Industrial and mining town planning layout optimization method and system based on multi-agent reinforcement learning

Abstract

The invention relates to the technical field of industrial and mining town planning, in particular to an industrial and mining town planning layout optimization method and system based on multi-agent reinforcement learning, and the method comprises the steps: obtaining geographic information system (GIS) data, population distribution data and functional area planning data of a target industrial and mining town; based on GIS data, constructing a digital map model of the industrial and mining town; dividing the industrial and mining town into a plurality of functional areas according to the population distribution data and the functional area planning data, and abstracting each functional area into an intelligent agent; constructing a multi-agent reinforcement learning environment based on a plurality of agents; a multi-agent reinforcement learning algorithm is utilized to optimize the planning layout of the industrial and mining town; the optimized mining town planning layout scheme is generated, a multi-agent reinforcement learning framework is adopted, and multiple planning targets such as economic benefits, environmental protection and social fairness can be considered at the same time. Each functional area serves as an intelligent agent, and overall optimization under a complex target is achieved through mutual cooperation and competition.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial and mining town planning, and more specifically, to an industrial and mining town planning layout optimization method and system based on multi-agent reinforcement learning. Background Art

[0002] With the acceleration of urbanization, especially the rapid development of industrial and mining towns, the optimization of urban planning layout has become an increasingly important issue. Traditional urban planning methods mainly rely on the experience and intuition of planners, which often makes it difficult to fully consider complex geographical environment, population distribution, economic development and other factors, and it is also difficult to cope with the growing demand for sustainable development.

[0003] In recent years, with the development of artificial intelligence technology, some intelligent urban planning methods have begun to emerge. For example, some researchers have tried to use genetic algorithms to optimize the layout of urban functional areas, and some scholars have proposed urban expansion simulation methods based on cellular automata. These methods have improved the efficiency and scientificity of planning to a certain extent, but there are still some obvious limitations.

[0004] First, most existing methods can only handle a single or a few optimization objectives, and it is difficult to fully consider the multiple factors in urban planning. For example, some methods may focus too much on economic benefits and ignore important aspects such as environmental protection and social equity. Second, existing methods usually use static optimization models, which are difficult to adapt to the dynamics and uncertainty of urban development. Furthermore, many methods lack consideration of long-term sustainability, which easily leads to a contradiction between short-term benefits and long-term development.

[0005] In addition, existing methods often face the challenge of low computational efficiency when dealing with large-scale and highly complex urban planning problems. In particular, for areas with special geographical environments and functional requirements such as industrial and mining towns, the applicability of existing methods is even more limited. Finally, existing methods generally lack effective human-computer interaction mechanisms, making it difficult to fully utilize the experience and judgment of human experts.

[0006] In view of the above problems, there is an urgent need for an industrial and mining town planning layout optimization method that can comprehensively consider multiple factors, adapt to dynamic changes, take into account long-term sustainability, has high computational efficiency, and can effectively combine human experience. Summary of the invention

[0007] The present invention is proposed to address these problems existing in the prior art. The present invention aims to solve a series of key technical problems in the optimization of industrial and mining town planning layout, such as multi-objective coordination, dynamic adaptation, sustainable development, computational efficiency and human-machine collaboration. By innovatively combining advanced technologies such as multi-agent reinforcement learning, graph neural networks and geographic information systems, the present invention provides a comprehensive, efficient and intelligent method and system for optimizing the planning layout of industrial and mining towns.

[0008] The present invention provides a method for optimizing the layout of industrial and mining town planning by multi-agent reinforcement learning, comprising:

[0009] The acquisition steps include:

[0010] Obtain geographic information system (GIS) data, population distribution data, and functional area planning data for target industrial and mining towns;

[0011] Processing steps include:

[0012] Based on the GIS data, a digital map model of industrial and mining towns is constructed;

[0013] According to the population distribution data and the functional area planning data, the industrial and mining town is divided into a plurality of functional areas, and each functional area is abstracted as an intelligent body;

[0014] Based on the multiple agents, construct a multi-agent reinforcement learning environment;

[0015] Use multi-agent reinforcement learning algorithms to optimize the planning and layout of industrial and mining towns;

[0016] Output steps include:

[0017] Generate optimized industrial and mining town planning layout plans.

[0018] Preferably, the processing step further comprises:

[0019] Based on the GIS data, construct an adjacency matrix between intelligent agents;

[0020] According to the adjacency matrix, defining interaction rules between agents;

[0021] The element A of the adjacency matrix ij Defined as:

[0022]

[0023] Among them, P i and P j Represent the position coordinates of agent i and agent j respectively.

[0024] Preferably, the method of optimizing the planning layout of industrial and mining towns by using a multi-agent reinforcement learning algorithm specifically includes:

[0025] Define the state space S of the agent, where S = {s|s∈[0,1]}, s(i) represents the planning state of the i-th functional area;

[0026] Define the action space A of the agent, where A = {a|a∈{0,1} n}, n represents the total number of functional areas; = define the agent's reward function R(t), where R(t) takes into account factors such as traffic accessibility, environmental carrying capacity, economic benefits, and social relations;

[0027] The deep Q network DQN algorithm is used to train the agent to learn the optimal planning strategy.

[0028] Preferably, the reward function R(t) is defined as:

[0029] R(t)=w1×R accessibility +w2×R environment +w3×R economy +w4×R social ,

[0030] Among them, R accessibility represents the traffic accessibility reward, R environment represents the environmental carrying capacity reward, R economy Represents economic benefit reward, R social represents social relationship rewards, and w1, w2, w3 and w4 are the corresponding weight coefficients.

[0031] Preferably, the processing step further comprises:

[0032] Based on a graph neural network, the GIS data is encoded into a graph structure representation;

[0033] Utilize graph convolution operations to extract spatial features of industrial and mining towns;

[0034] The extracted spatial features are input into a multi-agent reinforcement learning algorithm to optimize the planning layout.

[0035] Preferably, the processing step further comprises:

[0036] Based on the GIS data, evaluate the environmental impact of the planning scheme;

[0037] Adjust the reward function of the multi-agent reinforcement learning algorithm based on the results of the environmental impact assessment;

[0038] Iterate and optimize the planning layout until the preset environmental protection requirements are met.

[0039] Preferably, the processing step further comprises:

[0040] Construct urban development simulation models to predict the long-term impact of planning options;

[0041] Assess the sustainability of the planning options based on the long-term impact forecasts;

[0042] Further optimize the planning layout based on the sustainability assessment results.

[0043] Preferably, the processing step further comprises:

[0044] After each optimization iteration, a visual representation of the current planning layout is generated;

[0045] Receive feedback from human experts on the current planning layout;

[0046] Based on the feedback, adjust parameters or constraints of the multi-agent reinforcement learning algorithm.

[0047] Preferably, the output step further comprises:

[0048] Post-process the optimized planning layout to eliminate layout conflicts between functional areas;

[0049] Generate a variety of visualizations including vector plots, heat maps, and density maps;

[0050] Output quantitative evaluation indicators of planning layout, including land use efficiency, transportation network density and balanced distribution of functional areas.

[0051] The industrial and mining town planning layout optimization system using multi-agent reinforcement learning to implement the method comprises:

[0052] Data acquisition module, used to obtain geographic information system GIS data, population distribution data and functional area planning data of target industrial and mining towns;

[0053] A digital modeling module, used to construct a digital map model of industrial and mining towns based on the GIS data;

[0054] A functional area division module is used to divide the industrial and mining town into multiple functional areas according to the population distribution data and the functional area planning data, and abstract each functional area into an intelligent body;

[0055] A learning environment construction module, used to construct a multi-agent reinforcement learning environment based on the multiple agents;

[0056] The layout optimization module is used to optimize the planning layout of industrial and mining towns using multi-agent reinforcement learning algorithms;

[0057] Output module, used to generate optimized industrial and mining town planning layout plan;

[0058] Wherein, the layout optimization module further includes:

[0059] Graph neural network encoding unit, used to encode GIS data into graph structure representation;

[0060] A spatial feature extraction unit, used to extract spatial features of industrial and mining towns using graph convolution operations;

[0061] Environmental impact assessment unit, used to evaluate the environmental impact of planning schemes and adjust the reward function;

[0062] a sustainability assessment unit to forecast and assess the long-term sustainability of planning options;

[0063] A human-computer interaction unit, used to receive and process feedback from human experts;

[0064] The output module also includes:

[0065] A post-processing unit for eliminating layout conflicts between functional areas;

[0066] A visualization generation unit, used to generate a variety of visualization representations;

[0067] The evaluation index calculation unit is used to output the quantitative evaluation index of the planning layout.

[0068] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0069] First, the present invention adopts a multi-agent reinforcement learning framework, which can simultaneously consider multiple planning goals, such as economic benefits, environmental protection, social equity, etc. Each functional area acts as an intelligent agent, and through mutual cooperation and competition, it achieves overall optimization under complex goals. This method can not only balance the needs of all parties, but also discover synergy effects that are difficult to detect with traditional methods.

[0070] Secondly, the present invention introduces graph neural network technology to effectively capture the complexity of urban spatial structure. By encoding GIS data into a graph structure and using graph convolution operations to extract spatial features, this method can provide a deeper understanding of the geographical environment and functional relationships of towns, thereby making more reasonable planning decisions.

[0071] Furthermore, the present invention achieves the prediction and evaluation of the long-term impact of planning schemes by constructing an urban development simulation model. This forward-looking method effectively improves the sustainability of planning and avoids the long-term problems that may be caused by short-sighted decisions.

[0072] In addition, the method of the present invention has strong adaptability and scalability. By dynamically adjusting the reward function and learning parameters, the method can flexibly respond to various changes and uncertainties in the process of urban development. At the same time, the modular system design allows each functional unit to be independently optimized and upgraded, which provides convenience for future functional expansion.

[0073] Finally, the present invention designs an effective human-computer interaction mechanism. By visually displaying the intermediate results of the optimization process and receiving feedback from human experts, this method realizes the organic combination of artificial intelligence and human experience, greatly improving the reliability and practicality of the planning scheme.

[0074] In general, the method and system for optimizing the layout of industrial and mining towns by multi-agent reinforcement learning provided by the present invention not only overcomes many limitations of the prior art, but also achieves significant breakthroughs in multi-objective optimization, dynamic adaptation, sustainable development, computational efficiency, and human-machine collaboration. This innovative method provides strong technical support for the scientific planning and sustainable development of industrial and mining towns, and has important theoretical value and broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 The figure is a flow chart of the method of the present invention.

[0076] Figure 2 It is a logic block diagram of the layout optimization module of the present invention.

[0077] Figure 3 1 is a logic block diagram of an output module of the present invention. DETAILED DESCRIPTION

[0078] Please refer to Figure 1-3 The present invention provides a method and system for optimizing the planning layout of industrial and mining towns based on multi-agent reinforcement learning. The method realizes the intelligent optimization of the planning layout of industrial and mining towns by innovatively combining multi-agent reinforcement learning, graph neural network and geographic information system technology.

[0079] Specifically, the method of the present invention comprises the following steps:

[0080] First, in the acquisition step, the method obtains the geographic information system (GIS) data, population distribution data and functional area planning data of the target industrial and mining town. These data provide basic information for the subsequent optimization process. Preferably, the GIS data includes environmental factors such as topography, hydrology, and soil, the population distribution data reflects the spatial distribution characteristics of urban residents, and the functional area planning data contains a preliminary urban functional layout plan.

[0081] Next, in the processing step, the method first constructs a digital map model of industrial and mining towns based on the acquired GIS data. This step converts the actual geographical environment into a digital model that can be processed by a computer, laying the foundation for the subsequent optimization process. For example, a raster data model or a vector data model can be used to represent geographical elements such as terrain and water systems.

[0082] Then, based on the population distribution data and functional area planning data, the method divides the industrial and mining town into multiple functional areas, and abstracts each functional area into an intelligent agent. This abstraction enables the transformation of complex town planning problems into optimization problems of multi-agent systems. In one embodiment, the functional areas may include residential areas, industrial areas, commercial areas, green areas, etc. Each functional area, as an independent intelligent agent, can make decisions based on its own characteristics and the surrounding environment.

[0083] Based on these agents, this method constructs a multi-agent reinforcement learning environment. In this environment, each agent can perceive the surrounding environment, make decisions, and learn from the results of the decisions. For example, an agent representing a residential area may adjust its position and size based on factors such as the surrounding green space area and transportation convenience.

[0084] The core optimization process is to use multi-agent reinforcement learning algorithms to optimize the planning and layout of industrial and mining towns. This process involves the design of complex state space, action space and reward function, which will be described in detail in the subsequent instructions.

[0085] Finally, in the output step, the method generates an optimized industrial and mining town planning layout plan, which includes not only the spatial layout of each functional area, but also detailed information such as the road network and public facilities distribution.

[0086] Furthermore, the method also includes constructing an adjacency matrix between agents based on GIS data in the processing step, and defining the interaction rules between agents according to the adjacency matrix. This step is crucial for simulating the mutual influence between different functional areas in a town.

[0087] Specifically, the element A of the adjacency matrix ij The definition is as follows:

[0088]

[0089] Among them, P i and P j Represent the position coordinates of agent i and agent j respectively, ||P i -P j || represents the Euclidean distance between two agents, and D1 is a preset threshold.

[0090] In practical applications, the choice of threshold $D_1$ is critical. It determines which functional areas are considered "adjacent", thus affecting the interaction between them. As a rule of thumb, $D_1$ can be set to 1.5 to 2 times the diameter of the average functional area in a town. For example, if a typical functional area has a diameter of 1 km, then $D_1$ can be set to 1.5 to 2 km. Such a setting ensures that adjacent functional areas can interact effectively while not causing too much long-distance interaction, thereby maintaining computational efficiency.

[0091] In practical applications, the choice of threshold D1 is critical. It determines which functional areas are considered "adjacent", thus affecting the interaction between them. As a rule of thumb, D1 can be set to 1.5 to 2 times the diameter of the average functional area in a town. For example, if a typical functional area has a diameter of 1 km, then D1 can be set to 1.5 to 2 km. Such a setting ensures that adjacent functional areas can interact effectively while not causing too much long-distance interaction, thereby maintaining computational efficiency.

[0092] The present invention further refines the specific implementation of the multi-agent reinforcement learning algorithm. First, the state space S of the agent is defined. In this space, s∈[0,1] represents a standardized state value, and s(i) represents the planning state of the i-th functional area. This representation method allows different types of state information (such as land utilization, population density, etc.) to be unified into a standardized range.

[0093] The action space A is defined as A = {a|a∈{0,1} n}, where x represents the total number of functional areas. This means that each agent can choose to take action (1) or not take action (0) on each functional area at each time step. For example, in a system with 5 functional areas, a possible action may be 1, 0, 1, 0, 1, indicating taking action on the 1st, 3rd, and 5th functional areas.

[0094] The reward function R(t) is a key component in reinforcement learning. It takes into account several factors:

[0095] R(t)=w1·R accessibility +w2·R environment +w3·R economy +w4·R social ,

[0096] Among them, R acccessibility represents the traffic accessibility reward, R environment represents the environmental carrying capacity reward, R economy Represents economic benefit reward, R socialrepresents social relationship rewards, w1, w2, w3 and w4 are corresponding weight coefficients. The design of this reward function reflects the comprehensive consideration of various factors in urban planning by the present invention. For example, traffic accessibility can be quantified by calculating the average distance between functional areas, environmental carrying capacity can be evaluated by green space coverage and pollution index, economic benefits can be measured by land use efficiency and industrial agglomeration, and social relations can be represented by the balance of population distribution and the coverage of public facilities.

[0097] In practical applications, the selection of these weight coefficients needs to be determined according to the specific town development goals and local characteristics. For example, for a town with ecological livability as its main goal, the weight of w2 may be increased; while for an industrial town with economic development as its main goal, the weight of w3 may be increased.

[0098] The present invention uses the deep Q-network (DQN) algorithm to train the agent to learn the optimal planning strategy. The DQN algorithm combines the idea of ​​Q-learning with the powerful expression ability of deep neural networks and can effectively process high-dimensional state space. During the training process, the agent gradually optimizes its decision-making strategy by continuously interacting with the environment, and finally achieves the purpose of optimizing the town layout.

[0099] Through the above method, the present invention can effectively deal with the complex multi-objective optimization problem of industrial and mining town planning layout, and realize the overall optimization of town layout while considering multiple factors. Compared with traditional planning methods, this method has stronger adaptability and optimization ability, and can better meet the needs of modern town development.

[0100] In a preferred embodiment of the present invention, the definition of the reward function R(t) is further refined as follows:

[0101] R(t)=w1·R accessibility +w2·R environment +w3·R economy +w4·R social ,

[0102] The design of this reward function reflects the comprehensive consideration of various factors in town planning by the present invention. Each component represents a key aspect in town planning, and by adjusting the weight coefficients w1, w2, w3 and w4, it can flexibly adapt to the planning needs of different types of industrial and mining towns.

[0103] Specifically, R accessibility represents the traffic accessibility bonus. In practical applications, this can be quantified by calculating the average distance between functional areas or the connectivity of the transportation network. For example, the following formula can be used:

[0104]

[0105] Where n is the number of functional areas, d ij is the distance between functional areas i and j. This formula encourages the functional areas to maintain an appropriate distance, neither too crowded nor too scattered. environment Represents the environmental carrying capacity bonus. This can be calculated by evaluating factors such as green space coverage, water quality indicators, air quality, etc. In one embodiment of the present invention, the following formula can be used:

[0106] R environment =α·GreenCoverage+β·WaterQuality+γ·AirQuality,

[0107] Among them, α, β and γ are weight coefficients, which respectively reflect the importance of green space coverage, water quality and air quality in environmental assessment.

[0108] R economy It indicates economic benefit reward, which can be measured by evaluating indicators such as land use efficiency and industrial agglomeration. For example: R economy =λ·LandUseEfficiency+μ·IndustrialAgglomeration, where λ and μ are weight coefficients, reflecting the relative importance of land use efficiency and industrial agglomeration in economic benefit evaluation.

[0109] Finally, R social Represents social relationship rewards, which can be calculated by evaluating factors such as the balance of population distribution and the coverage of public facilities. A possible formula is: R social =δ·PopulationBalance+∈·PublicFacilityCoverage, where δ and ∈ are weight coefficients, reflecting the importance of population balance and public facility coverage in social relationship evaluation.

[0110] Through this detailed reward function design, the method of the present invention can comprehensively consider all key aspects of industrial and mining town planning, thereby achieving more reasonable and sustainable layout optimization.

[0111] Furthermore, the method of the present invention introduces graph neural network technology to better capture the complexity of urban spatial structure. Specifically, the method encodes GIS data into a graph structure representation based on graph neural network. In this graph structure, nodes can represent functional areas or other key geographical units, while edges represent spatial relationships or other interactions between them.

[0112] Subsequently, this method uses graph convolution operations to extract the spatial features of industrial and mining towns. The graph convolution operation can be expressed as:

[0113] H (l+1) =σ(D -1 / 2 AD -1 / 2 H (l) W (l) ),

[0114] Among them, H (l) is the node feature matrix of the lth layer, A is the adjacency matrix, D is the degree matrix, W (l) is a learnable weight matrix, and σ is a nonlinear activation function. Through multi-layer graph convolution operations, this method can effectively extract and aggregate spatial information and capture local and global structural features in town layout.

[0115] These extracted spatial features are then fed into a multi-agent reinforcement learning algorithm to optimize the planning layout. This combination enables the method of the present invention to better understand and utilize the spatial structure information of the town, thereby generating a more reasonable and efficient planning solution.

[0116] In another embodiment of the present invention, the method further comprises evaluating the environmental impact of the planning scheme based on GIS data. This step reflects the emphasis of the present invention on sustainable development. Specifically, the environmental impact assessment can consider multiple factors, such as land use change, ecosystem service value, carbon emissions, etc.

[0117] For example, the following formula can be used to assess environmental impact:

[0118]

[0119] Among them, I i is the current value of the i-th environmental indicator, is the reference value, and are the maximum and minimum values ​​of the indicator respectively, w i is the weight coefficient. This formula standardizes different environmental indicators and weights them together to get a comprehensive environmental impact score. This formula standardizes different environmental indicators and weights them together to get a comprehensive environmental impact score.

[0120] Based on the results of the environmental impact assessment, this method dynamically adjusts the reward function of the multi-agent reinforcement learning algorithm. For example, if a planning scheme leads to a large negative environmental impact, the corresponding reward value will be reduced. This dynamic adjustment mechanism ensures that this method can continuously consider and improve environmental impacts while optimizing the town layout.

[0121] Preferably, the method will iteratively optimize the planning layout until the preset environmental protection requirements are met. This process may require multiple adjustments and optimizations, but in the end a planning scheme that meets both the needs of urban development and environmental protection standards can be obtained.

[0122] The method of the present invention also introduces an urban development simulation model to predict the long-term impact of the planning scheme. This step reflects the importance of the present invention to the long-term effects of urban planning. The urban development simulation model can predict the development of towns in the next few years or even decades based on a variety of factors, such as population growth trends, economic development expectations, and technological changes.

[0123] In one embodiment of the present invention, the urban development simulation model can use methods such as Cellular Automata (CA) or Agent-Based Modeling (ABM). For example, when using the CA model, the town can be divided into several grid cells, and the state change of each cell is determined by its own state and the state of the neighboring cells. The state transition rule can be expressed as:

[0124] S t+1 =f(S t ,N t ,C t ),

[0125] Among them, S t is the cell state at time t, N t is the state set of neighboring units, C t is a global constraint, and f is a state transition function.

[0126] Based on the long-term impact prediction results, this method will evaluate the sustainability of the planning scheme. This evaluation process may involve multiple indicators, such as resource utilization efficiency, environmental carrying capacity, social equity, etc. The following formula can be used to calculate the comprehensive sustainability index:

[0127]

[0128] Among them, SI i is the value of the ith sustainability indicator, and are the minimum and maximum values ​​of the indicator, respectively, v i is the weight coefficient.

[0129] Based on the results of the sustainability assessment, the method will further optimize the planning layout. This may involve adjusting the location and size of functional areas, changing land use patterns, adding or reducing certain facilities, etc. In this way, the method of the present invention can find a balance between short-term optimization and long-term sustainability, providing scientific planning guidance for the long-term development of industrial and mining towns.

[0130] In another preferred embodiment of the present invention, the method introduces a human-computer interaction mechanism to further improve the reliability and applicability of the planning scheme. Specifically, after each round of optimization iteration, the method generates a visualization of the current planning layout. This visualization includes not only a two-dimensional plan view, but also a three-dimensional model to more intuitively display the spatial layout and effect of the planning scheme.

[0131] Preferably, the visualization adopts a multi-level design. At the macro level, the functional area distribution and main road network of the entire industrial and mining town are displayed; at the meso level, the building layout and public space within each functional area are presented; at the micro level, the landscape design and architectural style of a typical block can be displayed. This multi-level visualization method can comprehensively display all aspects of the planning scheme, which is convenient for human experts to evaluate and provide feedback.

[0132] This method will receive feedback from human experts on the current planning layout. These feedbacks may involve multiple aspects, such as the rationality of the functional area layout, the connectivity of the transportation network, the distribution of public facilities, etc. In order to systematically collect and process these feedbacks, the present invention designs a structured feedback form. For example:

[0133] 1. Functional area layout score (1-10 points): __

[0134] 2. Transportation network score (1-10 points): __

[0135] 3. Public facilities distribution score (1-10 points): __;

[0136] 4. Environmental friendliness score (1-10 points): __

[0137] 5. Overall coordination score (1-10 points): __

[0138] 6. Specific suggestions and comments: __;

[0139] Based on the feedback collected, this method adjusts the parameters or constraints of the multi-agent reinforcement learning algorithm. For example, if the expert believes that the distance between two functional areas is too close, a penalty term can be added to the reward function:

[0140] R penalty =-k·max(0,D min -dij ),

[0141] Among them, D min is the preset minimum distance, d ij is the actual distance between the two functional areas, and k is the penalty coefficient. In this way, this method can incorporate the experience and judgment of human experts into the optimization process, thereby obtaining a planning scheme that better meets actual needs.

[0142] The method of the present invention also includes a series of post-processing steps when outputting the final planning scheme. First, the method will post-process the optimized planning layout to eliminate the layout conflicts that may exist between functional areas. This process can be achieved through an iterative adjustment algorithm. For example:

[0143] 1. Identify all pairs of functional areas that overlap or are too close together.

[0144] 2. For each pair of conflicting functional areas, calculate their centroids and move them along the straight line connecting the two centroids until the minimum distance requirement is met.

[0145] 3. If the move results in a new conflict, repeat steps 1 and 2 until there is no conflict or the maximum number of iterations is reached.

[0146] This post-processing can ensure that the final planning scheme is feasible in terms of spatial layout and avoid unreasonable overlap or interference between functional areas.

[0147] Secondly, this method generates a variety of visualizations, including vector drawings, heat maps, and density maps. Vector drawings can clearly show the boundaries of each functional area and the main road network; heat maps can intuitively display information such as population density and economic activity intensity; and density maps can show the spatial distribution of indicators such as building density and green space coverage. These diverse visualizations can show the characteristics of planning schemes from different perspectives, making it easier for decision makers and stakeholders to understand and evaluate the schemes.

[0148] Finally, this method will also output quantitative evaluation indicators of planning layout, including land use efficiency, transportation network density, and functional area distribution balance. For example, land use efficiency can be calculated by the following formula:

[0149]

[0150] Among them, A i is the area of ​​the ith functional area, W i is the weight of the functional area (reflecting its economic or social value). The density of the transportation network can be expressed by the length of roads per unit area:

[0151]

[0152] The balance of functional area distribution can be measured by calculating the standard deviation from the center point of each functional area to the geometric center of the town:

[0153]

[0154] Among them, d i is the distance from the center of the ith functional area to the geometric center of the town, is the average of all distances.

[0155] These quantitative indicators provide an objective basis for evaluating and comparing different planning schemes, and help to select the optimal planning scheme.

[0156] The present invention also provides a multi-agent reinforcement learning industrial and mining town planning layout optimization system corresponding to the above method. The system includes multiple functional modules, each module is responsible for a specific step or function in the method.

[0157] Specifically, the data acquisition module 1 is used to obtain the geographic information system (GIS) data, population distribution data and functional area planning data of the target industrial and mining town. This module can collect necessary information from various data sources to provide basic data support for the subsequent optimization process.

[0158] Digital modeling module 2 is responsible for building digital map models of industrial and mining towns based on GIS data. This module converts complex geographic information into computer-processable digital models, providing a virtual working environment for optimization algorithms.

[0159] Functional area division module 3 divides industrial and mining towns into multiple functional areas based on population distribution data and functional area planning data, and abstracts each functional area into an intelligent agent. This module realizes the transformation from actual geographic space to abstract intelligent agent system, laying the foundation for the application of multi-agent reinforcement learning algorithm.

[0160] The learning environment construction module 4 is responsible for building a multi-agent reinforcement learning environment based on multiple agents. This module defines the state space, action space, and interaction rules between the agents, creating a virtual environment suitable for the operation of the reinforcement learning algorithm.

[0161] The layout optimization module 5 is the core of the system, responsible for optimizing the planning layout of industrial and mining towns using multi-agent reinforcement learning algorithms. This module contains several important sub-units:

[0162] The graph neural network encoding unit 51 is used to encode GIS data into a graph structure representation. This unit can capture the complexity of the urban spatial structure and provide richer input information for the subsequent optimization process.

[0163] The spatial feature extraction unit 52 uses graph convolution operations to extract the spatial features of industrial and mining towns. Through this unit, the system can understand the local and global spatial relationships in the town layout, so as to make more reasonable planning decisions.

[0164] The environmental impact assessment unit 53 is responsible for evaluating the environmental impact of the planning scheme and adjusting the reward function. This unit reflects the system's emphasis on sustainable development and ensures that the optimization results not only meet functional requirements but also meet environmental requirements.

[0165] The sustainability assessment unit 54 is used to predict and assess the long-term sustainability of planning solutions. By simulating the long-term development trend of the town, this unit can help the system make more forward-looking planning decisions.

[0166] The human-computer interaction unit 55 is responsible for receiving and processing the feedback from human experts. This unit enables the system to combine human experience and judgment to further improve the reliability and applicability of the planning scheme.

[0167] Finally, the output module 6 is responsible for generating the optimized industrial and mining town planning layout plan. This module includes three key sub-units:

[0168] The post-processing unit 61 is used to eliminate layout conflicts between functional areas. This unit ensures that the final output planning scheme is feasible in space and avoids unreasonable overlap or interference between functional areas.

[0169] The visualization generation unit 62 is responsible for generating various visualization representations, such as vector drawings, heat maps, and density maps, etc. These intuitive visualization results help decision makers and stakeholders better understand and evaluate planning options.

[0170] The evaluation index calculation unit 63 is used to output quantitative evaluation indexes of the planning layout, such as land use efficiency, traffic network density, and functional area distribution balance, etc. These objective numerical indexes provide a scientific basis for comparing and selecting the optimal planning scheme.

[0171] Through the collaborative work of these modules, the system can achieve complex industrial and mining town planning layout optimization tasks and provide town planners with intelligent decision support tools. The modular design of the system not only enables each functional unit to be optimized and upgraded independently, but also reserves space for possible functional expansion in the future.

[0172] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modification, replacement, and improvement made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A multi-agent reinforcement learning method for optimizing the layout of industrial and mining towns, characterized in that: include: The acquisition steps include: Obtain geographic information system (GIS) data, population distribution data, and functional area planning data for target industrial and mining towns; Processing steps include: Based on the GIS data, a digital map model of industrial and mining towns is constructed; According to the population distribution data and the functional area planning data, the industrial and mining town is divided into a plurality of functional areas, and each functional area is abstracted as an intelligent body; Based on the multiple agents, construct a multi-agent reinforcement learning environment; Use multi-agent reinforcement learning algorithms to optimize the planning and layout of industrial and mining towns; Output steps include: Generate optimized industrial and mining town planning layout plans.

2. The method according to claim 1, characterized in that The processing steps also include: Based on the GIS data, construct an adjacency matrix between intelligent agents; According to the adjacency matrix, defining interaction rules between agents; The element A of the adjacency matrix ij Defined as: Among them, P i and P j Represent the position coordinates of agent i and agent j respectively.

3. The method according to claim 1, characterized in that The method of optimizing the planning layout of industrial and mining towns by using a multi-agent reinforcement learning algorithm specifically includes: Define the state space S of the agent, where S = {s|s∈[0,1]}, s(i) represents the planning state of the i-th functional area; Define the action space A of the agent, where A = {a|a∈{0,1} n }, n represents the total number of functional areas; = define the agent's reward function R(t), where R(t) takes into account factors such as traffic accessibility, environmental carrying capacity, economic benefits, and social relations; The deep Q network DQN algorithm is used to train the agent to learn the optimal planning strategy.

4. The method according to claim 3, characterized in that The reward function R(t) is defined as: R(t)=w1×R accessibility +w2×R environment +w3×R economy +w4×R social , Among them, R accessibility represents the traffic accessibility reward, R environment represents the environmental carrying capacity reward, R economy Represents economic benefit reward, R social represents social relationship rewards, and w1, w2, w3 and w4 are the corresponding weight coefficients.

5. The method according to claim 1, characterized in that The processing steps also include: Based on a graph neural network, the GIS data is encoded into a graph structure representation; Utilize graph convolution operations to extract spatial features of industrial and mining towns; The extracted spatial features are input into a multi-agent reinforcement learning algorithm to optimize the planning layout.

6. The method according to claim 1, characterized in that The processing steps also include: Based on the GIS data, evaluate the environmental impact of the planning scheme; Adjust the reward function of the multi-agent reinforcement learning algorithm based on the results of the environmental impact assessment; Iterate and optimize the planning layout until the preset environmental protection requirements are met.

7. The method according to claim 1, characterized in that The processing steps also include: Construct urban development simulation models to predict the long-term impact of planning options; Assess the sustainability of the planning options based on the long-term impact forecasts; Further optimize the planning layout based on the sustainability assessment results.

8. The method according to claim 1, characterized in that The processing steps also include: After each optimization iteration, a visual representation of the current planning layout is generated; Receive feedback from human experts on the current planning layout; Based on the feedback, adjust parameters or constraints of the multi-agent reinforcement learning algorithm.

9. The method according to claim 1, characterized in that: The output step further comprises: Post-process the optimized planning layout to eliminate layout conflicts between functional areas; Generate a variety of visualizations including vector plots, heat maps, and density maps; Output quantitative evaluation indicators of planning layout, including land use efficiency, transportation network density and balanced distribution of functional areas.

10. A multi-agent reinforcement learning industrial and mining town planning layout optimization system that implements the method described in any one of claims 1 to 9, characterized in that: include: Data acquisition module, used to obtain geographic information system GIS data, population distribution data and functional area planning data of target industrial and mining towns; A digital modeling module, used to construct a digital map model of industrial and mining towns based on the GIS data; A functional area division module is used to divide the industrial and mining town into multiple functional areas according to the population distribution data and the functional area planning data, and abstract each functional area into an intelligent body; A learning environment construction module, used to construct a multi-agent reinforcement learning environment based on the multiple agents; The layout optimization module is used to optimize the planning layout of industrial and mining towns using multi-agent reinforcement learning algorithms; Output module, used to generate optimized industrial and mining town planning layout plan; Wherein, the layout optimization module further includes: Graph neural network encoding unit, used to encode GIS data into graph structure representation; A spatial feature extraction unit, used to extract spatial features of industrial and mining towns using graph convolution operations; Environmental impact assessment unit, used to evaluate the environmental impact of planning schemes and adjust the reward function; a sustainability assessment unit to forecast and assess the long-term sustainability of planning options; A human-computer interaction unit, used to receive and process feedback from human experts; The output module also includes: A post-processing unit for eliminating layout conflicts between functional areas; A visualization generation unit, used to generate a variety of visualization representations; The evaluation index calculation unit is used to output the quantitative evaluation index of the planning layout.

Images