Chemical industrial park safety management system and method based on Internet of Things

Abstract

The invention provides a chemical industry park safety management system and method based on the Internet of Things. The method comprises the following steps: collecting multi-source monitoring data in a chemical industry park to generate a dynamic monitoring data set; applying space-time causal constraints to multi-hop connection paths in the topological graph of the equipment in the chemical industrial park to generate path confidence distribution; constructing entity feature embedding of each device in the chemical industry park, and determining a time-varying causal incidence matrix through all entity feature embedding and path confidence distribution; dividing risk equipment groups with different risk levels, and constructing a risk propagation tensor of the equipment risk partition topology network in the chemical industry park through each risk equipment group; and generating a dynamic risk vector through the time-varying causal incidence matrix and the risk propagation tensor, and carrying out early warning on the safety risk in the chemical industry park based on the dynamic risk vector. By adopting the scheme of the invention, time-space registration can be carried out on the multi-source monitoring data of the sensing nodes of the Internet of Things, and the time-varying causal relationship between the devices can be mined, so that dynamic safety risk early warning can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Internet of Things, more particularly, the present application relates to a chemical industrial park safety management system and method based on Internet of Things. BACKGROUND

[0002] As the core place of chemical production, chemical industrial park is characterized by dense equipment, complex process, flammable and explosive, toxic and harmful medium, and safety management is crucial. With the development of Internet of Things technology, real-time collection of equipment operation data by distributed sensing nodes has become an important means of park safety monitoring, but the core is how to transform multi-source heterogeneous monitoring data into accurate risk warning information to cope with the complex correlation between devices and the dynamic change of running state.

[0003] In the prior art, chemical industrial park safety management relies on single sensor data or static topology analysis, such as monitoring single-point parameters by independent sensors and threshold alarm, or simple risk path analysis based on fixed device connection graph, etc. Such methods are difficult to realize time synchronization and spatial correlation calibration of distributed Internet of Things sensing node collected data, resulting in time and space deviation of device state representation, affecting the reliability of data basis for risk assessment, and traditional methods do not mine the dynamic causal relationship of multi-hop connection path in device topology graph, which cannot quantify the causal strength of multi-hop path evolution over time to identify the time-varying dependence relationship between devices to adapt to the characteristics of safety risk evolution over time in chemical industrial park, which may cause early warning lag or misjudgment. Therefore, how to realize time and space registration of multi-source monitoring data of Internet of Things sensing nodes and mine the time-varying causal relationship between devices to realize dynamic safety risk warning has become a difficult problem in the industry. SUMMARY

[0004] The present application provides a chemical industrial park safety management system and method based on Internet of Things, which can realize time and space registration of multi-source monitoring data of Internet of Things sensing nodes and mine the time-varying causal relationship between devices to realize dynamic safety risk warning.

[0005] In a first aspect, the present application provides a park risk dynamic warning method based on Internet of Things, which is used for dynamic identification of safety risk of chemical industrial park and warning in a chemical industrial park safety management system, wherein the distributed Internet of Things is provided with communication support by narrowband Internet of Things, and the method comprises: Collecting multi-source monitoring data in the chemical industrial park by distributed Internet of Things sensing nodes, and then generating a set of dynamic monitoring data synchronized in time and space; Obtaining a device topology graph of the chemical industrial park, imposing a time and space causal constraint on the multi-hop connection path in the device topology graph, and generating a path confidence distribution between devices in the chemical industrial park; constructing entity feature embeddings of each device in the chemical industrial park based on the dynamic monitoring dataset, determining a time-varying causal correlation matrix between devices in the chemical industrial park through all the entity feature embeddings and the path confidence distribution; dividing the devices in the chemical industrial park into risk groups according to the multi-dimensional state vectors of the devices in the dynamic monitoring dataset, obtaining risk device groups of different risk levels, performing cross-group risk feature propagation on the entity feature embeddings of the devices in each risk device group, and further constructing a risk propagation tensor of the risk zoning topology network of the devices in the chemical industrial park; generating a dynamic risk vector of the chemical industrial park through the time-varying causal correlation matrix and the risk propagation tensor, and warning about the safety risk in the chemical industrial park based on the dynamic risk vector.

[0006] In some embodiments, imposing a spatio-temporal causal constraint on the multi-hop connection path in the device topology graph to generate a path confidence distribution between devices in the chemical industrial park specifically includes: performing multi-scale path mining on the device topology graph to further determine the causal frame chain of each multi-hop connection path in the device topology graph; calculating the time-varying causal value of each causal frame chain by using a time-varying Granger causality test algorithm; generating the path confidence distribution between devices in the chemical industrial park based on the time-varying causal values of all the causal frame chains.

[0007] In some embodiments, constructing entity feature embeddings of each device in the chemical industrial park based on the dynamic monitoring dataset specifically includes: performing feature engineering processing on the dynamic monitoring dataset to further construct a multi-dimensional initial feature set of each device in the chemical industrial park; for each device in the chemical industrial park, performing feature fusion on the multi-dimensional initial feature set of the device to generate a graph convolution feature representation of the device during operation; determining the initial feature embedding of the device through the graph convolution feature representation to further obtain the initial feature embedding of each device in the chemical industrial park; performing contrast enhancement on the initial feature embeddings of all the devices to generate the entity feature embeddings of each device in the chemical industrial park.

[0008] In some embodiments, determining a time-varying causal correlation matrix between devices in the chemical industrial park through all the entity feature embeddings and the path confidence distribution specifically includes: constructing a feature correlation matrix of the devices in the chemical industrial park through all the entity feature embeddings; extracting path confidence degrees between devices in the path confidence distribution; Convolution and fusion of all path confidence and the feature association matrix generates a time-varying causal association matrix between the equipment in the chemical industry park.

[0009] In some embodiments, the equipment in the chemical industry park is divided into risk groups according to the multi-dimensional state vector of the equipment in the dynamic monitoring data set, and the risk equipment groups of different risk levels specifically include: In the dynamic monitoring data set, the multi-dimensional state vector of each equipment in the chemical industry park is extracted; The multi-dimensional state vector is decomposed in space and time, and then the equipment state feature space of the chemical industry park is constructed; In the equipment state feature space, all the equipment is clustered and divided according to the risk propagation index of the equipment, and the risk equipment groups of different risk levels are obtained.

[0010] In some embodiments, the entity feature embedding of the equipment in each risk equipment group is propagated across the group risk feature, and then the risk propagation tensor of the risk zoning topology network of the equipment in the chemical industry park is constructed, which specifically includes: For each risk equipment group, a risk association matrix of risk propagation between equipment is constructed based on the entity feature embedding of the equipment in the risk equipment group; The risk association matrix is modeled in time sequence evolution to generate a spatiotemporal risk propagation map of the risk equipment group, and then the spatiotemporal risk propagation map of each risk equipment group is obtained; The spatiotemporal risk propagation maps of all risk equipment groups are fused to generate a risk zoning topology network of the chemical industry park; The risk propagation tensor is extracted in the risk zoning topology network.

[0011] In some embodiments, the dynamic risk vector of the chemical industry park is generated by the time-varying causal association matrix and the risk propagation tensor, which specifically includes: The time-varying causal association matrix and the risk propagation tensor are multi-modal embedded and fused to generate a risk state transition matrix of the equipment in the chemical industry park; The risk state transition matrix is used for safety risk prediction of the chemical industry park to obtain a dynamic risk vector of the chemical industry park.

[0012] In a second aspect, the present application provides a chemical industry park safety management system based on Internet of Things, which comprises a risk dynamic early warning unit for dynamically identifying the safety risk of the chemical industry park and giving early warning, wherein the distributed Internet of Things is provided with communication support by narrowband Internet of Things, and the risk dynamic early warning unit comprises: The acquisition module is used for acquiring multi-source monitoring data in the chemical industry park through the distributed Internet of Things sensing node, and then generating a spatiotemporally synchronized dynamic monitoring data set; The processing module is configured to acquire a device topology of the chemical industrial park, impose a space-time causal constraint on a multi-hop connection path in the device topology, and generate a path confidence distribution between devices in the chemical industrial park. The processing module is configured to construct entity feature embeddings of the devices in the chemical industrial park based on the dynamic monitoring data set, determine a time-varying causal correlation matrix between the devices in the chemical industrial park by using all the entity feature embeddings and the path confidence distribution, and generate a dynamic risk vector of the chemical industrial park by using the time-varying causal correlation matrix and the risk propagation tensor. The processing module is configured to divide the devices in the chemical industrial park into risk groups according to multi-dimensional state vectors of the devices in the dynamic monitoring data set, obtain risk device groups of different risk levels, perform cross-group risk feature propagation on the entity feature embeddings of the devices in each risk device group, and further construct a risk propagation tensor of a risk zoning topology network of the devices in the chemical industrial park. The execution module is configured to generate a dynamic risk vector of the chemical industrial park by using the time-varying causal correlation matrix and the risk propagation tensor, and perform early warning on a safety risk in the chemical industrial park based on the dynamic risk vector.

[0013] In a third aspect, a computer device is provided, which includes a memory and a processor. The memory stores a code. The processor is configured to acquire the code and execute the above-mentioned method for early warning of a risk in a park based on Internet of Things.

[0014] In a fourth aspect, a computer readable storage medium is provided, which stores a computer program. The computer program is executed by a processor to implement the above-mentioned method for early warning of a risk in a park based on Internet of Things.

[0015] The technical scheme provided by the embodiments of the present application has the following beneficial effects: The chemical industry park safety management system and method based on the Internet of Things provided in the application first collects multi-source monitoring data in the chemical industry park through distributed Internet of Things sensing nodes, and then generates a spatio-temporal synchronous dynamic monitoring data set; an equipment topology graph of the chemical industry park is obtained, a spatio-temporal causal constraint is imposed on a multi-hop connection path in the equipment topology graph, a path confidence distribution between equipment in the chemical industry park is generated; entity feature embeddings of each equipment in the chemical industry park are constructed based on the dynamic monitoring data set, a time-varying causal correlation matrix between equipment in the chemical industry park is determined through all entity feature embeddings and the path confidence distribution; the equipment in the chemical industry park is divided into risk groups according to the multi-dimensional state vector of the equipment in the dynamic monitoring data set, and risk equipment groups of different risk levels are obtained; the entity feature embeddings of the equipment in each risk equipment group are subjected to cross-group risk feature propagation, and then a risk propagation tensor of the risk zoning topology network of the equipment in the chemical industry park is constructed; a dynamic risk vector of the chemical industry park is generated through the time-varying causal correlation matrix and the risk propagation tensor, and the safety risk in the chemical industry park is warned based on the dynamic risk vector.

[0016] It can be seen that the application carries out early warning on the safety risk in the chemical industry park based on the dynamic risk vector; first, the path confidence distribution is determined to obtain the path probability distribution representing the reliability of all devices on the multi-hop connection path in risk propagation, and the determination of the path confidence distribution quantifies the causal strength of the multi-hop path changing dynamically over time, solves the problem that the static analysis in the prior art cannot depict the time-varying dependence relationship between devices, provides a probabilistic path reliability basis for subsequent construction of the time-varying causal correlation matrix, and makes the subsequent risk assessment model more in line with the uncertainty characteristics of a complex system; then, the time-varying causal correlation matrix is determined to obtain the matrix quantifying the time-varying nature of the causal relationship between devices in the chemical industry park, the determination of the time-varying causal correlation matrix can quantify the time-varying nature of the causal relationship between devices by capturing the implicit correlation between devices changing over time, not only solves the defect that the static topology graph in the prior art cannot reflect the dynamic change of the device correlation over time, but also eliminates the distortion of the correlation analysis caused by the time-space bias of the multi-source data through the feature fusion of the time-space dimension, provides a structured carrier for dynamic modeling of the implicit dependence relationship between devices, and can more accurately identify abnormal propagation paths and key risk nodes, thereby improving the real-time performance and early warning accuracy of the safety monitoring of the chemical industry park; finally, the risk propagation tensor is determined to obtain a trend matrix representing the trend of the risk propagation strength between different risk groups in the chemical industry park changing over time, the determination of the risk propagation tensor not only integrates the spatial correlation between the risk device groups, but also captures the dynamic evolution of the risk propagation over time, solves the defect that the static analysis in the prior art cannot depict the time-space coupling characteristics of the risk propagation, and realizes the dynamic quantification of the risk diffusion law by intercepting the time series data through a sliding time window to extract the propagation strength, thereby providing a structured data carrier for modeling the time-space evolution of the risk propagation, and upgrading the risk analysis from "static single-point evaluation" to "dynamic global tracking"; in summary, based on the above scheme, the multi-source monitoring data of the Internet of Things sensing nodes can be time-space registered, and the time-varying causal relationship between devices can be mined to realize dynamic safety risk early warning. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is an example flowchart of a method for dynamic early warning of park risks based on the Internet of Things according to some embodiments of the application; Figure 2 is an example flowchart of determining entity feature embedding according to some embodiments of the application; Figure 3 is an operation flowchart of determining a risk device group according to some embodiments of the application; Figure 4 is a structural schematic diagram of a risk dynamic early warning unit according to some embodiments of the application; Figure 5 is an internal structure diagram of a computer device for implementing a method for dynamic early warning of park risks based on the Internet of Things according to some embodiments of the application. DETAILED DESCRIPTION

[0018] In order to better understand the technical solutions of the present application, the technical solutions of the present application will be described in detail below in combination with the drawings of the specification and specific embodiments.

[0019] Reference Figure 1 The figure is an exemplary flow chart of an Internet of Things-based park risk dynamic early warning method according to some embodiments of the present application, which mainly includes the following steps: In step 101, multi-source monitoring data in the chemical industry park is collected by distributed Internet of Things sensing nodes, and then a time-space synchronous dynamic monitoring data set is generated.

[0020] It should be noted that in the present application, the dynamic monitoring data set is a time-space synchronous data set composed of multiple monitoring data in the chemical industry park. The dynamic monitoring data set can provide standardized data support for the extraction of equipment risk characteristics and relationship modeling in the chemical industry park, ensuring the consistency of different sensor data in time series and spatial coordinates, and avoiding risk misjudgment caused by time-space deviation. In specific implementation, the following methods can be used to generate the time-space synchronous dynamic monitoring data set by collecting multi-source monitoring data in the chemical industry park through distributed Internet of Things sensing nodes, that is, first, for each device in the chemical industry park, the combustible gas concentration, operating temperature, operating pressure and medium flow of the device can be collected every preset collection time interval (such as half an hour) through the sensors (such as catalytic combustion gas sensor, thermocouple temperature sensor, piezoelectric pressure sensor and electromagnetic flowmeter) in the distributed Internet of Things sensing nodes deployed in the device, and the collection of all combustible gas concentration, operating temperature, operating pressure and medium flow is used as multi-source monitoring data of the device. Through the above steps, the multi-source monitoring data of each device in the chemical industry park can be obtained. Then, the multi-source monitoring data of all devices can be time calibrated by using an improved Kalman filter algorithm (such as adaptive extended Kalman filter), that is, time alignment is realized by setting a sliding time window (such as a window size of 5 minutes and a sliding step of 1 minute), so that time-synchronized multi-source monitoring data is obtained. The time-space attention mechanism-based graph neural network is used to calibrate the spatial calibration of all time-synchronized multi-source monitoring data, and the time-space synchronous dynamic monitoring data set is obtained. The graph neural network based on the time-space attention mechanism constructs a graph structure based on the spatial distribution of the distributed Internet of Things sensing nodes, automatically learns the spatial correlation of multi-source monitoring data of different devices through an attention weight matrix, and outputs the time-space synchronous dynamic monitoring data set. The multi-source monitoring data is a data set covering the combustible gas concentration, operating temperature, operating pressure and medium flow during the operation of the equipment in the chemical industry park. The multi-source monitoring data can provide multi-dimensional information input basis for risk analysis in the chemical industry park, comprehensively cover equipment status, environmental parameters and other dimensions, eliminate the monitoring blind area of single data perspective, improve the integrity of risk identification, and the distributed Internet of Things is supported by narrowband Internet of Things for communication.

[0021] In step 102, the device topology graph of the chemical industry park is obtained, time-space causal constraints are imposed on the multi-hop connection path in the device topology graph, and the path confidence distribution between the devices in the chemical industry park is generated.

[0022] In a specific implementation, the device topology graph of the chemical industrial park can be obtained in the following manner: the physical layout information of the devices in the chemical industrial park can be obtained by reading the computer-aided design drawings of the chemical industrial park, and a graph database (such as Neo4j) can be used to construct a topology graph according to the physical layout information of the devices in the chemical industrial park, the devices (such as reaction kettles, pipelines, valves, etc.) are defined as nodes, and the physical connection relationship (such as material flow direction, control signal transmission path, etc.) is defined as an edge, thereby obtaining the device topology graph of the chemical industrial park; wherein the device topology graph is a directed graph representing the spatial layout and connection relationship of the devices in the chemical industrial park, the device topology graph accurately presents the material flow and energy flow paths between the devices by fusing the physical layout and industrial control connection data, provides a structured model basis for causal constraint path analysis, and makes the risk propagation path identification more consistent with the actual process logic.

[0023] In some embodiments, the path confidence distribution between the devices in the chemical industrial park can be obtained by imposing a spatiotemporal causal constraint on the multi-hop connection paths in the device topology graph in the following steps: Multi-scale path mining is performed on the device topology graph, and then the causal frame chain of each multi-hop connection path in the device topology graph is determined; The time-varying causal value of each causal frame chain is calculated using a time-varying Granger causality test algorithm; The path confidence distribution between the devices in the chemical industrial park is generated based on the time-varying causal values of all causal frame chains.

[0024] In a specific implementation, the multi-scale path mining of the device topology graph and the determination of the causal frame chain of each multi-hop connection path in the device topology graph can be achieved in the following manner. First, the device topology graph can be processed in layers, divided into three abstract levels of physical layer, control layer and information layer, the physical connection relationship between devices is extracted based on computer-aided design drawings to construct an undirected base graph in the physical layer, the control signal flow between devices is obtained through industrial protocols (such as Modbus protocol) to construct a directed control graph in the control layer, and the semantic association between devices is defined and a semantic network graph is constructed in the information layer. Then, an improved search algorithm (such as depth-first search algorithm) is used in combination with chemical field knowledge rules (such as material flow direction constraints) to mine multi-hop connection paths in the graph structures of the three abstract levels in parallel, to obtain all multi-hop connection paths in the device topology graph. Finally, for each multi-hop connection path, a time window mechanism (such as a window size of 5 minutes and a sliding step of 1 minute) can be introduced to divide the multi-hop connection path into continuous time segments, the path state in each time segment constitutes a causal frame, and adjacent causal frames are connected in time sequence to form a causal frame chain. Through the above steps, the causal frame chain of each multi-hop connection path in the device topology graph can be obtained. The causal frame chain is a causal relationship time sequence representing the risk propagation time evolution process of the multi-hop connection path in the device topology graph. The causal frame chain accurately captures the time sequence dependence of risk propagation between devices through the coupling modeling of time slicing and path state, avoids the distortion of causal relationship caused by static analysis, and provides structured data support with time sequence dimension for dynamic risk assessment. The multi-hop connection path refers to an indirect connection path from a starting device node to a target device node in the device topology graph of the chemical industry park, which needs to pass through two or more intermediate device nodes.

[0025] In a specific implementation, the time-varying causality value of each causality frame chain can be calculated by using a time-varying Granger causality test algorithm in the following manner: for each causality frame chain, first, the causality frame chain is processed by using a sliding window mechanism, the window size can be set to 20 time points and the sliding step size can be set to 5 time points, a vector autoregressive model is constructed based on the causality frame chain in each window (i.e., the current device state vector is equal to the sum of the product of the state vectors at multiple previous time points and the coefficient matrix plus an error term), and the optimal lag order of the vector autoregressive model is automatically determined by using the Bayesian information criterion; then, Granger causality test is performed based on all vector autoregressive models, the null hypothesis is that "the historical value of device A cannot cause the current value of device B", the probability of the hypothesis being true is evaluated by using the F test statistic, the F test statistic is converted into a causality strength value (i.e., the logarithm of the negative F test statistic), and the causality strength value is taken as the time-varying causality value of the causality frame chain. Through the above steps, the time-varying causality value of each causality frame chain can be obtained. The time-varying causality value is a dynamic indicator representing the causality influence strength between devices in a multi-hop connection path, and can reflect the causality correlation degree of device state changes at different time points. By quantifying the time-varying characteristics of the causality relationship in real time, the dynamic fluctuations of the device operating parameters in the chemical industrial park are adapted, the problem that the traditional fixed-parameter causality analysis cannot capture instantaneous causality changes is solved, and the real-time performance and accuracy of risk propagation path identification are improved.

[0026] In a specific implementation, the path confidence distribution between devices in the chemical industrial park can be generated based on the time-varying causality values of all causality frame chains in the following manner: a Dirichlet process mixture model can be used to model the time-varying causality values of all causality frame chains to obtain the path confidence distribution between devices in the chemical industrial park. The Dirichlet process mixture model can convert all time-varying causality values into a high-dimensional feature vector (i.e., a vector containing statistical characteristics such as the mean, variance, and trend slope of all time-varying causality values), assume that the high-dimensional feature vector follows a Gaussian mixture distribution and automatically determine the number of mixture components by using a Dirichlet process, estimate the model parameters by using a variational inference algorithm, and iteratively update the variational parameters until convergence. After convergence, each causality frame chain is assigned to a different cluster, the cluster center represents a typical causality pattern, the posterior probability of each causality frame chain belonging to each cluster is calculated, and the posterior probability is taken as a parameter of the path confidence distribution. The path confidence distribution of the probabilistic causality between devices is generated by using Monte Carlo sampling.

[0027] It should be noted that in the present application, the path confidence distribution is a causal path probability distribution representing the reliability degree of all devices on the multi-hop connection path in risk propagation. The path confidence distribution automatically determines the confidence level of the causal path in a data-driven manner, avoids the subjectivity of artificial experience assignment, provides probabilistic risk propagation weights for time-varying causal correlation matrix construction, and makes the subsequent risk assessment model more consistent with the uncertainty characteristics of complex systems.

[0028] In step 103, the entity feature embedding of each device in the chemical industrial park is constructed based on the dynamic monitoring data set, and the time-varying causal correlation matrix between the devices in the chemical industrial park is determined through all the entity feature embeddings and the path confidence distribution.

[0029] In some embodiments, with reference to Figure 2 The figure is an exemplary flow chart for determining entity feature embedding according to some embodiments of the present application. The construction of the entity feature embedding of each device in the chemical industrial park based on the dynamic monitoring data set can be implemented by the following steps in the present application: In step 1031, the dynamic monitoring data set is subjected to feature engineering processing, and then a multi-dimensional initial feature set of each device in the chemical industrial park is constructed; In step 1032, for each device in the chemical industrial park, the multi-dimensional initial feature set of the device is subjected to feature fusion to generate a graph convolution feature representation of the device in operation; In step 1033, the initial feature embedding of the device is determined through the graph convolution feature representation, and then the initial feature embedding of each device in the chemical industrial park is obtained; In step 1034, the initial feature embeddings of all devices are compared and enhanced to generate the entity feature embedding of each device in the chemical industrial park.

[0030] In a specific implementation, the feature engineering processing on the dynamic monitoring data set and the construction of the multi-dimensional initial feature set of each device in the chemical industrial park can be realized in the following manner, that is, first, for each device in the chemical industrial park, the multi-source monitoring data of the device is extracted from the dynamic monitoring data set, and multi-dimensional feature extraction is performed on the multi-source monitoring data, that is, the time domain features (including mean, standard deviation, and peak factor) of the combustible gas concentration, operating temperature, operating pressure, and medium flow in the multi-source monitoring data are calculated to represent the parameter fluctuation characteristics, and the frequency domain features (such as main frequency amplitude and harmonic energy proportion) of the combustible gas concentration, operating temperature, operating pressure, and medium flow in the multi-source monitoring data are extracted by fast Fourier transform to capture periodic abnormal signals; then, the set of all time domain features and all frequency domain features is taken as the multi-dimensional initial feature set of the device, and the multi-dimensional initial feature set of each device in the chemical industrial park can be obtained through the above steps; wherein the multi-dimensional initial feature set is a basic feature set formed by representing the multi-dimensional indexes of the time domain and the frequency domain of the device in the chemical industrial park, which can provide multi-perspective basic data input for device state analysis, covering the time sequence fluctuation and frequency characteristics of device operation, avoiding the one-sidedness of single-dimensional features, and providing comprehensive original feature materials for subsequent space-time feature fusion.

[0031] In a specific implementation, the feature fusion on the multi-dimensional initial feature set of the device can be realized in the following manner, that is, an existing graph convolution network model (such as a space-time graph convolution network model) can be loaded, and the feature fusion on the multi-dimensional initial feature set of all devices is performed through the graph convolution network model to generate the graph convolution feature representation of the device operation; wherein the graph convolution network model defines a graph structure based on the device topology graph of the chemical industrial park, the nodes are devices, and the edges are physical connection relationships, and the graph Laplacian matrix is used to represent the spatial correlation of the devices, in the time dimension, a one-dimensional convolution kernel (such as a kernel size of 10 time steps) is applied to the multi-dimensional initial feature set of the device to capture the evolution law of the parameters over time, and the space neighborhood features and the time sequence features are alternately processed through the space-time convolution layer to realize feature fusion, and the feature matrix containing space-time coupling information is output as the graph convolution feature representation; wherein the graph convolution feature representation is a feature matrix representing the space-time correlation characteristics of the device operation state, each element in the graph convolution feature representation represents the space-time correlation state of the device at the corresponding time step, the spatial connection relationship of the device is modeled through the graph structure, and the parameter dynamics are captured through the time sequence convolution, which can effectively extract the space-time propagation law of the device state change and solve the problem that the traditional method is difficult to handle complex spatial correlation.

[0032] In a specific implementation, the initial feature embedding of the device determined by the graph convolution feature representation can be achieved in the following manner: the graph convolution feature representation can be input into the fully connected layer of the graph convolution network model for nonlinear mapping to generate an initial feature embedding vector of the device; wherein the fully connected layer automatically focuses on risk-sensitive feature dimensions such as hyperthermia and hyperbaric oxygen through weighted aggregation of the risk-sensitive feature dimensions after soft-max normalization of the graph convolution feature representation, and performs splicing and linear transformation on the output result to generate an initial feature embedding containing individual state and neighborhood interaction information of the device (such as 128 dimensions); wherein the initial feature embedding refers to a low-dimensional vector containing individual state and neighborhood interaction information of the device, which focuses on risk-sensitive feature dimensions such as hyperthermia and hyperbaric oxygen, suppresses irrelevant noise interference, and makes the device features more suitable for safety risk analysis requirements by strengthening the expression of key parameters for risk assessment.

[0033] It should be noted that in this application, the entity feature embedding is a high-discrimination feature vector representing the individual state and neighborhood interaction information of the device. The entity feature embedding improves the feature discrimination ability by maximizing the similarity of similar devices and minimizing the difference between dissimilar devices, thereby solving the problem of insufficient discrimination of traditional feature embedding and making the features of devices with different risk levels significantly separated in space, providing a more reliable feature basis for subsequent risk group division and propagation analysis.

[0034] In a specific implementation, the initial feature embedding of all devices is compared and enhanced to generate the entity feature embedding of each device in the chemical industrial park in the following manner: first, the same device positive sample pairs can be generated according to the process unit (such as reaction unit and distillation unit) and spatial position (such as within a radius of 50 meters) to which the device belongs as positive sample screening conditions, and negative sample pairs can be generated by adding Gaussian noise with a mean of 0 and a standard deviation of 0.1 to the entity feature embedding; then, the cosine similarity of the entity feature embedding of each device with the positive and negative sample embeddings is calculated using an existing contrast loss function (such as an information noise contrast loss function), and the optimization goal is set to maximize the similarity of the positive sample pairs to 1 and minimize the similarity of the negative sample pairs to -1. After 100 rounds of iterative training, the entity feature embedding corresponding to each device is generated.

[0035] In some embodiments, the time-varying causal correlation matrix between devices in the chemical industrial park can be determined by all entity feature embeddings and the path confidence distribution in the following steps: Construct a feature correlation matrix of devices in the chemical industrial park by all entity feature embeddings; Extract the path confidence between each device in the path confidence distribution; Convolve and fuse all path confidences with the feature correlation matrix to generate a time-varying causal correlation matrix between devices in the chemical industrial park.

[0036] In a specific implementation, the feature correlation matrix of the equipment in the chemical industrial park can be constructed by all entity feature embeddings in the following manner: first, the multi-dimensional correlation calculation of the entity feature embeddings of all equipment can be performed by using a multi-head attention mechanism, that is, the entity feature embeddings of each equipment are respectively mapped into query vectors, key vectors and value vectors, the attention scores between different equipment are calculated by dot product operation in multiple (such as 8) parallel attention heads, and the attention scores of all attention heads are weighted and summed to obtain the multi-dimensional correlation scores between different equipment; then, a blank matrix is constructed by taking the sequence composed of all equipment as the row and column of the matrix, and all multi-dimensional correlation scores are filled into the corresponding positions in the matrix to obtain the feature correlation matrix of the equipment in the chemical industrial park; wherein, the feature correlation matrix is a matrix representing the running state correlation closeness between the equipment in the chemical industrial park, and the feature correlation matrix models the correlation relationship based on the equipment feature and state difference, and provides a similarity measurement basis for the time-varying causal correlation matrix of the equipment individual state.

[0037] In a specific implementation, the path confidence between each equipment can be extracted in the path confidence distribution in the following manner: for each pair of equipment in the chemical industrial park, all possible multi-hop connection paths between the two equipment can be traversed in the path confidence distribution, the posterior probability of each multi-hop connection path is obtained as the confidence of each multi-hop connection path, and the average value of all multi-hop connection path confidences is taken as the path confidence between the two equipment, so that the path confidence between each pair of equipment in the chemical industrial park can be obtained by the above steps, thereby obtaining the path confidence between each equipment; wherein, the path confidence is an index representing the reliability of the risk propagation path between two equipment in the time sequence dynamics, which can reflect the uncertainty and time-varying characteristics of the equipment causal relationship in the complex pipe network of the chemical industrial park, and improve the credibility of the risk propagation path analysis.

[0038] In a specific implementation, all path confidences are convoluted with the feature association matrix to generate the time-varying causal association matrix between devices in the chemical industrial park. The following method can be used to achieve this: a pre-existing convolutional neural network (such as a three-dimensional convolutional neural network) can be used to convolute the feature association matrix and all path confidences to generate the time-varying causal association matrix between devices in the chemical industrial park. In the convolutional neural network, a two-dimensional separable convolution kernel can be used to convolute the feature association matrix and the path confidence matrix, i.e., a 1x3 convolution kernel is used to perform row convolution to extract the dependence in the row direction, and then a 3x1 convolution kernel is used to perform column convolution to extract the dependence in the column direction. Two one-dimensional convolutions are used to replace the standard two-dimensional convolution to reduce the computational complexity. Finally, the convolution result is normalized by a soft-max normalization function to generate a time-varying causal association matrix with a dimension of device number x device number. Each element in the time-varying causal association matrix represents the time variation of the causal relationship between the corresponding device pair.

[0039] It should be noted that in this application, the time-varying causal association matrix is a matrix that quantifies the time variation of the causal relationship between devices in the chemical industrial park. By capturing the time-varying implicit association between devices, the time-varying causal association matrix can more accurately identify abnormal propagation paths and key risk nodes, thereby improving the real-time performance and early warning accuracy of safety monitoring in the chemical industrial park.

[0040] In step 104, the devices in the chemical industrial park are divided into risk groups according to the multi-dimensional state vectors of the devices in the dynamic monitoring data set, and risk device groups of different risk levels are obtained. The entity feature embedding of the devices in each risk device group is propagated across the groups, and then a risk propagation tensor of the risk zoning topology network of the devices in the chemical industrial park is constructed.

[0041] In some embodiments, with reference to Figure 3 The figure is an operation flow chart for determining the risk device groups according to some embodiments of the present application. In this application, the devices in the chemical industrial park are divided into risk groups according to the multi-dimensional state vectors of the devices in the dynamic monitoring data set, and risk device groups of different risk levels are obtained. The following steps can be used to achieve this: The multi-dimensional state vectors of all devices in the chemical industrial park are extracted from the dynamic monitoring data set; The multi-dimensional state vectors are decomposed in space and time to construct a device state feature space for the chemical industrial park; All devices are clustered and divided according to their risk propagation indicators in the device state feature space, and risk device groups of different risk levels are obtained.

[0042] In a specific implementation, the multi-dimensional state vector of each device in the chemical industrial park can be extracted from the dynamic monitoring data set in the following manner: for each device in the chemical industrial park, the multi-source monitoring data of the device can be extracted from the dynamic monitoring data set, and the vector composed of the data (such as combustible gas concentration, operating temperature, operating pressure, and medium flow) in each time window can be taken as a time slice, and the multi-dimensional vector composed of the time slices in all time windows can be taken as the multi-dimensional state vector of the device, so that the multi-dimensional state vector of each device in the chemical industrial park can be obtained through the above steps; wherein the multi-dimensional state vector is a three-dimensional data structure organized by the multi-source monitoring data of the device in the dynamic monitoring data according to the time dimension, and the multi-dimensional state vector can provide standardized input for the spatio-temporal feature decomposition of high-dimensional complex data, and solve the problem that traditional table data is difficult to capture multi-dimensional coupling relationship.

[0043] In a specific implementation, the spatio-temporal feature decomposition of all multi-dimensional state vectors and the construction of the device state feature space of the chemical industrial park can be implemented in the following manner: first, the existing feature decomposition technology (such as tensor train decomposition) can be used to reduce the dimension and extract the features of the multi-dimensional state vector of all devices, decompose the original multi-dimensional vector into the product form of multiple low-dimensional vectors, and perform principal component analysis on the decomposed low-dimensional vectors to extract principal components, so as to obtain the feature vector describing the device state; then, the device state feature space with the feature vector of the state as the coordinate axis is constructed through the position of each device in the space in the device topology diagram, so that the device state feature space of the chemical industrial park is obtained; wherein the device state feature space is a low-dimensional feature vector space mapping the device operating state with the key feature vector as the coordinate axis, the device state feature space eliminates redundant information and retains core features through the dimension reduction technology, compresses the high-dimensional data to a computable dimension, provides efficient and representative feature input for subsequent clustering analysis, and improves the calculation efficiency and accuracy of risk group division.

[0044] In a specific implementation, the following method can be used to obtain risk equipment groups of different risk levels by clustering and dividing all the equipment in the equipment state feature space according to the risk propagation indicators of the equipment: first, the risk propagation indicators of each equipment, including the equipment state fluctuation amplitude (i.e., the standard deviation of the operating temperature) and the change trend (i.e., the slope of the operating pressure time series), can be extracted in the equipment state feature space; then, the existing clustering algorithm (such as the improved density peak clustering algorithm) is used to calculate the weighted distance between the equipment with the risk propagation indicators as the weight, and the clustering center is automatically identified through the decision diagram, and all the equipment is clustered and divided to obtain risk equipment groups of different risk levels; for each clustering center, the density peak clustering algorithm can calculate its risk propagation entropy based on the information entropy theory as a quantitative indicator of the risk level, divide the equipment into different risk levels, and use semi-supervised learning for secondary classification of the boundary equipment, adjust the classification boundary using a small number of samples labeled by domain experts, and obtain risk equipment groups of different risk levels.

[0045] It should be noted that in this application, the risk equipment group is a cluster of equipment with similar risk propagation patterns, and through the risk equipment group, hierarchical management of the equipment in the chemical industrial park can be realized, the safety resources can be focused on the high-risk group, and the problem of strong subjectivity and low efficiency of traditional manual grading can be solved.

[0046] In some embodiments, the entity feature embedding of the equipment in each risk equipment group is used for cross-group risk feature propagation, and then the risk propagation tensor of the risk zoning topology network of the equipment in the chemical industrial park can be constructed by the following steps: For each risk equipment group, a risk correlation matrix of the risk propagation between the equipment in the risk equipment group is constructed based on the entity feature embedding of the equipment in the risk equipment group; The risk correlation matrix is modeled for time evolution to generate a spatiotemporal risk propagation atlas of the risk equipment group, and then the spatiotemporal risk propagation atlas of each risk equipment group is obtained; The spatiotemporal risk propagation atlases of all risk equipment groups are fused to generate a risk zoning topology network of the chemical industrial park; A risk propagation tensor is extracted in the risk zoning topology network.

[0047] In a specific implementation, the risk correlation matrix of the risk propagation between devices in the risk device group can be constructed based on the entity feature embedding of the devices in the risk device group, which can be implemented in the following manner: first, input the entity feature embedding of all devices in the risk device group into a graph attention network, map the entity feature embedding of each device into a query vector, a key vector, and a value vector through the multi-head attention mechanism of the graph attention network, and calculate the risk propagation strength between different devices through dot product; then, construct a blank matrix with the sequence of all devices as the row and column of the matrix, and fill the obtained risk propagation strength into the corresponding position in the matrix to obtain the risk correlation matrix of the risk propagation between devices in the risk device group; wherein the risk correlation matrix is a matrix quantifying the risk propagation strength between devices in the risk device group, which dynamically learns the risk propagation relationship based on the real-time running features of the devices, can reflect the risk conduction correlation driven by the state features of the devices, accurately capture the risk conduction path of nonlinear coupling, and solve the problem that the static weight cannot adapt to the change of the device state.

[0048] In a specific implementation, the risk correlation matrix is modeled for time evolution to generate the spatiotemporal risk propagation graph of the risk device group, which can be implemented in the following manner: a time sequence pulsation factor can be introduced to model the risk correlation matrix for time evolution using a gated recurrent unit, the risk correlation matrix at each time step is taken as an input sequence, and the spatiotemporal risk propagation graph containing time sequence information is generated by controlling the transmission and update of risk information through the forget gate and update gate of the gated recurrent unit; wherein the spatiotemporal risk propagation graph is a graph that integrates the risk propagation strength between devices and the dynamic change law of the time dimension, which retains historical risk propagation information and captures dynamic patterns such as attenuation and enhancement of risk features over time through the forget gate and update gate mechanism, and provides structured model support for real-time prediction and trend analysis of the risk of the chemical industrial park in the time sequence dimension.

[0049] In a specific implementation, the spatiotemporal risk propagation graphs of all risk device groups are fused to generate the risk zoning topology network of the chemical industrial park, which can be implemented in the following manner: existing graph fusion algorithms (such as the graph community discovery algorithm based on modularity optimization) can be used to fuse all spatiotemporal risk propagation graphs to generate the risk zoning topology network of the chemical industrial park; wherein the graph community discovery algorithm based on modularity optimization first constructs a global risk propagation graph, the nodes are risk device groups, and the edge weight is the risk propagation strength between risk device groups (i.e. the similarity between the spatiotemporal risk propagation graphs of risk device groups), then the community is divided by maximizing the modularity in each iteration of the graph community discovery algorithm, and a spatial constraint term is introduced to ensure that the zoning conforms to the physical layout of the chemical industrial park, through iterative optimization, the devices are divided into different risk communities to generate the risk zoning topology network.

[0050] It should be noted that in the present application, the risk partition topology network is a network composed of all equipment in the chemical industry park by dividing the risk propagation mode, which divides the areas with similar risk propagation modes in a modular clustering manner, combines the physical layout of the chemical industry park, and divides the complex system into hierarchical risk communities to provide a network structure basis for "partitioning measures" of park safety management.

[0051] In a specific implementation, the risk propagation tensor in the risk partition topology network can be extracted in the following manner: the time sequence data in the risk partition topology network can be intercepted by using a sliding time window mechanism (for example, the window size is set to 10 time steps, and the sliding step size is set to 2 time steps), and the risk propagation intensity between any two risk groups in each time window is extracted, and then all the extracted risk propagation intensities are organized according to the three-dimensional dimensions of "risk group pair-time step-propagation intensity" to obtain the risk propagation tensor; wherein the first dimension of the risk propagation tensor is the index of the risk group pair, the second dimension is the time step, and the third dimension is the risk propagation intensity in the corresponding window.

[0052] It should be noted that in the present application, the risk propagation tensor is a trend matrix representing the change of risk propagation intensity between different risk groups in the chemical industry park over time, which integrates the spatial correlation and time sequence dynamic characteristics of risk propagation, provides structured data support for accurately capturing risk diffusion rules and generating dynamic risk vectors, and improves the comprehensiveness and timeliness of park risk early warning.

[0053] In step 105, a dynamic risk vector of the chemical industry park is generated by the time-varying causal correlation matrix and the risk propagation tensor, and a safety risk in the chemical industry park is warned based on the dynamic risk vector.

[0054] In some embodiments, the dynamic risk vector of the chemical industry park can be generated by the time-varying causal correlation matrix and the risk propagation tensor in the following steps: The time-varying causal correlation matrix and the risk propagation tensor are subjected to multi-modal embedding fusion to generate a risk state transition matrix of the equipment in the chemical industry park; The safety risk of the chemical industry park is predicted by the risk state transition matrix to obtain the dynamic risk vector of the chemical industry park.

[0055] It should be noted that in the present application, the dynamic risk vector is a multi-dimensional parameter vector dynamically depicting the spatio-temporal evolution state of the risk of the chemical industry park, including risk intensity and propagation speed, which not only reflects the current risk level of the chemical industry park, but also predicts the propagation trend and influence range, providing a full-chain data support of "situation awareness-trend prediction-emergency response" for park safety management, and solving the problem that the traditional static risk assessment cannot capture dynamic changes.

[0056] In a specific implementation, the multi-modal embedding fusion of the time-varying causal correlation matrix and the risk propagation tensor can be implemented in the following manner to generate the risk state transition matrix of the equipment in the chemical industrial park: first, the time-varying causal correlation matrix can be expanded by rows into a vector sequence, the time sequence dependency between the equipment is calculated through a multi-head self-attention mechanism, and the causal feature representation of all equipment is outputted, then a convolution kernel (such as a 3x3x3 convolution kernel) in a three-dimensional convolutional neural network is used to perform three-dimensional convolution operation on the risk propagation tensor to extract the spatio-temporal local features in the risk propagation tensor, then the causal feature representation is taken as a query vector, the spatio-temporal local features are taken as key-value pairs, the attention weight is calculated and weighted summation is performed to realize the interactive fusion of the two modal features, and the fused features are inputted into the full connection network of the three-dimensional convolutional neural network to generate the risk state transition matrix through an activation function, each element in the risk state transition matrix represents the probability of the equipment transitioning from the current state to other states, finally, a forgetting gate mechanism is introduced to update the historical state transition matrix exponentially to ensure that the risk state transition matrix can reflect the latest risk propagation trend; wherein, the risk state transition matrix is a probability matrix quantifying the transition probability of the equipment in the chemical industrial park between different risk states, the risk state transition matrix provides a structured state evolution basis for dynamic risk prediction, accurately captures the risk transfer law, and improves the scientificity and forward-looking of the safety risk early warning.

[0057] In a specific implementation, the safety risk prediction of the chemical industrial park through the risk state transition matrix can be implemented in the following manner to obtain the dynamic risk vector of the chemical industrial park: the risk propagation process of the chemical industrial park can be modeled based on the risk state transition matrix using a continuous-time Markov chain, and the transfer probability density function of the continuous-time Markov chain is approximately solved using Chebyshev polynomials to accelerate matrix exponential operation to generate a dynamic risk vector containing risk intensity values and propagation speed (i.e. the number of risk diffusion nodes per unit time).

[0058] In a specific implementation, the safety risk in the chemical industry park can be prewarned based on the dynamic risk vector in the following manner: a multi-dimensional safety prewarning system of the chemical industry park can be constructed based on the dynamic risk vector to prewarn the safety risk in the chemical industry park, that is, the indexes of each dimension in the dynamic risk vector are evaluated in real time by a risk threshold determination module, for the risk intensity dimension, the risk intensity value is compared with the preset three-level threshold (for example, the safety threshold is 0.3, the prewarning threshold is 0.6, and the high-risk threshold is 0.8), and the corresponding level of prewarning is triggered; for the propagation speed dimension, the change rate of the number of risk diffusion nodes per unit time is calculated, and when the change rate exceeds 0.5 for three consecutive time steps, a trend prewarning is triggered; and based on the dynamic risk vector, a long short-term memory network is used to predict the change trend of the risk index in the next 4 hours, and when the predicted value exceeds 120% of the current threshold, an advanced prewarning is triggered, and finally three-dimensional prewarning information including the real-time risk level, the evolution trend, and the influence range is formed, and then visualized on the electronic map of the park by a geographic information system.

[0059] In addition, another aspect of the present application, in some embodiments, the present application provides an Internet of Things-based safety management system for a chemical industry park, which comprises a risk dynamic prewarning unit, which refers to Figure 4 The figure is a structural schematic diagram of a risk dynamic prewarning unit according to some embodiments of the present application. The Internet of Things-based safety management system for a chemical industry park risk dynamic prewarning unit 400 comprises a collection module 401, a processing module 402, and an execution module 403, which are described as follows: The collection module 401 is mainly used to collect multi-source monitoring data in the chemical industry park through distributed Internet of Things sensing nodes, and then generate a spatiotemporally synchronized dynamic monitoring data set. The processing module 402 is mainly used to obtain the device topology graph of the chemical industry park, impose spatiotemporal causal constraints on the multi-hop connection path in the device topology graph, and generate the path confidence distribution between devices in the chemical industry park. It should be noted that the processing module 402 is also used to construct entity feature embeddings of each device in the chemical industry park based on the dynamic monitoring data set, and determine the time-varying causal correlation matrix between devices in the chemical industry park through all entity feature embeddings and the path confidence distribution. In addition, it should be noted that the processing module 402 is also used to divide the devices in the chemical industry park into risk groups according to the multi-dimensional state vector of the devices in the dynamic monitoring data set, obtain risk device groups of different risk levels, perform cross-group risk feature propagation on the entity feature embeddings of the devices in each risk device group, and then construct the risk propagation tensor of the device risk zoning topology network in the chemical industry park. The execution module 403 is mainly used for generating a dynamic risk vector of the chemical industry park through the time-varying causal correlation matrix and the risk propagation tensor, and warning the safety risk in the chemical industry park based on the dynamic risk vector.

[0060] The modules in the safety management system for the chemical industry park based on the Internet of Things can be implemented by software, hardware, or a combination thereof. The modules can be embedded in or independent of a processor in a computer device in hardware form, or stored in a memory in the computer device in software form, so as to be called and executed by the processor.

[0061] In addition, in an embodiment, the present application provides a computer device, which can be a server, and the internal structure diagram thereof can be as shown in Figure 5 The computer device includes a processor, a memory, and a network interface connected through 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 operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the data of the park risk dynamic warning method based on the Internet of Things. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement a park risk dynamic warning method based on the Internet of Things.

[0062] Those skilled in the art can understand that Figure 5 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. A specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0063] In an embodiment, a computer device is also provided, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-mentioned park risk dynamic warning method based on the Internet of Things.

[0064] In an embodiment, a computer readable storage medium is provided, which stores a computer program. The computer program is executed by a processor to implement the steps in the above-mentioned park risk dynamic warning method based on the Internet of Things.

[0065] In one embodiment, a computer program product or computer program is provided, which includes computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device performs the steps in the above-mentioned IoT-based park risk dynamic early warning method embodiments.

[0066] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, the computer program can include the processes of the above-mentioned embodiments. Any reference to memory, storage, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. The non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory. The volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, the RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0067] The technical features of the above embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, but as long as the combinations of the technical features do not exist, they should be considered as the scope of the present application.

[0068] The above embodiments only express several implementation manners of the present application, and the description is specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be pointed out that for those of ordinary skill in the art, without departing from the concept of the present application, some modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of the patent of the present application should be subject to the appended claims.

Claims

1. An IoT-based dynamic risk early warning method for chemical industrial parks, used by the safety management system of chemical industrial parks to dynamically identify safety risks and issue early warnings, wherein... The distributed Internet of Things (IoT) is supported by narrowband IoT for communication, characterized by the following steps: Multi-source monitoring data within the chemical industrial park is collected through distributed IoT sensing nodes, thereby generating a dynamic monitoring dataset that is synchronized in time and space. Obtain the equipment topology map of the chemical industrial park, apply spatiotemporal causal constraints to the multi-hop connection paths in the equipment topology map, and generate the path confidence distribution between equipment in the chemical industrial park; Based on the dynamic monitoring dataset, entity feature embeddings of each device in the chemical industrial park are constructed. The time-varying causal association matrix between devices in the chemical industrial park is determined by all entity feature embeddings and the path confidence distribution. Based on the multidimensional state vectors of the equipment in the dynamic monitoring dataset, the equipment in the chemical industrial park is divided into risk groups to obtain risk equipment groups with different risk levels. The entity feature embeddings of the equipment in each risk equipment group are used to propagate cross-group risk features, thereby constructing a risk propagation tensor of the risk partitioning topology network of equipment in the chemical industrial park. A dynamic risk vector for the chemical industrial park is generated using the time-varying causal correlation matrix and the risk propagation tensor, and an early warning of safety risks within the chemical industrial park is provided based on the dynamic risk vector.

2. The method as described in claim 1, characterized in that, Applying spatiotemporal causal constraints to the multi-hop connection paths in the equipment topology graph to generate the path confidence distribution between equipment within the chemical industrial park specifically includes: Multi-scale path mining is performed on the device topology graph to determine the causal frame chain of each multi-hop connection path in the device topology graph; The time-varying causality value of each causal frame chain is calculated using the time-varying Granger causality test algorithm; The path confidence distribution between equipment in the chemical industrial park is generated based on the time-varying causal values ​​of all causal frame chains.

3. The method as described in claim 1, characterized in that, The specific steps for constructing entity feature embeddings for various equipment within the chemical industrial park based on the aforementioned dynamic monitoring dataset include: The dynamic monitoring dataset is subjected to feature engineering processing to construct a multi-dimensional initial feature set for each device in the chemical industrial park. For each piece of equipment in the chemical industrial park, feature fusion is performed on the multidimensional initial feature set of the equipment to generate a graph convolution feature representation of the equipment during operation; The initial feature embedding of the equipment is determined by the graph convolution feature representation, thereby obtaining the initial feature embedding of each piece of equipment in the chemical industrial park. The initial feature embeddings of all equipment are compared and enhanced to generate the entity feature embeddings of each piece of equipment in the chemical industrial park.

4. The method as described in claim 1, characterized in that, Determining the time-varying causal correlation matrix between equipment within the chemical industrial park through all entity feature embeddings and the path confidence distribution specifically includes: Construct a feature association matrix for equipment within the chemical industrial park by embedding all entity features; Extract the path confidence scores between each device from the path confidence distribution; All path confidences are convolved and fused with the feature association matrix to generate a time-varying causal association matrix between equipment in the chemical industrial park.

5. The method as described in claim 1, characterized in that, Based on the multidimensional state vectors of the equipment in the dynamic monitoring dataset, the equipment in the chemical industrial park is divided into risk groups, resulting in risk equipment groups with different risk levels, specifically including: Extract the multidimensional state vectors of each piece of equipment in the chemical industrial park from the dynamic monitoring dataset; Spatiotemporal feature decomposition is performed on all multidimensional state vectors to construct the equipment state feature space of the chemical industrial park; In the device state feature space, all devices are clustered according to the risk propagation index of the devices to obtain risk device groups with different risk levels.

6. The method as described in claim 1, characterized in that, The risk propagation tensor for constructing a risk zoning topology network of equipment risk partitioning within a chemical industrial park involves embedding the entity features of equipment in each risk equipment group and propagating risk features across groups. Specifically, this includes: For each risk equipment group, a risk association matrix for risk propagation between equipment is constructed based on the entity feature embedding of the equipment in the risk equipment group; The risk correlation matrix is ​​modeled in a time series evolution to generate a spatiotemporal risk propagation map of risk equipment groups, thereby obtaining the spatiotemporal risk propagation map of each risk equipment group; The spatiotemporal risk propagation maps of all risky equipment groups are fused to generate a risk zoning topology network for the chemical industrial park. Extract the risk propagation tensor from the risk partitioning topology network.

7. The method as described in claim 1, characterized in that, The generation of the dynamic risk vector for the chemical industrial park through the time-varying causal correlation matrix and the risk propagation tensor specifically includes: Multimodal embedding fusion is performed on the time-varying causal correlation matrix and the risk propagation tensor to generate a risk state transition matrix for equipment within the chemical industrial park; The risk state transition matrix is ​​used to predict the safety risks of the chemical industrial park, resulting in a dynamic risk vector for the chemical industrial park.

8. A safety management system for chemical industrial parks based on the Internet of Things (IoT), comprising a dynamic risk early warning unit for dynamically identifying safety risks in the chemical industrial park and issuing early warnings, wherein... The distributed Internet of Things (IoT) is supported by narrowband IoT for communication, characterized in that the dynamic risk early warning unit includes: The data acquisition module is used to collect multi-source monitoring data in the chemical industrial park through distributed Internet of Things sensing nodes, and then generate a dynamic monitoring dataset that is synchronized in time and space. The processing module is used to obtain the equipment topology map of the chemical industrial park, apply spatiotemporal causal constraints to the multi-hop connection paths in the equipment topology map, and generate the path confidence distribution between equipment in the chemical industrial park. The processing module is used to construct entity feature embeddings of each device in the chemical industrial park based on the dynamic monitoring dataset, and determine the time-varying causal association matrix between devices in the chemical industrial park through all entity feature embeddings and the path confidence distribution. The processing module is used to divide the equipment in the chemical industrial park into risk groups based on the multidimensional state vectors of the equipment in the dynamic monitoring dataset, obtain risk equipment groups with different risk levels, perform cross-group risk feature propagation on the entity feature embedding of the equipment in each risk equipment group, and then construct the risk propagation tensor of the risk partitioning topology network of equipment in the chemical industrial park. The execution module is used to generate a dynamic risk vector for the chemical industrial park through the time-varying causal correlation matrix and the risk propagation tensor, and to provide early warning of safety risks within the chemical industrial park based on the dynamic risk vector.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the IoT-based dynamic early warning method for park risks as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the IoT-based dynamic early warning method for park risks as described in any one of claims 1 to 7.

Images