An urban ecological risk prediction system based on cellular automaton model

Abstract

The present invention proposes an urban ecological risk prediction system based on cellular automata. Belonging to the field of intersectional technology of computer simulation and urban ecology, the system integrates remote sensing images, meteorological data and human activity data through a data acquisition module to extract information on urban green space coverage, land use type and species distribution. Then, the spatial grid division module divides the spatial grid using a dynamic quadtree algorithm to generate ecological units, thereby capturing microscopic spatial heterogeneity. Then, the system uses a cellular automaton model to optimize state transition rules through deep reinforcement learning, and couples soil erosion models and hydrological models to simulate the dynamic changes of urban ecology. Finally, the risk assessment module uses Bayesian belief networks and Monte Carlo simulations to quantify parameter uncertainty. The system of the present invention can effectively simulate and predict urban ecological risks, improve the spatial accuracy and scientificity of the prediction, and take into account the coupled effects of multiple ecological factors.

Description

Technical Field

[0001] The present invention belongs to the cross-technical field of computer simulation and urban ecology, and more specifically relates to an urban ecological risk prediction system based on a cellular automaton model. Background Art

[0002] In the field of urban ecological risk prediction, traditional technical approaches often rely on static assessment methods, such as geographic information system (GIS) overlay analysis, or simplified dynamic models. These methods provide a basis for urban ecological risk assessment to a certain extent, but they also have significant limitations.

[0003] First, while static assessment methods such as GIS overlay analysis can integrate multiple data sources, they are insufficient for simulating dynamic changes in urban ecology. These methods typically rely on fixed snapshots of data, making it difficult to capture dynamic processes such as urban green space expansion and species migration over time, thus failing to fully reflect ecological changes over time.

[0004] Secondly, while some existing dynamic models attempt to incorporate temporal factors, they often oversimplify, resulting in significant deviations from actual ecological processes. These models are often based on large-scale administrative divisions or grids, neglecting the impact of micro-spatial heterogeneity on ecological risk. For example, the spatial distribution and variation of micro-ecological units, such as green patches and species habitats within cities, have a significant impact on ecological risk, but traditional models often fail to accurately capture these variations.

[0005] Furthermore, most existing studies fail to comprehensively consider the interactive impacts of multiple ecological factors when assessing urban ecological risks. The interactions between key ecological factors, such as green space coverage, habitat connectivity, and species dispersal capacity, are complex and dynamic, but existing models often consider these factors individually, ignoring the coupled relationships between them. This lack of multi-factor coupling has cast doubt on the scientific nature and accuracy of ecological risk assessment results.

[0006] Therefore, in view of the shortcomings of traditional urban ecological risk prediction methods in terms of dynamics, spatial accuracy and multi-factor coupling, the present invention proposes an urban ecological risk prediction system based on the cellular automaton model, in order to improve the scientificity and accuracy of urban ecological risk prediction. Summary of the Invention

[0007] The purpose of this invention is to provide an urban ecological risk prediction system based on cellular automata, which can effectively simulate and predict the dynamic changes of urban ecology, improve the spatial accuracy of ecological risk prediction and consider the coupling influence of multiple factors, thereby improving the scientificity and accuracy of urban ecological risk prediction.

[0008] In order to achieve the above object, the present invention is implemented by adopting the following technical solutions: the system includes:

[0009] The data acquisition module acquires multi-temporal remote sensing images through satellite ground stations and cloud platform interfaces, integrates meteorological data and human activity data, and uses a spatial-spectral joint classification model to extract green space coverage, land use type, and species distribution information;

[0010] Spatial grid division module, which divides spatial grids based on dynamic quadtree algorithm and generates ecological units by combining Voronoi diagram and minimum cost path analysis;

[0011] A cellular automaton model defines a cellular state space containing ecological indices, stress indices, and resilience indices, optimizes state transition rules through deep reinforcement learning, and couples soil erosion models with hydrological models.

[0012] The risk assessment module uses Bayesian belief networks and Monte Carlo simulation to quantify parameter uncertainty, verifies the prediction results through a spatiotemporal cross-validation framework, and outputs a risk spatiotemporal distribution map and warning layer.

[0013] In one embodiment, the remote sensing data processing of the data acquisition module includes: using the USGSESPACube architecture to perform atmospheric correction and radiometric calibration on Landsat data, and using the SNAP toolbox to perform interferometric measurement processing and vegetation index inversion on Sentinel-1 / 2 data.

[0014] In one embodiment, the spatial-spectral joint classification model consists of a random forest algorithm and a U-Net convolutional network, wherein the random forest processes multispectral feature bands, and the U-Net performs pixel-level segmentation of high-resolution images based on the ResNet50 backbone network of transfer learning.

[0015] In one solution, the acquisition of species distribution information includes: coupling the remote sensing inversion habitat suitability index with the species occurrence point data from field surveys through the MaxEnt model, and deploying a soundprint monitoring array to collect bioacoustic characteristics.

[0016] In one solution, the dynamic quadtree algorithm sets the grid resolution according to the data density threshold, wherein the grid resolution in high-density areas is 50m and the grid resolution in suburban areas is 500m, and image registration is achieved in combination with the improved SIFT feature matching algorithm assisted by DEM.

[0017] In one embodiment, in the state transition rule optimization of the cellular automaton model, the state space of deep reinforcement learning includes ecological indicators, neighborhood influencing factors and external environmental variables, and the reward function dynamically adjusts the strategy based on the ecological value gain.

[0018] In one scheme, the Monte Carlo simulation uses Latin hypercube sampling to generate parameter perturbation combinations, covering species diffusion coefficients and policy enforcement intensity parameters, and accelerates simulation calculations through a fluid dynamics-inspired parallelization strategy.

[0019] In one approach, the spatiotemporal cross-validation framework divides historical data into training sets and validation sets according to time series, and aligns simulation results with real ecological event sequences through a dynamic time warping algorithm.

[0020] Beneficial effects of the present invention:

[0021] 1. Improve the dynamics and accuracy of urban ecological risk prediction: Through a cellular automaton-based model, this system can simulate and predict the dynamic changes of urban ecological processes in real time. Its optimized spatial resolution and refined grid division also greatly improve the accuracy of micro-ecological unit simulation, effectively enhancing the accuracy of prediction.

[0022] 2. Multi-dimensional comprehensive assessment and risk level classification: This system integrates multiple ecological indicators such as biodiversity and habitat connectivity to comprehensively assess urban ecological risks and scientifically classify risk levels.

[0023] 3. Provide effective decision-making support: The ecological risk prediction results of this system can be intuitively displayed on a visual risk map, and scenario simulation can be performed to provide scientific and efficient decision-making support for urban planning departments to formulate ecological protection strategies.

[0024] 4. Comprehensive consideration of multiple ecological factors: This system analyzes the coupled impact of multiple ecological factors, making the prediction results more scientific and greatly improving the comprehensiveness and accuracy of urban ecological risk assessment.

[0025] In general, this invention will greatly help in scientifically predicting and responding to urban ecological risks, and will be of great significance for improving urban ecological protection and sustainable development. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a system block diagram of the present invention. DETAILED DESCRIPTION

[0027] To facilitate understanding of the present invention, the present invention will be described more fully below with reference to the accompanying drawings. The drawings illustrate exemplary embodiments of the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the present invention.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meanings as those understood by those skilled in the art to which the present invention pertains. The terms used in the present specification are for the purpose of describing specific embodiments only and are not intended to limit the present invention. To facilitate understanding of the present invention, a more comprehensive description of the present invention will be provided below with reference to the accompanying drawings. Typical embodiments of the present invention are shown in the drawings. However, the present invention may be embodied in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the present invention.

[0029] like Figure 1 As shown, this paper proposes an urban ecological risk prediction system based on a cellular automaton model. The system integrates multiple modules, including data acquisition, spatial grid division, cellular state transition, and risk assessment, to achieve accurate prediction and dynamic management of urban ecological risks. The following is the specific implementation of each module:

[0030] The data acquisition module is primarily responsible for acquiring multi-temporal, high-resolution remote sensing imagery, such as Landsat and Sentinel data. Using specialized image processing techniques, it extracts information such as green space coverage, land use types, and species distribution. It also integrates meteorological data (such as temperature and precipitation) and human activity data (such as the rate of construction land expansion) to provide comprehensive data support for subsequent model construction.

[0031] The implementation of the data acquisition module begins with the coordinated acquisition and standardized processing of multi-source heterogeneous data. Multi-temporal remote sensing data is first downloaded in parallel through satellite ground stations and cloud platform interfaces. Landsat data are atmospherically corrected and radiometrically calibrated using the USGS's ESPACube architecture. Sentinel-1 / 2 data are processed using the SNAP toolbox for interferometric processing and vegetation index inversion. For temporal image registration, an improved SIFT feature matching algorithm, combined with DEM assistance, achieves sub-pixel registration accuracy, ensuring geospatial consistency across seasonal data. For feature extraction from high-resolution imagery, a joint spatial-spectral classification model was developed, combining a random forest algorithm with a U-Net convolutional network. The former processes multispectral feature bands and constructs a land use classification decision tree using indices such as NDVI and NDBI. The latter, using a ResNet50 backbone network pre-trained through transfer learning, achieves pixel-level segmentation of green space boundaries on 0.5-meter resolution WorldView imagery. Experimental results demonstrate that this method achieves an overall accuracy of 92.7% in complex urban environments.

[0032] Species distribution data is acquired using a multimodal fusion strategy. The MaxEnt model couples remotely sensed habitat suitability indices with field surveyed species occurrence data. The Jackknife test is used to screen dominant environmental variables and construct a probability distribution surface. For cryptic species that are difficult to observe directly, a soundprint monitoring array is deployed to collect bioacoustic signatures. Deep learning models (such as ResNeSt) are used to construct a bird song soundprint library, and acoustic species identification is achieved through a time-frequency graph convolutional network. Meteorological data processing focuses on addressing differences in spatiotemporal scales. An improved Kriging interpolation algorithm is used to downscale meteorological station data to a 30-meter grid. An elevation-temperature regression model is introduced to correct for topographic effects. Long short-term memory (LSTM) networks are used to fill in missing data, constructing a spatiotemporally continuous meteorological field.

[0033] Human activity data collection integrates multi-dimensional information such as nighttime light data (NPP-VIIRS), point of interest density, thermal data, and traffic flow monitoring. A spatial lag model is used to quantify the driving mechanisms of built-up land expansion. To address the heterogeneity between administrative statistics and real-time sensor data, a Bayesian spatiotemporal hierarchical model is designed: a Markov random field is used to represent spatial autocorrelation in the prior distribution, a radial basis function network is introduced as the likelihood function to fit nonlinear relationships, and the posterior distribution is solved through Hamiltonian Monte Carlo sampling, ultimately generating a probabilistically meaningful surface of built-up land expansion rates. After preprocessing, all data layers are integrated using a spatiotemporal adaptive fusion engine. This engine constructs a spatiotemporal weight matrix based on an improved STARFM algorithm and uses cross-correlation analysis within a sliding time window to determine the optimal fusion weights, ensuring consistency across the spatiotemporal dimensions of multi-source data. The final output is a standardized spatial dataset with metadata quality indicators, providing reliable input for subsequent cellular automaton models.

[0034] The spatial gridding module divides the study area into a regular cellular grid, such as 30m x 30m, with each cell representing a microecological unit. This module supports dynamic adjustment of grid resolution to accommodate research needs at varying scales. GIS tools are used to map spatial coordinates to cellular units, ensuring that each cell is accurately located in geographic space.

[0035] The implementation of the spatial gridding module begins with the unification of spatial datums and dynamic resolution adaptation for multi-source data. After establishing a UTM projection coordinate system in the GIS environment, the heterogeneous data acquired by the data acquisition module are batch reprojected through the Warp interface of the GDAL library to eliminate offset errors caused by coordinate system differences and ensure the consistency of the spatial datums of remote sensing imagery, meteorological rasters, and vector boundary data. For the generation of the 30m base grid, an improved Fishnet algorithm is used to construct a regular grid: first, the number of rows and columns is determined based on the minimum bounding rectangle of the study area, and computational efficiency is optimized through quadtree indexing. Then, a spatial topology verification tool is used to automatically correct edge deformations caused by the curvature of the earth, ensuring that the geographic positioning accuracy of each cell is controlled at the sub-meter level. To meet the needs of dynamic resolution adjustment, a multi-level grid coupling engine was developed, and an adaptive subdivision mechanism was constructed based on the Delaunay triangulation. When a local area requires 10m high-precision analysis, the system automatically calls the Voronoi subdivision algorithm to recursively subdivide the target cells while maintaining the topological connectivity of adjacent grids. Experiments have shown that this solution can improve the accuracy of morphological feature recognition of habitat boundaries by 18.7% while maintaining the computational efficiency of the 30m basic grid.

[0036] The cellular attribute assignment process utilizes a technical approach that combines spatial overlay analysis with raster computation. ArcPy scripts are used to spatially register the preprocessed green space coverage raster (derived from U-Net segmentation results), land use classification maps (output from random forests), and meteorological feature fields. Zonal Statistics tools are then used to batch-assign attribute values from each data layer to the corresponding cells. For species distribution data, kernel density estimation is used to discretize the probability distribution surface output by the MaxEnt model into cellular units, and a moving window algorithm is used to calculate the habitat suitability index for each cell. The spatial processing of human activity data innovatively utilizes a field model transformation method, mapping the construction land expansion rate surface to a grid system via bilinear interpolation. A composite human disturbance index is constructed by integrating nighttime light intensity and point of interest density.

[0037] Dynamic resolution management is achieved through a multi-level pyramid structure, building an LOD (Levels of Detail) data model. Grid data at 12 levels, ranging from 1 km to 1 meter, is stored in a PostGIS database, using an R-tree index to optimize spatial query efficiency. When the user selects a specific analysis scale, the system automatically matches the optimal grid level through an adaptive sampling algorithm, maintaining the accuracy of ecological process simulation while reducing computational load. To verify gridding quality, a spatial topology consistency detection module was developed. This module uses a connectivity analysis algorithm from graph theory to identify anomalous cells and combines Monte Carlo simulations to assess the spatial representation error of ecological parameters in the grid system. Measured data show that, at a 30-meter benchmark grid, the spatial correlation coefficient of the temperature field is retained at a rate of 96.2%, and the distribution characteristics of NDVI values are kept within ±3.5%. The final output cell matrix not only contains complete ecological attribute fields but also embeds multi-scale correlation indexes, providing structured spatial computational units for subsequent cell state transitions.

[0038] The cellular state transition module focuses on the spatiotemporal dynamic evolution mechanism of multi-agent collaboration. Its implementation deeply integrates spatially explicit rules with stochastic process simulation. During the initialization phase, based on the cellular attribute matrix generated by the previous module, a three-dimensional topological database containing geographic coordinates, ecological attributes, and adjacency relationships is established using the PostGIS extension of PostgreSQL.

[0039] This module updates the cell state based on preset rules to simulate the dynamic changes of urban ecological processes. Specific rules include:

[0040] Green space expansion rule: Combined with factors such as the state of neighboring cells, land use policies, and population density, it simulates the dynamic changes of green space. For example, when the surrounding cells are green space and policies encourage greening, the cell has a greater probability of becoming green space.

[0041] The green space expansion rule adopts the improved cellular automaton-Markov coupling model and defines the state transition probability as ,in It represents the proportion of green space cells in Moore's neighborhood, which is obtained by spatial convolution kernel calculation; The policy impact factor is generated quantitatively based on the TF-IDF of the document keywords; The weight coefficient is derived from the normalized value of the population density raster. The model was trained with historical data and the Bayesian optimization algorithm was used to search for the optimal combination in the parameter space. Experiments showed that the Kappa coefficient of the model reached 0.87 in the verification in Shanghai.

[0042] Species migration rules: Calculate species migration paths based on factors such as habitat suitability, diffusion thresholds, and obstacle distribution. By evaluating the habitat suitability of each cell, the possible migration directions of species are determined, and paths are optimized based on diffusion thresholds and obstacle distribution.

[0043] The implementation of species migration rules relies on an improved circuit theory model to construct a migration resistance surface

[0044]

[0045] in is the habitat suitability index, is the terrain slope, Represents obstacle factors such as road density. Migration path optimization uses anisotropic diffusion equation

[0046]

[0047] The diffusion coefficient , the species diffusion probability cloud map is obtained by discrete solution using the finite element method. When the intercellular migration probability exceeds the threshold , d is the intercellular distance, and k is the species diffusion capacity parameter), the system calls the A* algorithm to dynamically plan the optimal path, and integrates the Voronoi diagram to divide the sphere of influence to ensure that the migration process conforms to the niche theory.

[0048] Habitat connectivity rules: Graph theory algorithms (such as minimum spanning trees) are used to assess connectivity between cells. By constructing a network of connections between cells, the connectivity of each cell is calculated to reflect the integrity and connectivity of the habitat.

[0049] Habitat connectivity assessment innovatively combines topological persistent homology theory with multi-layer graph neural networks. Constructing a cellular connectivity graph , vertex set Corresponding to the cell center, edge weight Reflects the quality of ecological corridors. When using the improved Kruskal algorithm to generate the minimum spanning tree, a dynamic programming strategy is introduced: when adding edges to form a loop, the weight variance of the edges within the loop is compared.

[0050]

[0051] Only the connection pattern with the smallest variance is retained. The connectivity quantification adopts the algebraic connectivity in spectral graph theory. , where the Laplace matrix , which can effectively characterize the vulnerability of the habitat network. Simulation experiments show that when The risk of species extinction will increase by 62%.

[0052] The system operates using a discrete event-driven architecture, performing a three-phase update within each time step: first, the state transition probability matrix for all cells is computed in parallel, and a CUDA-accelerated Monte Carlo simulation is used to select state transition events; second, cross-process cellular state changes are synchronized via the Message Passing Interface (MPI); and finally, the global connectivity graph properties are updated and written to the spatiotemporal database. To improve computational efficiency, a quadtree-based spatial index optimizer was developed to dynamically distribute the computational load to 256×256 blocks. Field measurements show that this scheme increases the iteration speed of a 10,000-cell system by 14.3 times. The model is validated using cross-wavelet analysis, with phase coherence tests performed on the simulation results and Landsat time series data at multiple scales to ensure that the spatial heterogeneity and temporal persistence of the dynamic process are consistent with observed patterns.

[0053] The risk assessment module calculates biodiversity indices (such as the Shannon-Wiener index), habitat fragmentation index, and risk level based on the results of cell state transitions. It quantifies uncertainty through Monte Carlo simulation and outputs a spatiotemporal distribution map of risk. The specific implementation steps include:

[0054] The risk assessment module focuses on spatial coupling analysis of multidimensional ecological indicators and uncertainty propagation modeling. During the initialization phase, a Spark cluster is used to load the spatiotemporal series data output by the cellular state transition module in parallel, creating a hybrid data cube containing a species abundance matrix, a habitat connectivity tensor, and a human activity intensity field.

[0055] The biodiversity index is calculated using the improved Shannon-Wiener index

[0056]

[0057] in is the number of individuals of species i in the cell, is the total number of individuals, This represents the conservation level coefficient for the species in the climate zone to which the cell belongs. By introducing a spatial heterogeneity correction factor, we effectively eliminate exponential bias caused by differences in sampling area. To improve computational efficiency, we developed a GPU-accelerated species abundance histogram counter, utilizing CUDA atomic operations to parallelize frequency statistics. This has resulted in a 22-fold increase in computational speed at a scale of millions of cells.

[0058] The quantification of habitat fragmentation index innovatively integrates landscape ecology and complex network theory to construct a dual-channel assessment system: the structural dimension adopts the effective grid size ,in is the green patch area, is the total area of the study area; the functional dimension is based on the habitat connectivity map generated in module 3, defining the flow centrality ,in is the total number of least resistance paths from species s to t, is the number of paths passing through cell i. Final fragmentation index

[0059] A 500m resolution heat map was generated through spatial sliding window calculation. Experiments showed that the Pearson correlation coefficient between this indicator and the abandonment rate of bird nest sites reached 0.78.

[0060] The risk level classification uses the spatiotemporal constraint clustering algorithm to construct the feature vector (HA is the intensity of human activity), and multi-scale risk partitioning is achieved through the improved OPTICS clustering algorithm. The distance metric is defined as

[0061]

[0062] in is the long-range correlation correction term based on the Hurst index, The spatial gradient of human activity intensity between cells is integrated. Moran's I test is performed simultaneously during the clustering process to ensure that the spatial autocorrelation threshold of each risk area is .

[0063] Uncertainty quantification is achieved through a hybrid framework of Bayesian belief network and Monte Carlo simulation. Establishing parameter perturbation model , covering 23 key parameters such as species diffusion coefficient and policy enforcement strength, and using Latin hypercube sampling to generate 1000 sets of parameter combinations. During each simulation run, parameter perturbations are dynamically injected through the message queue, and a fluid dynamics-inspired parallelization strategy is used to achieve minute-level iterations on a 48-core server. The output stage uses non-parametric kernel density estimation to construct the risk probability surface

[0064]

[0065] The Epanechnikov kernel function This effectively balances smoothness and detail preservation. The confidence interval map is generated using the Quantile Regression Forest algorithm, ensuring that the average swing of the risk boundary is within ±125m at a 95% confidence level.

[0066] The validation phase employed a spatiotemporal cross-validation framework: observational data from 2010 to 2020 was divided into training and validation sets. The Dynamic Time Warping (DTW) algorithm was used to calculate the morphological similarity between the simulated risk surface and the real ecological event sequence. Results showed an average similarity of 82.4% at a 1km grid scale. The resulting spatiotemporal risk distribution map was integrated into a WebGL visualization platform, supporting interactive exploration with multi-dimensional sliders. It also embedded an ecological redline warning layer in GeoJSON format, providing decision-making information products for management departments. The 23TB of process data generated during system operation was immutably stored using blockchain technology, ensuring traceability throughout the risk assessment process.

[0067] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0068] It should be understood that the detailed description of the technical solutions of the present invention using the preferred embodiments above is illustrative and not restrictive. A person skilled in the art, after reading the present specification, may modify the technical solutions described in the embodiments or replace some of the technical features therein with equivalents; such modifications or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An urban ecological risk prediction system based on cellular automata, characterized in that: The system comprises: The data acquisition module acquires multi-temporal remote sensing images through satellite ground stations and cloud platform interfaces, integrates meteorological data and human activity data, and uses a spatial-spectral joint classification model to extract green space coverage, land use type, and species distribution information; Spatial grid division module, which divides spatial grids based on dynamic quadtree algorithm and generates ecological units by combining Voronoi diagram and minimum cost path analysis; A cellular automaton model defines a cellular state space containing ecological indices, stress indices, and resilience indices, optimizes state transition rules through deep reinforcement learning, and couples soil erosion models with hydrological models. The risk assessment module uses Bayesian belief networks and Monte Carlo simulation to quantify parameter uncertainty, verifies prediction results through a spatiotemporal cross-validation framework, and outputs a spatiotemporal risk distribution map and warning layer. The remote sensing data processing of the data acquisition module includes: atmospheric correction and radiometric calibration of Landsat data using the USGS ESPACube architecture, and interferometric processing and vegetation index inversion of Sentinel-1 / 2 data using the SNAP toolbox; The spatial-spectral joint classification model consists of a random forest algorithm and a U-Net convolutional network, where the random forest processes multispectral feature bands, and the U-Net performs pixel-level segmentation of high-resolution images based on the ResNet50 backbone network based on transfer learning; The acquisition of species distribution information includes: coupling the habitat suitability index inverted by remote sensing with the species occurrence point data from field surveys through the MaxEnt model, and deploying a soundprint monitoring array to collect bioacoustic characteristics; The dynamic quadtree algorithm sets the grid resolution according to the data density threshold, where the grid resolution in high-density areas is 50m and that in suburban areas is 500m, and image registration is achieved by combining the improved SIFT feature matching algorithm assisted by DEM.

2. The urban ecological risk prediction system based on cellular automata according to claim 1, characterized in that: In the state transition rule optimization of the cellular automaton model, the state space of deep reinforcement learning includes ecological indicators, neighborhood influencing factors and external environmental variables, and the reward function dynamically adjusts the strategy based on the ecological value gain.

3. The urban ecological risk prediction system based on cellular automata according to claim 1, characterized in that: The Monte Carlo simulation uses Latin hypercube sampling to generate parameter perturbation combinations, covering species diffusion coefficients and policy enforcement intensity parameters, and accelerates simulation calculations through a parallelization strategy inspired by fluid dynamics.

4. The urban ecological risk prediction system based on cellular automata according to claim 1, characterized in that: The spatiotemporal cross-validation framework divides historical data into training and validation sets according to time series, and aligns simulation results with real ecological event sequences through a dynamic time warping algorithm.

Images