Multi-objective optimization method of standard system optimization coping strategy based on artificial intelligence

Abstract

The invention discloses a multi-objective optimization method for a standard system optimization coping strategy based on artificial intelligence, and relates to the technical field of artificial intelligence, and the method comprises the steps: collecting multi-source data of a standard system, and fusing the multi-source data into a structured data set; based on the structured data set, constructing a causal influence network of a standard system through a causal discovery algorithm; establishing a mathematical model of a standard system multi-objective optimization problem through a causal influence network, and generating a personalized optimization strategy set through anti-factual reasoning in combination with personalized feature data of a decision maker; and based on the personalized optimization strategy set, performing multi-scene simulation verification on each optimization strategy through a standard system simulation unit, and dynamically adjusting the optimization strategy according to a simulation verification result. According to the method, the problems that a traditional method depends on statistical correlation to cause strategy failure and an optimization result is disjointed with the demand of a decision maker are solved.

Description

A Multi-Objective Optimization Method for Optimizing Response Strategies Based on Artificial Intelligence Standard System Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a multi-objective optimization method for optimizing response strategies based on artificial intelligence standard systems. Background Technology

[0002] In fields such as intelligent manufacturing and the industrial internet, the scientific management of standard systems is crucial for ensuring product quality and improving operational efficiency. Existing technologies have developed multi-objective optimization methods based on artificial intelligence for standard system optimization. By collecting standard execution data, environmental parameters, and performance indicators, machine learning models are used to establish statistical correlation models between standard parameters and multiple performance objectives. Evolutionary algorithms and other optimization methods are then used to solve these models, generating a set of Pareto optimal solutions. This provides decision-makers with alternative solutions for multi-objective trade-offs, and to a certain extent, achieves automated generation from data to optimization strategies.

[0003] Existing technologies still face the dual challenges of ensuring the scientific validity of strategies and adapting them to individual needs in practical applications. The optimization process mainly relies on statistical correlation rather than causal mechanisms. There are complex causal networks among the elements in the standard system. Modeling based solely on correlation may lead to optimization strategies being built on pseudo-correlation, causing the strategies to fail to achieve the expected results after actual deployment, or even producing unexpected negative effects. Existing methods lack a deep integration with the decision-maker's individual needs after generating the optimization solution set. The massive number of technical solutions makes it difficult for decision-makers to choose and they cannot understand the degree of matching between each solution and their own value preferences, resulting in a disconnect between the decision-making process and the optimization process. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a multi-objective optimization method based on artificial intelligence to optimize response strategies according to a standard system, addressing the problem of strategy failure and decision-making disconnect caused by neglecting causal relationships and individualized factors.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: This invention provides a multi-objective optimization method for optimizing response strategies in a standard system based on artificial intelligence. The method includes: collecting multi-source data from the standard system and fusing the multi-source data into a structured dataset; constructing a causal influence network for the standard system using a causal discovery algorithm based on the structured dataset; establishing a mathematical model of the multi-objective optimization problem of the standard system using the causal influence network, and generating a personalized set of optimization strategies through counterfactual reasoning, combined with the decision-maker's personalized characteristic data; performing multi-scenario simulation verification on each optimization strategy using a standard system simulation unit, and dynamically adjusting the optimization strategies based on the simulation verification results; deploying the verified and adjusted optimization strategies into the actual standard system for execution, and collecting strategy execution process data in real time through a monitoring unit; and evaluating the implementation effect of the optimization strategies from multiple dimensions based on the strategy execution process data, and obtaining a feedback-driven adaptive optimization mechanism by updating the causal influence network, the mathematical model, and the counterfactual reasoning process using the evaluation results.

[0007] As a preferred embodiment of the multi-objective optimization method for standard system optimization strategies based on artificial intelligence as described in this invention, the method involves: collecting multi-source data of the standard system and fusing the multi-source data into a structured dataset, including the following steps: collecting multi-source data of the standard system from a standard document library, a historical database, and monitoring equipment; performing data cleaning operations on the multi-source data of the standard system to remove missing and outlier values; performing format standardization and alignment operations on the multi-source data of the standard system after data cleaning to unify timestamps and measurement units; and storing the multi-source data of the standard system after format standardization and alignment operations according to a predefined pattern to generate a structured dataset.

[0008] As a preferred embodiment of the multi-objective optimization method for optimizing the standard system based on artificial intelligence as described in this invention, the method comprises the following steps: constructing a causal influence network of the standard system based on the structured dataset using a causal discovery algorithm; extracting observational data sequences corresponding to standard variables, environmental variables, and performance variables from the structured dataset; inputting the observational data sequences corresponding to standard variables, environmental variables, and performance variables into the causal discovery algorithm, using the PC algorithm to calculate the conditional independence between variables, and obtaining the conditional independence test results between variables; constructing a sketch of the causal influence network of the standard system representing the preliminary causal relationship between variables based on the conditional independence test results between variables; performing directional inference and structural optimization on the sketch of the causal influence network of the standard system, determining the direction of causal edges and removing false edges; and performing parameter learning on the causal influence network of the standard system after directional inference and structural optimization, estimating the effect strength of each causal edge, and obtaining a parameterized causal influence network of the standard system.

[0009] As a preferred embodiment of the multi-objective optimization method for standard system optimization strategies based on artificial intelligence as described in this invention, the method involves: establishing a mathematical model of the multi-objective optimization problem of the standard system through a causal influence network; combining the decision-maker's personalized characteristic data; and generating a personalized set of optimization strategies through counterfactual reasoning. This includes the following steps: extracting the decision-maker's implicit preferences and risk tolerance for objective weights from their historical interaction records to constitute personalized characteristic data; defining a personalized utility function reflecting the decision-maker's preferences; integrating the personalized utility function with multiple objective functions in the mathematical model; and maximizing the personalized utility that integrates the decision-maker's personalized characteristics within the decision variable space and constraint set defined by the mathematical model. Guided by a function, candidate optimization strategies are generated through a search. For each candidate optimization strategy, a causal intervention scenario for implementing the strategy and a causal control scenario for maintaining the baseline state are constructed. Based on the objective function derived from the causal influence network in the mathematical model, the evolution trajectory of each performance objective is deduced under the causal intervention scenario and the causal control scenario for maintaining the baseline state. By comparing the differences in the terminal values ​​of the evolution trajectories, the causal effect of the candidate optimization strategy is quantified. Based on the causal effect of the candidate optimization strategy, combined with the preference structure and risk propensity contained in the decision-maker's personalized characteristic data, the candidate optimization strategies are ranked in multiple dimensions and screened for contextual adaptability, generating a personalized set of optimization strategies that deeply matches the decision-maker's individual decision-making logic.

[0010] As a preferred embodiment of the multi-objective optimization method for optimizing response strategies based on artificial intelligence standard systems according to the present invention, the method includes: for the personalized optimization strategy set, multi-scenario simulation verification of each optimization strategy is performed through a standard system simulation unit, including the following steps: selecting the current optimization strategy to be verified from the personalized optimization strategy set and outputting it to the standard system simulation unit; constructing a benchmark operating scenario based on the current environmental state in the standard system simulation unit; deriving verification scenarios with different disturbance conditions based on predefined rules on the basis of the benchmark operating scenario; loading the optimization strategy into each verification scenario sequentially in the standard system simulation unit, driving the simulation execution, and recording and outputting the simulation execution process data and terminal performance indicators of the optimization strategy in each verification scenario.

[0011] As a preferred embodiment of the multi-objective optimization method for standard system optimization response strategy based on artificial intelligence as described in this invention, the following steps are included: dynamically adjusting the optimization strategy based on simulation verification results: identifying performance-vulnerable scenarios and parameter-sensitive intervals of the optimization strategy based on simulation execution process data and terminal performance indicators of the optimization strategy in each verification scenario, and formulating corresponding strategy parameter compensation rules to generate a strategy parameter adjustment scheme; modifying the adjustable parameters in the optimization strategy according to the strategy parameter adjustment scheme to generate a modified optimization strategy; inputting the modified optimization strategy into the standard system simulation unit and re-simulating in the performance-vulnerable scenario to obtain the simulation verification results of the modified optimization strategy; confirming that the performance of the optimization strategy meets the preset performance threshold based on the simulation verification results of the modified optimization strategy, and using the confirmed optimization strategy as the verified and adjusted optimization strategy.

[0012] As a preferred embodiment of the multi-objective optimization method for optimizing the standard system based on artificial intelligence as described in this invention, the following steps are included in deploying the verified and adjusted optimization strategy to the actual standard system for execution: converting the verified and adjusted optimization strategy into executable instructions or configuration changes for the actual standard system; and deploying the converted verified and adjusted optimization strategy in stages within the controlled environment of the actual standard system.

[0013] As a preferred embodiment of the multi-objective optimization method for optimizing response strategies based on the standard system of artificial intelligence described in this invention, the method includes: real-time collection of strategy execution process data through a monitoring unit, comprising the following steps: real-time collection of status data and performance data during the execution process of the optimized strategy after verification and adjustment; time-stamp alignment and formatting of the status data and performance data to generate preliminary strategy execution process data; and real-time association and labeling of the preliminary strategy execution process data with predefined expected strategy effect indicators to obtain a complete structure.

[0014] As a preferred embodiment of the multi-objective optimization method for optimizing response strategies based on the standard system of artificial intelligence described in this invention, the following steps are included: Based on strategy execution process data, a multi-dimensional evaluation of the implementation effect of the optimization strategy is conducted, comprising the following steps: extracting a sequence of actual performance indicators directly related to each performance objective of the optimization strategy from the strategy execution process data; comparing the sequence of actual performance indicators directly related to each performance objective of the optimization strategy with historical benchmark performance data before the implementation of the optimization strategy, quantitatively describing the absolute and relative changes in performance of each dimension after the implementation of the optimization strategy; assessing the performance volatility and stability of the optimization strategy during actual execution by combining the risk tolerance in the decision-maker's personalized characteristic data; analyzing abnormal events and adjustment operations recorded in the strategy execution process data to assess the adaptability and maintenance cost of the optimization strategy to environmental disturbances; and generating a multi-dimensional quantitative evaluation report on the implementation effect of the optimization strategy by comprehensively considering the absolute change, relative change, performance volatility and stability, adaptability, and maintenance cost.

[0015] As a preferred embodiment of the multi-objective optimization method for optimizing response strategies based on artificial intelligence standard systems as described in this invention, the following steps are included: using the updated causal influence network, the mathematical model, and the counterfactual reasoning process based on the evaluation results to obtain a feedback-driven adaptive optimization mechanism: using a multi-dimensional quantitative evaluation report on the implementation effect of the optimization strategy to identify causal edges in the causal influence network that do not match actual observations; based on the identified causal relationships to be corrected, updating the effect strength parameters of the corresponding causal edges in the causal influence network, and adding or deleting causal edges to obtain an updated causal influence network; integrating the updated causal influence network into the mathematical model, updating the objective function and constraint set in the mathematical model to obtain an updated mathematical model; inputting the performance stability and adaptability data from the multi-dimensional quantitative evaluation report on the implementation effect of the optimization strategy into the counterfactual reasoning process to calibrate the scenario confidence and effect estimation parameters in the counterfactual reasoning; and using the updated causal influence network, the updated mathematical model, and the calibrated counterfactual reasoning process to obtain a feedback-driven adaptive optimization mechanism.

[0016] The beneficial effects of this invention are as follows: It collects and integrates multi-source data to form a structured dataset; constructs a causal influence network revealing the causal relationships between elements within a standard system through a causal discovery algorithm; establishes a mathematical model for a multi-objective optimization problem based on the network; integrates personalized characteristic data of decision-makers; generates a highly adaptable set of personalized optimization strategies through counterfactual reasoning; verifies and dynamically adjusts the strategies in multiple scenarios using a simulation unit; then deploys and executes the optimization strategies and monitors them in real time; and updates the causal influence network, mathematical model, and reasoning process based on the evaluation results of the execution effect, forming a self-evolving adaptive optimization closed-loop mechanism. By combining causal inference with personalized counterfactual reasoning, it solves the problems of strategy failure due to reliance on statistical correlation in traditional methods, and the disconnect between optimization results and decision-makers' needs. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 is a flowchart of a multi-objective optimization method for optimizing response strategies based on an artificial intelligence-based standard system.

[0019] Figure 2 is a flowchart of the method for dynamically adjusting the optimization strategy based on the simulation verification results. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "example" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the invention. The appearance of an embodiment in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments.

[0023] Referring to Figures 1 and 2, an embodiment of the present invention is provided. This embodiment provides a multi-objective optimization method for optimizing response strategies based on artificial intelligence standard systems, including the following steps: S1, collecting multi-source data of the standard system and fusing the multi-source data into a structured dataset.

[0024] S1.1 Collect multi-source data of the standard system from standard document libraries, historical databases and monitoring equipment.

[0025] Furthermore, a multi-dimensional, multi-modal data acquisition framework covering the entire lifecycle and all elements of the standard system has been constructed. Traditional data acquisition typically focuses on a single type of data source, such as only paying attention to real-time streaming data from monitoring equipment or transaction records in historical databases, leading to a one-sided perspective in data analysis. This framework collaboratively acquires structured text containing rules and knowledge from the standard document library, relational data recording processes and results from the historical database, and time-series signals reflecting real-time status and environment from monitoring equipment. For example, when evaluating the optimization effect of a production process standard, it is necessary to simultaneously acquire the technical parameter specifications of the process from the standard document library, the execution records and quality inspection results of past batches from the historical database, and the real-time readings of temperature, pressure, and flow rate from the monitoring equipment on the current production line. This multi-source parallel acquisition strategy is crucial.

[0026] Specifically, the normative requirements, historical performance, and real-time status of the standard system are synchronously acquired and aligned. By configuring connectors or calling application programming interfaces, standard clauses, version information, and associated rules are extracted from the standard document library. Past standard execution records, audit logs, and change history are extracted from the historical database. Real-time operating status, performance indicators, and environmental parameter data are continuously acquired from monitoring devices deployed in the physical environment or information system, thus completing the comprehensive collection of multi-source data of the standard system.

[0027] S1.2 Perform data cleaning operations on the multi-source data of the standard system to remove missing and outlier values. Perform format standardization and alignment operations on the multi-source data of the standard system after data cleaning to unify timestamps and measurement units.

[0028] Furthermore, for the specific needs of causal discovery and multi-objective optimization applications, general data cleaning often only focuses on data integrity. The quality requirements of data used to construct causal influence networks and perform precise optimization calculations far exceed those of general data analysis. Removing missing and outlier values ​​is not only for data integrity, but also to prevent the introduction of unexplained confusion or the conclusion of incorrect causal directions in causal inference. For example, an abnormal pressure reading caused by a temporary sensor malfunction, if not identified and removed, may be misjudged as a cause of product quality fluctuations in causal discovery, or mistakenly regarded as a key variable that needs to be controlled in the optimization model. Further format standardization and alignment operations realize the fusion and computability of multi-source heterogeneous data in terms of time and measurement dimensions, so that the rule effective time from the document library, the record time in the database, and the sampling time of equipment monitoring can be placed on the same time axis. This is crucial for analyzing the temporal causal relationship between standard adjustments, historical events, and real-time status. Unified measurement units eliminate the dimensional differences caused by different data sources.

[0029] S1.3 Store multi-source data of a standard system that has undergone format standardization and alignment operations according to a predefined pattern to generate a structured dataset.

[0030] Furthermore, multi-source data is integrated into a knowledge base that can be directly used for complex analyses. Traditional database designs are typically tailored to specific business functions, resulting in fragmented data schemas. In contrast, this method predefines a schema designed to construct a global, relational data view. It not only defines basic fields such as standard numbers, parameter values, timestamps, and equipment IDs, but more importantly, it predefines semantic relationships such as constraints between standard entries and process parameters, membership relationships between historical batches and real-time monitoring points, and potential influence relationships between environmental variables and quality indicators.

[0031] S2. Based on structured datasets, construct a standard causal influence network using causal discovery algorithms.

[0032] S2.1 Extract the observation data sequences corresponding to standard variables, environmental variables, and performance variables from the structured dataset.

[0033] Furthermore, input data with clear causal semantic distinctions is prepared for the causal discovery algorithm. This is a key prerequisite for moving from general data analysis to causal inference. Traditional multivariate analysis often treats all observation data equally, ignoring the potential differences in the roles of different variables in the causal logic chain. Based on the inherent attributes of variables and their role in the standard system, they are pre-classified into standard variables, environmental variables, and performance variables. For example, a standard variable may be the control threshold of a certain process parameter, an environmental variable may be the ambient temperature of the workshop, and a performance variable may be the pass rate of the final product. This guides the algorithm to focus more on the possible causal paths from standard variables and environmental variables to performance variables, rather than performing an indiscriminate exhaustive search among all variables. This greatly improves the efficiency and accuracy of causal discovery and ensures that the final causal influence network is interpretable and has direct guiding significance in business.

[0034] Specifically, by precisely extracting the corresponding observation data sequences from the structured dataset according to this classification, causal discovery can focus on revealing the core business question of what performance results will be caused by adjusting what standards or how the environment changes. Based on predefined classification rules, the structured dataset is used to select the observation sequences of standard variables representing standard adjustable parameters, environmental variables representing external uncontrollable factors, and performance variables representing the final output results.

[0035] S2.2 Input the observation data sequences corresponding to the standard variables, environmental variables and performance variables into the causal discovery algorithm, use the PC algorithm to calculate the conditional independence between variables, and obtain the test results of the conditional independence between variables.

[0036] Furthermore, the PC algorithm is used to calculate the conditional independence between variables, obtaining the test results of conditional independence between variables. This constraint-based causal discovery method reveals the true dependency structure between variables, providing a statistical foundation for moving beyond correlation and establishing causal models. The PC algorithm tests whether any two variables remain independent given a set of other variables, thereby gradually eliminating spurious associations by traversing the entire set of variables. Consider all unordered pairs of variables and all possible subsets of conditions. Through calculation Statistical decision-making is used to determine conditional independence. For example, when analyzing the relationship between workshop humidity and product defect rate, directly obtaining the data may show a correlation, but the PC algorithm will test whether the two are still correlated under the given ambient temperature condition. If conditional independence holds, the observed correlation may be a spurious correlation caused by the common cause of ambient temperature.

[0037] Specifically, the most likely direct associations between variables are initially identified from the observational data, excluding indirect associations caused by third-party variables. This provides solid, statistically based evidence for constructing an undirected causal backbone network. This is the first crucial step in deriving the causal structure from the data. The observational data sequences corresponding to the standard variables, environmental variables, and performance variables are input into the PC algorithm. This algorithm iteratively increases the size of the condition set, performs conditional independence tests on each pair of variables, and records the results of the tests. And based on the significance level, the independent or non-independent decision conclusions are derived, ultimately outputting a complete set of conditional independence test results.

[0038] The expression for the result of the conditional independence test between variables is: ;in, For the results of the conditional independence test, The first variable to be tested selected from the entire set of variables. The second variable to be tested is selected from the entire set of variables. For conditional subsets, This is a statistical measure for testing conditional independence. This is a binary judgment conclusion. The set of all variables to be analyzed. This is the first variable to be tested. For the second variable to be tested, S2.3, based on the results of the conditional independence test between variables, construct a draft of the causal influence network of the standard system characterizing the preliminary causal relationship between variables.

[0039] Furthermore, the abstract results of statistical tests are transformed into an intuitive and visual network structure representation, serving as a concrete bridge from data conclusions to knowledge models. The Conditional Independence (CI) test results themselves are a set of test conclusions, and an undirected graph is constructed based on these conclusions: if two variables are determined to be non-independent under all tested condition sets, then an undirected edge is connected between them. For example, if equipment maintenance frequency and equipment failure rate show a dependency relationship in all tests, then there will be an edge connecting these two nodes in the sketch. The significance of the sketch is that it discards indirect edges generated by confounding variables and retains the connections that are most likely to represent direct causal relationships.

[0040] Specifically, the purpose of constructing the sketch is to provide a clear framework for directional inference, clarify which pairs of variables have potential direct causal relationships that need further investigation, narrow the search space for subsequent directional inference, and make the entire causal discovery process more efficient and focused. Analyze the results of the conditional independence test, establish an undirected connection edge between all pairs of variables whose decisions are not independent, and all variables and the undirected edges between them together constitute an undirected graph. The undirected graph is a sketch of the causal influence network of the standard system representing the preliminary causal relationship between variables.

[0041] S2.4 Perform directional inference and structural optimization on the sketch of the causal influence network of the standard system, determine the direction of the causal edges and remove the pseudo edges.

[0042] Furthermore, we assign direction to the undirected causal framework, refining it into a more accurate causal structure. This is a key step in upgrading the relationship to a causal mechanism. The sketch only tells us that there is a direct relationship between variables, but it does not indicate the causal direction. We use the conditional independence pattern, temporal information, or domain constraints in the data to infer the direction of the edges. For example, by using the order of variable values, we can determine that a variable that is later in time cannot be the cause of a variable that was earlier in time. By checking the conditional independence pattern, we can infer the unique causal direction. The structural optimization further removes the false edges found in the direction inference process. False edges may be retained in the initial sketch due to data noise or verification errors.

[0043] Specifically, the fuzzy network of associations is transformed into a directed acyclic graph with a clear causal direction, clearly indicating the potential causal driving path. For example, improving the purity of raw materials leads to an increase in the yield rate. The sketch of the causal influence network is combined with the conditional independence pattern between variables, the known temporal relationship between variables, or the constraint on causal irreversibility in the domain knowledge to infer a possible direction for each undirected edge. The possible conflicts or loops that may be generated after the inference are checked and adjusted, and edges that do not conform to the directional inference logic or lack sufficient statistical evidence are removed, resulting in a directed acyclic graph as a refined causal structure.

[0044] S2.5. Perform parameter learning on the causal influence network of the standard system after directional inference and structural optimization, estimate the effect strength of each causal edge, and obtain the parameterized causal influence network of the standard system.

[0045] Furthermore, transforming qualitative causal structures into quantitative causal models allows causal effects to be measured, making it an essential step in applying causal knowledge to numerical optimization decision-making. A structural graph with only causal directions can only answer whether an effect occurs, but not how significant. Based on a defined directed acyclic graph, a linear regression method is used to estimate the magnitude of the causal effect represented by each directed edge in the graph using observed data sequences. For example, after determining the causal edge from training duration to operational error rate, parameter learning can estimate the specific numerical value by which the operational error rate can be reduced on average for each additional unit of training duration.

[0046] Specifically, the parameterized causal influence network reveals only the causal topology between variables and quantifies the strength of causal effects. It provides crucial functional relationships for establishing multi-objective optimization mathematical models: the specific mathematical form of how performance variables in the objective function change with standard and environmental variables. Based on these estimated causal effect strengths, the model is constructed to ensure that the optimization model is based on real causal mechanisms rather than spurious correlations. According to the determined directed causal structure, the causal relationship between variables is expressed in the form of structural equations or regression equations. The parameters in the model are estimated using the observation data sequences corresponding to standard, environmental, and performance variables to obtain the effect coefficients corresponding to each causal edge, forming a standard system of causal influence networks.

[0047] S3. By establishing a causal influence network, a mathematical model of a standard system multi-objective optimization problem is established. Combined with the personalized characteristic data of decision-makers, a personalized set of optimization strategies is generated through counterfactual reasoning.

[0048] S3.1 Extract the decision-maker's implicit preference for target weights and risk tolerance from the decision-maker's historical interaction records to form the decision-maker's personalized characteristic data. Define a personalized utility function that reflects the decision-maker's preferences and integrate the personalized utility function with multiple objective functions in the mathematical model.

[0049] Furthermore, this approach explicitly and quantifies the subjective value judgments that are difficult to articulate during the decision-making process, using them as a guideline to drive optimization search. This enables a shift from general optimization to personalized decision support. Traditional multi-objective optimization typically outputs a general Pareto front, treating all objectives equally or pre-setting weights, ignoring the unique trade-offs of different decision-makers in specific situations. By analyzing the interaction records of decision-makers' choices, evaluations, and adjustments when faced with historical optimization solutions, this approach uses preference learning techniques to inversely infer the decision-makers' true level of importance they place on each performance objective, i.e., implicit weights, as well as their risk attitude under uncertainty. For example, a decision-maker may consistently prefer to choose solutions that excel in cost and delivery time objectives but only meet the minimum requirements for quality objectives. This reveals the decision-maker's high sensitivity to cost and time and relative tolerance for quality. The personalized utility function is the mathematical encapsulation of this implicit preference. It may be a weighted sum of the objective functions or a more complex function that includes a risk term.

[0050] Specifically, transforming the complex multi-dimensional trade-off problem into a single-objective search problem in a personalized utility space not only greatly simplifies the complexity of the optimization process, but more importantly, ensures that the final strategy set naturally aligns with the decision-maker's value orientation. This solves the pain point of the disconnect between decision-making and optimization in traditional methods. By analyzing the decision-maker's historical interaction records, identifying their selection and evaluation patterns for different optimization schemes in different contexts, and using methods such as regression analysis or learning ranking to extract the implicit target weight vector and risk aversion coefficient.

[0051] S3.2 Within the decision variable space and constraint set defined by the mathematical model, candidate optimization strategies are searched and generated, guided by maximizing the personalized utility function that integrates the decision-maker's individual characteristics.

[0052] Furthermore, by utilizing optimization algorithms, within a feasible domain that conforms to causal logic and realistic constraints, the algorithm proactively explores potential strategy points that best meet the personalized value demands of decision-makers. The key leap from analysis to generation involves a mathematical model defining the rules of the game for strategies—which criteria can be adjusted, how, and where the boundaries of adjustment lie. The personalized utility function defines the winning criteria—what strategy combinations are good. The algorithm proactively searches for strategies most likely to win, using the decision-maker's win as the standard. For example, for a decision-maker who highly values ​​compliance risk, their personalized utility function will assign a very high weight to compliance objectives. The optimization algorithm will then prioritize searching the decision variable space for strategy regions that maximize compliance while also considering other objectives. The search process employs variants of multi-objective evolutionary algorithms such as genetic algorithms and particle swarm optimization.

[0053] Specifically, a pool of candidate strategies customized for decision-makers is generated. The strategies in this pool are not only technically feasible, but also inherently carry value attributes that are highly correlated with the decision-makers' preferences, laying a high-quality foundation for subsequent refined screening and interpretation. Optimization search algorithms, such as multi-objective evolutionary algorithms, are applied to iteratively generate a series of technically feasible and potentially personalized optimization strategies within the range of decision variables defined by the mathematical model, on the premise of satisfying all constraints, with the goal of maximizing the value of the personalized utility function.

[0054] S3.3 For each candidate optimization strategy, construct a causal intervention scenario for implementing the candidate optimization strategy and a causal control scenario for maintaining the baseline state.

[0055] Furthermore, a counterfactual experiment is created for each candidate strategy to assess its true value. This is key to shifting from traditional predictive assessment to causal assessment. Traditional strategy assessments typically only predict the future state after implementing the strategy, but lack a reliable baseline for comparison of what would happen if the strategy were not implemented. For each candidate optimization strategy, two parallel worlds are explicitly defined: one is the intervention world, which assumes that we implemented the candidate strategy and the standard variables were adjusted according to the strategy recommendations; the other is the control world, which assumes that we keep the status quo unchanged and everything operates according to the existing baseline. These two worlds share all other initial conditions and external environments except for the intervention variable, providing a logical framework for removing confounding factors and assessing the net effect of the strategy.

[0056] Specifically, by comparing the differences in outcomes between two worlds, the effects of implementing a strategy can be distinguished from changes caused by other concurrent factors. For example, when evaluating a strategy to increase equipment maintenance frequency, it is necessary to distinguish how much of the yield improvement is truly caused by the increase in maintenance frequency, and how much may be due to the simultaneous replacement of raw materials with higher quality ones. The premise for achieving causal inference is to create a benchmark scenario that is exactly the same as the current standard system's true state for each candidate optimization strategy, as well as a causal intervention scenario that only applies the values ​​of the standard variables specified by the strategy to the current state. These two scenarios constitute a pair of analytical foundations for counterfactual reasoning.

[0057] S3.4 Based on the objective function derived from the causal influence network in the mathematical model, the evolution trajectory of each performance objective is deduced in the causal intervention scenario and the causal control scenario of maintaining the baseline state. By comparing the differences in the terminal values ​​of the evolution trajectory, the causal effect of the candidate optimization strategy is quantified.

[0058] Furthermore, by utilizing the established causal model, rigorous counterfactual inferences are performed to quantify the true contribution of each candidate strategy, rather than its accompanying effects. The objective function in the mathematical model is not a simple statistical fitting model, but a predictive model based on a causal influence network, which contains the causal mechanism between variables. Using this causal predictive model as a simulator, with causal intervention scenarios and causal control scenarios as inputs, the complete trajectory of the evolution of each performance objective over time or logical steps under different interventions is deduced. The deduction process strictly follows the transmission path and effect intensity defined by the causal network. By comparing the values ​​at the endpoints of the two deduced trajectories, the difference is defined as the causal effect of the candidate optimization strategy.

[0059] Specifically, for example, through causal model deduction, the yield rate eventually increases to a certain value in the intervention world, while in the control world, the yield rate remains at another value. The difference between the two can be attributed to the implementation of the strategy, providing a mechanism-based and interpretable effect assessment method. This not only tells decision-makers the expected results of the strategy but also clearly reveals the extent to which this result is brought about by the strategy itself, avoiding the risk of misjudging correlation as causation and making wrong decisions. The initial values ​​of variables and the values ​​of strategy variables in the causal intervention scenario and the causal control scenario are respectively input into a mathematical model constructed based on the causal influence network. Through iterative simulation, two sets of evolutionary trajectories describing the changes of each performance objective over time are obtained. The numerical difference between the two sets of evolutionary trajectories at the final time point or in a steady state is obtained, which is quantified as the causal effect of the candidate optimization strategy.

[0060] S3.5 Based on the causal effects of candidate optimization strategies, and combined with the preference structure and risk propensity contained in the decision-maker's personalized characteristic data, the candidate optimization strategies are ranked in multiple dimensions and screened for contextual adaptability, generating a set of personalized optimization strategies that are deeply matched with the decision-maker's individual decision-making logic.

[0061] Furthermore, from the causal set of effective strategies to the final transformation of value-aligned personalized recommendations, the causal effects of candidate optimization strategies provide objective and scientific performance predictions. However, this is not the whole story of decision-making. By using the decision-maker's personalized characteristic data as a value lens, we can examine the causal effects of each strategy. Multi-dimensional value ranking not only considers the absolute improvement of each objective but also combines the weight preferences in personalized characteristic data for comprehensive scoring. Contextual suitability screening further considers whether the risk characteristics of the strategy match the decision-maker's risk tolerance.

[0062] Specifically, for example, a strategy that brings high returns but also has large performance fluctuations may be suitable for risk-seeking decision-makers but not for risk-averse decision-makers. Through this comprehensive evaluation and screening that integrates objective causal effects and subjective value preferences, a personalized set of optimization strategies is generated. This is a carefully selected strategy menu, which reduces the cognitive load of decision-makers, improves decision-making efficiency and quality, and makes the optimization results truly understandable, acceptable, and implementable. Based on the improvement value of each objective in the causal effect of the candidate optimization strategies, combined with the objective weights in the personalized feature data, the comprehensive utility value of each strategy is ranked. The volatility or probability of achieving the objective in the causal effect is analyzed and compared with the risk tolerance in the personalized feature data. Strategies that exceed the acceptable risk range are screened out. The combined ranking and screening results output a personalized set of optimization strategies that is optimized in terms of both scientific rigor and personal suitability, and is scalable.

[0063] S4. Based on the set of personalized optimization strategies, each optimization strategy is simulated and verified in multiple scenarios through a standard system simulation unit.

[0064] S4.1 Select the current optimization strategy to be verified from the set of personalized optimization strategies and output it to the standard system simulation unit. Construct a benchmark operation scenario based on the current environmental state in the standard system simulation unit.

[0065] Furthermore, a repeatable virtual experimental environment based on the real world was established for simulation of theoretical optimization strategies, thereby assessing their feasibility before investing actual resources. Strategies are directly selected from a personalized set of optimization strategies for simulation verification, seamlessly integrating the optimization and verification processes. The current optimization strategy to be verified is output to the simulation unit, essentially transforming a set of abstract parameter settings into configuration instructions recognizable by the simulation environment. A baseline operating scenario based on the current environmental state is constructed within the standard system simulation unit, creating initial conditions and background highly consistent with the real world. The baseline scenario replicates the current state of the standard system before implementing any new strategy, including a current snapshot of all equipment status, resource levels, process parameters, and external environmental factors.

[0066] Specifically, for example, for an optimization strategy for a production line, the baseline scenario simulates the current capacity configuration, material inventory, personnel scheduling, and market demand of the production line. The scenario is built based on the current state, making the simulation verification highly relevant to reality and timely, and able to reflect the expected performance of the strategy in the current environment. A strategy is selected from the set of personalized optimization strategies in sequence or according to preset priority as the current optimization strategy to be verified. The adjustment values ​​or control logic of the defined standard parameters are configured into the standard system simulation unit. The environmental variables and performance variables data collected from the actual standard system that reflect the latest state are used as initial conditions and input into the standard system simulation unit, thereby constructing a baseline operating scenario that is completely synchronized with the starting point of the real world.

[0067] S4.2. Based on the baseline operating scenario, verification scenarios with different disturbance conditions are derived based on predefined rules.

[0068] Furthermore, we construct a set of stress test environments that include diverse and challenging conditions. The real world is full of uncertainty, and the baseline scenario is only one of countless possible futures. Instead of verifying a single scenario, we systematically apply various disturbances described by predefined rules to generate a series of possible future scenarios. The predefined rules can be formulated based on historical fluctuation patterns and risk analysis to simulate changes in key variables. For example, we can generate verification scenarios with different disturbance conditions such as a sharp drop in market demand, an increase in the price of key raw materials, an increased probability of sudden failure of a core piece of equipment, or the early implementation of new environmental regulations. This can proactively expose the potential vulnerabilities of optimization strategies under non-ideal conditions. Strategies that perform well in the baseline scenario may completely fail or even have negative effects in a certain disturbance scenario.

[0069] Specifically, by conducting multi-scenario verification during the simulation phase, strategies that not only perform well but are also robust enough are selected, or strategies that require additional fault tolerance or emergency mechanisms are identified. This reduces the risk of strategies failing due to sudden environmental changes after actual deployment, and enhances the practical value and reliability of the strategies. Based on the parameters and conditions of the established benchmark operating scenario, specific variables or parameters in the scenario are modified according to a predefined perturbation rule library. For example, the values ​​of environmental variables are changed according to a certain pattern, the impact of external events is simulated, or the constraints of internal resources are adjusted. This systematically generates a series of verification scenarios that share the core logic with the benchmark scenario but have different external conditions.

[0070] S4.3 In the standard system simulation unit, the optimization strategy is loaded into each verification scenario in sequence, and the simulation is driven to execute. The simulation execution process data and terminal performance indicators of the optimization strategy in each verification scenario are recorded and output.

[0071] Furthermore, through simulation experiments, complete performance data of each optimization strategy under diverse future scenarios are collected. The optimization strategy is then loaded into the verification scenario and implemented in the simulation environment. This involves adjusting the corresponding standard parameters or control rules in the simulation model according to the strategy definition. Driving the simulation execution means running the virtual standard system under the given strategy and scenario conditions according to its inherent dynamic logic for a period of time. An independent simulation experiment is performed for each strategy-scenario combination, and the simulation execution process data and the final terminal performance indicators are recorded throughout. The process data may include the temporal changes of key intermediate variables, the satisfaction of constraints, the dynamic process of resource consumption, etc. The terminal performance indicators are the objective function values ​​calculated at the end of the simulation, such as total cost, total output, and average quality, generating a multi-dimensional and comparable report on each optimization strategy.

[0072] Specifically, by comparing the performance of the same strategy in different verification scenarios, its robustness can be evaluated. By comparing the performance of different strategies in the same verification scenario, their relative advantages and disadvantages can be assessed. The detailed process data recorded can also help diagnose why the strategy succeeds or fails in certain scenarios, providing valuable insights for subsequent dynamic adjustment of the strategy. Comprehensive simulation-based testing is the most thorough and low-risk effective evaluation method before deploying the strategy in the real world. For each generated verification scenario, the configuration of the current optimization strategy to be verified is fully applied to the simulation model of that scenario. The simulation engine is started and run for a sufficiently long simulation duration to observe the complete effect of the strategy. During the simulation, simulation execution process data reflecting the dynamic operation of the standard system is continuously collected and recorded. At the end of the simulation, various terminal performance indicators reflecting the final optimization effect are calculated and recorded, thereby generating a complete set of strategy performance records for each verification scenario.

[0073] S5. Dynamically adjust the optimization strategy based on the simulation verification results.

[0074] S5.1 Based on the simulation execution process data and terminal performance indicators of the optimization strategy in each verification scenario, identify the performance vulnerability scenarios and parameter sensitivity ranges of the optimization strategy, formulate corresponding strategy parameter compensation rules, and generate strategy parameter adjustment schemes.

[0075] Furthermore, by transforming simulation testing from a passive performance evaluation to an active process of vulnerability diagnosis and remediation solution generation, we achieve in-depth analysis and targeted enhancement of optimization strategies. Traditional simulation verification often stops at outputting performance reports. By deeply mining simulation execution process data and terminal performance indicators, we can identify two key issues: first, performance vulnerability scenarios, i.e., under which specific disturbance conditions the optimization strategy performs significantly worse than expected or fails; second, parameter sensitivity ranges, i.e., which key parameters in the optimization strategy will cause a sharp drop in performance when there are slight changes. For example, by analyzing simulation data, we may find that an optimization strategy aimed at improving efficiency actually reduces efficiency significantly under the scenario of fluctuating raw material supply, which identifies a performance vulnerability scenario for the strategy. Further analysis reveals that the equipment operating rate parameter set in the strategy is effective within a narrow range, but fails outside the range, which identifies the parameter sensitivity range. For example, for the above vulnerability scenario, the compensation rule may be to automatically reduce the equipment operating rate to a robust range when the raw material supply fluctuation is detected to exceed the threshold, and generate a strategy parameter adjustment scheme. This kind of diagnostic insight is transformed into a specific and executable set of parameter modification instructions.

[0076] Specifically, this transforms the optimization strategy from a static and fragile solution into one with the potential to self-adjust based on environmental feedback. By identifying vulnerabilities and sensitivities, adjustment plans are generated accordingly, injecting adaptive logic to cope with uncertainty into the strategy. By comparing and analyzing the terminal performance indicators of the optimization strategy under different verification scenarios, scenarios where the performance indicators are significantly lower than the benchmark or preset threshold are marked as performance-vulnerable scenarios. The correlation between the dynamic changes of key parameters in the simulation execution process data and the performance indicators is analyzed to identify the intervals in which small parameter changes lead to large performance fluctuations, i.e., parameter-sensitive intervals. For each performance-vulnerable scenario or parameter-sensitive interval, one or more if-then logical rules are designed to specify which strategy parameter should be adjusted in which direction under what conditions. The set of these rules constitutes the strategy parameter adjustment plan.

[0077] S5.2. Based on the strategy parameter adjustment scheme, modify the adjustable parameters in the optimization strategy, generate the modified optimization strategy, input the modified optimization strategy into the standard system simulation unit, re-simulate in the performance fragile scenario, and obtain the simulation verification results of the modified optimization strategy.

[0078] Furthermore, targeted modifications based on vulnerability analysis results are possible. For example, if a strategy is identified as failing in high-temperature environments, and the parameter sensitivity range is coolant flow rate, the modification might involve adjusting the coolant flow rate setting from X to Y when the ambient temperature exceeds a certain threshold. Generating a modified optimization strategy means that the strategy itself has been upgraded from a set of fixed parameters to a more complex set of rules that includes conditional judgments and parameter adaptive logic. The modified optimization strategy is then placed back into the performance vulnerability scenario that previously caused its failure for simulation verification. This allows for immediate verification of the effectiveness of the strategy parameter adjustment scheme and whether it truly compensates for the strategy's weaknesses, enabling it to recover compliant performance in the previously failed scenario. Through this targeted verification focused on weak links, the repair effect can be efficiently confirmed, avoiding the huge computational overhead of re-simulating and verifying across all scenarios. This reflects the engineering wisdom of strengthening weaknesses and improves the efficiency and targeting of strategy optimization.

[0079] Specifically, based on the rules defined in the strategy parameter adjustment scheme, the parameter values ​​or control logic of the current optimization strategy to be verified are modified to generate a new strategy version with enhanced adaptability to specific scenarios, namely the revised optimization strategy. The revised optimization strategy is then used as input and reloaded into the standard system simulation unit. Specifically, previously identified performance-vulnerable scenarios are selected as the simulation environment to drive the simulation execution again. After the simulation, the new simulation execution process data and terminal performance indicators of the revised optimization strategy in these specific scenarios are collected and recorded. The data constitute the simulation verification results of the revised optimization strategy.

[0080] S5.3 Based on the simulation verification results of the modified optimization strategy, confirm that the performance of the optimization strategy meets the preset performance threshold, and use the confirmed optimization strategy as the verified and adjusted optimization strategy.

[0081] Furthermore, the preset performance threshold is a multi-dimensional standard, which may include minimum requirements for key performance indicators, performance stability requirements in different scenarios, etc. For example, an optimization strategy may be required to ensure that the core performance indicators do not drop below a certain percentage in all verification scenarios, and that the performance in at least a few critical and vulnerable scenarios is improved above the baseline. The simulation verification results based on the modified optimization strategy are compared with these preset performance thresholds one by one. If the modified optimization strategy meets the thresholds in all required dimensions, then it passes the verification and is confirmed as the verified and adjusted optimization strategy.

[0082] Specifically, through a series of operations—identifying vulnerabilities, making targeted corrections, regression verification, and confirming compliance—the final output strategy is ensured to be not only theoretically optimal but also robustly enhanced and quality-tested, possessing a stronger ability to cope with the complexity and uncertainty of the real world. This reduces the risk of directly deploying potentially flawed optimization strategies into actual standard systems, improving the practicality and output quality of the entire optimization method. The terminal performance indicators in the simulation verification results of the corrected optimization strategy are compared and judged with the various performance acceptance standards pre-set for the optimization strategy. If the terminal performance indicators of the corrected optimization strategy reach or exceed the preset performance thresholds in all previously identified performance vulnerability scenarios and other necessary verification scenarios, the corrected optimization strategy is deemed to have passed verification and is marked as the verified and adjusted optimization strategy.

[0083] S6. Deploy the verified and adjusted optimization strategy into the actual standard system for execution.

[0084] S6.1. Convert the verified and adjusted optimization strategy into executable instructions or configuration changes in the actual standard system.

[0085] Furthermore, based on the different forms of the target execution terminal, the strategy is translated into a language that it can understand and execute. For example, if the optimization strategy suggests raising the upper limit of the temperature standard in a certain production process by a certain degree, then the conversion process needs to generate specific instructions to modify the setpoint of the temperature controller for that process. If the optimization strategy involves adjusting the sampling frequency of quality inspection, then the conversion process may need to generate instructions to update the configuration file in the quality management system, or generate work instructions to be issued to quality inspectors, thus realizing a closed loop from decision support to operation execution. It fully considers the heterogeneity and interface differences of downstream execution systems, ensuring that the optimization intent can be fully and accurately implemented, avoiding execution deviations caused by unclear expression or incompatible formats.

[0086] Specifically, the content of the optimized strategy after analysis, verification and adjustment is analyzed, and the standard parameters, control logic or operation rules that need to be adjusted are identified. According to the specifications and protocols of the corresponding execution unit or management interface in the actual standard system, the adjusted content is mapped and generated into specific operation instructions, configuration file updates, work order tasks or document revisions. The generated instruction or change set completes the conversion from strategy to executable action.

[0087] S6.2 In a controlled environment of the actual standard system, deploy the converted verification and adjusted optimization strategies in stages.

[0088] Furthermore, a gradual and risk-controlled strategy implementation mechanism is adopted to minimize the impact on existing stable operations and allow for final verification and fine-tuning in a real environment. Even after sufficient simulation verification, there are still risks in deploying the optimization strategy to the actual standard system all at once, because the simulation model cannot fully reproduce all the details and contingencies in the real world. Two key concepts are introduced: controlled environment and phased deployment. The controlled environment refers to a subset or specific part of the actual standard system, such as a single production line, a specific business unit, or a pilot area. The impact of deploying the strategy in this environment is limited, observable, and can be quickly rolled back. It is carried out step by step according to the plan. For example, the initial deployment and observation are carried out in a minimal controlled environment. After verification, the deployment scope is gradually expanded. Alternatively, it can be deployed in batches according to the different components of the strategy. For example, for a complex optimization strategy with multiple adjustment items, the part that is considered to have the lowest risk or the most obvious benefit is deployed first. After observing the effect, the remaining parts are deployed.

[0089] Specifically, the optimization strategy can be tested in real-world conditions, without causing a global impact, to collect real feedback data. If unforeseen problems are discovered in the early stages of phased deployment, the deployment plan can be stopped or adjusted in a timely manner to avoid large-scale losses. Phased deployment also helps organizations learn and adapt to changes, accumulating experience and confidence for subsequent full-scale rollout. A controlled environment that is isolated from or has limited impact on the actual operating environment should be selected, such as an independent test production line or a pilot business unit. The executable instructions or configuration changes generated by the converted, verified, and adjusted optimization strategy should be applied to this controlled environment. After running stably in the controlled environment for a period of time and confirming that the effect meets expectations, the strategy can be gradually promoted to a wider part of the actual standard system according to the predetermined deployment roadmap until full deployment is completed.

[0090] S7. Collect strategy execution process data in real time through the monitoring unit.

[0091] S7.1 Real-time collection, verification, and adjustment of status and performance data during the execution of the optimized strategy. The status and performance data are timestamped and formatted to generate preliminary strategy execution process data.

[0092] Furthermore, to optimize the effectiveness of the strategy in a real-world environment, two types of key data are collected: status data, reflecting the real-time operational status of the standard system under the influence of the strategy, such as equipment parameters, process progress, and resource consumption; and performance data, directly corresponding to optimization objectives, such as output quantity, quality indicators, and cost consumption. Collecting only these data is insufficient; timestamp alignment and formatting aim to unify raw data streams from different sources, frequencies, and formats into a coherent and orderly time-series framework. For example, temperature status data collected every second from sensors needs to be aligned and correlated with batch pass rate performance data collected every ten minutes on a unified time reference. This generates preliminary strategy execution process data, forming a complete and clean data tape that records at what point in time the system was in what state and what performance was generated. This provides indispensable raw materials for accurately assessing the true impact of the strategy and identifying unexpected side effects.

[0093] Specifically, through data interfaces and sensors deployed at key nodes of the standard system, process parameters reflecting the operational status and performance indicators reflecting the achievement of goals are continuously collected in a fixed or event-triggered manner. The collected status data and performance data are aligned in time according to a unified time base and converted into a pre-agreed data format and unit, and integrated to form a preliminary strategy execution process data record arranged in chronological order and containing status and performance dimensions.

[0094] S7.2. The preliminary strategy execution process data is correlated and labeled in real time with the predefined expected strategy effect indicators to obtain a complete structure.

[0095] Furthermore, evaluation semantics and comparison benchmarks are injected into the raw execution observation data, elevating it from a simple record into a context-rich, story-driven dataset that can be directly analyzed and evaluated. Initial strategy execution data is a record of objective facts, but lacks interpretation of its meaning. Predefined expected strategy performance indicators are introduced as benchmarks. These indicators originate from the goals set during the strategy generation phase, such as the expected percentage reduction in unit energy consumption or the expected increase in mean time between failures (MTBF). The real-time correlation and labeling process compares the collected actual performance data with these expected indicators and assigns semantic labels such as whether the data meets expectations, the direction and degree of deviation. For example, when the real-time collected unit energy consumption value is lower than the expected threshold, the corresponding data point is marked as better than expected; conversely, it may be marked as not meeting expectations and the deviation is recorded.

[0096] Specifically, the evaluation algorithm can directly perform statistics and analysis based on these tags without repeatedly tracing back to the original target and calculating deviations. Real-time correlation and tagging enable the monitoring itself to have immediate feedback capabilities, allowing it to issue timely alerts when strategy execution deviates from expectations. This provides a time window for possible dynamic intervention, transforming the monitoring unit from a passive data recorder into an active strategy execution dashboard with preliminary diagnostic capabilities. During data stream processing, the performance data portion of the preliminary strategy execution process data that arrives in real time is instantly calculated and compared with predefined strategy expected effect indicators. Based on the comparison results, corresponding semantic tags are assigned to the performance data of each time segment, and these tags are integrated as new metadata dimensions with the original status data and performance data to form strategy execution process data.

[0097] S8. Based on the data from the strategy execution process, conduct a multi-dimensional evaluation of the effectiveness of the optimization strategy.

[0098] S8.1 Extract the sequence of actual performance indicators directly related to the performance objectives of the strategy from the data of the strategy execution process.

[0099] Furthermore, based on the clearly defined performance objectives initially set by the optimization strategy, a sequence of actual performance indicators directly corresponding to these objectives is precisely extracted from the complex process data. For example, if one of the objectives of the optimization strategy is to reduce energy consumption per unit of product, then a sequence of actual measured values ​​of energy consumption per unit of product arranged in chronological order is extracted from the process data; if another objective is to shorten the average order processing cycle, then a sequence of average order processing cycles is extracted accordingly.

[0100] Specifically, it achieves strict alignment between the evaluation focus and the initial optimization intention, filters out irrelevant noise data, and ensures that subsequent evaluation work is carried out in accordance with the strategy's preset goals. This enables the evaluation conclusions to directly answer the core question of the extent to which the strategy has achieved its intended goals. This provides an accurate data foundation for objectively and fairly quantifying the strategy's effectiveness. Based on each performance goal clearly listed in the optimization strategy document, data fields that correspond perfectly to these goals in the strategy execution process data are located and filtered out. The values ​​of these fields are extracted and arranged in chronological order to form a series of time series. Each series corresponds to the actual performance trajectory of a performance goal during the strategy execution period.

[0101] S8.2 Compare the sequence of actual performance indicators directly related to each performance objective of the optimization strategy with the historical benchmark performance data before the implementation of the optimization strategy, and quantitatively describe the absolute and relative changes in performance of each dimension after the implementation of the optimization strategy.

[0102] Furthermore, historical benchmark performance data prior to the implementation of the optimization strategy is selected as the control baseline. This typically refers to performance data from a representative and stable historical period before the strategy deployment. By comparing the actual performance indicator sequence with this baseline, the impact of the strategy implementation can be clearly identified. Quantitative descriptions include calculating the absolute change magnitude, such as how many kilowatt-hours the average unit energy consumption decreased after implementation compared to before implementation; and the relative change magnitude, such as what percentage energy consumption was reduced. This provides a standardized measure of the strategy's effectiveness.

[0103] Specifically, absolute magnitude provides the specific scale of the effect, making it easier to understand the actual benefits or costs; relative magnitude eliminates the influence of differences in the dimensions of different indicators, making it easier to compare the strength of the strategy's effect across different objectives. This is much more scientific and powerful than simply presenting data after implementation or making subjective evaluations. For each performance objective, the statistical characteristics of the actual performance indicator sequence after implementation, such as the mean or median, are used to obtain the difference between this characteristic value and the historical benchmark performance data mean before the implementation of the optimization strategy to obtain the absolute magnitude of change. The percentage of this difference relative to the historical benchmark mean is then obtained to obtain the relative magnitude of change, thereby quantifying the performance improvement or degradation of each dimension.

[0104] S8.3. Combine the risk tolerance data in the decision-maker's personalized characteristics to evaluate the performance volatility and stability of the optimization strategy in the actual implementation process.

[0105] Furthermore, by using the decision-maker's subjective risk preference as one of the evaluation benchmarks for objective performance, personalized effect evaluation is achieved. It is recognized that stable and controllable returns and highly volatile returns have drastically different values ​​for different decision-makers. The actual performance indicator series not only includes trends and means, but also volatility. It goes beyond simply calculating statistical indicators of volatility, such as variance or standard deviation, and further correlates and compares these objective volatility indicators with the risk tolerance explicitly stated in the decision-maker's personalized characteristic data. For example, the personalized characteristic data of a risk-averse decision-maker may contain a low risk tolerance parameter. If an optimization strategy improves efficiency on average, but its performance series fluctuates wildly and frequently experiences troughs far below the average, then for this decision-maker, the stability evaluation result of the strategy may be poor.

[0106] Specifically, the effectiveness evaluation goes beyond a simple performance improvement perspective by incorporating a risk adaptability dimension. It answers not only whether the strategy is effective, but also whether the way this effect is achieved reassuring to decision-makers. This ensures that the evaluation results match the decision-maker's personal management style and risk tolerance, making the evaluation conclusions more valuable for supporting the final decision. Analyzing the fluctuations of each actual performance indicator sequence around its trend line allows for the identification of statistical measures that characterize volatility and stability, such as standard deviation or coefficient of variation. Then, the magnitudes of these statistical measures are compared with the risk tolerance thresholds extracted from the decision-maker's personalized characteristic data to determine whether the actual performance fluctuations of the optimized strategy are within the decision-maker's acceptable range, thus completing a personalized assessment of performance volatility and stability.

[0107] S8.4 Analyze the abnormal events and adjustment operations recorded in the data during the strategy execution process, and evaluate the adaptability of the optimization strategy to environmental disturbances and maintenance costs.

[0108] Furthermore, by analyzing the process logs of strategy execution, we can uncover its resilience and operability, assess its robustness and implementation costs in dynamic real-world environments, and not just focus on its static performance results. Optimization strategies may differ between their performance under ideal conditions and in complex real-world environments. By carefully examining the anomalies and adjustment operations recorded in the strategy execution process data, we can indirectly assess the strategy's adaptability and maintenance costs. Anomalies refer to unplanned situations that occur during strategy execution, such as sudden equipment failures or sudden changes in demand. Adjustment operations refer to manual or automatic interventions made to address these events or maintain strategy operation. Analyzing these two types of records can reveal whether the strategy is prone to failure in the face of disturbances, whether it requires frequent and complex interventions to maintain, and the cost of these interventions. For example, a strategy that frequently triggers alarms and requires manual intervention under slight disturbances has low adaptability and high maintenance costs.

[0109] Specifically, it provides a realistic assessment of the robustness and operational friendliness of a strategy, focusing on the strategy's long-term sustainability and total cost of ownership. This is crucial for determining whether a strategy is worth large-scale deployment. A high-performance but exceptionally fragile or difficult-to-maintain strategy may have a lower actual total value than a slightly less powerful but exceptionally robust strategy. It retrieves data on the strategy execution process, identifies and statistically analyzes the number, type, and severity of anomalous events related to strategy execution during the strategy's effective period, and identifies and statistically analyzes the frequency and complexity of adjustment operations performed to maintain strategy operation or correct deviations. It comprehensively assesses the optimization strategy's adaptability to environmental disturbances by considering the frequency and severity of anomalous events, and comprehensively assesses the maintenance cost of the optimization strategy by considering the frequency and complexity of adjustment operations.

[0110] S8.5. Based on the comprehensive absolute change range, relative change range, performance volatility and stability, as well as adaptability and maintenance costs, generate a multi-dimensional quantitative evaluation report on the effectiveness of the optimization strategy implementation.

[0111] Furthermore, a comprehensive and balanced evaluation framework is adopted to unify the value, risk, robustness, and cost of strategy effectiveness into a comprehensive evaluation conclusion. This avoids the one-sidedness of single-dimensional evaluation. While the strategy effectiveness is evaluated from different perspectives, decision-makers need a holistic conclusion. It is not simply a matter of listing these dimensions side by side, but rather aggregating them according to a predefined or adaptively generated comprehensive evaluation logic. This logic may need to weigh the importance of different dimensions. For example, for some key tasks, stability may be more important than absolute improvement; or, excessively high maintenance costs may offset the benefits of performance improvement. The generated multi-dimensional quantitative evaluation report will present the evaluation results of each dimension.

[0112] Specifically, it provides decision-makers with a comprehensive, three-dimensional, and ultimately comparable profile of the strategy's effectiveness. This report not only answers whether the strategy is effective but also details in which aspects it is effective, in which aspects it is insufficient, its overall value, and where the potential risks lie. This provides decision-makers with extremely rich and reliable decision-making basis for deciding whether to continue, adjust, or terminate the strategy, and how to select from multiple candidate strategies. It comprehensively summarizes and analyzes the absolute and relative changes in performance across various dimensions, the conclusions on performance volatility and stability, and the conclusions on adaptability and maintenance costs, according to a pre-set evaluation framework or scoring model, generating a structured document. This document is a multi-dimensional quantitative evaluation report on the effectiveness of the optimization strategy, which clearly describes the strategy's performance and overall evaluation across each evaluation dimension.

[0113] S9. By utilizing the causal influence network, mathematical model, and counterfactual reasoning process updated by the evaluation results, a feedback-driven adaptive optimization mechanism is obtained.

[0114] S9.1. Using a multi-dimensional quantitative evaluation report on the effectiveness of the optimization strategy, identify causal edges in the causal influence network that do not match actual observations.

[0115] Furthermore, the actual deployment effect of the optimization strategy is used as the most authoritative verification experiment to test and correct the causal theory previously derived from historical data, thereby initiating a learning cycle of model self-evolution. The causal influence network is essentially an assumption about the internal operating mechanism of the standard system. The multi-dimensional quantitative evaluation report provides actual result data after strategy intervention, which constitutes the gold standard for testing the causal hypothesis. Prediction-observation comparison is carried out: the prediction of the causal influence network on the effect of the strategy is compared in detail with the actual effect reflected in the evaluation report.

[0116] Specifically, for example, a causal influence network may predict that adjusting parameter A will significantly improve performance B through path X. However, the evaluation report shows that the actual improvement in performance B is far lower than predicted or not improved at all, or even decreases. This bias strongly suggests that the assumptions about the existence, direction, or effect strength of this causal edge from parameter A to performance B in the network may not match reality, and thus it is marked as a causal edge to be corrected.

[0117] S9.2. Based on the identified causal relationships to be corrected, update the effect strength parameters of the corresponding causal edges in the causal influence network, and add or delete causal edges to obtain the updated causal influence network.

[0118] Furthermore, different correction strategies are adopted based on the nature of the bias: for edges with the correct direction but inconsistent strength, their effect strength parameters are updated, for example, by refitting regression coefficients using actual observation data; for edges predicted to exist but not showing a clear causal effect in actual observations, they are deleted, as they may represent spurious associations or invalid historical relationships; for edges with strong observed effects but missing from the network, new causal edges need to be added, revealing important causal relationships that were not previously recognized. For example, if it is found that increasing training time has a much lower effect on reducing operational errors than the network predicts, the strength parameter of that edge is lowered; if it is found that environmental humidity affects product quality but is not included in the network, this edge is added.

[0119] Specifically, each implementation and evaluation of a strategy is a causal experiment, and its feedback is used to make the causal influence network more accurate and complete. This ensures that subsequent optimization modeling and counterfactual inference based on this network can be built on an increasingly reliable causal understanding. For each identified causal relationship to be corrected, if the causal relationship exists but the effect strength is inaccurate, the effect strength parameters of the causal edge are re-estimated using new observational data after the strategy implementation; if the causal relationship is proven to be unlikely to exist, the corresponding causal edge is deleted from the causal influence network; if the evaluation results strongly suggest the existence of a new, previously unidentified causal relationship, new causal edges are added to the causal influence network based on data analysis and their initial parameters are estimated, ultimately resulting in a corrected and updated causal influence network.

[0120] S9.3 Integrate the updated causal influence network into the mathematical model, update the objective function and constraint set in the mathematical model, and obtain the updated mathematical model.

[0121] Furthermore, when the causal influence network is updated, the mathematical model must be adjusted accordingly to maintain consistency. The fusion is direct and automatic: the changed effect strength parameters in the updated causal influence network are mapped to the corresponding coefficients or functional forms in the objective function of the mathematical model; the addition or deletion of causal edges will correspondingly lead to the expansion or simplification of the set of constraints in the mathematical model. For example, if a new edge from equipment maintenance frequency to product yield is added to the causal network and the effect strength is positive, then in the mathematical model, equipment maintenance frequency may be introduced as a new decision variable, and the product yield objective function will contain a new term related to this variable, ensuring that the worldview of the optimization engine evolves synchronously with the real world learned through practice.

[0122] Specifically, this makes optimization no longer based on a static model that may be outdated or erroneous, but on a model that can learn from experience and evolve dynamically, thereby continuously generating optimization strategies that are more in line with the current reality. Based on the latest causal relationships and effect strengths reflected in the updated causal influence network, the mathematical expression form and parameters of the objective function in the mathematical model are modified accordingly to accurately reflect the new causal understanding. At the same time, new decision variable constraints are added or outdated constraints are deleted according to changes in the causal network structure, thus obtaining an updated mathematical model that is logically completely consistent with the updated causal influence network.

[0123] S9.4 Input the performance stability and adaptability data from the multi-dimensional quantitative evaluation report on the effectiveness of the optimization strategy into the counterfactual reasoning process to calibrate the scenario confidence and effect estimation parameters in the counterfactual reasoning.

[0124] Furthermore, by leveraging the robustness of strategies demonstrated in actual execution, the quality of assumptions used in the counterfactual reasoning process for strategy generation can be corrected and optimized, thereby improving the practicality and reliability of future generated strategies. Counterfactual reasoning relies on simulating "if...then..." scenarios, and its accuracy is limited by the rationality of scenario construction and the accuracy of effect estimation. Performance stability and adaptability data in multidimensional quantitative evaluation reports provide first-hand evidence about the performance of strategies in real, complex, and dynamic environments. This evidence is fed back into the counterfactual reasoning process: if a certain type of strategy exhibits poor stability in practice, then when constructing similar scenarios or estimating similar effects, the counterfactual reasoning process can lower its scenario confidence level or apply a more conservative correction to the effect estimation. For example, if the evaluation report shows that the strategy generated under demand fluctuation scenarios has poor actual adaptability, then when simulating similar demand fluctuation scenarios in the future, the counterfactual reasoning can apply a discount factor to the effect estimate derived from that scenario or reduce the weight of that scenario in strategy selection.

[0125] Specifically, the feedback loop of "learning from practice" is introduced into the core stage of strategy generation: counterfactual reasoning. This makes the reasoning process no longer a purely theoretical deduction detached from the complexities of reality, but rather able to absorb the friction coefficients and uncertainties of the real world, thereby generating more grounded and resilient optimization strategies. In essence, this transforms the practical experience of strategies into the wisdom of generating better strategies. By analyzing data from multi-dimensional quantitative evaluation reports regarding the volatility of optimization strategy performance, its ability to cope with abnormal events, and the required maintenance costs, information reflecting the robustness of the strategy in the real environment is quantified into parameters that can be used to adjust the counterfactual reasoning process. For example, the confidence weight of simulated counterfactual scenarios that lead to strategy instability in reality can be reduced, or the prior parameters regarding uncertainty in the counterfactual effect estimation can be corrected, thereby calibrating the counterfactual reasoning process.

[0126] S9.5. The updated causal influence network, the updated mathematical model, and the calibrated counterfactual reasoning process are combined to obtain a feedback-driven adaptive optimization mechanism.

[0127] Furthermore, the updated causal influence network, updated mathematical model, and calibrated counterfactual reasoning process serve as the starting point for a new round of optimization iterations. When an optimization strategy needs to be generated again, the entire process will run based on these more accurate components that have absorbed the latest experience. For example, the next causal discovery will use a larger dataset that includes the latest implementation data; the next mathematical modeling will be based on a more accurate causal network; and the next counterfactual reasoning will use calibrated parameters. This forms a complete feedback-driven loop of strategy generation -> implementation -> evaluation -> model update -> strategy regeneration. This marks the transformation of the entire method from a static, one-off optimization tool into a dynamic, continuously evolving adaptive optimization mechanism. The mechanism can continuously improve its internal cognitive model and decision-making logic as the standard system evolves, the external environment changes, and its own experience accumulates. The self-iterative capability enables the optimization process to maintain high adaptability and effectiveness in the long term, truly achieving continuous improvement in the level of intelligence and continuous accumulation of knowledge assets.

[0128] Specifically, the updated causal influence network and the updated mathematical model, along with the calibrated counterfactual reasoning process, are integrated and encapsulated as a complete set of knowledge and decision-making components that incorporate the latest practical experience. When optimization needs are triggered again, the entire process will start with the use of this updated set of components, enabling the optimization capability to continuously adjust and improve itself based on feedback from the implementation of historical strategies. This cyclical closed-loop operation mode constitutes the feedback-driven adaptive optimization mechanism.

[0129] In summary, this invention collects and integrates multi-source data to form a structured dataset, constructs a causal influence network revealing the causal relationships between elements within a standard system using a causal discovery algorithm, establishes a mathematical model for a multi-objective optimization problem based on the network, and integrates personalized characteristic data of decision-makers. It then generates a highly adaptable set of personalized optimization strategies through counterfactual reasoning, verifies and dynamically adjusts these strategies in multiple scenarios using a simulation unit, deploys and executes the optimization strategies, and monitors them in real time. Based on the evaluation results of the execution, it updates the causal influence network, mathematical model, and reasoning process, forming a self-evolving adaptive optimization closed-loop mechanism. By combining causal inference with personalized counterfactual reasoning, this invention solves the problems of strategy failure due to reliance on statistical correlation in traditional methods, as well as the disconnect between optimization results and decision-makers' needs.

[0130] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A multi-objective optimization method for optimizing response strategies based on an artificial intelligence standard system, characterized by: This includes collecting multi-source data from a standard system and integrating the multi-source data into a structured dataset; Based on the structured dataset, a standard causal influence network is constructed using a causal discovery algorithm; By establishing a causal influence network, a mathematical model of a multi-objective optimization problem in a standard system is constructed. Combined with the personalized characteristic data of decision-makers, a set of personalized optimization strategies is generated through counterfactual reasoning. Based on the personalized optimization strategy set, each optimization strategy is simulated and verified in multiple scenarios through a standard system simulation unit. The optimization strategies are then dynamically adjusted based on the simulation verification results. The validated and adjusted optimization strategy is deployed to the actual standard system for execution, and the strategy execution process data is collected in real time through the monitoring unit; Based on data from the strategy execution process, the effectiveness of the optimization strategy is evaluated from multiple dimensions. The causal influence network, mathematical model, and counterfactual reasoning process updated by the evaluation results are used to obtain a feedback-driven adaptive optimization mechanism.

2. The multi-objective optimization method for optimizing response strategies based on an artificial intelligence standard system as described in claim 1, characterized in that: Collecting multi-source data from a standard system and integrating it into a structured dataset includes the following steps: collecting multi-source data from a standard document library, historical database, and monitoring equipment; performing data cleaning operations on the multi-source data of the standard system to remove missing and outlier values; performing format standardization and alignment operations on the cleaned multi-source data of the standard system to unify timestamps and measurement units; and storing the format-standardized and aligned multi-source data of the standard system according to a predefined pattern to generate a structured dataset.

3. The multi-objective optimization method for optimizing response strategies based on an artificial intelligence standard system as described in claim 2, characterized in that: Based on the structured dataset, a causal influence network of the standard system is constructed using a causal discovery algorithm, comprising the following steps: extracting observational data sequences corresponding to standard variables, environmental variables, and performance variables from the structured dataset; inputting the observational data sequences corresponding to standard variables, environmental variables, and performance variables into the causal discovery algorithm, using the PC algorithm to calculate the conditional independence between variables, and obtaining the conditional independence test results between variables; based on the conditional independence test results between variables, constructing a sketch of the causal influence network of the standard system representing the preliminary causal relationship between variables; performing directional inference and structural optimization on the sketch of the causal influence network of the standard system, determining the direction of causal edges and removing false edges; performing parameter learning on the causal influence network of the standard system after directional inference and structural optimization, estimating the effect strength of each causal edge, and obtaining the parameterized causal influence network of the standard system.

4. The multi-objective optimization method for optimizing response strategies based on an artificial intelligence standard system as described in claim 3, characterized in that: A mathematical model for a standard system of multi-objective optimization problems is established through causal influence networks. Combined with personalized characteristic data of decision-makers, a personalized set of optimization strategies is generated through counterfactual reasoning. This includes the following steps: extracting the decision-makers' implicit preferences for objective weights and risk tolerance from their historical interaction records to constitute their personalized characteristic data; defining a personalized utility function reflecting their preferences; integrating the personalized utility function with multiple objective functions in the mathematical model; within the decision variable space and constraint set defined by the mathematical model, searching and generating candidate optimization strategies guided by maximizing the personalized utility function that integrates the decision-makers' personalized characteristics; and for each... The system constructs a causal intervention scenario for implementing the candidate optimization strategies and a causal control scenario for maintaining the baseline state. Based on the objective function derived from the causal influence network in the mathematical model, the evolution trajectory of each performance objective is deduced under the causal intervention scenario and the causal control scenario for maintaining the baseline state. By comparing the differences in the terminal values ​​of the evolution trajectories, the causal effect of the candidate optimization strategies is quantified. Based on the causal effect of the candidate optimization strategies, and combined with the preference structure and risk propensity contained in the decision-maker's personalized characteristic data, the candidate optimization strategies are ranked in multiple dimensions and screened for contextual adaptability, generating a personalized set of optimization strategies that deeply matches the decision-maker's individual decision-making logic.

5. The multi-objective optimization method for optimizing response strategies based on an artificial intelligence standard system as described in claim 4, characterized in that: Based on the personalized optimization strategy set, each optimization strategy is simulated and verified in multiple scenarios using a standard system simulation unit. This includes the following steps: selecting the current optimization strategy to be verified from the personalized optimization strategy set and outputting it to the standard system simulation unit; constructing a baseline operating scenario based on the current environmental state within the standard system simulation unit; deriving verification scenarios with different disturbance conditions based on predefined rules within the baseline operating scenario; sequentially loading the optimization strategy into each verification scenario within the standard system simulation unit, driving simulation execution, and recording and outputting the simulation execution process data and terminal performance indicators of the optimization strategy in each verification scenario.

6. The multi-objective optimization method for optimizing response strategies based on an artificial intelligence standard system as described in claim 5, characterized in that: The optimization strategy is dynamically adjusted based on simulation verification results, including the following steps: Based on the simulation execution process data and terminal performance indicators of the optimization strategy in each verification scenario, the performance-vulnerable scenarios and parameter-sensitive intervals of the optimization strategy are identified, and corresponding strategy parameter compensation rules are formulated to generate a strategy parameter adjustment scheme; Based on the strategy parameter adjustment scheme, the adjustable parameters in the optimization strategy are modified to generate a modified optimization strategy; The modified optimization strategy is input into the standard system simulation unit, and the simulation is re-performed in the performance-vulnerable scenarios to obtain the simulation verification results of the modified optimization strategy; Based on the simulation verification results of the modified optimization strategy, it is confirmed that the performance of the optimization strategy meets the preset performance threshold, and the confirmed optimization strategy is used as the verified and adjusted optimization strategy.

7. The multi-objective optimization method for optimizing response strategies based on an artificial intelligence standard system as described in claim 6, characterized in that: Deploying the verified and adjusted optimization strategy to the actual standard system includes the following steps: converting the verified and adjusted optimization strategy into executable instructions or configuration changes in the actual standard system; and deploying the converted verified and adjusted optimization strategy in stages in the controlled environment of the actual standard system.

8. The multi-objective optimization method for optimizing response strategies based on an artificial intelligence standard system as described in claim 7, characterized in that: The monitoring unit collects strategy execution process data in real time, including the following steps: real-time collection of status data and performance data during the execution of the optimized strategy after verification and adjustment; time-stamp alignment and formatting of the status data and performance data to generate preliminary strategy execution process data; real-time association and labeling of the preliminary strategy execution process data with predefined expected strategy performance indicators to obtain a complete structure.

9. The multi-objective optimization method for optimizing response strategies based on an artificial intelligence standard system as described in claim 8, characterized in that: Based on strategy execution process data, a multi-dimensional evaluation of the implementation effect of the optimization strategy is conducted, including the following steps: Extracting the sequence of actual performance indicators directly related to each performance objective of the optimization strategy from the strategy execution process data; comparing the sequence of actual performance indicators directly related to each performance objective of the optimization strategy with historical benchmark performance data before the implementation of the optimization strategy, quantitatively describing the absolute and relative changes in performance of each dimension after the implementation of the optimization strategy; assessing the performance volatility and stability of the optimization strategy during actual implementation by combining the risk tolerance in the decision-maker's personalized characteristic data; analyzing abnormal events and adjustment operations recorded in the strategy execution process data to assess the optimization strategy's adaptability to environmental disturbances and maintenance costs; and generating a multi-dimensional quantitative evaluation report on the implementation effect of the optimization strategy by comprehensively considering the absolute and relative changes, performance volatility and stability, adaptability, and maintenance costs.

10. The multi-objective optimization method for optimizing response strategies based on an artificial intelligence standard system as described in claim 9, characterized in that: The feedback-driven adaptive optimization mechanism is obtained by using the updated causal influence network, mathematical model, and counterfactual reasoning process based on the evaluation results. This includes the following steps: using a multi-dimensional quantitative evaluation report on the effectiveness of the optimization strategy, identifying causal edges in the causal influence network that do not match actual observations; updating the effect strength parameters of the corresponding causal edges in the causal influence network based on the identified causal relationships to be corrected, and adding or deleting causal edges to obtain an updated causal influence network; integrating the updated causal influence network into the mathematical model, updating the objective function and constraint set in the mathematical model to obtain an updated mathematical model; inputting the performance stability and adaptability data from the multi-dimensional quantitative evaluation report on the effectiveness of the optimization strategy into the counterfactual reasoning process to calibrate the scenario confidence and effect estimation parameters in the counterfactual reasoning; and combining the updated causal influence network, the updated mathematical model, and the calibrated counterfactual reasoning process to obtain the feedback-driven adaptive optimization mechanism.

Images