A pollution reduction and carbon reduction synergistic effect evaluation system and method fusing multi-agent and internet of things

Abstract

This application relates to the field of computer technology and discloses a system and method for evaluating the synergistic effect of pollution reduction and carbon reduction by integrating multi-agent and Internet of Things (IoT). The method includes: real-time acquisition and preprocessing of multi-source heterogeneous data through an IoT sensing layer; construction of a dynamic game model involving government, enterprises, and residents; sequential execution of physical common-source correlation analysis, policy coupling effect quantification, and multi-objective resource allocation optimization to generate evaluation indicators; dynamic calibration of model parameters based on measured data using a Bayesian optimization algorithm; and output of evaluation results and optimization strategies. The system includes a sensing layer, a preprocessing module, a multi-agent engine, a computing engine, a feedback optimization module, and an output interface. This application achieves real-time perception, accurate analysis, and closed-loop adaptive optimization of the synergistic effect of pollution reduction and carbon reduction through the deep integration of IoT and multi-agent game theory.

Description

Technical Field

[0001] This invention belongs to the field of computer technology, specifically relating to a system and method for evaluating the synergistic effect of pollution reduction and carbon reduction by integrating multi-agent and Internet of Things. Background Technology

[0002] With the intensification of global climate change and the increasing pressure on the ecological environment, synergistic effects in pollution and carbon reduction have become a core strategic direction for promoting green and low-carbon transformation and achieving sustainable development. The field of environmental governance and energy system regulation encompasses multiple key dimensions, including air pollutant emission control, greenhouse gas emission reduction, energy structure optimization, and guidance of socio-economic behavior. The fundamental challenge lies in overcoming the inherent limitations of traditional assessment systems in terms of data timeliness, logical completeness, and modeling dynamism, and constructing a quantifiable mechanism for synergistic effects that can be perceived in real time, accurately analyzed, and adaptively optimized.

[0003] Existing technologies for assessing the synergistic effects of pollution reduction and carbon reduction generally suffer from the following structural defects: First, data collection heavily relies on annual statistical reports, quarterly monitoring reports, or low-frequency manual inspections, lacking the ability to fuse multi-source heterogeneous real-time data based on the Internet of Things. This results in assessment results lagging significantly behind actual emission dynamics, failing to support hourly or even minute-level emergency response and fine-grained control. Second, the assessment logic has long been fragmented, treating air pollution prevention and carbon emission management as independent administrative tasks, ignoring the shared emission sources, common energy carriers, and interactive policy tools in typical scenarios such as industrial combustion, motor vehicle operation, and building heating. This leads to a systematic underestimation or misjudgment of synergistic potential. Third, modeling methods often employ static regression, unidirectional causal inference, or single-agent optimization frameworks, making it difficult to characterize the strategic game and behavioral feedback of diverse stakeholders such as governments, enterprises, and residents under fiscal constraints, market signals, and health risks. This results in simulation systems that fail to adequately reproduce the complexity of reality and generally lack closed-loop feedback and adaptive learning mechanisms. They cannot dynamically calibrate model parameters or generate rolling optimization strategies based on continuously flowing measured data, leading to a disconnect between assessment conclusions and governance practices.

[0004] To address the key bottlenecks such as data lag, logical fragmentation, static modeling, and lack of intelligent decision support, there is an urgent need for a novel synergistic effect evaluation method and system that deeply integrates real-time IoT sensing, multi-agent dynamic game theory, and Bayesian self-calibration mechanisms. This would break through the structural limitations of existing technologies and provide a scientific, dynamic, and closed-loop technical path for synergistic efficiency improvement in pollution reduction and carbon reduction. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a system and method for evaluating the synergistic effect of pollution reduction and carbon reduction by integrating multi-agent and Internet of Things.

[0006] To achieve the above objectives, the present invention provides the following technical solution: This application provides a pollution reduction and carbon reduction synergy assessment system that integrates multi-agent and Internet of Things technologies, including: The IoT sensing layer is used to collect multi-source heterogeneous raw data streams in real time through a sensor network deployed in various monitoring areas, and to perform time synchronization processing on the data from each sensor through an edge computing gateway. The data preprocessing module is used to perform timestamp alignment, outlier removal, adaptive window filtering, and feature normalization on the raw data stream to generate a structured time-series dataset. A multi-agent modeling engine is used to construct a dynamic game model that includes government agents, enterprise agents, and resident agents. The government agent aims to maximize total social welfare, the enterprise agent aims to maximize profits and is constrained by total emissions, and the resident agent aims to optimize the combination of travel utility and health risk. The synergy calculation engine is used to sequentially perform physical common source correlation analysis, policy coupling effect quantification, and multi-objective resource allocation optimization to generate pollution reduction and carbon reduction synergy evaluation indicators. The closed-loop feedback optimization module is used to dynamically calibrate the model parameters in the synergistic effect calculation engine based on measured data and using a Bayesian optimization algorithm. The strategy generation and output interface is used to output evaluation results and optimization strategies to the decision support terminal.

[0007] Furthermore, the IoT sensing layer adopts a layered deployment architecture, with different types of sensors configured in different monitoring areas, including: flue gas emission monitoring sensors in industrial source areas, air quality and traffic flow monitoring sensors at traffic nodes, energy consumption and heat monitoring sensors for energy facilities, and environmental and population activity monitoring sensors in residential areas; all data collected by the sensors are processed by time synchronization through an edge computing gateway to eliminate timing deviations caused by transmission delays.

[0008] Furthermore, when the data preprocessing module performs adaptive window filtering, it monitors the fluctuation level of the data within the window in real time; when the fluctuation level exceeds a preset threshold, the sliding window length is shortened from the first length to the second length to retain transient features; when the data returns to stability, the window length is gradually restored to the first length; outlier removal adopts an algorithm based on neighborhood statistical features, and interpolation is performed to fill in outlier data points.

[0009] Furthermore, in the multi-agent modeling engine, the travel utility function of the resident agent is constructed using a discrete choice model, and its expression is: in, Indicates the utility of residents' travel. Indicates commute time. This indicates the comfort index of the mode of transportation. Indicates the exposure concentration. Indicates the duration of exposure. , , The parameters are preset behavioral parameters; the utility function explicitly includes a pollutant exposure term to simulate residents' behavioral feedback under changes in environmental quality; each agent interacts with the other through a message queue to exchange states and actions, and updates the strategy according to a preset time step.

[0010] Furthermore, the physical co-source correlation analysis module in the synergistic effect calculation engine incorporates a material flow coupling operator. This operator establishes a nonlinear dynamic proportional relationship between pollutant generation and greenhouse gas emissions based on the stoichiometric relationship of the combustion reaction, specifically including: in, , These represent sulfur dioxide production and carbon dioxide emissions, respectively. Fuel consumption The sulfur content of the fuel, For desulfurization efficiency, , These are the ash content and moisture content, respectively. To fix the carbon content, For combustion efficiency, , These are stoichiometric constants; The operator dynamically corrects the correlation between the generation intensity of different emissions based on real-time monitoring of fuel characteristics and operating conditions, in order to identify the physical coupling characteristics of pollution reduction and carbon reduction. Secondly, this application provides a method for evaluating the synergistic effect of pollution reduction and carbon reduction by integrating multi-agent and Internet of Things technologies, including: Step 1: Collect multi-source heterogeneous data in real time covering industrial, transportation, energy and residential life scenarios through the Internet of Things sensing layer, and perform time synchronization processing on the data; Step 2: Perform anomaly detection, adaptive median filtering, and normalization on the collected raw data stream to form a structured time-series dataset with a unified time granularity; Step 3: Initialize the state space and action space of the three types of intelligent agents: government, enterprise, and resident. Set the objective functions of maximizing total social welfare, maximizing profit, and maximizing comprehensive utility as the objective functions of each intelligent agent. Step 4: Perform physical co-source correlation analysis, construct a nonlinear dynamic coupling operator for pollutants and greenhouse gas generation based on material flow analysis, and quantify the synergistic emission reduction potential of shared emission sources; Step 5: Quantify the policy coupling effect by using the difference-in-differences logic and synthetic control method to construct a counterfactual benchmark and separate the cross-effects caused by policy tools. Step 6: Perform multi-objective resource allocation optimization, solve the equilibrium solution of the multi-party game under the constraint of total emissions, and introduce an incentive compatibility constraint mechanism to generate an optimal allocation scheme; Step 7: Continuously receive measured data, calculate the model output error, and use a Bayesian optimization algorithm combined with a robust acquisition function to perform rolling calibration of the model parameters; Step 8: Push the synergy effect evaluation indicators and optimization strategies to the decision-making terminal.

[0011] Furthermore, the synthetic control method described in step five solves for the optimal weight vector so that the mean square error between the emission trajectory of the virtual control group, which is a weighted composite of multiple control regions, and the treatment group before the policy implementation is minimized. After the policy implementation, the synergistic emission reduction effect or offsetting effect of the policy is identified by calculating the difference between the actual trajectory of the treatment group and the counterfactual trajectory of the virtual control group, thereby eliminating the interference of confounding factors.

[0012] Furthermore, the multi-objective resource allocation optimization described in step six is ​​achieved by solving a master-slave game equilibrium: the government agent, as the leader, first releases a policy combination, and the enterprise and resident agents, as followers, adjust their production and travel behaviors accordingly; the incentive compatibility constraint mechanism requires that the marginal cost of emission reduction for enterprises is not higher than the marginal subsidy benefit provided by the government, and that the reduction in residents' health risks is not lower than the loss of travel utility, so as to ensure the feasibility of the optimization scheme.

[0013] Furthermore, in step seven, the robust acquisition function introduces a decay factor based on data confidence when calculating the improved value of Bayesian optimization. This decay factor is expressed as: in, As the attenuation factor, The variance of the current monitoring data. The sensitivity coefficient is used; the attenuation factor dynamically adjusts the contribution weight of parameter updates based on the data variance to suppress erroneous adjustments caused by sensor noise or faults; Bayesian optimization uses a Gaussian process as a surrogate model and iterates to bring the root mean square error between the model output and the measured value to within a preset threshold.

[0014] Furthermore, the strategy generation and output interface described in step eight supports multi-granularity output, including short-term emergency emission reduction instructions, regional energy dispatch strategies, and long-term policy effect evaluation reports; the interface is integrated with the city's smart management platform through standardized protocols to achieve real-time interaction and closed-loop control of evaluation indicators and governance instructions.

[0015] Compared with the prior art, this application has the following beneficial effects: This invention achieves high-frequency, multi-source data fusion through an IoT sensing layer and an adaptive data preprocessing module, significantly improving the timeliness of assessments. By constructing a dynamic game model involving the government, enterprises, and residents and introducing an incentive-compatible constraint mechanism, it greatly enhances the simulation system's ability to reproduce complex real-world behavioral interactions, significantly reducing model output errors. Through the nonlinear dynamic coupling operator built into the physical co-source correlation analysis, it accurately quantifies the collaborative generation relationships of different emissions from typical emission sources, avoiding resource misallocation. By combining a policy coupling effect quantification module with dual-difference and synthetic control methods, it effectively separates the interference of confounding factors, improving the accuracy of identifying policy cross-influences. By adopting a self-calibrating closed-loop feedback mechanism, the system can complete model parameter optimization within a preset time period, ensuring that the assessment conclusions continuously align with the actual operating state, achieving rolling optimization and continuous improvement of collaborative effectiveness. This comprehensively solves the technical problems mentioned in the background technology, such as data lag, logical fragmentation, static modeling, and lack of adaptive capabilities. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall technical architecture of the pollution reduction and carbon reduction synergistic effect evaluation method and system that integrates multi-agent and Internet of Things proposed in this invention. Figure 2 This is a schematic diagram of the core principle framework of the multi-agent dynamic game modeling and Bayesian self-calibration mechanism in this invention; Figure 3 This is a logical flowchart of the data preprocessing and multi-source heterogeneous time-series data fusion in this invention; Figure 4 This is a logical framework diagram for calculating the synergistic effect of physical co-origin correlation analysis, policy coupling effect quantification, and multi-objective resource allocation optimization in this invention; Figure 5 This is a schematic diagram of the multi-level interaction relationship and data flow between the IoT sensing layer, edge computing gateway and decision terminal in this invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Furthermore, in this invention, an element referred to as fixed to or disposed on another element may be directly disposed on the other element, or there may be an intermediate element. When an element is considered to be connected to another element, it may be directly connected to the other element, or there may be an intermediate element present simultaneously. The terms vertical, horizontal, left, right, and similar expressions used herein are for illustrative purposes only and do not represent the only possible implementation.

[0019] See Figures 1-5 This application provides a pollution reduction and carbon reduction synergy assessment system integrating multi-agent and Internet of Things, characterized by comprising: The IoT sensing layer is used to collect multi-source heterogeneous raw data streams in real time through a sensor network deployed in various monitoring areas, and to perform time synchronization processing on the data from each sensor through an edge computing gateway. The data preprocessing module is used to perform timestamp alignment, outlier removal, adaptive window filtering, and feature normalization on the raw data stream to generate a structured time-series dataset. A multi-agent modeling engine is used to construct a dynamic game model that includes government agents, enterprise agents, and resident agents. The government agent aims to maximize total social welfare, the enterprise agent aims to maximize profits and is constrained by total emissions, and the resident agent aims to optimize the combination of travel utility and health risk. The synergy calculation engine is used to sequentially perform physical common source correlation analysis, policy coupling effect quantification, and multi-objective resource allocation optimization to generate pollution reduction and carbon reduction synergy evaluation indicators. The closed-loop feedback optimization module is used to dynamically calibrate the model parameters in the synergistic effect calculation engine based on measured data and using a Bayesian optimization algorithm. The strategy generation and output interface is used to output evaluation results and optimization strategies to the decision support terminal.

[0020] In this embodiment, the overall composition of the system is defined, including an IoT sensing layer, a data preprocessing module, a multi-agent modeling engine, a synergistic effect calculation engine, a closed-loop feedback optimization module, and a strategy generation and output interface.

[0021] In practical implementation, the IoT sensing layer adopts a layered deployment architecture, deploying various types of sensors in monitoring areas such as industrial sources, traffic nodes, energy facilities, and residential areas. For example, in industrial source areas, continuous emission monitoring systems (CEMS) and infrared gas analyzers are deployed to collect real-time data on sulfur dioxide, nitrogen oxides, particulate matter concentrations, and carbon dioxide volume fraction; at traffic nodes, roadside air quality monitoring stations and vehicle detectors are deployed to acquire PM2.5, ozone concentration, and traffic flow data; at energy facilities, smart meters and heat metering devices are deployed to record electricity consumption, heating output, and return water temperature; and in residential areas, low-power environmental sensors and population activity density probes are deployed to collect temperature, humidity, noise, and population distribution data. All sensors aggregate raw data streams to the edge computing gateway via preset communication protocols such as LoRaWAN and 5G. The edge computing gateway has a built-in precision time protocol module (such as IEEE 1588) to align data packets from different sensors with nanosecond-level timestamps, eliminating timing misalignments caused by network transmission delays.

[0022] After receiving the time-aligned raw data stream, the data preprocessing module first performs alignment processing based on nanosecond-level timestamps to ensure that the multi-source data are on the same time reference. Then, it uses the Local Outlier Factor (LOF) algorithm to identify and remove outliers that deviate from the normal distribution range: a threshold is set for the number of neighborhood points and the multiple of the standard deviation. When the LOF score of a data point exceeds the threshold, it is judged as an outlier, removed, and then filled with linear interpolation. Subsequently, a sliding window mid-range filtering algorithm with an adaptive window length mechanism is used to smooth data fluctuations: the initial window length is set to a first preset length (e.g., 60 sampling points), and the rate of change of the standard deviation of the data within the window is monitored in real time; when emission step characteristics are detected (e.g., the sudden increase in the standard deviation rate caused by equipment start-up and shutdown exceeds the preset rate of change threshold), the window length is automatically shortened to a second preset length (e.g., 10 sampling points) to preserve transient characteristics; the window length is gradually restored after the data stabilizes. Finally, through feature normalization processes such as Min-Max normalization or Z-score normalization, a structured time series dataset with a uniform time granularity (e.g., 1 minute) is generated and stored in a time series database (e.g., InfluxDB).

[0023] A multi-agent modeling engine constructs a dynamic game model consisting of government agents, enterprise agents, and resident agents. The government agent's state space includes air quality index, carbon emission intensity, fiscal subsidy balance, and public opinion index; its action space is a combination vector of carbon tax rate, environmental subsidy amount, and off-peak production instructions. Its objective function is defined as maximizing total social welfare, specifically including resident comprehensive utility, fiscal balance (carbon tax revenue minus subsidy expenditure), and health risk loss. The enterprise agent's state space covers production line load rate, fuel inventory, pollution discharge permit holdings, and product market price; its action space includes production plan adjustments, clean technology investment, and carbon quota trading strategies; its objective function is maximizing net profit while meeting total emission constraints. The resident agent's state space includes commuting distance, real-time PM2.5 concentration, outdoor temperature, and public transportation waiting time; its action space includes travel mode selection (private car, public transport, cycling, etc.); its objective function is maximizing comprehensive travel utility including health risk factors. The three types of intelligent agents interact with state information and action instructions through message queues, and perform policy updates according to a preset time step (such as every 5 minutes).

[0024] The synergistic effect calculation engine executes three core sub-tasks sequentially. First, physical common-source correlation analysis: For typical shared emission sources such as combustion equipment, a built-in material flow coupling operator based on stoichiometry and combustion efficiency is employed. This operator derives the nonlinear dynamic proportional relationship between pollutant generation and greenhouse gas emissions based on real-time monitored fuel industry analysis data (moisture, ash, volatile matter, fixed carbon content), furnace temperature, and excess air coefficient, according to the combustion reaction equation. Second, policy coupling effect quantification: A counterfactual baseline is constructed using dual-difference logic and synthetic control method. The policy implementation area is selected as the treatment group, and multiple similar areas unaffected by the policy are selected as potential control groups. The optimal weight vector is solved through regression analysis, and a weighted virtual control group is synthesized, minimizing the mean square error between its pollutant emission trajectory before policy implementation and that of the treatment group. After policy implementation, the difference between the actual trajectory of the treatment group and the counterfactual trajectory of the virtual control group is calculated to separate the cross-effects caused by specific policy tools. Third, multi-objective resource allocation optimization: Under the hard constraints of meeting the regional total emission constraints and environmental quality targets, solve the equilibrium solution of the multi-party game among the government, enterprises and residents (such as Stackelberg-Nash equilibrium), and introduce an incentive compatibility constraint mechanism to ensure that the marginal costs and marginal benefits of each subject in the equilibrium state are consistent, and generate an optimal allocation scheme.

[0025] The closed-loop feedback optimization module continuously receives measured data from the IoT sensing layer and calculates the error index (such as root mean square error, RMSE) between the current model evaluation result and the measured value. A Bayesian optimization algorithm based on a robust acquisition function is employed, using a Gaussian process as a surrogate model. This automatically adjusts the weight coefficients in the synergistic effect calculation engine (such as the weights in total social welfare), model hyperparameters (such as SCM weight regularization parameters), and compensation coefficients in the material flow coupling operator (such as desulfurization efficiency compensation coefficients). The robust acquisition function introduces a decay factor based on data confidence when calculating the expected improvement value. This factor is dynamically adjusted according to the variance of the monitoring data in the current period: when the data variance increases, the decay factor reduces the contribution weight of that data point to parameter updates, thereby suppressing parameter misadjustments caused by sensor malfunctions or random noise. Iterative optimization continues until the root mean square error between the model output and the measured value converges to within a preset threshold range.

[0026] The strategy generation and output interface converts the final generated synergy evaluation indicators (such as the synergistic emission reduction potential index and policy cross-elasticity coefficient) and optimization strategies into visualized data, application programming interface (API) call requests, or equipment control instructions, and outputs them to the decision support terminal. Output formats include, but are not limited to, JSON-formatted API pushes, hourly emergency emission reduction recommendations, and monthly policy effectiveness evaluation reports, supporting decision-makers in dynamically adjusting governance measures.

[0027] In one specific implementation, the IoT sensing layer adopts a layered deployment architecture, with different types of sensors configured in different monitoring areas, including: flue gas emission monitoring sensors in industrial source areas, air quality and traffic flow monitoring sensors at traffic nodes, energy consumption and heat monitoring sensors for energy facilities, and environmental and population activity monitoring sensors in residential areas; all data collected by the sensors undergo time synchronization processing via an edge computing gateway to eliminate timing deviations caused by transmission delays.

[0028] In this embodiment, the layered deployment architecture of the IoT sensing layer is further defined.

[0029] In practical implementation, the sensor network adopts a layered deployment architecture: At industrial sources, continuous emission monitoring systems and infrared gas analyzers are deployed to acquire the concentrations of various pollutants (such as SO2, NOx, and particulate matter) and greenhouse gas fluxes (such as CO2 and CH4) at flue gas ducts and exhaust outlets; at traffic nodes, roadside air quality monitoring stations and vehicle detectors are deployed to acquire roadside pollutant concentrations (such as PM2.5, O3, and CO) and vehicle flow data (such as traffic volume, vehicle speed, and vehicle type classification); at energy facilities, smart meters and heat metering devices are deployed to acquire data on electricity consumption, heating output, and return water temperature; and at residential areas, environmental sensors (such as temperature, humidity, and noise sensors) and population activity density probes (such as Wi-Fi probes or infrared counting devices) are deployed to acquire local environmental parameters and population distribution data. Data collected by sensors at each level undergoes time synchronization processing through an edge computing gateway: the gateway has a built-in Precision Time Protocol (PTP) module that adds a unified nanosecond-level timestamp to each data packet, eliminating timing misalignments caused by network transmission delays and ensuring time consistency for subsequent multi-source data fusion.

[0030] In one specific implementation, when the data preprocessing module performs adaptive window filtering, it monitors the fluctuation level of the data within the window in real time; when the fluctuation level exceeds a preset threshold, the sliding window length is shortened from a first length to a second length to retain transient features; when the data returns to stability, the window length is gradually restored to the first length; outlier removal adopts an algorithm based on neighborhood statistical features, and interpolation is performed to fill in outlier data points.

[0031] In this embodiment, the adaptive sliding window mid-value filtering and outlier removal in the data preprocessing module are further defined.

[0032] In practical implementation, when the data preprocessing module performs sliding window mid-range filtering, it monitors the rate of change of the standard deviation of the data within the window in real time. The initial window length is set to a first preset length (e.g., 60 sampling points, corresponding to 1 hour of data). The standard deviation of the data within the window is continuously calculated, and its rate of change is tracked. When the rate of change of the standard deviation exceeds a preset threshold (e.g., 5% per minute), it is determined that the data exhibits emission abrupt changes (such as boiler start-up / shutdown, chemical plant start-up / shutdown, or sudden traffic accidents). At this time, the data preprocessing module automatically shortens the length of the sliding window from the first preset length to a second preset length (e.g., 10 sampling points, corresponding to 10 minutes) to quickly respond to and preserve the transient characteristics of the sudden event. When the data returns to a stable state (i.e., the rate of change of the standard deviation remains below another preset threshold for a certain period), the length of the sliding window gradually recovers from the second preset length to the first preset length. The recovery process can employ linear increment or exponential smoothing to avoid introducing new noise from abrupt changes in window length. During the outlier removal process, the data preprocessing module determines the outlier threshold based on the number of neighboring points (e.g., 7 neighboring points) and the standard deviation multiple threshold (e.g., 3.0 times the standard deviation). Linear interpolation is then performed on the data points that are identified as outliers, which means that the normal data points adjacent to the outlier are used for linear interpolation to ensure the continuity of the time series.

[0033] In one specific implementation, in the multi-agent modeling engine, the travel utility function of the resident agent is constructed using a discrete choice model, and its expression is: in, Indicates the utility of residents' travel. Indicates commute time. This indicates the comfort index of the mode of transportation. Indicates the exposure concentration. Indicates the duration of exposure. , , The parameters are preset behavioral parameters; the utility function explicitly includes a pollutant exposure term to simulate residents' behavioral feedback under changes in environmental quality; each agent interacts with the other through a message queue to exchange states and actions, and updates the strategy according to a preset time step.

[0034] In this embodiment, the travel utility function of the resident agents and the interaction mechanism between agents in the multi-agent modeling engine are further defined. In specific implementation, when constructing resident agents, the multi-agent modeling engine introduces a travel utility function based on a discrete choice model, the expression of which is consistent with the disclosure document: in, Indicates the overall travel utility of residents. This indicates commute time (in minutes). This represents the comfort index of different modes of transportation (different preset values ​​correspond to different modes of transportation, such as 0.8 for private cars, 0.5 for public transportation, and 0.6 for cycling). This represents the average PM2.5 concentration along the route (in μg / m³). Indicates exposure duration (in hours). , , These represent preset behavioral parameters (calibrated using historical travel data, for example...). =2.5、 =1.2、 =0.8). This utility function explicitly includes a pollutant exposure concentration term. With exposure duration By calculating commuting time costs, comfort indices, and health risk losses due to pollutant exposure under different travel modes, the system simulates residents' behavioral feedback logic under specific weather conditions and environmental quality. For example, when air quality deteriorates, leading to increased PM2.5 concentrations along the route, the health risk loss component in residents' travel utility increases, thus automatically favoring travel modes with lower exposure risks (such as switching from cycling to public transportation). The multi-agent modeling engine uses message queues (such as RabbitMQ or Kafka) to facilitate the exchange of state information and action commands between government agents, enterprise agents, and resident agents, and executes policy updates according to preset time steps (such as 5 minutes), ensuring that the decisions of each agent can respond in real time to environmental changes and policy adjustments by other stakeholders.

[0035] In one specific implementation, the physical co-source correlation analysis module in the synergistic effect calculation engine incorporates a material flow coupling operator. This operator establishes a nonlinear dynamic proportional relationship between pollutant generation and greenhouse gas emissions based on the stoichiometric relationship of the combustion reaction, specifically including: in, , These represent sulfur dioxide production and carbon dioxide emissions, respectively. Fuel consumption The sulfur content of the fuel, For desulfurization efficiency, , These are the ash content and moisture content, respectively. To fix the carbon content, For combustion efficiency, , These are stoichiometric constants; The operator dynamically corrects the correlation between the generation intensity of different emissions based on real-time monitoring of fuel characteristics and operating conditions, in order to identify the physical coupling characteristics of pollution reduction and carbon reduction.

[0036] In this embodiment, the physical co-origin correlation analysis module and the material flow coupling operator in the synergistic effect calculation engine are further defined.

[0037] In practical implementation, the physical co-source correlation analysis module in the synergistic effect calculation engine incorporates a built-in mass flow coupling operator for combustion equipment (such as coal-fired boilers and gas turbines). This operator first acquires industrial analysis data of the fuel, including moisture content. Ash content Volatile matter content and fixed carbon content Furthermore, by combining the real-time operating temperature and excess air coefficient of the combustion equipment, a nonlinear dynamic proportional relationship between pollutant generation and greenhouse gas emissions is derived based on the combustion reaction equation. Specifically, for typical coal-fired equipment, the operator uses the following two formulas: in, This indicates the amount of sulfur dioxide generated (in kg / h). This indicates carbon dioxide emissions (unit: kg / h). This indicates the amount of coal consumed (in kg / h). This indicates the sulfur content (mass fraction, %) in the fuel. This indicates the desulfurization efficiency (value range 0~1). This indicates the ash content (mass fraction, %). Moisture content (mass fraction, %) This indicates the fixed carbon content (mass fraction, %). Indicates combustion efficiency (value range 0~1), , They represent stoichiometric constants (e.g.) =2, =3.67). The material flow coupling operator dynamically corrects the correlation between the generation intensity of different emissions based on real-time monitored changes in fuel type (such as coal type switching) and fluctuations in combustion efficiency (such as changes in combustion efficiency caused by changes in oxygen content). For example, when the sulfur content of the coal entering the furnace increases, This increases the risk of energy consumption, and if the amount of desulfurizing agent injected is increased to ensure compliance, it may lead to increased energy consumption and thus affect performance. The operator can capture this coupling relationship in real time, thereby identifying the physical coupling characteristics of pollution reduction and carbon reduction in the production process.

[0038] Secondly, this application provides a method for evaluating the synergistic effect of pollution reduction and carbon reduction by integrating multi-agent and Internet of Things technologies, including: Step 1: Collect multi-source heterogeneous data in real time covering industrial, transportation, energy and residential life scenarios through the Internet of Things sensing layer, and perform time synchronization processing on the data.

[0039] In this embodiment, multi-source heterogeneous environmental and socio-economic data covering industrial, transportation, energy, and residential scenarios are collected in real time through the IoT sensing layer. The data includes pollutant concentrations (SO2, NOx, PM2.5, etc.), greenhouse gas fluxes (CO2, CH4, etc.), energy consumption indicators (electricity and heat consumption), meteorological parameters (wind speed, wind direction, temperature, humidity), traffic flow (vehicle volume and speed), and population density indicators. During the data collection process, the edge computing gateway performs time synchronization processing based on a precision time protocol on data from different sources, adding a unified timestamp to each data packet.

[0040] Step 2: Perform anomaly detection, adaptive median filtering, and normalization on the collected raw data stream to form a structured time-series dataset with a unified time granularity; In this embodiment, an anomaly detection module is used to perform anomaly detection on the acquired raw data stream, identifying and removing outliers that exceed the logical range (e.g., using the LOF algorithm or physical model-based verification). A median filtering algorithm with adaptively adjusted window length is used to smooth the data: the initial window length is set to a first preset length, and the degree of data fluctuation is monitored in real time. When a sudden change is detected, the window length is automatically shortened to preserve step characteristics, and then gradually restored after stabilization. Finally, feature normalization processing (e.g., Min-Max normalization) is performed to form a structured time-series dataset with a uniform time granularity (e.g., 1 minute).

[0041] Step 3: Initialize the state space and action space of the three types of intelligent agents: government, enterprise, and resident. Set the objective functions of maximizing total social welfare, maximizing profit, and maximizing comprehensive utility as the objective functions of each intelligent agent. In this embodiment, the state space and action space of three types of agents—government, enterprise, and resident—are initialized by a multi-agent modeling engine. The objective function of the government agent is set as maximizing total social welfare, which can be specifically expressed as: The meanings of the symbols are as described above. The objective function of the enterprise agent is set as maximizing net profit while satisfying emission constraints, i.e. =Income Production costs Emissions reduction costs Carbon tax expenditures plus subsidy revenues are constrained by the condition that total pollutant emissions do not exceed the approved limits. The objective function for the resident intelligent agent is set as maximizing the comprehensive travel utility, including health risk factors.

[0042] Step 4: Perform physical co-source correlation analysis, construct a nonlinear dynamic coupling operator for pollutants and greenhouse gas generation based on material flow analysis, and quantify the synergistic emission reduction potential of shared emission sources; In this embodiment, a synergistic effect calculation engine is used to perform physical co-source correlation analysis. Based on the material flow analysis logic, a nonlinear dynamic coupling operator for pollutant and greenhouse gas generation is constructed for the combustion process or chemical reaction stage. The synergistic emission reduction potential of shared emission sources (such as coal-fired boilers and motor vehicle engines) under different operating conditions (load rate, fuel type, temperature, etc.) is quantified, and the synergistic generation intensity coefficient of each emission source is output.

[0043] Step 5: Quantify the policy coupling effect by using dual difference logic and synthetic control method to construct a counterfactual benchmark and separate the cross-effects caused by policy tools.

[0044] In this embodiment, the policy coupling effect is quantified using a synergistic effect calculation engine. A difference-in-differences logic is employed to compare the emission changes before and after policy implementation between the policy implementation area and the control area. A synthetic control method is introduced, constructing a counterfactual benchmark by weighting historical data from multiple unaffected areas. Specifically, the optimal weight vector is solved to minimize the mean square error between the pollutant emission trajectory of the virtual control group (weighted from multiple control areas) before policy implementation and the treatment group. After policy implementation, the difference between the actual emission trajectory of the treatment group and the counterfactual trajectory of the virtual control group is calculated. This difference represents the cross-effect (synergistic emission reduction effect or offsetting effect) caused by the specific policy tool, thus eliminating the influence of confounding factors such as economic fluctuations and meteorological interference on the assessment results.

[0045] Step 6: Perform multi-objective resource allocation optimization, solve the equilibrium solution of the multi-party game under the constraint of total emissions, and introduce an incentive compatibility constraint mechanism to generate an optimal allocation scheme; In this embodiment, a synergy effect calculation engine is used to perform multi-objective resource allocation optimization. Under the hard constraints of regional total emission limits (e.g., annual SO2 emissions not exceeding 5,000 tons) and environmental quality targets (e.g., AQI compliance rate), the equilibrium solution of the multi-party game involving the government, enterprises, and residents is sought. This equilibrium solution can be achieved by solving a master-follower game (Stackelberg game): the government, as the leader, first releases a policy combination vector including carbon tax and subsidies, and enterprises and residents, as followers, adjust their production load and travel behavior according to this policy combination vector. The optimization process introduces an incentive compatibility constraint mechanism to ensure that the marginal costs and marginal benefits of each subject are consistent in the equilibrium state. Specifically, this includes requiring that the marginal cost of enterprises implementing emission reduction actions is not higher than the marginal subsidy benefits provided by the government, and requiring that the reduction in health risks caused by changes in residents' behavior is sufficient to compensate for the loss of their travel utility, thereby generating feasible optimized allocation schemes (e.g., monthly production load allocation for each enterprise, allocation of government subsidy funds, adjustment of public transportation schedules, etc.).

[0046] Step 7: Continuously receive measured data, calculate the model output error, and use a Bayesian optimization algorithm combined with a robust acquisition function to perform rolling calibration of the model parameters; In this embodiment, a closed-loop feedback optimization module continuously receives monitoring data and calculates the error index (such as root mean square error RMSE) between the current model evaluation result and the measured value. A Bayesian optimization algorithm combined with a robust acquisition function is used to automatically adjust parameters in physical co-source analysis (such as the desulfurization efficiency compensation coefficient), weights in policy coupling quantification (such as SCM weight regularization parameters), and objective function weights in resource allocation optimization (such as health risk weights in total social welfare). This enables rolling calibration of model parameters.

[0047] Step 8: Push the synergy effect evaluation indicators and optimization strategies to the decision-making terminal.

[0048] In this embodiment, the final generated synergistic effect evaluation indicators, emergency emission reduction suggestions, and energy dispatch instructions are pushed to the decision-making terminal through a strategy generation and output interface. The output format provides multi-granular results based on decision-making needs, including short-term emergency emission reduction instructions for sudden pollution events (e.g., "increase the amount of ammonia injected into the denitrification system by 15% within the next 2 hours"), load dispatch strategies for regional energy systems (e.g., output adjustment plans for combined heat and power units), and policy effect evaluation reports for long-term planning (e.g., monthly or quarterly reports). The interface integrates with the city's smart management platform through standardized data exchange protocols (e.g., RESTful API) to achieve real-time interaction and closed-loop control of evaluation indicators and governance instructions.

[0049] In one specific implementation, the synthetic control method described in step five solves for the optimal weight vector so that the mean square error between the emission trajectory of the virtual control group, which is a weighted composite of multiple control regions, and the treatment group before the policy is implemented is minimized. After the policy is implemented, the synergistic emission reduction effect or offsetting effect of the policy is identified by calculating the difference between the actual trajectory of the treatment group and the counterfactual trajectory of the virtual control group, thereby eliminating the interference of confounding factors.

[0050] In this embodiment, the synthetic control method solves for the optimal weight vector by performing regression analysis (such as Lasso regression or ordinary least squares regression). ,in Number of control areas and This weight vector minimizes the mean square error between the pollutant emission trajectory of the virtual control group (composed of multiple control regions) before policy implementation and the treatment group, i.e.: in, For the period prior to the policy implementation, For the processing group during the period The emission values, For the first The control region in the period The emission values. After the policy was implemented (during the period) to The difference between the actual emission trajectory of the treatment group and the counterfactual trajectory of the virtual control group was calculated. Identify the synergistic emission reduction effects of policies on non-target emissions (if...) (Indicates synergistic emission reduction) or offsetting effect (if) (This indicates coordinated emission increases). This method effectively eliminates the influence of confounding factors such as economic fluctuations and weather conditions on the assessment results.

[0051] In one specific implementation, the multi-objective resource allocation optimization in step six is ​​achieved by solving a master-slave game equilibrium: the government agent, as the leader, first releases a policy combination, and the enterprise and resident agents, as followers, adjust their production and travel behaviors accordingly; the incentive compatibility constraint mechanism requires that the marginal cost of emission reduction for enterprises is not higher than the marginal subsidy benefit provided by the government, and that the reduction in residents' health risks is not lower than the loss of travel utility, so as to ensure the feasibility of the optimization scheme.

[0052] In this embodiment, multi-objective resource allocation optimization is achieved by solving a master-slave game (Stackelberg game) equilibrium. The government agent, acting as the leader, releases a policy combination vector in the first phase of the game. ,in For carbon tax rates, This includes considerations such as environmental subsidies. Enterprises and residents, as followers, observe government policies and, in the second stage, adjust their production load and travel behavior based on the policy combination vector to maximize their own utility. The equilibrium solution of the game is obtained through backward induction: first, the optimal reaction function of the followers under a given policy is solved; then, the leader selects the optimal policy based on the followers' reaction functions. The incentive compatibility constraint mechanism requires that the marginal cost of enterprises implementing emission reduction actions be considered. Not higher than the marginal subsidy income provided by the government ,Right now At the same time, it requires residents to reduce their health risks due to behavioral changes. This is sufficient to compensate for the loss of travel utility. ,Right now These constraints ensure that the resource allocation scheme is acceptable to all parties and can be implemented stably in reality, avoiding the problem of being "mathematically optimal but practically infeasible".

[0053] In one specific implementation, the robust acquisition function in step seven introduces a decay factor based on data confidence when calculating the improved value of Bayesian optimization. The decay factor is expressed as: in, As the attenuation factor, The variance of the current monitoring data. The sensitivity coefficient is used; the attenuation factor dynamically adjusts the contribution weight of parameter updates based on the data variance to suppress erroneous adjustments caused by sensor noise or faults; Bayesian optimization uses a Gaussian process as a surrogate model and iterates to bring the root mean square error between the model output and the measured value to within a preset threshold.

[0054] In practical implementation, the Bayesian optimization algorithm in the closed-loop feedback optimization module uses a robust acquisition function to calculate the Expected Improvement (EI). This acquisition function introduces a data confidence attenuation factor based on the traditional EI, and its expression is consistent with the disclosure document: in, Indicates the attenuation factor. This represents the variance of the monitoring data for the current time period. The sensitivity coefficient (preset as a positive number, such as...) The attenuation factor is dynamically adjusted based on the variance of the monitoring data for the current period: when the variance of the monitoring data increases (e.g., due to sensor dampness or communication interference causing drastic fluctuations in readings), The attenuation factor is reduced, thus decreasing the contribution weight of the data point to the model parameter update; conversely, when the data variance is small, the attenuation factor is close to 1, and the data point participates in the parameter update normally. This mechanism effectively suppresses parameter misadjustment caused by sensor failure or random noise. The Bayesian optimization algorithm uses a Gaussian process as a surrogate model. Through iterative optimization (each iteration collects new measured data points, updates the Gaussian process posterior, and calculates the parameter combination corresponding to the maximum value of the collection function), the root mean square error (RMSE) between the model output and the measured value converges to within a preset threshold range (e.g., 0.05), completing the rolling calibration of the model parameters.

[0055] In one specific implementation, the strategy generation and output interface in step eight supports multi-granularity output, including short-term emergency emission reduction instructions, regional energy dispatch strategies, and long-term policy effect evaluation reports; the interface is integrated with the city's smart management platform through a standardized protocol to achieve real-time interaction and closed-loop control of evaluation indicators and governance instructions.

[0056] In practical implementation, the strategy generation and output interface provides multi-granular output results based on decision-making needs. Specifically, this includes: short-term emergency emission reduction instructions for sudden pollution events (such as a red alert for heavy pollution), such as "increase the amount of ammonia injected into denitrification systems by 15% within the next two hours" or "implement odd-even license plate restrictions for motor vehicles"; load dispatch strategies for regional energy systems, such as "reduce the output of combined heat and power units by 10%, supplementing heating with gas-fired boilers" or "shift electric vehicle charging load to off-peak hours at night"; and policy effectiveness evaluation reports for long-term plans, such as monthly or quarterly pollution reduction and carbon reduction synergy evaluation reports, including trends in the synergistic emission reduction potential index and analysis of policy cross-elasticity coefficients. The strategy generation and output interface integrates with the city's smart management platform through standardized data exchange protocols (such as RESTful API, MQTT, or OPCUA) to achieve real-time interaction and closed-loop control of evaluation indicators and governance instructions. For example, evaluation indicators can be directly pushed to the decision support system of the smart city brain, and governance instructions can be automatically issued to traffic signal control systems, power plant distributed control systems, or mobile travel apps.

[0057] Example 2 Based on Example 1, this example uses a typical industrial park as the application scenario, which includes a coal-fired combined heat and power unit, a chemical production line, a centralized heating network, and supporting staff living quarters. The specific application process of the present invention will be described in detail below with reference to this scenario.

[0058] First, a layered deployment architecture is adopted to build the IoT sensing layer within the park. A Continuous Emission Monitoring System (CEMS) is deployed in the flue gas ducts of coal-fired boilers to collect real-time data on sulfur dioxide, nitrogen oxides, particulate matter concentrations, and carbon dioxide volume fraction, with a sampling frequency of once per minute. Infrared gas analyzers are installed at the exhaust outlets of chemical reactors to monitor volatile organic compounds (VOCs) and methane flux, with a sampling frequency of once every 5 minutes. Roadside air quality monitoring stations are set up at intersections of the park's main roads, integrating PM2.5 and ozone sensors and microwave vehicle detectors to simultaneously acquire pollutant concentration and heavy truck traffic data, with a time granularity of 10 seconds. Intelligent heat metering devices are installed at the outlets of cogeneration units and the entrances of each building to record heat supply and return water temperature, with a sampling cycle of 15 minutes. Low-power environmental sensor nodes are deployed in the staff dormitory area to collect temperature, humidity, noise, and population activity density probe data, with a reporting interval of 30 minutes. All sensors transmit raw data streams to the edge computing gateway deployed in the substation via the LoRaWAN protocol. The gateway has a built-in time synchronization module that uses the IEEE 1588 precision time protocol to align data packets from different sensors with nanosecond-level timestamps, eliminating timing misalignments caused by network latency.

[0059] After receiving the time-aligned raw data stream, the data preprocessing module first performs outlier removal. For pollutant concentration sequences, a local outlier (LOF) algorithm based on a sliding window is used, with a neighborhood number of 7 and a standard deviation multiple threshold of 3.0. When the LOF score of a data point exceeds the threshold, it is determined to be a sensor malfunction or transient interference, and is removed and filled with linear interpolation. Subsequently, adaptive sliding window mean filtering is performed: the initial window length is set to 60 sampling points (corresponding to 1 hour), and the standard deviation of the data within the window is calculated in real time; when a sudden increase in the standard deviation is detected to exceed the 5% change rate threshold per minute (e.g., a step change in emissions caused by boiler start-up and shutdown), the window length is automatically shortened to 10 sampling points (10 minutes) to preserve the transient characteristics of sudden events; after the data stabilizes, the window length is gradually restored to 60 points. Finally, all features are normalized using Min-Max to map pollutant concentrations to the [0,1] interval, and energy consumption data is scaled according to the historical maximum value to generate a structured time series dataset with a uniform time granularity of 1 minute, which is then stored in the time series database InfluxDB.

[0060] The multi-agent modeling engine initializes three types of agents based on the aforementioned structured dataset. The state space of the government agent includes the park's real-time air quality index (AQI), carbon emission intensity, remaining fiscal subsidies, and public opinion index; its action space is a combination vector of carbon tax rate, environmental protection subsidy amount, and off-peak production instructions; its objective function is defined as maximizing total social welfare. ,in For the overall benefit of residents, For carbon tax revenue, To subsidize expenditures, For health risk losses, , , The enterprise agent's state space includes production line load rate, fuel inventory, pollution discharge permit holdings, and product market price; its action space includes production plan adjustments, clean technology investment, and carbon quota trading strategies; its objective function is to maximize net profit, with the constraint that total pollutant emissions do not exceed the limits approved by the environmental protection department. The resident agent's state space includes commuting distance, real-time PM2.5 concentration, outdoor temperature, and public transportation waiting time; its action space is travel mode selection (private car, public transport, cycling); its utility function explicitly incorporates a pollutant exposure term. ,in For commuting time, The comfort index of travel mode The average PM2.5 concentration along the route. For exposure duration, , , These are behavioral parameters; the three types of intelligent agents interact with state information and action instructions through message queues, and the policy is updated every 5 minutes.

[0061] The synergistic effect calculation engine executes three core sub-tasks sequentially. First, physical co-source correlation analysis: For the material flow coupling operator built into the coal-fired boiler based on stoichiometry and combustion efficiency, it dynamically calculates the theoretical air volume and actual excess air coefficient based on real-time coal quality industrial analysis data (moisture, ash, volatile matter, fixed carbon content) and furnace temperature fed back by CEMS, and derives the nonlinear proportional relationship between sulfur dioxide generation and carbon dioxide emissions based on the combustion reaction equation. ,in, , These represent sulfur dioxide production and carbon dioxide emissions, respectively. Fuel consumption The sulfur content of the fuel, For desulfurization efficiency, , These are the ash content and moisture content, respectively. To fix the carbon content, For combustion efficiency, , These are stoichiometric constants. By adjusting these parameters in real time, the operator accurately quantifies the synergistic generation intensity of pollutants and greenhouse gases per unit of coal consumption. Second, the policy coupling effect is quantified: a dual difference (DID) and synthetic control (SCM) framework is adopted, selecting parks implementing carbon trading pilots as the treatment group, and selecting 5 similar parks not included in the pilot as potential control groups; using the SCM algorithm, the optimal weight vector is solved through Lasso regression, and a virtual control group is synthesized with weights to minimize the mean square error of its pollutant emission trajectory in the 12 months before the policy is implemented compared with the treatment group; then, the difference between the changes in pollutant and carbon emissions in each month after the implementation of the policy is calculated between the treatment group and the virtual control group, separating the cross-effect of the carbon trading policy on sulfur dioxide emission reduction, and eliminating the interference of confounding factors such as economic cycles and meteorological conditions. Third, multi-objective resource allocation optimization: Under the hard constraints of meeting the annual total sulfur dioxide emissions in the region not exceeding 5,000 tons and the carbon emission intensity not exceeding 0.8 tons / 10,000 yuan of GDP, the Stackelberg-Nash equilibrium solution of the three-party intelligent agents is solved; the government, as the leader, first issues carbon tax and subsidy policies, and enterprises and residents, as followers, respond simultaneously; the optimization process introduces an incentive compatibility constraint mechanism, requiring that the marginal cost of emission reduction for enterprises equals the marginal benefit of government subsidies, and that the reduction in health risks for residents is not less than the loss of travel utility, ensuring that the equilibrium solution can be accepted by all parties in reality and implemented stably; the optimization results are output as monthly production load allocation plans for each enterprise, the allocation ratio of government subsidy funds, and suggestions for adjusting public transportation schedules.

[0062] The closed-loop feedback optimization module continuously receives new data streams pushed by the edge gateway and calculates the root mean square error (RMSE) between the model output and the measured values ​​every 24 hours. When a boiler fault is detected causing drastic fluctuations in emission data, the system automatically activates the robust acquisition function, which introduces a decay factor based on data confidence when calculating the expected improvement (EI) of Bayesian optimization. This represents the variance of the data for that period. This is the sensitivity coefficient. The impact of high-variance data points on parameter updates is automatically reduced. The Bayesian optimizer uses a Gaussian process as a surrogate model to iteratively adjust the desulfurization efficiency compensation coefficient in the physical commonality operator. SCM weight regularization parameters in policy coupling quantification and social welfare weights in resource allocation optimization The process continues until the RMSE converges to below the preset threshold of 0.05.

[0063] Finally, the strategy generation and output interface packages the final evaluation indicators (such as the synergistic emission reduction potential index and the policy cross-elasticity coefficient) and optimization strategies into JSON format and pushes them to the park's smart management platform via RESTful API. At the same time, it generates hourly emergency emission reduction suggestions (such as "increase the amount of ammonia sprayed for denitrification by 15% in the next 2 hours") and monthly policy effect evaluation reports to support decision-makers in dynamically adjusting governance measures.

[0064] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0065] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A system for evaluating the synergistic effect of pollution reduction and carbon reduction by integrating multi-agent technology and the Internet of Things, characterized in that, include: The IoT sensing layer is used to collect multi-source heterogeneous raw data streams in real time through a sensor network deployed in various monitoring areas, and to perform time synchronization processing on the data from each sensor through an edge computing gateway. The data preprocessing module is used to perform timestamp alignment, outlier removal, adaptive window filtering, and feature normalization on the raw data stream to generate a structured time-series dataset. A multi-agent modeling engine is used to construct a dynamic game model that includes government agents, enterprise agents, and resident agents. The government agent aims to maximize total social welfare, the enterprise agent aims to maximize profits and is constrained by total emissions, and the resident agent aims to optimize the combination of travel utility and health risk. The synergy calculation engine is used to sequentially perform physical common source correlation analysis, policy coupling effect quantification, and multi-objective resource allocation optimization to generate pollution reduction and carbon reduction synergy evaluation indicators. The closed-loop feedback optimization module is used to dynamically calibrate the model parameters in the synergistic effect calculation engine based on measured data and using a Bayesian optimization algorithm. The strategy generation and output interface is used to output evaluation results and optimization strategies to the decision support terminal.

2. The system according to claim 1, characterized in that, The IoT sensing layer adopts a layered deployment architecture, with different types of sensors configured in different monitoring areas, including: flue gas emission monitoring sensors in industrial source areas, air quality and traffic flow monitoring sensors at traffic nodes, energy consumption and heat monitoring sensors for energy facilities, and environmental and population activity monitoring sensors in residential areas; all data collected by the sensors are processed by time synchronization through an edge computing gateway to eliminate timing deviations caused by transmission delays.

3. The system according to claim 1, characterized in that, When the data preprocessing module performs adaptive window filtering, it monitors the fluctuation of the data within the window in real time. When the fluctuation exceeds a preset threshold, the sliding window length is shortened from the first length to the second length to preserve transient features. When the data returns to stability, the window length is gradually restored to the first length. Outlier removal adopts an algorithm based on neighborhood statistical features, and interpolation is performed to fill out outlier data points.

4. The system according to claim 1, characterized in that, In the multi-agent modeling engine, the travel utility function of the resident agent is constructed using a discrete choice model, and its expression is: in, Indicates the utility of residents' travel. Indicates commute time. This indicates the comfort index of the mode of transportation. Indicates the exposure concentration. Indicates the duration of exposure. , , The parameters are preset as behavioral parameters; the utility function explicitly includes a pollutant exposure term to simulate residents' behavioral feedback under changes in environmental quality; the agents interact with each other through message queues to exchange states and actions, and update strategies according to preset time steps.

5. The system according to claim 1, characterized in that, The physical co-source correlation analysis module in the synergistic effect calculation engine incorporates a material flow coupling operator. This operator establishes a nonlinear dynamic proportional relationship between pollutant generation and greenhouse gas emissions based on the stoichiometric relationship of the combustion reaction, specifically including: in, , These represent sulfur dioxide production and carbon dioxide emissions, respectively. Fuel consumption The sulfur content of the fuel, For desulfurization efficiency, , These are the ash content and moisture content, respectively. To fix the carbon content, For combustion efficiency, , These are stoichiometric constants; The operator dynamically corrects the correlation between the generation intensity of different emissions based on real-time monitoring of fuel characteristics and operating conditions, in order to identify the physical coupling characteristics of pollution reduction and carbon reduction.

6. A method for evaluating the synergistic effect of pollution reduction and carbon reduction by integrating multi-agent and Internet of Things, characterized in that, include: Step 1: Collect multi-source heterogeneous data in real time covering industrial, transportation, energy and residential life scenarios through the Internet of Things sensing layer, and perform time synchronization processing on the data; Step 2: Perform anomaly detection, adaptive median filtering, and normalization on the collected raw data stream to form a structured time-series dataset with a unified time granularity; Step 3: Initialize the state space and action space of the three types of intelligent agents: government, enterprise, and resident. Set the objective functions of maximizing total social welfare, maximizing profit, and maximizing comprehensive utility as the objective functions of each intelligent agent. Step 4: Perform physical co-source correlation analysis, construct a nonlinear dynamic coupling operator for pollutants and greenhouse gas generation based on material flow analysis, and quantify the synergistic emission reduction potential of shared emission sources; Step 5: Quantify the policy coupling effect by using dual difference logic and synthetic control method to construct a counterfactual benchmark and separate the cross-effects caused by policy tools; Step 6: Perform multi-objective resource allocation optimization, solve the equilibrium solution of the multi-party game under the constraint of total emissions, and introduce an incentive compatibility constraint mechanism to generate an optimal allocation scheme; Step 7: Continuously receive measured data, calculate the model output error, and use a Bayesian optimization algorithm combined with a robust acquisition function to perform rolling calibration of the model parameters; Step 8: Push the synergy effect evaluation indicators and optimization strategies to the decision-making terminal.

7. The method according to claim 6, characterized in that, The synthetic control method described in step five solves for the optimal weight vector so that the mean square error between the emission trajectory of the virtual control group, which is a weighted composite of multiple control regions, and the treatment group before the policy implementation is minimized. After the policy implementation, the synergistic emission reduction effect or offsetting effect of the policy is identified by calculating the difference between the actual trajectory of the treatment group and the counterfactual trajectory of the virtual control group, thereby eliminating the interference of confounding factors.

8. The method according to claim 6, characterized in that, The multi-objective resource allocation optimization described in step six is ​​achieved by solving a master-slave game equilibrium: the government agent, as the leader, first releases a policy combination, and the enterprise and resident agents, as followers, adjust their production and travel behaviors accordingly; the incentive compatibility constraint mechanism requires that the marginal cost of emission reduction for enterprises is not higher than the marginal subsidy benefit provided by the government, and that the reduction in residents' health risks is not lower than the loss of travel utility, so as to ensure the feasibility of the optimization scheme.

9. The method according to claim 6, characterized in that, In step seven, the robust acquisition function introduces a decay factor based on data confidence when calculating the improved value of Bayesian optimization. This decay factor is expressed as: in, As the attenuation factor, The variance of the current monitoring data. The sensitivity coefficient is used; the attenuation factor dynamically adjusts the contribution weight of parameter updates based on the data variance to suppress erroneous adjustments caused by sensor noise or faults; Bayesian optimization uses a Gaussian process as a surrogate model and iterates to bring the root mean square error between the model output and the measured value to within a preset threshold.

10. The method according to claim 6, characterized in that, The strategy generation and output interface described in step eight supports multi-granularity output, including short-term emergency emission reduction instructions, regional energy dispatch strategies, and long-term policy effect evaluation reports. The interface is integrated with the city's smart management platform through standardized protocols to achieve real-time interaction and closed-loop control of evaluation indicators and governance instructions.

Images