Multi-system mutual feeding method based on causal driving progressive reasoning model

Abstract

The application discloses a multi-system mutual feedback method based on a causal driving progressive reasoning model, and belongs to the technical field of reasoning model application under a multi-system mutual feedback environment. The application solves the problems of poor scene application flexibility and large reasoning result error of a traditional multi-system mutual feedback method based on a reasoning model in the prior art. According to the analyzed scene description and environment analysis, the application identifies observation variables of reasoning system evolution. By introducing latent variables, combining the observation variables, constructing a structural equation model, and establishing the causal relationship between the variables, a neighbor matrix is constructed according to the causal relationship between the variables, the influence of the variables is calculated, the variables are sorted in descending order of influence, and the sorting is used as a reasoning order. According to the reasoning order, gradual reasoning is calculated, and the final dependent variable is analyzed and obtained. The application effectively improves the reasoning result accuracy under the multi-system mutual feedback environment, and can be applied to a complex causal scene between cross-domain multi-systems.

Description

Technical Field

[0001] This invention relates to a multi-system feedback method, and more particularly to a multi-system feedback method based on a causal-driven progressive inference model, belonging to the technical field of inference model applications in a multi-system feedback environment. Background Technology

[0002] With the acceleration of urbanization and the continuous expansion of urban scale, the connections between various systems within cities are becoming increasingly close and complex. Cities encompass multiple key systems such as transportation, energy, water resources, information, ecology, economy, and society. These systems are interdependent and mutually influential, forming a vast and complex urban mega-system. For example, the operation of the transportation system depends on energy supply, while traffic congestion affects energy consumption and the efficiency of economic activities. The water resource system is closely related to the ecosystem; the rational allocation of water resources affects the health of the ecosystem, while the ecosystem plays a crucial role in the purification and conservation of water resources.

[0003] In such a multi-system interactive urban environment, urban management and development face numerous challenges, and traditional management methods are insufficient to effectively address the complex interactions and dynamic changes between systems. To achieve efficient urban operation, sustainable development, and precise management decision-making, a technological means capable of handling complex relationships and uncertain information is needed. Inference models have emerged to address this need, with logical reasoning models, rule-based reasoning models, probabilistic reasoning models, and fuzzy reasoning models demonstrating unique advantages in multi-system interactive urban environments. The precision and interpretability of logical reasoning models help make accurate judgments under clearly defined rules and conditions; rule-based reasoning models can intuitively utilize domain expert knowledge for rapid decision-making; probabilistic reasoning models excel at handling uncertain information and integrating multi-source data to provide effective analysis of the dynamic changes in multi-system urban environments; and fuzzy reasoning models can effectively handle fuzzy concepts and uncertain situations, simulating human thinking for decision-making. These reasoning models provide strong support for the research and application of multi-system interactive urban environments, helping urban managers better understand and regulate the operation of urban systems, and improve the overall performance and competitiveness of cities.

[0004] The advantages and disadvantages of existing reasoning models combined with urban traffic management are as follows: (1) The logic-based reasoning model is a model built on the basis of various reasoning rules in logic. It uses methods such as deductive reasoning, inductive reasoning, and analogical reasoning to formally represent known propositions and facts. It derives necessary conclusions from given premises based on strict logical rules. It has the characteristics of certainty and precision and can clearly show the reasoning process and basis. In urban traffic management, the logic-based reasoning model can make reasoning decisions based on traffic regulations and real-time traffic conditions. For example, if it is detected that the vehicle speed on a certain road section is lower than a certain threshold and the vehicle density is higher than a certain standard (premise). Based on the logical reasoning model, it can be concluded that the road segment is congested, and then, according to the preset rules, it can be deduced that the green light time of the road segment should be extended to alleviate traffic pressure; however, the logical reasoning model is highly dependent on the premises, and the premises for obtaining accurate and definite conclusions must be accurate and complete, otherwise the reasoning results may be wrong. In practical applications, it is often difficult to obtain completely accurate and complete premises, and it lacks flexibility. Once the logical rules are determined, they are difficult to adapt to complex and uncertain actual situations, and the ability to process fuzzy and uncertain information is limited; (2) The rule-based reasoning model represents and stores domain knowledge and experience in the form of rules. Rules usually consist of two parts: conditions and conclusions. The reasoning process is to represent and store the domain knowledge and experience in the form of rules. Rule-based reasoning models match input facts or data with corresponding rules. If the conditions are met, the rules are triggered to arrive at the corresponding conclusion. Rule-based reasoning models are intuitive, easy to understand, and easy to implement, enabling rapid reasoning and judgment based on known conditions. For example, in an urban energy management system, the following rule can be set: If the industrial electricity consumption in a certain area exceeds the historical average by 20% within a specific time period (condition), then an investigation into the electricity consumption of high-energy-consuming enterprises in that area is initiated, and it is recommended that enterprises optimize their production schedules to stagger electricity consumption (conclusion). Through this rule-based reasoning process, the rational allocation and management of urban energy can be achieved. However, rule-based reasoning models require manual summarization and refinement of rules. For complex domains, the acquisition of rules can be challenging. The workload of collecting and sorting is large, and omissions or inconsistencies are likely to occur. Once the rules are determined, the rule-based reasoning model can only reason according to the established rules. It is difficult to automatically learn and update the rules according to new situations and data, and its adaptability is poor. (3) The probability-based reasoning model uses the principles of probability theory and statistics to model and reason about uncertain information and events. It represents uncertainty by calculating the probability of an event. Based on the known probability distribution and observation data, it uses Bayes' theorem and other methods to update the probability estimate of the event, thereby drawing the most likely conclusion. The probability-based reasoning model can handle uncertainty and incomplete information well and has strong reasoning ability in complex situations with multiple possibilities.In urban water resource management, based on probabilistic reasoning models, the probability distribution of water demand in different areas of the city in the next week can be predicted based on historical meteorological data, water consumption data, etc. For example, if the probability of rainfall in the next week is predicted to be 70%, the water demand of residents in the corresponding area may decrease by a certain percentage. Based on this, the water supply department can reasonably adjust the water supply plan and optimize the allocation of water resources. However, based on probabilistic reasoning models, a large amount of data is needed to estimate the probability distribution and parameters. The quality and quantity of data directly affect the accuracy of the reasoning results. If the data is insufficient or there is a deviation, the reasoning results may be inaccurate. When dealing with complex problems, the computational workload of probabilistic reasoning is large, especially when multiple variables and complex probability distributions are involved, the computational complexity increases exponentially, which may lead to low reasoning efficiency. (4) Fuzzy reasoning models are reasoning models based on fuzzy sets and fuzzy logic. They allow concepts and judgments to have fuzziness and uncertainty. Membership functions are used to describe the degree to which things belong to a certain fuzzy set. Through steps such as fuzzification, fuzzy rule reasoning and defuzzification, the fuzzy input information is transformed. Fuzzy reasoning transforms information into precise output results, simulating human thinking and decision-making processes in fuzzy and uncertain environments. In urban ecological environment assessment, fuzzy concepts such as "good air quality" and "high vegetation coverage" can be comprehensively evaluated using fuzzy reasoning models. For example, based on data such as the concentration of various pollutants in air quality monitoring and urban green area, a comprehensive evaluation result of the urban ecological environment can be determined using a fuzzy reasoning model. If the membership degree of air quality is 0.8 (indicating good quality) and the membership degree of vegetation coverage is 0.7 (indicating high coverage), fuzzy reasoning can lead to the conclusion that the overall urban ecological environment is at a good level, providing a basis for decision-making in urban ecological construction and environmental protection. However, the determination of the membership function in fuzzy reasoning models is often subjective; different people may give different membership functions based on different experiences and understandings, thus affecting the consistency and accuracy of the reasoning results. Compared to logic-based reasoning models, fuzzy reasoning has a relatively weaker logical foundation, and the interpretability of the reasoning process and results is not as clear as that of logical reasoning models, which may affect people's trust in the reasoning results.

[0005] Therefore, refer to Figure 2 Existing technical solutions are all typical single-step reasoning methods. For complex systems with multiple domains, a large model or rule-based method is used to perform unified calculations based on multiple variables to directly obtain the final calculation results, which has many problems.

[0006] In summary, a multi-system feedback method based on a causal-driven progressive inference model is needed. Summary of the Invention

[0007] A brief overview of the invention is given below to provide a basic understanding of certain aspects of it. It should be understood that this overview is not an exhaustive summary of the invention. It is not intended to identify key or essential parts of the invention, nor is it intended to limit the scope of the invention. Its purpose is merely to present certain concepts in a simplified form as a prelude to the more detailed description that follows.

[0008] In view of this, in order to solve the problems of poor application flexibility and large error in inference results of traditional multi-system feedback methods based on inference models in the prior art, this invention provides a multi-system feedback method based on a causal-driven progressive inference model.

[0009] The technical solution is as follows: a multi-system feedback method based on a causal-driven progressive inference model, comprising the following steps:

[0010] S1. Based on the scene description and environmental analysis, identify the observed variables of the inference system evolution;

[0011] S2. By introducing latent variables and combining them with observed variables, a structural equation model is constructed to establish causal relationships between variables;

[0012] S3. Based on the causal relationship between variables, construct an adjacency matrix, calculate the influence of the variables, sort them in descending order of influence, and use this as the order of inference;

[0013] S4. Calculate step by step according to the reasoning order, perform progressive reasoning, and analyze to obtain the final dependent variable.

[0014] Furthermore, step S1 includes the following steps:

[0015] S11. Collect multi-source data as variables;

[0016] S12. Conduct exploratory data analysis on multi-source data, that is, use descriptive statistical analysis and visualization tools to identify observed variables and observe the relationships between variables and data distribution;

[0017] In S11, the multi-source data includes road construction data, traffic flow data, and public transportation operation data obtained from the transportation database; air quality data and noise data obtained from the environmental protection database; regional GDP, industrial output value, and employment data obtained from the economic database; and social media data.

[0018] S13. Calculate the correlation coefficients between the variables, perform correlation analysis on the calculation results, and determine the variables that are highly correlated with the research objectives as the observed variables.

[0019] Furthermore, step S2 includes the following steps:

[0020] S21. Perform Granger causality tests on the observed variables, construct their respective regression models, and determine whether there is a cause between the observed variables;

[0021] S22. For observed variables with existing causes, introduce latent variables, construct a structural equation model, estimate it, and obtain the direct and indirect causal effects between variables;

[0022] S23. Analyze the relationship between observed variables and time, capture the time delay effect, and integrate the causal relationships between variables.

[0023] Furthermore, step S3 includes the following steps:

[0024] S31. Construct an adjacency matrix based on the causal relationships of the variables obtained in step S2;

[0025] Specifically: elements of the adjacency matrix Indicates the first variable For the second variable Does it have a direct impact? If so, then ,otherwise ;

[0026] S32. Based on the adjacency matrix, perform cluster analysis to divide the variables into different groups;

[0027] S33. Calculate the number of surrounding variables affected by each variable, i.e., the sum of the elements in each row of the adjacency matrix, to obtain the influence of the variable. Arrange the influences in descending order and use them as the order of reasoning.

[0028] Furthermore, step S4 includes the following steps:

[0029] S41. According to the reasoning order determined by the grouping and sorting, the variables are grouped, and the variables in each group are calculated step by step in turn;

[0030] S42. After completing the calculation of one set of variables, update the calculation results to the calculation of subsequent sets, and consider the information transfer between sets to determine the final dependent variable.

[0031] The beneficial effects of this invention are as follows: Unlike existing technologies, this invention does not use a single model to process several variables simultaneously and directly deduce a conclusion in one step. Instead, it takes into account the mutual influence between variables themselves, making it suitable for complex causal scenarios involving multiple systems across domains. This invention relies on data from the research scenario for analysis and uses a step-by-step analysis and reasoning scheme, which does not require professional adaptation, greatly improving the flexibility of scenario application. This invention identifies and analyzes variables based on raw data, eliminating the problem of subjective human factors affecting the accuracy of reasoning results. This invention decomposes complex causal problems into several finitely related causal problems, avoiding the difficulty in constructing reasoning models due to multiple dependencies within complex systems. It also limits the content of reasoning at each step, solving the problem of model reasoning failure caused by data confusion bias. This invention solves the problem of deep binding in rule setting, improving the flexibility of scenario application. This invention no longer relies primarily on manual summarization, solving the problem of multiple reasoning dependencies under complex causal conditions. This invention solves the problem of evolutionary model failure and inaccurate reasoning results caused by data confusion bias. This invention reduces the influence of human subjective consciousness on the accuracy of model reasoning results. Attached Figure Description

[0032] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0033] Figure 1 This is a flowchart illustrating a multi-system feedback method based on a causal-driven progressive inference model.

[0034] Figure 2 This is a flowchart illustrating a single-step reasoning method.

[0035] Figure 3 This is a schematic diagram of an embodiment of a multi-system feedback method based on a causal-driven progressive inference model. Detailed Implementation

[0036] To make the technical solutions and advantages of the embodiments of the present invention clearer, the exemplary embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not an exhaustive list of all embodiments. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0037] refer to Figures 1-3 This embodiment describes a multi-system feedback method based on a causal-driven progressive inference model, specifically including the following steps:

[0038] S1. Based on the scene description and environmental analysis, identify the observed variables of the inference system evolution;

[0039] S2. By introducing latent variables and combining them with observed variables, a structural equation model is constructed to establish causal relationships between variables;

[0040] S3. Based on the causal relationship between variables, construct an adjacency matrix, calculate the influence of the variables, sort them in descending order of influence, and use this as the order of inference;

[0041] S4. Calculate step by step according to the reasoning order, perform progressive reasoning, and analyze to obtain the final dependent variable;

[0042] Specifically, in step S1, variable identification not only considers directly related variables, but also needs to explore potential indirectly influencing variables. At the same time, based on the reasoning requirements and objectives, it is necessary to accurately identify intervention variables, other observed variables, and dependent variables. In addition, the nature, value range, and possible trend of each variable are preliminarily defined.

[0043] In step S2, during the process of establishing causal relationships, the complex characteristics of causal relationships, such as possible time delays and feedback mechanisms, are comprehensively considered to confirm the causal relationships between variables, including both qualitative relationships and specific quantitative relationships of dependent variables changing with independent variables.

[0044] In step S3, during the grouping and sorting process, the intervention variable is used as the core variable. The variables that directly affect it are grouped together, and all variables in the system are divided into several groups. They are sorted according to the number of variables that affect the surrounding variables to determine the inference order.

[0045] In step S4, calculations are performed step by step in a determined order. Groups less affected are analyzed first. During the reasoning process, the state and value of variables are continuously updated, and information transmission and interaction between different groups are considered until the final dependent variable is accurately analyzed.

[0046] Furthermore, step S1 includes the following steps:

[0047] S11. Collect multi-source data as variables;

[0048] S12. Conduct exploratory data analysis on multi-source data, that is, use descriptive statistical analysis (calculate mean, median, standard deviation, etc.) and visualization tools (draw line graphs, scatter plots, heat maps) to identify observed variables and observe the relationships between variables and data distribution;

[0049] In S11, the multi-source data includes road construction data (such as road length and number of lanes), traffic flow data (hourly traffic flow of each road segment), public transportation operation data (number of bus routes and operating hours) obtained from the transportation database; air quality data (concentration of pollutants such as PM2.5, PM10, and sulfur dioxide) and noise data obtained from the environmental protection database; regional GDP, output value of various industries, employment data, etc. obtained from the economic database; and social media data to understand residents' evaluation of transportation and the environment and their travel intentions.

[0050] S13. Calculate the correlation coefficients (such as Pearson correlation coefficients) between variables, perform correlation analysis on the calculation results, and determine the variables that are highly correlated with the research objectives as the observed variables.

[0051] Specifically, in this embodiment, daily traffic flow data for one month is collected. PM2.5 concentration Retail sales revenue in the same region on the same day Traffic flow was obtained by calculating the Pearson correlation coefficient from the data. With PM2.5 concentration The Pearson correlation coefficient between them;

[0052] Traffic flow With PM2.5 concentration Pearson correlation coefficient between Represented as:

[0053] ;

[0054] If traffic flow With PM2.5 concentration Pearson correlation coefficient between A value close to 1 or -1 indicates traffic flow. and PM2.5 concentration With a strong correlation, and referring to the above calculations, the traffic flow can be further calculated. Retail industry revenue Pearson correlation coefficient between If traffic flow is detected With PM2.5 concentration Retail industry revenue They are all strongly correlated and can be listed as observed variables;

[0055] Based on the research objectives, if the goal is to study how to improve the environment and promote economic development through transportation policies, the intervention variables can be determined as traffic control measures (such as the intensity of traffic restriction policies), the observed variables can be traffic flow and public transportation usage, and the dependent variables can be air quality indicators and economic growth indicators.

[0056] Regarding the nature of the variables, traffic flow is a continuous variable, and its value range, based on historical data, is between 0 and the maximum traffic flow that the road segment can carry. The intensity of the traffic restriction policy can be defined as a discrete variable, such as a value of 0 (indicating no control), 1 (indicating light control), 2 (indicating heavy control), etc. The trend of change can be preliminarily judged through time series analysis, such as calculating the moving average.

[0057] Furthermore, step S2 includes the following steps:

[0058] S21. Perform Granger causality tests on the observed variables, construct their respective regression models, and determine whether there is a cause between the observed variables. This is used to determine whether one time series variable is the cause of the change in another time series variable.

[0059] S22. For observed variables with existing causes, introduce latent variables, construct a structural equation model (SEM), estimate them, and obtain the direct and indirect causal effects between variables;

[0060] S23. By using methods such as the Autoregressive Integrated Moving Average (ARIMA) model, we can analyze the relationship between observed variables and time, capture the time delay effect, and integrate the causal relationships between variables.

[0061] Specifically, in this embodiment, a traffic flow regression model and a PM2.5 concentration regression model are constructed;

[0062] The traffic flow regression model is expressed as:

[0063] ;

[0064] in, This indicates the traffic flow at the current moment. This represents the traffic flow at time i in the past. This represents the regression coefficient of the i-th term of traffic flow itself. This represents the PM2.5 concentration at time i in the past. This represents the lag i-term influence coefficient of PM2.5 concentration on traffic flow. Represents the random error term;

[0065] The PM2.5 concentration regression model is expressed as follows:

[0066] ;

[0067] in, This represents the regression coefficient of the lagged i-th term of PM2.5 concentration. This represents the lag i-th term's influence coefficient on PM2.5 concentration, This represents the second random error term;

[0068] Null hypothesis Represented as traffic flow regression model This is not a Granger cause in the PM2.5 concentration regression model, i.e. The F-test (joint hypothesis test) is used to determine whether to reject the null hypothesis. If the null hypothesis is rejected Therefore, it is believed that the traffic flow regression model is the Granger cause of the PM2.5 concentration regression model;

[0069] Structural equation modeling can handle complex causal relationships between multiple variables simultaneously, taking into account latent variables and measurement errors. It assumes that traffic flow... Air quality affects air quality, and air quality in turn affects tourism revenue. For economic indicators, a structural equation model containing latent and observed variables is constructed.

[0070] The specific structure of the structural equation model is as follows:

[0071] Potential variable: Environmental quality (considered by PM2.5 concentration) Observed variables such as noise data (reflected together through a measurement model); Observed variable: Traffic flow PM2.5 concentration Noise data, tourism revenue ;

[0072] The causal path is as follows:

[0073] Direct effects: Traffic flow → PM2.5 concentration (direct impact); Environmental quality → Tourism revenue (indirect impact of potential variables on economic indicators);

[0074] Indirect effects: Traffic flow → PM2.5 concentration → Environmental quality → Tourism revenue (chain causality);

[0075] The measurement model (the relationship between latent variables and observed variables) is expressed as follows:

[0076] ;

[0077] The structural model (causal relationship between variables) is represented as follows:

[0078] ;

[0079] The structural equation model was fitted using the maximum likelihood estimation method to obtain specific effect values: traffic flow. For every 10% increase, PM2.5 concentration A 2% increase (direct effect); PM2.5 concentration For every 1% increase in environmental quality, the environmental quality decreases by 0.8 (standardized coefficient, calculated using a measurement model); for every unit decrease in environmental quality, tourism revenue... Reduced by 0.5% (indirect effect).

[0080] In this embodiment, by introducing the latent variable "environmental quality", the structural equation model can quantify the cross-level impact of traffic pollution on economic indicators (tourism revenue): increased traffic flow → deterioration of PM2.5 concentration → reduced environmental quality → decreased tourism attractiveness → reduced tourism revenue.

[0081] Furthermore, step S3 includes the following steps:

[0082] S31. Construct an adjacency matrix based on the causal relationships of the variables obtained in step S2;

[0083] Specifically: elements of the adjacency matrix Indicates the first variable For the second variable Does it have a direct impact? If so, then ,otherwise ;

[0084] S32. Based on the adjacency matrix, perform cluster analysis, that is, use methods such as hierarchical clustering and K-means clustering to divide the variables into different groups;

[0085] S33. Calculate the number of surrounding variables affected by each variable, i.e., the sum of the elements in each row of the adjacency matrix, to obtain the influence of the variable. Arrange the influences in descending order and use them as the order of reasoning.

[0086] Specifically, in this embodiment, it is based on five variables, namely traffic flow. Traffic restriction policy PM2.5 concentration Tourism revenue and public transportation usage Construct an adjacency matrix A;

[0087] The adjacency matrix A is represented as:

[0088] ;

[0089] Refer to adjacency matrix A, which represents traffic flow. Affecting PM2.5 concentration Traffic restriction policy Impact on traffic flow and public transportation usage PM2.5 concentration Impact on tourism revenue Calculate the sum of elements in each row of adjacency matrix A to obtain the traffic flow. The influence is 1, traffic restriction policy The influence is 2, PM2.5 concentration The influence is 1, and the tourism revenue is... and public transportation usage The influence is 0, with intervention variables (such as traffic restriction policies) With ) as the core, the traffic flow directly affected by it will be and public transportation usage Divide them into groups, sort them according to their influence, and the resulting order of reasoning is expressed as follows: .

[0090] Furthermore, step S4 includes the following steps:

[0091] S41. According to the reasoning order determined by the grouping and sorting, the variables are grouped, and the variables in each group are calculated step by step in turn;

[0092] S42. After completing the calculation of a set of variables, update the calculation results to the calculation of subsequent sets, and consider the information transfer between sets to determine the final dependent variable;

[0093] Specifically, let's assume the reasoning order is the first group (traffic restriction policy). Group 2 (Traffic Flow) PM2.5 concentration Group 3 (tourism revenue) Public transportation usage rate );

[0094] The step-by-step calculation process is as follows:

[0095] First, the first set of data is known. Using this first set of data and data directly inferred from it, the second set of data can be deduced, i.e., the known traffic restriction policy. Finally, the traffic flow was deduced. With PM2.5 concentration The reasoning here is performed in two steps;

[0096] Step 1: Calculate the traffic restriction policy based on historical data and causal relationship models (such as structural equation modeling). Intensity changes on traffic flow The direct impact, for example, traffic restriction policies. The intensity changes from 0 to 1, and the traffic flow is calculated using SEM. A decrease of 15%;

[0097] Step 2: Update the traffic flow The data was input into the air quality causal model to calculate the PM2.5 concentration. The changes.

[0098] For example, traffic flow A 15% decrease led to a decrease in PM2.5 concentration. Decrease of 5%;

[0099] Secondly, the third set of data is inferred from the first and second sets of data, i.e., the known traffic restriction policy. Traffic flow PM2.5 concentration Reasoning tourism revenue Public transportation usage rate The reasoning here is also performed in two steps;

[0100] The specific process of constructing the causal model of air quality is as follows:

[0101] (1) Variable selection

[0102] Dependent variable: PM2.5 concentration (target variable).

[0103] Independent variables: traffic flow, industrial emissions, wind speed, humidity, temperature, etc.

[0104] Potential variables (optional): such as "environmental governance intensity" reflected by observed variables such as policy intervention and environmental protection investment;

[0105] (2) Model structure design:

[0106] Based on prior knowledge or statistical tests (such as Granger causality tests), the direction of the causal relationship between variables is determined. The structure of the air quality causal model is exemplarily represented as follows:

[0107] ;

[0108] in, Traffic flow variables The regression coefficients, Industrial emissions variable The regression coefficients, Wind speed variable The regression coefficients, For the error term, For PM2.5 concentration variables;

[0109] (3) Parameter estimation and verification:

[0110] Fit the model using historical data (such as sensor monitoring data) and solve the coefficients using maximum likelihood estimation or Bayesian methods;

[0111] Verify the model's rationality (e.g., residual test, goodness of fit R²);

[0112] (4) Application and prediction:

[0113] Input new data (such as changes in traffic flow) to predict changes in PM2.5 concentration;

[0114] Step 3: Update the PM2.5 concentration Data Substitution for Tourism Revenue A causal model can be used to obtain tourism revenue. For example, a 5% decrease in PM2.5 concentration led to a 3% increase in tourism revenue;

[0115] The fourth step is to simultaneously implement traffic restriction policies. Intensity Substituted into Public Transportation Utilization Causal models can be used to calculate public transportation usage rates. Changes, such as traffic restriction policies The change in intensity from 0 to 1 directly led to a decrease in public transportation usage. Up 10%;

[0116] Through step-by-step calculations and information updates, the dependent variables (tourism revenue and public transportation usage rate) were finally identified.

[0117] In addition to the above steps, uncertainty analysis and reliability verification of the model inference results are also required to obtain more accurate dependent variable results based on causal relationships.

[0118] Although the invention has been described with reference to a limited number of embodiments, those skilled in the art will understand from the foregoing description that other embodiments are conceivable within the scope of the invention described herein. Furthermore, it should be noted that the language used in this specification has been chosen primarily for readability and instructional purposes, and not for the purpose of interpreting or limiting the subject matter of the invention. Therefore, many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the appended claims. The disclosure of the invention is illustrative and not restrictive, and the scope of the invention is defined by the appended claims.

Claims

1. A multi-system feedback method based on a causal-driven progressive inference model, characterized in that, Includes the following steps: S1. Based on the scene description and environmental analysis, identify the observed variables of the inference system evolution; S2. By introducing latent variables and combining them with observed variables, a structural equation model is constructed to establish causal relationships between variables; S3. Based on the causal relationship between variables, construct an adjacency matrix, calculate the influence of the variables, sort them in descending order of influence, and use this as the order of inference; S4. Calculate step by step according to the reasoning order, perform progressive reasoning, and analyze to obtain the final dependent variable; S1 includes the following steps: S11. Collect multi-source data as variables; S12. Conduct exploratory data analysis on multi-source data, that is, use descriptive statistical analysis and visualization tools to identify observed variables and observe the relationships between variables and data distribution; In S11, the multi-source data includes road construction data, traffic flow data, and public transportation operation data obtained from the transportation database; air quality data and noise data obtained from the environmental protection database; regional GDP, industrial output value, and employment data obtained from the economic database; and social media data. S13. Calculate the correlation coefficients between the variables, conduct correlation analysis on the calculation results, and determine the variables that are highly correlated with the research objectives as the observed variables; S2 includes the following steps: S21. Perform Granger causality tests on the observed variables, construct their respective regression models, and determine whether there is a cause between the observed variables; S22. For observed variables with existing causes, introduce latent variables, construct a structural equation model, estimate it, and obtain the direct and indirect causal effects between variables; S23. Analyze the relationship between the observed variables and time, capture the time delay effect, and integrate the causal relationship between the variables; S3 includes the following steps: S31. Construct an adjacency matrix based on the causal relationships of the variables obtained in step S2; Specifically: elements of the adjacency matrix Indicates the first variable For the second variable Does it have a direct impact? If so, then ,otherwise ; S32. Based on the adjacency matrix, perform cluster analysis to divide the variables into different groups; S33. Calculate the number of surrounding variables affected by each variable, i.e., the sum of the elements in each row of the adjacency matrix, to obtain the influence of the variable. Arrange the influences in descending order and use them as the order of reasoning.

2. The multi-system feedback method based on a causal-driven progressive inference model according to claim 1, characterized in that, S4 includes the following steps: S41. According to the reasoning order determined by the grouping and sorting, the variables are grouped, and the variables in each group are calculated step by step in turn; S42. After completing the calculation of one set of variables, update the calculation results to the calculation of subsequent sets, and consider the information transfer between sets to determine the final dependent variable.

Images