Hazardous chemical substance storage accident risk analysis method, system and equipment and storage medium

Abstract

The invention relates to the technical field of hazardous chemical substance safety management and control, and particularly provides a hazardous chemical substance storage accident risk analysis method, which comprises the following steps: collecting and preprocessing time sequence data and spatial topological data of a storage area; inputting the time sequence data into a time dynamic risk prediction model, and extracting time sensitive features to output a risk level time sequence curve with a risk storage unit label; labeling a high-risk period window in the curve; constructing a spatial diffusion risk assessment model based on the spatial topological data, extracting risk diffusion features of a target storage unit, associating adjacent units, and outputting a dynamic risk spatial distribution result; performing space-time coupling analysis on the high-risk time period and the spatial distribution, and outputting early warning information after correction; and finally, calling the management and control rule base to generate a risk management and control decision instruction. According to the method, time-space two dimensions are covered, dynamic risk assessment, early warning and management and control are realized, the comprehensiveness and management and control accuracy of hazardous chemical substance storage risk analysis are effectively improved, and a scientific support is provided for safety management of a storage area.

Description

Technical Field

[0001] The present invention relates to the technical field of hazardous chemical safety management and control, and in particular to a method and system for analyzing the risk of hazardous chemical storage accidents. Background Art

[0002] Hazardous chemicals (such as gasoline, diesel, and corrosive chemicals) are flammable, explosive, toxic, and harmful. Their storage is a core scenario for safety management. The operating status of storage units (tanks, warehouses), environmental changes, and spatial layout are all related and may cause major accidents such as leakage, explosion, and poisoning. Scientific risk analysis methods are needed to identify risks in advance and formulate control measures to avoid accidents or reduce the consequences.

[0003] Currently, mainstream analysis methods in the industry are mainly divided into two categories: one is the traditional static assessment method represented by the Hazard and Operability Analysis (HAZOP) and static risk matrix. By pre-setting risk factors (such as storage medium toxicity and tank pressure) and fixed weights, the risk level of the storage condition at a specific moment is determined and a static result of "high / medium / low" is output; the other is a single-dimensional dynamic method, such as using conventional LSTM to predict the time series changes of the risk of a single tank and using CFD to simulate the spatial diffusion range of a single leak. Such methods only focus on one dimension of "time" or "space" and do not form a linkage.

[0004] Although existing methods have certain applications, they are limited by the "static and isolated" design logic and cannot adapt to the "dynamic evolution and spatiotemporal linkage" characteristics of risks in actual storage scenarios: in the time dimension, the results are output only based on the operating data at a certain moment, ignoring the differences in day and night inspections and the risk accumulation patterns caused by seasonal humidity changes, failing to capture the "time-sensitive effect", and prone to misjudgment; in the spatial dimension, for storage areas with multi-unit cluster layouts, the risks of individual units are assessed in isolation, ignoring the chain effect of the spread of hazardous substances, resulting in "fragmented" results; at the spatiotemporal coordination level, time and space analysis are separated, and the "when + where" coupled risk information cannot be output, which in turn causes delayed warnings and insufficient targeted management and control, making it difficult to effectively prevent accidents.

[0005] In summary, the current field of hazardous chemical storage accident risk analysis urgently needs technical solutions that break through the limitations of static assessment and isolated dimensions to solve the above problems. Summary of the Invention

[0006] In order to overcome the deficiencies of the prior art, the present invention provides a method and system for analyzing the risk of hazardous chemical storage accidents to solve the problems in the prior art.

[0007] One embodiment of the present invention provides a method for analyzing the risk of hazardous chemical storage accidents, comprising the following steps: Collecting and preprocessing multi-source data on hazardous chemical storage in the storage area, wherein the multi-source data includes time series data and spatial topology data; Input the preprocessed time series data into a pre-built time dynamic risk prediction model, extract the time-sensitive features in the time series data through the time dynamic risk prediction model, output a risk level time series curve for a preset time period in the future based on the time sensitive features, and associate and mark the storage units in the risk level time series curve that are within the preset risk threshold range; Mark the high-risk time period window in the risk level time series curve where the risk level exceeds the preset safety threshold; A spatial diffusion risk assessment model is constructed based on the preprocessed spatial topological data. When the risk level time series curve is associated with a storage unit within the preset risk threshold range, the storage unit is used as the target storage unit. The risk diffusion characteristics of the target storage unit are extracted through the spatial diffusion risk assessment model and associated with the adjacent storage units, and the dynamic risk spatial distribution results of the storage area are output; Conduct spatiotemporal coupling analysis on the high-risk period window and the dynamic risk spatial distribution results, modify the analysis results based on the preset risk association rules, and output spatiotemporal coupling risk warning information; Based on the spatiotemporal coupling risk warning information, a preset risk management and control rule library is called to generate risk management and control decision instructions for hazardous chemicals storage areas.

[0008] By adopting the above scheme, the time dynamic risk prediction model is used to output the risk level time series curve with risk storage unit labels and high-risk periods, which can capture the time evolution law of risks and solve the problem that static assessments are difficult to reflect dynamic accumulation and easy to misjudge; then the spatial diffusion risk assessment model is used to analyze the diffusion trend of risk storage units and associate them with adjacent storage units to cover the overall risk of the storage area, and solve the problem that isolated assessments ignore chain effects and fragmented results; finally, through spatiotemporal coupling analysis, early warnings are output and control instructions are generated to achieve spatiotemporal risk coordination, solve the problems of spatiotemporal fragmentation and early warning lag in existing methods, and overall improve the dynamics, comprehensiveness and control accuracy of hazardous chemicals storage risk analysis, providing scientific support for safety management.

[0009] In one embodiment, the construction of the time dynamic risk prediction model specifically includes the following steps: Obtain historical time series sample data in the storage area, including storage unit operating condition data, environmental dynamic data, and human intervention data; pre-process the historical time series sample data to obtain a standardized time series feature set; and label the standardized time series feature set with a risk level label based on historical accident data of the same type of storage units within a preset time period, where pre-accident data is labeled as high risk and normal operating data is labeled as low risk; Constructing a time series prediction model including a feature extraction layer for extracting time-sensitive features and a prediction output layer for outputting a risk level time series curve; The standardized time series feature set marked with risk level labels is input into the time series prediction model architecture, and the model parameters are adjusted using the preset training strategy to obtain the trained time dynamic risk prediction model; The preprocessed time series data is input into the trained time dynamic risk prediction model, and the time-sensitive features in the time series data are extracted through the feature extraction layer to form a sensitive feature vector set containing storage unit operating condition characteristics, environmental dynamic characteristics and human intervention characteristics. The sensitive feature vector set is input into the prediction output layer, and the risk level time series curve of the future preset time length is output through the prediction output layer.

[0010] By adopting the above scheme, the training basis is constructed with three types of historical time series sample data and risk level labels, which can solve the prediction bias problem caused by one-sided model training data and fuzzy labels; then the long short-term memory network architecture integrated with the attention mechanism is used to extract key time-sensitive features and control the training accuracy, solving the problems of key feature omission and overfitting failure of conventional time series models; finally, through the output of the structured sensitive feature vector set, the problems of feature extraction black box and spatiotemporal coupling input bias are solved, thereby improving the overall accuracy, reliability and scenario adaptability of time dynamic risk prediction, and providing more refined time dimension support for hazardous chemicals storage risk analysis.

[0011] In one embodiment, the step of inputting the sensitive feature vector set into the prediction output layer and outputting the risk level time series curve of the future preset time period through the prediction output layer specifically includes the following steps: The sensitive feature vector set is associated with the hazardous chemical type attribute of the corresponding storage unit to form an associated data set. Based on the associated data set, the sensitive feature vector set is grouped by hazardous chemical type, and the sensitive feature vectors in each group are sorted by time series to form a time series feature matrix stratified by hazardous chemical risk characteristics. The dynamic correlation relationship between adjacent vectors in each hierarchical time series feature matrix is ​​analyzed using a preset time series correlation algorithm. Based on the attention mechanism, accident-related feature vectors that match historical accident features are assigned a preset high-impact weight value, and other non-accident-related auxiliary feature vectors are assigned a preset basic weight value; wherein the historical accident features are extracted based on historical accident data of the same type of storage units within a preset time period; Based on the preset hazardous chemical type risk level mapping rules, combined with the risk level trend correction factor, the high impact weight value of the accident-related feature vector and the basic weight value of the auxiliary feature vector, the risk level value of each period within the preset time period in the future is calculated; The risk level values ​​of each time period are continuously arranged along the time axis to generate a risk level time series curve, wherein the risk level time series curve includes labels of storage units within a preset risk threshold range and spatially associated units of the storage units.

[0012] By adopting the above scheme, the sensitive feature vector set is associated with the hazardous chemical type attributes, grouped and time-series sorted to form a hierarchical matrix, which can solve the problems of mixed analysis of different hazardous chemical characteristics and masking of key risk characteristics in the existing prediction output layer; then the dynamic correlation relationship is analyzed by the time series association algorithm, and the accident-related feature vector and the auxiliary feature vector are differentially weighted through the attention mechanism. The risk level is calculated by combining the hazardous chemical type-specific mapping rules and the time series trend correction factor, which solves the problems of subjective weight setting in the conventional output layer, undifferentiated risk calculation standards and gradual risk omission; finally, by generating a risk level time series curve containing high-risk storage units and spatial association unit labels, the problem that the prediction output only reflects the risk in the time dimension and ignores the spatial chain effect, resulting in insufficient prevention and control targeting, is solved, and the accuracy, information integrity and prevention and control guidance of the risk level time series curve output are improved as a whole, providing more targeted technical support for the prediction output link of hazardous chemical storage risk analysis.

[0013] In one embodiment, the method for determining the preset risk threshold comprises the following steps: Based on the hazardous chemical type attribute of the storage unit, the basic safety threshold of the corresponding type of hazardous chemicals is retrieved as the initial risk threshold; Extract the risk level critical value of the corresponding type of hazardous chemicals when the historical accident occurred from the historical accident characteristics, and modify the initial risk threshold based on the risk level critical value to obtain the type-specific benchmark threshold; Real-time dynamic impact factors are extracted from the sensitive feature vector set, and the type-specific benchmark threshold is dynamically adjusted based on the extracted real-time dynamic impact factors to determine the preset risk threshold of the corresponding storage unit.

[0014] By adopting this solution and taking the basic safety threshold corresponding to the type attribute of hazardous chemicals as the initial standard, the problem of misjudgment caused by the disconnection between the existing preset risk threshold and the characteristics of hazardous chemicals and excessive universality can be solved; combined with the historical accident risk critical value correction benchmark of the same type of hazardous chemicals, the problem of lack of actual accident data support in threshold setting can be solved; finally, the dynamic influencing factor of the sensitive feature vector is introduced for real-time adjustment to solve the problem that the static threshold cannot adapt to real-time working condition changes, and the overall pertinence, scientificity and dynamic adaptability of the preset risk threshold are improved, providing an accurate judgment basis for marking high-risk storage units on the risk level time series curve.

[0015] In one embodiment, the step of constructing a spatial diffusion risk assessment model based on preprocessed spatial topological data, when a risk level time series curve is associated with a storage unit within a preset risk threshold range, taking the storage unit as the target storage unit, extracting the risk diffusion characteristics of the target storage unit through the spatial diffusion risk assessment model and associating it with adjacent storage units, and outputting the dynamic risk spatial distribution result of the storage area, specifically includes the following steps: The pre-processed spatial topology data is analyzed hierarchically to extract the spatial correlation parameters and physical barrier levels between the target storage unit and adjacent storage units. Based on the diffusion characteristic parameters of hazardous chemicals in the target storage unit, a spatial diffusion risk assessment model is constructed. When the risk level time series curve is associated with a storage unit within the preset risk threshold range, the real-time risk level value of the target storage unit is obtained and input into the spatial diffusion risk assessment model. The risk diffusion characteristics of the target storage unit are extracted through the spatial diffusion risk assessment model. The risk diffusion characteristics include the risk diffusion rate, diffusion range radius and hazardous substance concentration attenuation coefficient of the target storage unit; Based on the extracted risk diffusion characteristics and the spatial association parameters between the target storage unit and the adjacent storage units, the risk diffusion characteristics of the target storage unit are associated with the corresponding adjacent storage units according to the preset association rules. The risk level of each adjacent storage unit affected by the diffusion is calculated, and an association relationship table containing the target storage unit, adjacent storage units, diffusion paths, and affected risk levels is generated. Based on the risk change trend of the risk level time series curve, the association table is superimposed and matched with the spatial map of the storage area to generate and output the dynamic risk spatial distribution results of the storage area that are dynamically updated over time. The dynamic risk spatial distribution results include the actual risk level of each storage unit in different time periods and the diffusion impact range labeling.

[0016] By adopting the above scheme, a spatial diffusion risk assessment model that couples spatial association and diffusion characteristics is constructed using hierarchical analytical spatial topological data. The risk diffusion rate, range radius, and concentration attenuation coefficient are extracted as quantitative features. This can address the weak assessment foundation caused by the existing spatial diffusion model's generality (relying only on unit distance and ignoring connectivity and physical barriers) and abstract diffusion characteristics (lack of clear quantitative parameters). Adjacent units are then associated according to the diffusion range coverage priority + connectivity influence weighted rule, and the affected risk level is calculated using the concentration attenuation coefficient. This addresses the existing problems of inaccurate association and grade deviation caused by the rough association of adjacent units (selecting only by distance) and the lack of quantitative basis for risk level calculation. Finally, the spatial distribution results are dynamically updated using the risk level time series curve. This addresses the problem of dynamic control lag caused by the inability of existing static spatial assessments to adapt to real-time changes in operating conditions (such as risk escalation / attenuation). This overall improves the accuracy, dynamic adaptability, and information integrity of the dynamic spatial distribution of risks in storage areas, providing precise spatial dimension support for the positioning of high-risk areas, the allocation of emergency resources, and the formulation of real-time prevention and control strategies.

[0017] In one embodiment, the step of extracting the risk diffusion characteristics of the target storage unit using the spatial diffusion risk assessment model further includes the following steps: Obtain the physical properties of the target storage unit and the type of hazardous chemicals inside it, and determine whether the target storage unit belongs to a special scenario based on a preset judgment threshold; When the target storage unit belongs to a special scene, calling a preset special scene correction coefficient table to match the correction coefficient corresponding to the target storage unit, wherein the special scene correction coefficient table includes correction coefficients corresponding to several special scenes; The extracted risk diffusion feature is dynamically modified according to a modification coefficient corresponding to the target storage unit to obtain a modified risk diffusion feature.

[0018] By adopting the above scheme, based on the physical properties of the target storage unit and the type of hazardous chemicals, combined with the preset judgment threshold to identify special scenarios, it is possible to solve the problem that the existing spatial diffusion assessment has no clear definition for special scenarios (such as low-temperature / high-pressure units, hazardous chemicals that are prone to spontaneous combustion / highly corrosive) and is only evaluated according to general rules, resulting in insufficient scenario adaptability. Then, by calling the preset special scenario correction coefficient table to match the corresponding correction coefficient, the problem of no quantitative basis for risk diffusion feature correction in special scenarios and subjective correction parameters is solved. Finally, the risk diffusion characteristics are dynamically corrected according to the matched correction coefficient to solve the problem of assessment distortion caused by the use of the same diffusion characteristic parameters for special and conventional scenarios (such as overestimation of the diffusion range of low-temperature units and misjudgment of the concentration attenuation of hazardous chemicals that are prone to spontaneous combustion). This overall improves the accuracy of risk diffusion feature extraction and scenario adaptability in special scenarios, provides technical support that is more in line with actual working conditions for hazardous chemical storage space diffusion risk assessment, and avoids risk assessment deviations and subsequent improper management and control caused by misjudgment of special scenarios.

[0019] In one embodiment, the step of performing spatiotemporal coupling analysis on the high-risk time window and the dynamic risk spatial distribution results, correcting the analysis results based on preset risk association rules, and outputting spatiotemporal coupling risk warning information specifically includes the following steps: Extracting time boundary parameters of the high-risk period window, the time boundary parameters including the high-risk starting time, duration and risk level peak time; Map the time boundary parameters of the high-risk period window to the dynamic risk spatial distribution results in the space-time dimension, match the risk spatial distribution state of the storage area corresponding to each time node, and form an initial space-time coupling matrix. The initial space-time coupling matrix contains the associated data of time nodes, storage units, risk levels, and diffusion ranges. Modifying the initial spatiotemporal coupling matrix based on preset risk association rules, wherein the preset risk association rules include a time priority weight rule, a spatial correlation rule, and a risk superposition rule; Based on the revised space-time coupling matrix, space-time coupling risk warning information is generated and output.

[0020] By adopting the above scheme, the time boundary parameters (start time, duration, peak time) of the high-risk period window are taken as time anchor points, which can solve the problem of insufficient spatio-temporal correlation accuracy caused by the fuzzy description of the high-risk period (only defining the period range without key nodes); then, the initial coupling matrix containing time-space-risk-diffusion data is constructed through spatio-temporal dimension mapping, which solves the problem of weak coupling analysis basis caused by the split of time and space risk data (each is independently presented without explicit corresponding relationship); finally, the matrix is revised based on the time priority weight, spatial correlation and risk superposition rule and the early warning information is output, which solves the problem of poor pertinence of early warning and insufficient guidance of control caused by the simple splicing of data in the conventional spatio-temporal coupling (without distinguishing risk priority and actual harm difference), and improves the accuracy, priority distinguishing ability and early warning practicability of spatio-temporal coupling risk analysis as a whole, providing decision support with time sensitivity and spatial targeting for dynamic risk collaborative control of hazardous chemical storage area.

[0021] The application also relates to a hazardous chemical storage accident risk analysis system, comprising: a data acquisition module configured to acquire and preprocess hazardous chemical storage multi-source data of a storage area, wherein the hazardous chemical storage multi-source data comprises time series data and spatial topology data; a risk prediction module configured to input the preprocessed time series data into a pre-constructed time dynamic risk prediction model, extract time sensitive features in the time series data through the time dynamic risk prediction model, output a risk level time series curve of a future preset time length according to the time sensitive features, and associate and label storage units in the risk level time series curve located in a preset risk threshold range; a labeling module configured to label a high-risk period window in the risk level time series curve whose risk level exceeds a preset safety threshold; a risk assessment module configured to construct a spatial diffusion risk assessment model based on the preprocessed spatial topology data, take a storage unit as a target storage unit when the risk level time series curve is associated with a storage unit located in the preset risk threshold range, extract risk diffusion features of the target storage unit through the spatial diffusion risk assessment model and associate the risk diffusion features to adjacent storage units, and output a dynamic risk spatial distribution result of the storage area; a coupling analysis module configured to perform spatio-temporal coupling analysis on the high-risk period window and the dynamic risk spatial distribution result, revise the analysis result based on a preset risk correlation rule, and output spatio-temporal coupling risk early warning information; a decision generation module configured to call a preset risk control rule library based on the spatio-temporal coupling risk early warning information, and generate a risk control decision instruction for the hazardous chemical storage area.

[0022] The present application also relates to a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned method for analyzing the risk of hazardous chemical storage accidents when executing the computer program.

[0023] The present application also relates to a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps of the above-mentioned method for analyzing the risk of hazardous chemical storage accidents.

[0024] The hazardous chemicals storage accident risk analysis method and system provided in the above embodiments have the following beneficial effects: By using a time-dynamic risk prediction model to output a time-series curve of risk levels with risk storage unit labels and high-risk periods, the temporal evolution of risks can be captured, solving the problem that static assessments are difficult to reflect dynamic accumulation and prone to misjudgment. The spatial diffusion risk assessment model is then used to analyze the diffusion trend of risk storage units and link them to adjacent storage units to cover the overall risk of the storage area, solving the problem of isolated assessments ignoring chain effects and fragmented results. Finally, through spatiotemporal coupling analysis, early warnings are output and control instructions are generated to achieve spatiotemporal risk coordination, solve the problems of spatiotemporal fragmentation and early warning lag in existing methods, and overall improve the dynamics, comprehensiveness and control accuracy of hazardous chemical storage risk analysis, providing scientific support for safety management. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 A flowchart of a method for analyzing the risk of hazardous chemical storage accidents provided by an embodiment of the present invention; Figure 2 A block diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings.

[0027] Reference Figure 1 One embodiment of the present invention provides a method for analyzing the risk of hazardous chemical storage accidents, comprising the following steps: S10. Collect and pre-process multi-source data on hazardous chemical storage in the storage area, wherein the multi-source data includes time series data and spatial topology data.

[0028] In this embodiment, the core objective is to provide complete, standardized foundational data for subsequent temporal dynamic risk prediction and spatial diffusion risk assessment, avoiding subsequent analysis biases due to missing data or disorganized formats. For data acquisition, considering the explosion and corrosion protection requirements of hazardous chemical storage areas, an explosion-proof IoT sensor array (such as flameproof temperature sensors, intrinsically safe pressure transmitters, and electrochemical gas concentration sensors) is used to collect real-time time series data. Core data includes storage unit operating parameters (such as temperature, pressure, and liquid level), environmental parameters (such as temperature and humidity, and toxic / flammable gas concentrations), and personnel inspection records. Spatial topological data is obtained through CAD drawing analysis or on-site distance measurement. Core data includes storage unit locations, unit spacing, connectivity and separation relationships such as pipelines and firebreaks, and the distribution of surrounding sensitive targets (such as water sources and residential areas). For data preprocessing, conventional denoising, missing value imputation, and format standardization are performed on the time series data (which can be achieved by conventional means by those skilled in the art). Spatial topological data is deduplicated and standardized into a format suitable for spatial analysis. The preprocessed time series data are used for subsequent temporal dynamic risk prediction, and the preprocessed spatial topology data are used for subsequent spatial diffusion risk assessment.

[0029] S20. Input the preprocessed time series data into a pre-built time dynamic risk prediction model, extract the time sensitive features in the time series data through the time dynamic risk prediction model, output the risk level time series curve of the future preset time length according to the time sensitive features, and associate and mark the storage units in the risk level time series curve that are within the preset risk threshold range.

[0030] In this embodiment, the core goal is to dynamically predict future risks in storage areas based on standardized time series data. This provides a time-based risk benchmark for subsequent high-risk period identification and spatial diffusion analysis, avoiding the risk prediction lags caused by relying solely on historical static data. For input data, the preprocessed time series data from S10 (including storage unit operating parameters, environmental parameters, and personnel inspection records) is used. Data format verification ensures that it matches the input interface of the time-dynamic risk prediction model, allowing direct use without secondary conversion.

[0031] When the model is used to automatically analyze time series data, time-sensitive features that are strongly related to risk evolution are extracted as the focus (such as hourly rising trend of temperature parameters, abnormal fluctuation frequency of gas concentration, and duration of pressure exceeding the standard, etc.); based on the extracted features, the model outputs a risk level time series curve for a future preset time period (the preset time period can be flexibly set according to actual management needs, such as 24 hours for daily monitoring and 72 hours for major holidays). The curve contains double key information: one is the overall risk level of the storage area at each time node (reflecting the overall safety state of the area), and the other is the individual risk level of each storage unit at each time node (locating specific high-risk units), for example: if the preset time period is 24 hours and the time nodes are divided by 1 hour, the curve will show that at 10:00, the overall risk level of the area is grade II (medium risk), among which unit A (storing ethanol) has a risk level of grade III (high risk), and unit B (storing sodium hydroxide) has a risk level of grade I (low risk), and at 15:00, the overall risk level of the area rises to grade III, unit A has a risk level of grade IV (high risk), and unit B still maintains grade I.

[0032] At the same time, according to the preset risk threshold (such as setting "risk level ≥ grade III" as the high-risk judgment standard), the storage units with risk level exceeding the standard are marked by special markers (such as red dots and bold lines) in the risk level time series curve, and the distribution characteristics of high-risk units at different time nodes are clearly presented.

[0033] It should be noted that the risk level judgment rules and preset thresholds (such as the quantitative index boundaries of grades I-IV and the high-risk judgment standard "≥ grade III") can be flexibly set by those skilled in the art according to the actual operation of the storage area (such as the type of hazardous chemicals, the distribution of surrounding sensitive targets, and the requirements of enterprise safety management). Such adaptive adjustments are routine technical means and are not rigidly limited here.

[0034] The output risk level time series curve and high-risk unit marking information will be directly used in subsequent S30 (high-risk period window marking) to "lock the time interval where the high-risk set is concentrated", and in S40 (spatial diffusion risk assessment) to "determine the target storage units that need to be analyzed", providing a unified time dimension benchmark for cross-link risk analysis.

[0035] S30, marking the high-risk period window in the risk level time series curve where the risk level exceeds the preset safety threshold.

[0036] In this embodiment, the core purpose is to accurately locate the time interval of risk exceeding from the risk level time sequence curve, to provide clear high-risk time boundary for subsequent space-time coupling analysis, and to avoid insufficient risk control due to ambiguous time dimension. For input data, the risk level time sequence curve (including overall and individual risk level) output by S20 is used as the judgment basis combined with the preset safety threshold.

[0037] In specific operation, through the risk level change trend of the time sequence curve, the time interval continuously or intermittently exceeding the preset safety threshold is identified: if the risk level of a single time node exceeds the threshold and the adjacent nodes (such as within 1 hour before and after) continuously exceed the threshold, they are merged into a continuous high-risk period window; if there are multiple independent exceeding intervals (such as 9-11 am and 15-17 pm), they are marked as independent windows respectively. For example, based on the preset threshold of "risk level ≥ level III", two high-risk period windows "8:00-10:30 (region overall and unit A continuously level III)" and "14:00-16:00 (region overall level II but unit A level IV)" are identified from the 24-hour time sequence curve output by S20, and each window is marked with the starting time, ending time and core exceeding unit in the corresponding period. The output high-risk period window information will be directly used as the key input for time dimension mapping in S50 (space-time coupling analysis) to match the dynamic risk spatial distribution state in the corresponding period and provide time coordinate support for accurately locking the high-risk space-time overlapping area.

[0038] S40, based on the pre-processed spatial topology data, a spatial diffusion risk assessment model is constructed. When the risk level time sequence curve is associated with a storage unit located within the preset risk threshold range, the storage unit is taken as a target storage unit, the risk diffusion characteristics of the target storage unit are extracted through the spatial diffusion risk assessment model and associated with adjacent storage units, and the dynamic risk spatial distribution result of the storage area is output.

[0039] In this embodiment, the core purpose is to convert the high-risk unit identified in the time dimension into spatial diffusion risk analysis, to construct the association logic of "unit risk-space diffusion", and to provide the risk distribution benchmark in the spatial dimension for subsequent space-time coupling analysis, avoiding the omission of prevention and control range caused by focusing only on single unit risk and ignoring the surrounding chain effect. For input data, on the one hand, the pre-processed spatial topology data (including storage unit position, unit distance, connected and separated relationship of pipeline / fire dike, surrounding sensitive target distribution) in S10 is used as the spatial basis for model construction; on the other hand, the risk level time sequence curve output by S20 is combined to take the storage units (such as unit A and unit C with risk level ≥ level III) located within the preset risk threshold range as target storage units, and the analysis object is clear.

[0040] The specific operation consists of two steps: First, a spatial diffusion risk assessment model is constructed based on spatial topological data. The model integrates unit physical properties (such as protective facility type), hazardous chemical diffusion characteristics (such as volatilization rate), and spatial correlations (such as unit spacing and barrier conditions) to provide a computational framework for diffusion feature extraction. Second, the model extracts risk diffusion characteristics of the target storage unit (such as diffusion rate, diffusion range radius, and hazardous substance concentration attenuation coefficient). These characteristics are then associated with adjacent storage units using the principle of "diffusion range coverage priority + connectivity impact weighting." For example, if target unit A is an ethanol tank area with a diffusion range radius of 50 meters, the model will associate this diffusion characteristic with units B and D, which are within 50 meters and unobstructed by firebreaks, and calculate the risk level of units B and D based on the concentration attenuation coefficient.

[0041] The final output, the dynamic spatial distribution of storage area risk, is a spatial map containing the target storage unit location, the risk level of adjacent units affected by the diffusion, diffusion path annotations (e.g., diffusion direction from "unit A to unit B"), and diffusion range boundaries. For example, the map will highlight target unit A in red, while affected units B (risk level II) and D (risk level III) will be highlighted in yellow. A 50-meter diffusion range will be outlined with a dashed line. This result will be directly used in the S50 (temporal-spatial coupling analysis) and matched with the high-risk time window (S30 annotations) to form the foundation for a three-dimensional "time-space-risk" analysis.

[0042] S50: Perform spatiotemporal coupling analysis on the high-risk period window and the dynamic risk spatial distribution results, modify the analysis results based on preset risk association rules, and output spatiotemporal coupling risk warning information.

[0043] In this embodiment, the core goal is to break the disconnect between the temporal dimension (high-risk periods) and the spatial dimension (diffusion distribution), enabling a three-dimensional collaborative analysis of "when, where, and what kind of risk." This avoids the omission of critical overlapping risks (such as the overlapping areas of high-risk periods and high diffusion ranges) in early warnings due to analyzing the temporal and spatial dimensions separately, providing a precise basis for risk location for the subsequent generation of control instructions. For input data, the high-risk period window (including start / end times and core exceeding units) output by S30 and the dynamic risk spatial distribution results (including unit location, diffusion range, and affected risk level) output by S40 are directly called, ensuring that the two types of data are accurately linked through the "storage unit ID" and "time node."

[0044] The specific operation is divided into two steps: the first step is the spatiotemporal coupling analysis, mapping and matching each high-risk time window with the dynamic risk spatial distribution of the corresponding time node; for example, the 8:00-10:30 (unit A high risk) time period marked in S30 corresponds to the spatial status of unit A with a diffusion range of 50 meters, unit B (risk level II affected by diffusion), and unit D (risk level III affected by diffusion) in the 8:00-10:30 time period in S40, forming an initial coupling matrix of "time-space-risk level", which clearly presents the differences in risk spatial distribution in different time periods. The second step is to modify the analysis results based on the preset risk association rules. The preset rules are set around the "risk superposition effect", such as: time priority weight rule (the peak risk level moment during the high-risk period is marked as one level higher in the corresponding spatial risk level), spatial association rule (if the unit affected by the diffusion is close to sensitive targets such as water sources / residential areas, the risk level will be additionally increased), and risk superposition rule (the same unit is continuously in the diffusion impact area during multiple high-risk periods and is marked as a "key focus unit"). Through rule modification, "low-priority risks" (such as slight diffusion during non-peak periods) are eliminated, and "high-priority risks" (such as peak period + diffusion near sensitive targets) are highlighted.

[0045] The final output of spatiotemporal coupled risk warning information, presented as a visual report or structured data, includes the following core elements: 1. High-risk areas with spatiotemporal overlap (e.g., "8:30-9:30 (peak high-risk period) + within 50 meters of the diffusion range of Unit A (including Unit D and the water source to the north)"), 2. Warning levels for corresponding areas (e.g., "Blue Warning - General Concern," "Orange Warning - Key Prevention and Control," and "Red Warning - Emergency Disposal" based on the degree of risk overlap), and 3. Risk impact indicators (e.g., "Red warning areas may affect residential areas within 300 meters; response required within 1 hour"). This warning information directly serves as the core input for S60 (risk control decision instruction generation), ensuring that subsequent control measures accurately address "high-risk spatiotemporal overlap areas" and avoiding indiscriminate control measures that lead to resource waste.

[0046] S60: Based on the spatiotemporal coupling risk warning information, a preset risk management and control rule library is called to generate risk management and control decision instructions for the hazardous chemicals storage area.

[0047] In this embodiment, the core purpose is to transform the "high-risk spatiotemporal superposition information" derived from spatiotemporal coupling analysis into implementable control measures, avoiding the disconnect between warning information and actual control, leading to the problem of "warning without action," and ensuring that risks can be controlled promptly and accurately. For input data, on the one hand, the spatiotemporal coupling risk warning information output by S50 (including spatiotemporal superposition high-risk areas, warning levels, and risk impact prompts) is used as the core basis for matching control measures. On the other hand, a preset risk control rule library is invoked. This rule library is a "warning level-control measure" mapping system constructed based on hazardous chemical storage safety regulations and historical accident handling experience. For example, it is divided into grades by warning level (blue / orange / red) and classified by risk type (diffusion risk, unit-specific risk, sensitive target-related risk), ensuring that measures are precisely adapted to risk characteristics.

[0048] The specific operation is divided into two steps: the first step is rule matching, which calls the rule library according to the key parameters of the spatiotemporal coupling risk warning information; for example, if the warning information is "red warning (unit A, 8:30-9:30, the diffusion range includes the water source on the north side)", then the control rules corresponding to the "red warning + sensitive target (water source) associated risk" are matched from the rule library, such as "immediately start the emergency ventilation system of unit A, set up an isolation zone within 50 meters of the diffusion range, and dispatch emergency monitoring personnel to arrive at the water source within 10 minutes to detect the concentration"; the second step is instruction generation, which converts the matched rules into structured control decision instructions. The instructions clearly define the control objects (unit A, the area around the water source), specific measures (ventilation, isolation, personnel dispatch), execution time limit (within 10 minutes, until the warning is lifted), and responsible parties (on-site operation and maintenance team, emergency monitoring team) to avoid execution deviations caused by vague measures.

[0049] The final output of risk management and control decision instructions can be synchronized to various execution departments (such as operations and emergency response posts) through the system platform. These instructions come in the form of text instructions and visual task lists (with a map of the control area and the priority of the measures). For example, the instruction list may include the following: "Priority 1: The emergency monitoring team must arrive at the water source on the north side of Unit A before 9:00 and conduct gas concentration testing every 5 minutes; Priority 2: The operations and maintenance team must activate Unit A's explosion-proof ventilation equipment before 8:40 and continue operating until the warning is lifted." This instruction is used directly to guide on-site safety management operations, achieving a closed-loop management system of "warning-to-action" and minimizing the possibility of risk escalation.

[0050] In one embodiment, in step S20, the construction of the time dynamic risk prediction model specifically includes the following steps: S21. Acquire historical time series sample data in the storage area, including storage unit operating condition data, environmental dynamic data, and human intervention data; preprocess the historical time series sample data to obtain a standardized time series feature set; and label the standardized time series feature set with a risk level label based on historical accident data of the same type of storage units within a preset time period, wherein pre-accident data is labeled as high risk and normal operating data is labeled as low risk; S22, constructing a time series prediction model including a feature extraction layer for extracting time-sensitive features and a prediction output layer for outputting a risk level time series curve; S23. Input the standardized time series feature set marked with risk level labels into the time series prediction model architecture, adjust the model parameters using a preset training strategy, and obtain a trained time dynamic risk prediction model; S24. Input the preprocessed time series data into the trained time dynamic risk prediction model, extract the time sensitive features in the time series data through the feature extraction layer, form a sensitive feature vector set including storage unit operating condition features, environmental dynamic features and human intervention features, and input the sensitive feature vector set into the prediction output layer, and output the risk level time series curve of the future preset time length through the prediction output layer.

[0051] In this embodiment, as described in step S21, the core purpose is to provide high-quality historical sample data and labels for the training of the time-dynamic risk prediction model to ensure that the model can learn the laws of risk evolution and avoid low prediction accuracy due to insufficient training data or ambiguous labels. For input data, focus on three types of historical time series samples: storage unit operating condition data (such as equipment parameters such as tank pressure, liquid level, and pump operating status), environmental dynamic data (such as real-time temperature and humidity, wind speed, atmospheric pressure and other external environmental parameters in the storage area), and human intervention data (such as abnormal feedback in inspection records, start and stop records in operation logs, maintenance time nodes, etc.). The data coverage must include the normal operating cycle and the period before and after the accident to ensure sample diversity.

[0052] Specifically, the first step is data preprocessing, which involves cleaning the collected historical time series samples (eliminating abnormal jump data caused by sensor failures), normalizing (unifying parameters of different magnitudes to the [0, 1] range, such as converting temperatures from -20°C to standardized values), and time series alignment (integrating multi-source data by timestamp to form a continuous sequence in minutes). Ultimately, a structured standardized time series feature set is obtained. The second step is risk level labeling. Based on historical accident data of similar storage units (such as ethanol tank leaks and sodium hydroxide corrosion accidents) within a preset time period (such as 72 hours before and 24 hours after the accident), the standardized time series feature set is labeled. For example, if an ethanol tank leaks on June 10, 2023, its historical data from 00:00 on June 7 to 08:00 on June 10 will be labeled as "high risk," and the normal operating data from May 1 to May 31, 2023 will be labeled as "low risk." The labels and feature sets are precisely associated through timestamps. The output standardized time series feature set and corresponding risk level labels will be directly used for the training of the time dynamic risk prediction model (such as model parameter optimization and feature weight learning), providing corresponding "feature-label" samples for the model to identify risk evolution patterns, and ensuring that the risk level time series curve output by the subsequent model has reliable historical data support.

[0053] As described in step S22, the core goal is to build a model architecture that accurately captures the temporal evolution of risk, providing a foundational framework for time-dynamic risk prediction and avoiding incomplete extraction of time-sensitive features or biased prediction outputs due to poorly designed model structures. Model construction must incorporate the characteristics of the standardized time series feature set output by S21 (including multi-dimensional time series data and risk level labels) to ensure the architecture is adaptable to the dual needs of "time series analysis + risk level prediction."

[0054] Specific operation focuses on two-layer core structure design: one is the feature extraction layer, mainly responsible for capturing time-sensitive features related to risk from time series data, using network structures that can handle long-term temporal dependencies (such as long short-term memory networks with attention mechanisms), such as assigning higher weights to key features such as "temperature rising trend" and "abnormal pressure fluctuation frequency" through attention mechanisms, to enhance the ability to extract risk precursor signals; two is the prediction output layer, based on the key features output by the feature extraction layer, to construct a mapping relationship to generate a risk level time series curve, using a fully connected network combined with time series smoothing processing (such as sliding window averaging), to ensure that the output curve can reflect the dynamic changes of risk over time and clearly distinguish the individual risk differences of different storage units (such as through multiple output nodes corresponding to the risk levels of each unit). The constructed time series prediction model will serve as the basic architecture for subsequent model training (such as parameter optimization combined with the standardized time series feature set and labels of S21), ensuring that the model has the ability to learn the risk evolution pattern from historical data, providing reliable model support for the "input time series data and output risk level time series curve" in S20.

[0055] As described in step S23, the core purpose is to optimize model parameters through labeled historical sample data, so that the time series prediction model has accurate risk level prediction ability, avoiding prediction bias caused by unreasonable initial model parameter values (such as missing high-risk periods or mislabeling low-risk units), providing a performance-standard model tool for subsequent dynamic risk prediction in S20. For input data, on the one hand, call the "standardized time series feature set with risk level labels" output by S21 (including feature data and corresponding high / low risk labels), and on the other hand, use the "time series prediction model architecture" constructed in S22 (including feature extraction layer and prediction output layer), to ensure that the data and model input dimensions and formats are completely matched.

[0056] The specific operation is divided into two steps: the first step is data division and training initialization, and the labeled standardized time series feature set is divided into a training set (used for parameter learning) and a verification set (used for performance verification) according to a predetermined proportion (such as 7:3), for example, 70% of the "ethanol storage tank historical data (including high-risk labels before accidents and normal low-risk labels)" is taken as the training set, and 30% is taken as the verification set, and the model parameters (such as the attention weight of the feature extraction layer and the connection weight of the prediction output layer) are initialized; the second step is to adjust the parameters by using a predetermined training strategy, and the training strategy is designed around "improving prediction accuracy + avoiding overfitting", for example: the optimizer is Adam (adaptive learning rate optimizer, which can dynamically adjust the learning step), the loss function is cross-entropy loss (which measures the deviation between the model predicted label and the true label), early stopping strategy is added during training (stop training when the prediction error of the verification set does not decrease for 5 consecutive rounds to prevent the model from overfitting the training data), and the risk level prediction accuracy of the model on the verification set is calculated after each training (such as requiring an accuracy of ≥90%), until the parameters are adjusted to meet the performance of the verification set.

[0057] Finally, a trained time dynamic risk prediction model is obtained, which meets the core performance indicators (such as a risk level prediction accuracy of ≥90% on the verification set and a high-risk period missed rate of ≤5%), and can be directly used to input the preprocessed real-time time series data into the model in step S20, which can stably output the risk level time series curve for a predetermined length of time in the future, providing reliable model support for time dimension risk prediction.

[0058] As described in step S24, the core purpose is to apply the trained time dynamic risk prediction model (output by S23) to actual time series data, realize the complete landing from "data input" to "feature extraction" to "risk prediction", avoid the disconnection between model training and actual data application, and ensure that the time series curve reflecting the real risk evolution can be stably output, providing time dimension basis for subsequent high-risk identification and spatial analysis. For input data, on the one hand, the preprocessed real-time time series data (including storage unit operating parameters, environmental parameters, personnel inspection records, etc., which have been denoised and standardized) is called, and on the other hand, the trained model (including optimized feature extraction layer and prediction output layer parameters) is enabled, ensuring that the data format and model input interface are completely matched without the need for secondary processing.

[0059] The specific operation is divided into two steps: the first step is feature extraction and vector set construction. Through the feature extraction layer of the model (such as the long short-term memory network integrated with the attention mechanism), three types of time-sensitive features are extracted from the input time series data: storage unit operating condition characteristics (such as the fluctuation amplitude of the tank pressure within 1 hour, the interval between pump start and stop), environmental dynamic characteristics (such as the 3-hour rising trend of ambient temperature, the influence coefficient of wind speed on gas diffusion), and human intervention characteristics (such as the response time after the abnormality is discovered during the inspection, and the recovery trend of equipment parameters after maintenance). These features are integrated according to the "timestamp-feature type-feature value" structure to form a sensitive feature vector set (for example: "10:00-operating condition characteristics-pressure fluctuation 0.2M The second step is to output a time-series curve for risk levels. The sensitive feature vector set is input into the prediction output layer (a fully connected network structure). Based on the feature-risk mapping relationship learned during the training phase, the model calculates the risk level for each time point within a preset time period (which can be set as needed, such as 24 hours for daily monitoring or 48 hours for extreme weather). The model then outputs a time-series curve. For example, the curve might show the dynamic changes from "12:00 - Regional Overall Risk Level I (Low Risk), Unit C Risk Level II; 15:00 - Regional Overall Risk Level II, Unit C Risk Level III (High Risk)" and distinguish between overall and individual unit risks. The output risk level time-series curve is directly used in subsequent S30 (marking high-risk time windows) and S40 (identifying target storage units), becoming a key bridge connecting temporal dynamic prediction and spatial diffusion analysis, ensuring that subsequent steps are based on accurate temporal risk data.

[0060] In one embodiment, in step S24, the sensitive feature vector set is input into the prediction output layer, and the prediction output layer outputs a risk level time series curve for a preset time period in the future, which specifically includes the following steps: S241. Associate the sensitive feature vector set with the hazardous chemical type attribute of the corresponding storage unit to form an associated data set; based on the associated data set, group the sensitive feature vector set by hazardous chemical type, and sort the sensitive feature vectors in each group by time series to form a time series feature matrix stratified by hazardous chemical risk characteristics; S242. Analyze the dynamic association relationship between adjacent vectors in each hierarchical time series feature matrix using a preset time series association algorithm, and assign a preset high-impact weight value to accident-related feature vectors that match historical accident features based on an attention mechanism, and assign a preset basic weight value to other non-accident-related auxiliary feature vectors; wherein the historical accident features are extracted based on historical accident data of the same type of storage unit within a preset time period; S243. Based on the preset hazardous chemical type risk level mapping rules, combined with the risk level trend correction factor, the high impact weight value of the accident-related feature vector and the basic weight value of the auxiliary feature vector, calculate the risk level value for each period within the preset future time period; S244. Continuously arrange the risk level values ​​of each time period along the time axis to generate a risk level time series curve, wherein the risk level time series curve includes labels of storage units within a preset risk threshold range and spatially associated units of the storage units.

[0061] In this embodiment, as described in step S241, the core purpose is to bind the sensitive feature vector set to the hazardous chemical type attribute to achieve a precise association between "features and material properties", avoid the mixed analysis of risk characteristics of different types of hazardous chemicals, and thus obscure key characteristics (for example, the risk evolution patterns of flammable liquids and corrosive solids are quite different), and provide structured input for the differentiated calculation of the subsequent prediction output layer. For the input data, on the one hand, the sensitive feature vector set extracted in S24 (including three types of features: operating conditions, environment, and human intervention) is used, and on the other hand, the hazardous chemical type attributes of each storage unit are called (such as "Unit A - Ethanol (Flammable Liquid)", "Unit B - Sodium Hydroxide (Corrosive Solid)", and "Unit C - Ammonia (Toxic Gas)") to ensure that the feature vectors can correspond one-to-one with the specific hazardous chemical type.

[0062] The specific operation is divided into three steps: the first step is to associate and form a data set, bind each sensitive feature vector (such as the pressure fluctuation feature of unit A at 10:00) with the hazardous chemical type attribute of the unit (such as ethanol), and form an associated data set of "feature vector-hazardous chemical type", for example: "feature: [pressure fluctuation 0.2MPa, temperature rise 2℃]; type: ethanol"; the second step is to group by type, and divide the sensitive feature vector set into several subgroups based on the associated data set. The feature vectors of the same type of hazardous chemicals are in the same subgroup. For example, the feature vectors of all "ethanol" units are grouped together. The quantity is classified as "flammable liquid group" and the "sodium hydroxide" unit is classified as "corrosive solid group"; the third step is time series sorting and matrix construction. Within each group, the sensitive feature vectors are arranged in chronological order according to timestamps (such as sorted by hour from 9:00 to 18:00) to form a structured time series feature matrix. For example, the rows of the "flammable liquid group" matrix represent time nodes (9:00, 10:00...), the columns represent feature types (pressure fluctuations, temperature trends, etc.), and the matrix elements are corresponding eigenvalues, which clearly show the evolution of the characteristics of the same type of hazardous chemicals over time.

[0063] The output of the time sequence feature matrix stratified according to the risk characteristics of the hazardous chemicals will be directly input into the subsequent calculation link of the prediction output layer, so that the model can perform differential analysis on the risk characteristics (such as paying more attention to temperature / concentration for flammable substances and paying more attention to equipment pressure / tightness for corrosive substances) of different types of hazardous chemicals (such as flammable, corrosive, and toxic), and lay a foundation for accurately outputting the risk level time curve.

[0064] As described in step S242, the core purpose is to strengthen the model's ability to identify "risk precursor features". By quantifying the dynamic correlation between features and differentiating the weight distribution, the key accident features are prevented from being overwhelmed by a large number of routine features (such as distinguishing between sudden pressure rise features directly related to accidents and ordinary environmental temperature and humidity features), providing a more accurate basis for the importance of features for subsequent risk level calculation. For input data, on the one hand, the "time sequence feature matrix stratified according to the type of hazardous chemicals" (such as the feature matrix of the flammable liquid group and the corrosive solid group) output by S241 is used, and on the other hand, the "historical accident features of the same type of storage unit" (such as the "pressure rises by 0.5 MPa within 1 hour + gas concentration exceeds the standard" combination feature extracted from ethanol tank leakage accident data) is called to ensure that the weight distribution is supported by actual accident data.

[0065] The specific operation is divided into two steps: the first step is dynamic correlation analysis, which captures the dependency relationship between adjacent time vectors in the stratified matrix through a preset time sequence correlation algorithm (such as sliding window correlation analysis, time sequence pattern mining algorithm); for example, in the flammable liquid group matrix, the correlation strength between "t time pressure fluctuation feature" and "t+1 time gas concentration feature" (such as the probability of concentration exceeding the standard within 10 minutes after pressure surge) is analyzed, and typical risk evolution chains such as "pressure anomaly → concentration exceeds the standard" are identified; the second step is to assign weights based on the attention mechanism, and compare the current feature vector with the historical accident feature. If they match (such as the current "ethanol unit pressure rises by 0.4 MPa within 1 hour" matches the historical accident feature "pressure rise ≥0.3 MPa"), it is determined as an "accident-related feature vector" and given a high impact weight value (such as a weight value of 0.8), while the auxiliary feature vector that does not match (such as normal fluctuation of environmental humidity) is given a basic weight value (such as a weight value of 0.2). For example: in the feature matrix of the ethanol unit, the "pressure fluctuation vector" is given a high weight because it matches the historical leakage accident feature, and the "daily inspection without anomaly vector" is given a basic weight as an auxiliary feature. The output "weighted stratified time sequence feature matrix" clearly marks the impact weight of each feature vector, not only retaining the dynamic correlation information between features, but also highlighting the key features related to accidents through weight differences, which will be directly used in the subsequent risk level calculation link to ensure that the model prioritizes high-weight risk precursor features when predicting, thereby improving the early warning sensitivity of the risk level time curve.

[0066] As described in step S243, the core purpose is to convert the weighted feature vector into a quantifiable risk level value, achieving the implementation from feature analysis to risk determination. This avoids distortion in risk level calculation (such as using a unified standard to measure the risk of flammable and corrosive hazardous chemicals) caused by the calculation logic being out of touch with the characteristics of the hazardous chemicals or ignoring the risk evolution trend. This provides accurate quantitative data support for the subsequent output of the risk level time series curve. For the input data, on the one hand, the "weighted hierarchical time series feature matrix" output by S242 (including the high-impact weights of accident-related features and the basic weights of auxiliary features) is called. On the other hand, two key parameters are introduced: the preset hazardous chemical type risk level mapping rules (the "feature-risk" correspondence standard customized according to the characteristics of the hazardous chemicals) and the risk level trend correction factor (a dynamic adjustment parameter reflecting the change of risk over time) to ensure that the calculation logic is consistent with the actual risk evolution law.

[0067] The specific operation involves three steps. The first step is to define mapping rules. These pre-set mapping rules are customized for the risk characteristics of different hazardous chemical types. For example, the mapping rule for "ethanol (flammable liquid)" is "combustible gas concentration ≥ 15% LEL (high impact weight 0.8) → basic risk score 3 points; ambient temperature ≥ 35°C (basic weight 0.2) → basic risk score 1 point." The mapping rule for "sodium hydroxide (corrosive solid)" is "storage tank seal leakage (high impact weight 0.8) → basic risk score 3 points; ambient humidity ≥ 60% (basic weight 0.2) → basic risk score 1 point." This avoids confusion between risk standards across hazardous chemical types. The second step is to calculate a preliminary risk value. This is done by weighting the characteristic vectors for each time period using the logic of "characteristic basic risk score × corresponding weight." For example, for a certain time period for an ethanol unit, the preliminary risk value for "gas concentration 18% LEL (basic score 3 × 0.8 = 2.4) + temperature 36°C (basic score 1 × 0.2 = 0.2)" is 2.4 + 0.2 = 2.6 points. The third step is to introduce the trend correction factor adjustment. If the characteristics before and after the period show a "continuous deterioration trend" (for example, the gas concentration rises from 12%LEL to 18%LEL for three consecutive periods), the correction factor is 1.2 (risk increase); if it shows a "stable trend", it is 1.0. The final risk level value = preliminary risk value × correction factor (for example, 2.6×1.2=3.12 points, which corresponds to Level III risk according to the preset classification standard).

[0068] The output "risk level value for each time period within the preset time period in the future" (such as the risk level value for 24 time periods at 1-hour intervals in the next 24 hours) will be directly used in the subsequent S24 links; these values ​​​​are integrated according to the time series to form a complete risk level time series curve, ensuring that the risk level of each time node in the curve has a dual calculation basis of "feature adaptation + trend dynamics", thereby improving the accuracy and credibility of risk prediction.

[0069] As described in step S244, the core purpose is to convert the dispersed time period risk level values into an intuitive and continuous time sequence curve, and through key unit labeling, let the curve carry both "time risk change" and "spatially associated risk" information, avoid missing spatially associated risks (such as sensitive units near high-risk units) due to the presentation of only single time dimension data, and provide a visual carrier for "spatiotemporal information fusion" for the final output of S20, high-risk time period labeling of S30, and target unit determination of S40. For the input data, on the one hand, the "risk level value of each time period in the future preset time length" (such as 24 risk level quantization values for every 1 hour in the future 24 hours) output by S243 is called, and on the other hand, two types of key information are associated: the storage unit basic information (unit ID, storage dangerous chemical type) after preprocessing by S10, and the "storage unit-spatially associated unit" correspondence in the spatial topology data of S10 (such as adjacent units B and C of unit A, or associated facilities such as fire barriers and ventilation pipes close to the unit), while the preset risk threshold (such as risk level ≥ level III as the high-risk judgment standard) is used.

[0070] The specific operation is divided into two steps: the first step is curve generation, which arranges the risk level values of each time period in the order of time axis (such as from 0 o'clock to 24 o'clock) to construct a two-dimensional risk level time sequence curve: the horizontal axis is the time node (such as 0 o'clock, 1 o'clock... 24 o'clock), and the vertical axis is the risk level (such as level I to level IV), the values of each time period are connected by a smooth curve to clearly present the dynamic change trend of the risk with time (such as stable risk level I from 0 o'clock to 8 o'clock, rising to level II at 9 o'clock, and reaching level III at 12 o'clock); the second step is key unit labeling, which filters out "risk level exceeding period" based on the preset risk threshold, and labels two types of labels at the curve position of the corresponding time node: one is the "storage unit located in the preset risk threshold range" (such as unit A with risk level III at 12 o'clock, labeled as "unit A (ethanol) - level III risk" with a red triangular marker), and the other is the "spatially associated unit" of the unit (such as adjacent units B and C of unit A, labeled as "associated unit B (toluene) - potential risk" and "associated unit C (fire hydrant) - protective facility" with yellow dot markers), and the unit ID, storage material and risk association type (directly exceeding / latent impact) are marked in the labels. The finally output risk level time sequence curve not only intuitively presents the risk evolution law in the time dimension, but also clearly presents the exceeding unit and its spatially associated objects through the labels, realizing the preliminary fusion of "time risk-spatial association". The curve will be directly used as the core output result of step S20, providing key data support for subsequent S30 to accurately lock the high-risk time period, and S40 to determine the target storage unit and associated unit that need to be analyzed.

[0071] In one of the embodiments, the determination of the preset risk threshold includes the following steps: S201. Based on the hazardous chemical type attribute of the storage unit, retrieve the basic safety threshold of the corresponding type of hazardous chemicals as the initial risk threshold; S202. Extract the risk level critical value of the corresponding type of hazardous chemicals at the time of the historical accident from the historical accident characteristics, and modify the initial risk threshold based on the risk level critical value to obtain a type-specific benchmark threshold; S203: extracting real-time dynamic impact factors from the sensitive feature vector set, dynamically adjusting the type-specific benchmark threshold based on the extracted real-time dynamic impact factors, and determining a preset risk threshold for the corresponding storage unit.

[0072] In this embodiment, as described in steps S201-S203 above, the core purpose is to establish a three-layer threshold determination logic of "type adaptation + historical verification + real-time dynamics" to avoid the problems of "high-risk misjudgment (threshold too high)" or "low-risk omission (threshold too low)" caused by the use of fixed thresholds. This ensures that the risk level determination criteria in steps S20 and S24 are both consistent with the characteristics of hazardous chemicals and can adapt to changes in actual scenarios. The three steps are connected in sequence, from basic thresholds to dynamic adjustment, gradually forming a precise preset risk threshold, as follows: Step S201 aims to establish initial threshold benchmarks tailored to the characteristics of different hazardous chemicals, avoiding the use of a single standard to measure the risks of significantly different hazardous chemicals (e.g., the safety thresholds for flammable gases and corrosive liquids differ significantly). Input data uses the hazardous chemical type attributes of the storage unit (e.g., "Unit A - Ethanol (Flammable Liquid)" and "Unit B - Ammonia (Toxic Gas)"). This can be retrieved from the spatial topology data preprocessed in S10). During operation, the corresponding basic safety thresholds are retrieved from a pre-set "Hazardous Chemical Safety Threshold Library" based on the type attributes. This library is based on national and industry standards. For example, the basic safety thresholds for ethanol are "Combustible Gas Concentration ≥ 20% LEL (Lower Explosive Limit)" and "Temperature ≥ 40°C," while the basic safety thresholds for ammonia are "Gas Concentration ≥ 30 ppm" and "Pressure ≥ 0.6 MPa." The output initial risk thresholds provide a "type-specific" starting point for subsequent revisions.

[0073] In step S202, the initial threshold is verified and optimized using historical accident data to ensure that it aligns with actual accident evolution patterns (to avoid a disconnect between the standard threshold and on-site accident characteristics). The input data is based on the initial risk threshold outputted in S201. Furthermore, the risk threshold (the actual risk parameter at the time of the accident) is extracted from the "historical accident characteristics of similar storage units" in S21. This threshold is the actual risk parameter at the time of the accident. For example, in a historical ethanol leak, the actual pre-explosion gas concentration threshold was 15% LEL, lower than the initial threshold of 20% LEL. During operation, the threshold is compared with the initial threshold. If the threshold is lower than the initial threshold (indicating an earlier risk trigger point), the initial threshold is adjusted down to a value near the threshold (e.g., the ethanol concentration threshold is adjusted from 20% LEL to 16% LEL). If the threshold is higher than the initial threshold, the initial threshold is maintained or fine-tuned. The resulting type-specific baseline threshold integrates standard requirements and historical accident experience, making it more applicable to the field than the initial threshold.

[0074] In step S203, real-time factors are incorporated to dynamically calibrate the baseline threshold, avoiding the inability of fixed thresholds to address unexpected scenarios (such as temporary risk-increasing factors like extreme weather and equipment aging). For input data, the type-specific baseline threshold output from S202 is used. Furthermore, real-time dynamic influencing factors (such as sudden increases in ambient temperature and humidity, equipment expiration, and the presence of nearby fire sources) are extracted from the sensitive feature vectors extracted in S24. During operation, adjustment rules are set based on the degree of influence of these factors. For example, in high-temperature weather (ambient temperature ≥ 35°C), the ethanol concentration threshold is adjusted by 20% (to 12.8% LEL) from the baseline value of 16% LEL. For equipment aging (over 10 years of operation), the pressure threshold is lowered by 15%. The resulting preset risk threshold retains the type-specific and historical validation foundation while incorporating real-time scenario variables. This serves as the direct criterion for determining whether the risk level exceeds the standard in S20 and S24, ensuring accurate and timely risk identification.

[0075] In one embodiment, step S40 specifically includes the following steps: S41. Perform hierarchical analysis on the preprocessed spatial topology data to extract the spatial correlation parameters and physical barrier levels between the target storage unit and adjacent storage units. Based on the diffusion characteristic parameters of hazardous chemicals in the target storage unit, a spatial diffusion risk assessment model is constructed. S42. When the risk level time series curve is associated with a storage unit within a preset risk threshold, a real-time risk level value of the target storage unit is obtained and input into a spatial diffusion risk assessment model. The risk diffusion characteristics of the target storage unit are extracted using the spatial diffusion risk assessment model. The risk diffusion characteristics include the risk diffusion rate, diffusion range radius, and hazardous substance concentration attenuation coefficient of the target storage unit. S43, based on the extracted risk diffusion features and the spatial correlation parameters of the target storage unit and the adjacent storage units, associating the risk diffusion features of the target storage unit to the corresponding adjacent storage units according to a preset correlation rule, calculating the risk levels of each adjacent storage unit affected by diffusion, and generating a correlation table containing the target storage unit, the adjacent storage units, the diffusion path and the affected risk level; S44, according to the risk change trend of the risk level time curve, superimposing and matching the correlation table with the spatial map of the storage area, generating and outputting a dynamic risk spatial distribution result of the storage area dynamically updated over time, the dynamic risk spatial distribution result containing the actual risk level and the diffusion influence range label of each storage unit in different time periods.

[0076] In this embodiment, as described in step S41, the core purpose is to build a model framework that can accurately simulate the diffusion path and influence range of hazardous chemicals, providing a calculation basis for subsequent analysis of the conduction rules of target unit risk to the surrounding, avoiding the distortion of diffusion range prediction (such as underestimating the diffusion speed of flammable gas and overestimating the influence radius of liquid leakage) due to the neglect of spatial correlation details (such as unit spacing, barrier facilities) or hazardous chemical diffusion characteristics (such as density, volatilization rate) by the model. For input data, on the one hand, the spatial topology data after preprocessing by S10 (including storage unit position, spacing, pipeline / fire dike connection and separation relationship), on the other hand, the diffusion characteristic parameters of hazardous chemicals in the target storage unit (such as the volatilization rate of ethanol, the diffusion coefficient of ammonia gas, the flow viscosity of sulfuric acid, etc., which can be retrieved from the hazardous chemical safety technical specification), ensure that the model input covers both spatial structure and material properties.

[0077] The specific operation is divided into three steps: the first step is to analyze the spatial topology data in layers, and disassemble the data according to the "unit layer-association layer-barrier layer"; the unit layer focuses on the physical properties of the target storage unit and adjacent units (such as tank volume and material); the association layer extracts the spatial association parameters between units (such as the straight-line distance between the target unit and the adjacent unit, the pipeline connection method (direct connection / indirect connection), and the relative height difference); the barrier layer determines the physical barrier level (such as the height and fire resistance level of the fire dike, the material of the wall (concrete / iron sheet), and the coverage of the ventilation facilities, which are divided into "strong barrier (level 3)-medium barrier (level 2)-weak barrier (level 1)" according to the protection capability); the second step is parameter integration, which combines the spatial association Parameters (e.g., 50-meter distance, direct pipeline connection), physical barrier levels (e.g., Level 3 firebreaks), and diffusion characteristic parameters (e.g., ethanol evaporation rate of 0.5 kg / h) are linked to form the model input parameter set. The third step is model construction. Based on this parameter set, an appropriate diffusion algorithm is selected (e.g., a "gravity flow + evaporation diffusion" model is used for flammable liquids, and a "Gaussian diffusion model" is used for toxic gases). The barrier attenuation coefficient is also embedded (e.g., a Level 3 barrier can reduce the diffusion impact by 60%). For example, the model for the ethanol storage tank (target unit) calculates the concentration of unit B 50 meters away affected by diffusion in the absence of a barrier, multiplying it by the attenuation coefficient of 0.4 for the Level 3 firebreak to obtain the actual impact concentration. The resulting spatial diffusion risk assessment model, which includes the following functions: "Input target unit information → Calculate diffusion range, impact concentration, and impact level of adjacent units." This model will be directly used in the subsequent S40 step: "When a target storage unit is marked as high-risk, the model quickly extracts its risk diffusion characteristics, providing the core calculation tool for outputting dynamic risk spatial distribution results."

[0078] As described in step S42, the core purpose is to convert the high-risk units identified in the time dimension (the results of S20 annotation) into quantifiable spatial diffusion parameters, providing key characteristic basis for analyzing the transmission pattern of risks to the surrounding areas, avoiding only knowing that "the unit risk is high" without knowing "how the risk spreads and the extent of the impact", which leads to a lack of targeted subsequent spatial prevention and control. For input data, on the one hand, the "high-risk storage unit associated with the annotation in the risk level time series curve" (i.e., the target storage unit, such as unit A with a risk level ≥ Level III) and its real-time risk level value (e.g., unit A's current risk level is Level IV) output by S20 are called, and on the other hand, the spatial diffusion risk assessment model constructed in S41 (including spatial association parameters, physical barrier levels, and hazardous chemical diffusion characteristic parameters) is activated to ensure that the input matches the format and dimensions required by the model.

[0079] The specific operation consists of two steps. The first is triggering and data acquisition. When the system detects a storage unit within the preset risk threshold in the risk level time series curve (for example, Unit A's risk level reaches Level IV, exceeding the threshold of Level III), the diffusion analysis process is automatically triggered, and the real-time risk level value (reflecting the current risk severity) of the target unit, the unit ID, and the corresponding hazardous chemical type (for example, Unit A stores ethanol) are simultaneously obtained. The second step is model calculation and feature extraction. This data is input into the spatial diffusion risk assessment model. Based on a preset algorithm (for example, combining parameters such as the ethanol volatilization rate and current wind speed), the model calculates and extracts three core risk diffusion characteristics: risk diffusion rate (for example, ethanol vapor diffuses at a speed of 0.3 m / s), diffusion radius (for example, the maximum diffusion range within 1 hour is 50 meters), and hazardous substance concentration attenuation coefficient (for example, the ethanol concentration decreases by 30% for every 10 meters from Unit A). For example, the diffusion characteristic calculation result for Unit A (ethanol, Level IV risk) is "rate 0.3 m / s, radius 50 meters, attenuation coefficient 0.3 / 10 meters," clearly reflecting the spatial propagation of risk. The output risk diffusion characteristics (rate, range, attenuation coefficient) will be directly used in the subsequent steps of S40, associated with adjacent storage units, and the risk level of the affected adjacent units will be calculated, providing quantitative diffusion parameter support for the final output of "dynamic risk spatial distribution results of storage areas", ensuring that spatial risk analysis knows both the "risk source" and the "diffusion path and impact degree".

[0080] As described in step S43, the core purpose is to establish a risk transmission association between the "target unit and adjacent units" and quantify the extent to which adjacent units are affected by diffusion. This avoids focusing solely on the target unit's own risks while ignoring surrounding chain risks (e.g., after a target unit leaks, adjacent units are quickly affected due to their proximity or lack of barriers). This provides structured correlation data support for the subsequent generation of dynamic risk spatial distribution results. For the input data, on the one hand, the risk diffusion characteristics extracted in S42 (diffusion rate, range radius, and concentration attenuation coefficient of the target unit) are used, and on the other hand, the spatial correlation parameters analyzed in S41 (the spacing between the target unit and adjacent units, relative position, pipeline connection relationship, physical barrier level, etc.) are used. At the same time, preset association rules are introduced (defining the logic of "which adjacent units should be associated" and "how to calculate the risk level") to ensure that the association and calculation logic are quantifiable and reproducible.

[0081] The specific operation is divided into three steps: the first step is to define the association range. According to the preset association rules, the adjacent units that are "within the diffusion range radius of the target unit" and "the spatial association parameters meet the conduction conditions" are screened. For example, the diffusion range radius of the target unit A (ethanol) is 50 meters. The rule sets "units with a spacing of ≤50 meters and no insurmountable barriers (such as concrete walls above 3 meters) as associated objects". Then, unit B (toluene, only separated by a 1-meter-high iron wall) at 30 meters and unit C (empty tank area, no barrier) at 45 meters are screened out, and unit D at 60 meters (out of range) is excluded; the second step is to calculate the affected risk level. Combining the diffusion characteristics and spatial parameters, according to "concentration attenuation × barrier The logical calculation of "attenuation": Unit B is 30 meters away, the concentration attenuation coefficient is "30% attenuation every 10 meters" (90% attenuation at 30 meters, 10% remaining), and the physical barrier (iron wall) is level 2 (40% attenuation). Then the actual affected concentration = initial concentration × 10% × (1-40%) = initial concentration × 6%. According to the ethanol diffusion risk level standard (concentration ≥ 5% corresponds to Level II risk), the affected level of Unit B is Level II; Unit C is 45 meters away and has no barrier. The concentration attenuates to 15% × 100% = 15%, corresponding to Level III risk. The third step is to generate an association relationship table, which is structured according to "target unit - adjacent unit - diffusion path - affected level". For example: Table 1: Association table The output correlation table clearly presents the specific chain and impact degree of the target unit risk transmission to the surrounding areas, which will be directly used in the subsequent steps of S40: integration into the dynamic risk spatial distribution results of the storage area, providing core data for the visual display of "risk source-diffusion path-affected area", and ensuring that the spatial risk analysis is both relevant and quantitative.

[0082] As described in step S44, the core goal is to deeply integrate the "correlationships of risk diffusion" with "spatial geographic information" and "temporal trends" to form an intuitive and dynamic spatial risk landscape. This avoids the problems of "spatial location ambiguity" and "lack of temporal dynamics" (e.g., the inability to intuitively determine the movement trajectories of high-risk areas during different time periods) that arise from describing risk distribution solely through tables or text. This provides "visible and traceable" spatial dimension dynamic data support for the spatiotemporal coupling analysis in S50. For input data, the correlation table output by S43 (including the target unit, adjacent units, diffusion paths, and affected levels) is used. Furthermore, two key pieces of information are introduced: the risk level time series curve output by S20 (reflecting the temporal trend of risk, such as "the target unit's risk continues to increase from 10:00 to 12:00") and the spatial map of the storage area (generated based on the spatial topology data preprocessed by S10, containing geographic information such as unit location coordinates, roads, and protective facilities). This ensures that the input data simultaneously covers the three dimensions of "correlationships," "temporal trends," and "spatial geography."

[0083] The specific operation is divided into three steps: the first step is dynamic matching of the time dimension. According to the time nodes of the risk level time series curve (such as 10:00, 11:00, and 12:00 divided by 1 hour intervals), the risk data in the association table are split according to the corresponding time periods. For example, the "diffusion characteristics of unit A at 10:00" are bound to the "association relationship of the 10:00 time period", and the "diffusion characteristics of 11:00" are bound to the "association relationship of the 11:00 time period" to ensure that the risk data of each time node are independently corresponding; the second step is spatial map overlay, mapping the association relationship data of each time period (target unit position, adjacent unit position, diffusion path, affected level) to the corresponding coordinates of the spatial map, and distinguishing the risk levels by visual symbols (such as red The third step is dynamic update and integration, which connects the spatial superposition results of each time period in chronological order to form a dynamic view that automatically updates over time. For example, at 10:00, the map displays "unit A (red), unit B (yellow), and a diffusion range of 50 meters"; at 11:00, due to the escalation of the risk, it displays "unit A (dark red), unit B (orange), and a newly added unit C (yellow), with a diffusion range of 60 meters", which intuitively presents the expansion trend of the risk in space.

[0084] The output of the dynamic spatial distribution of risk is presented as a dynamic map or time-series slices. Its core components are a "timeline + spatial map + risk annotation" framework. The timeline can be dragged to select any time period, and the spatial map displays in real time the actual risk level of each storage unit (including its own risk and the risk affected by diffusion), the diffusion impact range boundaries, and key diffusion paths. This result is directly used in the S50 spatiotemporal coupling analysis, accurately matching it with the high-risk time window (S30). This provides a visual basis for identifying overlapping areas of "high-risk time periods + high-risk spatial regions," ensuring that subsequent risk warnings and control measures can be precisely targeted at the right time and location.

[0085] In one embodiment, step S42 further includes the following steps: S421. Obtain the physical properties of the target storage unit and the type of hazardous chemicals inside it, and determine whether the target storage unit belongs to a special scenario based on a preset determination threshold; S422: When the target storage unit belongs to a special scene, calling a preset special scene correction coefficient table to match the correction coefficient corresponding to the target storage unit, wherein the special scene correction coefficient table includes correction coefficients corresponding to several special scenes; S423 , dynamically correcting the extracted risk diffusion feature according to the correction coefficient corresponding to the target storage unit to obtain a corrected risk diffusion feature.

[0086] In this embodiment, S421-S423 are supplementary optimization steps for extracting the risk diffusion characteristics of the target storage unit in S42. Their core purpose is to modify the diffusion characteristics for special scenarios (such as storage tanks with heating devices and units storing highly toxic substances). This avoids ignoring the risk amplification / attenuation effects of special scenarios, which can lead to distortion in the calculation of parameters such as diffusion rate and range (for example, heated storage tanks accelerate the volatilization of flammable liquids, so the diffusion rate should be higher than that of ordinary storage tanks). This ensures that the risk diffusion characteristics are more closely aligned with actual scenarios. These three steps sequentially implement the logical closed loop of determining special scenarios → matching correction coefficients → modifying diffusion characteristics, as detailed below: Step S421 aims to accurately identify the storage units that require special treatment, provide a basis for subsequent corrections, and avoid mixing common scenarios with special scenarios. For input data, on the one hand, the physical properties of the target storage unit are retrieved (such as whether the tank material is thermal insulation type, whether it is equipped with heating / cooling devices, whether the tank volume exceeds 500 , whether it is located underground or semi-underground), on the other hand, the type attributes of hazardous chemicals in the unit are retrieved (such as whether it is a highly toxic substance, a self-igniting substance, or a water-reactive substance, which can be obtained from the hazardous chemicals safety technical manual); at the same time, a preset judgment threshold is introduced, which is based on the "special scenario judgment standard" defined in industry safety regulations, such as: "storage tanks equipped with heating devices", "units storing highly toxic substances (LD50≤50mg / kg)", "underground storage tanks with a volume of more than 1000 " are all judged as special scenarios.

[0087] During the specific operation, the physical properties and hazardous chemical type properties of the target unit are compared with the preset judgment thresholds one by one: if any threshold condition is met, it is judged as a special scenario; if none of them are met, it is judged as a normal scenario. For example: Target unit E is a "methanol storage tank equipped with a steam heating device (methanol is a flammable liquid, not highly toxic)", and its physical property "with heating device" meets the preset threshold, so it is judged as a special scenario; target unit F is an "ethanol storage tank made of ordinary carbon steel (no heating device, volume 300 )", it is determined to be a normal scene. The output result is "yes / no special scene" and the corresponding special scene type (such as "heating tank scene" and "highly toxic substance scene").

[0088] Step S422 aims to provide a quantitative basis for corrections for different special scenarios, avoiding subjective correction logic (e.g., relying on empirical judgments about the impact of heating devices). For input data, the "special scenario type" output from S421 (e.g., "heating tank scenario," "highly toxic substance scenario") is used. Furthermore, a pre-set special scenario correction coefficient table is used. This table, constructed based on historical experimental data and accident cases, is divided into "scenario type-diffusion characteristic dimension" to clearly define the correction ratios for "diffusion rate, diffusion range radius, and concentration attenuation coefficient" for different scenarios. An example is shown below: Table 2: Correction coefficient table for special scenarios During specific operations, the correction coefficients for the corresponding rows in the table are precisely matched based on the scenario type determined in S421. For example, if S421 determines that unit E is a "heating tank scenario (temperature 55°C)," the matching results are "diffusion rate coefficient 1.2, range coefficient 1.1, attenuation coefficient 0.9." If the scenario is a "highly toxic substance scenario," the matching results are "rate coefficient 1.0, range coefficient 1.3, attenuation coefficient 0.8." The output results are three sets of correction coefficients corresponding to this specific scenario (one for diffusion rate, one for range radius, and one for concentration attenuation coefficient).

[0089] Step S423 implements the optimization for specific scenarios. This involves adjusting the original diffusion characteristics using quantitative coefficients to ensure that the results align with the actual risk patterns of specific scenarios. For the input data, the three sets of correction coefficients output by S422 are used, while the "initial risk diffusion characteristics" extracted from the original step S42 (e.g., an uncorrected diffusion rate of 0.3 m / s, a range radius of 50 m, and a concentration attenuation coefficient of 0.3 / 10 m) are used.

[0090] During the specific operation, make corrections one by one according to the logic of "initial eigenvalue × corresponding correction coefficient": Corrected diffusion rate = initial diffusion rate × diffusion rate correction factor (e.g. 0.3 m / s × 1.2 = 0.36 m / s); Corrected diffusion range radius = initial range radius × range radius correction factor (e.g. 50m × 1.1 = 55m); Corrected concentration attenuation coefficient = initial attenuation coefficient × concentration attenuation coefficient correction factor (e.g. 0.3 / 10 m × 0.9 = 0.27 / 10 m).

[0091] For example, the initial diffusion characteristics for unit E are "rate 0.3 m / s, range 50 m, and attenuation 0.3 / 10 m." After correction using the heating scenario coefficient, they become "rate 0.36 m / s, range 55 m, and attenuation 0.27 / 10 m." This better reflects the reality of methanol volatilization, which accelerates under heating, expands its diffusion range, and slows its concentration decay. The output result, the "corrected risk diffusion characteristics" (including the corrected rate, range, and attenuation coefficient), directly replaces the original initial characteristics and is used to calculate the risk level of affected adjacent units in step S43, ensuring more accurate spatial diffusion analysis in specific scenarios.

[0092] In one embodiment, step S50 specifically includes the following steps: S51. Extracting time boundary parameters of the high-risk period window, wherein the time boundary parameters include the high-risk starting time, duration, and risk level peak time; S52: Mapping the time boundary parameters of the high-risk time window with the dynamic risk spatial distribution results in a spatiotemporal dimension, matching the storage area risk spatial distribution state corresponding to each time node, and forming an initial spatiotemporal coupling matrix. The initial spatiotemporal coupling matrix includes associated data of time nodes, storage units, risk levels, and diffusion ranges. S53, modifying the initial spatiotemporal coupling matrix based on preset risk association rules, wherein the preset risk association rules include a time priority weight rule, a spatial correlation rule, and a risk superposition rule; S54. Generate and output space-time coupling risk warning information based on the corrected space-time coupling matrix.

[0093] In this embodiment, S51-S54 are the core sub-steps in S50 that implement spatiotemporal coupling analysis and warning output. The core purpose is to accurately associate high-risk periods in the time dimension with risk distribution in the spatial dimension through a closed-loop process of "time parameter extraction → spatiotemporal mapping → rule modification → warning generation." Ultimately, the output of risk warning information with both spatiotemporal accuracy avoids "management timing misalignment" or "regional positioning deviation" (e.g., only knowing the high-risk period but not the high-risk area corresponding to the period) caused by the fragmentation of spatiotemporal analysis. The four steps are progressive, gradually achieving the transformation from "data association" to "decision support," as follows: Step S51 aims to identify key time nodes during high-risk periods, provide precise time coordinates for subsequent spatiotemporal mapping, and avoid mapping deviations due to ambiguous time boundaries (e.g., the inability to determine the specific high-risk period corresponding to a certain spatial state). For input data, the "high-risk period window" output by S30 is used (e.g., "8:00-10:30 (Unit A high risk)" and "14:00-16:00 (Unit A high risk)"). During operation, three core time boundary parameters are extracted from each window: the high-risk starting time (e.g., 8:00, 14:00, marking the time point when the risk begins), the duration (e.g., 2.5 hours, 2 hours, reflecting the time span of the risk), and the peak risk level moment (e.g., 9:15, 15:00, based on the risk level time series curve of S20, the time point with the highest risk).

[0094] For example, for the "8:00-10:30" window, the extracted parameters are "starting at 8:00, lasting 2.5 hours, peaking at 9:15." The output time boundary parameters will serve as the reference coordinates of the "time dimension" in S52, ensuring that each time node accurately corresponds to the spatial state.

[0095] Step S52 aims to establish a one-to-one correspondence between time nodes and spatial states, integrating the fragmented time and space data into a structured matrix to avoid the loss of associated information caused by analyzing the spatiotemporal dimensions separately (e.g., not knowing the high-risk spatial distribution corresponding to the peak time of 9:15). For the input data, on the one hand, the time boundary parameters extracted by S51 (key nodes such as the starting time and peak time) are called, and on the other hand, the "dynamic risk spatial distribution results" output by S44 (including the unit risk level and diffusion range in different time periods) are used; during operation, the mapping is performed according to the logic of "time node → corresponding spatial state": for the starting time (e.g., 8:00), the spatial distribution at that time is matched (e.g., unit A risk level III, diffusion range 40 meters, unit B affected risk level II); for the peak time (e.g., 9:15), the spatial distribution at that time is matched (e.g., unit A risk level IV, diffusion range 50 meters, units B and C affected risk level III); and finally, the initial spatiotemporal coupling matrix is ​​integrated, as shown below: Table 3: Initial spatiotemporal coupling matrix The output initial spatiotemporal coupling matrix clearly presents the risk status and diffusion range of each storage unit at different time nodes, providing a basic data framework for subsequent corrections.

[0096] Step S53 aims to optimize the matrix through rules, highlighting high-priority risks (such as the combined risks of peak times and sensitive areas) and preventing low-priority information from interfering with decision-making (such as the slight risk of diffusion during non-peak periods). For the input data, on the one hand, the initial spatiotemporal coupling matrix output by S52 is used, and on the other hand, the "preset risk association rules" (three core rules) are called. During operation, the matrix is ​​modified according to the rules: Time priority weighting rule: Peak risk levels are weighted 1.2 times (e.g., Unit B at 9:15 is affected from Level III to Level III+), while non-peak risk levels (e.g., 8:30) are weighted 0.8 times (Level II to Level II-), highlighting the times when risks are most severe. Spatial correlation rule: If the affected unit is close to a sensitive target (such as a water source or residential area), the risk level is raised by one level (for example, unit C was affected at level III at 9:15 and is close to a water source → revised to level IV); Risk superposition rule: If the same unit is continuously affected during multiple high-risk periods (e.g., unit B is Level II or above during 8:00-10:30 and 14:00-16:00), it will be marked as a "key focus unit."

[0097] For example, after corrections, the risk level of Unit C (near the water source) at 9:15 AM was raised from Level III to Level IV, and Unit B was marked as "key concern" due to continued impact. The output of the corrected spatiotemporal coupling matrix eliminates redundant information, focuses on high-priority risks, and provides accurate data for early warning generation.

[0098] Step S54 aims to transform the revised matrix into intuitive and actionable early warning information, achieving the transition from "data matrix" to "decision-making basis" and preventing on-site personnel from quickly understanding risks due to complex data formats. For input data, the spatiotemporal coupling matrix revised in S53 is used; during operation, the information is integrated according to the structure of "spatiotemporal superposition area → warning level → impact prompt": Identify high-risk areas with overlapping time and space (e.g., "9:15-9:45 (peak period) + within 50 meters of the diffusion range of unit A (including unit C and the water source to the north)"); Classify warning levels according to the degree of risk superposition (e.g., "red warning - emergency response" and "orange warning - key prevention and control"); Additional risk impact prompts (such as "red alert area may affect surrounding residential areas within 30 minutes, immediate isolation should be initiated").

[0099] The output spatiotemporal coupling risk warning information can be presented in the form of a visual map (annotating high-risk areas and time periods), text instructions (clearly indicating disposal requirements), etc., directly as the core input for S60 to generate control decision instructions, ensuring that control measures can accurately match "high-risk spatiotemporal overlapping areas" and improve risk disposal efficiency.

[0100] In one of the embodiments, a hazardous chemical storage accident risk analysis system is provided, which corresponds to the hazardous chemical storage accident risk analysis method in the above-mentioned embodiments. The hazardous chemical storage accident risk analysis system comprises: A data acquisition module for acquiring and preprocessing hazardous chemical storage multi-source data of a storage area, wherein the hazardous chemical storage multi-source data comprises time series data and spatial topology data; A risk prediction module for inputting the preprocessed time series data into a pre-constructed time dynamic risk prediction model, extracting time-sensitive features in the time series data through the time dynamic risk prediction model, outputting a risk level time series curve for a future preset time length according to the time-sensitive features, and associating and labeling storage units located within a preset risk threshold range in the risk level time series curve; A labeling module for labeling high-risk time period windows in the risk level time series curve where the risk level exceeds a preset safety threshold; A risk assessment module for constructing a spatial diffusion risk assessment model based on the preprocessed spatial topology data, when the risk level time series curve is associated with a storage unit located within the preset risk threshold range, taking the storage unit as a target storage unit, extracting risk diffusion features of the target storage unit through the spatial diffusion risk assessment model and associating them to adjacent storage units, and outputting a dynamic risk spatial distribution result of the storage area; A coupling analysis module for spatiotemporal coupling analysis of the high-risk time period window and the dynamic risk spatial distribution result, correcting the analysis result based on a preset risk association rule, and outputting spatiotemporal coupling risk warning information; A decision generation module for generating risk control decision instructions for the hazardous chemical storage area based on the spatiotemporal coupling risk warning information and calling a preset risk control rule library.

[0101] The specific definition of a hazardous chemical storage accident risk analysis system can be found in the definition of a hazardous chemical storage accident risk analysis method described above and will not be further elaborated here. Each module in the aforementioned hazardous chemical storage accident risk analysis system can be implemented in whole or in part through software, hardware, or a combination thereof. Each of these modules can be embedded in or independent of a processor in a computer device in hardware form, or stored in a computer device memory in software form, allowing the processor to call and execute the corresponding operations of each module.

[0102] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 2 As shown. The computer device includes a processor, memory, network interface and database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used for data storage, data processing, predictive analysis, etc. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, it implements a method for analyzing the risk of hazardous chemical storage accidents.

[0103] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, a method for analyzing the risk of hazardous chemical storage accidents is implemented.

[0104] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, a method for analyzing the risk of hazardous chemical storage accidents is implemented.

[0105] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0106] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0107] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. All equivalent structural transformations made by using the contents of the present invention description and drawings under the inventive concept of the present invention, or direct / indirect application in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A method for analyzing the risk of hazardous chemical storage accidents, characterized in that: The steps include: Collecting and preprocessing multi-source data on hazardous chemical storage in the storage area, wherein the multi-source data includes time series data and spatial topology data; Input the preprocessed time series data into a pre-built time dynamic risk prediction model, extract the time-sensitive features in the time series data through the time dynamic risk prediction model, output a risk level time series curve for a preset time period in the future based on the time sensitive features, and associate and mark the storage units in the risk level time series curve that are within the preset risk threshold range; Mark the high-risk time period window in the risk level time series curve where the risk level exceeds the preset safety threshold; A spatial diffusion risk assessment model is constructed based on the preprocessed spatial topological data. When the risk level time series curve is associated with a storage unit within the preset risk threshold range, the storage unit is used as the target storage unit. The risk diffusion characteristics of the target storage unit are extracted through the spatial diffusion risk assessment model and associated with the adjacent storage units, and the dynamic risk spatial distribution results of the storage area are output; Conduct spatiotemporal coupling analysis on the high-risk period window and the dynamic risk spatial distribution results, modify the analysis results based on the preset risk association rules, and output spatiotemporal coupling risk warning information; Based on the spatiotemporal coupling risk warning information, a preset risk management and control rule library is called to generate risk management and control decision instructions for hazardous chemicals storage areas.

2. The method for analyzing the risk of hazardous chemical storage accidents according to claim 1, wherein: The construction of the time dynamic risk prediction model specifically includes the following steps: Obtain historical time series sample data in the storage area, including storage unit operating condition data, environmental dynamic data, and human intervention data; pre-process the historical time series sample data to obtain a standardized time series feature set; and label the standardized time series feature set with a risk level label based on historical accident data of the same type of storage units within a preset time period, where pre-accident data is labeled as high risk and normal operating data is labeled as low risk; Constructing a time series prediction model including a feature extraction layer for extracting time-sensitive features and a prediction output layer for outputting a risk level time series curve; The standardized time series feature set marked with risk level labels is input into the time series prediction model architecture, and the model parameters are adjusted using the preset training strategy to obtain the trained time dynamic risk prediction model; The preprocessed time series data is input into the trained time dynamic risk prediction model, and the time-sensitive features in the time series data are extracted through the feature extraction layer to form a sensitive feature vector set containing storage unit operating condition characteristics, environmental dynamic characteristics and human intervention characteristics. The sensitive feature vector set is input into the prediction output layer, and the risk level time series curve of the future preset time length is output through the prediction output layer.

3. The method for analyzing the risk of hazardous chemical storage accidents according to claim 2, wherein: The step of inputting the sensitive feature vector set into the prediction output layer and outputting the risk level time series curve of the future preset time period through the prediction output layer specifically includes the following steps: The sensitive feature vector set is associated with the hazardous chemical type attribute of the corresponding storage unit to form an associated data set. Based on the associated data set, the sensitive feature vector set is grouped by hazardous chemical type, and the sensitive feature vectors in each group are sorted by time series to form a time series feature matrix stratified by hazardous chemical risk characteristics. The dynamic correlation relationship between adjacent vectors in each hierarchical time series feature matrix is ​​analyzed using a preset time series correlation algorithm. Based on the attention mechanism, accident-related feature vectors that match historical accident features are assigned a preset high-impact weight value, and other non-accident-related auxiliary feature vectors are assigned a preset basic weight value; wherein the historical accident features are extracted based on historical accident data of the same type of storage units within a preset time period; Based on the preset hazardous chemical type risk level mapping rules, combined with the risk level trend correction factor, the high impact weight value of the accident-related feature vector and the basic weight value of the auxiliary feature vector, the risk level value of each period within the preset time period in the future is calculated; The risk level values ​​of each time period are continuously arranged along the time axis to generate a risk level time series curve, wherein the risk level time series curve includes labels of storage units within a preset risk threshold range and spatially associated units of the storage units.

4. The method for analyzing the risk of hazardous chemical storage accidents according to claim 3, wherein: The method for determining the preset risk threshold includes the following steps: Based on the hazardous chemical type attribute of the storage unit, the basic safety threshold of the corresponding type of hazardous chemicals is retrieved as the initial risk threshold; Extract the risk level critical value of the corresponding type of hazardous chemicals when the historical accident occurred from the historical accident characteristics, and modify the initial risk threshold based on the risk level critical value to obtain the type-specific benchmark threshold; Real-time dynamic impact factors are extracted from the sensitive feature vector set, and the type-specific benchmark threshold is dynamically adjusted based on the extracted real-time dynamic impact factors to determine the preset risk threshold of the corresponding storage unit.

5. The method for analyzing the risk of hazardous chemical storage accidents according to claim 1, wherein: The step of constructing a spatial diffusion risk assessment model based on the preprocessed spatial topological data, and when a risk level time series curve is associated with a storage unit within a preset risk threshold range, taking the storage unit as a target storage unit, extracting the risk diffusion characteristics of the target storage unit through the spatial diffusion risk assessment model and associating it with adjacent storage units, and outputting a dynamic risk spatial distribution result of the storage area specifically includes the following steps: The pre-processed spatial topology data is analyzed hierarchically to extract the spatial correlation parameters and physical barrier levels between the target storage unit and adjacent storage units. Based on the diffusion characteristic parameters of hazardous chemicals in the target storage unit, a spatial diffusion risk assessment model is constructed. When the risk level time series curve is associated with a storage unit within the preset risk threshold range, the real-time risk level value of the target storage unit is obtained and input into the spatial diffusion risk assessment model. The risk diffusion characteristics of the target storage unit are extracted through the spatial diffusion risk assessment model. The risk diffusion characteristics include the risk diffusion rate, diffusion range radius and hazardous substance concentration attenuation coefficient of the target storage unit; Based on the extracted risk diffusion characteristics and the spatial association parameters between the target storage unit and the adjacent storage units, the risk diffusion characteristics of the target storage unit are associated with the corresponding adjacent storage units according to the preset association rules. The risk level of each adjacent storage unit affected by the diffusion is calculated, and an association relationship table containing the target storage unit, adjacent storage units, diffusion paths, and affected risk levels is generated. Based on the risk change trend of the risk level time series curve, the association table is superimposed and matched with the spatial map of the storage area to generate and output the dynamic risk spatial distribution results of the storage area that are dynamically updated over time. The dynamic risk spatial distribution results include the actual risk level of each storage unit in different time periods and the diffusion impact range labeling.

6. The method for analyzing the risk of hazardous chemical storage accidents according to claim 5, wherein: The step of extracting the risk diffusion characteristics of the target storage unit by using the spatial diffusion risk assessment model further includes the following steps: Obtain the physical properties of the target storage unit and the type of hazardous chemicals inside it, and determine whether the target storage unit belongs to a special scenario based on a preset judgment threshold; When the target storage unit belongs to a special scene, calling a preset special scene correction coefficient table to match the correction coefficient corresponding to the target storage unit, wherein the special scene correction coefficient table includes correction coefficients corresponding to several special scenes; The extracted risk diffusion feature is dynamically modified according to a modification coefficient corresponding to the target storage unit to obtain a modified risk diffusion feature.

7. The method for analyzing the risk of hazardous chemical storage accidents according to claim 1, wherein: The step of performing spatiotemporal coupling analysis on the high-risk time window and the dynamic risk spatial distribution result, correcting the analysis result based on the preset risk association rules, and outputting spatiotemporal coupling risk warning information specifically includes the following steps: Extracting time boundary parameters of the high-risk period window, the time boundary parameters including the high-risk starting time, duration and risk level peak time; Map the time boundary parameters of the high-risk period window to the dynamic risk spatial distribution results in the space-time dimension, match the risk spatial distribution state of the storage area corresponding to each time node, and form an initial space-time coupling matrix. The initial space-time coupling matrix contains the associated data of time nodes, storage units, risk levels, and diffusion ranges. Modifying the initial spatiotemporal coupling matrix based on preset risk association rules, wherein the preset risk association rules include a time priority weight rule, a spatial correlation rule, and a risk superposition rule; Based on the revised space-time coupling matrix, space-time coupling risk warning information is generated and output.

8. A hazardous chemical storage accident risk analysis system, used to implement the steps of a hazardous chemical storage accident risk analysis method according to any one of claims 1 to 7, characterized in that: include: A data acquisition module, configured to collect and preprocess multi-source data on hazardous chemicals stored in storage areas, wherein the multi-source data includes time series data and spatial topology data; A risk prediction module is configured to input the preprocessed time series data into a pre-built time dynamic risk prediction model, extract time-sensitive features from the time series data through the time dynamic risk prediction model, output a risk level time series curve for a preset time period in the future based on the time sensitive features, and associate and mark storage units in the risk level time series curve that are within a preset risk threshold range; A marking module, used for marking high-risk time period windows in the risk level time series curve where the risk level exceeds a preset safety threshold; The risk assessment module is used to construct a spatial diffusion risk assessment model based on the preprocessed spatial topological data. When the risk level time series curve is associated with a storage unit within the preset risk threshold range, the storage unit is used as the target storage unit. The risk diffusion characteristics of the target storage unit are extracted through the spatial diffusion risk assessment model and associated with the adjacent storage units, and the dynamic risk spatial distribution results of the storage area are output; The coupling analysis module is used to conduct spatiotemporal coupling analysis on the high-risk period window and the dynamic risk spatial distribution results, correct the analysis results based on the preset risk association rules, and output spatiotemporal coupling risk warning information; The decision generation module is used to call a preset risk management and control rule library based on the spatiotemporal coupling risk warning information to generate risk management and control decision instructions for hazardous chemical storage areas.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the hazardous chemical storage accident risk analysis method as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of a hazardous chemical storage accident risk analysis method as described in any one of claims 1 to 7 are implemented.

Images