Urban carbon emission optimization model algorithm based on complex adaptation system theory

Abstract

The invention provides an urban carbon emission optimization model algorithm aiming at the urgent demand of green and low-carbon transformation of an urban industrial structure. According to the algorithm, a multi-agent model comprising a public management mechanism agent and an enterprise agent is constructed based on a complex adaptive system theory. The core lies in that an evolution path of urban carbon emission and industrial upgrading is dynamically deduced by simulating an adjacent learning effect between enterprises and a regulation and control mechanism of a public management mechanism based on a pre-decision. The enterprise agent decides the output value speed increase and the carbon emission intensity change rate according to the self state, the surrounding enterprise state and the public management mechanism predetermined target; and the government agent formulates and adjusts industry and emission reduction policies according to the overall goal of the city and enterprise pre-decision. According to the invention, through dynamic coupling of the microscopic enterprise decision and the macroscopic city system, low-cost and high-efficiency simulation and optimization of the city carbon emission path under different target scenes are realized, and a scientific quantitative analysis tool is provided for city carbon emission.

Description

Technical Field

[0001] This invention belongs to the field of urban environmental management and strategy simulation technology, and relates to a multi-agent model based on complex adaptive systems theory, which includes intelligent agents of public management agencies and intelligent agents of enterprises. Background Technology

[0002] As a major source of carbon emissions and the primary battleground for emission reduction, the scientific rigor and effectiveness of urban carbon emission decisions are crucial. Currently, urban carbon emission decisions largely rely on macroeconomic statistical data, static models, or empirical judgments, which have the following limitations: First, the lack of micro-mechanisms; traditional models struggle to characterize the decision-making behavior and complex interactions of micro-entities such as public management agencies and enterprises, failing to explain the micro-causes of macro-carbon emission trends. Second, insufficient dynamic feedback; static models cannot simulate the chain reactions and long-term effects resulting from market entities dynamically adjusting their behavior after strategy implementation. Third, difficulties in target pre-simulation: it is difficult to simulate various scenarios with different strategy combinations and implementation intensities in a low-cost and efficient manner before targets are achieved, leading to a degree of blindness in target setting. Furthermore, spatial factors are often neglected, with less consideration given to the impact of proximity learning and technology diffusion effects caused by the spatial distribution of enterprises on regional carbon emissions. Multi-agent simulation technology provides an effective means of simulating complex systems, but its application in urban carbon emission management is still immature. Existing technologies typically simplify agent behavior rules, failing to effectively integrate key real-world factors such as public management agency target management, inter-enterprise learning effects, and target cost constraints, resulting in limited model realism and decision support capabilities.

[0003] In recent years, research on urban carbon emissions and green and low-carbon transformation has deepened. Some scholars have constructed a three-layer analytical model of urban power systems based on complex systems theory, including the element layer, analysis layer, and result layer, to obtain the influencing factors, mechanisms, and development direction of urban power system evolution (Xiao Jinxing et al., 2023). However, significant differences in industrial structures lead to different green and low-carbon transformation paths. Other scholars have used system dynamics models to simulate carbon emissions in urban agglomerations in 2030 and 2060. Under the baseline scenario, carbon emission targets are difficult to achieve, and the carbon emission system is on the verge of collapse. However, under the high-quality development scenario, except for the central urban agglomeration, other urban agglomerations can achieve their carbon emission targets (Zeng Peng et al., 2022). However, a city is an "adaptive complex mega-system" containing several subsystems, including natural, social, and economic systems. It is necessary to accurately grasp the interaction mechanisms between these subsystems and conduct research from a multivariate, multi-feedback perspective using appropriate methods.

[0004] A search revealed several invention patents related to urban carbon emission control. One patent, application number CN202210110326.4, constructs a calculation model for urban-scale industrial carbon emissions, including a carbon emission prediction model based on an improved environmental impact assessment method. Another patent, application number CN202310524278.8, constructs an urban energy carbon emission monitoring system based on carbon emission targets. This system includes a basic information collection module, an electricity usage structure layer division module, a carbon emission monitoring module for designated industrial parks, a carbon emission assessment center, a management prompt cloud platform, and an SQL database. Patent application CN202410060957.9 describes an invention based on a regional tree-based carbon accounting model. It creates a regional carbon accounting tree structure according to the characteristics of the region, with the final branches corresponding to specific spatial objects. The data import module is configured to import carbon accounting-related data from outside the regional tree-based model. Patent application CN202510664823.2 describes an invention that obtains a multidimensional data vector sequence of urban smart energy at the current sampling time. This multidimensional data vector sequence is used as input to an LSTM model to obtain the hidden state data vector and multidimensional activation vector at the current sampling time. No patents related to urban carbon emission optimization algorithms were found.

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide an urban carbon emission optimization model algorithm. This algorithm can more realistically simulate the decision-making interaction between public management agencies and enterprises in the urban system, as well as the spatial interaction between enterprises, thereby realizing the dynamic deduction and optimization of different carbon emission target scenarios. Summary of the Invention

[0006] An algorithm for optimizing urban carbon emissions is proposed. Its core lies in constructing a multi-agent system comprising public management agency agents and enterprise agents, and simulating their decision-making and interaction processes. The algorithm mainly includes: a system initialization phase, which primarily inputs urban enterprise data (coordinates, industry, output value) and assigns initial carbon emission intensity to each enterprise based on greenhouse gas inventory data; a pre-decision-making cycle phase, which preliminarily determines the next year's output growth rate and carbon emission intensity change rate based on each enterprise's own state (natural growth rate) and the state of neighboring enterprises (through a proximity learning algorithm); and finally, after summarizing the pre-decision decisions of all enterprises, an environmental pre-simulation is performed to obtain the urban carbon emission rate under the condition of not imposing new targets. Economic and carbon emission forecasts; in the target regulation and coordination stage, the public management agency's intelligent agent compares the pre-simulation results with the city's carbon emission targets (such as output targets and peak carbon emission targets). If there are deviations, the public management agency formulates and releases regulation targets (such as restricting the capacity of high-emission industries, encouraging the development of low-carbon industries, and mandatory emission reductions). These targets are quantified as adjustments to the pre-decision decisions of enterprises. Each enterprise corrects its pre-decision decisions based on the targets preset by the public management agency, forming the final decision; in the environmental update and iteration stage, based on the final decisions of enterprises, the state of the entire city system (total output, total carbon emissions, etc.) is updated, and the next time step is entered, repeating the above process to achieve multi-year dynamic simulation. Attached Figure Description

[0007] Figure 1 Overall process of urban carbon emission optimization model algorithm Figure 2 The relationship between multi-agent interactions and environmental systems Figure 3 Detailed process of coordinating corporate pre-decision-making with public management agency regulation. Figure 4 Inter-firm proximity learning effect Detailed Implementation

[0008] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0009] This invention designs an urban carbon emission optimization model algorithm based on complex adaptive systems theory. This algorithm integrates a multi-agent model, which deconstructs the environmental, economic, and social subsystems into three types of agents: public management agencies, industries, and populations, each with different behavioral characteristics and interactions. At the environmental subsystem level, public management agencies, influenced by enterprise revenue and resident evaluations, control carbon emissions by setting environmental indicators and indirectly affect environmental carbon emissions by setting optimization goals and social indicators to regulate industry and population behavior. At the economic subsystem level, various industries, regulated by indicators and corresponding target tools set by public management agencies and driven by population employment and consumption, continuously transform their production methods, scale, and structure, thus affecting carbon emissions. At the social subsystem level, under the combined influence of public management agency guidance and industry job and service supply, population size, structure, and lifestyles constantly change, also affecting carbon emissions. Finally, the optimal path for modeling the characteristics and interrelationships of each subsystem in complex adaptive systems theory—that is, the adaptation process mechanism—is given, including the following steps: (1) Model building Phase S1 involves building the model environment and initializing the intelligent agents: initializing the urban system environmental variables E~t~, including the initial output value, carbon emission intensity, spatial coordinates, and industry of each enterprise; and building enterprise intelligent agents and public management agency intelligent agents. Phase S2 is the enterprise pre-decision: Each enterprise agent calculates the pre-decision D'~t~ based on the current environmental variable E~t~, its natural growth rate of output value, natural change rate of carbon emission intensity, and the status of spatially neighboring enterprises through the proximity learning algorithm. The pre-decision includes the predetermined growth rate of output value and the change rate of carbon emission intensity. Phase S3 is the environmental pre-simulation: the enterprise's pre-decision D'~t~ is input into the environmental impact function I, and the pre-simulated environmental variable E'~t+1~ for the next year is obtained; Phase S4 is the regulation and coordination decision-making of public management agencies: the public management agency's intelligent agent formulates regulation targets based on the city's output and carbon emission targets, compared with the pre-simulated environmental variables E'~t+1~; each enterprise's intelligent agent coordinates and modifies its pre-decision based on the public management agency's preset targets, forming the final decision D~t~; Phase S5 is for environmental state update: Based on the final decisions D~t~ of all enterprise agents, the environmental variables are updated through the environmental impact function I to obtain the environmental state E~t+1~ for the next period; Phase S6 is an iterative simulation: taking E~t+1~ as the new current environmental state, repeating phases S1-S5, performing multi-cycle iterative simulations until the preset termination conditions are reached, and outputting the simulated path of urban carbon emissions and industrial upgrading. (2) Algorithm design 1) In the S2 stage, the specific process of the proximity learning algorithm is as follows: for any enterprise agent, identify enterprises in the same industry or other specified categories within its spatial proximity range; calculate the average or weighted average of the output growth rate and carbon emission intensity change rate of the neighboring enterprises; the output growth rate and carbon emission intensity change rate in the enterprise agent's pre-decision will approach the average of its neighboring enterprises. 2) In the S4 stage, the regulatory targets set by the public management agency intelligent agent include output value regulation targets and carbon emission intensity regulation targets; the targets have intensity attributes, and the cost is related to the target intensity through an exponential function. The greater the target intensity, the higher the cost. 3) The algorithm execution process for the regulation and coordination decision-making of public management agencies in the S4 stage is as follows: The public management agency's intelligent agent obtains the enterprise's pre-decision D'~t~ and the pre-simulated environmental variable E'~t+1~; compares the city's total output and total carbon emissions in E'~t+1~ with the preset target; if the pre-simulated result deviates from the target, the public management agency's intelligent agent generates output adjustment instructions and carbon emission intensity adjustment instructions for specific industries or the entire industry. Each enterprise's intelligent agent receives the instructions from the public management agency, superimposes its pre-decision D'~t~ with the instructions from the public management agency, and forms the final decision D~t~. 4) The carbon emissions of the enterprise intelligent agent are calculated through the improved Kaya identity model. Specifically, the enterprise carbon emissions are obtained by multiplying the enterprise output value and the enterprise carbon emission intensity. The enterprise carbon emission intensity is jointly determined by its initial intensity, natural rate of change, adjustment amount of proximity learning effect and adjustment amount of public management agency regulation. 5) The initial data for the algorithm comes from real data of urban enterprises, including enterprise geographical coordinates, industry classification, and historical output value; the initial carbon emission intensity of enterprises is generated through the following steps: based on the urban greenhouse gas emission inventory, determine the average carbon emission intensity of each inventory industry; using the average carbon emission intensity of the respective inventory industry as the mean, and one-tenth of the mean as the standard deviation, assign an initial carbon emission intensity to each enterprise through a normal distribution random function.

[0010] Example Taking a certain region as an example, the algorithm was implemented on the Anylogic simulation platform using real data from 5,531 enterprises.

[0011] (1) Data preparation and initialization 1) Collect enterprise data in the region, including enterprise name, industry, geographical latitude and longitude, and 2019 output value, and save it into an Excel file; 2) Determine the correspondence between the listed industries (such as power and heat, building materials, chemical industry, etc.) and the objectives of public management agencies; 3) Calculate the average carbon emission intensity for each industry in the inventory list. For the i-th industry in the inventory list, its emission intensity IndCatEF is... i It is obtained by dividing the industry's total emissions by the industry's total output. 4) Generate the initial carbon emission intensity EF for the k-th enterprise. k Using a normally distributed random function, the average emission intensity IndCatEF of the industry to which the enterprise belongs is used. i The mean is represented by IndCatEF. i / 10 is used to randomly assign a value to the standard deviation; 5) Set city carbon emission targets, such as annual output growth targets and total carbon emission control targets, and save them to another Excel file; (2) Model building and agent definition 1) Enterprise Intelligent Agent: Each enterprise is an intelligent agent with attributes including output value (GDP), carbon emission intensity (EF), output value growth rate (GDPRate), carbon emission intensity change rate (EFRate), and spatial coordinates. The pre-decision function calculates the predetermined output value growth rate (GDPRate') and carbon emission intensity change rate (EFRate'). The output value growth rate is obtained by summing GDPRateN (natural growth rate) and GDPRateS (clustering effect adjustment), and the carbon emission intensity change rate is obtained by summing EFRateN (natural change rate) and EFRateS (clustering effect adjustment). GDPRateS and EFRateS are calculated using a proximity learning algorithm: the enterprise searches for enterprises in the same industry within a certain distance, calculates the average of the GDPRate and EFRate of these neighboring enterprises, and then adjusts its own pre-decision value to approach this average.

[0012] 2) Public Management Agency Intelligent Agent: Attributes include output target, carbon emission target, and decision-making cost. It reads the pre-decision D'_t and pre-simulated environment E'_{t+1} of all enterprises. If the pre-simulated total output is lower than the target or the total carbon emissions are higher than the target, then a control target is set. Different parameter values ​​are set according to different industries: for the steel industry, GDPRateG is assigned a value of -10% to limit its capacity growth; for the building materials industry, EFRateG is assigned a value of -10% to force it to increase emission reduction efforts; for low-carbon industries (such as other industrial sectors), GDPRateG is assigned a value of +5% to encourage its development.

[0013] (3) Model running and simulation The model's time step is set to 1 year. Within each time step, the following loop is executed: In the pre-decision stage, all enterprises execute pre-decision decisions in parallel, generating D'_t; in the pre-deduction stage, the system calculates E'_{t+1} based on I(E_t, D'_t); in the coordination decision stage, public management agencies formulate objectives based on E'_{t+1}, and enterprises revise their decisions based on the objectives to obtain the final decision D_t; in the state update stage, the system calculates E_{t+1} based on I(E_t, D_t) and updates the state of all enterprises and cities; the model runs until the target year (e.g., 2060), recording all process data.

[0014] (4) Results analysis and target optimization By observing real-time trends in carbon emissions, output value, and target costs, and comparing simulation results under different strategy combinations, such as scenarios without pre-set targets and optimized scenarios, the effectiveness, economic impact, and cost of various strategies can be clearly assessed, thereby selecting the optimal carbon emission optimization scheme. The results show that compared to the scenario without pre-set targets, implementing a combination of targets in 2024 (strict control of steel, industry-wide emission reduction, structural adjustment, and special emission reduction for high-emission industries) could advance the peak time of carbon emissions in the region from 2045 to around 2031, and achieve a balance between carbon emissions and carbon absorption before 2060.

Claims

1. An urban carbon emission optimization model algorithm based on complex adaptive systems theory, characterized in that: The multi-agent model deconstructs the environmental, economic, and social subsystems into three types of agents: public management agencies, industries, and populations, each with different behavioral characteristics and interactions. At the environmental subsystem level, public management agencies, influenced by enterprise revenue and resident evaluations, control carbon emissions by setting environmental indicators and indirectly affect environmental carbon emissions by regulating industry and population behavior through optimization goals and social indicators. At the economic subsystem level, various industries, regulated by indicators and corresponding target tools set by public management agencies and driven by population employment and consumption, continuously transform their production methods, scale, and structure, thus influencing carbon emissions. At the social subsystem level, under the combined influence of public management agency guidance and industry job and service supply, population size, structure, and lifestyles constantly change, also affecting carbon emissions. Finally, the optimal path for modeling the characteristics and interrelationships of each subsystem in complex adaptive systems theory—that is, the adaptive process mechanism—is presented, including the following steps: (1) Model building Phase S1 involves building the model environment and initializing the intelligent agents: initializing the urban system environmental variables E~t~, including the initial output value, carbon emission intensity, spatial coordinates, and industry of each enterprise; and building enterprise intelligent agents and public management agency intelligent agents. Phase S2 is the enterprise pre-decision: Each enterprise agent calculates the pre-decision D'~t~ based on the current environmental variable E~t~, its natural growth rate of output value, natural change rate of carbon emission intensity, and the status of spatially neighboring enterprises through the proximity learning algorithm. The pre-decision includes the predetermined growth rate of output value and the change rate of carbon emission intensity. Phase S3 is the environmental pre-simulation: the enterprise's pre-decision D'~t~ is input into the environmental impact function I, and the pre-simulated environmental variable E'~t+1~ for the next year is obtained; Phase S4 is the regulation and coordination decision-making of public management agencies: the public management agency agents formulate regulation targets based on the city's output and carbon emission targets, compared with the pre-simulated environmental variables E'~t+1~; each enterprise agent coordinates and modifies its pre-decision based on the public management agency's predetermined targets, forming the final decision D~t~. Phase S5 is for environmental state update: Based on the final decisions D~t~ of all enterprise agents, the environmental variables are updated through the environmental impact function I to obtain the environmental state E~t+1~ for the next period; Phase S6 is an iterative simulation: taking E~t+1~ as the new current environmental state, repeating phases S1-S5, performing multi-cycle iterative simulations until the preset termination conditions are reached, and outputting the simulated path of urban carbon emissions and industrial upgrading. (2) Algorithm design 1) In the S2 stage, the specific process of the proximity learning algorithm is as follows: for any enterprise agent, identify enterprises in the same industry or other specified categories within its spatial proximity range; calculate the average or weighted average of the output growth rate and carbon emission intensity change rate of the neighboring enterprises; the output growth rate and carbon emission intensity change rate in the enterprise agent's pre-decision will approach the average of its neighboring enterprises. 2) In the S4 stage, the regulatory targets set by the public management agency intelligent agent include output value regulation targets and carbon emission intensity regulation targets; the targets have intensity attributes, and the cost is related to the target intensity through an exponential function. The greater the target intensity, the higher the cost. 3) The algorithm execution process for the regulation and coordination decision-making of public management agencies in the S4 stage is as follows: The public management agency's intelligent agent obtains the enterprise's pre-decision D'~t~ and the pre-simulated environmental variable E'~t+1~; compares the city's total output and total carbon emissions in E'~t+1~ with the preset target; if the pre-simulated result deviates from the target, the public management agency's intelligent agent generates output adjustment instructions and carbon emission intensity adjustment instructions for specific industries or the entire industry. Each enterprise's intelligent agent receives the instructions from the public management agency, superimposes its pre-decision D'~t~ with the instructions from the public management agency, and forms the final decision D~t~. 4) The carbon emissions of the enterprise intelligent agent are calculated through the improved Kaya identity model. Specifically, the enterprise carbon emissions are obtained by multiplying the enterprise output value and the enterprise carbon emission intensity. The enterprise carbon emission intensity is jointly determined by its initial intensity, natural rate of change, adjustment amount of proximity learning effect and adjustment amount of public management agency regulation. 5) The initial data for the algorithm comes from real data of urban enterprises, including enterprise geographical coordinates, industry classification, and historical output value; the initial carbon emission intensity of enterprises is generated through the following steps: based on the urban greenhouse gas emission inventory, determine the average carbon emission intensity of each inventory industry; using the average carbon emission intensity of the respective inventory industry as the mean, and one-tenth of the mean as the standard deviation, assign an initial carbon emission intensity to each enterprise through a normal distribution random function.

Images