An ai spatiotemporal multi-element information collection and analysis system and method for urban renewal project generation field

Abstract

This invention relates to the fields of smart city and urban planning technology, specifically disclosing an AI-based spatiotemporal multivariate information acquisition and analysis system and method for urban renewal project generation. The method includes: collecting and standardizing urban spatiotemporal multivariate information to generate spatiotemporal feature sequences; encoding the feature sequences into urban spatial state sequences using unsupervised learning; constructing a directed weighted network dynamics model for cross-spatial state propagation based on causal analysis between state sequences; defining renewal intervention operations in the model and simulating the network cascading diffusion process of anchored states at different intervention points to calculate the long-term global optimization effect; integrating the optimization effect and cost calculation to determine intervention effectiveness, thereby generating recommended urban renewal project schemes with key intervention spaces and strategies. The core of this invention lies in constructing and simulating a networked propagation model of urban spatial states, thereby achieving quantitative extrapolation and identification of the long-term effects and key lever nodes of renewal interventions.

Description

Technical Field

[0001] This invention relates to the field of smart city and urban planning technology, and more specifically, to an AI spatiotemporal multi-dimensional information collection and analysis system and method for urban renewal project generation. Background Technology

[0002] As urban development shifts from incremental expansion to stock renewal, scientifically and accurately identifying areas with renewal potential and formulating effective intervention strategies has become a core issue in urban governance. The urban system is a complex spatiotemporal dynamic system, and its renewal needs are embedded in the evolutionary patterns of multi-source heterogeneous data on buildings, the environment, and pedestrian flow. Utilizing artificial intelligence technology to deeply collect and analyze diverse spatiotemporal information about cities, thereby providing data-driven decision support for the generation of renewal projects, has significant practical implications.

[0003] Current technologies primarily identify redevelopment areas by overlaying and indexing multi-source geospatial data. However, this approach typically suffers from the following shortcomings: First, the information integration methods are superficial, often relying on spatial registration and layer overlay, lacking in-depth exploration of the spatiotemporal correlations and causal relationships between data points, and failing to construct computational models that reflect the inherent operational patterns of cities. Second, the analysis focuses on static status quo evaluation or simple time-series comparisons, often assessing regional conditions by constructing a multi-dimensional indicator system encompassing socio-economic and material environmental factors and assigning weighted scores. This method fails to characterize the inherent mechanisms of dynamic evolution in urban spatial states and their cross-spatial transmission effects. Finally, the decision support logic leans towards "status quo diagnosis" rather than "intervention simulation." Existing methods often output a ranking of current "problem areas" but cannot quantitatively simulate the chain reactions and long-term global effects that specific redevelopment interventions (such as renovating a particular neighborhood) may trigger, resulting in a lack of forward-looking and systematic simulation basis for redevelopment strategy formulation.

[0004] Therefore, this paper proposes an AI-based spatiotemporal multivariate information collection and analysis system and method for urban renewal project generation. The problem to be solved is: how to break through the static and fragmented evaluation model in the current urban renewal analysis, and provide an analytical framework that can deeply explore the inherent dynamic laws of spatiotemporal data and simulate and evaluate the long-term and global effects of renewal intervention strategies. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an AI-based spatiotemporal multivariate information acquisition and analysis system and method for urban renewal project generation, to address the problems mentioned in the background art, comprising the following steps: S1: Through multiple heterogeneous data source interfaces, the spatiotemporal multi-dimensional information of the target area within a set historical time range is collected synchronously to form an original dataset. The information categories include at least building entity information, human settlement environment perception information, and dynamic human flow activity information, so as to provide a multi-dimensional data foundation covering urban static and dynamic elements for subsequent analysis. S2: Perform spatiotemporal alignment, scale unification, and feature extraction on the original dataset to generate a standardized spatiotemporal feature sequence indexed by basic geospatial units and arranged in chronological order, thereby integrating multi-source heterogeneous data into a structured input that can be directly processed by machine learning models; S3: Based on the spatiotemporal feature sequence, an unsupervised machine learning algorithm is used to abstractly represent the comprehensive status of each basic spatial unit at different time points, and generate a corresponding urban spatial state code for it. This code is used to discretize high-dimensional, continuous spatiotemporal features into spatial state categories with semantic meaning. S4: Based on the time series of the urban spatial state codes of all basic spatial units, analyze the sequential influence relationship of state evolution between units, and construct a directed weighted network model that reflects the law of cross-geographical spatial state propagation, where nodes represent basic spatial units and directed edges represent the direction and intensity of state influence. S5: In the directed weighted network model, an update intervention operation is defined, the core of which is to forcibly set the current state of one or more specified nodes (corresponding to basic spatial units) in the network to a preset anchor state that represents the ideal development goal. S6: For multiple candidate sets of update intervention units, simulate the execution of the intervention operation in the network model respectively, and iteratively calculate the cascading diffusion process of the anchor state along the network connection edge at multiple discrete time steps. Finally, quantitatively evaluate the long-term global space optimization effect that each candidate set can induce beyond the local range. S7: The long-term global optimization effect obtained by fusion calculation and the estimated cost required to implement physical updates for the candidate set are used to calculate the update intervention efficiency evaluation value for each candidate set through a preset efficiency evaluation function. This value aims to measure the global benefits that can be obtained per unit cost. S8: Prioritize all candidate sets according to the evaluation value of the effectiveness of the update intervention, and automatically generate urban renewal project plans that include the geographical scope of the recommended key intervention space, targeted update strategy suggestions, and expected impact range based on the ranking results.

[0006] Furthermore, step S3 specifically includes: S31: Input the spatiotemporal feature sequence of each basic spatial unit within the set historical time range into a temporal encoder model based on an attention mechanism; this model compresses and maps the high-dimensional feature sequence into a low-dimensional temporal embedding vector containing temporal information by capturing the temporal dependencies within the sequence, thereby achieving feature dimensionality reduction and information condensation. S32: Clustering algorithm is used to perform clustering analysis on the set of time-series embedding vectors of all basic spatial units at all time points; based on the proximity of vectors in the feature space, this process automatically discovers and summarizes a limited number of representative urban spatial state patterns, and the center vector of each cluster is defined as an urban spatial state prototype. S33: For each temporal embedding vector, calculate its similarity to each of the aforementioned urban spatial state prototypes, and classify it into the prototype category with the highest similarity; the unique identifier corresponding to the prototype is used as the urban spatial state encoding of the basic spatial unit at that point in time, thereby completing the classification from continuous features to discrete states.

[0007] Furthermore, the specific method for constructing the directed weighted network model in step S4 is as follows: S41: For any two different basic spatial units within the target area, extract their respective urban spatial status coding time series within the historical time range; S42: Using Granger causal analysis based on statistical hypothesis testing, analyze whether there is a statistically significant predictive or guiding relationship between the state sequence of the first unit and the state sequence of the second unit; if the test results show that there is a significant influence, then establish a directed edge in the network model from the node representing the first unit to the node representing the second unit to characterize the potential direction of the influence. S43: The statistical values ​​obtained from the Granger causality analysis and used to quantify the influence intensity are normalized and converted into a value between 0 and 1. This value is used as the initial weight of the corresponding directed edge to characterize the relative strength of the influence. S44: Traverse all possible basic spatial unit pairs, repeat the analysis and assignment process of steps S42 and S43, and finally integrate all nodes, directed edges and weights to construct a complete initial directed weighted network topology.

[0008] Furthermore, step S6, which involves simulating the cascade diffusion process and calculating the long-term global optimization effect, specifically includes: S61: For a selected set of candidate update intervention units, at the initial moment of the network simulation, the state values ​​of all corresponding nodes in the set are set to 1 to indicate that these nodes are in the anchored state due to the intervention; at the same time, the state values ​​of all other nodes in the network are initialized to 0 to indicate that they are in the non-anchored state. S62: At each simulation time step t, for each node i in the network whose current state value is still 0, calculate the total network influence input value received by node i by weighted summation based on the state values ​​of all its upstream neighbor nodes (i.e., all nodes that point to node i through directed edges) at the current time step and the weights of the corresponding directed edges. S63: Based on a preset state transition function with nonlinear characteristics, the total network influence input value received by node i is mapped to the update probability that the state value of the node changes from 0 to 1 in the next time step t+1; based on this probability, it is determined by random sampling whether the state of node i is updated to 1 in the next time step. S64: Repeat steps S62 and S63 to perform iterative simulation for N time steps, where N is a preset integer greater than 5, to simulate multiple stages of state propagation in the network; S65: After the simulation process is completed, count the total number of nodes in the entire directed weighted network model whose state value is finally 1, and calculate the proportion of them to the total number of nodes. This proportion is quantified as the long-term global optimization effect value caused by this specific candidate intervention set corresponding to this simulation, reflecting the radiation range of the intervention.

[0009] Furthermore, in step S62, for node At time step Total impact input value received The calculation is based on the following network dynamics formula: ; in, This represents all directed edges in the network that point to a node. neighboring nodes A set; Indicates from node Pointing to node The weight of the directed edge, which is derived from the causal strength quantization in step S43; Represents a node Simulated time step State value at time ( or The physical meaning of this formula lies in the nodes. The impact is the weighted sum of the current states of all its upstream neighbor nodes, with the weights reflecting the differences in the importance of the impact of different neighbors.

[0010] Furthermore, in step S7, the fusion evaluation function used to calculate the updated intervention efficacy evaluation value E adopts the following mathematical form: ; in, This represents the long-term global optimization effect value calculated through step S6; This represents the estimated total cost of implementing the actual physical modification and upgrade of the candidate set of update intervention units; and The preset index adjustment parameters are used to adjust the relative importance of effects and costs in the final evaluation value, and their typical value range is [value range missing]. , This function aims to describe the balance between pursuing global optimization effects and cost control under conditions of limited resources.

[0011] Furthermore, the automated generation of urban renewal project plans in step S8 specifically includes: S81: Based on the updated intervention efficacy evaluation value calculated in step S7, sort all candidate updated intervention units participating in the evaluation in descending order from high to low. S82: Select the top K candidate sets from the sorted list, where K is an integer not less than 1 preset by the user or system; spatially merge, connect and regularize the geographical areas covered by these candidate sets to finally determine one or more coherent "key intervention spaces" that are recommended for priority implementation; S83: For each basic spatial unit within the key intervention space, compare and analyze the differences between its original spatiotemporal characteristics and the ideal characteristic pattern corresponding to the anchoring state; based on these differences, generate a set of targeted update measures that include specific transformation directions, measure types, and intensity suggestions; S84: Integrate the geographical boundary information of the "key intervention space" identified above, the generated set of targeted update measures, and the expected state propagation and impact range obtained through simulation and prediction in step S6 into a structured urban renewal project plan document that can be directly used for decision-making reference or project initiation.

[0012] An AI spatiotemporal multivariate information acquisition and analysis system for implementing the above method, the system comprising the following functional modules: The data acquisition and preprocessing module is configured to execute steps S1 and S2, and is responsible for acquiring information from diverse heterogeneous data sources and completing data cleaning, alignment and standardization feature sequence generation. The state encoding and network modeling module is configured to execute steps S3 and S4. Its core function is to use an unsupervised learning algorithm to realize the abstract encoding of urban spatial state and to construct a network dynamics model reflecting spatial influence relationships based on the state sequence. The intervention simulation and efficacy evaluation module is configured to perform steps S5, S6 and S7, and is responsible for defining and simulating the updated intervention in the network model, calculating the cascading effects caused by the intervention, and evaluating efficacy by taking into account the overall costs. The scheme generation and output module is configured to execute step S8, automatically generate and output a structured urban renewal project scheme based on the performance evaluation results.

[0013] Furthermore, the state coding and network modeling module further includes: The temporal coding unit, which embeds an encoder model based on the Transformer architecture or Long Short-Term Memory (LSTM) network, is specifically designed to perform step S31, which efficiently encodes the spatiotemporal feature sequence into a low-dimensional temporal embedding vector. The state clustering unit integrates KMeans or hierarchical clustering algorithms to execute steps S32 and S33, realizing cluster analysis of temporal embedding vectors and allocation of state codes; The network inference unit integrates Granger causality test or transfer entropy calculation package to execute steps S41 and S44, automatically analyzing causal relationships between state sequences and constructing a directed weighted network.

[0014] Furthermore, the intervention simulation and efficacy evaluation module also includes an online model learning unit; this unit is configured to perform the following adaptive optimization process: Continuously collect information on urban renewal projects that have actually been implemented in the target area from external databases or sensor networks, as well as relevant spatial status monitoring data for a period of time after the completion of the projects; The actual state propagation effect and scope caused by the updated project are compared and analyzed with the results of the previous simulation prediction based on the network model in this module, and the prediction error is calculated. Based on the prediction error, the weight parameters of the relevant directed edges in the directed weighted network model are dynamically and slightly adjusted using the gradient descent algorithm or other backpropagation optimization techniques, and the parameters of the state transition function are fine-tuned. This allows the network model to continuously optimize the accuracy of its simulation predictions based on actual feedback data, thereby achieving a gradual approximation of the dynamic evolution of the real urban system.

[0015] The technical effects and advantages of this invention are as follows: Compared to existing technologies that rely on static weighted evaluation of multiple indicators, leading to fragmented decision-making dimensions, this invention constructs a unified metric, "urban spatial state," and quantifies renewal potential based on state transition costs. First, unsupervised learning is used to abstract state codes representing the comprehensive spatial condition from raw multivariate time-series data, transforming complex multidimensional indicators into discrete state sequences. Then, by analyzing the state sequences, the minimum expected cost required to migrate from the current state to the ideal target state is calculated, and its reciprocal is defined as the renewal potential. This embeds goal orientation and cost constraints into the assessment of renewal potential, providing a direct and quantitative basis for resource allocation.

[0016] Compared to the limitations of existing technologies in revealing the transmission paths and long-term effects of updating measures, this invention constructs and simulates a networked dynamics model based on state propagation to deduce intervention strategies. Based on the causal relationships between state sequences in historical data, a directed weighted network is constructed with basic spatial units as nodes and influence relationships as edges. By simulating the dynamic process of the cascading diffusion of anchor states along the network edges after an intervention (resetting the states of certain nodes to anchor states) is applied within this network, the long-term global spatial optimization effects triggered by different combinations of intervention points can be quantitatively assessed. This allows decision-makers to proactively compare different site selection schemes and identify key spatial nodes that can induce significant positive changes with relatively small interventions.

[0017] Compared to existing technologies that provide static decision support results and lack feedback and optimization capabilities, this invention enables the system to continuously evolve through an integrated model online learning mechanism. During operation, the system continuously collects monitoring data on actual implemented update projects and their subsequent impacts, comparing the actual impacts with simulated predictions. Based on these differences, the system dynamically adjusts the connection weights and state transition rule parameters in the network model using optimization algorithms. This mechanism allows the system's analytical model to continuously approximate the operational patterns of real urban systems over time and with data accumulation, thereby improving the accuracy of long-term predictions and making the generated update project recommendations more adaptable and reliable. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the overall workflow of the method of the present invention.

[0019] Figure 2 This is a flowchart of the state coding and network modeling process of the present invention.

[0020] Figure 3 A decision flowchart is generated for the solution of this invention. Detailed Implementation

[0021] The technical solutions of 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.

[0022] Example 1 As attached Figures 1 to 3 This embodiment presents an AI-based spatiotemporal multivariate information acquisition and analysis system and method for urban renewal project generation. The method is based on the understanding that a city is a complex, dynamically evolving spatial network system with nonlinear interactions between its internal units. Traditional evaluation methods based on static indicator superposition struggle to reveal these dynamic interactions and cannot predict the global and long-term spatial evolution that local interventions may trigger. The purpose of this method is to construct a data-driven urban spatial dynamics model, shifting urban renewal decision-making from "empirical judgment" based on current status evaluation to "effect prediction" based on network simulation and deduction, thereby enabling the quantitative comparison and optimization of renewal intervention strategies.

[0023] This method systematically integrates spatiotemporal big data processing, unsupervised machine learning, causal inference, complex network dynamics, and multi-objective decision analysis. Its complete process begins with the collection and standardization of multi-source heterogeneous urban data, followed by the abstraction of urban spatial states through unsupervised learning, and the construction of a spatial influence network based on causal analysis of the state sequences. On this network model, update intervention operations are defined and their cascading diffusion effects are simulated, the long-term global benefits and costs of different intervention schemes are quantitatively evaluated, and finally, recommended urban renewal project schemes are automatically generated. This process constitutes a closed loop from data perception to intelligent decision-making.

[0024] Furthermore, the method begins with the systematic collection and structured processing of multi-dimensional spatiotemporal information about the target area. Specifically, it involves accessing urban planning databases, geographic information systems, Internet of Things (IoT) sensor networks, and social media platforms with geographic location information via application programming interfaces (APIs) to collect historical data covering dimensions such as building entities, environmental perception, and pedestrian activity. This data must be aligned and normalized in terms of temporal scale and spatial reference.

[0025] For example, building data stored as polygon vectors, pedestrian trajectory data stored as point sequences, and public opinion data existing as text streams can be uniformly aggregated into regular geographic grid cells, with the grid size set according to the required analytical precision. or And generate a multidimensional feature sequence for each grid at a uniform time frequency (e.g., monthly). The dimensionality of the feature sequence... The value is typically between 10 and 30, depending on the available data source. The standardized spatiotemporal feature sequence produced in this step is the foundational data carrier for all subsequent machine learning and model building, and its quality directly determines the reliability of the analysis results.

[0026] Furthermore, deep feature learning and state abstraction are performed on the standardized feature sequences described above. This step employs an attention-based temporal encoder model, such as the Transformer or Long Short-Term Memory network, to encode the long-term feature sequences of each geographic grid. The encoder aims to capture the temporal patterns and dependencies of each grid's own development, such as identifying functional mutation points caused by the opening of major infrastructure.

[0027] The model outputs a low-dimensional, dense temporal embedding vector, which is a general representation of the historical development trajectory of the grid. Typically much smaller than the original feature dimension. ,For example .

[0028] Subsequently, a clustering algorithm, such as KMeans, was used to perform unsupervised clustering of the embedding vectors generated by all grids at all time points. The number of clusters... It can be determined based on indicators such as the profile coefficient, usually in Between 15 and 15. The clustering process automatically discovers a limited number of representative urban spatial development patterns from the data, and each cluster center is defined as an "urban spatial state prototype".

[0029] Finally, by calculating cosine similarity, each grid time point is mapped to the best-matching state prototype, thereby obtaining its corresponding discrete "urban spatial state code". ,in This process transforms high-dimensional, continuous, and noisy raw data into a set of discrete, semantically interpretable state label sequences, enabling a computable, dimensionality-reduced description of the operational modes of complex urban systems.

[0030] Furthermore, based on the obtained time series of urban spatial state codes from all geographic grids, a network dynamics model is constructed to characterize the mutual influence relationships between spatial units.

[0031] The key to this step is to objectively infer causal relationships from the data, rather than pre-determining connections. For any two grid cells... and Extract its state coding time series and The Granger causality test was then applied for analysis.

[0032] Granger causality test passes the test sequence Can historical values ​​significantly improve the performance of sequences? The predictive power of future values ​​is used to determine whether the former is a statistically significant cause of the latter. If the test results show that the significance level is within a certain range... The following exists from arrive Significant Granger causality is then established in the model from node Pointing to node The directed edges. Edge weights. The strength of the effect is determined by normalizing the test statistic (such as the F-value). After traversing all grid pairs, a directed weighted complex network is obtained. ,in It is a set of nodes (mesh). It is a set of directed edges. It is a set of weights.

[0033] The network's topology and weight distribution are learned entirely from historical data. It quantitatively reveals the implicit functional connections, radiation effects, and dependencies within the city, such as the impact of commercial centers on surrounding residential areas and the guidance of activity distribution by transportation hubs, providing a realistic path map for simulating the transmission of intervention measures.

[0034] Furthermore, update interventions are defined on the constructed network model, and their long-term effects are simulated. An update intervention is defined as forcibly setting the current state of one or more specific nodes (corresponding to the target geographic grid) in the network to a preset "anchored state" representing the ideal development goal. , .

[0035] To evaluate a candidate intervention (i.e., a grid set to be updated) To determine the potential of the intervention, the chain reaction triggered by its implementation needs to be simulated within a network. The simulation process is based on iterative calculations at discrete time steps. At the initial time... For all nodes Set its simulation state like And its target state after intervention is ,otherwise .

[0036] At each time step For each nodes Calculate its value from all states. The total impact input value received by the upstream neighbor nodes is calculated using the following formula: ; in, It is a node The set of all upstream neighbors, It is the connection weight. Neighboring nodes exist Simulated state at time step. This formula simulates the cumulative effect of positive influences along network connections.

[0037] node Does the state at the next time step change to...? This is determined by a probability transition function. This function typically takes the form of a sigmoid function: ; in, Let the state transition probability be... For sensitivity parameters ( ), Activation threshold ( This design reflects the uncertainty of influence propagation in the real world: only when the received influence input is strong enough (exceeding a threshold) will it be effective. A node has a higher probability of undergoing a state transition only when [the node is in a certain state]. This is achieved through multiple iterations (total number of iterations is [number]). ,For example The anchored state starts from the intervention origin and spreads according to the network structure and probability rules. After the simulation, the final state in the statistical network is... The proportion of nodes is the "long-term global optimization effect value" of this intervention. , It quantifies the potential ripple effect of the intervention strategy.

[0038] Furthermore, a comprehensive effectiveness evaluation of candidate intervention programs should be conducted. Decision-making must balance benefits and costs. Therefore, obtaining the long-term global optimal effect value is crucial. Then, it needs to be compared with the estimated cost of implementing the intervention. Combined. Cost The effectiveness can be comprehensively estimated based on factors such as the area involved, the type of renovation, the difficulty of demolition, and the investment in infrastructure construction, and then normalized to a positive real number. The effectiveness assessment is calculated using a fusion function: ; in, To update the intervention efficacy evaluation values, and This is an adjustable exponential parameter, typically ranging from [value range missing]. , This function form allows policymakers to adjust their focus based on policy preferences: if emphasizing the multiplier effect, the function can be increased. If cost control is emphasized, it can be increased. By calculating all candidate solutions By valuing and sorting these values, we can objectively and quantitatively identify the set of optimal or suboptimal intervention schemes that maximize overall benefits per unit cost, thereby achieving precise allocation of limited public resources.

[0039] Furthermore, based on the effectiveness evaluation results, executable urban renewal project plans are automatically generated. First, several candidate intervention plans with high effectiveness evaluation values ​​(E) are selected, and their corresponding geographic spatial ranges are merged and their boundaries are regularized to form a spatially coherent "key intervention area".

[0040] Next, for each geographic grid within the key intervention area, the system compares its original detailed feature data with its "anchored status". By comparing and analyzing the ideal feature templates, specific gap dimensions are identified. Based on a pre-built planning strategy knowledge base or rule engine, targeted update measures suggestions are automatically generated, such as "increasing public green space," "renovating old housing," and "implementing cultural facilities," and these suggestions are linked to specific spatial locations.

[0041] Finally, the system integrates the spatial boundaries of key intervention areas, the list of update measures for each plot, the priority of investment estimates, and the simulated radiation impact range map to generate a structured project proposal report. This report can be directly used for planning approval, project initiation, and subsequent design, realizing the automatic conversion of data analysis conclusions into planning management language.

[0042] Based on the foregoing detailed description of the technical solution, the complete implementation process of this invention can be systematically summarized into the following six core steps. This process is data-driven, model-based, and ultimately outputs the generation of an executable update solution.

[0043] S100: Multi-source spatiotemporal data acquisition and standardization processing The system first automatically accesses and collects multi-source heterogeneous data from the target area within a set historical time range (e.g., T=5 years). This data includes at least: (1) Building entity data (such as outline, height, and age); (2) Human living environment perception data (such as noise and air quality sensor readings, or sentiment index obtained from social media texts through NLP analysis). (3) Dynamic pedestrian activity data (such as mobile phone signaling density and public transportation card swipe records).

[0044] Subsequently, all the above data are unified into a spatial reference frame and aggregated onto a regular geographic grid (e.g., with a side length of L=100 meters), forming a multidimensional feature sequence with the grid as the basic spatial unit and sampled at a fixed time frequency (e.g., monthly). The feature vector of each unit at time t is denoted as , where d is the feature dimension. This step outputs a structured, standardized spatiotemporal feature dataset.

[0045] S200: Unsupervised coding of urban spatial status The feature dataset output by S100 is fed into a pre-trained unsupervised temporal coding model (e.g., a Transformer-based encoder). This model processes long sequences of each basic spatial unit. Mapped to a low-dimensional temporal embedding vector (usually) ).

[0046] Subsequently, a clustering algorithm (such as KMeans++) was used to cluster the set of embedding vectors generated by all units at all time points, resulting in K cluster centers. Each center is defined as a prototype of urban spatial state. Finally, by calculating similarity, each unit time point is assigned to the most similar prototype to obtain its corresponding discrete state code. The entire process discretizes the continuous, high-dimensional original feature space into a finite, semantically meaningful state space.

[0047] S300: Construction of Causal Inference for Spatial Influence Networks State-encoded time series of all basic spatial units obtained based on S200 Perform pairwise Granger causality tests. For any two units i and j, if the test rejects the null hypothesis "sj is not a Granger cause of si" at a significance level α (e.g., α = 0.05), then create a directed edge in the network from node j to node i. The initial weight of the edge... It is obtained by normalizing the test statistic (such as the F-value). After traversing all unit pairs, a directed weighted complex network model is constructed. Where V is the set of nodes (units), E is the set of directed edges, and W is the set of weights. This network objectively reveals the statistical causal relationships of state evolution among spatial units from the data.

[0048] S400: Dynamic Simulation and Effectiveness Equivalence of Updated Interventions First, define the update intervention in the network model G: select a set of candidate units. The state is forcibly set to the preset anchoring state. ( The simulation process is performed iteratively in discrete time steps: Initialization: for ,like Then let ;otherwise .

[0049] Iteration (for steps t=0 to N1): For each For node v, calculate the total impact input it receives. .in It is the set of all upstream neighbors of v.

[0050] State transition: Calculate the probability that state v changes to state 1 at time t+1. ,in k and θ are preset parameters. The probability is determined accordingly. Is it 1?

[0051] After the simulation, the long-term global optimization effect value is calculated. At the same time, based on The physical properties are used to estimate the intervention cost C. Finally, the updated intervention efficacy evaluation value of the candidate program is calculated. , where α and β are adjustment parameters.

[0052] S500: Multi-option comparison and key intervention space identification Repeat step S400 for multiple (e.g., M) candidate update unit sets proposed by the planner. Parallel simulation and performance calculation are performed to obtain the corresponding performance values. The system sorts all candidate schemes in descending order based on the E value. The top R (e.g., R=3) schemes are selected, and their corresponding spatial unit sets are geospatially merged and their boundaries are normalized, outputting a coherent "key intervention area" spatial polygon.

[0053] S600: Generation of Structured Update Project Proposals For the key intervention areas identified by S500, the system conducts a refined analysis: for each basic spatial unit within the area, its original characteristics are traced back. , and anchoring state By comparing the corresponding ideal feature patterns, the main gap dimensions are identified. Combined with a pre-built planning measures knowledge base, targeted update measures suggestions are automatically generated (such as "increasing the public green space ratio to X%" or "adjusting the building function to mixed use"). Finally, the system automatically integrates the spatial boundaries of key intervention areas, the list of update measures for each unit, the investment estimation framework, and the state diffusion expectation map generated by simulation, packaging them into a structured, richly illustrated urban renewal project plan document, completing the full output from data analysis to decision support.

[0054] As can be seen from the detailed description of the above specific implementation methods, this method, by introducing the analytical framework of "state-coded causal network dynamic simulation", realizes the transformation of urban renewal intervention strategies from qualitative judgment to quantitative simulation, and from local consideration to global evaluation.

[0055] Finally, the following points should be noted: First, in the description of this application, it should be noted that, unless otherwise specified and limited, the terms "installation", "connection", and "linkage" should be interpreted broadly, and can be mechanical or electrical connections, or internal connections between two components, or direct connections. "Up", "down", "left", "right", etc. are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may change. Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for collecting and analyzing spatiotemporal multivariate information using AI in the field of urban renewal project generation, characterized in that, Includes the following steps: S1: Collect spatiotemporal multi-dimensional information of the target area, including building information, environmental perception information and pedestrian activity information; S2: Standardize the spatiotemporal multivariate information to generate a spatiotemporal feature sequence indexed by basic spatial units; S3: Based on the spatiotemporal feature sequence, an unsupervised machine learning algorithm is used to generate urban spatial status codes for each basic spatial unit at different time points; S4: Based on the time series of the urban spatial state code, construct a directed weighted network model representing the dynamics of cross-spatial state propagation; S5: In the directed weighted network model, an update intervention operation is defined to set the state of a specified basic spatial unit to a preset anchor state. S6: Simulate the cascading diffusion process of the anchored state in the network model after the update intervention operation is performed on the candidate update intervention unit set, and calculate the resulting long-term global optimization effect value; S7: By combining the long-term global optimization effect value and the estimated cost, the updated intervention efficacy evaluation value is calculated; S8: Generate an urban renewal project plan based on the aforementioned evaluation value of the effectiveness of the intervention.

2. The AI-based spatiotemporal multi-dimensional information collection and analysis method for urban renewal project generation as described in claim 1, characterized in that, Step S3 specifically includes: S31: Input the spatiotemporal feature sequence of each basic spatial unit into the temporal encoder to obtain the corresponding temporal embedding vector; S32: Cluster the set consisting of all temporal embedding vectors, and define each cluster center as a prototype of urban spatial state; S33: Assign a corresponding urban spatial state code to each temporal embedding vector based on its similarity to the prototype of each urban spatial state.

3. The AI-based spatiotemporal multi-dimensional information collection and analysis method for urban renewal project generation as described in claim 1, characterized in that, The construction of the directed weighted network model in step S4 is specifically as follows: S41: For any two basic spatial units, extract their urban spatial status coding time series; S42: Analyze the causal relationship between the two time series. If there is a significant influence from the first unit to the second unit, then establish a directed edge in the network model from the node of the first unit to the node of the second unit. S43: Quantify the strength of the causal relationship into the initial weight of the corresponding directed edge; S44: Based on the influence relationships and initial weights between all basic spatial unit pairs, construct a complete network model.

4. The AI-based spatiotemporal multi-dimensional information collection and analysis method for urban renewal project generation as described in claim 1, characterized in that, Step S6 specifically includes: S61: Initialize the network model, set the state values ​​of all corresponding nodes in the candidate update intervention unit set to the first value, representing the anchoring state, and set the state values ​​of the remaining nodes to the second value. S62: In multiple consecutive time steps, for each node whose state value is the second value, calculate the network influence input value it receives based on the current state values ​​of all its upstream neighbor nodes and the weights of the corresponding directed edges. S63: According to the preset state transition rules, the network influence input value determines whether the state of the node in the next time step is updated to the first value; S64: After the simulation ends, the proportion of nodes in the network whose state value is the first value is counted, and this proportion is used as the long-term global optimization effect value.

5. The AI-based spatiotemporal multi-dimensional information collection and analysis method for urban renewal project generation as described in claim 4, characterized in that, The state transition rule is as follows: when the network influence input value received by a node exceeds a preset threshold, the probability that the node's state will be updated to the first value in the next time step increases significantly.

6. The AI-based spatiotemporal multi-dimensional information collection and analysis method for urban renewal project generation as described in claim 1, characterized in that, In step S7, the formula for calculating the updated intervention effectiveness evaluation value is the ratio of the long-term global optimization effect value raised to the power of α to the estimated cost raised to the power of β, where α and β are preset positive real number parameters.

7. The AI-based spatiotemporal multi-dimensional information collection and analysis method for urban renewal project generation as described in claim 1, characterized in that, Step S8 specifically includes: S81: Sort the candidate set according to the updated intervention efficacy evaluation value; S82: Select at least one candidate set that ranks highly and determine the geographic spatial range it covers as the key intervention space; S83: Analyze the spatiotemporal characteristics of each basic spatial unit within the key intervention space and generate corresponding update measure suggestions; S84: Output an urban renewal project plan that includes the key intervention spatial boundaries, renewal measure recommendations, and expected impact range.

8. An AI spatiotemporal multivariate information acquisition and analysis system for implementing the method of any one of claims 1 to 7, characterized in that, The system includes: The data acquisition and preprocessing module is used to perform steps S1 and S2; The state coding and network modeling module is used to perform steps S3 and S4; An intervention simulation and efficacy evaluation module is used to perform steps S5, S6, and S7. The scheme generation module is used to execute step S8.

9. The AI ​​spatiotemporal multi-source information acquisition and analysis system according to claim 8, characterized in that, The state coding and network modeling module includes: Temporal coding unit, used to encode spatiotemporal feature sequences into temporal embedding vectors; State clustering unit, used to cluster temporal embedding vectors to generate urban spatial state codes; The network inference unit is used to analyze the causal relationship between state-encoded sequences to construct the directed weighted network model.

10. The AI ​​spatiotemporal multi-source information acquisition and analysis system according to claim 8, characterized in that, The intervention simulation and effectiveness evaluation module includes an online model learning unit, which is configured to dynamically adjust the weights of the directed edges in the directed weighted network model based on the differences between the subsequent impact data of actual urban renewal projects and the simulation prediction results.

Images