Risk assessment early warning method and device based on mechanism and data fusion

Abstract

The invention provides a risk assessment early warning method and device based on mechanism and data fusion, equipment and a medium, and the method comprises the steps: obtaining historical data, including variables of a production field sensing part, a logic part and an execution part; the method comprises the following steps of: preprocessing data, performing alignment and fusion processing on the preprocessed data, determining a causal relationship among factors in a production process through a causal analysis method, describing a quantitative relationship among nodes by estimating conditional probability density among the factors, and constructing a causal probability graph of a system; constructing a risk assessment model; and inputting the risk data with the risk label and the risk node label and normal data into the risk assessment model for training until a convergence condition is met. According to the method, a high-precision and high-robustness model is established, and the risk mapping relationship is clarified by mining the joint features of the data under different dimensions and scales including time, so that comprehensive security risk perception and decision support are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of information processing and analysis, and in particular to a risk assessment and early warning method, device, equipment and medium based on mechanism and data fusion. Background Art

[0002] my country's process industries (such as chemical, petroleum, steel, etc.) are at a critical stage of intelligent upgrading, facing challenges such as improving the scientific nature of production and operation management, enhancing the accuracy of risk analysis to cope with complex process mechanisms, strong variable coupling and process dynamics. The existing risk management and control systems at home and abroad often show that the risk management and control system is not stable in dealing with the complexity and dynamics of the process industry, and the risk perception, intelligent analysis and safety collaborative innovation level of all elements of the production and operation of the process industry are not high. In particular, there is a lack of unified standards among various types of data, resulting in serious information islands and low efficiency in data integration and sharing. Therefore, in the integration of functional safety and information security, the risk management and control system is not stable. There are obvious deficiencies in the multi-safety coordinated protection of safety and physical security, real-time risk monitoring and early warning, and intelligent decision-making support; the risks in the production and operation process of the re-process industry have the characteristics of dynamic evolution, and are affected by multiple uncertain factors such as environmental changes, fluctuations in raw material quality and equipment status. At present, there is a lack of accurate extraction of key characteristics, which makes it impossible to accurately monitor the risk trends in the reaction process. For example, potential safety hazards cannot be identified in time when the reaction temperature rises sharply; at the same time, many evaluation models fail to effectively consider the time lag of the process (such as feedback delay of temperature changes) and uncertainty (such as instability of raw material composition), which affects the reliability of risk prediction.

[0003] In view of this, there is an urgent need to establish a full-factor risk perception and intelligent analysis system, a multi-security collaborative protection technology architecture and intelligent management and control technology to achieve real-time monitoring and early warning of safety risks in the process industry. Summary of the invention

[0004] In order to overcome the problems existing in the related art, the present disclosure provides a risk assessment and early warning method, device, equipment and medium based on mechanism and data fusion to solve the technical problems in the related art.

[0005] One or more embodiments of this specification provide a risk assessment and early warning method based on mechanism and data fusion, including the following steps:

[0006] Data collection: Obtain historical data collected during the system production process, including variables of the sensing part, logic part and execution part of the production site;

[0007] Data preprocessing: preprocess the multi-factor, multi-modal, multi-source heterogeneous data collected, and align and fuse the preprocessed data;

[0008] Relationship model construction: Based on the processed data, the causal relationship between various factors in the production process is determined by the Granger causal analysis method, and the quantitative relationship between nodes is described by estimating the conditional probability density between various factors, and the causal probability diagram of the system is constructed;

[0009] Constructing a risk assessment model; the risk assessment model includes a feature extraction module and a risk assessment and classification module, wherein the feature extraction module is a graph convolutional network, and the risk assessment and classification module is a spatiotemporal graph convolutional neural network or a graph convolutional network; and

[0010] Model training: input the data with risk node labels and risk causal probability diagram, as well as normal data, into the risk assessment model for training until the convergence conditions are met, thereby obtaining a model for monitoring and early warning of production risks in the process industry;

[0011] Risk prediction and early warning: obtain real-time data during the system production process, and monitor and warn of process industry production risks through risk assessment models.

[0012] Preferably, the risk assessment and classification module includes a graph convolutional network layer including L layers of graph convolutional network layers GCN and multi-layer perceptron layers MLP, the activation function of the GCN layer is the ReLU function, and the activation function of the MLP layer is the tanh activation function.

[0013] Preferably, further data preprocessing specifically includes the following steps:

[0014] Through data quality management, data standard management and data desensitization;

[0015] The data is then cleaned, including noise reduction, normalization, missing value filling, and the fuzzy stacked autoencoder network is used to reduce dimension, remove noise and retain important information;

[0016] Through data interpolation, resampling and synchronization algorithms, the cleaned data is accurately aligned and fused across dimensions and frequencies.

[0017] Preferably, it also includes constructing a kinetic mechanism model of the process industry for data expansion.

[0018] Preferably, the construction of a kinetic mechanism model of a process industry process for data expansion specifically comprises the following steps:

[0019] Constructing conceptual dynamic models of process industries;

[0020] Based on the constructed conceptual dynamic model of the process industry process and analysis of industrial data, the coupling relationship between the data-driven method and the various factors in the process is determined, the adaptive dynamic update mechanism and update method of the risk model are confirmed and implemented, and then based on the finite element numerical simulation method, numerical simulation software is used to establish a dynamic numerical model of the process industry process and construct a dynamic mechanism model of the process industry process, which can be used to simulate the cause-effect relationship between the various factors in the production process.

[0021] Preferably, it also includes determining the root cause of the fault by reverse reasoning based on the constructed causal probability graph and risk fault data, and the specific steps are as follows:

[0022] Based on the risk failure data, the node where the failure occurs is determined, and the cause-effect relationship represented by the arrow is reversely inferred to the previous associated node, and the conditional probability density is calculated to determine whether a failure occurs, until the root node in the probability graph model; if the probability density of the monitored sample is within the range of the upper and lower confidence limits, it is identified as a normal sample; otherwise, it is identified as a faulty sample.

[0023] Preferably, the adaptive threshold introduced in the risk assessment and classification module includes an adaptive global threshold and an adaptive local threshold, and the classified target domain data is set to different thresholds through the adaptive threshold. The threshold selection adopts a parameter-free adaptive threshold selection method, and the risk level is determined according to different thresholds.

[0024] One or more embodiments of this specification provide a risk assessment and early warning device based on mechanism and data fusion, including:

[0025] The data acquisition module is used to obtain historical data collected during the system production process, including variables of the production site sensor part, logic part and execution part;

[0026] The data preprocessing module is used to preprocess the multi-factor, multi-modal, multi-source heterogeneous data collected by the data acquisition module, and to align and fuse the preprocessed data;

[0027] The relationship model building module is used to determine the causal relationship between various factors in the production process through the Granger causal analysis method based on the data processed by the data preprocessing module, describe the quantitative relationship between nodes by estimating the conditional probability density between various factors, and build the causal probability diagram of the system;

[0028] Constructing a risk assessment model module, which is used to construct the risk assessment model module, including a feature extraction module and a risk assessment and classification module, wherein the feature extraction module is a graph convolutional network, and the risk assessment and classification module is a spatiotemporal graph convolutional neural network or a graph convolutional network;

[0029] A model training module, which is used to input the data for setting risk node labels and the risk causal probability diagram, as well as normal data, into the risk assessment model for training until the convergence conditions are met, thereby obtaining a model for monitoring and early warning of production risks in the process industry; and

[0030] Risk prediction and early warning module: used to obtain real-time data during the system production process, and monitor and warn of process industry production risks through risk assessment models.

[0031] One or more embodiments of the present specification provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the risk assessment and early warning method based on mechanism and data fusion as described in any one of the above items when executing the computer program.

[0032] One or more embodiments of the present specification provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the risk assessment and early warning method based on mechanism and data fusion as described in any one of the above items.

[0033] The risk assessment and early warning method, device, equipment and medium based on mechanism and data fusion provided by the present disclosure have the advantages of constructing a multi-dimensional risk perception model, and realizing comprehensive coverage of production and operation risks by integrating multi-source data such as personnel, process, environment, equipment, network, etc.; secondly, using signal processing and feature extraction technology to process and analyze multimodal data in real time and extract key risk features; then, based on the method of mechanism and data fusion, a dynamic risk analysis model is established, and the risk mapping relationship is clarified by mining the joint features of data in different dimensions and scales including time, so as to realize comprehensive safety risk perception and decision support; it provides intelligent and scientific risk decision-making and optimization solutions for the production and operation of process industries, and significantly improves the safety, stability and operation efficiency of production systems. The method of this embodiment establishes a high-precision and high-robustness risk prediction and monitoring model, and clarifies the risk mapping relationship by mining the joint features of data in different dimensions and scales including time, so as to realize comprehensive safety risk perception and decision support. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate one or more embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1 A flow chart of a risk assessment and early warning method based on mechanism and data fusion provided for one or more embodiments of this specification;

[0036] Figure 2 A schematic diagram of a mechanism model combined with a data dynamic risk analysis model provided in one or more embodiments of this specification;

[0037] Figure 3 A risk assessment model training flow chart provided for one or more embodiments of this specification;

[0038] Figure 4 A block diagram of a risk assessment and early warning device based on mechanism and data fusion provided for one or more embodiments of this specification; and

[0039] Figure 5 A schematic diagram of the structure of a computer device provided for one or more embodiments of this specification. DETAILED DESCRIPTION

[0040] In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below in conjunction with the drawings in one or more embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0041] The present invention is described in detail below in conjunction with specific implementation methods and the accompanying drawings.

[0042] Method Embodiment

[0043] According to an embodiment of the present invention, a risk assessment and early warning method based on mechanism and data fusion is provided. Figure 1 As shown, it is a flow chart of the risk assessment and early warning method based on mechanism and data fusion provided in this embodiment. The risk assessment and early warning method based on mechanism and data fusion according to the embodiment of the present invention comprises the following steps:

[0044] Step S1, data collection; obtain historical data collected during the system production process, including variables of the production site sensor part, logic part and execution part, define the time-varying variables in each part as dynamic variables, and define the remaining variables as static variables.

[0045] Specific data collection includes key factors such as personnel, technology, environment, equipment, and network involved in process industry production. Personnel may include factors such as personnel information, personnel behavior, operation records, and location; technology includes process parameters; environment includes factors such as temperature and humidity; equipment includes factors such as PLC, DSC, LIMS, and sensors; and network includes factors such as network intrusion and network attack. Multimodal data is monitored through a real-time monitoring system to ensure the accuracy, comprehensiveness, and timeliness of the data, providing a basis for subsequent analysis and decision-making.

[0046] By deploying sensors, video surveillance equipment, industrial Internet of Things acquisition terminals and real-time databases at the production site, such as real-time images from surveillance cameras, radar imaging data from material flow monitoring, maintenance logs and accident reports filled out by operators, temperature and flow rate read by sensors, etc., data on the above factors are collected to form information data related to functional safety, information security and physical safety, covering operation records, personnel information, personnel location behavior data, equipment status, environmental parameters, key process parameters and network security information, etc. Multimodal data is monitored through a real-time monitoring system to ensure the accuracy, comprehensiveness and timeliness of the data, providing a basis for subsequent analysis and decision-making.

[0047] Step S2, data preprocessing: preprocess the multi-factor, multi-modal, multi-source heterogeneous data collected in step S1, and align and fuse the preprocessed data.

[0048] In this embodiment, since the production and operation data of the process industry has multi-source heterogeneous characteristics, including sensor data, equipment status data, production environment data, operation record data, etc.; the production and operation data format of the process industry has multimodal characteristics, including real-time images from monitoring cameras, radar imaging data from material flow monitoring, maintenance logs and accident reports filled out by operators and other texts, general values ​​such as temperature and flow rate read by sensors, etc., but there is coupling and redundancy between the data, the safety data fluctuates slightly but has high repeatability, and multiple sensors have significant correlation when measuring the same process, and the data needs to be reduced and compressed; there is invalid and erroneous data, the sensor may cause abnormal readings due to failure, and the manually recorded data may be wrong due to operator errors, which requires verification and filtering; and there is a lack of unified standards between various types of data, resulting in serious information islands and low efficiency in data integration and sharing; in addition, the differences in data formats of different modes also increase the difficulty of integration. Therefore, further, the collected data is processed as follows:

[0049] (1) First, through data quality management, data standard management and data desensitization, multi-source data is pre-processed to ensure the integrity, consistency and security of the data.

[0050] (2) The data is then cleaned, including noise reduction, normalization, missing value filling, etc., and a fuzzy stacked autoencoder network is used for dimensionality reduction to effectively remove noise and retain important information.

[0051] (3) By designing efficient data interpolation, resampling and synchronization algorithms, the cleaned data is accurately aligned and fused across dimensions and frequencies to ensure the coherence and consistency of the data in the time dimension. The fusion of multimodal data can be achieved, for example, through the self-attention mechanism and Transformer to generate splicing features with higher information density, which is convenient for subsequent processing and analysis.

[0052] It is also possible to achieve precise alignment and fusion of data across dimensions and frequencies through generative learning algorithms, transfer learning, deep learning and other methods. A feature extraction technology based on the fusion of data and knowledge is proposed for complex process technologies to characterize the complex characteristics of process data such as strong coupling, nonlinearity, non-stationarity, and multimodality. Other methods will not be described in detail here.

[0053] Step S3, relationship model construction; based on the data processed in step S2, the causal relationship between various factors in the production process is determined by the Granger causal analysis method, and the quantitative relationship between nodes is described by estimating the conditional probability density between various factors to construct a system directed acyclic graph model or knowledge graph, wherein the directed acyclic graph model or knowledge graph is a causal probability graph.

[0054] In this embodiment, the risk in the process of production and operation of the process industry has dynamic evolution characteristics, which is affected by various uncertain factors such as environmental changes, raw material quality fluctuations and equipment status. At present, the lack of accurate extraction of key features has led to the inability to accurately monitor the risk trend in the reaction process, such as the failure to timely identify potential safety hazards when the reaction temperature rises sharply; at the same time, many evaluation models fail to effectively consider the time lag of the process (such as feedback delay of temperature change) and uncertainty (such as instability of raw material composition), which affects the reliability of risk prediction. Therefore, the influence relationship between individual factors is represented by a directed acyclic graph, which effectively solves the problem of information island phenomenon leading to analysis results. In this embodiment, the directed acyclic graph is used to display the causal structure between multivariate variables (factors), the nodes in the figure represent the variables monitored in the production process, and the arrows in the figure represent the qualitative relationship between the monitored variables; the branch composed of the related fault nodes obtained by the reverse arrow of the causal relationship diagram is the fault propagation path; if a one-to-many relationship is formed in the node, it branches into multiple fault propagation paths. For observation data with time continuity, non-parametric estimation methods are used to determine the probability relationship between nodes in the causal model structure to represent the dependency relationship between variables. Whether an abnormal event occurs is determined by whether the current data distribution deviates from the normal distribution. The quantitative propagation path of the fault is determined by forward analysis of the causal probability graph, and reverse reasoning can be used to determine the root cause of the fault.

[0055] Step S4, constructing a risk assessment model; constructing a risk assessment model includes a feature extraction module and a risk assessment and classification module, wherein the feature extraction module may be a graph convolutional network (Graph Convolutional Networks GCN) suitable for image feature extraction, and the risk assessment and classification module may be a spatio-temporal graph convolutional neural network or (Spatio-Temporal Graph Convolutional Networks, STGCN). In this embodiment, if the risk assessment and classification module may include a graph convolutional network layer of L layers GCN and a multi-layer perceptron layer MLP, the activation function of the GCN layer is the ReLU function, and the activation function of the MLP layer is the tanh activation function of the graph convolutional network.

[0056] Step S5, model training; input the data for setting risk node labels, risk causal probability diagram, and normal data into the risk assessment model for training until the convergence conditions are met, thereby obtaining a model for monitoring and warning of production risks in process industries.

[0057] Step S6, based on obtaining real-time data during the system production process, monitor and warn of process industry production risks through risk assessment models.

[0058] The risk assessment and early warning method based on mechanism and data fusion provided in this embodiment firstly builds a multi-dimensional risk perception model, and realizes comprehensive coverage of production and operation risks by integrating multi-source data such as personnel, process, environment, equipment, and network; secondly, adopts signal processing and feature extraction technology to process and analyze multimodal data in real time and extract key risk features; then, based on the method of mechanism and data fusion, establishes a dynamic risk analysis model, and clarifies the risk mapping relationship by mining the joint features of data in different dimensions and scales including time, so as to realize comprehensive safety risk perception and decision support; it provides intelligent and scientific risk decision-making and optimization solutions for the production and operation of process industries, and significantly improves the safety, stability and operation efficiency of production systems. The method of this embodiment establishes a high-precision and high-robustness risk prediction and monitoring model, and clarifies the risk mapping relationship by mining the joint features of data in different dimensions and scales including time, so as to realize comprehensive safety risk perception and decision support.

[0059] In this embodiment, the risk assessment model is used to predict risk failure types, obtain the probability of occurrence of each risk failure type, and sort them according to the size of the occurrence probability to generate a failure risk list.

[0060] In this embodiment, in the process industry, such as oil and gas, chemical industry, and steel, there is a large amount of data in the typical process of the process industry, including automatic data collected by sensors and equipment and manually collected data recorded by humans. However, there are small samples and imbalance problems in the data, with abundant safety data but missing accident data. When constructing a data set, it is necessary to balance the proportion of different types of data; it is necessary to solve the problem of small sample learning based on the inherent characteristics of process industry process data and effectively deal with the data imbalance phenomenon.

[0061] In one embodiment, it also includes constructing a kinetic mechanism model of a process industry process for data expansion, which specifically includes the following implementation steps:

[0062] A1, construct a dynamic conceptual model of process industry processes.

[0063] A2. Based on the constructed conceptual dynamics model of the process industry and analysis of industrial data, determine the data-driven approach and the coupling relationship between various factors in the process, confirm and implement the adaptive dynamic update mechanism and update method of the risk model, and then based on the finite element numerical simulation method, use numerical simulation software to establish a dynamic numerical model of the process industry process and construct a dynamic mechanism model of the process industry process, which can be used to simulate the cause-effect relationship between various factors in the production process.

[0064] In this embodiment, based on the conceptual model of the dynamics of the process industry process, numerical simulation software is used to construct a 3D grid, input the process data of the historical process industry operation, and use the MODFLOW module of the numerical simulation software for calculation to obtain a preliminary dynamic unsteady flow numerical model of the process industry process; and adjust the numerical model parameters with the collected process data of the process industry to obtain the mechanism model of the dynamics of the process industry process.

[0065] A3, using the mechanism model of the dynamics of the process industry process to expand the risk data to generate generated data that conforms to the physical laws and has a similar distribution to the real signal, and use the generated data to expand the original training samples.

[0066] In this specific embodiment, refer to Figure 2 As shown in the figure, in complex process, there are many process links and complex variable coupling, and risks often present the characteristics of nonlinearity, large time lag and fuzzy uncertainty that are difficult to effectively characterize. Conventional risk assessment methods are usually based on empirical formulas or statistical models, which are difficult to fully capture the dynamic changes in the process, and have limited ability to describe nonlinear and multi-dimensional process. Focusing on the fields of oil and gas, chemical industry and metallurgy, taking catalytic cracking units, ethylene units and blast furnace smelting units as typical process unit cases, risk mechanism modeling is carried out, integrating historical operation data, product test data of each period, environmental data and risk events and other data sets, and studying data-driven methods to realize the adaptive dynamic update mechanism and update method of risk model, and realize real-time dynamic risk analysis. In view of the complexity of process mechanism and the difficulty of risk perception and prediction caused by multiple production conditions, as well as the imbalance, heterogeneity and time-varying nature of data caused by multi-rate sampling. The study will analyze the process mechanism to establish a preliminary risk model, use multi-rate sampling data for risk assessment, and use machine learning to identify abnormal risk state information and predict future risk dynamic trends. In view of the uneven distribution and missing problems of field data, the data enhancement method based on mechanism model is studied to expand the sample area.

[0067] This embodiment aims to solve the problems of small samples and data imbalance by generating high-quality synthetic data or transferring knowledge in related fields, significantly alleviating the model bias and insufficient generalization ability caused by data limitations, thereby achieving comprehensive and in-depth mining and maximizing the value of process industry production and operation data. This innovative framework not only solves the technical bottleneck in current data processing, but also provides a reference solution for data-driven decision-making in similar complex industrial scenarios.

[0068] refer to Figure 3As shown, this is the risk assessment model training flow chart provided in this embodiment. When the risk assessment model is trained through the training set and the validation set and the dynamic risk analysis industry validation data set is constructed, it is necessary to integrate multi-rate sampling and multi-source heterogeneous data. Research on data preprocessing suitable for mixed dynamic and static data in multiple working conditions and multi-source and multi-type data conditions is a reliable guarantee for data quality. On this basis, the quality assessment and verification methods of the data set are studied, including effective verification in terms of robustness, integrity and representativeness.

[0069] In this embodiment, in order to further ensure the safety and stability of production operation, it also includes online verification of SIL level safety parameters of the process industry system. The specific steps are as follows:

[0070] (1) Determine the parameters required in the SIL verification process, which parameters include dynamic parameters and static parameters. In this embodiment, the required parameters involve variables of the sensing part, the logic part, and the execution part. The time-varying variables in each part are defined as dynamic variables, and the remaining variables are defined as static variables.

[0071] (2) Determine the dynamic parameter values ​​and static parameter values, and calculate the PFD and structural constraints of the SIS by combining the intelligent memory network algorithm to obtain the SIL verification level value of the SIS;

[0072] (3) Obtaining a dynamic SIL verification curve based on the SIL verification level value.

[0073] This embodiment provides a safe operation management solution for the case where both the SIL verification result of the current SIS and the SIL verification result of the fitting prediction can meet the SIL verification requirements of the SIS, and analyzes the dynamic trend of the PFD and SFF;

[0074] If the SIL verification results of the current SIS or the fitted predicted SIL verification results fail to meet the SIL verification requirements of the SIS, a system warning will be issued for the time nodes that fail to meet the SIL verification requirements, and a traceability analysis will be performed on the variables that fail to meet the requirements, providing a predictive maintenance strategy for the SIS.

[0075] In this embodiment, reverse reasoning is also performed based on the constructed causal probability graph and risk failure data to determine the root cause of the failure. The specific steps are as follows:

[0076] Based on the risk failure data, the node where the failure occurs is determined, and the cause-effect relationship represented by the arrow is reversely inferred to the previous associated node, and the conditional probability density is calculated to determine whether a failure occurs, until the root node in the probability graph model; if the probability density of the monitored sample is within the range of the upper and lower confidence limits, it is identified as a normal sample; otherwise, it is identified as a faulty sample.

[0077] In this embodiment, a fault tracing algorithm based on Bayesian network reasoning is used to trace the root cause of the fault and guide maintenance by constructing a Bayesian network.

[0078] In the preferred embodiment of the present invention, a risk assessment model is constructed, and an adaptive threshold is introduced in the risk assessment and classification module, including an adaptive global threshold and an adaptive local threshold. The classified target domain data is set to different thresholds through the adaptive threshold. The threshold selection adopts a parameter-free adaptive threshold selection method, and the risk level is determined according to different thresholds.

[0079] In this embodiment, a parameter-free adaptive threshold selection method is used, which does not rely on additional parameters and actual anomaly labels of the data. It can automatically search for possible threshold ranges based on the mean and standard deviation of the anomaly scores, and select the most appropriate threshold based on the number of point anomalies and sequence anomalies within the threshold range. It is more robust than the direct mean standard deviation method.

[0080] Device Embodiment

[0081] According to an embodiment of the present invention, a risk assessment and early warning device based on mechanism and data fusion is provided. Figure 4 As shown, it is a block diagram of a risk assessment and early warning device based on mechanism and data fusion provided in this embodiment. The risk assessment and early warning device based on mechanism and data fusion according to an embodiment of the present invention includes:

[0082] The data acquisition module 10 is used to obtain historical data collected during the system production process, including variables of the production site sensor part, logic part and execution part. The time-varying variables in each part are defined as dynamic variables, and the remaining variables are defined as static variables.

[0083] The data preprocessing module 20 is used to preprocess the multi-factor, multi-modal, multi-source heterogeneous data collected by the data collection module 10, and to align and fuse the preprocessed data.

[0084] Furthermore, the collected data is processed as follows:

[0085] (1) First, through data quality management, data standard management and data desensitization, multi-source data is pre-processed to ensure the integrity, consistency and security of the data.

[0086] (2) The data is then cleaned, including noise reduction, normalization, missing value filling, etc., and a fuzzy stacked autoencoder network is used for dimensionality reduction to effectively remove noise and retain important information.

[0087] (3) By designing efficient data interpolation, resampling and synchronization algorithms, the cleaned data is accurately aligned and fused across dimensions and frequencies to ensure the coherence and consistency of the data in the time dimension. The fusion of multimodal data can be achieved, for example, through the self-attention mechanism and Transformer to generate splicing features with higher information density, which is convenient for subsequent processing and analysis.

[0088] The relationship model building module 30 is used to determine the causal relationship between various factors in the production process by Granger causal analysis method based on the data processed by the data preprocessing module 20, and to describe the quantitative relationship between nodes by estimating the conditional probability density between various factors, so as to build a directed acyclic graph model or knowledge graph of the system, wherein the directed acyclic graph model or knowledge graph is a causal probability graph;

[0089] Constructing a risk assessment model module 40, for constructing a risk assessment model module, including a feature extraction module and a risk assessment and classification module, wherein the feature extraction module may be a graph convolutional network suitable for image feature extraction, and the risk assessment and classification module may be a spatiotemporal graph convolutional neural network or a graph convolutional network. In this embodiment, if the risk assessment and classification module may include a graph convolutional network layer GCN with L layers and a multi-layer perceptron layer MLP, the activation function of the GCN layer is the ReLU function, and the activation function of the MLP layer is the tanh activation function of the graph convolutional network.

[0090] The model training module 50 is used to input the data for setting risk node labels, the risk causal probability diagram, and normal data into the risk assessment model for training until the convergence conditions are met, thereby obtaining a model for monitoring and warning of production risks in the process industry.

[0091] The risk prediction and early warning module 60 is used to obtain real-time data during the system production process and monitor and warn of process industry production risks through risk assessment models.

[0092] The risk assessment and early warning method based on mechanism and data fusion provided in this embodiment firstly builds a multi-dimensional risk perception model, and realizes comprehensive coverage of production and operation risks by integrating multi-source data such as personnel, process, environment, equipment, and network; secondly, adopts signal processing and feature extraction technology to process and analyze multimodal data in real time and extract key risk features; then, based on the method of mechanism and data fusion, establishes a dynamic risk analysis model, and clarifies the risk mapping relationship by mining the joint features of data in different dimensions and scales including time, so as to realize comprehensive safety risk perception and decision support; it provides intelligent and scientific risk decision-making and optimization solutions for the production and operation of process industries, and significantly improves the safety, stability and operation efficiency of production systems. The method of this embodiment establishes a high-precision and high-robustness risk prediction and monitoring model, and clarifies the risk mapping relationship by mining the joint features of data in different dimensions and scales including time, so as to realize comprehensive safety risk perception and decision support.

[0093] In this embodiment, the risk assessment model is used to predict risk failure types, obtain the probability of occurrence of each risk failure type, and sort them according to the size of the occurrence probability to generate a failure risk list.

[0094] The embodiment of the present invention is an apparatus embodiment corresponding to the above-mentioned method embodiment. The specific operations of the processing steps of each module can be understood by referring to the description of the method embodiment, which will not be repeated here.

[0095] like Figure 5 As shown, the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the risk assessment and early warning method based on mechanism and data fusion in the above-mentioned embodiment is implemented.

[0096] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the risk assessment and early warning method based on mechanism and data fusion in the above-mentioned embodiment is implemented; or when the computer program is executed by a processor, the risk assessment and early warning method based on mechanism and data fusion in the above-mentioned embodiment is implemented.

[0097] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0098] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and the contents not described in detail in the specification of the present invention belong to the common knowledge of those skilled in the art.

Claims

1. A risk assessment and early warning method based on mechanism and data fusion, characterized in that The following steps are involved: Data collection; Obtain historical data collected during the system production process, including variables of the sensing part, logic part and execution part of the production site; Data preprocessing; Preprocess the multi-factor, multi-modal, multi-source heterogeneous data collected, and align and fuse the preprocessed data; Relationship model construction: Based on the processed data, the causal relationship between various factors in the production process is determined by the Granger causal analysis method, and the quantitative relationship between nodes is described by estimating the conditional probability density between various factors, and the causal probability diagram of the system is constructed; Constructing a risk assessment model; the risk assessment model includes a feature extraction module and a risk assessment and classification module, wherein the feature extraction module is a graph convolutional network, and the risk assessment and classification module is a spatiotemporal graph convolutional neural network or a graph convolutional network; and Model training: input the data with risk node labels and risk causal probability diagram, as well as normal data, into the risk assessment model for training until the convergence conditions are met, thereby obtaining a model for monitoring and early warning of production risks in the process industry; Risk prediction and early warning: obtain real-time data during the system production process, and monitor and warn of process industry production risks through risk assessment models.

2. The risk assessment and early warning method based on mechanism and data fusion as claimed in claim 1, characterized in that: The risk assessment and classification module includes a graph convolutional network layer GCN with L layers and a multi-layer perceptron layer MLP, the activation function of the GCN layer is the ReLU function, and the activation function of the MLP layer is the tanh activation function.

3. The risk assessment and early warning method based on mechanism and data fusion as claimed in claim 1, characterized in that: Further data preprocessing includes the following steps: Through data quality management, data standard management and data desensitization; The data is then cleaned, including noise reduction, normalization, missing value filling, and the fuzzy stacked autoencoder network is used to reduce dimension, remove noise and retain important information; Through data interpolation, resampling and synchronization algorithms, the cleaned data is accurately aligned and fused across dimensions and frequencies.

4. The risk assessment and early warning method based on mechanism and data fusion as claimed in claim 1, characterized in that: It also includes building a kinetic mechanism model of process industry processes for data expansion.

5. The risk assessment and early warning method based on mechanism and data fusion as claimed in claim 4, characterized in that: The construction of the kinetic mechanism model of the process industry process for data expansion specifically includes the following steps: Constructing conceptual dynamic models of process industries; Based on the constructed conceptual dynamic model of the process industry process, industrial data is analyzed to determine the data-driven method and the coupling relationship between various factors in the process, and the adaptive dynamic update mechanism and update method of the risk model are confirmed and implemented. Then, based on the finite element numerical simulation method, numerical simulation software is used to establish a dynamic numerical model of the process industry process and construct a dynamic mechanism model of the process industry process, which can be used to simulate the cause-effect relationship between various factors in the production process.

6. The risk assessment and early warning method based on mechanism and data fusion as claimed in claim 1, characterized in that: It also includes reverse reasoning based on the constructed causal probability diagram and risk failure data to determine the root cause of the failure. The specific steps are as follows: Based on the risk failure data, the node where the failure occurs is determined, and the cause-effect relationship represented by the arrow is reversely inferred to the previous associated node, and the conditional probability density is calculated to determine whether a failure occurs, until the root node in the probability graph model; if the probability density of the monitored sample is within the range of the upper and lower confidence limits, it is identified as a normal sample; otherwise, it is identified as a faulty sample.

7. The risk assessment and early warning method based on mechanism and data fusion as claimed in claim 1, characterized in that: The adaptive thresholds in the risk assessment and classification module include an adaptive global threshold and an adaptive local threshold. The classified target domain data is set to different thresholds through the adaptive thresholds. The threshold selection adopts a parameter-free adaptive threshold selection method to determine the risk level according to different thresholds.

8. A risk assessment and early warning device based on mechanism and data fusion, characterized in that include: The data acquisition module is used to obtain historical data collected during the system production process, including variables of the production site sensor part, logic part and execution part; The data preprocessing module is used to preprocess the multi-factor, multi-modal, multi-source heterogeneous data collected by the data acquisition module, and to align and fuse the preprocessed data; The relationship model building module is used to determine the causal relationship between various factors in the production process through the Granger causal analysis method based on the data processed by the data preprocessing module, describe the quantitative relationship between nodes by estimating the conditional probability density between various factors, and build the causal probability diagram of the system; Constructing a risk assessment model module, which is used to construct the risk assessment model module, including a feature extraction module and a risk assessment and classification module, wherein the feature extraction module is a graph convolutional network, and the risk assessment and classification module is a spatiotemporal graph convolutional neural network or a graph convolutional network; A model training module, which is used to input the data for setting risk node labels and the risk causal probability diagram, as well as normal data, into the risk assessment model for training until the convergence conditions are met, thereby obtaining a model for monitoring and early warning of production risks in the process industry; and Risk prediction and early warning module: used to obtain real-time data during the system production process, and monitor and warn of process industry production risks through risk assessment models.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the risk assessment and early warning method based on mechanism and data fusion as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the risk assessment and early warning method based on mechanism and data fusion as described in any one of claims 1 to 7 is implemented.

Images