Land utilization optimization method and system coupling bionic intelligent algorithm and multiple agents

Abstract

The invention provides a land utilization optimization method and system coupled with a bionic intelligent algorithm and multiple agents, constructs an economic, social and ecological multi-target scene, and provides an optimization model coupled with spatial leapfrog and multi-stage multi-agents. According to the method, the cooperative competition mechanism of multiple subjects in the agent model is enhanced, and the reasonability of the agent model in the microscopic simulation level is enhanced; on the basis of neighborhood knowledge, opening and closing strategies and reinforcement learning, heterogeneous multi-level main body space perception is mined, and the information transmission process of upper and lower level main bodies is described; based on a scheme-oriented optimization mode, the global optimization capability of the land utilization optimization model of the leapfrog algorithm is improved; the internal and external nested model mode realizes spatialization coupling of multiple models, reinforces the interpretability of the seamless integration process, and provides a new thought for coupling microscopic decision and global optimization land utilization optimization modeling.

Description

Technical Field

[0001] The present invention belongs to the field of geographic information science and technology, and relates to a land use optimization model that combines macroscopic spatial optimization with microscopic decision-making simulation, and couples a bionic intelligent algorithm with multi-agents. Background Art

[0002] The United Nations has included promoting urbanization and sustainable cities as one of its 17 Sustainable Development Goals, aiming to promote green and low-carbon urban development through strengthened urban planning and management. In this context, optimizing land use allocation for multiple objectives, coordinating regional economic and social development with low-carbon ecological conservation and achieving sustainable development, has become a hot topic in urban research. Focusing on the intertwined land use in urban areas, building a comprehensive land use optimization model that balances multiple conflicting objectives to support detailed land use adjustments, has important guiding value for fully tapping land use potential and achieving harmonious development between people and land.

[0003] Land use spatial optimization modeling, as a means of regulating the quantitative structure and spatial layout of land use and guiding the rational spatial allocation of land, requires addressing three issues: 1. Spatial expression and optimal solution search at the macro level; 2. Process expression and agent-driven decision-making simulation at the micro level; and 3. Dynamic coordination between the macro and micro levels. First, macro-level optimization often utilizes biomimetic intelligent algorithms. "Top-down" design and heuristic operator construction, combining geographic information systems with biomimetic intelligent mathematical theory, have shown high performance in solving high-dimensional, nonlinear, and multi-solution land use combination optimization problems. However, existing biomimetic intelligent models employ mathematical knowledge to interpret geographic phenomena. While they offer significant advantages in searching for possible solutions, they oversimplify or even ignore a significant portion of the inherent patterns of land use change during the solution process. Furthermore, these methods lack the ability to express spatially and simulate microscopic evolution, which can easily lead to optimization results that deviate from the common sense framework of planning. Second, at the micro level, cellular automata (CA) and agent-based models (ABMs), with their "bottom-up" approach, are more suitable for expressing microscopic land use change processes. While CA focuses on representing natural processes, ABM emphasizes the behaviors and decisions of multiple stakeholders involved in land allocation. The resulting multi-agent models are widely used to simulate complex systems involving the interactions between diverse human behaviors and heterogeneous environmental landscapes. However, existing multi-agent models inadequately represent the hierarchical and subordinate guidance relationships between agents at different levels. They often focus on modeling decision-making behaviors, rarely designing at the level of the decision-making object, neglecting the spatial non-stationarity of land use development and failing to capture the heterogeneous decision-making receptive fields of multi-level agents. Finally, coupling multiple models to compensate for the shortcomings of a single model is a common approach to multi-dimensional collaborative optimization problems. The coupling method determines the model's stability and interpretability. Existing coupling methods focus on quantitative structure and the effects of parameter adjustment, insufficiently representing synergies at the spatial level, and to some extent hindering the coordinated optimization of multiple models.

[0004] Therefore, it is imperative to establish a land use optimization model that strengthens the spatial expression of macro-optimization, refines the multi-agent behavioral decision-making of micro-models, and achieves dynamic coordination between macro- and micro-scale, global- and local-space. Based on the differentiated perception knowledge and decision-making behaviors of multi-level agents, combined with efficient spatial optimization using biomimetic intelligence, this approach can provide refined optimization strategies for future urban land use planning, thereby offering a possible path to achieving sustainable regional land use spatial planning. Summary of the Invention

[0005] To overcome the deficiencies of the above-mentioned prior art, the present invention provides a land use optimization method and system that couples a bionic intelligent algorithm and multi-agent, aiming to coordinate the multi-objective land use configuration needs that combine regional nature and humanities, simulate the land use decision-making system of multi-level stakeholders, and collaboratively optimize the complex land use structure of urban areas from a global and local, macro and micro dynamic perspective.

[0006] According to one aspect of the present invention, a land use optimization method coupling a biomimetic intelligent algorithm and a multi-agent is provided, comprising:

[0007] Step 1: Collect and process study area data, including land use data, suitability evaluation variable data, and target parameter determination data. Establish a land use optimization decision database.

[0008] Step 2: Clarify the system elements such as knowledge, behavior and rules of multi-agents, and build a multi-level multi-agent model that maps decisions and objects and interacts with multi-level agents.

[0009] Step 3: Based on the suitability evaluation variable data, calculate the land use decision suitability of the multi-level intelligent agents and create a static suitability perception knowledge base of the multi-agents.

[0010] Step 4: Based on theories such as adaptive neighborhood, morphological opening and closing, and reinforcement learning, by constructing spatial perception operators and inherited perception operators, the neighborhood benefits and interactive perception of multi-level intelligent agents are calculated to create a dynamic environmental perception knowledge base for multiple intelligent agents.

[0011] Step 5: Based on the behavior and decision-making characteristics of multiple agents, calculate the expected utility of the public agent, the competitiveness of the department agent, and the decision-making utility of the government agent as the basis for the agent's decision-making behavior.

[0012] Step 6: Calculate the selection probability of each agent, quantify the mathematical expression of the agent's decision-making behavior, and build a probabilistic selection mechanism.

[0013] Step 7: Spatially express the existing land use spatial layout plan and initialize the bionic intelligent group (frog group).

[0014] Step 8: Through the spatial dynamic learning operator, the heuristic process of the spatialized frog leaping algorithm is constructed, and the optimal solution search and position update with the participation of multiple intelligent agents are carried out from both global and local dimensions, forming a learning optimization stage.

[0015] Step 9: Construct the mutation optimization operator of the spatialized frog leaping algorithm to form the mutation optimization stage to deal with position updates outside the search domain.

[0016] Step 10: Construct an optimized multi-objective function from economic, social, and ecological perspectives and determine the target coefficients.

[0017] Step 11: Establish a constraint system for land use optimization from three aspects: quantity structure, laws, regulations and policies, and land use conversion rules.

[0018] Step 12: Combine the optimization objectives and the constraint system, alternately run the learning operator and mutation operator of the leapfrog algorithm, and output the results of the land use spatial layout optimization.

[0019] As a preference, step 1 of data processing mainly includes: (1) performing unified projection transformation on spatial data, converting it into a unified resolution through resampling method, and forming a data set with consistent spatial coordinate system and resolution; (2) performing secondary classification and sorting on land use data, simplifying the model, and highlighting coupling design.

[0020] As a preferred method, step 2 of constructing the multi-agent knowledge, behavior, and rule systems mainly includes two core parts:

[0021] (1) Expression of the mapping relationship between decision-making and objects. Focusing on the functions of stakeholders in the decision-making process, we abstracted the subject of land use and designed a decision-making agent to simulate the information interaction of decision-making among stakeholders in real scenarios. Based on the idea of ​​self-organizing neighborhoods, we designed an object agent to represent the differentiated perceptions of stakeholders at the local scale. Decision-making and object agents at the same level directly interact with each other, forming a strong mapping relationship. Information interaction between agents at different levels is a weak mapping relationship.

[0022] (2) Multi-level agent division and information transmission. Based on the hierarchical relationship within the agent, the decision-making and object agents are divided into multi-level relationships mapped to "individual-department-government" and "unit-type-scheme" respectively. There is information exchange between the hierarchical and sub-level agents in the form of land use schemes.

[0023] As a preferred method, the land use decision suitability evaluation method of the multi-level intelligent agent in step 3 is as follows:

[0024] Suitability refers to the degree to which a land unit is suitable for the land use of a specific subject, and is calculated using a weighted approach. The formula is as follows:

[0025] K suit =K T＝k (i,j)=α k *A1+β k *A2+…+γ k *A m ,α k +β k +…+γ k =1

[0026] In the above formula, public suitability evaluation is defined as K T＝k (i, j), represents the influence of adaptive knowledge on the land use T = k selected by land unit (i, j), which is determined by multiple decision factors A m The difference in the attention paid by different types of public entities to land use forms a decision-making preference. Therefore, a corresponding decision-making factor weight system is formulated based on different entities, namely α k wait.

[0027] As a preference, the dynamic environment perception knowledge base of the multi-level intelligent agent in step 4 relies on the calculation of the spatial perception operator and the inheritance perception operator. The operation of these two operators is as follows:

[0028] (1) Spatial perception operator. In order to optimize the spatial aggregation of land use spatial configuration, the present invention constructs four spatial perception operators at two levels: unit and type. The unit spatial function operator is mainly oriented towards the grid unit of land use, and realizes the dynamic perception of the neighborhood environment of the public intelligent agent in the decision-making subject function. It can be divided into quantitative structure perception operator and spatial topology perception operator. The formula is as follows:

[0029]

[0030] In the above formula, the number structure of units and the perception of spatial topology are respectively represented by K num and K topology Indicates that N k (i, j, length) is the number of land units with land use k in the neighborhood with land unit (i, j) as the center and side length as length, D k represents the distance from the center unit to all land units of land use k in the neighborhood, D max The spatial perception operator re-emphasizes the shape and neighborhood relationship of the patch, and enhances the aggregation by adjusting the edge units to reduce the effect of landscape fragmentation. It can be divided into patch core edge operator and patch morphology opening and closing operator according to the operation mode:

[0031]

[0032] K Open-Close =K Open *K Clos e

[0033]

[0034] In the above formula, the patch core edge operator and the morphological opening and closing operator are respectively represented by K Core-Edge and K Open-CloseIndicates that Dis is the shortest distance from the grid to the edge of the patch, and Limitation is the distance threshold for activating the core edge operator. K Open and K Close The opening operation is mainly used to fill the gaps and cracks between the patches, so its magnification coefficient C Open It is often greater than 1, and the closing operation is longer than removing scattered patches, and its corresponding reduction coefficient C Close Between 0 and 1, and satisfy C Open *C Close =1.

[0035] (2) Inherited perception operator. The inherited perception operator adjusts the decision-making behavior by characterizing the expected benefits of the agent for making decisions. Benefit is a psychological process. Individuals will draw on past experience as the main basis for evaluating current benefits. This feature is very suitable for characterization by reinforcement learning. The present invention adopts the reinforcement learning idea of ​​Roth-Erev, constructs an improved reward function Reward, expands the process of spatial knowledge inheritance learning, and then corrects the current agent's perception of the benefits of behavior, which includes the decision results of the previous plan and the current benefits. Since the heuristic operator depends on the differences in the decisions of different subjects, the previous decision results will directly guide the current decision. According to the first law of geography, adjacent elements have similar characteristics, and the behavior of neighboring units is also instructive. Similar to the herd mentality, the decision-making subject also tends to make decisions similar to his neighbors. Areas with large differences will become hot spots for land changes in this stage. In order to reflect the impact of environmental knowledge on the evaluation of the agent's benefits, the study introduces the quantitative structure of units and the spatial topology perception operator to modify the benefit function. The formula is expressed as follows:

[0036]

[0037] Among them, EB t (a i ) represents the expected benefit of land use type i in unit j at time node t, and Reward is the benefit function that includes environmental knowledge. is the legacy parameter, which indicates the degree to which the agent retains its past experience; δ∈[0,1] represents the degree of influence of the hypothetical benefits of the unselected strategy, which characterizes the learning intensity of the agent on other learning object schemes; I(i,y) is the selection index parameter, which is 1 when the land use selection tendency y is the same as the current land type i, otherwise it is 0. i Indicates the decision result of land use type i in the current unit in the previous scenario. If the decision exists, it is 1, otherwise it is 0. K num and K topologyRespectively, they represent the quantitative structure and spatial topology of the environment surrounding the decision. At t = 0, the agent's expected benefit of land use type i in unit j is derived from its perception of the multi-objective benefits, corresponding to the objective function. Furthermore, a random interference term δ is introduced into the formula to reflect the blindness of subjective decision-making.

[0038] As a preferred approach, the behavior and decision-making knowledge system of the multi-level agents in step 5 can be summarized as follows: Based on the perceptual knowledge in the knowledge base, the multi-agents calculate the expected utility U in a weighted manner, use a discrete choice model to form the selection probability P, generate decision-making behavior D under the guidance of the selection mechanism, generate decision plans, and interact with the object agents to realize land use changes at the spatial level. The sources of the knowledge base include static adaptive knowledge, dynamic perceptual environment knowledge E, inherited knowledge K, and restrictive conditions Res. Therefore, the expected utility U is calculated as follows:

[0039] U=α*K suit +β*E agent +γ*K inherit +δ*U others

[0040] In the above formula, α and others are weights, U others represents the expected utility or competitiveness of other agents. Because each agent has a different decision-making model, several elements in the formula are selectively activated, as shown in Table 1.

[0041] Table 1 Multi-level multi-agent decision-making behavior and knowledge system

[0042]

[0043]

[0044] Preferably, the selection probability of the multi-level agent in step 6 is based on the expected utility, and the random influence factor ε conforms to the Weibull distribution. Therefore, the present invention adopts a discrete choice model to evaluate the selection probability, and the calculation method is as follows:

[0045]

[0046] In the above formula, U T =k(i, j, t)≥U T = k(i′, j′, t) represents the probability that agent k's expected utility for land unit (i, j) is greater than or equal to that of other selectable land units (i′, j′). This paper addresses the differences in decision-making behaviors among different agents by constructing two decision-making mechanisms: Monto Carlo random selection and roulette wheel.

[0047] Preferably, the initialization of the spatialized frog leaping algorithm in step 7 primarily involves performing a mutation operation on the current land use situation to generate several differentiated land use scenarios as the original frog swarm. To facilitate information exchange between frogs, the present invention employs the principle of mixed sorting and grouping for initialization. Specifically, before each iteration, the frog swarm is divided into several subgroups based on the order of the objective function values ​​corresponding to each frog, thereby organizing a local optimization within a limited search range. The frog swarm reorganizes after iterative optimization, and the process is repeated until convergence.

[0048] As an optimization step, the learning optimization phase of the spatialized frog leaping algorithm in step 8 includes a local optimization phase and a global optimization phase:

[0049] (1) Local optimization stage: Local optimization is the optimization within the subgroup. The learning operator is used to obtain the knowledge of the local optimal frog to improve the position of the worst frog. The present invention uses a multi-level agent spatial optimization process to improve the random update strategy of the original learning operator and form a spatial dynamic learning operator. The improved worst solution position optimization formula is:

[0050]

[0051] Among them, P new represents the updated position of the worst frog, P best and P worst Corresponding to the spatial solutions of the local best frog and the local worst frog respectively, Step represents the step length. If the new position after local optimization is better than the previous position, the worst frog individual is updated. Otherwise, the model will use G best Instead of P best , entering the global optimization stage.

[0052] (2) Global optimization stage: Global optimization is the optimization among frog populations. The improved spatial dynamic learning operator is also used to obtain the knowledge of the global optimal frog to improve the position of the worst frog. The improved worst solution position optimization formula is:

[0053]

[0054] The parameters of global optimization are similar to those of local optimization, and the learning object is modified to the global optimal frog G best If the result of global optimization still cannot improve the frog's target benefit, it will enter the mutation optimization stage.

[0055] As a preferred approach, the mutation optimization phase of the spatialized frog leaping algorithm in step 9 primarily involves generating a new frog within the search range to replace the global optimal solution as the learning object, known as mutation. Integrating this with the land use optimization task, a mutation optimization operator was developed that randomly selects several cells for mutation. To prevent excessive fragmentation, the mutated land use type is the one that occurs most frequently in the Moore neighborhood of the current grid cell.

[0056] As a preference, the construction of the objective function for land use optimization in step 10 includes maximizing economic, social and ecological benefits, and the calculation method is:

[0057]

[0058] Among them, f Economic It reflects the financial benefits of land use in the actual production process and promotes regional development by maximizing the economic benefits of land use. Eco(k) represents the economic output corresponding to land use type k (yuan / hm2). 2 ), Area(k) is the corresponding land area. Social Maximize land compactness, thereby improving the utilization rate of infrastructure. This paper uses the neighborhood method to quantify social benefits, Ω represents the Moore neighborhood of the unit, A ijk represents the number of type k land uses in the neighborhood, x ijk is a binary variable representing the discrimination of unit land use type k, and N Ω is the total number of cells in the neighborhood Ω. Ecological As an ecological benefit maximization equation, it reflects the positive and negative impacts of quantitative structure and project layout on the natural environment. We have formulated a hybrid ecological goal that combines ecosystem service value and low carbon. Among them, ESV(k) represents the unit ecosystem service value corresponding to land use type k, Carbon(k) is the low carbon coefficient (tC / hm 2 ).

[0059] As an optimal method, the constraints in step 11 include three parts: quantity structure, laws, regulations and policies, and land use conversion rules. The mathematical expression and calculation method are as follows:

[0060]

[0061] In the above formula, the mathematical expression of the constraints, Res, is divided into mandatory and non-mandatory constraints. Mandatory constraints are mainly related to the management of quantity structure and laws and regulations, while non-mandatory constraints are mainly the adjustment of selection probabilities by conversion rules, with the adjustment coefficient P ranging from 0 to 1.

[0062] According to one aspect of the present invention, a land use optimization system coupling a bionic intelligent algorithm and a multi-agent is provided, which is used to implement the steps of the land use optimization method coupling a bionic intelligent algorithm and a multi-agent.

[0063] According to one aspect of the present invention, a non-transitory computer-readable storage medium is provided, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions enable the computer to execute the steps of the coupled bionic intelligent algorithm and multi-agent land use optimization method.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] The present invention constructs an economic, social and ecological multi-objective scenario, and proposes a coupled spatialized frog leaping and multi-level multi-agent optimization model (SSFLA-MLAS). Specifically, from the two dimensions of decision-making behavior and objects, a multi-level multi-agent model (MLAS) in which mapping and multi-level information transmission coexist is proposed; based on a solution-oriented optimization model, a spatialized frog leaping algorithm (SSFLA) is constructed by reconstructing the heuristic operator strategy; and a dynamic coupling method of internal and external nested models with a bionic intelligent algorithm as the basis and multi-agent as the core is proposed.

[0066] The present invention proposes a multi-level multi-agent model in which mapping and multi-level information transmission coexist, which strengthens the cooperative competition mechanism of multiple agents and enhances its rationality at the micro-simulation level. Based on neighborhood knowledge and opening and closing strategies, the present invention explores the spatial perception of heterogeneous multi-level agents; at the same time, the application of reinforcement learning in inheritance perception depicts the information transmission process between superior and subordinate agents; based on the solution-oriented optimization model, the quantitative structure optimization of the frog leaping algorithm is extended to the spatial layout level, improving the global optimization capability; the proposed dynamic coupling method of internal and external nested models with the bionic intelligent algorithm as the basis and multi-level multi-agent as the core realizes the spatial coupling of multiple models and enhances the interpretability of the seamless integration process. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, a brief introduction will be given below to the drawings used in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0068] Figure 1 An overall flow chart of the coupled bionic intelligent algorithm and multi-agent land use optimization method provided by an embodiment of the present invention;

[0069] Figure 2The land use distribution map of Huangpi District, Wuhan City provided in the embodiment of the present invention;

[0070] Figure 3 A diagram of a multi-level multi-agent structure framework provided by an embodiment of the present invention;

[0071] Figure 4 (a)-(d) are multi-level multi-agent spatial perception operation operators provided by an embodiment of the present invention;

[0072] Figure 5 A technical roadmap for multi-level, multi-agent implementations of the present invention;

[0073] Figure 6 A technical roadmap for the spatialized frog leaping algorithm provided by an embodiment of the present invention;

[0074] Figure 7 A schematic diagram of a spatial dynamic learning operator provided by an embodiment of the present invention;

[0075] Figure 8 The three-district and three-line division of Huangpi District, Wuhan City provided in the embodiment of the present invention;

[0076] Figure 9 (a)-(b) are land use optimization target and multi-model comparison curves provided by the embodiment of the present invention, where (a) is the optimization multi-target convergence curve and (b) is the optimization multi-model comparison curve;

[0077] Figure 10 This is a land use optimization spatial layout diagram provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0078] It should be noted that:

[0079] The terms "including" and "having" and any variations thereof in the description and claims of the present invention and the above-mentioned drawings are intended to cover non-exclusive inclusions, for example, a process, method, system, product or apparatus that includes a series of steps or units is not necessarily limited to the steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0080] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically separate entities. That is, these functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices. The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily need to be executed in the order described. For example, some operations / steps may be further decomposed, while others may be combined or partially combined, so the actual execution order may vary depending on the actual situation.

[0081] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In addition, the technical features in the various embodiments or single embodiments provided by the present invention are arbitrarily combined with each other to form a new technical solution. This combination is not restricted by the sequence of steps and / or structural composition mode, but must be based on the ability of ordinary technicians in this field to implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that this combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0082] See also Figure 1 , the technical solution adopted by the present invention includes the following steps:

[0083] Step 1: Take Huangpi District, Wuhan City, Hubei Province as the research area and 2020 as the research time point. Figure 2 As shown. Land use data, suitability evaluation variable data and target parameter determination data were collected. The adaptability evaluation variable data include 29 spatial variable data from natural and human aspects such as topography, soil endowment, and transportation accessibility, as shown in Table 2. The target parameter determination data include statistical data, net primary productivity (NPP), meteorological and soil data, mainly from statistical yearbooks and remote sensing and geographic data platforms. The data processing mainly includes: (1) using Arcgis10.2 software, uniformly projecting spatial raster data to the CGCS2000 projection system and converting it to a spatial resolution of 30m; (2) performing a secondary classification of land use and merging it into six types of land use: cultivated land, forest land, grassland, construction land, water area, and unused land.

[0084] Table 2 Data sources

[0085]

[0086]

[0087] Step 2: Mapping relationship between decision and object and information transmission between multi-level agents, such as Figure 3 shown.

[0088] (1) The mapping relationship between decision-making and objects can be expressed as follows: the behavioral intention of the decision-making subject is expressed spatially as the change of grid cells, and the environmental change feedback caused by the evolution of land use space forms the update of the decision-making subject's knowledge base at the functional level. The two types of agents at the same level will directly interact with each other and form a strong mapping relationship. The information interaction between agents at different levels affects the decision-making in an indirect way. For example, in addition to the patch aggregation effect of the focusing type agent, the unit effect of the patch will also provide a reference for the decision-making of the department agent through the decision information transmission of the public agent, forming a weak interaction. In order to strengthen the environmental perception connection in the mapping, the model designs a spatial perception operator representation, see step 5.

[0089] (2) The information transmission of multi-level agents is divided into two levels: decision-making and object. In the decision-making agent, based on the actual approval process of land changes, the government subject in a broad sense is split and refined, and divided into functional departments mainly responsible for formulating planning schemes and government approval departments responsible for approving and coordinating non-land resources according to the specific work functions. According to their respective responsibilities, departmental agents can be further divided into agricultural departments that mainly manage cultivated land, housing and construction departments that are responsible for construction land, and forestry and grassland departments that are responsible for forest land and grassland. MLAS abstracts the public entity as the actual user on the land unit, including land-using units, developers, practitioners and residents. Based on the above theory, the present invention defines a three-level decision-making agent system of public-department-government. Among them, the public agent represents the local stakeholders, mainly relying on its own perception of the local environment to make decisions and pursue the maximization of its own individual interests; the departmental agent is at the middle level, responsible for organizing the land use function zoning and formulating land use layout plans suitable for its department; the government agent is the overall decision-making subject, playing a control and coordination role in the multi-level multi-agent system. The lower-level intelligent body will feedback its decision-making needs to the upper-level in the form of a plan, and the upper-level intelligent body will distribute the land use planning after the decision step by step to guide the public entities to complete the actual land use changes.

[0090] According to the scale effect of the spatial layout of land use, the present invention defines a three-level object agent system of unit-type-scheme. Among them, the unit agent is oriented to the optimization of micro-grid units, perceives the characteristics and potential of land units, and feeds back to the public agent; the type agent is oriented to the optimization of land use patches, reveals the agglomeration characteristics of similar land patches at the meso-scale, perceives the overall regional environment under a specific type, and forms domain knowledge as the basis for departmental agent decision-making. The scheme agent is oriented to the global scheme optimization at the macro-scale, coordinates land use conflicts between types and between units and types, perceives the dynamic changes in quantity structure and spatial position, considers the constraints of laws, regulations, policies and conversion rules, and forms expected utility to feed back to the government agent. In the object agent, the subordinate agent is a component of the superior agent, and the superior agent needs to coordinate conflicts between multiple subordinates.

[0091] Step 3: In determining the agent's suitability knowledge, this invention selects 29 spatial variables from natural and human aspects, such as topography, soil endowment, and transportation accessibility, as the basis for multi-agent decision-making. Topography, soil endowment, and natural climate are closely related to vegetation cover and are the main factors affecting the spatial distribution of cultivated land, forests, and grasslands. Topography and slope characteristics determine the suitability of land for agricultural cultivation or ecological protection. The quality of soil endowment directly affects vegetation growth, thereby reflecting the land's agricultural potential. The healthy development of agriculture and forestry is highly dependent on the water and heat supply of the natural climate. In urban planning, some studies have shown that economic and social development potential, mainly based on GDP and population density, is the primary driving force in determining city size. The level of infrastructure construction, such as hospitals and schools, and transportation accessibility determine the intensity of regional development, and accessibility reflects the quality of the living environment. In addition, due to the differences in perception among multi-level agents, a differentiated design is adopted for the suitability evaluation of the same type of land use. Agents at the public level are influenced by individual preferences and focus on knowledge related to their own interests. The convenience of farming brought about by the interaction between farming and housing is the focus of individual farmers' concern. Urban residents pay more attention to the transportation convenience of housing. Intelligent entities at the department level have more macro and comprehensive land use evaluation criteria. With the intensification of the negative impact of urbanization, ecological security has become the primary goal of sustainable development. Ecosystem services characterize the relationship between socioeconomic and ecological factors in urban renewal from multiple angles, and are of great significance to improving urban health and enhancing vegetation cover. Therefore, it has become an important indicator for forestry and grassland departments to evaluate ecological land. Its calculation method can be found in the ecosystem service coefficient in the determination of the target coefficient in step 9. The present invention uses the entropy weight method to determine the weight of each indicator. The specific parameters are shown in Table 3.

[0092] Table 3 Multi-level intelligent agent land use suitability weight coefficient table

[0093]

[0094]

[0095] Step 4: Multi-level multi-agent spatial perception operator, such as Figure 4 As shown in (a)-(d). The unit space perception operator uses a 7×7 neighborhood as the range of unit neighborhood perception. The type space perception operator designs a patch edge operator based on the edge effect theory to distinguish the core and edge of the patch, and land use changes are only allowed to occur in the edge area. The threshold for distinguishing between the boundary and the core in the invention is 1, that is, a Moore field representation at the patch level is constructed. In order to improve the morphological structure of the patch and construct a more compact land use pattern, the present invention applies the expansion and erosion theory in morphology, optimizes the patch morphology through opening and closing operations, and constructs a type morphological opening and closing operator.

[0096] The inherited perception operator of multiple levels and multiple agents forms the corresponding inherited perception knowledge, which is one of the foundations of agent decision-making. It also reflects that agents acquire inherited and better knowledge through information exchange with agents at other levels. This interaction can be information transfer between physical agents, such as departmental agents adopting public opinion in their decisions, or it can be an abstract vision, such as the public's ideal demand for a work and living environment, which corresponds to the optimal solution in the frog leaping algorithm. This paper refers to the sensitivity verification structure in reinforcement learning in some related research and determines the learning intensity δ = 0.4.

[0097] Step 5: The behavior and decision-making knowledge system of multi-level multi-agents can be summarized as a set of paradigms, in which each agent uses an equal weight design to evaluate its multi-knowledge when calculating its expected utility or competitiveness U. Due to the differences between agents, different decision-making systems are formed, which are shown in Figure 5 Specifically:

[0098] (1) The public agent will transform the land use unit type based on its own knowledge and preferences in the optimization process, with the goal of maximizing its own interests. Therefore, the public agent's decision-making knowledge base includes the public suitability knowledge K that represents its own preferences. suit , environmental perception knowledge E obtained by interacting with the unit agent PUB and the inherited perceptual knowledge K obtained by interacting with the intelligent agent of the learning object in the frog leap model inherit After completing its own land use selection decision, each public agent compiles an application plan and submits it to the department agent, waiting for the department agent's feedback. Public adaptability is calculated using a weighted approach, see step 3. The unit agent's environmental perception is calculated using the unit space perception operator, with the following structure:

[0099]

[0100] In the above formula, Indicates that in the t-th iteration cycle, the unit agent k centered on the land unit (i, j) perceives the environment, and the quantitative structure perception K num and spatial topology-aware K topology The calculation of the two parts is shown in step 4. The inherited perception knowledge of the public agent comes from the optimal decision solution in the historical space.

[0101] (2) Based on the evaluation results of the competitiveness of the departments, the departmental agent pursues the goal of maximizing the interests of the departments, plans key development areas and submits multiple land use layout plans to the government to guide the type transformation and spatial migration of land use. The departmental agent is divided into three categories: agricultural department, housing and construction department, and ecological department, and the departments compete with each other. The agricultural department agent is responsible for formulating the layout plan of cultivated land and gardens; the construction department agent is responsible for formulating the layout plan of urban and rural construction land; the ecological department agent is responsible for formulating the layout plan of forest land and grassland. The decision-making knowledge base of the departmental agent includes the departmental suitability knowledge K suit , environmental perception knowledge E obtained by interacting with type agents DEP and the inherited perception knowledge K obtained by interacting with public agents inherit Each type of departmental agent will form a land use plan that meets its own demands, summarize it into a set of application plans and submit it to the government agent, waiting for the government agent's feedback. Departmental suitability mainly represents the degree of adaptability of land use from the perspective of a specific land-using department. Due to the differences in departmental subject domain knowledge and functions, the tendency of land use layout is generated. Each department will form its own evaluation system for the type layout it is interested in, which is reflected in the design of decision factor weights. Since departments have more comprehensive domain knowledge than the public, their functions also require departments to consider the interactions between contiguous spatial units. Therefore, the environmental perception of departmental agents focuses on considering the agglomeration effect at the map level. The environmental perception of type agents is calculated using the type space perception operator, and the structure is as follows:

[0102]

[0103] In the above formula, Represents the type of intelligent agent's perception of the environment k, which is represented by the core-edge operator K core-edge and morphological opening and closing operator K open-close Two-part calculation. From the perspective of departmental agents, departmental decisions need to take into account public sentiment. Therefore, their decisions inherit public preference knowledge. In the specific interaction process, departments review land use applications submitted by the public and formulate planning schemes based on their own demands.

[0104] (3) The decision-making behavior of the government agent is to review the land use applications submitted by the department agents based on the decision utility and coordinate the land use application conflicts of multiple departments. The government agent is the highest-level decision-maker for regional land use optimization. There is only one government agent. The knowledge base of the government agent includes the background knowledge of other subjects' decisions: the public's expected utility U PUB Competitiveness with controversial sectors U DEP , and we need to consider many constraints K res Evaluation indicators of regional overall development goals O, land use applications submitted by the public and departmental intelligent bodies D PUB and D DEP etc., will serve as a guide for government decision-making GOV The government agent will review the land use requirements unit by unit, approve the most appropriate plan and reject the remaining requests. The government agent will also guide departments and the public to implement land use changes based on the final plan. The final optimal land use plan will be used as the updated frog P new Continue to iterate the optimization.

[0105] Step 6: The multi-level multi-agent selection mechanism is associated with the decision-making behavior characteristics of the agents. The decision logic of the public and departments adopts the Monto Carlo random selection process, which is as follows:

[0106]

[0107] In the expression, D PUB (i, j, t) is the behavior decision of the public agent on the land unit (i, j) in the t-th iterative cycle process, D DEP (i, j, t) is the decision result of the department agent, 1 means applying to change the land use type, 0 means applying to maintain the current land use type, r is a random number in the range of 0-1, k now is the land use type of the land unit (i, j) at the current time t, P max (i, j, t) represents the probability of selecting the land class with the highest utility among all land classes. Due to their tendency to maximize benefits, the public will only apply for land change when the benefits exceed their existing benefits. Unlike public decision-making, due to competition, each department serves its own functions and independently demarcates key development areas, thus preventing the comparison of conditions across multiple land classes within the same unit. The government is responsible for coordinating land use conflicts, and its decision-making must reflect fairness in the selection of alternatives. Therefore, the roulette wheel decision-making approach adopted in the model is more reasonable. In a roulette wheel decision-making process, land use types with higher probabilities have a stronger allocation advantage, while other land use types with lower probabilities still retain some allocation possibilities.

[0108] Step 7, Step 8, Step 9: The present invention uses Python3.8 programming language and multiple open source libraries (including NumPy, SciPy and Pandas) to realize the model construction of SSFLA-MLAS. The case study selected 2020 as the research time point, extracted Huangpi District of Wuhan City as the research area, and generated a grid containing 1468×2570 units with an accuracy of 30m*30m. The example determined the basic parameters of the frog leaping algorithm: the population size is M=50, the number of individuals in the population is N=32, and the frog group size of P=M×N=1600 was established. After multiple observations and adjustments, the number of evolutionary times within the group was finally determined. P =10, global iteration number Epoch G =600. The mutation rate of the mutation optimization operator is set to 0.1%. The technical route of the spatialized frog leaping algorithm is shown in Figure 6 In the case of the frog leaping algorithm coupled with the spatial dynamic learning operator of the multi-level intelligent agent, Figure 7 shown.

[0109] Step 10: The target coefficients for land use optimization must be accurately determined based on actual conditions. Economic objectives reflect the financial benefits of land use options. Based on insights from statistical yearbooks and other relevant research, the economic target coefficients are determined by summarizing the economic benefits per unit area of ​​land use and the gross domestic product of the secondary and tertiary industries. Social objectives focus on land compactness. Compact land use patterns can increase access to public services and improve resource efficiency, thereby promoting large-scale economic development and enhancing the quality of life. Ecological objectives reflect the impact of the quantity structure and project layout on the natural environment. Ecosystem service value is a commonly used method to measure the ecological benefits of land use. The ecosystem service coefficient is based on the research results of Xie Gaodi, and the specific calculation method is shown in Table 4.

[0110] Table 4 Ecosystem service calculation methods

[0111]

[0112] Based on the carbon cycle mechanism, low-carbon development goals can be divided into two parts: emission reduction and carbon sink increase. Therefore, the low-carbon calculation of urban areas should include two aspects: CO2 emissions and CO2 sinks. In terms of carbon emission assessment, carbon emissions are mainly in the form of industrial and agricultural production emissions, which are specifically manifested in the human intervention of crops during their growth cycle and the burning of fossil energy. Carbon emissions from agricultural sources are mainly reflected in the use of fertilizers, pesticides, agricultural machinery, agricultural irrigation, and mulch film consumption. The inventory method is adopted for statistics in this invention, and the calculation formula is expressed as follows:

[0113] E g =A*G f +B*N p+(C*A m +D*W m )+E*P i +F*M a

[0114] Where E g It represents the carbon emissions per unit of cultivated land (tC), and A~F are the carbon emission coefficients of related agricultural activities, and each coefficient corresponds to a carbon emission estimation indicator. Carbon emissions from construction land are mainly from the combustion of fossil fuels. The carbon emissions caused by this are estimated using the energy inventory method. Referring to the energy emission coefficients of the IPCC, nine fuels including raw coal, coke, crude oil, gasoline, kerosene, diesel, fuel oil, liquefied petroleum gas and natural gas are selected. Based on the annual fuel consumption in the 2020 Statistical Yearbook of Wuhan City, the carbon emissions of fuel consumption are calculated by converting it into standard coal. The calculation method is as follows:

[0115]

[0116] Where CE represents the carbon emissions of energy consumption (10 6 tC); E i is the annual consumption of type i fuel (10 6 t); = represents the conversion coefficient for standard coal; Ki is the carbon emission coefficient of the fuel. Carbon sinks are calculated using the net ecosystem productivity (NEP), and this study is based on NEP inversion from remote sensing images. To be consistent with carbon emissions calculations, carbon sinks need to be converted into CO2 sinks. The calculation formula is as follows:

[0117]

[0118] R h =0.4679×R s +114.42

[0119]

[0120] in It represents the CO2 sink of a region (tC), which is calculated from the NEP of each unit in the region. NPP is the net primary productivity of terrestrial ecosystems, and R h Represents the heterotrophic respiration of soil, which is determined by the soil respiration R s The model is fitted by the monthly mean temperature T (℃), monthly precipitation P (cm) and soil organic carbon density SOCD (kg / m 2) driven by carbon emissions. According to the low-carbon calculation formula, Huangpi District's carbon emissions in 2020 were 1.281 million tons, of which 61.54% came from fuel combustion, totaling 0.789 million tons. Raw coal produced the largest emissions, followed by coke and crude oil. Based on this, the low-carbon development target parameters were determined, as shown in Table 5. The summarized coefficients for each target are shown in Table 6. To ensure synergy in multi-objective optimization, the case study assigned equal weights to each sub-target.

[0121] Table 5 Low-carbon development target parameters

[0122]

[0123]

[0124] Table 6 Land use optimization target coefficient table

[0125]

[0126] Step 11: Constraints can provide clear restrictions and guidance for the formulation and implementation of land use planning. Quantity structure constraints and spatial constraints are planned in this invention. Quantity structure constraints are formulated based on the social and economic development goals of Huangpi District, Wuhan City during the planning period, and spatial constraints are set based on spatial planning zoning or planning policies, see Table 7. The quantitative structure constraints of land use are used to control the extent of changes in various types of land for the sake of land resource protection and minimizing the cost of land use changes. In order to minimize the economic losses and environmental damage caused by land changes, the study controls the upper and lower limits of quantitative structure changes within 10%. The constraints of laws, regulations and policies better reflect the planning policy guidance under real scenarios. Based on the 2020 land use master plan of the Wuhan Municipal Government, the future development direction of the region has been formulated, including the central city optimization construction area, key towns and industrial construction areas, basic farmland concentration areas and ecological protection areas, such as Figure 8 As shown in Table 8, land use conversion rules reflect the difficulty of land use change under the guidance of the inherent attributes of land use. They are represented by a transfer matrix, with larger values ​​indicating greater conversion difficulty. The coefficients were determined based on the experience of local urban planning experts and relevant scholars, as shown in Table 8.

[0127] Table 7 Land use spatial constraint knowledge

[0128]

[0129]

[0130] Table 8 Land use transfer matrix

[0131] arable land woodland grassland waters construction land Unused land arable land - 0.3 0.6 1 0.1 1 woodland 0.7 - 0.89 1 0.97 1 grassland 0.6 0.2 - 1 0.65 1 waters 1 1 1 - 1 1 construction land 0.85 0.7 0.9 1 - 1 Unused land 0.65 0.4 0.4 1 0.2 -

[0132] Step 12: Output the land use layout optimization results. The SSFLA-MLAS model has a significant optimization effect, such as Figure 9 (a)-(b) show. The economic, social and ecological target benefits increased by 5.01%, 2.87% and 0.74% respectively, and the regional net carbon sink increased by 1.38%, indicating that under the multi-objective guidance of low-carbon development, the coupling model achieved ideal optimization results. The optimization effect and efficiency of SSFLA-MLAS are better than those of a single algorithm, and the improvement of multi-objective benefits is 3.4% and 32.5% higher than that of SSFLA and MLAS, respectively, demonstrating the rationality of the nested coupling design at the functional and result levels. After optimization, the land use landscape pattern is more compact and reasonable, as shown in Table 9. The spatial layout presents a scenario that matches the regional development strategy, see Figure 10 , which shows that MLAS containing mapping relationships and multi-level transmission mechanisms can effectively improve the support capability for regional planning decisions.

[0133] Table 9 Comparison of landscape patterns of multi-model optimization results

[0134]

[0135]

[0136] It should be understood that parts not elaborated in detail in this specification belong to the prior art.

[0137] The implementation of each embodiment of the present invention is based on programmed processing by a device with processor functionality. Therefore, in practical engineering, the technical solutions and functions of each embodiment of the present invention are encapsulated into various modules. Based on this reality, and in addition to the aforementioned embodiments, an embodiment of the present invention provides a coupled bionic intelligent algorithm and multi-agent land use optimization system. This system is used to implement the coupled bionic intelligent algorithm and multi-agent land use optimization method described in the aforementioned method embodiments.

[0138] The system includes: a first main module for collecting and processing study area data, including land use data, suitability evaluation variable data and target parameter determination data;

[0139] The second main module is used to clarify the system elements of multi-agents, build a multi-level multi-agent model that maps decisions and objects and information interaction between multi-level agents. Based on the constructed multi-level multi-agent model, it also includes: calculating the suitability of land use decisions of multi-level agents based on suitability evaluation variable data, and creating a static suitability perception knowledge base of multi-agents; based on the theories of adaptive neighborhood, morphological opening and closing, and reinforcement learning, by building spatial perception operators and inherited perception operators, calculating the neighborhood benefits and interactive perception of multi-level agents, and creating a dynamic environmental perception knowledge base of multi-agents; based on the behavior and decision-making characteristics of multiple agents, calculating the expected utility of public agents, the competitiveness of departmental agents, and the decision utility of government agents as the basis for the decision-making behavior of agents; calculating the selection probability of each agent, quantifying the mathematical expression of the decision-making behavior of agents, and building a probabilistic selection mechanism;

[0140] The third main module is used to spatially express the existing land use spatial layout plan based on land use data and initialize the bionic intelligent group. Through the spatial dynamic learning operator, it constructs the heuristic process of the spatial frog leaping algorithm, searches for the optimal solution and updates the position from both global and local dimensions with the participation of multiple intelligent agents, forming a learning optimization phase. It also constructs the mutation optimization operator of the spatial frog leaping algorithm, forming a mutation optimization phase to deal with position updates outside the search domain.

[0141] The fourth main module is used to determine data based on target parameters, construct an optimized multi-objective function from multiple perspectives such as economic, social, and ecological, and determine target coefficients; and establish a constraint system for land use optimization from three aspects: quantitative structure, laws, regulations, and policies, and land use conversion rules;

[0142] The fifth main module is used to combine the optimization objectives and constraint system, alternately run the learning operator and mutation operator of the frog leaping algorithm, and output the results of land use spatial layout optimization.

[0143] The coupled bionic intelligent algorithm and multi-agent land use optimization system provided by the embodiment of the present invention adopts the above-mentioned several modules, coordinates the multi-objective land use configuration requirements that combine regional nature and humanities, simulates the land use decision-making system of multiple levels of stakeholders, and collaboratively optimizes the complex land use structure of urban areas from a dynamic perspective that combines the global and local, macro and micro perspectives.

[0144] It should be noted that the system embodiments provided by the present invention are not only used to implement the methods in the above-mentioned method embodiments, but also used to implement the methods in other method embodiments provided by the present invention. The only difference lies in the setting of corresponding functional modules, and the principles thereof are basically the same as the principles of the above-mentioned system embodiments provided by the present invention. As long as those skilled in the art refer to the specific technical solutions in other method embodiments on the basis of the above-mentioned system embodiments, obtain corresponding technical means and technical solutions composed of these technical means by combining technical features, and on the premise of ensuring the practicality of the technical solutions, improve the modules in the above-mentioned system embodiments to obtain corresponding system class embodiments for implementing the methods in other method class embodiments.

[0145] Based on the same inventive concept as the above-mentioned embodiment, an embodiment of the present invention also provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions enable the computer to execute the steps of the coupled bionic intelligent algorithm and multi-agent land use optimization method.

[0146] In summary, the present invention relates to a land use optimization model that combines macro-spatial optimization with micro-decision-making simulation, and couples a bionic intelligent algorithm and multi-agent. The present invention constructs an economic, social and ecological multi-objective scenario, and proposes an optimization model (SSFLA-MLAS) that couples spatial frog leaping and multi-level multi-agent. From the two dimensions of decision-making behavior and object, a multi-level multi-agent model (MLAS) in which mapping and multi-level information transmission coexist is proposed; based on the solution-oriented optimization model, a spatial frog leaping algorithm (SSFLA) is constructed by reconstructing the strategy of the heuristic operator; and a dynamic coupling method of internal and external nested models with a bionic intelligent algorithm as the base and multi-agent as the core is proposed. The present invention strengthens the cooperative competition mechanism of multiple agents in the intelligent agent model and enhances its rationality at the micro-simulation level; based on neighborhood knowledge, opening and closing strategies and reinforcement learning, it explores the heterogeneous spatial perception of multi-level agents and depicts the information transmission process of superior and subordinate agents; based on the solution-oriented optimization model, it improves the global optimization ability of the frog leaping algorithm land use optimization model; the internal and external nested model model realizes the spatial coupling of multiple models and enhances the interpretability of the seamless integration process, providing a new idea for the land use optimization modeling that couples micro-decision-making and global optimization.

[0147] It should be understood that parts not elaborated in detail in this specification belong to the prior art.

[0148] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A land use optimization method that couples a biomimetic intelligent algorithm and a multi-agent approach, characterized in that: include: Collect and process study area data, including land use data, suitability evaluation variable data, and target parameter determination data; Clarify the system elements of multi-agents, construct a multi-level multi-agent model that maps decisions and objects and interacts with multi-level agents. Based on the constructed multi-level multi-agent model, it also includes: calculating the suitability of land use decisions of multi-level agents based on suitability evaluation variable data, and creating a static suitability perception knowledge base of multi-agents; based on the theories of adaptive neighborhood, morphological opening and closing, and reinforcement learning, by constructing spatial perception operators and inherited perception operators, calculating the neighborhood benefits and interactive perceptions of multi-level agents, and creating a dynamic environmental perception knowledge base of multi-agents; based on the behavior and decision-making characteristics of multiple agents, calculating the expected utility of public agents, the competitiveness of departmental agents, and the decision utility of government agents as the basis for the decision-making behavior of agents; calculating the selection probability of each agent, quantifying the mathematical expression of the decision-making behavior of agents, and constructing a probabilistic selection mechanism; Based on land use data, the existing land use spatial layout plan is spatially expressed and a bionic intelligent group is initialized. A spatial dynamic learning operator is used to construct a heuristic process for the spatialized frog leaping algorithm, which searches for optimal solutions and updates positions with multi-agent coupling from both global and local dimensions, forming a learning optimization phase. A mutation optimization operator is constructed for the spatialized frog leaping algorithm, forming a mutation optimization phase to handle position updates outside the search domain. Based on the target parameter determination data, an optimized multi-objective function is constructed from multiple perspectives, including economic, social, and ecological perspectives, and target coefficients are determined. A constraint system for land use optimization is established from three aspects: quantitative structure, laws, regulations, and policies, and land use conversion rules. Combining the optimization objectives and constraint system, the learning operator and mutation operator of the leapfrog algorithm are run alternately to output the results of land use spatial layout optimization.

2. The land use optimization method coupled with a biomimetic intelligent algorithm and a multi-agent according to claim 1 is characterized in that: Collect and process study area data, including: The spatial data is uniformly projected and transformed, and converted to a uniform resolution through resampling to form a data set with consistent spatial coordinate system and resolution; the land use data is secondary classified and sorted to simplify the model and highlight the coupling design.

3. The land use optimization method coupled with a biomimetic intelligent algorithm and a multi-agent according to claim 1 is characterized in that: Construct a multi-level multi-agent model that maps decisions to objects and interacts with multi-level agents, including: The mapping relationship between decision-making and objects is expressed as follows: Based on the functions of stakeholders in the decision-making process, the subject of land use is abstracted and a decision-making agent is designed to simulate the decision-making information interaction of various stakeholders in real scenarios. Based on the idea of ​​self-organizing neighborhoods, an object agent is designed to represent the differentiated perceptions of stakeholders at the local scale. Decision-making and object agents at the same level directly interact with each other, forming a strong mapping relationship, while information interaction between agents at different levels is a weak mapping relationship. Multi-level agent division and information transmission: Based on the hierarchical relationship within the agent, the two types of agents, decision-making and object agents, are divided into multi-level relationships mapped by "individual-department-government" and "unit-type-scheme", respectively. Information exchange in the form of land use plans occurs between the superior and subordinate agents.

4. The land use optimization method coupled with a biomimetic intelligent algorithm and a multi-agent according to claim 1 is characterized in that: Compute the suitability of land use decisions for multi-level agents, including: K suit =K T＝k (i,j)=a k *A1+b k *A2+…+c k *A m ,a k +b k +…+c k =1 Among them, K T＝k (i, j) is the public suitability evaluation, which indicates the influence of adaptive knowledge on the land use selection T = k of land unit (i, j), A1, A2, ..., A m represents the decision factor, α k , β k …, γ k Represents the decision factor weight coefficients of different subjects.

5. The land use optimization method coupled with a biomimetic intelligent algorithm and a multi-agent according to claim 1 is characterized in that: Construct spatial perception operators and inherited perception operators, including: Construct spatial perception operators at two levels: unit and type. The unit spatial perception operator is oriented towards the grid unit of land use, realizing the dynamic perception of the neighborhood environment of the public intelligent agent in the decision-making function; the type spatial perception operator focuses on the morphology and neighborhood relationship of the patches, and enhances the aggregation by adjusting the units at the edge, thereby reducing the effect of fragmented landscape. The inheritance perception operator adjusts the decision-making behavior by characterizing the expected benefits of the intelligent agent for making decisions. It adopts the reinforcement learning idea of ​​Roth-Erev to construct an improved reward function Reward to expand the process of spatial knowledge inheritance learning, and then corrects the current intelligent agent's perception of the benefits of behavior.

6. The land use optimization method coupled with a biomimetic intelligent algorithm and a multi-agent according to claim 1 is characterized in that: A knowledge system of multi-level intelligent agent behavior and decision-making, including: Based on the perceptual knowledge in the knowledge base, multi-agents calculate the expected utility in a weighted manner, use a discrete choice model to form the selection probability, generate decision-making behaviors under the guidance of the selection mechanism, generate decision plans and interact with object agents to realize land use changes at the spatial level; among them, the sources of the knowledge base include static adaptive knowledge, dynamic perceptual environment knowledge, inherited knowledge and restrictive conditions.

7. The land use optimization method coupled with a biomimetic intelligent algorithm and a multi-agent according to claim 1 is characterized in that: The learning optimization stage includes the local optimization stage and the global optimization stage, in which: The formula for finding the worst solution position in the local optimization stage is: Among them, P new represents the updated position of the worst frog, P best and P worst Corresponding to the spatial solutions of the local best frog and the local worst frog respectively, Step is expressed as the step length; if the new position after local optimization is better than the previous position, the worst frog individual is updated, otherwise the model adopts G best Instead of P best , enter the global optimization stage; The formula for finding the worst solution position in the global optimization stage is: G best Corresponding to the spatial solution of the global best frog, if the result of global optimization cannot improve the frog's target benefit, it will enter the mutation optimization stage.

8. The land use optimization method coupled with a biomimetic intelligent algorithm and a multi-agent according to claim 1 is characterized in that: From the perspectives of economy, society and ecology, we build an optimized multi-objective function, including: Among them, f Economic represents the financial benefits of the land use scheme in the actual production process, Eco(k) represents the economic output corresponding to land use type k, Area(k) represents the corresponding land area, and f Social represents social benefits, Ω represents the Moore neighborhood of the unit, A ijk represents the number of type k land uses in the neighborhood, x ijk is a binary variable representing the discrimination of unit land use type k, N Ω is the total number of cells in the neighborhood Ω, M and N represent the number of rows and columns of the grid area, and f Ecological As an equation for maximizing ecological benefits, it reflects the positive and negative impacts of quantitative structure and project layout on the natural environment, and formulates a hybrid ecological goal that combines ecosystem service value and low carbon. ESV(k) represents the unit ecosystem service value corresponding to land use type k, Carbon(k) is the low carbon coefficient, and α and β correspond to the weights of the two ecological sub-goals, respectively.

9. A land use optimization system that couples biomimetic intelligent algorithms and multi-agents, characterized by: Steps for implementing the coupled bionic intelligent algorithm and multi-agent land use optimization method described in any one of claims 1 to 8.

10. A non-transitory computer-readable storage medium, characterized in that The non-transitory computer-readable storage medium stores computer instructions, which enable the computer to execute the steps of the coupled bionic intelligence algorithm and multi-agent land use optimization method described in any one of claims 1 to 8.

Images