Cloud-native spatio-temporal fusion analysis method based on spatio-temporal asset catalog

Abstract

The spatio-temporal asset catalog-based cloud-native spatio-temporal fusion analysis method of the present application belongs to the technical field of data indexing and fusion analysis, and the steps include: S1, obtaining the core elements of the source asset and constructing a spatio-temporal influence potential model to obtain the spatio-temporal influence potential; S2, based on the spatio-temporal influence potential, calculating the influence gradient and comparing the module length of the influence gradient with the preset threshold to generate a coupling relationship evaluation result; S3, in response to the coupling relationship evaluation result, using a state evolution model to perform evolution calculation on the current core indicators of the target asset to predict and generate future core indicators; S4, based on the future core indicators, generating an intervention strategy, and after executing the intervention strategy, using the prediction error to adaptively calibrate the parameters in the state evolution model, the present application realizes the cognitive leap from fuzzy correlation to quantitative causality.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data indexing and fusion analysis, in particular to a cloud-native spatio-temporal fusion analysis method based on a spatio-temporal asset catalog. BACKGROUND

[0002] In the operation and management of urban complex systems, traditional data analysis and decision-making methods mainly rely on isolated and static correlation analysis of historical data. Such methods usually treat spatio-temporal data as independent and static information points, and seek to find the fuzzy correlation between different events or assets through post-event analysis. When intervention decisions are needed, they often rely on human experience to judge and dispose, forming a passive emergency response mode.

[0003] The above status and deficiencies mainly result from the limitations of analysis models and technical links:

[0004] Limitations of analysis models: Traditional techniques cannot quantify the potential impact of events, and can only draw vague conclusions such as "event A is related to asset B", without accurately describing the strength and direction of the impact. In addition, their prediction models are mostly simple linear extrapolations, which cannot simulate the saturation effect or bottleneck constraints that exist in the real world, such as parking capacity, leading to unrealistic infinite growth predictions under strong influence.

[0005] Limitations of technical links: Existing technologies lack an automated closed-loop system from data perception, causal reasoning, state prediction to decision intervention; analysis models are usually fixed and static, and cannot be self-corrected and evolved according to real-world feedback, resulting in a lack of continuous improvement in their cognitive and predictive accuracy.

[0006] As a result, when facing sudden events in the city, managers cannot transform isolated data into predictable and intervenable dynamic causal relationships; this leads to insufficient efficiency and foresight in emergency response, and the decision-making mode relying on human experience is difficult to quickly generate scientific and cost-optimal intervention strategies in complex situations.

[0007] The above information disclosed in the background section is only used to strengthen the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0008] The purpose of the present application is to provide a cloud-native spatio-temporal fusion analysis method based on a spatio-temporal asset catalog to solve the problems raised in the background section.

[0009] The technical solution of the present application is as follows, the specific steps include:

[0010] S1, acquire the core elements of the source asset, and construct a spatio-temporal influence potential model to obtain the spatio-temporal influence potential;

[0011] S2, based on the spatio-temporal influence potential, calculate the influence gradient, and compare the module length of the influence gradient with the preset threshold to generate the coupling relationship evaluation result;

[0012] S3, in response to the coupling relationship evaluation result, use the state evolution model to perform evolution calculation on the current core indicators of the target asset to predict and generate future core indicators;

[0013] S4, based on the future core indicators, generate an intervention strategy, and after executing the intervention strategy, use the prediction error to adaptively calibrate the parameters in the state evolution model.

[0014] Preferably, S1 includes:

[0015] Acquire the core elements of the source asset from the spatio-temporal asset catalog, including spatio-temporal coordinates and core business indicators;

[0016] Normalize the core business indicators to obtain the event source intensity; combine the event source intensity, the preset time decay coefficient, and the preset regularization length to construct the spatio-temporal influence potential model; and use the spatio-temporal influence potential model to obtain the spatio-temporal influence potential.

[0017] Preferably, the initial calibration method of the regularization length includes calibrating the core action range or equivalent radius of the asset as the regularization length according to the type and physical properties of the asset; and the initial calibration method of the time decay coefficient includes extracting historical similar event influence decay data and using an exponential decay model to fit the data to obtain the time decay coefficient.

[0018] Preferably, S2 includes:

[0019] Calculate the negative gradient of the spatio-temporal influence potential to obtain the influence gradient; collect the gradient module length of events in history that did not cause the core indicators of the target asset to change, and calibrate the high percentile point as the monitoring response threshold; collect the gradient module length of events in history that caused the core indicators of the target asset to mutate, and calibrate the low percentile point as the critical early warning threshold.

[0020] Preferably, the step of generating the coupling relationship evaluation result includes:

[0021] When the module length of the influence gradient is less than or equal to the monitoring response threshold, determine that the coupling relationship evaluation result is weak coupling;

[0022] When the module length of the influence gradient is greater than the monitoring response threshold and less than or equal to the critical early warning threshold, determine that the coupling relationship evaluation result is strong coupling;

[0023] When the length of the influence gradient is greater than the critical early warning threshold, the coupling relationship evaluation result is determined as critical coupling.

[0024] Preferably, S3 comprises:

[0025] For the asset pair with the coupling relationship evaluation result as strong coupling or critical coupling, a state evolution model is used, and based on the influence gradient, the sensitivity vector of the target asset and the capacity parameter, the current core indicator of the target asset is evolved to predict the future core indicator.

[0026] Preferably, the capacity parameter is calibrated by obtaining the physical properties or historical data extreme value of the target asset; the calibration method of the sensitivity vector includes: based on the historical data of the functional layout of the target asset, a weighted average calculation is performed to determine the direction; and by performing linear regression analysis on the historical influence gradient and the core indicator change amount, the length is determined.

[0027] Preferably, S4 comprises:

[0028] The future core indicator is compared with the preset business alarm threshold to identify the risk asset; the system instability index is calculated in combination with the business weight of the risk asset, the influence gradient and the monitoring response threshold; based on the system instability index and the preset business safety threshold, and taking the minimum total intervention cost as the principle, an intervention strategy is solved and generated;

[0029] The business weight is quantitatively calibrated by the analytic hierarchy process or the contribution degree analysis based on historical data.

[0030] Preferably, the step of self-adaptive calibration comprises:

[0031] After the execution of the intervention strategy, the actual core indicator of the target asset is tracked; the prediction error between the actual core indicator and the future core indicator is calculated; and the sensitivity vector in the state evolution model is modified by using the prediction error.

[0032] The cloud-native spatio-temporal fusion analysis method based on the spatio-temporal asset catalog is provided by improving the present application, compared with the prior art, has the following improvements and advantages:

[0033] 1. The present application realizes the cognitive leap from fuzzy correlation to quantitative causality; the core is to build a unified spatio-temporal influence potential model, which solves the singularity problem existing at the source of the traditional physical field model by introducing the regularization length, ensures the robustness of the calculation, and enables the system to automatically classify complex interactions into weak coupling, strong coupling and critical coupling, providing clear and quantifiable trigger conditions for subsequent enabling different levels of analysis and intervention.

[0034] 2.The application transcends the simple linear extrapolation prediction of the prior art, introduces a state evolution model that is more in line with the law of nonlinear dynamics, ensures that the prediction logic closely matches the real response characteristics of the asset, and the system automatically solves and generates the optimal intervention strategy with the principle of minimizing the total intervention cost, which marks the transition from passive and artificial emergency response mode to active and automatic intelligent operation mode; the closed-loop feedback enables the entire analysis system to have learning ability, continuously improve its cognition and prediction accuracy of complex spatio-temporal systems, and distinguish it from all static and fixed analysis models, becoming an intelligent decision execution body that can evolve with the environment. BRIEF DESCRIPTION OF DRAWINGS

[0035] The application will be further explained in conjunction with the accompanying drawings and embodiments:

[0036] Figure 1 is a flowchart of the method of the application. DETAILED DESCRIPTION

[0037] To make the purpose, technical scheme and advantages of the application clearer, the application will be further described in detail below in conjunction with specific embodiments.

[0038] Embodiment 1:

[0039] Please refer to Figure 1 The application provides a cloud-native spatio-temporal fusion analysis method based on a spatio-temporal asset catalog, and the specific steps include:

[0040] S1, obtaining the core elements of the source asset and constructing a spatio-temporal influence potential model to solve and obtain the spatio-temporal influence potential;

[0041] S2, based on the spatio-temporal influence potential, calculating the influence gradient and comparing the module length of the influence gradient with the preset threshold to generate a coupling relationship evaluation result;

[0042] S3, in response to the coupling relationship evaluation result, using a state evolution model to perform evolution calculation on the current core indicators of the target asset to predict and generate future core indicators;

[0043] S4, based on the future core indicators, generating an intervention strategy, and after executing the intervention strategy, using the prediction error to adaptively calibrate the parameters in the state evolution model

[0044] The cloud-native spatio-temporal fusion analysis method based on the spatio-temporal asset catalog disclosed in this embodiment is constructed as a closed-loop system containing four core steps to ensure the automation and intelligence of the data perception to decision execution;

[0045] The flow of the method starts with obtaining the core elements of the source asset and constructing a spatio-temporal influence potential model; the core elements are structured data representing the key characteristics of a spatio-temporal asset, and their role is to provide basic input for subsequent quantitative analysis, and the data comes from a pre-constructed spatio-temporal asset directory;

[0046] In a preferred embodiment, the spatio-temporal asset directory can be constructed as a relational database or a graph database; the core data table or node attribute should at least include: asset unique identifier, asset type, such as traffic incident, commercial facility, sensor, four-dimensional spatio-temporal coordinates, , core business indicators, such as congestion index, passenger flow, and related physical or business attributes, such as road grade, parking capacity ; The data of the directory can be imported in real time from the city's Internet of Things sensor network, public transportation information platform, weather service, etc. Data sources, supplemented by manual entry and regular maintenance of static assets such as building information to ensure the accuracy and timeliness of the data;

[0047] In this scenario, the system extracts the traffic accident, the spatio-temporal coordinates of the source asset A and its core business indicators, such as congestion index, from the directory; based on these elements, a spatio-temporal influence potential model describing its influence range and intensity is constructed; the purpose of this model is to transform isolated events into measurable, dynamically changing influence fields in the four-dimensional spatio-temporal continuum, thereby obtaining a quantitative spatio-temporal influence potential;

[0048] Based on the spatio-temporal influence potential obtained in the previous step, the system calculates the influence gradient and compares the length of the influence gradient with the preset threshold to generate the coupling relationship evaluation result; the influence gradient, as a measure of the rate and direction of change of the spatio-temporal influence potential in space, serves to accurately quantify the strength and directionality of the influence exerted by the source asset on the target asset; by performing vector operations on the spatio-temporal influence potential, an influence gradient pointing to the shopping center B can be obtained; the preset threshold referred to here is a set of calibration values used to evaluate the strength of the influence, and the internal logic of its establishment lies in objectively distinguishing the boundaries between background noise influence, significant effective influence and critical dangerous influence through historical data statistics; by comparing the length of the calculated influence gradient with these thresholds, the system can objectively determine the coupling relationship strength of the traffic accident on the shopping center B, such as strong coupling or critical coupling;

[0049] In response to the coupling relationship evaluation result, the system utilizes the state evolution model to perform evolution calculation on the current core indicators of the target asset to generate future core indicators; when the evaluation result indicates that there is a significant impact, a state evolution model is activated; the model aims to simulate the dynamic changes of the state of the target asset over time under external influences; rather than simple linear accumulation, it comprehensively considers the influence intensity, sensitive characteristics of the target asset, and its own carrying limit; in this scenario, the model receives the influence gradient as input to deduce the current passenger flow, parking lot occupancy rate, and other core indicators of shopping center B, and further predicts the future core indicators such as a 60% decrease in main entrance passenger flow within the next 15 minutes;

[0050] Based on the predicted future core indicators, the system generates intervention strategies, and after the strategy is executed, the prediction error is used to adaptively calibrate the parameters in the state evolution model; the so-called intervention strategy is a set of optimized operation instructions aimed at avoiding risks or taking advantage of opportunities; the system automatically solves an optimal intervention combination based on the disruptive prediction generated in the previous step, such as parking lot congestion rate exceeding the alarm line; after the strategy is executed, the system will continuously track the actual passenger flow changes of shopping center B and compare the actual value with the predicted value of the model to produce a prediction error; this error is used as a feedback signal to drive adaptive calibration of the internal parameters of the state evolution model, forming a complete learning closed loop;

[0051] Through the above steps, the method constructs a complete technical link from spatio-temporal influence quantification to coupling relationship evaluation, to future state prediction and closed-loop adaptive intervention; it transforms originally isolated and static spatio-temporal data into predictable and intervenable dynamic causal relationships, achieving automated, precise, and intelligent operation and management of complex urban system events, greatly improving the efficiency and foresight of emergency response.

[0052] Embodiment 2

[0053] S1 includes:

[0054] Collecting the core elements of the source asset from the spatio-temporal asset catalog, the core elements including spatio-temporal coordinates and core business indicators;

[0055] Normalizing the core business indicators to obtain the event source intensity; combining the event source intensity, a preset time decay coefficient, and a preset regularization length, a spatio-temporal influence potential model is constructed; using the spatio-temporal influence potential model, the spatio-temporal influence potential is solved.

[0056] The construction and solution process of the spatio-temporal influence potential model are described;

[0057] The core elements of the source asset are collected from the spatio-temporal asset catalog, including spatio-temporal coordinates and core business indicators; in this scenario, the system accurately extracts the spatio-temporal coordinates of accident A from the spatio-temporal asset catalog database and its core business indicators ; the indicators are quantitative values that best represent the asset state, such as the congestion index here, which serves as the original input for impact strength;

[0058] The core business indicators are normalized to obtain event source strength; the spatio-temporal influence potential model is constructed by combining the event source strength, the preset time decay coefficient, and the preset regularization length; the spatio-temporal influence potential is obtained by using the spatio-temporal influence potential model; the event source strength is a dimensionless parameter obtained after normalization of the core business indicators , which allows the influence of different types of assets to be compared in a unified framework, ensuring the universality of the model;

[0059] The normalization process preferably uses the min-max normalization method; for a certain type of asset, the event source strength can be calculated by the following formula:

[0060]

[0061] wherein is the actual value of the currently collected core business indicator, and and are the maximum and minimum values of the core business indicators recorded in the historical database for this type of asset; by this method, business indicators of different dimensions can be uniformly mapped to the interval to , thereby obtaining the standardized event source strength ;

[0062] To construct the model, two preset core parameters are introduced: the time decay coefficient and the regularization length ; which describes the natural weakening of event influence over time; and which solves the problem of numerical singularity of the physical model near the source point due to the distance approaching zero; based on the above parameters, the spatio-temporal influence potential model is constructed, and the calculation is given by the following formula:

[0063]

[0064] wherein is the spatio-temporal influence potential, and the input parameters are derived from: is the core business indicator normalized calculated event source intensity, is the spatial distance between the source asset and the target asset, is the time difference between the two, is a preset time decay coefficient, is a preset regularization length; is the base of the natural logarithm;

[0065] By substituting the value of accident A , and the spatial distance and the time difference between it and shopping center B, an accurate value can be obtained, which quantifies the influence potential size of shopping center B from accident A at the current time;

[0066] This embodiment defines the space-time influence potential model by introducing normalized event source intensity , time decay coefficient and regularization length ; in particular, the introduction of the regularization length solves the singularity problem of the traditional physical model in the near distance calculation, ensuring the good calculation robustness of the model at any spatial scale; this makes the quantification process of the influence field not only unified and standard, but also more stable and reliable in calculation;

[0067] Compared with the existing technology which usually stays in isolated and static correlation analysis of historical data, the present application realizes the cognitive leap from fuzzy correlation to quantitative causality; the core lies in constructing a unified space-time influence potential model; the model solves the singularity problem of the traditional physical field model at the source point by introducing the regularization length , ensuring the robustness of the calculation; the mathematical expression is:

[0068]

[0069] wherein, : space-time influence potential, : event source intensity, : spatial distance, : time difference, : time decay coefficient, : regularization length; the physical meaning of this formula is that it abstracts the core business indicators of any source asset into a potential field source that continuously decays with time and space, and the potential generated by it provides a unified and singularity-free mathematical basis for subsequent analysis; by substituting and Initial calibration based on physical properties and historical data eliminates the black box of model parameter sources, making the construction of the influence field source objective and interpretable.

[0070] Example 3

[0071] The initial calibration method for the regularization length includes defining the core scope or equivalent radius of the asset as the regularization length based on the asset's type and physical attributes; the initial calibration method for the time decay coefficient includes extracting historical data on the decay of the influence of similar events and fitting the data using an exponential decay model to obtain the time decay coefficient.

[0072] To ensure the objectivity and interpretability of the model parameters, the initial calibration method for two key preset parameters in the model is further disclosed.

[0073] The initial method for determining the regularization length is described as follows: based on the asset's type and physical attributes, the core effective range or equivalent radius of the asset is defined as the regularization length. Regularization length Physically, it represents the core area of ​​an asset's influence, i.e., a meaningful minimum interaction distance; the calibration principle is to anchor abstract mathematical parameters to the physical reality of the asset; for example, for point-in-the-moment assets like traffic accidents, the core area of ​​physical influence is relatively small. It can be objectively defined as a small value, such as 5 meters; for area assets like large shopping malls, It can then be calibrated based on the equivalent radius calculated from its building footprint, such as 100 meters;

[0074] The initial calibration method for the time decay coefficient is described as follows: extract historical data on the decay of the influence of similar events, and fit the data using an exponential decay model to obtain the time decay coefficient. The working principle is to establish a mathematical model for the natural laws governing the fading of the influence of different types of events over time; for a specific type of event, multiple samples are extracted from a historical database; for each sample, data is collected at different time points after the event. Influence indicators, such as the actual observed congestion index ; these observation data points Fit to exponential decay model In; among them, Initial influence metrics; The decay rate, after being solved through fitting, is the time decay coefficient. : Subscript, used to identify different data collection time points, such as The attenuation rate is obtained by optimization algorithms such as the least squares method. The initial value of the type of event to be calibrated; usually, the impact of traffic accidents decays quickly, and the value is large; while the impact of environmental pollution events decays slowly, and the value is small. The initial value of the type of event to be calibrated; usually, the impact of traffic accidents decays quickly, and the value is large; while the impact of environmental pollution events decays slowly, and the value is small. The initial value of the type of event to be calibrated; usually, the impact of traffic accidents decays quickly, and the value is large; while the impact of environmental pollution events decays slowly, and the value is small. The initial value of the type of event to be calibrated; usually, the impact of traffic accidents decays quickly, and the value is large; while the impact of environmental pollution events decays slowly, and the value is small.

[0075] By providing an objective calibration method based on physical properties and historical data statistics for and , the embodiment ensures the transparency and reproducibility of the model's basic parameters; this method closely combines the construction of the model with the physical laws and historical experience of the real world, significantly improving the accuracy, interpretability, and credibility of the model in practical applications, and avoiding the problem of model uncertainty caused by unknown parameter sources.

[0076] Embodiment 4

[0077] S2 comprises:

[0078] The negative gradient of the spatiotemporal impact potential is calculated to obtain the impact gradient; the gradient module length generated by the events in history that did not cause changes in the core indicators of the target asset is collected, and the high percentile point is calibrated as the monitoring response threshold; the gradient module length generated by the events in history that caused mutations in the core indicators of the target asset is collected, and the low percentile point is calibrated as the critical early warning threshold;

[0079] The step of generating the coupling relationship evaluation result comprises:

[0080] When the module length of the impact gradient is less than or equal to the monitoring response threshold, it is determined that the coupling relationship evaluation result is weak coupling;

[0081] When the module length of the impact gradient is greater than the monitoring response threshold and less than or equal to the critical early warning threshold, it is determined that the coupling relationship evaluation result is strong coupling;

[0082] When the module length of the impact gradient is greater than the critical early warning threshold, it is determined that the coupling relationship evaluation result is critical coupling.

[0083] The impact gradient calculation and coupling relationship evaluation process are specified; a complete evaluation and grading process is formed;

[0084] The process starts with calculating the negative gradient of the spatiotemporal impact potential to obtain the impact gradient; the impact gradient is a vector pointing in the direction of the fastest change of the impact potential, and the module length represents the strength of the impact; the spatiotemporal impact potential obtained in the previous step is applied to the gradient operator and the negative value is taken to solve it;

[0085]

[0086] in, To influence the gradient vector, the input for its calculation is the spatiotemporal influence potential obtained from the previous step. ;

[0087] Calculation results It precisely indicates the direction and intensity of the impact of traffic accident A on shopping mall B, for example, towards B's main vehicle entrance;

[0088] To objectively classify the intensity of impact, two core thresholds need to be defined. First, the gradient magnitudes of historical events that have not caused changes in the core indicators of the target asset are collected, and the high percentiles are defined as the monitoring response threshold; this is the monitoring response threshold. The calibration process; its principle is to identify and filter the impact of background noise; the system analyzes all historical events that have affected shopping center B, and filters out those events that have not caused meaningful changes in its core indicators such as customer traffic, for example, events with fluctuations less than the preset business sensitivity. The gradient magnitude is set to the magnitude of the effects of these invalid events. The 95th percentile ensures that only statistically significant effects are systematically considered;

[0089] Secondly, the gradient magnitudes generated by historical events that caused sudden changes in the core indicators of the target asset are collected, and low percentage points are marked as critical warning thresholds; this is the critical warning threshold. The calibration process; the principle is to identify high-risk couplings that may cause a chain reaction in the business system; the system screens out all historical events that have caused sudden changes in the core indicators of shopping mall B, such as a month-on-month change in customer traffic exceeding 30%; The gradient magnitude is set to correspond to these high-impact events. The 10th percentile; this low percentile is used to ensure the sensitivity of the warning, ensuring that even the smallest impact that has historically caused a mutation is sufficient to trigger a high-level warning.

[0090] Based on the aforementioned threshold, the step of generating the coupling relationship evaluation results is performed; the magnitude of the influence gradient is calculated. Compare with two objectively calibrated thresholds:

[0091] like The coupling relationship assessment result indicates weak coupling, meaning the impact is negligible and the system will not take any action.

[0092] like The coupling relationship assessment result is determined to be strong coupling, indicating the existence of a clear influence transmission relationship. The system will then activate the subsequent causal inference module.

[0093] like , the coupling relationship evaluation result is critical coupling, indicating that the target asset may have a state mutation, and the system will trigger high-level early warning while activating causal deduction;

[0094] The embodiment converts the abstract influence into a gradient vector that can be accurately calculated , and establishes a double threshold (threshold , ) calibration and evaluation system based on the historical data of the asset itself; this enables the system to strictly and objectively quantify the interaction strength and direction between assets, providing clear and unambiguous input conditions for subsequent activation of which level of prediction and intervention, and realizing the cognitive upgrade from fuzzy correlation to quantitative conduction relationship;

[0095] The present application further defines the influence gradient , wherein : influence gradient vector, : gradient operator, : space-time influence potential, which converts the scalar influence potential into a vector, accurately describing the strength and direction of the influence; this is in sharp contrast to the existing technology which can only draw a fuzzy conclusion that A is related to B; more valuable is that the present application establishes a double evaluation system based on monitoring response threshold and critical early warning threshold; these two thresholds are not artificially set, but are objectively calibrated through statistical analysis of the historical data of the target asset, corresponding to the minimum influence that produces a meaningful response and the critical influence that causes a state mutation; this design enables the system to automatically classify complex interactions into weak coupling, strong coupling and critical coupling, providing clear and quantifiable trigger conditions for subsequent activation of different levels of analysis and intervention.

[0096] Embodiment 5

[0097] S3 includes:

[0098] For the asset pair with a coupling relationship evaluation result of strong coupling or critical coupling, the state evolution model is used, and based on the influence gradient, the sensitivity vector and the capacity parameter of the target asset, the current core index of the target asset is evolved to predict the future core index.

[0099] The state evolution and prediction process is described;

[0100] For the asset pair with strong coupling or critical coupling evaluation result, the state evolution model is used to perform evolution calculation on the current core indicator of the target asset based on the influence gradient, the sensitivity vector of the target asset and the capacity parameter, so as to predict the future core indicator; this step is triggered only when there is significant influence between assets, so as to improve the system efficiency; the state evolution model is a nonlinear dynamic model for predicting the future state of the target asset; the model introduces two key parameters: the sensitivity vector and the capacity parameter ; The response tendency and intensity of the target asset to the influence from a specific direction are described; The physical or business upper limit that can be reached by the core indicator of the target asset is represented; based on these parameters, the core indicator of the target asset at the next time is predicted by the following nonlinear evolution formula:

[0101]

[0102] Wherein, is the predicted core indicator at the next time, and the input parameters come from: is the core indicator of the target asset collected at the current time, is the influence gradient vector calculated in the previous step, and are the pre-labeled sensitivity vector and capacity parameter of the target asset respectively, is the preset time step;

[0103] The value of the time step can be determined according to the emergency response time requirement of the business scenario or the data refresh frequency of the core indicator, for example, in the traffic congestion prediction scenario requiring fast response, it can be preferably set to 1-5 minutes; while in the commercial district planning scenario with relatively gentle passenger flow change, it can be set to 15 minutes or longer to balance the calculation resource consumption and prediction accuracy;

[0104] In this scenario, the model uses the calculated gradient , the preset and values of shopping center B to perform iterative calculation on the current passenger flow , so as to accurately predict the nonlinear phenomena such as aggravation of parking lot congestion and abnormal increase of passenger flow in the catering area;

[0105] This embodiment introduces a capacity limit ​The nonlinear state evolution model greatly improves the authenticity of the prediction; the model can scientifically simulate the saturation effect and bottleneck restriction which exist universally in the real world, and avoids the linear model which will draw the conclusion of unlimited growth under strong influence; this makes the system be able to promote from static relationship strength analysis to dynamic and more physical law-conformed causal effect prediction, and provides more accurate and reliable decision basis for risk early warning and resource allocation;

[0106] The application overcomes the simple linear extrapolation prediction of the prior art, and introduces a state evolution model which is more in line with the nonlinear dynamics law; the mathematical form is Wherein, : predicted core index, : current core index, : influence gradient, : sensitivity vector, : capacity parameter, : time step; the essence of the formula is to introduce the capacity parameter which represents the physical or business upper limit of the core index of the asset, such as the total parking space of a parking lot; this parameter enables the model to accurately predict the saturation effect or bottleneck restriction which exist universally in the real world, for example, the prediction of road congestion will not be infinitely intensified, but will enter a saturation state when reaching the maximum traffic capacity ; this is completely beyond the reach of the traditional linear model; at the same time, the sensitivity vector in the model is calibrated through regression analysis of historical data and weighted calculation of functional layout, so as to ensure that the prediction logic closely matches the real response characteristics of the asset.

[0107] Embodiment 6

[0108] The capacity parameter is calibrated by obtaining the physical properties or extreme value of historical data of the target asset; the calibration method of the sensitivity vector includes: based on the historical data of the functional layout of the target asset, the weighted average calculation is performed to determine the direction; and the linear regression analysis is performed on the historical influence gradient and the change amount of the core index to determine the module length.

[0109] In this embodiment, the initial calibration method of the two core parameters in the state evolution model, the capacity parameter and the sensitivity vector is specified, aiming to eliminate the source uncertainty of the model parameters;

[0110] The capacity parameter is calibrated by obtaining the physical properties or extreme value of historical data of the target asset; the capacity parameter represents the upper limit or lower limit of the core index of the target asset; the calibration method directly comes from the objective data of the asset, so as to ensure the physical meaning of the parameter; for example, for the shopping center B, the capacity parameter It could be the maximum number of people allowed under fire safety regulations, the total number of parking spaces in its associated parking lot, or the highest single-day passenger flow record in history;

[0111] The calibration method for the sensitivity vector is decomposed into the separate calibration of direction and magnitude;

[0112] Firstly, direction The calibration, in its underlying logic, lies in using a weighted average calculation based on historical data of the target asset's functional layout to determine the direction. This method transforms abstract attributes such as functional zoning into calculable vector directions. For example, the sensitivity direction of shopping mall B is not subjectively set, but rather derived by using a weighted average calculation of historical passenger flow data from its multiple entrances, such as the main entrance and parking lot entrances. Each entrance... This corresponds to a unit vector pointing from the outside in. Its weight The historical daily average foot traffic percentage of this entrance; the final sensitivity direction. Calculated using the following formula:

[0113]

[0114] in, The sensitivity direction vector, whose input parameters are derived from: The weights of each entry point are calculated based on historical data. This represents the unit direction vector corresponding to each entrance; Subscript: refers to the various entry points of the target asset;

[0115] Secondly, the length of the mold The calibration technical solution involves determining the magnitude by performing linear regression analysis on historical impact gradients and changes in core indicators; the magnitude quantifies the severity of the asset response; the calibration process collects historical impact-response data pairs, i.e., the gradient vector each time asset B is affected. and the resulting actual changes in core indicators According to the linear approximation of the state evolution formula The independent variable can be made dependent variable By analyzing these data points Perform linear regression analysis without intercept. The obtained regression coefficients That is, the initial magnitude that is objectively defined as the sensitivity vector. ;

[0116] This embodiment describes two core parameters of the state evolution model. and A completely transparent, reproducible quantitative calibration scheme is provided; it deeply binds model parameters with the physical specifications, functional layout and historical response behavior of assets, completely eliminates the subjectivity and randomness of parameter setting, and ensures the rationality and accuracy of the prediction model in the cold start stage.

[0117] Embodiment 7

[0118] S4 comprises:

[0119] The future core indicators are compared with the preset business alarm threshold to identify risk assets; the system instability index is calculated in combination with the business weight of the risk assets, the impact gradient suffered and the monitoring response threshold; based on the system instability index and the preset business safety threshold, and taking the minimum total intervention cost as the principle, an intervention strategy is solved and generated;

[0120] The business weight is quantitatively calibrated by the analytic hierarchy process or contribution analysis based on historical data.

[0121] The intervention strategy generation process is specified.

[0122] The future core indicators are compared with the preset business alarm threshold to identify risk assets; the system compares the future core indicators of shopping center B predicted by the evolutionary model with the preset business alarm threshold , such as the predicted parking lot congestion rate reaching 95% ; the threshold is determined, usually based on statistical analysis of the critical point of indicators in historical data that cause a significant decline in service quality, or directly referencing relevant industry operation safety standards; once the predicted value breaks through the threshold, shopping center B is identified as a risk asset;

[0123] The system instability index is calculated in combination with the business weight of the risk assets, the impact gradient suffered and the monitoring response threshold; to quantify the severity of the overall risk, the system introduces a system instability index ; the calculation of the index requires the business weight ; the weight is a parameter quantitatively calibrated by the analytic hierarchy process or contribution analysis based on historical data, reflecting the importance of a single asset in the entire business network;

[0124] Among them, the contribution analysis based on historical data can include: defining a system-level key performance indicator that can measure the overall running state of the system, for example, the average traffic efficiency of the regional traffic network; building a multiple linear regression model with the system-level KPI as the dependent variable and the core indicators of each risk asset as the independent variable; the absolute value of the standardized regression coefficient corresponding to each asset core indicator in the model is directly calibrated as the business weight of the asset The coefficient value objectively reflects the influence degree of single asset state change on the system overall state;

[0125] Unstable index The influence excess degree of all risk assets is weighted and summed according to its importance by the following formula:

[0126]

[0127] Wherein, The input parameters of the system unstable index are as follows: The total number of identified risk assets, The pre-labeled business weight of each risk asset, The influence gradient of each risk asset, The monitoring response threshold of each risk asset; : subscript, representing each risk asset from 1 to N;

[0128] Based on the system unstable index and the preset business safety threshold, and taking the minimum total intervention cost as the principle, the intervention strategy is solved and generated; the system compares the calculated Value with a business safety threshold The setting of the threshold Is to ensure that the system returns to an acceptable low-risk state after intervention, and the value can be derived by analyzing the highest Value in history without chain reaction; if , the system will start the intervention strategy optimization process, and the optimization goal is to find a combination of measures with the minimum total intervention cost Under the premise of reducing the unstable index After intervention to Below; the cost of each measure Is derived from a cost library established based on historical resource consumption or expert evaluation;

[0129]

[0130] Wherein, : the cost of each intervention measure in the measure library, as indicated by subscript j; : subscript, referring to each intervention measure in the measure library; : unstable index after executing intervention; : preset business safety threshold; the system automatically generates and recommends the cost-effective intervention strategy by simulating the influence of each operation in the measure library on the source asset intensity;

[0131] To ensure the objectivity of the cost, the establishment of the cost library is preferably based on quantifiable data; for intervention measures involving material consumption, such as traffic cone deployment, the cost is the product of the market price and the quantity of the material; for measures involving human dispatch, such as increasing the number of traffic coordinators, the cost is the product of the standard man-hour cost and the estimated input working hours; for measures requiring the calling of third-party services, such as the use of towing services, the cost is the market public quotation thereof; by structurally storing these quantifiable cost data, an objective cost library can be formed;

[0132] The embodiment introduces a system instability index and a business weight , converts the fuzzy system risk into a quantifiable decision index capable of reflecting the business priority; further, by constructing an optimization problem with the cost-benefit optimization as the target, the automatic and optimized generation of the intervention strategy is realized, which replaces the traditional mode of relying on artificial experience for decision-making, so that the emergency response is more rapid, scientific and economical;

[0133] The progress of the present application lies in that it does not stop at prediction, but constructs a complete prediction-decision-intervention-learning closed loop; it quantifies the multiple and concurrent future risks into a single decision index through the system instability index , wherein, : the system instability index, : the total number of risk assets, : the business weight, : the influence gradient, : the monitoring response threshold; based on this index, the system automatically solves and generates the optimal intervention strategy with the lowest total intervention cost as the principle; this marks the transition from the passive and artificial emergency response mode to the active and automatic intelligent operation mode.

[0134] Embodiment 8

[0135] The adaptive calibration step includes:

[0136] After the execution of the intervention strategy, the actual core indicators of the target assets are tracked; the prediction error between the actual core indicators and the future core indicators is calculated; the prediction error is used to modify the sensitivity vector in the state evolution model.

[0137] The embodiment specifically realizes the adaptive calibration step after the execution of the intervention strategy, and forms a complete and self-evolving learning closed loop through a dimensionally rigorous feedback mechanism.

[0138] After the execution of the intervention strategy, the system continuously tracks the actual core indicators of the target assets. After the system executes the intervention strategy, the process does not terminate, but continuously tracks and records the actual core indicators of the core indicators of the shopping center B in the subsequent time through the sensor or the data interface .

[0139] The prediction error between the actual core indicator and the future core indicator is calculated, and the system compares the real data collected with the future core indicator generated by the state evolution model before the intervention , i.e. the predicted value, to obtain a quantitative prediction error ; wherein, : prediction error; the error accurately reflects the deviation between the model prediction and the real world;

[0140] Using the prediction error, the sensitivity vector in the state evolution model is corrected; the prediction error is used as a feedback signal to drive the automatic calibration of the core parameters of the state evolution model; to ensure the rigor of the calibration process in mathematical and physical dimensions, the sensitivity vector is updated according to the following gradient descent correction rule:

[0141]

[0142] wherein, is the updated sensitivity vector, and the input for calculation is: is the sensitivity vector before updating, is the actual core indicator tracked after the intervention, is the core indicator predicted by the model, is the time step used by the state evolution model, is the impact gradient vector calculated in the previous step, is the square of the length of the impact gradient, and is a dimensionless machine learning hyperparameter, i.e. learning rate, preset to balance the convergence speed and stability;

[0143] In the preferred embodiment, the initial value of the learning rate can be set in the conventional range of to ; those skilled in the art can dynamically adjust the value of by cross-validation method or monitoring the convergence of the prediction error during the actual operation of the model to seek the optimal model convergence speed and stability;

[0144] The physical meaning of this correction formula is that it takes the prediction error as the core driving force for correction, and makes the adjustment amount of the sensitivity vector proportional to the gradient vector that causes the impact; this means that the sensitivity vector is adjusted more strongly by the gradient vector The resulting error will cause a larger correction, which is consistent with the logical robust learning mechanism.

[0145] The present embodiment converts the entire analysis system from a static one-time prediction tool to an intelligent learning system capable of self-evolution by introducing a feedback calibration mechanism based on the predicted error rate, which is dimensionally consistent. This self-adaptive calibration capability ensures that the model parameters can continuously and automatically approach the physical laws of the real world, not only in mathematics, but also in physical interpretability, thereby continuously improving the accuracy of future predictions and the effectiveness of intervention decisions, and achieving truly closed-loop adaptive control.

[0146] The present method realizes self-evolution through an adaptive calibration mechanism. After the implementation of the intervention strategy, the system will use the observed actual core indicators and the predicted values of the model to correct the model parameters. For example, the correction of the sensitivity vector is completed through a dimensionally consistent gradient descent rule: wherein, is the new sensitivity vector, is the old sensitivity vector, is the dimensionless learning rate, is the actual indicator, is the predicted indicator, is the time step, is the influence gradient; is the square of the length of the influence gradient; this correction method converts the original prediction error into a change rate error with more physical meaning, and through the normalization processing of the gradient term, it ensures that the learning rate is a standard dimensionless hyperparameter, which solves the problem of inconsistent dimensions of the model; this closed-loop feedback enables the entire analysis system to have learning ability, continuously improve its cognition and prediction accuracy of complex spatio-temporal systems, and distinguish it from all static and fixed analysis models, becoming an intelligent decision execution body that can evolve with the environment.

[0147] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and all should be covered in the scope of the claims of the present application.​

Claims

1. A cloud-native spatiotemporal fusion analysis method based on a spatiotemporal asset catalog, characterized in that, The specific steps include: S1. Obtain the core elements of the source asset and construct a spatiotemporal influence potential model to solve for the spatiotemporal influence potential; S2. Based on the spatiotemporal influence potential, calculate the influence gradient and compare the magnitude of the influence gradient with a preset threshold to generate a coupling relationship evaluation result; S3. In response to the coupling relationship assessment results, the current core indicators of the target asset are calculated using the state evolution model to predict and generate future core indicators. S4. Based on future core indicators, generate intervention strategies, and after implementing the intervention strategies, use the prediction error to adaptively calibrate the parameters in the state evolution model. S1 includes: collecting the core elements of source assets from the spatiotemporal asset catalog, including spatiotemporal coordinates and core business indicators; The core business indicators are normalized to obtain the event source intensity; the spatiotemporal influence potential model is constructed by combining the event source intensity, the preset time decay coefficient and the preset regularization length; the spatiotemporal influence potential model is used to solve for the spatiotemporal influence potential. The source asset is a traffic event, and its core elements include the spatiotemporal coordinates of the traffic event and its core business indicators. The core business indicator is the congestion index corresponding to the traffic event. The formula for the intensity of the event source is: in, For the intensity of the event source, This represents the actual value of the congestion index. and These are the maximum and minimum values ​​of the congestion index recorded in the historical database, respectively. The formula for the spatiotemporal influence potential is: in, As a potential influenced by spacetime, For the intensity of the event source, The spatial distance between the source asset and the target asset. The time difference between the two The preset time decay coefficient, The preset regularization length; is the base of the natural logarithm; and It is calculated based on the spatiotemporal coordinates of the core elements of the source asset and the spatiotemporal coordinates of the target asset obtained from the spatiotemporal asset catalog.

2. The cloud-native spatiotemporal fusion analysis method based on a spatiotemporal asset catalog according to claim 1, characterized in that, The initial calibration method for regularization length includes calibrating the core scope of action or equivalent radius of the asset as the regularization length based on the asset's type and physical attributes; The initial calibration method for the time decay coefficient includes extracting historical data on the decay of the influence of similar events and fitting the data with an exponential decay model to obtain the time decay coefficient.

3. The cloud-native spatiotemporal fusion analysis method based on spatiotemporal asset catalog according to claim 1, characterized in that, S2 include: Calculate the negative gradient of the spatiotemporal influence potential to obtain the influence gradient; Collect the gradient magnitudes of historical events that did not cause changes in the core indicators of the target asset, and mark the high percentiles as the monitoring response thresholds; collect the gradient magnitudes of historical events that caused sudden changes in the core indicators of the target asset, and mark the low percentiles as the critical warning thresholds.

4. The cloud-native spatiotemporal fusion analysis method based on a spatiotemporal asset catalog according to claim 3, characterized in that, The steps to generate coupling relationship evaluation results include: When the magnitude of the gradient is less than or equal to the monitoring response threshold, the coupling relationship evaluation result is determined to be weak coupling. When the magnitude of the gradient is greater than the monitoring response threshold and less than or equal to the critical warning threshold, the coupling relationship evaluation result is determined to be strong coupling. When the magnitude affecting the gradient exceeds the critical warning threshold, the coupling relationship evaluation result is determined to be critical coupling.

5. The cloud-native spatiotemporal fusion analysis method based on spatiotemporal asset catalog according to claim 1, characterized in that, S3 include: For asset pairs whose coupling relationship assessment results are strong coupling or critical coupling, a state evolution model is used, and based on the influence gradient, the sensitivity vector of the target asset, and the capacity parameter, the current core indicators of the target asset are calculated to predict and generate future core indicators.

6. The cloud-native spatiotemporal fusion analysis method based on a spatiotemporal asset catalog according to claim 5, characterized in that, Capacity parameters are calibrated by obtaining the physical attributes of the target asset or extreme values ​​of historical data; the calibration methods for sensitivity vectors include: weighted average calculation based on historical data of the functional layout of the target asset to determine the direction; and linear regression analysis of historical influence gradients and changes in core indicators to determine the magnitude.

7. The cloud-native spatiotemporal fusion analysis method based on a spatiotemporal asset catalog according to claim 1, characterized in that, S4 includes: The system compares future core indicators with preset business alarm thresholds to identify risky assets; it calculates the system instability index by combining the business weight of the risky assets, the impact gradient, and the monitoring response threshold; and it generates an intervention strategy based on the system instability index and preset business safety thresholds, with the principle of minimizing total intervention cost. Business weights are quantified using the analytic hierarchy process (AHP) or contribution analysis driven by historical data.

8. The cloud-native spatiotemporal fusion analysis method based on a spatiotemporal asset catalog according to claim 1, characterized in that, The steps of adaptive calibration include: After the intervention strategy is implemented, the actual core indicators of the target asset are tracked; the prediction error between the actual core indicators and the future core indicators is calculated; and the sensitivity vector in the state evolution model is corrected using the prediction error.

Images