A safety patrol path acquisition method based on dynamic risk profiling

Abstract

The application discloses a kind of based on dynamic risk portrait safety inspection path acquisition method, belong to safety management technical field, method is: obtaining standard four-dimensional risk correlation data, and construct causal relationship chain based on it;Based on the above data and preset risk assessment model, obtain comprehensive weight vector and construct time series risk prediction model, and based on the above data, obtain time series risk grade set to generate three-dimensional risk heat map and identify risk aggregation area and hidden danger heat core, and then based on the above data, obtain cell initial state matrix;Based on the above data, obtain risk diffusion prediction result and dynamic risk portrait data;Based on the above data, generate inspection work order and then generate inspection path, complete safety inspection.Therefore, by implementing the present application, the risk prediction lag and the problem of inaccurate hidden danger investigation caused by data isolation can be avoided, accurate risk prediction is realized, thereby effectively improving the execution efficiency and comprehensive coverage of safety inspection.

Description

Technical Field

[0001] This invention relates to the field of security management technology, and in particular to a method for obtaining security patrol paths based on dynamic risk profiling. Background Technology

[0002] In modern, complex production and operational environments, safety inspections are a crucial link in ensuring the safety of personnel, assets, and the environment. Traditional safety inspection management systems often rely on human experience and fixed inspection procedures, making it difficult to cope with constantly changing risk situations. With the rapid development of technologies such as the Industrial Internet of Things, big data, and artificial intelligence, breaking through the limitations of traditional inspection models and achieving intelligent, precise, and adaptive safety inspections has become an inevitable trend in the current industry development, and also provides technical support for solving the pain points of traditional inspections.

[0003] Under the current technological background, existing security inspection systems generally suffer from the following problems, making it difficult to meet the needs of efficient and comprehensive inspection work: First, data isolation is a prominent issue, with data from different sources being stored in a scattered manner, lacking effective integration and correlation analysis, and failing to form a comprehensive risk understanding; Second, risk prediction has a significant lag, often relying on post-event analysis, failing to issue timely warnings and take intervention measures when risks are in their infancy, which can easily lead to the expansion and spread of risks; Third, the inspection mode is rigid, adopting a one-size-fits-all inspection strategy, unable to flexibly adjust according to the risk differences of different areas, equipment, and time periods, resulting in unreasonable allocation of inspection resources, insufficient inspection of key areas and failure to promptly identify hidden dangers, while excessive inspection of low-risk areas, which wastes resources and reduces the overall efficiency of inspection work. Summary of the Invention

[0004] This invention aims to provide a method for obtaining security patrol paths based on dynamic risk profiles, in order to solve the above-mentioned technical problems, avoid the problems of delayed risk prediction and inaccurate hidden danger investigation caused by isolated data, achieve accurate risk prediction, and effectively improve the execution efficiency and comprehensive coverage of security patrols.

[0005] To address the aforementioned technical problems, this invention provides a method for obtaining security patrol paths based on dynamic risk profiling, comprising: Obtain standard four-dimensional risk correlation data and construct causal relationship chains based on the standard four-dimensional risk correlation data; Based on standard four-dimensional risk correlation data, causal relationship chain and preset risk assessment model, obtain comprehensive weight vector and construct time series risk prediction model, and obtain time series risk level set of target area based on time series risk prediction model and standard four-dimensional risk correlation data; Based on the target area time-series risk level set and the preset target area map, a three-dimensional risk heat map is generated and risk clustering areas and hidden danger heat cores are identified. Based on the risk clustering areas, hidden danger heat cores and the three-dimensional risk heat map, multi-dimensional state values ​​are assigned to the preset cell space to obtain the cell initial state matrix. Based on causal relationship chains, comprehensive weight vectors, standard four-dimensional risk association data, cellular initial state matrices, and preset cellular automata models, risk diffusion prediction results and dynamic risk profile data are obtained. Based on dynamic risk profile data, patrol task orders for target areas are generated, and patrol routes are generated based on the patrol task orders for target areas and the risk diffusion prediction results.

[0006] In the above scheme, by constructing a causal relationship chain, the isolated and scattered state of multi-source data is broken, and a standard four-dimensional risk correlation data correlation system is established. This enables the tracing of risk factors across dimensions, allowing subsequent risk prediction to be based not only on single-dimensional data, thus solving the problems of delayed risk prediction and inaccurate hazard identification caused by isolated data. Next, by calculating a comprehensive weight vector and constructing a time-series risk prediction model, the risk level of the target area can be predicted, solving the problem of delayed risk prediction. The resulting time-series risk level set of the target area can better match the actual risk situation of the target area, improving the accuracy of risk prediction. Then, by combining the time-series risk level set of the target area with a preset target area map to generate a three-dimensional risk heat map, risk clustering areas and hazard heat cores are identified, thereby obtaining the initial state matrix of cells. This provides an initial state that fits the actual hazard distribution for accurate simulation of subsequent risk diffusion, further solving the pain point of inaccurate hazard identification. Subsequently, by correcting and optimizing the preset cellular automata model, risk diffusion prediction results are obtained. This enables precise determination of the risk level and spatial location of potential hazards in the current target area, achieving comprehensive and dynamic accurate risk prediction. The resulting dynamic risk profile data aggregates all risk information for the target area, fundamentally improving the comprehensiveness and accuracy of risk prediction. Finally, patrol paths are generated based on the dynamic risk profile data and risk diffusion prediction results. This avoids over-patrolling low-risk areas while ensuring comprehensive coverage of high-risk areas, solving the problems of delayed risk prediction and inaccurate hazard investigation caused by isolated data. This effectively improves the execution efficiency and comprehensive coverage of safety patrols.

[0007] Furthermore, the step of acquiring standard four-dimensional risk association data of the target area and constructing a causal relationship chain based on the standard four-dimensional risk association data includes: Acquire raw equipment status data, raw environmental risk data, raw human error data, and raw management defect data for the target area; The original equipment status data, original environmental risk data, original human error data, and original management defect data are sequentially subjected to conflict fusion, missing data filling, and noise filtering to obtain standard equipment status data, standard environmental risk data, standard human error data, and standard management defect data, which are then used to form standard four-dimensional risk correlation data. A causal relationship chain is constructed based on standard four-dimensional risk association data.

[0008] In the above scheme, by acquiring raw equipment status data, raw environmental risk data, raw human error data, and raw management defect data of the target area, complete and comprehensive basic data can be provided for subsequent data preprocessing and causal chain construction. Next, by sequentially performing conflict fusion, missing data filling, and noise filtering on the raw equipment status data, raw environmental risk data, raw human error data, and raw management defect data, conflicts, data gaps, and noise interference from multiple sources of raw data can be eliminated. Simultaneously, data format and unit unification are achieved, resulting in standard equipment status data, standard environmental risk data, standard human error data, and standard management defect data, which are then integrated into standard four-dimensional risk correlation data. This breaks the isolation of multi-source data and provides high-quality and correlated standardized data for subsequent causal chain construction. Then, causal chains are constructed using the standard four-dimensional risk correlation data, establishing the inherent causal relationships between four-dimensional risk factors, enabling cross-dimensional risk factor correlation analysis. This provides causal correlation basis for subsequent risk assessment, time-series risk prediction, and risk diffusion simulation, solving the problems of delayed risk prediction and inaccurate hazard identification caused by isolated data.

[0009] Furthermore, the construction of a causal relationship chain based on standard four-dimensional risk association data includes: Extract several entities from standard four-dimensional risk association data, and extract several relationships between these entities; Based on the preset IF-THEN deterministic causal rule, several relationships are filtered to obtain the initial causal chain backbone; Based on several entities, several relationships, and a preset four-dimensional Bayesian network, the fault causal contribution of the relationships is calculated, and the relationships with a fault causal contribution greater than or equal to the preset fault causal contribution threshold are taken as the initial implicit causal relationships. Verify the initial implicit causal relationship through counterfactual reasoning to obtain the valid implicit causal relationship; The initial causal chain backbone and effective implicit causal relationships are filtered based on a preset causal chain length threshold and a preset path confidence threshold to obtain the standard causal chain backbone and standard implicit causal relationships. A causal relationship chain is constructed based on the standard causal chain backbone and the standard implicit causal relationship.

[0010] In the above scheme, several risk-related entities are extracted from standard four-dimensional risk association data, and several relationships between these entities are extracted, providing support for the construction of causal relationship chains and breaking down the barriers between four-dimensional risk data. Next, the relationships are screened using a preset IF-THEN deterministic causal rule, ensuring the accuracy and reliability of the initial causal chain backbone and providing a core framework for the subsequent improvement of causal relationships. Then, using several entities, several relationships, and a preset four-dimensional Bayesian network, the causal contribution of equipment failures corresponding to each relationship is calculated. Relationships with a failure causal contribution greater than or equal to a preset failure causal contribution threshold are identified as initial implicit causal relationships, thereby uncovering implicit causal relationships, filling the coverage gaps of the initial implicit causal relationships, and making the causal relationships more comprehensive. Subsequently, counterfactual reasoning is used to verify the initial implicit causal relationships, eliminating false relationships and obtaining valid implicit causal relationships, ensuring that the obtained valid implicit causal relationships are real, improving the accuracy of causal chain construction, and avoiding the interference risk of invalid relationships. Next, by pre-setting a causal chain length threshold and a pre-setting path confidence threshold, the initial causal chain backbone and effective implicit causal relationships are dually screened to obtain standard causal chain backbones and standard implicit causal relationships. This optimizes the structure and quality of the causal chain, ensuring its simplicity and reliability. Finally, through the standard causal chain backbone and standard implicit causal relationships, a causal relationship chain is constructed, forming a complete causal association system for risk factors. This enables cross-dimensional risk factor tracing and cross-validation, solving the problems of delayed risk prediction and inaccurate hazard identification caused by isolated data at the data association level.

[0011] Furthermore, the step of obtaining a comprehensive weight vector based on standard four-dimensional risk correlation data, causal relationship chains, and a preset risk assessment model, constructing a time-series risk prediction model, and obtaining a target area time-series risk level set based on the time-series risk prediction model and standard four-dimensional risk correlation data includes: Calculate the information entropy for the standard four-dimensional risk correlation data to obtain the objective weight vector; Based on the preset subjective weight vector, preset dynamic adjustment factor and objective weight vector, a comprehensive weight vector is calculated, and the weight is adapted based on the comprehensive weight vector and the preset risk assessment model to obtain a standard risk assessment model. A time-series risk prediction model is constructed based on causal relationship chains and standard risk assessment models, and a set of time-series risk levels for the target area is obtained based on the time-series risk prediction model and standard four-dimensional risk correlation data.

[0012] In the above scheme, information entropy is calculated for each standard four-dimensional risk correlation data to obtain an objective weight vector. This avoids weight bias caused by relying solely on a preset subjective weight vector, providing an objective basis for calculating the comprehensive weight vector. Furthermore, leveraging the cross-dimensional correlation of the standard four-dimensional risk correlation data further breaks down data silos, laying an objective foundation for risk quantification assessment. Next, a comprehensive weight vector, balancing subjective expertise and data objectivity, is calculated by pre-setting a subjective weight vector, a preset dynamic adjustment factor, and an objective weight vector. Based on this comprehensive weight vector and a preset risk assessment model, a weight adaptation standard is applied, resulting in a risk assessment model with dynamic adaptation capabilities. This provides a precise quantitative assessment framework for subsequent time-series risk prediction. Then, a time-series risk prediction model is constructed using causal chains and the standard risk assessment model. Utilizing the time-series risk prediction model's ability to capture long-term dependencies in time-series data, combined with the standard four-dimensional risk correlation data, a time-series risk level set for the target area is obtained. This provides a precise time-series quantitative basis for subsequent three-dimensional risk heat map generation and risk cluster area identification, further improving the targeting of hazard investigation.

[0013] Furthermore, the construction of a time-series risk prediction model based on causal relationship chains and standard risk assessment models, and the acquisition of a time-series risk level set for the target area based on the time-series risk prediction model and standard four-dimensional risk correlation data, includes: An initial model for time-series risk prediction was constructed based on causal chains and standard risk assessment models. Obtain historical time-series risk data and actual risk level data corresponding to the historical time-series risk data; Based on historical time-series risk data and an initial time-series risk prediction model, risk level prediction data is obtained. The risk level error is calculated based on the risk level prediction data and the actual risk level data. The initial model for time series risk prediction is trained with the goal of minimizing the risk level error until the preset first convergence condition is met, and the time series risk prediction model is obtained. The time-series risk level set of the target area is obtained based on the time-series risk prediction model and standard four-dimensional risk correlation data.

[0014] In the above scheme, an initial time-series risk prediction model is constructed through a causal chain and a standard risk assessment model, laying the foundation for subsequent model training and optimization. Next, by acquiring historical time-series risk data and the corresponding actual risk level data, labeled sample data is provided to support the training of the initial time-series risk prediction model, ensuring that the trained model's predictions closely match real-world scenarios. Then, risk level prediction data is obtained using the historical time-series risk data and the initial time-series risk prediction model, providing data for subsequent calculation of risk level errors and optimization of model parameters. Subsequently, the risk level error is calculated using the risk level prediction data and the actual risk level data, and the initial time-series risk prediction model is trained with the goal of minimizing the risk level error until a preset first convergence condition is met, resulting in a time-series risk prediction model. This model ensures that the prediction results of the time-series risk prediction model closely match the actual risk level data corresponding to the historical time-series risk data, thereby effectively improving the model's risk prediction accuracy and solving the risk prediction bias problem before model optimization. Finally, by incorporating standard four-dimensional risk correlation data into the time-series risk prediction model, the risk levels of the target area in different time dimensions can be dynamically predicted, and a time-series risk level set of the target area can be output. This upgrades risk prediction from static level determination to dynamic time-series prediction, solving the technical problem of lagging risk prediction. At the same time, it provides accurate time-series quantitative basis for subsequent three-dimensional risk heat map generation, risk cluster area identification, and hidden danger thermal core labeling, further improving the targeting and accuracy of hidden danger investigation.

[0015] Furthermore, based on the target area time-series risk level set and a preset target area map, a three-dimensional risk heat map is generated, and risk clustering areas and hidden danger heat cores are identified. Then, based on the risk clustering areas, hidden danger heat cores, and the three-dimensional risk heat map, multi-dimensional state assignments are performed on the preset cell space to obtain the cell initial state matrix, including: Based on the time-series risk level set of the target area and the preset target area map, a three-dimensional risk heat map is generated and risk clustering areas and hidden danger heat cores are identified; The preset cell space is divided based on the preset grid granularity to obtain several cell grids, and the preset target area map and cell grids are bound together to obtain the target area grid distribution map. Based on the aforementioned risk clustering area, hidden danger thermal core, and three-dimensional risk heat map, multi-dimensional state assignment is performed on the grid distribution map of the target area to obtain the initial state matrix of the cells.

[0016] In the above scheme, by using a set of temporal risk levels for the target area and a preset target area map, abstract temporal risk level values ​​can be transformed into a spatially visualized 3D risk heat map. Simultaneously, it accurately locates high-risk density risk clusters and hazard heat cores superimposed with high-frequency historical accident points, completing spatial targeted risk labeling. This solves the accuracy problem of traditional hazard investigation lacking clear spatial orientation, providing a realistic and accurate spatial risk distribution basis for subsequent state assignment in the cellular space. Next, by dividing the preset cellular space with a preset grid granularity, several cellular grids are obtained. The preset target area map and the cellular grids are spatially bound, transforming the target area into a model-recognizable and computable cellular grid space, obtaining a grid distribution map of the target area, and providing a structured spatial foundation for subsequent multi-dimensional state assignment. Then, by assigning multi-dimensional state values ​​to the grid distribution map of the target area through the risk clustering area, the hidden danger heat core and the three-dimensional risk heat map, each cell grid can be given multi-dimensional state characteristics such as risk level, hidden danger type and diffusion potential, so that the initial state of the cell closely matches the actual risk distribution and the cell initial state matrix is ​​obtained. This can provide accurate initial parameter support for subsequent risk diffusion path simulation and prediction, realize the transition from static risk space positioning to dynamic risk evolution prediction, and further improve the comprehensiveness and accuracy of risk prediction.

[0017] Furthermore, the standard four-dimensional risk association data includes standard environmental risk data; the acquisition of risk diffusion prediction results and dynamic risk profile data based on causal relationship chains, comprehensive weight vectors, standard four-dimensional risk association data, cellular initial state matrices, and preset cellular automata models includes: Based on the causal relationship chain, standard four-dimensional risk association data are filtered to obtain a standard indicator set of diffusion characteristics; Based on the standard index set of diffusion characteristics and the comprehensive weight vector, a comprehensive diffusion score set is obtained, and the comprehensive diffusion score set is mapped to a diffusion potential level set according to a preset interval. Based on the diffusion potential level set and the preset dynamic neighbor rules, standard environmental risk data are matched to obtain a cell neighbor weight matrix; Based on the cell initial state matrix, cell neighbor weight matrix, preset cell state transition triggering conditions, standard four-dimensional risk association data and comprehensive weight vector, obtain the cell state transition probability set; Based on the cell initial state matrix, cell state transition probability set, standard four-dimensional risk association data, and preset cellular automata model, risk diffusion prediction results and dynamic risk profile data are obtained.

[0018] In the above scheme, targeted screening of standard four-dimensional risk-related data through causal relationship chains can extract risk indicators highly correlated with risk diffusion, eliminate irrelevant and redundant data, and obtain a standard indicator set of diffusion characteristics. This provides a foundation for subsequent quantitative calculation of risk diffusion capacity, avoiding invalid data from interfering with the accuracy of risk diffusion simulation. Next, a comprehensive diffusion score set is obtained by calculating the diffusion score set and the comprehensive weight vector, making the calculation of diffusion scores more closely reflect real-world scenarios. Then, the comprehensive diffusion score set is mapped to a diffusion potential level set according to preset intervals, enabling quantitative assessment and level classification of risk diffusion capacity at various locations in the target area. This makes the representation of risk diffusion capacity more intuitive and provides a basis for subsequent matching of preset dynamic neighbor rules and construction of the cell neighbor weight matrix. Finally, the standard environmental risk data is dynamically matched using the diffusion potential level set and preset dynamic neighbor rules to obtain the cell neighbor weight matrix. This corrects the influence weights of neighbors in each direction, ensuring that the cell neighbor weight matrix conforms to actual environmental physical laws and differences in risk diffusion capacity. Subsequently, by using the initial state matrix of cells, the neighbor weight matrix of cells, preset cell state transition triggering conditions, standard four-dimensional risk correlation data, and comprehensive weight vector, the probability of each cell transitioning from the current risk state to a higher-level risk state was quantified, and a cell state transition probability set was obtained. Finally, dynamic simulation calculations of risk diffusion were performed using the initial state matrix of cells, the cell state transition probability set, standard four-dimensional risk correlation data, and a preset cellular automata model to obtain risk diffusion prediction results and dynamic risk profile data. This allows for accurate prediction of the future risk diffusion direction, scope, and trend in the target area. Simultaneously, it integrates all risk information to form dynamic risk profile data, realizing the transformation from static state judgment to dynamic evolution prediction of risks. This solves the problem of lagging risk prediction and provides comprehensive and accurate dynamic risk basis for the generation of subsequent target area inspection task work orders and the planning of inspection routes, fundamentally improving the accuracy of hazard investigation and the pertinence of inspection work.

[0019] Furthermore, the step of obtaining risk diffusion prediction results and dynamic risk profile data based on the cell initial state matrix, cell state transition probability set, standard four-dimensional risk association data, and preset cellular automata model includes: Obtain historical causes of risk diffusion and corresponding risk diffusion data for those causes. Based on historical causes of risk diffusion and a pre-set cellular automata initial model, simulated diffusion data is obtained. The diffusion error is calculated based on simulated diffusion data and risk diffusion data. The initial cellular automaton model is trained with the goal of minimizing the diffusion error until the preset second convergence condition is met, and the preset cellular automaton model is obtained. Based on the cell initial state matrix, cell state transition probability set, standard four-dimensional risk association data, and preset cellular automata model, risk diffusion prediction results and dynamic risk profile data are obtained.

[0020] In the above scheme, by acquiring historical causes of risk diffusion and the corresponding risk diffusion data, it provides labeled sample data to support the training and parameter optimization of the initial cellular automata model, ensuring that the simulated diffusion data obtained by the trained model closely matches the actual evolution and causes of risk diffusion. Next, by acquiring simulated diffusion data using historical causes of risk diffusion and the preset initial cellular automata model, it provides reference data for subsequent calculation of diffusion error and targeted optimization of model parameters. Then, by calculating the diffusion error and iteratively training the initial cellular automata model with the goal of minimizing the diffusion error, it ensures that the simulated diffusion data output by the model closely matches the actual risk diffusion data, effectively improving the risk diffusion simulation accuracy of the cellular automata model, until a preset second convergence condition is met, obtaining the preset cellular automata model. This solves the problem of potential risk diffusion simulation bias and distorted prediction results due to unoptimized models. Finally, by using the initial state matrix of cells, the probability set of cell state transitions, standard four-dimensional risk correlation data, and a preset cellular automata model to dynamically simulate risk diffusion, the system obtains risk diffusion prediction results and dynamic risk profile data. This enables accurate prediction of the future direction, scope, and trend of risk diffusion in the target area. Simultaneously, it integrates all risk information to form dynamic risk profile data, realizing the transformation from static state judgment to dynamic evolution prediction of risks. This solves the problem of delayed risk prediction and provides comprehensive and accurate dynamic risk basis for the generation of subsequent target area inspection task orders and the planning of inspection routes, fundamentally improving the accuracy of hazard investigation and the pertinence of inspection work.

[0021] Furthermore, the step of generating target area patrol task orders based on dynamic risk profile data, and generating patrol paths based on target area patrol task orders and risk diffusion prediction results, includes: Based on dynamic risk profile data, extract a standard four-dimensional risk parameter vector; Based on the standard four-dimensional risk parameter vector and the preset inspection mode, an IF-THEN risk matching rule base is constructed. Perform rule matching and verification processing on the standard four-dimensional risk parameter vector and the IF-THEN risk matching rule base to obtain the rule engine matching result; Based on the standard four-dimensional risk parameter vector and the preset Bayesian network model, obtain the Bayesian network probability matching result; Based on the matching results of the rule engine and the probability matching results of the Bayesian network, the posterior probability of the patrol mode is calculated, and the posterior probability that meets the preset sorting conditions is determined as the appropriate patrol mode. Based on the adaptive patrol mode and dynamic risk profile data, generate patrol task work orders for the target area; Based on the target area patrol task work order and risk diffusion prediction results, a patrol route is generated.

[0022] In the above scheme, a standard four-dimensional risk parameter vector is extracted from dynamic risk profile data. This enables the structured refinement and integration of multi-source heterogeneous risk data in the dynamic risk profile, eliminating differences in data format and dimension, and providing standardized data support for subsequent inspection pattern matching. Next, an IF-THEN risk matching rule base is constructed using the standard four-dimensional risk parameter vector and preset inspection patterns. This establishes the logical association between risk parameters and inspection patterns, laying a structured rule foundation for rapid and accurate matching of inspection patterns and improving the logic and execution efficiency of pattern matching. Then, by performing rule matching verification processing on the standard four-dimensional risk parameter vector and the IF-THEN risk matching rule base, the initial inspection pattern for risk scenarios can be determined. For scenarios that conform to clear rules, the rule engine matching results are directly output, significantly improving the efficiency of pattern matching in conventional risk scenarios. Simultaneously, the standard four-dimensional risk parameter vector for ambiguous or conflicting risk scenarios can be processed by subsequent Bayesian network models. Subsequently, Bayesian network probability matching results were obtained using a standard four-dimensional risk parameter vector and a pre-set Bayesian network model. This solved the problem of judging fuzzy or conflicting risk scenarios that the rule engine could not handle, overcoming the limitations of pure rule matching and improving the comprehensiveness and accuracy of pattern matching. Next, the posterior probability of the inspection mode was calculated using the rule engine matching results and the Bayesian network probability matching results. The posterior probability that met the pre-set ranking conditions was determined as the adapted inspection mode, ensuring that the adapted inspection mode matched the actual risk scenario and providing accurate pattern basis for the subsequent generation of inspection task work orders. Then, by generating target area inspection task work orders using the adapted inspection mode and dynamic risk profile data, the core information in the dynamic risk profile can be combined to focus inspection tasks on high-risk areas and core hidden danger points, improving the targeting of inspection tasks. Finally, by generating patrol paths based on target area patrol task work orders and risk diffusion prediction results, patrol paths can prioritize coverage of high-risk predicted areas, avoiding over-patrolling of low-risk areas, effectively improving the execution efficiency of security patrols and achieving comprehensive security patrols. Finally, patrols can be completed according to the optimized patrol paths, thereby completely solving the problems of low patrol efficiency, incomplete patrol path coverage, and inaccurate investigation caused by data isolation and delayed risk prediction.

[0023] Further, obtaining the Bayesian network probability matching result based on the standard four-dimensional risk parameter vector and the preset Bayesian network model includes: Obtain historical patrol data and the corresponding actual pattern data; An initial Bayesian network model is constructed based on a standard four-dimensional risk parameter vector, a preset inspection mode, and a causal relationship chain. Based on the historical patrol data and the initial Bayesian network model, predictive pattern data of the historical patrol data is obtained; The prediction consistency rate is calculated based on the predicted pattern data and the actual pattern data. The initial Bayesian network model is trained with the goal of maximizing the prediction consistency rate until the preset third convergence condition is reached, and the preset Bayesian network model is obtained. Based on the standard four-dimensional risk parameter vector and the preset Bayesian network model, the Bayesian network probability matching result is obtained.

[0024] In the above scheme, by acquiring historical patrol data and the corresponding actual pattern data, labeled sample data is provided to support the training and parameter optimization of the initial Bayesian network model. Next, an initial Bayesian network model is constructed using a standard four-dimensional risk parameter vector, a preset patrol pattern, and a causal relationship chain, providing a model framework with risk causal logic support for subsequent iterative training. Then, using the historical patrol data and the initial Bayesian network model, predicted pattern data from the historical patrol data is obtained, providing reference data for calculating the prediction consistency rate. Subsequently, the prediction consistency rate is calculated using the predicted pattern data and the actual pattern data, and the initial Bayesian network model is iteratively trained with the goal of maximizing the prediction consistency rate. This continuously improves the model's prediction accuracy and matching fit for patrol patterns until a preset third convergence condition is met, resulting in a preset Bayesian network model. This solves the problems of large prediction bias and incompatibility with actual scenarios in the initial Bayesian network model, providing reliable and accurate model support for subsequent patrol pattern probability matching. Finally, by obtaining Bayesian network probability matching results through standard four-dimensional risk parameter vectors and preset Bayesian network models, we can provide a scientific probabilistic basis for determining inspection patterns in fuzzy or conflicting risk scenarios that rule engines cannot handle, make up for the limitations of pure rule matching, and improve the comprehensiveness, accuracy and adaptability of inspection pattern matching. Attached Figure Description

[0025] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0026] Figure 1 A flowchart illustrating a method for obtaining security patrol paths based on dynamic risk profiling, provided in an embodiment of the present invention; Figure 2This is an architecture diagram of a security patrol path acquisition system based on dynamic risk profiling, provided as an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0029] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0030] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0031] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0032] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0033] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0034] See Figure 1 To avoid delays in risk prediction and inaccurate hazard identification due to isolated data, and to achieve accurate risk prediction, thereby effectively improving the efficiency and comprehensiveness of security patrols, this embodiment provides a method for obtaining security patrol paths based on dynamic risk profiles. The flowchart of this method can be found in [link to flowchart]. Figure 1 ,include: Step S1: Obtain standard four-dimensional risk association data and construct a causal relationship chain based on the standard four-dimensional risk association data; Step S2: Based on standard four-dimensional risk correlation data, causal relationship chain and preset risk assessment model, obtain comprehensive weight vector and construct time series risk prediction model, and obtain target area time series risk level set based on time series risk prediction model and standard four-dimensional risk correlation data; Step S3: Based on the target area time-series risk level set and the preset target area map, generate a three-dimensional risk heat map and identify risk clustering areas and hidden danger heat cores. Then, based on the risk clustering areas, hidden danger heat cores and the three-dimensional risk heat map, assign multi-dimensional state values ​​to the preset cell space to obtain the cell initial state matrix. Step S4: Based on the causal relationship chain, comprehensive weight vector, standard four-dimensional risk association data, cellular initial state matrix and preset cellular automata model, obtain risk diffusion prediction results and dynamic risk profile data; Step S5: Generate a target area patrol task work order based on dynamic risk profile data, and generate a patrol route based on the target area patrol task work order and risk diffusion prediction results.

[0035] In this embodiment, by constructing a causal relationship chain, the isolated and scattered state of multi-source data is broken, and a standard four-dimensional risk correlation data association system is established. This enables the tracing of risk factors across dimensions, allowing subsequent risk prediction to be based not only on single-dimensional data, thus solving the problems of delayed risk prediction and inaccurate hazard identification caused by isolated data. Next, by calculating a comprehensive weight vector and constructing a time-series risk prediction model, the risk level of the target area can be predicted, solving the problem of delayed risk prediction. The resulting time-series risk level set of the target area can better match the actual risk situation of the target area, improving the accuracy of risk prediction. Then, by combining the time-series risk level set of the target area with a preset target area map to generate a three-dimensional risk heat map, risk clustering areas and hazard heat cores are identified, thereby obtaining the initial state matrix of cells. This provides an initial state that fits the actual hazard distribution for accurate simulation of subsequent risk diffusion, further solving the pain point of inaccurate hazard identification. Subsequently, by correcting and optimizing the preset cellular automata model, risk diffusion prediction results are obtained. This enables precise determination of the risk level and spatial location of potential hazards in the current target area, achieving comprehensive and dynamic accurate risk prediction. The resulting dynamic risk profile data aggregates all risk information for the target area, fundamentally improving the comprehensiveness and accuracy of risk prediction. Finally, patrol paths are generated based on the dynamic risk profile data and risk diffusion prediction results. This avoids over-patrolling low-risk areas while ensuring comprehensive coverage of high-risk areas, solving the problems of delayed risk prediction and inaccurate hazard investigation caused by isolated data. This effectively improves the execution efficiency and comprehensive coverage of safety patrols.

[0036] Furthermore, the step of acquiring standard four-dimensional risk association data of the target area and constructing a causal relationship chain based on the standard four-dimensional risk association data includes: Acquire raw equipment status data, raw environmental risk data, raw human error data, and raw management defect data for the target area; The original equipment status data, original environmental risk data, original human error data, and original management defect data are sequentially subjected to conflict fusion, missing data filling, and noise filtering to obtain standard equipment status data, standard environmental risk data, standard human error data, and standard management defect data, which are then used to form standard four-dimensional risk correlation data. A causal relationship chain is constructed based on standard four-dimensional risk association data.

[0037] In this embodiment, IoT sensors are deployed on various devices to monitor the raw equipment status data of the target area in real time. Meteorological sensors collect raw environmental risk data such as temperature, humidity, wind speed, and rainfall in real time. Smart safety helmets integrate positioning functions and GPS or BeiDou satellite navigation systems to track personnel location information in real time, ensuring personnel remain within designated work areas. Simultaneously, artificial intelligence algorithms are used to analyze work videos, obtaining raw data on human error and extracting raw management defect data through office automation (OA) systems. This provides complete and comprehensive foundational data for subsequent data preprocessing and causal chain construction. The IoT sensors include vibration sensors, temperature sensors, and pressure sensors: vibration sensors monitor the vibration amplitude and frequency of equipment in real time; temperature sensors accurately measure the temperature of key parts of the equipment; and pressure sensors monitor internal or external pressure changes. Combined with Building Information Modeling (BIM), the coordinates of the equipment in space can be accurately located. By combining the equipment's operating parameters with spatial location information, comprehensive monitoring of the equipment's entire lifecycle status can be achieved. The meteorological sensors are mounted on drones, leveraging their high mobility and wide field of view to quickly survey large areas, acquiring data on terrain changes, building conditions, and other information. This data allows for the construction of a three-dimensional environmental risk field model, visually displaying the distribution of environmental risks and helping safety managers quickly identify high-risk areas. The original human error data includes instances of personnel not wearing protective equipment, unauthorized operations, and dangerous behaviors such as unauthorized climbing or hot work without approval. The original management defect data includes deviations in workflow approval records, task assignments, hazard rectification records, and safety logs. Next, by sequentially applying the DS evidence theory to conflict fusion of raw equipment status data, raw environmental risk data, raw human error data, and raw management defect data, and using an improved KNN algorithm for missing data filling and a noise-reducing autoencoder (DAE) for noise filtering, the problems of conflict, data gaps, and noise interference in multi-source raw data can be eliminated. Simultaneously, data format and units are unified, resulting in standard equipment status data, standard environmental risk data, standard human error data, and standard management defect data, which are then integrated into standard four-dimensional risk correlation data. This breaks the isolation of multi-source data and provides high-quality and correlated standardized data for the subsequent construction of causal relationship chains. Then, causal relationship chains are constructed using the standard four-dimensional risk correlation data to establish the intrinsic causal relationships between four-dimensional risk factors, enabling cross-dimensional risk factor correlation analysis. This provides causal correlation evidence for subsequent risk assessment, time-series risk prediction, and risk diffusion simulation, solving the problems of delayed risk prediction and inaccurate hazard identification caused by data isolation.

[0038] Furthermore, the construction of a causal relationship chain based on standard four-dimensional risk association data includes: Extract several entities from standard four-dimensional risk association data, and extract several relationships between these entities; Based on the preset IF-THEN deterministic causal rule, several relationships are filtered to obtain the initial causal chain backbone; Based on several entities, several relationships, and a preset four-dimensional Bayesian network, the fault causal contribution of the relationships is calculated, and the relationships with a fault causal contribution greater than or equal to the preset fault causal contribution threshold are taken as the initial implicit causal relationships. Verify the initial implicit causal relationship through counterfactual reasoning to obtain the valid implicit causal relationship; The initial causal chain backbone and effective implicit causal relationships are filtered based on a preset causal chain length threshold and a preset path confidence threshold to obtain the standard causal chain backbone and standard implicit causal relationships. A causal relationship chain is constructed based on the standard causal chain backbone and the standard implicit causal relationship.

[0039] In this embodiment, several risk-related entities are extracted from standard four-dimensional risk association data, and several relationships between these entities are extracted, providing support for the construction of a causal chain and breaking down barriers between four-dimensional risk data. For structured data in the standard four-dimensional risk association data, entities are extracted based on rule-driven extraction; for unstructured data, entities are extracted based on named entity recognition models, such as BERT-NER combined with a domain dictionary. The relationships include temporal, spatial, and attribute relationships. Next, the relationships are filtered using a preset IF-THEN deterministic causal rule, retaining high-confidence relationships to ensure the accuracy and reliability of the initial causal chain backbone, providing a core framework for the subsequent refinement of causal relationships. The preset IF-THEN deterministic causal rule is formulated by combining knowledge of safety management and historical accident experience. Then, using several entities, several relationships, and a preset four-dimensional Bayesian network, the causal contribution of each relationship to device failure can be calculated using Shapley values. Relationships with a failure causal contribution greater than or equal to a preset failure causal contribution threshold are identified as initial implicit causal relationships, thereby uncovering implicit causal associations, filling the coverage gaps of the initial implicit causal relationships, and making the causal relationships more comprehensive. The preset failure causal contribution threshold can be 0.5. Subsequently, by performing counterfactual reasoning verification on the initial implicit causal relationships, false relationships are eliminated and valid implicit causal relationships are obtained, ensuring that the obtained valid implicit causal relationships are real, improving the accuracy of causal chain construction, and avoiding the interference risk of invalid relationships. Next, using a preset causal chain length threshold and a preset path confidence threshold, depth-first search (DFS) is used to double-filter the initial causal chain backbone and valid implicit causal relationships, obtaining standard causal chain backbones and standard implicit causal relationships. This optimizes the structure and quality of the causal chain, ensuring its simplicity and reliability. The preset causal chain length threshold can be 3, and the preset path confidence threshold can be 0.75. Finally, by constructing a causal relationship chain through the standard causal chain backbone and standard implicit causal relationship, a complete causal relationship system for risk factors is formed. When equipment failure is discovered, the causal relationship chain can be used to quickly trace possible environmental causes, human errors or management loopholes, providing a basis for formulating targeted improvement measures. This achieves the tracing and cross-verification of cross-dimensional risk factors and solves the problems of delayed risk prediction and inaccurate hidden danger investigation caused by isolated data from the perspective of data correlation.

[0040] Furthermore, the step of obtaining a comprehensive weight vector based on standard four-dimensional risk correlation data, causal relationship chains, and a preset risk assessment model, constructing a time-series risk prediction model, and obtaining a target area time-series risk level set based on the time-series risk prediction model and standard four-dimensional risk correlation data includes: Calculate the information entropy for the standard four-dimensional risk correlation data to obtain the objective weight vector; Based on the preset subjective weight vector, preset dynamic adjustment factor and objective weight vector, a comprehensive weight vector is calculated, and the weight is adapted based on the comprehensive weight vector and the preset risk assessment model to obtain a standard risk assessment model. A time-series risk prediction model is constructed based on causal relationship chains and standard risk assessment models, and a set of time-series risk levels for the target area is obtained based on the time-series risk prediction model and standard four-dimensional risk correlation data.

[0041] In this embodiment, information entropy is calculated for each standard four-dimensional risk correlation data using the entropy weight method. An objective weight vector is obtained based on the degree of variation in the standard four-dimensional risk correlation data, avoiding weight bias caused by relying solely on a preset subjective weight vector. This provides an objective basis for calculating the comprehensive weight vector. Furthermore, leveraging the cross-dimensional correlation of the standard four-dimensional risk correlation data further breaks down data silos, laying an objective foundation for risk quantification assessment. The preset subjective weight vector is determined using the analytic hierarchy process (AHP) based on expert judgments of the relative importance of the standard four-dimensional risk correlation data. Next, a comprehensive weight vector, balancing subjective expertise and data objectivity, is calculated using the preset subjective weight vector, a preset dynamic adjustment factor, and the objective weight vector. Based on this comprehensive weight vector and a preset risk assessment model, a weight adaptation standard is applied, resulting in a risk assessment model with dynamic adaptation capabilities. This provides a precise quantitative assessment framework for subsequent time-series risk prediction. Then, a time-series risk prediction model is constructed through causal relationship chains and standard risk assessment models. By utilizing the long-term dependence capture capability of the time-series risk prediction model on time series data and combining it with standard four-dimensional risk correlation data, a time-series risk level set for the target area is obtained. This provides accurate time-series quantitative basis for subsequent generation of three-dimensional risk heat maps and identification of risk cluster areas, further improving the targeting of hidden danger investigation.

[0042] Furthermore, the construction of a time-series risk prediction model based on causal relationship chains and standard risk assessment models, and the acquisition of a time-series risk level set for the target area based on the time-series risk prediction model and standard four-dimensional risk correlation data, includes: An initial model for time-series risk prediction was constructed based on causal chains and standard risk assessment models. Obtain historical time-series risk data and actual risk level data corresponding to the historical time-series risk data; Based on historical time-series risk data and an initial time-series risk prediction model, risk level prediction data is obtained. The risk level error is calculated based on the risk level prediction data and the actual risk level data. The initial model for time series risk prediction is trained with the goal of minimizing the risk level error until the preset first convergence condition is met, and the time series risk prediction model is obtained. The time-series risk level set of the target area is obtained based on the time-series risk prediction model and standard four-dimensional risk correlation data.

[0043] In this embodiment, an initial time-series risk prediction model is constructed based on LSTM using causal relationship chains and standard risk assessment models, laying the foundation for subsequent model training and optimization. LSTM can effectively process time-series data, capturing long-term dependencies within the data. Furthermore, the introduction of an attention mechanism within LSTM allows the constructed initial time-series risk prediction model to automatically focus on key risk factors in the causal relationship chains, ignoring secondary information, thus improving prediction accuracy and ultimately outputting the risk level for each region, intuitively displaying the risk status of different areas. The hyperparameters for the initial time-series risk prediction model are set as follows: learning rate = 0.001, batch size = 32, number of training epochs = 50, and number of LSTM hidden units = 64. Next, by acquiring historical time-series risk data and the corresponding actual risk level data, labeled sample data is provided to support the training of the initial time-series risk prediction model, ensuring that the trained model's predictions closely match real-world scenarios. Then, risk level prediction data is obtained through historical time-series risk data and an initial time-series risk prediction model, providing data for subsequent calculation of risk level error and optimization of model parameters. Subsequently, the risk level error is calculated using the cross-entropy loss function on the risk level prediction data and the actual risk level data. The initial time-series risk prediction model is then trained using the backpropagation algorithm with the goal of minimizing the risk level error until a preset first convergence condition is met, resulting in a time-series risk prediction model. This model ensures that the prediction results closely match the actual risk level data corresponding to the historical time-series risk data, effectively improving the model's risk prediction accuracy and resolving the risk prediction bias problem before model optimization. Finally, by substituting standard four-dimensional risk correlation data into the time-series risk prediction model, the risk levels of the target area in different time dimensions can be dynamically predicted, outputting a time-series risk level set for the target area. This upgrades risk prediction from static level determination to dynamic time-series prediction, solving the technical problem of lagging risk prediction. Simultaneously, it provides accurate time-series quantitative basis for subsequent three-dimensional risk heat map generation, risk cluster area identification, and hazard thermal core labeling, further improving the targeting and accuracy of hazard investigation.

[0044] Furthermore, based on the target area time-series risk level set and a preset target area map, a three-dimensional risk heat map is generated, and risk clustering areas and hidden danger heat cores are identified. Then, based on the risk clustering areas, hidden danger heat cores, and the three-dimensional risk heat map, multi-dimensional state assignments are performed on the preset cell space to obtain the cell initial state matrix, including: Based on the time-series risk level set of the target area and the preset target area map, a three-dimensional risk heat map is generated and risk clustering areas and hidden danger heat cores are identified; The preset cell space is divided based on the preset grid granularity to obtain several cell grids, and the preset target area map and cell grids are bound together to obtain the target area grid distribution map. Based on the aforementioned risk clustering area, hidden danger thermal core, and three-dimensional risk heat map, multi-dimensional state assignment is performed on the grid distribution map of the target area to obtain the initial state matrix of the cells.

[0045] In this embodiment, by using a target area temporal risk level set and a preset target area map, GIS technology can be used to transform abstract temporal risk level values ​​into a spatially visualized 3D risk heat map, displaying risk levels on the map with different colors and brightness. Simultaneously, the density clustering algorithm (DBSCAN) accurately locates high-risk density risk clusters and overlays high-frequency historical accident points into hazard heat cores, completing spatial targeted risk labeling. This solves the accuracy problem of traditional hazard investigation lacking clear spatial orientation, providing a realistic and accurate spatial risk distribution basis for subsequent cellular space state assignment. Next, using a preset grid granularity, the preset cellular space is automatically divided using Python's GDAL library or ArcGIS tools, obtaining several cellular grids. Each grid is assigned a unique cell ID, and its basic attributes are associated with it. A unified spatial coordinate system is established, spatially binding the preset target area map and the cellular grids. This transforms the target area into a model-recognizable and computable cellular grid space, obtaining a target area grid distribution map, providing a structured spatial foundation for subsequent multi-dimensional state assignment. The unified spatial coordinate system can adopt a Gauss-Kruger projection coordinate system. The cellular grid serves as the spatial carrier of the cellular automata (CA) model and must match the physical characteristics and risk distribution of the target inspection area. Therefore, the preset grid granularity needs to balance simulation accuracy and computational efficiency, and can be set differently according to risk type: small granularity (1m×1m) is suitable for high-risk, small-scale scenarios, accurately capturing local risk diffusion; medium granularity (5m×5m) is suitable for conventional industrial scenarios, taking into account equipment distribution and environmental impact range; large granularity (20m×20m) is suitable for open areas, reducing redundant calculations. Then, the grid distribution map of the target area is assigned multi-dimensional state values ​​through the risk clustering area, hidden danger heat core, and three-dimensional risk heat map. The grid corresponding to the hidden danger heat core is set as the initial risk cell, which can assign multi-dimensional state characteristics such as risk level, hidden danger type, and diffusion potential to each cellular grid, so that the initial state of the cell closely matches the actual risk distribution. The initial state matrix of the cell is obtained, which can provide accurate initial parameter support for subsequent risk diffusion path simulation and prediction, realizing the transition from static risk spatial positioning to dynamic risk evolution prediction, and further improving the comprehensiveness and accuracy of risk prediction. The multi-dimensional state assignment specifically employs a three-dimensional state coding system, assigning the following scores: Risk level S includes 0 (no risk), 1 (low risk), 2 (medium risk), 3 (high risk), and 4 (extremely high risk); Hazard type T includes 0 (no hazard), 1 (equipment failure type), 2 (environmentally induced type), 3 (human error type), and 4 (management deficiency type); Diffusion potential P includes 0 (non-diffusion), 1 (weak diffusion), 2 (medium diffusion), and 3 (strong diffusion). For example, the cellular state code "3-2-3" in a chemical industrial park represents "high risk - environmentally induced type - strong diffusion potential". Furthermore, the standard four-dimensional risk association data includes standard environmental risk data; the acquisition of risk diffusion prediction results and dynamic risk profile data based on causal relationship chains, comprehensive weight vectors, standard four-dimensional risk association data, cellular initial state matrices, and preset cellular automata models includes: Based on the causal relationship chain, standard four-dimensional risk association data are filtered to obtain a standard indicator set of diffusion characteristics; Based on the standard index set of diffusion characteristics and the comprehensive weight vector, a comprehensive diffusion score set is obtained, and the comprehensive diffusion score set is mapped to a diffusion potential level set according to a preset interval. Based on the diffusion potential level set and the preset dynamic neighbor rules, standard environmental risk data are matched to obtain a cell neighbor weight matrix; Based on the cell initial state matrix, cell neighbor weight matrix, preset cell state transition triggering conditions, standard four-dimensional risk association data and comprehensive weight vector, obtain the cell state transition probability set; Based on the cell initial state matrix, cell state transition probability set, standard four-dimensional risk association data, and preset cellular automata model, risk diffusion prediction results and dynamic risk profile data are obtained.

[0046] In this embodiment, by selectively screening standard four-dimensional risk-related data through causal relationship chains, risk indicators highly correlated with risk diffusion can be extracted, irrelevant and redundant data can be eliminated, a standard indicator set of diffusion characteristics can be obtained, and the original data of the standard indicator set of diffusion characteristics can be converted into standardized scores of 0-1 to eliminate the influence of unit differences and numerical ranges. This provides a foundation for the subsequent quantitative calculation of risk diffusion capabilities, so as to avoid invalid data interfering with the accuracy of risk diffusion simulation. Next, a comprehensive diffusion score set is obtained by calculating the standard index set of diffusion characteristics and the comprehensive weight vector. This makes the calculation of diffusion scores more consistent with the actual scenario. Then, the comprehensive diffusion score set is mapped to a diffusion potential level set according to a preset interval: Score [0, 0.2) Diffusion Level 0 (non-diffusion), the risk is limited to the current cell, with no propagation ability; Score [0.2, 0.5) Diffusion Level 1 (weak diffusion), the risk only affects directly adjacent cells (4 neighbors); Score [0.5, 0.8) Diffusion Level 2 (medium diffusion), the risk affects 3×3 grid cells (8 neighbors); Score [0.8, 1.0] Diffusion Level 3 (strong diffusion), the risk affects 5×5 grid cells and above (extended neighbors). This allows for the quantitative assessment and level classification of the risk diffusion ability at each location in the target area, making the representation of risk diffusion ability more intuitive and providing a basis for the matching of subsequent preset dynamic neighbor rules and the construction of the cell neighbor weight matrix. Then, standard environmental risk data is dynamically matched using a diffusion potential level set and preset dynamic neighbor rules to obtain a cell neighbor weight matrix. Specifically, since neighbors in different directions are affected by diffusion to varying degrees, weights are assigned based on the diffusion potential level set. For example, at a wind speed of 3 m / s, the weight of the north neighbor of a gas leak cell is 0.1 (headwind), the weight of the south neighbor is 0.8 (tailwind), and the weight of the east and west neighbors is 0.05 (crosswind). A weight matrix is ​​constructed based on the standard environmental risk data, and the weight calculation formula is as follows: ,in, This represents the neighbor weights of cell (i,j). This indicates the wind speed influence coefficient, which is preset. The wind speed component in that direction is represented by n, obtained from standard environmental risk data, ensuring the sum of weights is 1. This allows for the correction of the influence weights of neighbors in each direction, ensuring the cell-neighbor weight matrix aligns with actual environmental physics and differences in risk diffusion capabilities. The preset dynamic neighbor rules include: Von Neumann neighborhood (4-neighborhood) suitable for linear diffusion scenarios, considering only four adjacent cells (up, down, left, and right); Moore neighborhood (8-neighborhood) suitable for planar diffusion scenarios, considering eight adjacent cells; and extended neighborhood (e.g., 12-neighborhood) suitable for strong diffusion risks, extending the neighbor range to cells outside the 3×3 grid. Subsequently, the probability of each cell transitioning from its current risk state to a higher-level risk state is quantified using the cell initial state matrix, cell neighbor weight matrix, preset cell state transition trigger conditions, standard four-dimensional risk correlation data, and comprehensive weight vector, obtaining the cell state transition probability set. Finally, dynamic simulation calculations of risk diffusion are performed using the cell initial state matrix, cell state transition probability set, standard four-dimensional risk correlation data, and a pre-set cellular automata model to obtain risk diffusion prediction results and dynamic risk profile data, specifically: ,in, This represents the probability that a cell will transition from its current state to a higher-level risk state. This represents the comprehensive weight of the standard four-dimensional risk correlation data. This represents the normalized value of standard four-dimensional risk correlation data. This indicates the conversion threshold (which can be calibrated based on historical accident data; high-risk cells...). A value of 0.6 is acceptable for low-risk cells. (A weight of 0.8 can be used). For example, the weights are: equipment failure 0.3, environmental parameters 0.25, personnel intervention 0.25, and management measures 0.2. The initial state of the current cell is 1-1-2 (low risk - equipment failure - medium diffusion), and the trigger condition is that the equipment vibration value exceeds the threshold. =0.7) + No personnel intervention ( =0.9), the probability is calculated. The value is approximately 0.72, and the transformed state is 2-1-2 (medium risk). This allows for accurate prediction of the future risk diffusion direction, scope, and trend in the target area. Simultaneously, it integrates all risk information to form dynamic risk profile data, realizing the transformation from static state judgment to dynamic evolution prediction. This solves the problem of delayed risk prediction and provides comprehensive and accurate dynamic risk basis for the generation of subsequent target area inspection task orders and the planning of inspection routes, fundamentally improving the accuracy of hazard investigation and the targeting of inspection work. The preset cell state transition triggering conditions include: Condition 1: Activation of its own risk source; Condition 2: Influenced by neighboring risk cells; Condition 3: External intervention. The preset cellular automata model is a discrete model in time, space, and state. By defining the state and state transition rules of each cell, it can simulate the dynamic evolution process.

[0047] Furthermore, the step of obtaining risk diffusion prediction results and dynamic risk profile data based on the cell initial state matrix, cell state transition probability set, standard four-dimensional risk association data, and preset cellular automata model includes: Obtain historical causes of risk diffusion and corresponding risk diffusion data for those causes. Based on historical causes of risk diffusion and a pre-set cellular automata initial model, simulated diffusion data is obtained. The diffusion error is calculated based on simulated diffusion data and risk diffusion data. The initial cellular automaton model is trained with the goal of minimizing the diffusion error until the preset second convergence condition is met, and the preset cellular automaton model is obtained. Based on the cell initial state matrix, cell state transition probability set, standard four-dimensional risk association data, and preset cellular automata model, risk diffusion prediction results and dynamic risk profile data are obtained.

[0048] In this embodiment, by acquiring historical causes of risk diffusion and the corresponding risk diffusion data, labeled sample data is provided to support the training and parameter optimization of the initial cellular automaton model. This ensures that the simulated diffusion data obtained by the trained model closely matches the actual evolution and causes of risk diffusion. Next, by acquiring simulated diffusion data using historical causes of risk diffusion and the preset initial cellular automaton model, reference data is provided for subsequent calculation of diffusion error and targeted optimization of model parameters. Then, by calculating the diffusion error and iteratively training the initial cellular automaton model with the goal of minimizing the diffusion error, the simulated diffusion data output by the model closely matches the actual risk diffusion data, effectively improving the risk diffusion simulation accuracy of the cellular automaton model until a preset second convergence condition is met, thus obtaining the preset cellular automaton model. This solves the problem of potential risk diffusion simulation bias and distorted prediction results due to unoptimized models. The preset second convergence condition is a diffusion error of less than or equal to 15%. Finally, by using the initial state matrix of cells, the probability set of cell state transitions, standard four-dimensional risk correlation data, and a preset cellular automata model to dynamically simulate risk diffusion, the system obtains risk diffusion prediction results and dynamic risk profile data. This enables accurate prediction of the future direction, scope, and trend of risk diffusion in the target area. Simultaneously, it integrates all risk information to form dynamic risk profile data, realizing the transformation from static state judgment to dynamic evolution prediction of risks. This solves the problem of delayed risk prediction and provides comprehensive and accurate dynamic risk basis for the generation of subsequent target area inspection task orders and the planning of inspection routes, fundamentally improving the accuracy of hazard investigation and the pertinence of inspection work.

[0049] Furthermore, the step of generating target area patrol task orders based on dynamic risk profile data, and generating patrol paths based on target area patrol task orders and risk diffusion prediction results, includes: Based on dynamic risk profile data, extract a standard four-dimensional risk parameter vector; Based on the standard four-dimensional risk parameter vector and the preset inspection mode, an IF-THEN risk matching rule base is constructed. Perform rule matching and verification processing on the standard four-dimensional risk parameter vector and the IF-THEN risk matching rule base to obtain the rule engine matching result; Based on the standard four-dimensional risk parameter vector and the preset Bayesian network model, obtain the Bayesian network probability matching result; Based on the matching results of the rule engine and the probability matching results of the Bayesian network, the posterior probability of the patrol mode is calculated, and the posterior probability that meets the preset sorting conditions is determined as the appropriate patrol mode. Based on the adaptive patrol mode and dynamic risk profile data, generate patrol task work orders for the target area; Based on the target area patrol task work order and risk diffusion prediction results, a patrol route is generated.

[0050] In this embodiment, by extracting a standard four-dimensional risk parameter vector from dynamic risk profile data, the multi-source heterogeneous risk data in the dynamic risk profile can be structurally refined and integrated, eliminating differences in data format and dimension, and providing standardized and consistent data support for the matching of subsequent inspection modes. For example, equipment status S1: equipment risk level (0 low, 1 medium, 2 high, 3 extremely high), failure frequency, core component temperature (exceeding threshold = 1, not exceeding = 0); environmental risk S2: environmental risk level (0-3), harmful gas concentration (exceeding threshold = 1, not exceeding = 0), wind speed; human error S3: number of violations (0, 1-2, ≥3, corresponding to codes 0, 1, 2 respectively), frequency of not wearing protective equipment; management defects S4: number of rectification timeouts (0, 1, ≥2, corresponding to codes 0, 1, 2 respectively), inspection plan non-completion rate (non-completion rate ≥30% is coded as 1, otherwise it is 0), and integrated into an input vector. For example, the risk parameters of a certain scenario are "high equipment risk (S1=2), high environmental risk (S2=2), 2 violations (S3=1), 1 rectification timeout (S4=1)", which is converted into a vector [S1=2,S2=2,S3=1,S4=1]. Next, an IF-THEN risk matching rule base is constructed using a standard four-dimensional risk parameter vector and preset inspection modes. Specifically, the preset inspection modes include comprehensive inspections and specialized inspections: the core objective of comprehensive inspection M1 is to cover the entire risk domain and verify the basic risk model. It involves configuring a multi-sensor inspection robot (without AR assistance), and the triggering scenario features are low / medium risk, no single-dimensional threshold exceeding, and no associated risks, with a label code of 0. The core objective of specialized inspection M2 is to target and investigate high-risk points (high / extremely high) and trace the causal chain. It involves configuring AR glasses, a human team, and a specialized testing instrument, and the triggering scenario features are high / extremely high risk, single-dimensional threshold exceeding, or associated anomalies, with a label code of 1. For complex scenarios, sub-modes can be further divided into the two modes, such as M2-1 equipment failure special and M2-2 environmental risk special, with the label codes expanded to 0 (M1), 1 (M2-1), and 2 (M2-2).

[0051] IF-THEN Risk Matching Rule Base: A1: Priority 1, condition is S1=3 OR S2=3 OR (S1=2 and S2=2 and S3≥2), output M2 (Special Inspection), suitable for scenarios with clearly defined high risks and urgent investigations required; A2: Priority 2, condition is (S1=1 and S2=1 and S3=0 and S4=0) OR all dimensions ≤1, output M1 (Comprehensive Inspection), suitable for scenarios with controllable risks and full coverage required; A3: Priority 3, condition is S4≥2 and S1≤1 and S2≤1, output M2 (Special Inspection - Management), suitable for scenarios with management issues as the main focus and special verification processes required. This establishes the logical association between risk parameters and inspection modes, laying a structured rule foundation for rapid and accurate matching of inspection modes, and improving the logic and execution efficiency of mode matching. Then, by performing rule matching verification on the standard four-dimensional risk parameter vector and the IF-THEN risk matching rule base, when rules conflict, if multiple rules are satisfied simultaneously, the rule with higher priority is executed first to determine the initial inspection pattern of the risk scenario. For scenarios that meet the explicit rules, the rule engine matching result is directly output, which greatly improves the efficiency of pattern matching in conventional risk scenarios. If the priorities are the same, they are marked as rule ambiguity areas, and the standard four-dimensional risk parameter vector of the ambiguity or conflict risk scenario is handed over to the subsequent Bayesian network model for processing. Subsequently, the Bayesian network probability matching result is obtained through the standard four-dimensional risk parameter vector and the preset Bayesian network model, which solves the problem of judging ambiguity or conflict risk scenarios that the rule engine cannot handle, makes up for the limitations of pure rule matching, and improves the comprehensiveness and accuracy of pattern matching.

[0052] Next, the posterior probability of the inspection pattern is calculated using the matching results from the rule engine and the probability matching results from the Bayesian network. Specifically, the posterior probability of the output node M can be calculated using the junction tree algorithm through inference. ; Wherein, P(M) represents the prior probability of the inspection mode, P(A,B,C,D|M) represents the conditional probability (i.e., likelihood) of the combination of risk parameters under a certain inspection mode, and P(A,B,C,D) represents the marginal probability (i.e., normalization factor) of the combination of risk parameters. If the posterior probability of an inspection mode satisfies the preset sorting condition, the corresponding inspection mode is determined as the suitable inspection mode, ensuring that the suitable inspection mode matches the actual risk scenario and providing accurate mode basis for the generation of subsequent inspection task work orders. If the rule engine matching result is directly output, the posterior probability of the inspection mode corresponding to the rule engine matching result calculated by the preset Bayesian network model is used. If the posterior probability is ≥0.8, the result is confirmed; if the posterior probability is <0.8, manual review is required. The preset sorting condition is the maximum probability among the posterior probabilities of each inspection mode. After obtaining the appropriate patrol mode, the confidence level of the mode is tested. If the confidence level is greater than or equal to 0.6, it is output; otherwise, it is manually reviewed. The appropriate patrol mode is selected based on historical experience to avoid model misjudgment. Then, patrol task work orders for the target area are generated using the appropriate patrol mode and dynamic risk profile data. The work orders are presented in the form of structured fields and task lists, including basic identification information (work order ID, task number, target area, patrol mode, generation time), core task content (patrol scope, key verification items, associated risk points), resource allocation list (personnel, equipment, tools), constraints and requirements (time window, safety precautions, execution specifications), and acceptance and feedback (completion standards, feedback methods). By combining the core information in the dynamic risk profile, the patrol tasks can focus on high-risk areas and core hidden danger points, improving the targeting of the patrol tasks. Finally, by generating patrol paths based on reinforcement learning algorithms using target area patrol task work orders and risk diffusion prediction results, the patrol paths can prioritize covering high-risk predicted areas, avoiding over-patrolling of low-risk areas, effectively improving the execution efficiency of security patrols and achieving comprehensive security patrols. The patrols can then be completed according to the optimized patrol paths, and the patrol results can be fed back. Risk areas identified during the patrols can be manually reviewed, and the entire patrol process data can be recorded and managed. This completely solves the problems of low patrol efficiency, incomplete patrol path coverage, and inaccurate investigation caused by isolated data and delayed risk prediction.

[0053] Further, obtaining the Bayesian network probability matching result based on the standard four-dimensional risk parameter vector and the preset Bayesian network model includes: Obtain historical patrol data and the corresponding actual pattern data; An initial Bayesian network model is constructed based on a standard four-dimensional risk parameter vector, a preset inspection mode, and a causal relationship chain. Based on the historical patrol data and the initial Bayesian network model, predictive pattern data of the historical patrol data is obtained; The prediction consistency rate is calculated based on the predicted pattern data and the actual pattern data. The initial Bayesian network model is trained with the goal of maximizing the prediction consistency rate until the preset third convergence condition is reached, and the preset Bayesian network model is obtained. Based on the standard four-dimensional risk parameter vector and the preset Bayesian network model, the Bayesian network probability matching result is obtained.

[0054] In this embodiment, by acquiring historical patrol data and the corresponding actual pattern data, labeled sample data is provided to support the training and parameter optimization of the initial Bayesian network model. Next, an initial Bayesian network model is constructed using a standard four-dimensional risk parameter vector, a preset patrol pattern, and a causal relationship chain, providing a model framework with risk causal logic support for subsequent iterative training. Then, using the historical patrol data and the initial Bayesian network model, predicted pattern data from the historical patrol data is obtained, providing reference data for calculating the prediction consistency rate. Subsequently, the prediction consistency rate is calculated using the predicted pattern data and the actual pattern data, and the initial Bayesian network model is iteratively trained with the goal of maximizing the prediction consistency rate. This continuously improves the model's prediction accuracy and matching fit for patrol patterns until a preset third convergence condition is reached, resulting in a preset Bayesian network model. This solves the problem of large prediction bias and incompatibility with actual scenarios in the initial Bayesian network model, providing reliable and accurate model support for subsequent patrol pattern probability matching. The preset third convergence condition can be set to 85%. Finally, by using a standard four-dimensional risk parameter vector and a preset Bayesian network model, the posterior probability of each inspection mode is calculated. The mode with the highest probability is selected as the Bayesian network probability matching result. This provides a scientific probabilistic basis for determining inspection modes in fuzzy or conflicting risk scenarios that rule engines cannot handle, making up for the limitations of pure rule matching and improving the comprehensiveness, accuracy and adaptability of inspection mode matching.

[0055] See Figure 2 This embodiment also provides a security patrol path acquisition system based on dynamic risk profiling, including a causal relationship chain construction module, a temporal risk prediction module, a cell initial state assignment module, a dynamic risk profiling construction module, and a path optimization execution module, wherein: The causal relationship chain construction module is used to obtain standard four-dimensional risk association data of the target area and construct a causal relationship chain based on the standard four-dimensional risk association data; The time-series risk prediction module is used to obtain a comprehensive weight vector and construct a time-series risk prediction model based on standard four-dimensional risk correlation data, causal relationship chain and preset risk assessment model, and to obtain a set of time-series risk levels for the target area based on the time-series risk prediction model and standard four-dimensional risk correlation data. The cell initial state assignment module is used to generate a three-dimensional risk heat map and identify risk clustering areas and hidden danger heat cores based on the target area time-series risk level set and the preset target area map, and to perform multi-dimensional state assignment on the preset cell space based on the risk clustering areas, hidden danger heat cores and the three-dimensional risk heat map to obtain the cell initial state matrix. The dynamic risk profile construction module is used to obtain risk diffusion prediction results and dynamic risk profile data based on causal relationship chains, comprehensive weight vectors, standard four-dimensional risk association data, cell initial state matrices and preset cellular automata models. The path optimization execution module is used to generate target area patrol task work orders based on dynamic risk profile data, and generate patrol paths based on the target area patrol task work orders and risk diffusion prediction results.

[0056] It is understood that the above system item embodiments correspond to the method item embodiments of the present invention, and can realize the security patrol path acquisition method based on dynamic risk profile provided by any of the above method item embodiments of the present invention.

[0057] It should be noted that the system embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0058] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for obtaining security patrol paths based on dynamic risk profiles, characterized in that, include: Obtain standard four-dimensional risk correlation data for the target area, and construct a causal relationship chain based on the standard four-dimensional risk correlation data; Based on standard four-dimensional risk correlation data, causal relationship chain and preset risk assessment model, obtain comprehensive weight vector and construct time series risk prediction model, and obtain time series risk level set of target area based on time series risk prediction model and standard four-dimensional risk correlation data; Based on the target area time-series risk level set and the preset target area map, a three-dimensional risk heat map is generated and risk clustering areas and hidden danger heat cores are identified. Based on the risk clustering areas, hidden danger heat cores and the three-dimensional risk heat map, multi-dimensional state values ​​are assigned to the preset cell space to obtain the cell initial state matrix. Based on causal relationship chains, comprehensive weight vectors, standard four-dimensional risk association data, cellular initial state matrices, and preset cellular automata models, risk diffusion prediction results and dynamic risk profile data are obtained. Based on dynamic risk profile data, patrol task orders for target areas are generated, and patrol routes are generated based on the patrol task orders for target areas and the risk diffusion prediction results.

2. The method for obtaining security patrol paths based on dynamic risk profiles according to claim 1, characterized in that, The acquisition of standard four-dimensional risk correlation data for the target area, and the construction of a causal relationship chain based on the standard four-dimensional risk correlation data, includes: Acquire raw equipment status data, raw environmental risk data, raw human error data, and raw management defect data for the target area; The original equipment status data, original environmental risk data, original human error data, and original management defect data are sequentially subjected to conflict fusion, missing data filling, and noise filtering to obtain standard equipment status data, standard environmental risk data, standard human error data, and standard management defect data, which are then used to form standard four-dimensional risk correlation data. A causal relationship chain is constructed based on standard four-dimensional risk association data.

3. The method for obtaining security patrol paths based on dynamic risk profiles according to claim 2, characterized in that, The construction of a causal relationship chain based on standard four-dimensional risk association data includes: Extract several entities from standard four-dimensional risk association data, and extract several relationships between these entities; Based on the preset IF-THEN deterministic causal rule, several relationships are filtered to obtain the initial causal chain backbone; Based on several entities, several relationships, and a preset four-dimensional Bayesian network, the fault causal contribution of the relationships is calculated, and the relationships with a fault causal contribution greater than or equal to the preset fault causal contribution threshold are taken as the initial implicit causal relationships. Verify the initial implicit causal relationship through counterfactual reasoning to obtain the valid implicit causal relationship; The initial causal chain backbone and effective implicit causal relationships are filtered based on a preset causal chain length threshold and a preset path confidence threshold to obtain the standard causal chain backbone and standard implicit causal relationships. A causal relationship chain is constructed based on the standard causal chain backbone and the standard implicit causal relationship.

4. The method for obtaining security patrol paths based on dynamic risk profiles according to claim 1, characterized in that, The process involves obtaining a comprehensive weight vector based on standard four-dimensional risk correlation data, causal relationship chains, and a pre-defined risk assessment model, constructing a time-series risk prediction model, and obtaining a set of time-series risk levels for the target area based on the time-series risk prediction model and standard four-dimensional risk correlation data, including: Calculate the information entropy for the standard four-dimensional risk correlation data to obtain the objective weight vector; Based on the preset subjective weight vector, preset dynamic adjustment factor and objective weight vector, a comprehensive weight vector is calculated, and the weight is adapted based on the comprehensive weight vector and the preset risk assessment model to obtain a standard risk assessment model. A time-series risk prediction model is constructed based on causal relationship chains and standard risk assessment models, and a set of time-series risk levels for the target area is obtained based on the time-series risk prediction model and standard four-dimensional risk correlation data.

5. The method for obtaining security patrol paths based on dynamic risk profiles according to claim 4, characterized in that, The process involves constructing a time-series risk prediction model based on causal chains and standard risk assessment models, and obtaining a set of time-series risk levels for the target area based on the time-series risk prediction model and standard four-dimensional risk correlation data, including: An initial model for time-series risk prediction was constructed based on causal chains and standard risk assessment models. Obtain historical time-series risk data and actual risk level data corresponding to the historical time-series risk data; Based on historical time-series risk data and an initial time-series risk prediction model, risk level prediction data is obtained. The risk level error is calculated based on the risk level prediction data and the actual risk level data. The initial model for time series risk prediction is trained with the goal of minimizing the risk level error until the preset first convergence condition is met, and the time series risk prediction model is obtained. The time-series risk level set of the target area is obtained based on the time-series risk prediction model and standard four-dimensional risk correlation data.

6. The method for obtaining security patrol paths based on dynamic risk profiles according to claim 1, characterized in that, The process involves generating a 3D risk heatmap based on a time-series risk level set of the target area and a preset target area map, identifying risk clusters and hazard heat cores, and assigning multi-dimensional state values ​​to a preset cellular space based on the risk clusters, hazard heat cores, and the 3D risk heatmap to obtain an initial state matrix for the cells, including: Based on the time-series risk level set of the target area and the preset target area map, a three-dimensional risk heat map is generated and risk clustering areas and hidden danger heat cores are identified; The preset cell space is divided based on the preset grid granularity to obtain several cell grids, and the preset target area map and cell grids are bound together to obtain the target area grid distribution map. Based on the aforementioned risk clustering area, hidden danger thermal core, and three-dimensional risk heat map, multi-dimensional state assignment is performed on the grid distribution map of the target area to obtain the initial state matrix of the cells.

7. The method for obtaining security patrol paths based on dynamic risk profiles according to claim 1, characterized in that, The standard four-dimensional risk association data includes standard environmental risk data; the acquisition of risk diffusion prediction results and dynamic risk profile data based on causal relationship chains, comprehensive weight vectors, standard four-dimensional risk association data, cellular initial state matrices, and preset cellular automata models includes: Based on the causal relationship chain, standard four-dimensional risk association data are filtered to obtain a standard indicator set of diffusion characteristics; Based on the standard index set of diffusion characteristics and the comprehensive weight vector, a comprehensive diffusion score set is obtained, and the comprehensive diffusion score set is mapped to a diffusion potential level set according to a preset interval. Based on the diffusion potential level set and the preset dynamic neighbor rules, standard environmental risk data are matched to obtain a cell neighbor weight matrix; Based on the cell initial state matrix, cell neighbor weight matrix, preset cell state transition triggering conditions, standard four-dimensional risk association data and comprehensive weight vector, obtain the cell state transition probability set; Based on the cell initial state matrix, cell state transition probability set, standard four-dimensional risk association data, and preset cellular automata model, risk diffusion prediction results and dynamic risk profile data are obtained.

8. The method for obtaining security patrol paths based on dynamic risk profiles according to claim 7, characterized in that, The process of obtaining risk diffusion prediction results and dynamic risk profile data based on the cell initial state matrix, cell state transition probability set, standard four-dimensional risk association data, and preset cellular automata model includes: Obtain historical causes of risk diffusion and corresponding risk diffusion data for those causes. Based on historical causes of risk diffusion and a pre-set cellular automata initial model, simulated diffusion data is obtained. The diffusion error is calculated based on simulated diffusion data and risk diffusion data. The initial cellular automaton model is trained with the goal of minimizing the diffusion error until the preset second convergence condition is met, and the preset cellular automaton model is obtained. Based on the cell initial state matrix, cell state transition probability set, standard four-dimensional risk association data, and preset cellular automata model, risk diffusion prediction results and dynamic risk profile data are obtained.

9. The method for obtaining security patrol paths based on dynamic risk profiling according to claim 1, characterized in that, The process of generating target area patrol task orders based on dynamic risk profile data, and generating patrol paths based on target area patrol task orders and risk diffusion prediction results, includes: Based on dynamic risk profile data, extract a standard four-dimensional risk parameter vector; Based on the standard four-dimensional risk parameter vector and the preset inspection mode, an IF-THEN risk matching rule base is constructed. Perform rule matching and verification processing on the standard four-dimensional risk parameter vector and the IF-THEN risk matching rule base to obtain the rule engine matching result; Based on the standard four-dimensional risk parameter vector and the preset Bayesian network model, obtain the Bayesian network probability matching result; Based on the matching results of the rule engine and the probability matching results of the Bayesian network, the posterior probability of the patrol mode is calculated, and the posterior probability that meets the preset sorting conditions is determined as the appropriate patrol mode. Based on the adaptive patrol mode and dynamic risk profile data, generate patrol task work orders for the target area; Based on the target area patrol task work order and risk diffusion prediction results, a patrol route is generated.

10. A method for obtaining security patrol paths based on dynamic risk profiles according to claim 9, characterized in that, The process of obtaining Bayesian network probability matching results based on the standard four-dimensional risk parameter vector and the preset Bayesian network model includes: Obtain historical patrol data and the corresponding actual pattern data; An initial Bayesian network model is constructed based on a standard four-dimensional risk parameter vector, a preset inspection mode, and a causal relationship chain. Based on the historical patrol data and the initial Bayesian network model, predictive pattern data of the historical patrol data is obtained; The prediction consistency rate is calculated based on the predicted pattern data and the actual pattern data. The initial Bayesian network model is trained with the goal of maximizing the prediction consistency rate until the preset third convergence condition is reached, and the preset Bayesian network model is obtained. Based on the standard four-dimensional risk parameter vector and the preset Bayesian network model, the Bayesian network probability matching result is obtained.

Images