Real-time safety inspection system for chemical industrial park based on edge computing

Abstract

The application discloses a kind of based on edge computing's chemical industry park real-time safety inspection system, including data acquisition module: for collecting equipment operating condition data and environmental safety data;Risk analysis module: for executing analysis processing to collected data, generates risk state vector;Space mapping module: for mapping risk state vector to risk state space and determining position coordinates;State kernel construction module: for executing topological division to risk state space and constructing segmented affine state transition subkernel;State update module: for generating system state update result under state transition subkernel constraint;Parameter update module: for executing spectral structure splitting to complex parameter space and completing parameter update;Result output module: for outputting safety inspection state based on updated complex parameter and system state update result.This application realizes risk evolution controlled modeling and parameter stable update.

Description

Technical Field

[0001] This invention relates to the fields of edge computing and intelligent industrial safety monitoring technology, and in particular to a real-time safety inspection system for chemical industrial parks based on edge computing. Background Technology

[0002] Chemical industrial parks typically house numerous high-temperature, high-pressure, flammable, explosive, and toxic equipment. The equipment is complex, operating conditions are variable, and safety risks exhibit significant dynamic, spatially correlated, and sudden characteristics. Current safety inspections in chemical industrial parks primarily rely on fixed-cycle inspections, threshold alarms, and centralized monitoring platforms to collect and analyze equipment operating status and environmental safety data. This approach generally suffers from the following shortcomings in engineering applications: Firstly, traditional inspection models often employ uniform state assessment or rule-based judgment mechanisms, making it difficult to characterize the nonlinear evolution of risk states under different areas and operating conditions. Secondly, centralized computing architectures are heavily reliant on data transmission and computing resources, exhibiting response lag and insufficient real-time performance when faced with high-frequency, multi-source data input.

[0003] With the development of edge computing technology, extending security inspection computing capabilities to the edge nodes of industrial parks has become an important trend. However, most existing edge-side inspection methods still rely on global models or simple parameter update strategies, lacking a characterization of the risk state space structure and failing to reflect the propagation relationship of risks between spatial neighborhoods. Furthermore, during parameter updates, most methods perform overall updates based on the real-valued parameter space, failing to fully consider the inherent constraints of complex-valued parameters in the frequency domain structure. This can easily lead to coupling interference during parameter updates, affecting the stability and consistency of inspection results.

[0004] Therefore, there is an urgent need for a real-time safety inspection technology solution that can combine edge computing architecture to perform structured modeling of the risk status of chemical industrial parks, introduce spatial topology and segmented transition mechanisms at the state evolution level, and introduce spectral subspace constraints and complex value gradient calculation methods at the parameter update level, so as to meet the comprehensive requirements of real-time performance, structural consistency and computational controllability in the complex scenarios of chemical industrial parks. Summary of the Invention

[0005] One objective of this invention is to propose a real-time safety inspection system for chemical industrial parks based on edge computing. This invention fully integrates edge computing technology with the safety inspection needs of chemical industrial parks, uniformly processing equipment operating status and environmental safety data. It comprehensively describes the entire process of safety risk assessment, from data acquisition, status analysis, risk evolution to inspection result output, and features timely response, stable inspection results, and adaptability to complex operating conditions within the industrial park. According to an embodiment of this invention, a real-time safety inspection system for chemical industrial parks based on edge computing includes: Data acquisition module: used to collect equipment operating status data and environmental safety data at the edge nodes of the chemical industrial park; Risk analysis module: used to analyze and process the collected data to generate risk status vectors; Spatial mapping module: used to map risk state vectors to risk state space to determine corresponding location coordinates; State kernel construction module: used to generate risk region units in the risk state space to construct a piecewise affine state kernel set; State update module: used to select the state transition sub-core based on the positional change of the risk state vector to generate the system state update result; Parameter update module: used to perform complex-valued parameter updates in the target spectrum subspace under the constraints of the selected state transition subkernel; Result output module: used to generate safety inspection status output based on system status update results and complex value parameter update results.

[0006] Optionally, modules can be integrated using the following methods: S1. Collect equipment operation status data and environmental safety data at the edge nodes of the chemical industrial park, perform target detection calculation on video image data, perform time series difference calculation and trend slope calculation on gas concentration data, perform frequency domain feature extraction calculation on equipment vibration data, and form a unified time index data sequence. S2. Perform parsing and processing on equipment operation status data and environmental safety data to form a risk state vector containing risk intensity component, risk diffusion component and risk evolution component, and generate anomaly type identifier based on the numerical range of risk intensity component, risk diffusion component and risk evolution component; and generate warning level identifier based on the combined range of risk intensity component and risk diffusion component. S3. Map the risk state vector to the risk state space, and determine the position coordinates corresponding to the risk state vector in the risk state space; S4. In the risk state space, perform topological partitioning on the risk state space to generate several risk region units with adjacency relationships. For each risk region unit, construct a state transition sub-kernel in the form of a linear affine. During real-time operation, identify the risk evolution path based on the positional changes of the risk state vector between risk region units. Select a state transition sub-kernel based on the risk evolution path or perform restricted transition calculations between adjacent state transition sub-kernels to generate system state update results. S5. After the system state update results are generated, determine the complex value parameter space associated with the system state. S6. Under the constraint of the selected state transition kernel, perform a spectrum structure splitting operation on the complex parameter space to form several mutually orthogonal complex spectrum subspaces. In the several complex spectrum subspaces, determine the target spectrum subspace according to the selected state transition kernel. Project the Wirtinger gradient corresponding to the complex parameter into the target spectrum subspace to perform parameter update operation. S7. Based on the complex value parameters obtained from the parameter update calculation in S6, the system status update result, and the anomaly type identifier, the warning level identifier generates the safety inspection status output for the current time slice. When generating the safety inspection status output, the corresponding emergency response process step sequence is read according to the anomaly type identifier, and a partition warning record is generated in combination with the risk area unit identifier. The corresponding warning execution instructions are generated according to the warning level identifier. The warning execution instructions include on-site audible and visual alarm instructions, management platform pop-up instructions, mobile terminal push instructions, and emergency notification sending instructions. The warning level identifier, risk area unit identifier, and warning execution instructions are written into the current time slice output record.

[0007] Optionally, the parsing process in step S2 includes the following steps: S21. Read the equipment operating status data and environmental safety data, perform time index rearrangement, and form a unified time index data sequence; S22. Read data items along the unified time index data sequence, calculate the numerical offset between the current time index data item and the reference time index data item, calculate the difference between adjacent time index data items, and generate risk intensity components. S23. At the same time, index and read data items from different collection nodes, calculate the numerical difference, the consistency value of the direction of change, and the difference of the offset of the data items for the collection nodes, arrange them according to the combination order of the collection node identifiers to form a spatial correlation numerical sequence, and generate risk diffusion components. S24. Read data items at adjacent time index positions along the unified time index data sequence, perform trend calculations on the data items at adjacent time index positions, and generate risk evolution components; S25. Arrange the risk intensity component, risk diffusion component, and risk evolution component in the order of their preset positions to form a risk state vector.

[0008] Optionally, S3 specifically includes: S31. Read the risk state vector, read the coordinate axis component definition set of the risk state space, and record the risk state vector component position identifiers corresponding to each coordinate axis in the coordinate axis component definition set. S32. Read the component values ​​at the corresponding component positions in the risk state vector according to the coordinate axis component definition set, perform normalization processing on the component values, and form a normalized component value set. S33. Write the set of normalized component values ​​into the position coordinate vector in the order of the coordinate axes, and write the position coordinate vector into the position buffer of the risk state space to complete the determination of the position coordinates of the risk state vector in the risk state space.

[0009] Optionally, S4 specifically includes: S41. Read the position coordinate vector in the position buffer in the risk state space, and perform coordinate axis component parsing on the position coordinate vector according to the coordinate axis component definition set to obtain the coordinate axis component sequence; S42. Establish a topology partitioning parameter table in the risk state space. The topology partitioning parameter table records the value range and interval boundary identifier corresponding to each coordinate axis component. Perform interval assignment determination on the position coordinate vector according to the coordinate axis component sequence, and output the interval boundary identifier combination. The interval boundary identifier combination corresponds to the risk area unit identifier. S43. Generate a risk area unit identifier set based on the combination of interval boundary identifiers, perform boundary identifier difference operation on the risk area unit identifier set, and record the boundary identifier difference operation as an adjacency relationship identifier when the boundary identifier difference operation satisfies the single coordinate axis change condition. Write the adjacency relationship identifier into the adjacency relationship table. S44. Establish a state transition sub-kernel parameter table for the risk area unit identifier set. The state transition sub-kernel parameter table records the affine mapping matrix and affine bias vector corresponding to the risk area unit identifier, forming a linear affine form of the state transition sub-kernel. Several state transition sub-kernels constitute a piecewise affine state kernel set. S45. In the unified time index data sequence, advance the update of the location coordinate vector, read the risk area unit identifiers corresponding to adjacent time index positions in time index order, and form a risk evolution path identifier sequence. S46. Read the adjacent risk area unit identifier pairs along the risk evolution path identifier sequence. If there is an adjacency relationship identifier, read the corresponding state transition sub-core parameters and perform state transition mapping operation to generate system state update results. S47. When the adjacent risk area unit identifier is not recorded in the adjacency table, read the state transition sub-kernel parameter corresponding to the adjacent risk area unit identifier, perform restricted transition calculation on the state transition mapping operation, and generate the system state update result.

[0010] Optionally, S5 specifically includes: Read the system state update results and parse the state update component sequence according to the component position order of the system state update results; Read the parameter components at the corresponding positions in the complex parameter vector along the state update component sequence, and read the corresponding real and imaginary components for each parameter component; The read real and imaginary components are combined in order of their positions to form a subset of complex-valued parameters; According to the position order of the complex-valued parameter subsets in the complex-valued parameter vector, the complex-valued parameter subsets are mapped to the complex-valued parameter space representation structure, and the complex-valued parameter space representation structure records the coordinate position of the complex-valued parameter subsets in the parameter space; A subset of complex-valued parameters is written into the complex-valued parameter space representation structure to obtain the complex-valued parameter space corresponding to the system state update result.

[0011] Optionally, S6 specifically includes: Read the complex-valued parameter space representation structure corresponding to the system state update result, and parse the complex-valued parameter coordinate sequence according to the parameter position order recorded in the complex-valued parameter space representation structure; Read a subset of complex-valued parameters along the coordinate sequence of complex-valued parameters, and read the corresponding real and imaginary components for each complex-valued parameter, forming a sequence of complex-valued parameter components according to the parameter position order; A spectral structure splitting operation is performed on the complex-valued parameter component sequence. The spectral structure splitting operation performs orthogonal decomposition on the real component sequence and the imaginary component sequence respectively, resulting in several sets of mutually orthogonal complex-valued spectral components. A set of complex-valued spectrum subspaces is established based on the result of spectrum structure splitting. Each complex-valued spectrum subspace records the corresponding set of spectrum component index positions. Read the risk area unit identifier, and select the target complex value spectrum subspace from the complex value spectrum subspace set according to the correspondence between the risk area unit identifier and the spectrum component index position; Read the corresponding complex-valued parameter components along the target complex-valued spectrum subspace, calculate the corresponding Wirtinger gradient components for the complex-valued parameter components in the target complex-valued spectrum subspace, and obtain the gradient component sequence in the target spectrum subspace; The gradient component sequence is restricted to the range of parameter index positions corresponding to the target complex value spectrum subspace. Then, parameter update operations are performed on the complex value parameter components in the target complex value spectrum subspace to generate updated complex value parameter components. The updated complex-valued parameter components are written into the response value parameter space representation structure in the order of their positions, forming the updated complex-valued parameter space.

[0012] Optionally, the calculation of the corresponding Wirtinger gradient components specifically includes: The sequence of complex-valued parameter components in the target complex-valued spectrum subspace is split into a sequence of real components and a sequence of imaginary components, with the real component sequence and the imaginary component sequence corresponding one-to-one in the parameter position index; the system state update result is read, and the system state update component sequence is obtained by parsing according to the component position order; The real component sequence, the imaginary component sequence, and the system state update component sequence are combined in the order of parameter position index to form the gradient calculation input sequence. The real and imaginary partial derivative values ​​are calculated for each parameter position index along the gradient calculation input sequence. The real and imaginary partial derivative values ​​are combined in the order of parameter position index to form the Wirtinger gradient component sequence.

[0013] Optionally, S7 specifically includes: Read the complex-valued parameters obtained from the parameter update operation in S6, and parse them according to the parameter position order to obtain the complex-valued parameter component sequence; read the system state update result, and parse it according to the component position order to obtain the system state update component sequence; perform a concatenation operation on the complex-valued parameter component sequence and the system state update component sequence according to the preset component position order to form the inspection output vector; write the inspection output vector into the current time slice output record, and the current time slice output record records the time index and the inspection output vector; output the current time slice output record as the safety inspection status output.

[0014] The beneficial effects of this invention are: (1) This invention constructs a risk state vector on the edge node side and performs risk state space mapping, so that the equipment operation status and environmental safety data are parsed and the state evolution calculation is completed locally. This reduces the latency and load caused by centralized data transmission and centralized processing, and enables the safety inspection process to have stable real-time response capability. It is suitable for scenarios where multiple devices operate in parallel in chemical industrial parks.

[0015] (2) The present invention introduces a topological partitioning and segmented affine state transition sub-core mechanism in the risk state space, so that the risk state update process is constrained by spatial adjacency and evolution path, avoiding the overgeneralization of complex working conditions by a single global model, and can more accurately reflect the evolution differences of risk in different regional units, thereby enhancing the consistency and interpretability of the inspection state description.

[0016] (3) In the parameter update stage, the present invention introduces the complex parameter spectrum structure splitting and the Wirtinger gradient update mechanism in the target spectrum subspace, so that the parameter update process is limited to the spectrum subspace that matches the evolution of the current risk state, reducing the instability caused by parameter coupling and ensuring that the inspection output maintains smoothness and structural consistency between consecutive time slices. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a real-time safety inspection system for chemical industrial parks based on edge computing, as proposed in this invention. Figure 2 This is a schematic diagram of the risk state space topology partitioning and segmented affine state transition of a real-time safety inspection system for chemical industrial parks based on edge computing proposed in this invention. Figure 3 This is a schematic diagram of the splitting and restricted updating of the complex-valued parameter spectrum structure of a real-time safety inspection system for chemical industrial parks based on edge computing, as proposed in this invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0019] refer to Figure 1-3 A real-time safety inspection system for chemical industrial parks based on edge computing includes: Data acquisition module: used to collect equipment operating status data and environmental safety data at the edge nodes of the chemical industrial park; Risk analysis module: used to analyze and process the collected data to generate risk status vectors; Spatial mapping module: used to map risk state vectors to risk state space to determine corresponding location coordinates; State kernel construction module: used to generate risk region units in the risk state space to construct a piecewise affine state kernel set; State update module: used to select the state transition sub-core based on the positional change of the risk state vector to generate the system state update result; Parameter update module: used to perform complex-valued parameter updates in the target spectrum subspace under the constraints of the selected state transition subkernel; Result output module: used to generate safety inspection status output based on system status update results and complex value parameter update results.

[0020] In this embodiment, the modules are interconnected using the following method: S1. Collect equipment operation status data and environmental safety data at the edge nodes of the chemical industrial park, perform target detection calculation on video image data, perform time series difference calculation and trend slope calculation on gas concentration data, perform frequency domain feature extraction calculation on equipment vibration data, and form a unified time index data sequence. S2. Perform parsing and processing on equipment operation status data and environmental safety data to form a risk state vector containing risk intensity component, risk diffusion component and risk evolution component, and generate anomaly type identifier based on the numerical range of risk intensity component, risk diffusion component and risk evolution component; and generate warning level identifier based on the combined range of risk intensity component and risk diffusion component. S3. Map the risk state vector to the risk state space, and determine the position coordinates corresponding to the risk state vector in the risk state space; S4. Perform topological partitioning on the risk state space to form several risk region units with adjacency relationships; construct a linear affine state transition sub-kernel for each risk region unit; during real-time operation, identify the evolution path of the risk state between different risk region units based on the position change of the risk state vector in the risk state space; select the corresponding state transition sub-kernel along the risk evolution path, or perform restricted transition calculations between the state transition sub-kernels corresponding to adjacent risk region units to generate system state update results; S5. After the system state update results are generated, determine the complex value parameter space associated with the system state. S6. Under the constraint of the selected state transition kernel, perform a spectrum structure splitting operation on the complex parameter space to form several mutually orthogonal complex spectrum subspaces. In the several complex spectrum subspaces, determine the target spectrum subspace according to the selected state transition kernel. Project the Wirtinger gradient corresponding to the complex parameter into the target spectrum subspace to perform parameter update operation. S7. Based on the complex value parameters obtained from the parameter update calculation in S6, the system status update result, and the anomaly type identifier, the warning level identifier generates the safety inspection status output for the current time slice. When generating the safety inspection status output, the corresponding emergency response process step sequence is read according to the anomaly type identifier, and a partition warning record is generated in combination with the risk area unit identifier. The corresponding warning execution instructions are generated according to the warning level identifier. The warning execution instructions include on-site audible and visual alarm instructions, management platform pop-up instructions, mobile terminal push instructions, and emergency notification sending instructions. The warning level identifier, risk area unit identifier, and warning execution instructions are written into the current time slice output record.

[0021] In this embodiment, the parsing process in step S2 includes the following steps: S21. Read the equipment operating status data and environmental safety data, perform time index rearrangement, and form a unified time index data sequence; S22. Read data items along the unified time index data sequence, calculate the numerical offset between the current time index data item and the reference time index data item, calculate the difference between adjacent time index data items, and generate a risk intensity component; the numerical offset refers to the difference between the value corresponding to the current time index position and the value corresponding to the preset reference time index position for the same data item in the unified time index data sequence, and perform difference calculation on the two to obtain the numerical change of the current time index data item relative to the reference time index data item.

[0022] S23. At the same time, index and read data items from different collection nodes, calculate the numerical difference, the consistency value of the direction of change, and the difference of the offset of the data items for the collection nodes, arrange them according to the combination order of the collection node identifiers to form a spatial correlation numerical sequence, and generate risk diffusion components. S24. Read data items at adjacent time index positions along the unified time index data sequence, perform trend calculations on the data items at adjacent time index positions, and generate risk evolution components; S25. Arrange the risk intensity component, risk diffusion component, and risk evolution component in the order of their preset positions to form a risk state vector.

[0023] In this embodiment, the execution time index rearrangement specifically includes: Read the original timestamps corresponding to each data item in the equipment operation status data and environmental safety data, and extract the time index sequence of each data item respectively; A merging operation is performed on the time index sequence corresponding to the equipment operation status data and the time index sequence corresponding to the environmental safety data to generate a unified time index set covering the time range of the two types of data; The unified time index set is sorted according to chronological order to form an ordered unified time index sequence. Search for data items with the same time index in the equipment operation status data and environmental safety data respectively, following the ordered and unified time index sequence; When a certain time index does not have a corresponding data item in one type of data, write a null value or missing value flag at the corresponding position of that data. The data items of equipment operation status data and environmental safety data are rearranged according to a unified time index sequence to form a data sequence arranged according to a unified time index, thus completing the time index rearrangement process.

[0024] In this embodiment, the trend calculation specifically includes: Read data items corresponding to multiple consecutive time index positions along a unified time index data sequence, and form a data item change sequence according to the time index order; Perform a difference operation on the data items at adjacent time index positions in the data item change sequence to obtain a data item change value sequence between adjacent time index positions; Record the sign direction of each change value in the sequence of change values ​​of a data item to form a sequence of change direction. The sequence of change direction is used to represent the relationship of the change direction of the data item within the continuous time index interval. Perform cumulative or average operations on the absolute values ​​of each change value in the data item change value sequence to form a change magnitude sequence, which is used to represent the magnitude of change of the data item within a continuous time index interval; The sequence of change direction and the sequence of change magnitude are combined in the order of preset component positions to form risk evolution components, and the risk evolution components are written into the component positions corresponding to the risk state vector.

[0025] In this embodiment, S3 specifically includes: S31. Read the risk state vector, read the coordinate axis component definition set of the risk state space, and record the risk state vector component position identifiers corresponding to each coordinate axis in the coordinate axis component definition set. S32. Read the component values ​​at the corresponding component positions in the risk state vector according to the coordinate axis component definition set, perform normalization processing on the component values, and form a normalized component value set. S33. Write the set of normalized component values ​​into the position coordinate vector in the order of the coordinate axes, and write the position coordinate vector into the position buffer of the risk state space to complete the determination of the position coordinates of the risk state vector in the risk state space.

[0026] In this embodiment, S4 specifically includes: S41. Read the position coordinate vector in the position buffer in the risk state space, and perform coordinate axis component parsing on the position coordinate vector according to the coordinate axis component definition set to obtain the coordinate axis component sequence; S42. Establish a topology partitioning parameter table in the risk state space. The topology partitioning parameter table records the value range and interval boundary identifier corresponding to each coordinate axis component. Perform interval assignment determination on the position coordinate vector according to the coordinate axis component sequence, and output the interval boundary identifier combination. The interval boundary identifier combination corresponds to the risk area unit identifier. S43. Generate a risk area unit identifier set based on the combination of interval boundary identifiers, perform boundary identifier difference operation on the risk area unit identifier set, and record the boundary identifier difference operation as an adjacency relationship identifier when the boundary identifier difference operation satisfies the single coordinate axis change condition. Write the adjacency relationship identifier into the adjacency relationship table. S44. Establish a state transition sub-kernel parameter table for the risk area unit identifier set. The state transition sub-kernel parameter table records the affine mapping matrix and affine bias vector corresponding to the risk area unit identifier, forming a linear affine form of the state transition sub-kernel. Several state transition sub-kernels constitute a piecewise affine state kernel set. S45. In the unified time index data sequence, advance the update of the location coordinate vector, read the risk area unit identifiers corresponding to adjacent time index positions in time index order, and form a risk evolution path identifier sequence. S46. Read the adjacent risk area unit identifier pairs along the risk evolution path identifier sequence. If there is an adjacency relationship identifier, read the corresponding state transition sub-core parameters and perform state transition mapping operation to generate system state update results. S47. When the adjacent risk area unit identifier is not recorded in the adjacency table, read the state transition sub-kernel parameter corresponding to the adjacent risk area unit identifier, perform restricted transition calculation on the state transition mapping operation, and generate the system state update result.

[0027] In this embodiment, the execution of boundary marker differential operation specifically includes: After topological partitioning, the risk state space is divided into several risk region units. Each risk region unit corresponds to a set of interval boundary identifiers, which are composed of the value interval numbers corresponding to each coordinate axis component.

[0028] Specifically, for any two risk area units, the interval boundary identifier vectors are compared one by one according to the coordinate axis components, and the number of coordinate axes where the interval number changes is counted.

[0029] When the number of coordinate axes whose interval numbers change is equal to one, and the change in interval numbers on the coordinate axes corresponds to the relationship between adjacent interval numbers, the risk area unit identifier pair is recorded as satisfying the adjacency relationship condition; when the number of coordinate axes whose interval numbers change is greater than one, or the change in interval numbers does not satisfy the relationship between adjacent interval numbers, the risk area unit identifier pair is determined to not satisfy the adjacency relationship condition.

[0030] In this embodiment, the execution of restricted transition computation specifically refers to: In the risk evolution path identifier sequence, read the two risk region unit identifiers corresponding to adjacent time index positions. When the risk region unit identifier pair is not recorded in the adjacency table, read the state transition sub-kernel parameters corresponding to the risk region unit identifier respectively. The state transition sub-kernel parameters include an affine mapping matrix and an affine bias vector.

[0031] Substitute the system state vector at the current time index position into the affine mapping matrix and affine bias vector corresponding to the adjacent risk area unit identifiers to calculate multiple candidate state transition results; Transition constraints are introduced between multiple candidate state transition results. These transition constraints include constraints on the range of coordinate changes and the number of coordinate axis changes of the state transition results in the risk state space. Candidate state transition results that do not meet the transition constraints are eliminated. Among the remaining candidate state transition results, the state transition results that meet the constraints are selected as the restricted transition calculation results according to the coordinate axis change direction corresponding to the risk evolution path identifier sequence. The results of the restricted transition calculation are written into the system state update results and used as the system state update output at the current time index position.

[0032] In this embodiment, S5 specifically includes: Read the system state update results and parse the state update component sequence according to the component position order of the system state update results; Read the parameter components at the corresponding positions in the complex parameter vector along the state update component sequence, and read the corresponding real and imaginary components for each parameter component; The read real and imaginary components are combined in order of their positions to form a subset of complex-valued parameters; According to the position order of the complex-valued parameter subsets in the complex-valued parameter vector, the complex-valued parameter subsets are mapped to the complex-valued parameter space representation structure, and the complex-valued parameter space representation structure records the coordinate position of the complex-valued parameter subsets in the parameter space; A subset of complex-valued parameters is written into the complex-valued parameter space representation structure to obtain the complex-valued parameter space corresponding to the system state update result.

[0033] In this embodiment, S6 specifically includes: Read the complex-valued parameter space representation structure corresponding to the system state update result, and parse the complex-valued parameter coordinate sequence according to the parameter position order recorded in the complex-valued parameter space representation structure; Read a subset of complex-valued parameters along the coordinate sequence of complex-valued parameters, and read the corresponding real and imaginary components for each complex-valued parameter, forming a sequence of complex-valued parameter components according to the parameter position order; A spectral structure splitting operation is performed on the complex-valued parameter component sequence. The spectral structure splitting operation performs orthogonal decomposition on the real component sequence and the imaginary component sequence respectively, resulting in several sets of mutually orthogonal complex-valued spectral components. A set of complex-valued spectrum subspaces is established based on the result of spectrum structure splitting. Each complex-valued spectrum subspace records the corresponding set of spectrum component index positions. Read the risk area unit identifier, and select the target complex value spectrum subspace from the complex value spectrum subspace set according to the correspondence between the risk area unit identifier and the spectrum component index position; Read the corresponding complex-valued parameter components along the target complex-valued spectrum subspace, calculate the corresponding Wirtinger gradient components for the complex-valued parameter components in the target complex-valued spectrum subspace, and obtain the gradient component sequence in the target spectrum subspace; The gradient component sequence is restricted to the range of parameter index positions corresponding to the target complex value spectrum subspace. Then, parameter update operations are performed on the complex value parameter components in the target complex value spectrum subspace to generate updated complex value parameter components. The updated complex-valued parameter components are written into the response value parameter space representation structure in the order of their positions, forming the updated complex-valued parameter space.

[0034] In this embodiment, the calculation of the corresponding Wirtinger gradient components specifically includes: The sequence of complex-valued parameter components in the target complex-valued spectrum subspace is split into a sequence of real components and a sequence of imaginary components, with the real component sequence and the imaginary component sequence corresponding one-to-one in the parameter position index; the system state update result is read, and the system state update component sequence is obtained by parsing according to the component position order; The real component sequence, the imaginary component sequence, and the system state update component sequence are combined in the order of parameter position index to form the gradient calculation input sequence. The real and imaginary partial derivative values ​​are calculated for each parameter position index along the gradient calculation input sequence. The real and imaginary partial derivative values ​​are combined in the order of parameter position index to form the Wirtinger gradient component sequence.

[0035] In this embodiment, S7 specifically includes: Read the complex-valued parameters obtained from the parameter update operation in S6, and parse them according to the parameter position order to obtain the complex-valued parameter component sequence; read the system state update result, and parse it according to the component position order to obtain the system state update component sequence; perform a concatenation operation on the complex-valued parameter component sequence and the system state update component sequence according to the preset component position order to form the inspection output vector; write the inspection output vector into the current time slice output record, and the current time slice output record records the time index and the inspection output vector; output the current time slice output record as the safety inspection status output.

[0036] In this embodiment, the parameter update operation specifically includes: Along the range of parameter index positions corresponding to the target complex-valued spectrum subspace, read the complex-valued parameter components and the corresponding Wirtinger gradient components one by one in the target complex-valued spectrum subspace; For each parameter index position, a linear combination operation is performed on the complex parameter component and the corresponding Wirtinger gradient component according to the parameter update step size, wherein the complex parameter component and the Wirtinger gradient component correspond one-to-one at the parameter index position. The linear combination operation, according to a preset parameter update direction rule, subtracts the weighted result of its corresponding Wirtinger gradient component from the complex-valued parameter component to obtain the updated complex-valued parameter component. For complex-valued parameter components that do not belong to the range of the corresponding parameter index positions in the target complex-valued spectrum subspace, keep the original parameter values ​​unchanged; The updated complex-valued parameter components and the unupdated complex-valued parameter components in the target complex-valued spectrum subspace are merged in the order of their original parameter index positions to form an updated complex-valued parameter vector. The updated complex-valued parameter vector is then written into the return-valued parameter space representation structure.

[0037] In this embodiment, the emergency response process sequence specifically includes: It is a set of structured steps pre-stored in edge nodes, and the emergency response process step sequence corresponds one-to-one with the anomaly type identifier.

[0038] Each emergency response procedure sequence includes operation steps arranged in a preset order, and the operation steps include at least one of equipment control instruction items, personnel dispatch instruction items, area isolation instruction items, and communication notification instruction items; The equipment control instruction entry includes the corresponding equipment number and control action identifier; the personnel dispatch instruction entry includes the emergency personnel number and assembly point number; the communication notification instruction entry includes the notification type identifier and notification object number.

[0039] Example 1: To verify the feasibility and effectiveness of this invention in a real-world operating environment, it is applied to a routine safety inspection scenario in a chemical industrial park. Chemical industrial parks are characterized by densely packed equipment, complex equipment operating states, and significant temporal and spatial correlations in safety risks. Traditional inspection methods primarily rely on manual inspections and fixed threshold monitoring, which struggle to reflect the evolution of risks across different areas in a timely manner. Furthermore, with continuously increasing data volumes, inspection responses suffer from lag. This embodiment addresses these issues. In the application scenario, operational status data and environmental safety data from various types of equipment within the park are continuously collected through edge nodes. The data content encompasses information such as changes in operational load, fluctuations in environmental conditions, and changes in equipment correlations. The collected data is first parsed at the edge, aligning data from different sources using a unified time index to form a continuous data sequence. During parsing, the magnitude of data changes at different time points is extracted as a risk intensity component, the data correlations between different collection nodes are organized as a risk diffusion component, and the data change trends within continuous time periods are organized as a risk evolution component. These components are combined in a fixed order to form a risk state vector. The resulting risk state vector is mapped to a risk state space, where the trajectory of risk state changes over time can be clearly depicted. By topologically partitioning the risk state space, the risk states within the park are divided into multiple risk region units with adjacency relationships. Each risk region unit corresponds to a set of linear affine state transition kernels. In actual operation, the system identifies risk evolution paths based on the movement of risk states between different risk region units and selects the corresponding state transition kernel to update the system state, thus avoiding the error accumulation problem caused by using a single model to uniformly process all states. After the state update is completed, the system further determines the complex-valued parameter space associated with the current state and performs a spectral structure splitting operation on the complex-valued parameter space under the constraints of the selected state transition kernel. The complex-valued parameters are split into multiple mutually orthogonal complex-valued spectral subspaces, and the system only performs parameter updates within the target spectral subspace corresponding to the current risk evolution path. During the parameter update process, the Wirtinger gradient corresponding to the complex-valued parameters is used for calculation to ensure the consistency of the update process between the real and imaginary parts. The updated complex-valued parameters and the system state update results jointly participate in generating the safety inspection state output for the current time slice. Comparative analysis revealed that, after adopting the method of this invention, the inspection system responds more smoothly to changes in risk status, the spatial evolution path of risk status is clearer, and abnormal states are reflected in the inspection output results within a shorter time span. Compared with traditional inspection methods, this invention maintains a relatively stable state output structure over multiple consecutive operating cycles, reducing misjudgments caused by drastic parameter fluctuations. To further verify the effectiveness, statistical analysis was performed on the inspection results within the same operating cycle.Statistical results show that, under the same data input conditions, the risk status change amplitude of the present invention is more concentrated, the spatial diffusion trend is consistent with the correlation with the actual equipment, and the inspection output shows good continuity and consistency between consecutive time slices.

[0040] Table 1: Quantitative Comparison of Safety Inspection Outputs in Chemical Industrial Parks

[0041] Table 1 presents a quantitative comparison of the safety inspection output characteristics between traditional inspection methods and the method of this invention under the same data input conditions. All indicators in the table are calculated based on uniform data normalization processing, and the numerical ranges do not involve specific physical units. They are used to reflect the differences between different inspection methods in terms of risk state expression, spatial correlation characterization, and parameter update stability.

[0042] The width of the fluctuation range of the risk status values ​​reflects the overall distribution range of the risk status values ​​in the inspection output. The fluctuation range of the traditional inspection method is relatively large, indicating that the risk status changes unstablely between consecutive time slices; the fluctuation range of the method of this invention is significantly converged, indicating that the risk status output is concentrated in a relatively continuous numerical segment, which is beneficial to subsequent state evolution analysis.

[0043] The mean and variance of the state differences between adjacent time slices are used to describe the magnitude and fluctuation of risk states between consecutive time slices. The results in the table show that the method of this invention is significantly lower than the traditional method in both mean and variance statistics, indicating that the risk state update process is subject to structural constraints, avoiding large jumps and maintaining the continuity of the state evolution process.

[0044] The consistency rate of spatial correlation change direction is used to measure the degree of consistency in the direction of risk state change corresponding to different collection nodes. In traditional inspection methods, the index value is relatively low, reflecting the instability of spatial correlation; while in the method of this invention, the consistency rate is significantly improved, indicating that through the spatial topology partitioning of risk state and the constraint of state transition sub-kernels, the evolution path of risk in space is more consistent.

[0045] The mean parameter update magnitude reflects the intensity of model parameter changes in adjacent update stages during the inspection process. The significant reduction in the index in the method of this invention indicates that the parameter update process is constrained within a reasonable range by updating the spectral structure through spectral structure splitting and Wirtinger gradient within the target complex-valued spectral subspace, thereby reducing the fluctuation risk caused by parameter coupling.

[0046] The continuous time slice output offset rate is used to measure the stability of the inspection output over time. The results in the table show that the method of the present invention maintains a low level in terms of the index, indicating that the safety inspection status output remains consistent across continuous time slices, which is beneficial for forming a stable and traceable sequence of inspection results.

[0047] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A real-time safety inspection system for chemical industrial parks based on edge computing, characterized in that, include: Data acquisition module: used to collect equipment operating status data and environmental safety data at the edge nodes of the chemical industrial park; Risk analysis module: used to analyze and process the collected data to generate risk status vectors; Spatial mapping module: used to map risk state vectors to risk state space to determine corresponding location coordinates; State kernel construction module: used to generate risk region units in the risk state space to construct a piecewise affine state kernel set; State update module: used to select the state transition sub-core based on the positional change of the risk state vector to generate the system state update result; Parameter update module: used to perform complex-valued parameter updates in the target spectrum subspace under the constraints of the selected state transition subkernel; Result output module: used to generate safety inspection status output based on system status update results and complex value parameter update results.

2. The real-time safety inspection system for chemical industrial parks based on edge computing according to claim 1, characterized in that, The modules are connected in the following way: S1. Collect equipment operation status data and environmental safety data at the edge nodes of the chemical industrial park, perform target detection calculation on video image data, perform time series difference calculation and trend slope calculation on gas concentration data, perform frequency domain feature extraction calculation on equipment vibration data, and form a unified time index data sequence. S2. Perform parsing and processing on equipment operation status data and environmental safety data to form a risk state vector containing risk intensity component, risk diffusion component and risk evolution component, and generate anomaly type identifier based on the numerical range of risk intensity component, risk diffusion component and risk evolution component. The warning level identifier is generated based on the combined range of the risk intensity component and the risk diffusion component. S3. Map the risk state vector to the risk state space, and determine the position coordinates corresponding to the risk state vector in the risk state space; S4. In the risk state space, perform topological partitioning on the risk state space to generate several risk region units with adjacency relationships. For each risk region unit, construct a state transition sub-kernel in the form of a linear affine. During real-time operation, identify the risk evolution path based on the positional changes of the risk state vector between risk region units. Select a state transition sub-kernel based on the risk evolution path or perform restricted transition calculations between adjacent state transition sub-kernels to generate system state update results. S5. After the system state update results are generated, determine the complex value parameter space associated with the system state. S6. Under the constraint of the selected state transition kernel, perform a spectrum structure splitting operation on the complex parameter space to form several mutually orthogonal complex spectrum subspaces. In the several complex spectrum subspaces, determine the target spectrum subspace based on the state transition kernel selected according to the risk evolution path. Project the Wirtinger gradient corresponding to the complex parameter into the target spectrum subspace to perform parameter update operation. S7. Based on the complex value parameters obtained from the parameter update calculation in S6, the system status update result, and the anomaly type identifier, the warning level identifier generates the safety inspection status output for the current time slice. When generating the safety inspection status output, the corresponding emergency response process step sequence is read according to the anomaly type identifier, and a partition warning record is generated in combination with the risk area unit identifier. The corresponding warning execution instructions are generated according to the warning level identifier. The warning execution instructions include on-site audible and visual alarm instructions, management platform pop-up instructions, mobile terminal push instructions, and emergency notification sending instructions. The warning level identifier, risk area unit identifier, and warning execution instructions are written into the current time slice output record.

3. The real-time safety inspection system for chemical industrial parks based on edge computing according to claim 2, characterized in that, The parsing process in step S2 includes the following steps: S21. Read the equipment operating status data and environmental safety data, perform time index rearrangement, and form a unified time index data sequence; S22. Read data items along the unified time index data sequence, calculate the numerical offset between the current time index data item and the reference time index data item, calculate the difference between adjacent time index data items, and generate risk intensity components. S23. At the same time, index and read data items from different collection nodes, calculate the numerical difference, the consistency value of the direction of change, and the difference of the offset of the data items for the collection nodes, arrange them according to the combination order of the collection node identifiers to form a spatial correlation numerical sequence, and generate risk diffusion components. S24. Read data items at adjacent time index positions along the unified time index data sequence, perform trend calculations on the data items at adjacent time index positions, and generate risk evolution components; S25. Arrange the risk intensity component, risk diffusion component, and risk evolution component in the order of their preset positions to form a risk state vector.

4. The real-time safety inspection system for chemical industrial parks based on edge computing according to claim 3, characterized in that, S3 specifically includes: S31. Read the risk state vector, read the coordinate axis component definition set of the risk state space, and record the risk state vector component position identifiers corresponding to each coordinate axis in the coordinate axis component definition set. S32. Read the component values ​​at the corresponding component positions in the risk state vector according to the coordinate axis component definition set, perform normalization processing on the component values, and form a normalized component value set. S33. Write the set of normalized component values ​​into the position coordinate vector in the order of the coordinate axes, and write the position coordinate vector into the position buffer of the risk state space to complete the determination of the position coordinates of the risk state vector in the risk state space.

5. A real-time safety inspection system for chemical industrial parks based on edge computing as described in claim 4, characterized in that, S4 specifically includes: S41. Read the position coordinate vector in the position buffer in the risk state space, and perform coordinate axis component parsing on the position coordinate vector according to the coordinate axis component definition set to obtain the coordinate axis component sequence; S42. Establish a topology partitioning parameter table in the risk state space. The topology partitioning parameter table records the value range and interval boundary identifier corresponding to each coordinate axis component. Perform interval assignment determination on the position coordinate vector according to the coordinate axis component sequence, and output the interval boundary identifier combination. The interval boundary identifier combination corresponds to the risk area unit identifier. S43. Generate a risk area unit identifier set based on the combination of interval boundary identifiers, perform boundary identifier difference operation on the risk area unit identifier set, and record the boundary identifier difference operation as an adjacency relationship identifier when the boundary identifier difference operation satisfies the single coordinate axis change condition. Write the adjacency relationship identifier into the adjacency relationship table. S44. Establish a state transition sub-kernel parameter table for the risk area unit identifier set. The state transition sub-kernel parameter table records the affine mapping matrix and affine bias vector corresponding to the risk area unit identifier, forming a linear affine form of the state transition sub-kernel. Several state transition sub-kernels constitute a piecewise affine state kernel set. S45. In the unified time index data sequence, advance the update of the location coordinate vector, read the risk area unit identifiers corresponding to adjacent time index positions in time index order, and form a risk evolution path identifier sequence. S46. Read the adjacent risk area unit identifier pairs along the risk evolution path identifier sequence. If there is an adjacency relationship identifier, read the corresponding state transition sub-core parameters and perform state transition mapping operation to generate system state update results. S47. When the adjacent risk area unit identifier is not recorded in the adjacency table, read the state transition sub-kernel parameter corresponding to the adjacent risk area unit identifier, perform restricted transition calculation on the state transition mapping operation, and generate the system state update result.

6. A real-time safety inspection system for chemical industrial parks based on edge computing as described in claim 5, characterized in that, S5 specifically includes: Read the system state update results and parse the state update component sequence according to the component position order of the system state update results; Read the parameter components at the corresponding positions in the complex parameter vector along the state update component sequence, and read the corresponding real and imaginary components for each parameter component; The read real and imaginary components are combined in order of their positions to form a subset of complex-valued parameters; According to the position order of the complex-valued parameter subsets in the complex-valued parameter vector, the complex-valued parameter subsets are mapped to the complex-valued parameter space representation structure, and the complex-valued parameter space representation structure records the coordinate position of the complex-valued parameter subsets in the parameter space; A subset of complex-valued parameters is written into the complex-valued parameter space representation structure to obtain the complex-valued parameter space corresponding to the system state update result.

7. A real-time safety inspection system for chemical industrial parks based on edge computing as described in claim 6, characterized in that, S6 specifically includes: Read the complex-valued parameter space representation structure corresponding to the system state update result, and parse the complex-valued parameter coordinate sequence according to the parameter position order recorded in the complex-valued parameter space representation structure; Read a subset of complex-valued parameters along the coordinate sequence of complex-valued parameters, and read the corresponding real and imaginary components for each complex-valued parameter, forming a sequence of complex-valued parameter components according to the parameter position order; A spectral structure splitting operation is performed on the complex-valued parameter component sequence. The spectral structure splitting operation performs orthogonal decomposition on the real component sequence and the imaginary component sequence respectively, resulting in several sets of mutually orthogonal complex-valued spectral components. A set of complex-valued spectrum subspaces is established based on the result of spectrum structure splitting. Each complex-valued spectrum subspace records the corresponding set of spectrum component index positions. Read the risk area unit identifier, and select the target complex value spectrum subspace from the complex value spectrum subspace set according to the correspondence between the risk area unit identifier and the spectrum component index position; Read the corresponding complex-valued parameter components along the target complex-valued spectrum subspace, calculate the corresponding Wirtinger gradient components for the complex-valued parameter components in the target complex-valued spectrum subspace, and obtain the gradient component sequence in the target spectrum subspace; The gradient component sequence is restricted to the range of parameter index positions corresponding to the target complex value spectrum subspace. Then, parameter update operations are performed on the complex value parameter components in the target complex value spectrum subspace to generate updated complex value parameter components. The updated complex-valued parameter components are written into the response value parameter space representation structure in the order of their positions, forming the updated complex-valued parameter space.

8. A real-time safety inspection system for chemical industrial parks based on edge computing as described in claim 7, characterized in that, The calculated Wirtinger gradient components specifically include: The sequence of complex-valued parameter components in the target complex-valued spectrum subspace is split into a sequence of real components and a sequence of imaginary components, with the real component sequence and the imaginary component sequence corresponding one-to-one in the parameter position index; the system state update result is read, and the system state update component sequence is obtained by parsing according to the component position order; The real component sequence, the imaginary component sequence, and the system state update component sequence are combined in the order of parameter position index to form the gradient calculation input sequence. The real and imaginary partial derivative values ​​are calculated for each parameter position index along the gradient calculation input sequence. The real and imaginary partial derivative values ​​are combined in the order of parameter position index to form the Wirtinger gradient component sequence.

9. A real-time safety inspection system for chemical industrial parks based on edge computing as described in claim 8, characterized in that, Specifically, S7 includes: Read the complex-valued parameters obtained from the parameter update operation in S6, and parse them according to the parameter position order to obtain the complex-valued parameter component sequence; read the system state update result, and parse it according to the component position order to obtain the system state update component sequence; perform a concatenation operation on the complex-valued parameter component sequence and the system state update component sequence according to the preset component position order to form the inspection output vector; write the inspection output vector into the current time slice output record, and the current time slice output record records the time index and the inspection output vector; output the current time slice output record as the safety inspection status output.

Images