Land space planning multi-source data fusion decision support system

Abstract

The invention relates to the technical field of land space planning, and discloses a territorial space planning multi-source data fusion decision support system, which comprises a data foundation and preprocessing module used for processing multi-source heterogeneous data related to territorial space planning; the macroscopic causal knowledge construction module is used for constructing a space-time causal map representing a macroscopic causal relationship among territorial space elements from a unified data basis; a coupling simulation and deduction execution module; the conflict perception and scene differentiation module is used for monitoring the consistency between the macroscopic trend and the microscopic behavior trend; the multi-scene toughness strategy evaluation module is used for receiving a planning strategy; and the model adaptive calibration module is used for comparing the deviation between the virtual data of the future space-time state and the real world observation data. The space-time causal atlas is fed back and updated by comparing simulation and reality deviations, closed-loop learning is formed, and the problems that macroscopic and microscopic disjunction occurs, and deep uncertainty is difficult to deal with are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of land space planning, in particular to a land space planning multi-source data fusion decision support system. BACKGROUND

[0002] Land space planning is a guide for national spatial development and a spatial blueprint for sustainable development. Its scientific compilation and effective implementation are crucial for optimizing the development and protection pattern of land space and improving the ability of national governance. With the development of information technology, using geographic information systems (GIS), remote sensing data, social and economic statistical data, and other multi-source data to analyze and simulate land space systems has become an important technical means to assist in planning decisions.

[0003] The existing technical solutions still face deep-seated challenges when making planning decisions. Current analysis methods usually focus on statistical analysis and correlation mining of historical data, which makes it difficult for the model to reveal the deep-seated, time-dynamic evolution of the causal transmission mechanism between the elements of the land space system, resulting in that the prediction of future trends often stays at the level of phenomenon correlation, lacking deep insight into the internal logic of the system.

[0004] The existing planning simulation and deduction model is fragmented between the macro policy level and the micro subject (such as residents and enterprises) behavior level. How does macro-level planning intervention specifically affect the decisions of micro individuals, and how does the aggregation of a large number of micro behaviors emerge new macro phenomena? This two-way coupled dynamic process cannot be effectively simulated, resulting in a deviation between planning intentions and actual emerging effects. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a land space planning multi-source data fusion decision support system, which solves the problems of the existing land space planning decision support technology in dynamically coupling macro causality and micro behavior, actively discovering system endogenous uncertainty, and realizing closed-loop adaptive calibration.

[0006] To achieve the above purpose, the present application realizes the following technical solutions: a land space planning multi-source data fusion decision support system, comprising:

[0007] A data base and preprocessing module for processing multi-source heterogeneous data related to land space planning to build a unified data base;

[0008] A macro causal knowledge construction module connected with the data base and preprocessing module, for constructing a spatio-temporal causal graph representing the macro causal relationship between land space elements from the unified data base, and generating a macro trend based on the spatio-temporal causal graph;

[0009] The coupling simulation and deduction execution module is connected with the macroscopic causal knowledge construction module, and is used for deducing a specific national space planning intervention under the constraint of the space-time causal graph to generate virtual data of a future space-time state and a micro-behavior trend.

[0010] The conflict perception and scenario differentiation module is connected with the macroscopic causal knowledge construction module and the coupling simulation and deduction execution module, and is used for monitoring consistency between the macroscopic trend and the micro-behavior trend, and triggering generation of multiple future scenarios when a conflict is identified.

[0011] The multi-scenario resilience strategy evaluation module is connected with the conflict perception and scenario differentiation module and the coupling simulation and deduction execution module, and is used for receiving a planning strategy, scheduling the coupling simulation and deduction execution module to perform cross-deduction evaluation in the multiple future scenarios, and sorting the planning strategy based on a resilience evaluation function to output a decision suggestion.

[0012] The model self-adaptive calibration module is connected with the coupling simulation and deduction execution module and the macroscopic causal knowledge construction module, and is used for comparing deviation between virtual data of a future space-time state and real world observation data, attributing the deviation to a specific causal path in the space-time causal graph, and generating a calibration signal to feed back to the macroscopic causal knowledge construction module to update the space-time causal graph.

[0013] Preferably, the space-time causal graph is defined as a weighted directed acyclic graph evolving over time. :

[0014] In the formula, denotes a node set of a graph, wherein each node corresponds to a spatial variable in a unified data base; denotes a total number of variables;

[0015] denotes a directed edge set at time , and each directed edge denotes a variable has a direct causal effect on a variable at time

[0016] denotes a weight matrix at time , which quantifies strength and direction of a causal relationship, wherein a matrix element is a real number:

[0017] When denotes has a promoting effect on ;

[0018] When denotes right It has an inhibitory effect;

[0019] The larger the absolute value, the greater the influence intensity;

[0020] This represents the time delay matrix, whose elements are... ;

[0021] A non-negative integer represents a variable. Changes Only after a certain time step will the variables be adjusted. It has a significant impact.

[0022] Preferably, the macroscopic causal knowledge construction module further includes a spatiotemporal causal graph dynamic evolution unit, used to receive the calibration signal generated by the model adaptive calibration module, and update the weight matrix of the spatiotemporal causal graph according to the following update rules:

[0023] :

[0024] In the formula, Indicates the current moment Below, in the spacetime causal graph from the nodes Pointing to node causal path The updated weight values ​​reflect the strength of the causal relationship at the current moment;

[0025] Indicates the previous moment Time path The old weight values;

[0026] This represents the learning rate parameter;

[0027] Indicates the previous moment Time by structured calibration signal Provided for paths The calibration amount.

[0028] Preferably, the coupled simulation and inference execution module defines the micro-decision subjects in the territorial spatial system as a set of intelligent agents, and instantiates the set of intelligent agents based on a unified data foundation; each intelligent agent evaluates different decision options based on a utility function; the utility function contains a set of preference weight coefficients.

[0029] Preferably, the preference weight coefficient is dynamically correlated with the spatiotemporal causal graph; specifically, the preference weight coefficient... defined as a mapping function whose input is the set of weights associated with the in- and out- causal paths corresponding to the agent decision attribute in the spatio-temporal causal graph .

[0030] Preferably, the conflict perception and scenario differentiation module quantifies the dynamic gap between the macro trend and the micro behavior trend by computing a conflict metric; the formula of the conflict metric is:

[0031] :

[0032] wherein, denotes the dynamic conflict metric, denotes the key variable being monitored, denotes the time variable, denotes the micro behavior trend function, denotes the instantaneous change rate of the micro trend, denotes the macro trend function, the instantaneous change rate of the macro trend, denotes the two-norm.

[0033] Preferably, the conflict perception and scenario differentiation module sets a pre-defined threshold , and when the conflict metric exceeds the threshold, the current time step is identified as a scenario bifurcation point, and the generation of multiple future scenarios is automatically triggered.

[0034] Preferably, the multi-scenario resilience strategy evaluation module combines each alternative planning strategy with each of the multiple future scenarios to form strategy-scenario pairs, and dispatches the coupled simulation and deduction execution module to perform simulation and deduction for each pair.

[0035] Preferably, the resilience evaluation function is used to calculate the comprehensive resilience score of each planning strategy under all scenarios, and the formula is:

[0036] ;

[0037] wherein, denotes the final evaluation score under a given planning strategy ;

[0038] denotes the th planning strategy;

[0039] denotes a weight coefficient;

[0040] Represents a set of scenarios;

[0041] First item It is the bottom line guarantee of weight;

[0042] It is a utility function used to extract the most critical bottom-line indicators from a multi-dimensional evaluation indicator vector and ensure that their values ​​are as high as possible.

[0043] The operator then identifies the worst-case performance that the strategy can achieve across all inference scenarios;

[0044] Second item It represents the weight of stability;

[0045] It is a utility function used to extract indicators that measure the degree to which the core goals of the plan have been achieved;

[0046] The operator calculates the variance of the index across all scenarios. The smaller the variance, the less sensitive the implementation effect of the strategy is to future uncertainties, and the more stable the performance.

[0047] Preferably, the bias attribution method implemented by the model adaptive calibration module is as follows: calculating the partial derivative of the bias with respect to the weight of each specific causal path pointing to the bias variable in the spatiotemporal causal graph to generate the calibration signal; the calculation formula is:

[0048] :

[0049] In the formula, Indicates attribution to a specific causal path The calibration quantity or partial derivative coefficient, Indicates macroscopic deviation For weight parameters The partial derivatives, Indicates a point in time Above, macroeconomic variables The overall macroscopic deviation.

[0050] This invention provides a multi-source data fusion decision support system for land spatial planning. It has the following beneficial effects:

[0051] 1. This invention constructs a spatiotemporal causal map that characterizes the macroscopic causal relationships among land spatial elements, surpassing the limitations of traditional planning that relies on static data and correlation analysis. It can reveal the deep-seated causal transmission mechanism within the system. By setting up a model adaptive calibration module, it can compare the virtual data generated by simulation with real-world observation data, attribute the deviation to specific causal paths in the map, and generate calibration signals for feedback updates, forming a closed-loop learning and evolution system. This ensures that the system has continuous adaptability to the dynamically changing real world and long-term predictive accuracy.

[0052] 2. This invention defines the micro-decision subject as a set of intelligent agents by coupling the simulation and deduction execution modules, and dynamically associates the decision-making basis of the intelligent agents with the macro-spatiotemporal causal graph. This enables the system to not only assess the macro-impact of policies, but also simulate how these impacts are transmitted to the micro-level and trigger chain reactions in behavior. Thus, it can conduct a more comprehensive and realistic assessment of the comprehensive and systemic effects of policies.

[0053] 3. This invention utilizes a conflict perception and scenario differentiation module to monitor the dynamic gap between macro trends and micro behavioral trends in real time. It also employs conflict metrics to automatically identify scenario bifurcation points where internal conflicts escalate, thereby endogenously generating multiple future scenarios. Based on this, a multi-scenario resilience strategy evaluation module uses a resilience evaluation function to assess the robustness and bottom-line performance of candidate strategies under all identified critical scenarios, ultimately selecting highly resilient planning strategies. Attached Figure Description

[0054] Figure 1 This is a system framework diagram of the present invention;

[0055] Figure 2 This is a matrix diagram illustrating the multi-scenario resilience strategy evaluation of the present invention. Detailed Implementation

[0056] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] Please see the appendix Figure 1 and attached Figure 2 This invention provides a multi-source data fusion decision support system for land spatial planning, comprising:

[0058] The data foundation and preprocessing module is used to access, clean, spatiotemporally align, and standardize multi-source heterogeneous data related to land and space planning in order to build a unified data foundation.

[0059] Specifically, in this embodiment, the data foundation and preprocessing module is responsible for transforming the diverse and formatted original data related to land and space planning into a standardized dataset with unified structure, spatiotemporal alignment, and consistent dimensions, i.e., a unified data foundation.

[0060] The specific implementation of the data foundation and preprocessing module includes the following processing units:

[0061] Multi-source heterogeneous data access and cleaning unit. The data foundation and preprocessing module is equipped with multiple data interfaces for accessing various types of data required for land spatial planning analysis, including but not limited to: vector format land use status data, point of interest (POI) data, and road network data; raster format remote sensing image data and digital elevation model (DEM) data; and tabular format socio-economic statistics, population census data, and mobile signaling data. After data access, the cleaning unit will process the data quality, for example, using the 3-sigma principle or local anomaly factor algorithm to identify and remove extreme outliers in statistical data, using Kriging interpolation or time-series mean interpolation methods to fill missing values ​​in time series data, and correcting data with inconsistent geographic coordinate systems.

[0062] Unified Spatiotemporal Reference Construction Unit. To achieve spatiotemporal alignment of multi-source heterogeneous data, the data foundation and preprocessing module first constructs a unified spatiotemporal reference covering the entire study area. Spatially, a unified geographic coordinate system (e.g., WGS-84) and a two-dimensional Cartesian grid system are constructed, dividing the study area into a set of equal-sized, non-overlapping rectangular spatial cells. ,in The total number of spatial cells, each with a defined spatial resolution, such as 1 km × 1 km. In the time dimension, a discrete time step is defined. (e.g., a month or a quarter) to form a unified time series ,in This represents the total number of time steps.

[0063] Spatiotemporal aggregation unit. Based on the established unified spatiotemporal benchmark, the data foundation and preprocessing module performs spatiotemporal aggregation processing on all cleaned data. Appropriate aggregation functions are used for different data types. A variable In spatial units and time point aggregate value The calculation process is as follows;

[0064] :

[0065] In the formula, Representing variables In spatial units and time point The aggregate value;

[0066] This represents aggregate functions, which can take different forms depending on the data type.

[0067] For point data It can be a counting function or a kernel density estimation function;

[0068] For areal data (such as land use). It can be an area-weighted average or a dominance type extraction function;

[0069] For raster data, It can be a resampling function for pixel values;

[0070] Regarding statistical data related to administrative regions, This is a spatial weight-based allocation function;

[0071] This represents a set of original data points and their corresponding set of values. Each data point... Each corresponds to an observation value ;

[0072] Indicates the first One original data point;

[0073] Representing data points The observed values ​​or attribute values, such as intensity, quantity, density, etc.

[0074] Representing data points Spatial location coordinates;

[0075] Representing data points The timestamp, the observation or collection time of the data point;

[0076] Indicates the first Each spatial unit can be a regular grid unit, an administrative region unit, or other spatial division unit;

[0077] Indicates the first The starting time of a point in time or a time interval;

[0078] Indicates a time interval, starting from a specific point in time. Start to The previous time period was used to filter data points that belong to the time window;

[0079] Representing data points Spatial location belongs to spatial unit ;

[0080] Representing data points The time belongs to the time interval .

[0081] Data standardization unit. To eliminate the influence of differences in physical units and numerical ranges between different variables, the data foundation and preprocessing module standardizes all aggregated variable data. One specific implementation method is to use the min-max normalization method to standardize variable values. Linear mapping to interval;

[0082] ;

[0083] In the formula, It is the normalized value. It is the original value. and These are the minimum and maximum values ​​of the variable in the global observation dataset, respectively. In another embodiment, the Z-score standardization method can also be used to transform the data into a distribution with a mean of 0 and a standard deviation of 1. After processing by all the above units, the final data product generated is the unified data foundation.

[0084] The macro-causal knowledge construction module, connected to the data foundation and preprocessing module, is used to construct a spatiotemporal causal map representing the macro-causal relationship between land space elements from the unified data foundation, and to generate macro trends based on the spatiotemporal causal map;

[0085] Specifically, in this embodiment, the macro-causal knowledge construction module has the core function of receiving a unified data foundation output by the data foundation and preprocessing module, and constructing a spatiotemporal causal map that can dynamically evolve and represent the macro-causal relationships between various elements of the national land space. It also generates macro trends based on the constructed spatiotemporal causal map.

[0086] The spatiotemporal causal graph data structure definition unit abstracts and defines the macroscopic causal relationship between land space elements as a weighted directed acyclic graph that evolves over time, namely the spatiotemporal causal graph;

[0087] :

[0088] In the formula, Represents the set of nodes in a graph, where each node... This corresponds to a spatial variable in a unified data foundation.

[0089] Indicates the total number of variables;

[0090] Indicates time The set of directed edges, each directed edge Representing variables For variables There is a direct causal relationship at every moment;

[0091] Indicates time The weight matrix quantifies the strength and direction of causal relationships, where the matrix elements... It is a real number:

[0092] express right It has a promoting effect;

[0093] express right It has an inhibitory effect;

[0094] The larger the absolute value, the greater the influence intensity;

[0095] This represents the time delay matrix, whose elements are... ;

[0096] A non-negative integer represents a variable. Changes Only after a certain time step will the variables be adjusted. It has a significant impact.

[0097] The initial spatiotemporal causal graph construction unit, based on a unified data foundation, employs a causal discovery algorithm to construct the initial spatiotemporal causal graph. One specific implementation is to use a vector autoregression (VAR) model for spatiotemporal panel data combined with the Granger causality test. For any pair of variables... By testing variables Can historical time series significantly improve the variability The accuracy of future value predictions is used to determine causal relationships. Does it exist? If it is statistically significant, then create a directed edge in the graph. The model regression coefficients are used as their initial weights. The optimal lag order is used as the time delay. .

[0098] The Spatiotemporal Causality Graph Dynamic Evolution Unit. To ensure that the spatiotemporal causality graph can reflect the non-stationary changes in the real world, it possesses a dynamic evolution capability. The Spatiotemporal Causality Graph Dynamic Evolution Unit is equipped with a standard data input interface for receiving structured calibration signals. The calibration signal contains information about specific causal paths within the spectrum. calibration amount Upon receiving the signal, the spatiotemporal causal graph dynamic evolution unit updates the weight matrix of the spatiotemporal causal graph using incremental learning. The specific update rules are as follows:

[0099] :

[0100] In the formula, Indicates the current moment Below, in the spacetime causal graph from the nodes Pointing to node causal path The updated weight values ​​reflect the strength of the causal relationship at the current moment;

[0101] Indicates the previous moment Time path The old weight values;

[0102] This represents the learning rate parameter;

[0103] Indicates the previous moment Time by structured calibration signal Provided for paths The calibration amount.

[0104] Macroeconomic trend generation unit. After the spatiotemporal causal graph is constructed or updated, it is used as a prediction model to generate macroeconomic trends. For any node variable in the graph... In its future moments Predicted value It is obtained by weighted summation of the state values ​​of all its parent nodes before the corresponding time delay:

[0105] :

[0106] In the formula, Indicated in spatial unit Internal, node variables In the future The predicted value;

[0107] This represents all pointer nodes in the causal graph. The set of parent node variables, that is, those directly affecting The set of predecessor nodes;

[0108] Represents parent node variables In spatial units Internal, relative to the current time With time delay Observed values;

[0109] Indicates the current moment From node To the node The causal weight value;

[0110] ;

[0111] Indicates a spatial unit or region index, used to distinguish dynamic processes at different spatial locations;

[0112] Indicates a time step.

[0113] By forecasting all key variables, a set of multivariate future time series is generated, which represents the macroeconomic trend.

[0114] The coupled simulation and deduction execution module is connected to the macro-causal knowledge construction module and is used to deduce specific land space planning interventions under the constraints of the spatiotemporal causal graph in order to generate virtual data of future spatiotemporal states and micro-behavioral trends.

[0115] Specifically, in this embodiment, the coupled simulation and deduction execution module, under the macroscopic constraints of the spatiotemporal causal graph, uses a bottom-up multi-agent simulation method to deduce specific land spatial planning interventions (such as new land use policies or infrastructure construction) input by the user, and outputs virtual data including future spatiotemporal states and micro-behavioral trends.

[0116] The specific implementation of the coupled simulation and deduction execution module includes the following processing units:

[0117] The intelligent agent definition and initialization unit abstracts the micro-decision-making entities in the territorial spatial system into a set of intelligent agents with autonomous decision-making capabilities. ,in This represents the total number of intelligent agents. These agents are categorized into different types, such as resident agents, enterprise agents, and real estate development agents. Each intelligent agent... It has a set of internal state properties For example, the attributes of a resident intelligent agent may include age, income, occupation, and family structure; the attributes of a business intelligent agent may include industry type, size, and technological level. At the start of the simulation, the intelligent agent set is configured based on a unified data foundation (e.g., census data, business registration data) provided by the data foundation and preprocessing module. Perform instantiation and space distribution initialization.

[0118] The planning intervention scenario analysis unit receives land spatial planning intervention scenarios input by the user. These interventions can include establishing new land use zones, constructing transportation infrastructure (such as subway lines), or implementing industrial subsidy policies. The planning intervention scenario analysis unit interprets the planning intervention as a modification of simulated environmental parameters, such as modifying spatial units. Accessibility attributes, land use constraints, or decision-making costs of agents.

[0119] A macro-micro coupled decision-making model unit. The key to achieving macro-micro bidirectional coupling in this invention lies in the agent's decision-making model. At each simulation time step... intelligent agent Based on a utility function To evaluate different decision options (e.g., selecting the space unit to move into) The satisfaction level of an intelligent agent. For spatial units Addressing utility function For example:

[0120] :

[0121] In the formula, Represents intelligent agents At any moment For spatial units The overall utility value;

[0122] Indicates the first One spatial unit;

[0123] Represents a discrete time step (moment);

[0124] Representing spatial units At any moment The Each location attribute value;

[0125] The index representing the attribute or characteristic;

[0126] This represents the total number of location attributes.

[0127] Represents intelligent agents At any moment For the The preference weight coefficients for each attribute.

[0128] This represents a random disturbance term.

[0129] The coupling mechanism of this invention is reflected in the preference weight. On dynamic calculations. Instead of preset fixed parameters, they are related to the generated spatiotemporal causal graph. Dynamic association. Specifically, location attributes. Corresponding to a node in the spacetime causal graph Intelligent agents Preference weights for attributes It is defined as a mapping function. The output of the mapping function is the graph containing nodes. The set of weights for associated causal paths:

[0130] :

[0131] In the formula, Represents a mapping function. Represents the spacetime causal graph The middle corresponds to the location attribute The node, Indicates from node The set of weights for all outward causal paths originating from this point. Represents all pointer nodes The set of inward causal path weights Represents the causal path weight, starting from the node. To the node The edge at any moment The weight value, Represents another type of causal path weight, from node Pointing to node The edge at any moment The weight, Indicates time The set of causal edges below.

[0132] Simulation and state update unit. During the simulation process, all agents... The utility values ​​of all feasible options are calculated in parallel or in a specific order, and a probabilistic choice is made using decision rules (such as a multinomial logit model). The agent makes the decision (e.g., residents...). Decision from Relocated After that, the system will update the space unit. and Status attributes (e.g., population size, land occupancy status) will be considered during the simulation run at a specified time. After that, all spatial units All time steps The set of state attributes constitutes the virtual data of the future spatiotemporal state. .

[0133] Micro-behavioral trend aggregation unit. At each time step It aggregates the decision-making behaviors of all intelligent agents in real time to generate micro-behavioral trends. These micro-behavioral trends... For specific macroeconomic variables Emergent measures at the micro level (e.g., 'housing demand'). For example, the variable corresponding to 'housing demand'. Micro-behavioral trends It can be calculated as at time step All attempts to migrate into space units The total utility or total number of resident intelligent agents (but which failed due to capacity limitations).

[0134] The conflict perception and scenario differentiation module is connected to the macro causal knowledge construction module and the coupled simulation and deduction execution module. It is used to continuously monitor the consistency between the macro trend generated by the macro causal knowledge construction module and the micro behavioral trend generated by the coupled simulation and deduction execution module during the deduction process, and to trigger the generation of multiple future scenarios when a conflict is identified.

[0135] Specifically, in this embodiment, the conflict perception and scenario differentiation module continuously monitors the consistency between the predicted direction of macroscopic laws and the convergence trend of microscopic intelligent agent behavior during the deduction process, thereby endogenously discovering uncertainties in the system evolution path and triggering the generation of multiple future scenarios when major disagreements are identified.

[0136] The specific implementation of the conflict perception and context differentiation module includes the following processing units:

[0137] A unit that acquires both macro and micro trends in parallel. The conflict perception and scenario differentiation module performs this at each time step of the simulation. Two key data streams are received in parallel:

[0138] Macro trends generated by the macro causal knowledge building module It reflects variables inferred based on historical causal relationships. exist The predicted value or rate of change at any given time.

[0139] Microscopic behavioral trends aggregated in real time by coupled simulation and inference execution modules It reflects the state of all intelligent agents at any given moment. After the decision-making behaviors are aggregated, the variables are analyzed. The actual or potential cumulative impact and rate of change.

[0140] Dynamic Conflict Measurement Unit. The significant innovation of this invention lies in introducing a quantitative index to measure the dynamic inconsistency between macroscopic trends and microscopic behavioral trends, thereby enabling the automatic detection of uncertainties along the system's evolution path. This is achieved by calculating a conflict measurement index. to quantify variables The dynamic gap between the macroscopic expected rate of change and the microscopic emergent rate of change:

[0141] :

[0142] In the formula, This represents a dynamic conflict metric. This indicates the key variables being monitored. Represents a time variable. This represents a micro-behavioral trend function. The instantaneous rate of change representing micro-trends Function representing macroeconomic trends The instantaneous rate of change of macroeconomic trends This represents the L2 norm.

[0143] The scenario bifurcation point identification and differentiation unit sets a preset threshold. When any key variable Conflict metrics Exceeding the threshold At that time, that is The system will set the current time step This was identified as a scenario bifurcation point. The emergence of this conflict indicates that macro-historical patterns can no longer fully constrain or explain the collective emergence of micro-behaviors under current planning interventions, suggesting that the system may enter multiple different evolutionary paths in the future.

[0144] Once the scenario bifurcation point is identified, the scenario differentiation process is immediately triggered, automatically generating a set of scenarios containing multiple alternative futures. ,in This refers to the number of scenarios generated. The scenario differentiation mechanism, based on the nature and direction of the conflict, perturbs key parameters or model assumptions that lead to the conflict:

[0145] Scenario Differentiation Strategy 1 (Parameter Perturbation): Based on the direction of conflict, adjust the utility function in the coupled simulation and deduction execution module. random disturbance term The distribution parameters are systematically modified (e.g., by increasing or decreasing the standard deviation) to reflect the two opposing possibilities of weakening or strengthening macroscopic constraints, thereby generating two or more extreme scenarios. .

[0146] Scenario Differentiation Strategy Two (Knowledge Disturbance): Spatiotemporal Causal Mapping in the Macro-Causal Knowledge Construction Module Specific causal pathways in weight Positive and negative perturbations are applied to simulate different failure modes that macro-knowledge may exhibit when faced with new planning interventions.

[0147] Scenario set output and feedback unit. The scenario differentiation unit will generate a scenario set. As structured data output. Each scenario Each includes a set of modification settings for key model parameters or macroscopic causal knowledge that take effect after the bifurcation point, affecting the coupled simulation and deduction execution modules.

[0148] The multi-scenario resilience strategy evaluation module is connected to the conflict perception and scenario differentiation module and the coupled simulation and inference execution module. It is used to receive planning strategies, schedule the coupled simulation and inference execution module to perform cross-inference evaluation in the multiple future scenarios, and sort the planning strategies based on the resilience evaluation function to output decision suggestions.

[0149] Specifically, in this embodiment, the multi-scenario resilience strategy evaluation module receives user-defined alternative planning strategies and schedules the coupled simulation and deduction execution module to conduct comprehensive cross-deduction and evaluation across multiple future scenarios generated by the conflict perception and scenario differentiation module. Finally, the planning strategies are ranked based on a defined resilience evaluation function to output robust decision recommendations.

[0150] The specific implementation of the multi-scenario resilience strategy evaluation module includes the following processing units:

[0151] The planning strategy and evaluation index definition unit first receives a set of alternative planning strategies input by the user. ,in This represents the total number of strategies. Simultaneously, users or the system predefine a multi-dimensional evaluation index system to measure the effectiveness of the planning strategy implementation, such as economic efficiency indicators (GDP growth rate per unit area), social equity indicators (changes in the Gini coefficient), and ecological environment indicators (green coverage rate, carbon emissions).

[0152] The cross-simulation evaluation and scheduling unit receives a set of scenarios from the conflict perception and scenario differentiation module. For each alternative planning strategy The multi-scenario resilience strategy assessment module (Module 5) integrates it with the scenario set. Each scene Combine them to form a "strategy-scenario" pair. Subsequently, the multi-scenario resilience strategy assessment module, acting as a scheduling center, automatically invokes the coupled simulation and inference execution module for each paired scenario. As input, a complete simulation is performed. After the simulation, an evaluation function is used. Predefined evaluation indicators are extracted from the simulation results to form a multi-dimensional evaluation indicator vector. :

[0153] :

[0154] In the formula, Indicates that under a given planning strategy With context The resulting multi-dimensional evaluation index vector Indicates the first One alternative planning strategy Indicates the first This scenario Indicates time The spatiotemporal causal graph of moments This indicates the coupled simulation and deduction execution process. This represents the evaluation function.

[0155] The resilience assessment function calculation unit does not aim to find the optimal strategy under a single forecast scenario, but rather to identify the most resilient strategy in the face of uncertainty. Resilience is defined as the combined stability and bottom-line guarantee capability of a strategy across all possible scenarios. To this end, the resilience assessment function calculation unit introduces a resilience assessment function. Used to calculate each planning strategy Overall resilience score across all scenarios:

[0156] ;

[0157] In the formula, Indicates that under a given planning strategy The final evaluation score;

[0158] Representing the One planning strategy;

[0159] This represents a weighting coefficient;

[0160] Represents a set of scenarios;

[0161] First item It is the bottom line guarantee of weight;

[0162] It is a utility function used to extract the most critical bottom-line indicators (e.g., minimum economic growth rate or maximum unemployment rate) from a multi-dimensional evaluation indicator vector and ensure that its value is as high as possible;

[0163] The operator identifies the worst-case performance the strategy can achieve across all hypothetical scenarios. A higher score indicates a stronger resilience to risk and a greater ability to effectively avoid catastrophic consequences.

[0164] Second item It represents the weight of stability;

[0165] It is a utility function used to extract indicators that measure the degree to which core planning objectives are achieved (e.g., per capita public service level).

[0166] The operator calculates the variance of the index across all scenarios. The smaller the variance, the less sensitive the implementation effect of the strategy is to future uncertainties, and the more stable the performance.

[0167] Strategy ranking and decision recommendation output unit. This involves ranking all alternative planning strategies. Calculate its toughness score Then, all strategies are sorted from highest to lowest according to their resilience scores. Finally, the system outputs an ordered list of planning strategies, along with the corresponding resilience score, bottom-line guarantee score, and performance robustness score for each strategy, as final decision-making recommendations for planners.

[0168] The model adaptive calibration module is connected to the coupled simulation and inference execution module and the macroscopic causal knowledge construction module. It is used to compare the deviation between the virtual data of the future spatiotemporal state generated by the coupled simulation and inference execution module and the real-world observation data, attribute the deviation to a specific causal path in the spatiotemporal causal graph, and generate a calibration signal to be fed back to the macroscopic causal knowledge construction module to update the spatiotemporal causal graph.

[0169] Specifically, in this embodiment, the model adaptive calibration module periodically acquires real-world observation data, compares it with virtual data of future spatiotemporal states generated by the coupled simulation and inference execution module to calculate the deviation, and intelligently allocates the deviation to specific causal paths in the spatiotemporal causal graph through a causal path attribution algorithm. Finally, a calibration signal is generated and fed back to the macroscopic causal knowledge construction module to update the spatiotemporal causal graph.

[0170] The specific implementation of the model adaptive calibration module includes the following processing units:

[0171] Real-world observation data acquisition and alignment unit. Equipped with a data monitoring interface to acquire the latest real-world observation data from external data sources at specific points in time after the simulation (e.g., one quarter or one year later). The type and data basis of the real-world observation data are consistent with the data types accessed by the preprocessing module, such as the latest land use status interpreted from remote sensing images and the latest socio-economic statistics. The spatiotemporal benchmark construction and aggregation functions in the data basis and preprocessing module are invoked to ensure the newly acquired real-world observation data... Virtual data of future spatiotemporal states output by the coupled simulation and inference execution module They are completely identical in spatial resolution, time step and data structure, providing a comparable basis for subsequent deviation calculations.

[0172] Spatiotemporal deviation calculation unit. Receives virtual data on the future spatiotemporal state from the coupled simulation and deduction execution module. and real-world observation data processed from the previous unit The spatiotemporal deviation calculation unit calculates the spatial deviation by comparing the two. and time point Regarding variables macroscopic deviation :

[0173] :

[0174] In the formula, Indicates the macroscopic variable being compared, the first... An indicator or observation, Represents a spatial unit. Indicates a point in time. Indicates in variable Below, the virtual future spatiotemporal state data obtained from model deduction or simulation, i.e., model predictions, Indicates in variable Below, the corresponding spatiotemporal state data comes from real-world observations or measurements, i.e., the true values. Represents the loss function, when When taking the MSE format;

[0175] ;

[0176] Indicated in spatial unit Time point Below, variables Local macroscopic deviations;

[0177] Aggregate the deviations of all spatial units to obtain information about the variables. Overall macroeconomic deviation;

[0178] Indicates a point in time Above, variables Overall macro deviation, This represents the loss function.

[0179] The causal path attribution algorithm unit, when identifying macroscopic deviations... Instead of using simple black-box model tuning (e.g., adjusting agent parameters in the coupled simulation and inference execution module), the system traces the root cause of the deviation to the spatiotemporal causal graph in the macroscopic causal knowledge construction module. The incompleteness or obsolescence of the data. The causal path attribution algorithm unit implements a causal path attribution algorithm to calculate the overall macroscopic deviation. To what extent should the pointer variable be determined? Each specific causal path To take responsibility.

[0180] One specific implementation method utilizes backpropagation and sensitivity analysis. This is based on virtual data of future spatiotemporal states. It is generated through a coupled simulation and inference execution module, and the agent's decision-making in the coupled simulation and inference execution module is influenced by the weights in the spatiotemporal causal graph. Constraints. Therefore, macroscopic deviations. It can be viewed as all causal weights in a spacetime causal graph. A complex, non-linear function, the goal of the causal path attribution algorithm unit is to calculate the bias. For each pointer causal path weight The partial derivatives of , i.e., sensitivity;

[0181] :

[0182] In the formula, Indicates attribution to a specific causal path The calibration quantity or partial derivative coefficient, Indicates macroscopic deviation For weight parameters The partial derivatives, Indicates a point in time Above, macroeconomic variables The overall macroscopic deviation.

[0183] Quasi-signal generation and feedback unit. The causal path attribution algorithm unit calculates all relevant calibration values. Subsequently, the quasi-signal generation and feedback unit encapsulates these attribution results into a structured calibration signal. The calibration signal It is a set containing {path identifier, calibration value} pairs, for example The model adaptive calibration module then uses this calibration signal. The signal is sent to the dynamic evolution unit of the macro-causal knowledge construction module. Upon receiving the signal, the macro-causal knowledge construction module will, based on the signal (e.g., using...),... (As an update for gradient descent) to adjust the corresponding weights in the spatiotemporal causal graph. .

Claims

1. A multi-source data fusion decision support system for land spatial planning, characterized in that, include: The data foundation and preprocessing module is used to process multi-source heterogeneous data related to land and space planning in order to build a unified data foundation. The macro-causal knowledge construction module, connected to the data foundation and preprocessing module, is used to construct a spatiotemporal causal map representing the macro-causal relationships between land space elements from a unified data foundation, and to generate macro trends based on the spatiotemporal causal map. The coupled simulation and deduction execution module is connected to the macro-causal knowledge construction module. It is used to deduce specific land space planning interventions under the constraints of spatiotemporal causal graphs to generate virtual data of future spatiotemporal states and micro-behavioral trends. The conflict perception and scenario differentiation module is connected to the macro-causal knowledge construction module and the coupled simulation and inference execution module. It is used to monitor the consistency between the macro trend and the micro-behavioral trend, and to trigger the generation of multiple future scenarios when a conflict is identified. The multi-scenario resilience strategy evaluation module is connected to the conflict perception and scenario differentiation module and the coupled simulation and inference execution module. It is used to receive planning strategies, schedule the coupled simulation and inference execution module to perform cross-inference evaluation in multiple future scenarios, and rank the planning strategies based on the resilience evaluation function to output decision recommendations. The model adaptive calibration module, connected to the coupled simulation and inference execution module and the macroscopic causal knowledge construction module, is used to compare the deviation between virtual data of future spatiotemporal states and real-world observation data, attribute the deviation to specific causal paths in the spatiotemporal causal graph, and generate calibration signals to feed back to the macroscopic causal knowledge construction module to update the spatiotemporal causal graph.

2. The multi-source data fusion decision support system for land spatial planning according to claim 1, characterized in that, The spatiotemporal causal graph is defined as a weighted directed acyclic graph that evolves over time; : In the formula, Represents the set of nodes in a graph, where each node... This corresponds to a spatial variable in a unified data foundation. Indicates the total number of variables; Indicates time The set of directed edges, each directed edge Representing variables For variables There is a direct causal relationship at every moment; Indicates time The weight matrix quantifies the strength and direction of causal relationships, where the matrix elements... It is a real number: when express right It has a promoting effect; when express right It has an inhibitory effect; The larger the absolute value, the greater the influence intensity; This represents the time delay matrix, whose elements are... ; A non-negative integer represents a variable. Changes Only after a certain time step will the variables be adjusted. It has a significant impact.

3. The multi-source data fusion decision support system for land spatial planning according to claim 1, characterized in that, The macroscopic causal knowledge construction module also includes a spatiotemporal causal graph dynamic evolution unit, which receives the calibration signal generated by the model adaptive calibration module and updates the weight matrix of the spatiotemporal causal graph according to the following update rules: : In the formula, Indicates the current moment Below, in the spacetime causal graph from the nodes Pointing to node causal path The updated weight values ​​reflect the strength of the causal relationship at the current moment; Indicates the previous moment Time path The old weight values; This represents the learning rate parameter; Indicates the previous moment Time by structured calibration signal Provided for paths The calibration amount.

4. The multi-source data fusion decision support system for land spatial planning according to claim 1, characterized in that, The coupled simulation and inference execution module defines the micro-decision-making entities in the territorial spatial system as a set of intelligent agents, and instantiates the set of intelligent agents based on a unified data foundation; each intelligent agent evaluates different decision options based on a utility function; the utility function contains a set of preference weight coefficients.

5. A multi-source data fusion decision support system for land spatial planning according to claim 4, characterized in that, The preference weight coefficient is dynamically correlated with the spatiotemporal causal graph; specifically, the preference weight coefficient... Defined as a mapping function The output of the mapping function is the spatiotemporal causal graph related to the agent's decision attributes. corresponding nodes The weight set of associated introverted and extroverted causal paths.

6. The multi-source data fusion decision support system for land spatial planning according to claim 1, characterized in that, The conflict perception and scenario differentiation module quantifies the dynamic gap between the macro trend and the micro behavioral trend by calculating a conflict metric; the calculation formula for the conflict metric is as follows: : In the formula, This represents a dynamic conflict metric. This indicates the key variables being monitored. Represents a time variable. This represents a micro-behavioral trend function. The instantaneous rate of change representing micro-trends Function representing macroeconomic trends The instantaneous rate of change of macroeconomic trends This represents the L2 norm.

7. A multi-source data fusion decision support system for land spatial planning according to claim 1, characterized in that, The conflict perception and context differentiation module is set with a preset threshold. When the conflict metric When the threshold is exceeded, the current time step is identified as a scenario bifurcation point, and multiple future scenarios are automatically generated.

8. The multi-source data fusion decision support system for land spatial planning according to claim 1, characterized in that, The multi-scenario resilience strategy evaluation module combines each alternative planning strategy with each of the multiple future scenarios to form a strategy-scenario pair, and schedules the coupled simulation and deduction execution module to perform simulation and deduction for each pair.

9. A multi-source data fusion decision support system for land spatial planning according to claim 1, characterized in that, The resilience assessment function is used to calculate the comprehensive resilience score of each planning strategy across all scenarios, and its calculation formula is as follows: ； In the formula, Indicates that under a given planning strategy The final evaluation score; Representing the One planning strategy; This represents a weighting coefficient; Represents a set of scenarios; First item It is the bottom line guarantee of weight; It is a utility function used to extract the most critical bottom-line indicators from a multi-dimensional evaluation indicator vector and ensure that their values ​​are as high as possible. The operator then identifies the worst-case performance that the strategy can achieve across all inference scenarios; Second item It represents the weight of stability; It is a utility function used to extract indicators that measure the degree to which the core goals of the plan have been achieved; The operator calculates the variance of the index across all scenarios. The smaller the variance, the less sensitive the implementation effect of the strategy is to future uncertainties, and the more stable the performance.

10. A multi-source data fusion decision support system for land spatial planning according to claim 1, characterized in that, The bias attribution method implemented by the model adaptive calibration module is as follows: The partial derivative of the bias with respect to the weight of each specific causal path pointing to the bias variable in the spatiotemporal causal graph is calculated to generate the calibration signal; the calculation formula is as follows: : In the formula, Indicates attribution to a specific causal path The calibration quantity or partial derivative coefficient, Indicates macroscopic deviation For weight parameters The partial derivatives, Indicates a point in time Above, macroeconomic variables The overall macroscopic deviation.

Images