Third, according to the predicted deterioration rate of the equipment health state from the causal knowledge graph, the time span between the current time and the critical point at which action must be taken is determined, which is defined as the expiration time of the option .

[0100] At the same time, an externally input risk-free interest rate is introduced to represent the time value of capital.

[0101] Finally, and most importantly, the volatility of the underlying asset value , which quantifies the uncertainty faced by the decision, is calculated, which combines at least two sources:

[0102] First, the uncertainty of the deterioration rate of the equipment health state is derived by analyzing historical operation data and the physical model of the equipment;

[0103] Second, the uncertainty of the future economic environment is derived by analyzing external environmental data (such as electricity market prices, load forecasts, etc.).

[0104] After all the above parameters are determined, the decision option value quantification unit 30 uses the following real option value calculation model to calculate the real option value of each potential maintenance decision option:

[0105] ;

[0106] where:

[0107] ;

[0108] ;

[0109] where: is the real option value of the potential maintenance decision option; is the present value of the underlying asset, i.e. the expected present value of the avoidable future loss; is the exercise price, i.e. the cost of exercising the maintenance decision; is the time to expiration, i.e. the effective time window for exercising the decision; is the risk-free interest rate; is the volatility of the underlying asset value, quantifying the uncertainty of the decision environment; is the cumulative distribution function of the standard normal distribution; is the natural logarithm function; and are intermediate variables in the model calculation process.

[0110] Through the above steps, the decision option value quantification unit 30 assigns a quantified economic value to each potential maintenance decision option, taking into account future uncertainty and outputs these options along with their values to the portfolio decision optimization unit 40.

[0111] The portfolio decision optimization unit 40, whose function is to systematically integrate the multiple independent potential maintenance decision options and their corresponding real option values output by the decision option value quantification unit 30, in order to maximize the overall value under the condition of limited resources. This unit 40 receives the set of potential maintenance decision options and the real option value and exercise cost of each option.

[0112] First, the portfolio decision optimization unit 40 constructs all the received potential maintenance decision options into a decision portfolio. In this model, each potential maintenance decision option is considered as a project to be invested. A binary decision variable is introduced for each project to indicate whether the project is selected for execution.

[0113] Next, the portfolio decision optimization unit 40 defines the resources required to execute the decisions in the portfolio and formalizes them as a series of resource constraints. These constraints reflect the limitations faced in actual operations. In a specific embodiment, the defined resource constraints include:

[0114] budget constraint, i.e. the total cost of all selected decisions cannot exceed the preset total budget;

[0115] Spare parts constraint means that the total quantity of each specific type of spare parts required for all selected decisions cannot exceed the current inventory of that type of spare parts.

[0116] Human resource constraints mean that the total man-hours of specialized human resources required for each specific skill for all selected decisions cannot exceed the total available man-hours for that skill.

[0117] After completing portfolio modeling and constraint definition, the portfolio decision optimization unit 40 constructs and solves a portfolio decision optimization objective function. The goal of this optimization is to maximize the total real option value of the selected decision portfolio while satisfying all defined resource constraints. This portfolio optimization problem can be formalized into the following mathematical model:

[0118] Objective function:

[0119] ;

[0120] Constraints:

[0121] :

[0122] ;

[0123] ;

[0124] ;

[0125] ;

[0126] In the formula: For binary decision variables, when the first When one potential maintenance decision option is selected ,otherwise ; For the first The real option value of each potential maintenance decision option is calculated by the decision option value quantification unit 30; The total number of potential maintenance decision options; To execute the first The cost of each potential maintenance decision option; This is the preset total budget limit; To execute the first The first decision option requires the consumption of the first... The quantity of each type of spare part; For the first Total available inventory of each type of spare parts; This represents the total quantity of spare parts of each type. the first skill type of manpower consumed to execute the first decision option; the total man-hours available for the first skill type of manpower; the total number of skill types of manpower.

[0127] The combined decision optimization unit 40 solves the above model using an integer programming solver or other algorithms to obtain a set of optimal values of the decision variables . All the corresponding set of potential maintenance decision options, i.e. the final optimal decision combination with the maximum value under the current resource conditions. This optimal decision combination is then output to the maintenance decision voucher management unit 50.

[0128] In one specific embodiment, the solver used to solve the integer programming problem can employ a Branch-and-Bound based algorithm. This method finds the global optimal solution by systematically decomposing the original problem into a series of sub-problems and using the bounds of the solutions of the sub-problems to prune the search space that does not contain the optimal solution. In engineering implementation, a commercial optimization solver (such as Gurobi, CPLEX) or an open-source optimization solver (such as SCIP, GLPK) can be called to perform the solving process.

[0129] Referring to the attached Figure 5 , Figure 5 is a data structure diagram of the standardized maintenance decision voucher according to one embodiment of the present application. The maintenance decision voucher management unit 50 is responsible for converting the abstract and optimized decision results into a specific and standardized data voucher for subsequent management, execution and circulation.

[0130] The maintenance decision voucher management unit 50 receives the optimal decision combination output by the combined decision optimization unit 40. For each selected maintenance decision option (i.e. the option of the decision variable ) in the combination, the maintenance decision voucher management unit 50 generates a standardized maintenance decision voucher for it. The voucher is a structured data object, referring to the attached Figure 5 , its data structure specifically includes the following three modules:

[0131] The obligation module is used to clearly define the maintenance tasks to be performed. The data fields it contains include: task unique identifier, target device code, specific maintenance task content description generated according to the diagnosis results of the causal knowledge graph, and technical quality standards and acceptance procedures to be followed to perform the task.

[0132] ​​​The opportunity value module is used to record the economic value information associated with the maintenance task. The data fields it contains include the real option value associated with this maintenance decision calculated by the decision option value quantification unit 30 , and the core parameters used to calculate the value, such as the present value of the underlying asset , the execution price , the expiration time , and the volatility .

[0133] The provenance hash module is used to ensure the traceability and tamper resistance of the decision-making process. When the credential is generated, the system records an analysis snapshot of the causal knowledge graph used to generate this decision, which contains all relevant node states, causal relationships and parameters at that time. Subsequently, the system uses a preset hash algorithm (such as SHA-256) to operate on the complete data of the analysis snapshot, generating a unique hash value. The hash value is stored in the provenance hash module, providing an unalterable verification path pointing to the original analysis basis.

[0134] In an optional embodiment, in order to further enhance the tamper resistance, transparency and transferability of the credential, the standardized maintenance decision credential can be minted as a non-fungible token (NFT) and recorded on a blockchain (such as a consortium chain or a private chain). At this time, the data structure of the credential is stored as the metadata of the NFT, and the hash value in the provenance hash module can point to the detailed analysis snapshot stored in the distributed file system (such as IPFS), thereby realizing the on-chain management of the credential asset.

[0135] In an optional embodiment, the function of the maintenance decision credential management unit 50 can be further extended to build and manage a two-layer transaction platform. The platform takes the generated maintenance decision credential as the transaction object.

[0136] The internal market layer of the platform is aimed at different departments within the enterprise. In this layer, departments can view the generated maintenance decision credentials and their opportunity values. A department can exchange resources with another department based on the value of the credential through internal processes, such as using a high-value maintenance decision credential to apply for additional budget or transferring task execution rights to a department with more relevant resources.

[0137] The platform's external marketplace layer targets pre-certified third-party service providers outside the enterprise. Enterprises can list specific maintenance decision credentials here. Third-party service providers can then quote prices and bid based on the task content defined in the obligation module and the economic importance demonstrated in the opportunity value module within the credentials, thereby securing contracts to execute tasks. This process utilizes standardized credentials to achieve market-based procurement of maintenance services.

[0138] In an optional embodiment, refer to the appendix Figure 1 The system of the present invention may further include a fault diagnosis unit 60. The input of the fault diagnosis unit 60 is connected to the output of the causal knowledge graph construction unit 10. The function of this unit is to perform reasoning based on the causal knowledge graph when an abnormal state of the device is detected, so as to calculate and output the probability ranking of the root causes of the abnormality.

[0139] When a detected abnormal state node is used as input, the fault diagnosis unit 60 first traces backward along the defined causal relationships from the abnormal state node in the causal knowledge graph to determine the set of all potential root cause nodes that may have caused this abnormality.

[0140] For each potential root cause node in the set, the fault diagnosis unit 60 uses Bayesian inference to calculate its posterior probability as a true cause. This process utilizes prior knowledge and conditional probability information stored in the causal knowledge graph. Specifically, the following Bayesian inference model can be used to calculate the posterior probability of a potential root cause node as a true cause:

[0141] ;

[0142] In the formula: Let be the posterior probability, representing the probability of observing an abnormal state. Under the conditions, the first One potential root cause The probability that it is the real cause; These are abnormal state nodes that have been detected, such as excessive concentration of dissolved acetylene gas in oil; For the first Potential root cause nodes, such as local overheating or high-energy discharge; Let be the conditional probability or likelihood, representing the probability if the root cause is... An abnormal state was observed. The probability. This value is based on the causal knowledge graph from... arrive The strength of causal relationships along the path is determined; Let be the prior probability, representing the root cause in the absence of any observational information. The probability of occurrence itself. This value can be set based on historical failure statistics of the equipment or expert experience and stored in the knowledge graph; The marginal probability of evidence is the total probability of observing the abnormal state as a normalization factor. The calculation method is to sum the probability of each cause causing the anomaly for all potential root causes; The total number of potential root cause nodes determined in this diagnosis.

[0143] The fault diagnosis unit 60 repeatedly performs the above calculation for each candidate root cause node, thereby obtaining the posterior probability value of each candidate cause. Finally, the fault diagnosis unit 60 ranks all candidate root cause nodes in descending order according to the posterior probability values calculated by it, and outputs the probability ranking list. This list provides a data-supported diagnostic direction for the root cause of the fault for subsequent maintenance decisions.

[0144] In summary, the embodiment of the application provides a power equipment life cycle management system and method based on digital twin technology. Through the cooperative work of each unit, the whole process transformation and closed-loop management from multi-source heterogeneous data to standardized value certificates are realized.

[0145] The operation of the system starts from the causal knowledge graph construction unit 10, which integrates the design, operation, maintenance and external environment data of the equipment, and constructs a knowledge graph describing the internal causal logic between equipment entities, environmental factors and management activities through the combination of data-driven causal discovery and knowledge-driven rule injection. This graph provides a foundation for all subsequent advanced analysis.

[0146] On this basis, the strategy rule learning unit 20 takes historical maintenance events as samples, uses the latent outcome framework and counterfactual reasoning to quantitatively evaluate the utility gain of alternative decisions relative to actual decisions, thereby refining new and better strategy rules and feeding them back to the causal knowledge graph, so that the decision logic of the system has the ability of self-evolution and continuous optimization.

[0147] Subsequently, the decision option value quantification unit 30 uses the dynamically updated causal knowledge graph to identify potential maintenance decision options in the future, and by introducing a real option value calculation model, quantifies the future uncertainty (such as the rate of deterioration of equipment health, market price fluctuations) contained in each decision option into the option value of the decision. This process upgrades the traditional cost analysis to a dynamic value evaluation oriented to the future.

[0148] Then, the portfolio decision optimization unit 40 constructs all the potential decision options that have been quantified in value into a decision portfolio, and takes the budget, spare parts, manpower, etc. as resource constraints, and outputs an optimal decision portfolio that maximizes the total real option value by solving the portfolio decision optimization objective function under the premise of meeting all constraints, realizing the transition from single decision to global resource optimal allocation.

[0149] Finally, the maintenance decision voucher management unit 50 solidifies each selected item in the optimal decision portfolio into a standardized maintenance decision voucher. Through its obligation module, opportunity value module and traceability hash module, the voucher converts an abstract decision into a specific, value-measurable and process-traceable digital asset, providing a solid technical carrier for subsequent fine management, task allocation and even market transactions.

[0150] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A power equipment lifecycle management system based on digital twin technology, characterized in that, The system includes: The causal knowledge graph construction unit is used to acquire design data, operation data, maintenance data and external environment data of power equipment to form a full life cycle dataset, and to construct a causal knowledge graph containing causal relationships between equipment entities, environmental factors and management activities based on the full life cycle dataset. The strategy rule learning unit is used to perform counterfactual deduction on historical maintenance events based on the causal knowledge graph to determine the utility gain of alternative decisions, generate new strategy rules, and inject the new strategy rules into the causal knowledge graph for dynamic updates. The decision option value quantification unit is used to identify potential maintenance decision options based on a dynamically updated causal knowledge graph, and to calculate the real option value of each potential maintenance decision option using a real option model. The combined decision optimization unit is used to receive potential maintenance decision options and the real option value, and perform combined optimization under preset resource constraints to generate the optimal decision combination that maximizes the total real option value. The maintenance decision voucher management unit is used to generate standardized maintenance decision vouchers containing task obligations, opportunity value and traceability information for the selected maintenance decision options based on the optimal decision combination. The decision option value quantification unit is specifically used for: Each potential maintenance decision option is modeled as a real option; The value of the real option is determined by using the predicted present value of avoidable loss as the present value of the underlying asset, the maintenance and execution cost as the execution price, and combining the effective decision-making time window, the risk-free interest rate, and the quantitative uncertainty into a preset real option pricing model. Among them, the expected present value of future economic returns and the quantitative uncertainty are both calculated based on the dynamically updated causal knowledge graph. The calculation of the quantified uncertainty is specifically derived by taking a dynamically updated causal knowledge graph as input and comprehensively quantifying at least two of the following sources of uncertainty: The uncertainty of the rate of deterioration of equipment health status, determined by the equipment's own physical characteristics and historical operating data; Uncertainty in the external economic environment, determined by electricity market prices and future load forecasts.

2. The power equipment lifecycle management system based on digital twin technology according to claim 1, characterized in that, The causal knowledge graph construction unit is specifically used for: A constraint-based causal discovery algorithm is adopted to learn and determine the preliminary causal structure between data variables by performing a series of conditional independence tests on the operational data in the full life cycle dataset. The validated equipment failure analysis model and domain expert knowledge are transformed into formal causal rules, and the preliminary causal structure is verified, corrected or completed using the formal causal rules.

3. The power equipment lifecycle management system based on digital twin technology according to claim 1, characterized in that, The policy rule learning unit is specifically used for: Using a potential outcome framework, we simulate and calculate the actual outcomes of the decisions actually made in the historical maintenance events, as well as the counterfactual outcomes that might occur if alternative decisions were made. The utility gain is determined by calculating the difference between the utility of the counterfactual outcome and the utility of the actual outcome, and new policy rules are extracted and generated based on the utility gain.

4. The power equipment lifecycle management system based on digital twin technology according to claim 1, characterized in that, The combined decision optimization unit is specifically used for: Construct a decision portfolio from multiple potential maintenance decision options; Budget, spare parts inventory, and professional human resources are defined as the resource constraints of the decision portfolio; An integer programming algorithm is used to find the optimal decision portfolio that maximizes the total real option value of the decision portfolio under the condition of satisfying resource constraints.

5. The power equipment lifecycle management system based on digital twin technology according to claim 1, characterized in that, The maintenance decision credentials include: The obligation module is used to describe the specific content and technical quality standards of the maintenance task based on the diagnostic results of the causal knowledge graph. The opportunity value module is used to record the real option value associated with the maintenance task, calculated by the decision option value quantification unit. The traceability hash module stores a hash value calculated based on the analysis snapshot used to generate maintenance decision credentials, thereby establishing an immutable link between the maintenance decision credentials and the analysis snapshot.

6. The power equipment lifecycle management system based on digital twin technology according to claim 1, characterized in that, The maintenance decision credential management unit is also used for: Based on the maintenance decision credentials, a two-tier trading platform consisting of an internal market layer and an external market layer is constructed and managed. The internal market layer provides interfaces for querying voucher value and transferring internal resources to support resource exchange between different departments within the enterprise based on the value of maintenance decision vouchers. The external market layer provides credential listing, bidding, and contract generation functions to support the market-based procurement of maintenance tasks among pre-certified third-party service providers.

7. The power equipment lifecycle management system based on digital twin technology according to claim 1, characterized in that, The system also includes: The fault diagnosis unit is used to perform Bayesian backward probabilistic reasoning along the causal relationship from the detected abnormal state nodes in the causal knowledge graph, so as to calculate and output the probability ranking of the root cause nodes that lead to the abnormal state.

8. A method for full life-cycle management of power equipment based on digital twin technology, applied to the system described in any one of claims 1-7, characterized in that, The method includes the following steps: S1. Obtain design data, operation data, maintenance data and external environment data of power equipment to form a full life cycle dataset, and construct a causal knowledge graph containing causal relationships between equipment entities, environmental factors and management activities based on the full life cycle dataset; S2. Based on the causal knowledge graph, counterfactual inference is performed on historical maintenance events to determine the utility gain of alternative decisions, new policy rules are generated, and the new policy rules are injected into the causal knowledge graph for dynamic updates. S3. Identify potential maintenance decision options based on the dynamically updated causal knowledge graph, and calculate the real option value of each potential maintenance decision option using a real option model; S4. Receive the potential maintenance decision options and real option values, and perform combination optimization under preset resource constraints to generate the optimal decision combination that maximizes the total real option value. S5. Based on the optimal decision combination, generate standardized maintenance decision credentials for the selected maintenance decision options, including task obligations, opportunity value, and traceability information.

Images