Model-data double driven dynamic risk early warning method and system for hazardous chemical equipment

Abstract

The present application relates to the technical field of dangerous equipment, and particularly relates to a model-data dual driving type dynamic risk early warning method and system for dangerous equipment, which comprises the following steps: a data collection step, a pre-training model construction step, a first real-time data acquisition step, a first prediction step, a second real-time data acquisition step, a model reinforcement fine-tuning step, a second prediction step, a failure probability calculation step, a failure consequence calculation step, a risk grade determination step, a risk early warning step and a cyclic iteration step. The model-data dual driving method combining real-time data acquisition, deep learning pre-training and reinforcement fine-tuning overcomes the problems of poor real-time performance, low prediction accuracy and lack of model adaptability in the background technology, realizes dynamic risk early warning for dangerous equipment, has the advantages of strong real-time performance, accurate prediction and good model adaptability, and significantly improves the efficiency and reliability of risk assessment and early warning for dangerous equipment.

Description

Technical Field

[0001] This invention relates to the field of hazardous chemical equipment technology, and in particular to a model-data dual-driven dynamic risk early warning method and system for hazardous chemical equipment. Background Technology

[0002] Existing risk assessment methods for hazardous chemical equipment typically rely on historical data and manual analysis. For example, existing literature 1 (Publication No.: CN117829578A, Title: A Risk Assessment Method for a Surface Gathering and Transportation System of a Depleted Water-Inundated Gas Reservoir) proposes a risk assessment method based on historical failure information. This method collects historical failure information of the equipment, calculates the basic failure probability, and corrects it by incorporating multiple failure probability influencing factors, ultimately obtaining the risk level of the equipment. However, the method in literature 1 has the following technical problems:

[0003] It relies primarily on static historical data, lacking monitoring and analysis of real-time equipment operating status, thus failing to provide dynamic, real-time early warnings of equipment risks. The inability to promptly capture anomalies or emergencies during equipment operation leads to insufficient timeliness and accuracy in risk assessment, making it difficult to prevent accidents in a timely manner. Summary of the Invention

[0004] The purpose of this invention is to provide a model-data dual-driven dynamic risk early warning method and system for hazardous chemical equipment, addressing the shortcomings of existing technologies. This method aims to achieve dynamic assessment and real-time early warning of equipment risks by collecting real-time operational data and combining deep learning models and reinforcement learning algorithms. This approach can promptly detect abnormal situations during equipment operation, improve the accuracy and timeliness of risk prediction, and enhance the safety of hazardous chemical equipment.

[0005] This invention achieves the above objective through the following technical solution: a model-data dual-driven dynamic risk early warning method for hazardous chemical equipment, comprising the following steps:

[0006] Data collection steps: Collect historical operational data and / or simulation experiment data of hazardous chemical equipment and perform data preprocessing to obtain a sample training dataset;

[0007] Pre-trained model building steps: Use the sample training dataset to pre-train the deep learning model that combines residual network and bidirectional long short-term memory network, learn the mapping relationship between input data and device state parameters, obtain the initial model parameters, and build the deep learning pre-trained model;

[0008] The first real-time data acquisition step: Use sensors to collect the current real-time operating data of the hazardous chemical equipment that needs to be monitored and warned, and perform data preprocessing to obtain the first real-time input data that meets the model input requirements;

[0009] First prediction step: Input the first real-time input data obtained in the first real-time data acquisition step into the deep learning pre-trained model established in the pre-trained model building step. The deep learning pre-trained model outputs the first prediction parameters used to calculate the necessary input conditions for the failure probability and failure consequences of the device.

[0010] The second real-time data acquisition step: Real-time acquisition of the next real-time operating data of the current hazardous chemical equipment and preprocessing to obtain the second real-time input data that meets the model input requirements;

[0011] Model reinforcement and fine-tuning steps: The second real-time input data is used as the actual running parameters in the first prediction step of the model, and compared with the first prediction parameters. The comparison deviation is calculated, thereby performing reinforcement learning and fine-tuning on the deep learning pre-trained model, updating the model parameters, and obtaining the deep learning pre-trained-reinforcement model; specifically, the comparison deviation is calculated and feedback is given, and the model weights are adjusted based on the feedback so that it can more accurately reflect the operation of the device in the real environment.

[0012] Second prediction step: Input the second real-time input data obtained in the second real-time data acquisition step into the deep learning pre-trained-reinforcement model obtained in the model reinforcement and fine-tuning step. The deep learning pre-trained-reinforcement model outputs the second prediction parameters used to calculate the necessary input conditions for the failure probability and failure consequences of the device.

[0013] Failure probability calculation steps: Calculate the failure probability of the current hazardous chemical equipment using the second prediction parameters output from the second prediction step;

[0014] Failure consequence calculation steps: Calculate the failure consequences of the current hazardous chemical equipment using the second prediction parameters output from the second prediction step;

[0015] Risk level determination steps: Based on the calculated failure probability and failure consequences, and according to the preset failure probability level and failure consequence level, the current risk level of the equipment is determined using a risk matrix.

[0016] Risk warning procedure: If the current risk level reaches the preset warning threshold, the warning mechanism is triggered;

[0017] Iterative steps: Repeatedly execute the second real-time data acquisition step to the risk warning step.

[0018] As a further aspect of the present invention: the data collection step includes, prior to the data collection step:

[0019] Range Presetting Step: Based on the type of hazardous chemical equipment, preset the range of values ​​for failure probability and the range of values ​​for consequence characterization indicators. Assign corresponding failure probability levels to different ranges of values ​​for failure probability, and assign corresponding failure consequence levels to different ranges of values ​​for values ​​for consequences characterization indicators (e.g., area consequence CA of pressure vessels). Specifically, the range of values ​​for consequences characterization indicators is the range of values ​​for the final consequence area CA.

[0020] Risk matrix construction steps: Map the failure probability level and failure consequence level to the risk matrix level one by one, and match them with the corresponding risk level in the risk matrix to establish the mapping relationship between the failure probability level, failure consequence level and risk level.

[0021] Parameter determination steps: Determine the parameters that need to be collected that match the type of hazardous chemical equipment, and select the appropriate sensors;

[0022] The first real-time data acquisition step is as follows: using the sensor selected in the parameter determination step, the current real-time operating data of the hazardous chemical equipment that needs to be monitored and warned is collected in real time and the data is preprocessed to obtain the first real-time input data that meets the model input requirements.

[0023] The second real-time data acquisition step is as follows: using the sensor selected in the parameter determination step, the next real-time operating data of the current hazardous chemical equipment is acquired in real time and preprocessed to obtain the second real-time input data that meets the model input requirements;

[0024] The failure consequence calculation steps are as follows: using the second prediction parameters output from the second prediction step, calculate the consequence characterization index of the current hazardous chemical equipment (e.g., the area consequence CA of the pressure vessel);

[0025] The risk level determination steps are as follows: based on the calculated failure probability and consequence characterization index (e.g., area consequence CA), match the preset value range in the search range preset step, determine the failure probability level value and failure consequence level value, and determine the corresponding risk level in the risk matrix to obtain the current risk level.

[0026] The beneficial effects of this invention are:

[0027] Enhanced Real-Time and Dynamic Performance: In the first and second real-time data acquisition steps, this invention collects the current operational data of hazardous chemical equipment in real time and performs data preprocessing to obtain real-time input data that meets the model's input requirements. Through real-time data acquisition and processing, the model can promptly obtain the latest operating status of the equipment, enhancing the real-time and dynamic performance of risk warnings. It can quickly respond to changes in equipment status and achieve timely warnings of potential risks.

[0028] Improved Prediction Accuracy: In the pre-training model construction step, this invention utilizes a sample training dataset to pre-train a deep learning model combining residual networks and bidirectional long short-term memory networks, establishing a deep learning pre-trained model. Through the powerful feature extraction and learning capabilities of the deep learning model, the model can fully exploit the complex relationships in historical and real-time data, improving the prediction accuracy of key equipment parameters and thus more accurately calculating the failure probability and consequences.

[0029] Enhanced Model Adaptability and Accuracy: In the model reinforcement and fine-tuning step, this invention uses the second real-time input data as the actual operating parameters, compares them with the first prediction parameters, calculates the comparison deviation, and performs reinforcement learning and fine-tuning on the deep learning pre-trained model to update the model parameters, resulting in a deep learning pre-trained-reinforcement model. Through reinforcement learning fine-tuning, the model can continuously optimize based on feedback from real-time data, adaptively adjusting model parameters, thus improving the model's adaptability to the actual operating state of the equipment and its prediction accuracy.

[0030] Iterative optimization for continuous improvement: This invention incorporates iterative steps, repeatedly executing the entire process from second-level real-time data acquisition to risk warning. Through iterative iteration, the model continuously receives new real-time data, performs reinforcement learning and prediction, gradually optimizes model performance, and achieves continuous monitoring and early warning of equipment risks.

[0031] Improved Accuracy of Risk Warnings: In the risk level determination step, this invention determines the current risk level of the equipment based on the calculated failure probability and consequences, according to preset failure likelihood and consequence levels, using a risk matrix. Through precise risk level calculation and the application of the risk matrix, the risk status of the equipment can be assessed more accurately. When the risk level reaches a preset warning threshold, a timely warning mechanism is triggered, improving the accuracy and reliability of risk warnings.

[0032] In summary, this invention overcomes the problems of poor real-time performance, low prediction accuracy, and lack of model adaptability in existing technologies by combining real-time data acquisition, deep learning pre-training, and reinforcement fine-tuning with a model-data dual-drive approach. It achieves dynamic risk warning for hazardous chemical equipment, with advantages such as strong real-time performance, accurate prediction, and good model adaptability. This significantly improves the efficiency and reliability of risk assessment and early warning for hazardous chemical equipment, timely detects abnormal situations in equipment operation, and effectively prevents accidents. Attached Figure Description

[0033] Figure 1 This is a schematic block diagram of the tree structure of the device correction coefficients of the present invention.

[0034] Figure 2 This is a schematic diagram of the material matching curve structure for determining the brittle fracture secondary factor of the present invention.

[0035] Figure 3 This is a schematic diagram of the risk matrix of the present invention. Detailed Implementation

[0036] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0037] like Figures 1-3 As shown, this embodiment of the invention provides a model-data dual-driven dynamic risk early warning method for hazardous chemical equipment, including: a scope preset step, a risk matrix construction step, a parameter determination step, a data collection step, a pre-trained model construction step, a first real-time data acquisition step, a first prediction step, a second real-time data acquisition step, a model enhancement and fine-tuning step, a second prediction step, a failure probability calculation step, a failure consequence calculation step, a risk level determination step, a risk early warning step, and a cyclic iteration step.

[0038] The specific implementation method is as follows:

[0039] I. Sensor Data Acquisition: The sensor detects data such as wall thickness and cracks.

[0040] II. Sensor Data Conversion and Transmission: Data collected by sensors such as ultrasonic sensors is read into the pre-processing system of the dynamic risk warning system by using a Python program written with the gpiozero library. In the pre-processing system, the data is automatically converted into wall thickness, crack size, etc., which can be identified by the risk identification system.

[0041] Transformation method: A multi-damage detection technology for mobile pressure vessels based on ultrasonic guided waves is proposed. The propagation mechanism of circumferential and axial ultrasonic guided waves is studied, the damage response characteristics of their circumferential and axial guided wave modes are obtained, and the excitation mode of the circumferential and axial guided wave modes is studied to realize the identification and detection of damage forms such as cracks, thinning, and wear at the mouth and body of the mobile pressure vessel.

[0042] This study investigates the propagation characteristics of ultrasonic guided waves in mobile pressure vessels, obtains damage response features and ultrasonic excitation methods, and aims to achieve high-precision detection of multiple damages. The specific steps are as follows:

[0043] (1) Based on the material properties and structural form of the mobile pressure vessel, a wave equation is established, and the ultrasonic guided wave circumferential mode and axial mode are solved by numerical analysis.

[0044] (2) Consider damage in the circumferential and axial ultrasonic guided wave propagation models, analyze the response characteristics caused by damage, and use this as the basis for damage identification.

[0045] (3) Based on the propagation laws of circumferential and axial ultrasonic guided waves, select appropriate circumferential and axial guided wave modes and determine excitation parameters such as excitation frequency.

[0046] Multi-sensor data fusion processing for mobile pressure vessels based on heterogeneous graph deep learning: A multi-sensor data fusion framework for ultrasonic, temperature, pressure, and flow angle data of mobile pressure vessels is constructed based on heterogeneous graph deep learning. The specific steps are as follows:

[0047] A multi-source feature encoding network (NN-1 network) is constructed to encode multi-source data separately, obtaining the data embedding of each node to form a self-representation that differs from information from its neighbors. Information fusion is then performed through a similar-feature neighbor aggregation network (NN-2 network) and a heterogeneous feature hybrid network (NN-3 network). The aggregation of neighbor nodes of the same type and those of different types is divided into two levels. For aggregation of neighbor nodes of the same type, Bi-LSTM is still used to obtain an average pooled data representation (similar-feature neighbor aggregation network, NN-2 network); for different types of node sets that have different impacts on the current node, an attention mechanism is used to learn the importance of nodes, assigning greater weight to features with greater influence to meet the needs of the objective (heterogeneous feature hybrid network, NN-3 network). Finally, multi-source heterogeneous data fusion processing is achieved.

[0048] Safety Early Warning for Mobile Pressure Vessels Based on Deep Learning Pre-trained-Reinforcement Models: Based on historical data, simulation experimental data, and actual operational data, deep learning pre-trained-reinforcement models are constructed, including a safety early warning pre-training model based on residual networks and bidirectional long short-term memory, and a safety early warning reinforcement model based on incremental learning. The specific steps are as follows:

[0049] (1) A residual network model for extracting features of the state information is constructed based on historical data and simulated experimental data. The model is used to extract features of the input data. The preprocessed dataset is used as the input of the residual network, and the residual network is used to perform in-depth mining of the features of the sample data.

[0050] (2) Input the residual network output features into the bidirectional long short-term memory model. The bidirectional long short-term memory model can effectively capture the sequential dependencies in the sequence through its bidirectional structure.

[0051] (3) Initialize the reinforcement model using the pre-trained model, update the security warning reinforcement model based on the actual running data using the model fine-tuning strategy, and add a non-convex regularization term to the loss function of the reinforcement model to enhance the penalty characteristics by utilizing the non-convex characteristics and accelerate the convergence of the algorithm.

[0052] After model conversion, the data is fed into the subsequent processing module, ultimately resulting in readable wall thickness, crack conversion, and transmission of sensor waveform data.

[0053] III. Calculation of Equipment Failure Probability Level (Taking Storage Tanks as an Example)

[0054] Establish quantifiable risk identification criteria. Formula for the failure probability of pressure vessels and piping equipment:

[0055] F = F G ×F M ×F L ×F E

[0056] The wall thickness data obtained from preprocessing is adjusted using actual working scenarios and various standard correction factors to obtain F. G —Average failure probability of similar equipment, F M —Management system evaluation coefficient, F L —Influence coefficient of defects exceeding the standard, F E —Equipment correction factor, thus obtaining the failure probability F.

[0057] 1.F G —Calculation of average failure probability for similar equipment

[0058] Table 1

[0059]

[0060] Based on Table 1, F is obtained from the crack data measured by the sensor. G。

[0061] 2.F M —Management System Evaluation Rating In the formula: F M - Management system evaluation coefficient, x - Management system score.

[0062] Generally, the same F is used for all equipment within the same company. M

[0063] Table 2 shows the corresponding enterprise score x, from which F is obtained. M

[0064]

[0065] 3.F E —Equipment correction factor: see Figure 1 F E —The equipment correction factor consists of four parts: a) technical module factor, b) general condition factor, c) mechanical factor, and d) process factor. a. Technical module factor-Ⅰ: thinning sub-factor

[0066] If the hazardous chemical equipment is a container type: 0.0]; Where: A rtAn index representing the severity of corrosion thinning; t rd The minimum wall thickness measured in the most recent inspection and testing is expressed in millimeters (mm); 'a' represents the service life in years; 'r' represents the corrosion rate in millimeters per year (mm / year); 't' represents the corrosion rate in millimeters per year (mm / year). min The calculated wall thickness is expressed in millimeters (mm) (the wall thickness value obtained during design according to the calculation formula); CA represents the corrosion allowance, also expressed in millimeters (mm).

[0067] If the hazardous chemical equipment is a pipeline type:

[0068] In the formula: t represents the nominal thickness in millimeters (mm); the corrosion rate r is obtained based on the medium conditions (sulfuric acid), and the rest are obtained based on the data measured by the sensor and the equipment parameters.

[0069] Get A rt The thinning factor was then obtained by interpolation according to Table 3 below.

[0070] Table 3

[0071] A rt ]]> 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20 Thinning factor 1 1 1 1 1 1 1 2 7 10 A rt ]]> 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65 Thinning factor 15 25 70 110 160 210 260 360 640

[0072] II: Stress corrosion cracking secondary factor (using sulfuric acid medium, stainless steel tank)

[0073] First, determine the tank's susceptibility to stress corrosion cracking caused by the medium according to Table 4 below. For example, for sulfuric acid media, the stainless steel tank should be selected based on its medium susceptibility. Then, determine the severity index.

[0074] Table 4

[0075] Sensitivity none Low middle high Severity Index (PASCC) 1 50 500 5000

[0076] Then, determine the stress corrosion cracking subfactor according to Table 5 below.

[0077] Table 5

[0078] Severity Index (PASCC) 1 50 500 5000 Stress corrosion cracking secondary factor 1 1 10 125

[0079] III: High-Temperature Hydrogen Corrosion (HTHA) Sub-Factor

[0080] First, the material's HTHA damage sensitivity was determined; the time-related parameter P was used. V Sure

[0081]

[0082] In the formula: P v - Parameters relating temperature and hydrogen partial pressure to usage time; P H2 - Hydrogen partial pressure, in megapascals (MPa); T represents temperature, in degrees Celsius (°C);

[0083] t represents the running time in hours (h); the sensitivity is then determined by the equipment and materials (see Table 6 below):

[0084] Table 6

[0085]

[0086] Then, the magnitude of the high-temperature hydrogen corrosion (HTHA) subfactor is determined (see Table 7 below);

[0087] Table 7

[0088] Sensitivity none Low middle high Damage has been observed High-Temperature Hydrogen Corrosion (HTHA) Subfactor 1 18 1800 1800 2000

[0089] IV: Secondary factor of mechanical fatigue damage (acting on pipelines)

[0090] a. Determine the number of fatigue failures that have occurred previously and determine the baseline sensitivity according to the table (see Table 8 below);

[0091] Table 8

[0092] Previously invalid Benchmark sensitivity none 1 1 time 50 >1 time 500

[0093] b. Determine the noise level in the pipeline and multiply it by the adjustment factor to obtain the baseline sensitivity (see Tables 9 and 10 below);

[0094] Table 9

[0095] Vibration noise Benchmark sensitivity smaller 1 medium 50 serious 500

[0096] Table 10

[0097] Vibration duration / week Adjustment coefficient 0～<2 1 2～＜13 0.2 13～52 0.02

[0098] c. Determine the type of internal stress cycle in the pipeline and determine the baseline sensitivity (see Table 11 below, Cyclic Stress Types);

[0099] Table 11

[0100] Source of cyclic force within 15.24m Benchmark sensitivity Reciprocating machinery 50 Safety valve vibration 25 Valves with high pressure drop 10 none 1

[0101] d. Take the maximum value from a to c as the baseline sensitivity;

[0102] e. Baseline sensitivity multiplied by complexity adjustment factor (see Table 12 below);

[0103] Table 12

[0104] Complexity, per 15.240m of pipe Adjustment coefficient 0-5 branch pipes, pipe fittings, etc. 0.5 5 to 10 branch pipes, pipe fittings, etc. 1 More than 10 branch pipes, pipe fittings, etc.

[0105] f. Multiply by the answer or direct pipe adjustment factor (see Table 13 below);

[0106] Table 13

[0107] Connector type Adjustment coefficient Threaded connection 2 Socket welding 2 Convex saddle shape 2 Concave saddle shape 1 Small welded pipe 0.2 Reducing tube 0.02

[0108] g. Multiply by the pipeline condition adjustment factor (see Table 14 below);

[0109] Table 14

[0110] condition Adjustment coefficient Pipe support lost or damaged 2 The weight was not supported 2 The reinforcing plate broke. 2 The reinforcing plate / pipe support is welded directly onto the pipe. 2 good condition 1

[0111] h, multiplied by the branch pipe diameter adjustment coefficient (see Table 15 below);

[0112] Table 15

[0113]

[0114] i. Determine the mechanical fatigue damage sub-factor based on a~h (>=5000 indicates failure).

[0115] V: External damage subfactor (carbon steel)

[0116] a. Determine the atmospheric corrosion rate of carbon steel based on temperature and environment (see Table 16 below for recommended atmospheric corrosion rates of carbon steel);

[0117] Table 16

[0118]

[0119] b. Adjust the atmospheric corrosion initiation time according to the coating (see Table 17 below);

[0120] Table 17

[0121]

[0122] c. Pipe support compensation and adjustment (see Table 18 below);

[0123] Table 18

[0124] There is compensation No compensation Damage rate = Damage velocity × 2.0 Damage rate = Damage velocity × 1.0

[0125] d. Interface compensation adjustment (see Table 19 below);

[0126] Table 19

[0127] There is compensation No compensation Damage rate = Damage velocity × 2.0 Damage rate = Damage velocity × 1.0

[0128] e. Calculate the thinning factor;

[0129] If the hazardous chemical equipment is a container type: In the formula: A rt An index representing the severity of corrosion thinning; t rd The smallest wall thickness measured in the most recent inspection and testing is indicated by 'a', in millimeters (mm); 'a' represents the service life in years; 'r' represents the corrosion rate in millimeters per year (mm / year); 't' represents the corrosion rate.min Calculate wall thickness in millimeters (mm) (the wall thickness value obtained from the calculation formula during design); CA corrosion allowance in millimeters (mm);

[0130] If the hazardous chemical equipment is a pipeline type:

[0131] In the formula: t represents the nominal thickness, in millimeters (mm); the rest is the same as the formula above.

[0132] The external damage subfactor TMF is obtained from interpolation in Table 20 below.

[0133] Table 20

[0134] A rt ]]> 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20 Thinning factor 1 1 1 1 1 1 1 2 7 10 A rt ]]> 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65 Thinning factor 15 25 70 110 160 210 260 360 640

[0135] VI: Brittle fracture secondary factor

[0136] a. Determine the minimum operating temperature T of the equipment. min b. Determine the equipment thickness; c. Determine the impact test temperature T based on the material. ref ;

[0137] For example, the materials and curves are shown in Table 21 below, and the curves are as follows: Figure 2 ;

[0138] Table 21

[0139]

[0140] d. Select different tables depending on whether the device performs PWHT module processing (see Table 22 and Table 23 below);

[0141] Table 22

[0142]

[0143] Table 23

[0144]

[0145] e. If the equipment is free from fatigue or SCC damage and has been operating safely for a long time at normal operating temperature, the brittle fracture subfactor determined by module d can be divided by 100.

[0146] Finally, SUM (thinning factor + stress corrosion cracking factor + high-temperature hydrogen corrosion (HTHA) factor + mechanical fatigue damage factor (acting on pipelines) + external damage factor (carbon steel) + brittle fracture factor) is taken as the technical module factor F. E .

[0147] b. General condition factor

[0148] I: Sub-factors of factory conditions

[0149] Determined based on factory equipment conditions (see Table 24 below)

[0150] Table 24

[0151] Device conditions level Assignment Higher than current industry standards A -1.0 Meets the general requirements of current industrial standards B 0 Slightly below the general requirements of current industry standards C 1.5 Significantly lower than the general requirements of current industry standards D 4.0

[0152] II: Sub-factors of cold climate operation

[0153] Based on winter temperatures, the lowest average daily temperature was selected (see Table 25 below).

[0154] Table 25

[0155] Winter temperature Assignment Above 4℃ 0 7℃～4℃ 1.0 -29℃~-7℃ below 2.0 -29℃ or below 3.0

[0156] III: Secondary Factors of Seismic Activity

[0157] Determined by region (see Table 26 below)

[0158] Table 26

[0159] Characteristic periodic region of seismic response spectrum Assignment Zone 1 0 Zone 2 1.0 Zone 3 2.0

[0160] Ultimately, SUM (factory condition sub-factor + cold climate operation sub-factor + seismic activity sub-factor) was chosen as the general condition factor.

[0161] c. Mechanical factors

[0162] I: Structural complexity factor

[0163] 1. Equipment complexity sub-factor

[0164] All pipes and manholes with an outer diameter of not less than 50 mm, whether in use or out of use, are included in the calculation, as determined by Table 27 below.

[0165] Table 27

[0166]

[0167] 2. Pipeline complexity sub-factors

[0168] Calculation formula: n=(n1×10.0)+(n2×20.0)+(n3×3.0)+(n4×5.0);

[0169] In the formula: n represents the total complexity coefficient of the pipeline; n1 represents the number of flange interfaces, with a complexity coefficient of 10.0 for each flange interface; n2 represents the number of injection points, with a complexity coefficient of 20.0 for each injection point; n3 represents the number of straight pipes, with a complexity coefficient of 3.0 for pipelines connected in the form of tees; n4 represents the number of valves (excluding safety valves), with a complexity coefficient of 5.0 for each valve.

[0170] Next, determine the pipe length L, and then determine the pipe complexity sub-factor using n / L and Table 28 below:

[0171] Table 28

[0172]

[0173] 3. Equipment Result Rationality Sub-factor

[0174] The equipment result rationality factor is set to 0 by default. If there is obvious irrationality in the equipment structure, it is set to 5.

[0175] SUM (equipment complexity sub-factor + pipeline complexity sub-factor + equipment outcome rationality sub-factor) is taken as the structural complexity factor.

[0176] II: Construction Standard Sub-Factor

[0177] Determined from Table 29 below

[0178] Table 29

[0179]

[0180] III: Life Cycle Sub-Factor

[0181] It is determined by the ratio of service life to design life (see Table 30 below).

[0182] Table 30

[0183] Ratio of service life to design service life / % Assignment 0～7 2.0 8～75 0 76～100 1.0 ＞100 4.0

[0184] IV: Safety Factor

[0185] 1. Operating pressure sub-factor

[0186] The pressure is determined based on the ratio m of the operating pressure to the design pressure (see Table 31 below):

[0187] Table 31

[0188] m Assignment m＞1.0 5.0 09＜m≤1.0 1.0 0.7＜m≤0.9 0 0.5≤m≤0.7 -1.0 m＜0.5 -2.0

[0189] 2. Operating temperature factor

[0190] As determined in Table 32 below, values ​​outside the table range are assigned as 0;

[0191] Table 32

[0192] Material Category Operating temperature / °C Assignment carbon steel ＞288 2.0 1%–5% Cr steel ＞343 2.0 5%–9% Cr steel ＞399 2.0 3074 / 316 stainless steel ＞816 2.0 All steel ＜-20 1.0

[0193] 3. Take SUM (operating pressure factor + operating temperature factor) as the safety factor.

[0194] V: Vibration monitoring sub-factors are determined by the vibration source and monitoring program (see Table 33 below);

[0195] Table 33

[0196]

[0197] SUM (structural complexity sub-factor + construction specification sub-factor + life cycle sub-factor + safety factor sub-factor + vibration monitoring sub-factor) is taken as the mechanical factor.

[0198] d. Process factors

[0199] I: Process continuity sub-factor

[0200] 1. Planned parking sub-factor

[0201] Based on the number of planned parking sessions per year over the past three years, determine the number of sessions as shown in Table 34 below;

[0202] Table 34

[0203] Number of planned parking sessions per year Assignment 0～1 -1.0 1.1～3 0 3.1～6 1.0 ＞6 1.5

[0204] 2. Unplanned parking sub-factor

[0205] Based on the number of unplanned parking sessions per year over the past three years, determine the number of parking sessions according to Table 35 below;

[0206] Table 35

[0207] Number of unplanned parking sessions per year Assignment 0～1 -1.5 1.1～3 0 3.1～6 2.0 ＞6 3.0

[0208] Take sUM (planned shutdown sub-factor + unplanned shutdown sub-factor) as the process continuity sub-factor.

[0209] II: Sub-factor of process stability

[0210] Determined by Table 36 below;

[0211] Table 36

[0212] Stability evaluation Assignment More stable than average process -1.0 Comparable to average process stability 0 Lower than average process stability 1.0 Significantly lower than average process stability 2.0

[0213] III: Safety Valve Status Sub-Factor

[0214] 1. Maintain program sub-factors

[0215] Determined by Table 37 below;

[0216] Table 37

[0217] Safety valve maintenance status category Assignment Safety valves requiring delayed maintenance account for less than 5%. A -1.0 The percentage of safety valves requiring delayed maintenance is greater than or equal to 5% but less than 15%. B 0 The percentage of safety valves requiring delayed maintenance is greater than or equal to 15% but less than 25%. c 1.0 Safety valves that require delayed maintenance account for 25% or more. D 2.0

[0218] 2. Fouling State Factor

[0219] Determined by Table 38 below;

[0220] Table 38

[0221] Scaling trend category Assignment No excessive scaling A The presence of polymers or other scaling substances has led to occasional accumulation in certain parts of the system. B 20 It is prone to scaling, and there has been a history of frequent deposit buildup on safety valves and / or other components of the system. c 4.0

[0222] 3. Corrosion state sub-factor

[0223] Determined by Table 39 below;

[0224]

[0225] 4. Cleanliness State Sub-factor

[0226] For process fluids that have no tendency to scale, no corrosive substances or other contaminants, the values ​​in Table 40 below shall be used; otherwise, the value shall be 0.

[0227] Table 40

[0228] Cleaning status Assignment yes -1.0 no

[0229] The sub-factors sUM (maintenance procedure sub-factor + fouling state sub-factor + corrosion state sub-factor + cleanliness sub-factor) are used as the safety valve state sub-factors. The sub-factors sUM (process continuity sub-factor + process stability sub-factor + safety valve state sub-factor) are used as the process factors. A tree structure is used to divide the calculation parameters into blocks, and an automatic calculation program is embedded in each module.

[0230] The final equipment correction factor F was determined. E = SUM(Technical Module Factor + General Condition Factor + Mechanical Factor + Process Factor).

[0231] 4.F L —Influence coefficient of defects exceeding standards

[0232] I: Determination of the influence coefficient of defects exceeding the standard when the equipment does not have a time-related degradation mechanism

[0233] 1. Take F when there are no defects in the equipment. L =1.0;

[0234] 2. When there are defects exceeding the standard in the equipment:

[0235] a. If the defects exceeding the standard are within the allowable range of the equipment's periodic inspection procedures or pass the usage evaluation, and the safety factor is greater than 2, then F is selected. L =1.0;

[0236] b. Use products that have passed evaluation and have a safety factor of less than 2, F L =1.5;

[0237] c. Evaluation results are unsatisfactory, but acceptable. (F) L =500.

[0238] II: Determination of the influence coefficient of defects exceeding standards when equipment exhibits time-related degradation mechanisms

[0239] According to the planned service time T n (h) and remaining lifetime T SL The ratio Q of (h) is determined as shown in Table 41 below:

[0240] Table 41

[0241]

[0242] After obtaining the failure probability, it is then classified according to the failure likelihood level, as shown in Table 42 below:

[0243] Table 42

[0244]

[0245]

[0246] IV. Calculation of Failure Consequence Level (Taking Storage Tanks as an Example)

[0247] The failure consequence level is determined based on the usage scenario and the severity of the hazard (see Table 43 below).

[0248] Table 43

[0249]

[0250] Various equipment are characterized by their corresponding consequences, such as pressure vessels using area consequences.

[0251] 1. Final equipment damage area (CA) cmd ), final area of ​​personal injury (CA) inf ):

[0252]

[0253] Where: CA cmd -Final equipment damage area; CA inf -Final area of ​​personal injury; -Area of ​​equipment damage resulting from the fire; -Area of ​​personal injury;

[0254]

[0255] In the formula: -Area of ​​equipment damaged by the fire; gff n - The probability of equipment failure of the same type corresponding to the nth type of leakage hole; gff total -The sum of the average failure probabilities of the same type of equipment corresponding to all leak holes; - The nth type of hole leakage is determined by the area of ​​equipment damage caused by combustion consequences based on the leakage type and the self-ignition temperature equilibrium.

[0256] Ⅰ: Calculate gff n

[0257] Select according to the size of the leakage hole as shown in Table 44;

[0258] Table 44

[0259]

[0260] >0mm~6mm >6mm~25mm >25mm~100mm >100mm 8 x 10 -6 ]] 2 x 10 -5 ]] <![CDATA[2×10 -6 ]]> <![CDATA[6×10 -7 ]]>

[0261] II: Calculate gff total

[0262]

[0263] III: Calculation

[0264] 1. Determine the leak hole number, and select it according to the size of the leak hole as shown in Table 45 below;

[0265] Table 45

[0266]

[0267] Determine its leakage area A n :

[0268]

[0269] 2. Determine the leakage rate

[0270] For the liquid phase, the leakage rate W for each type of leak hole n :

[0271]

[0272] In the formula: W n - Leakage rate for each type of leak hole; C d - Leakage coefficient is taken as 0.61; ρ1 - density of the leaking medium; A n - Leakage area; g c - Mechanical constant, taken as 1.0 (kg·m); P s - Pressure inside the tank; P atm-Atmospheric pressure.

[0273] 3. Calculate the potential maximum media leakage mass of the nth type of leak. avail·n :

[0274] mass add·n =180×W n mass avail·n =mass comp +mass add·n ;

[0275] Where: mass avail·n - The maximum potential media leakage rate of the nth type of leak; mass comp - The amount of media stored in the equipment; W n - Leakage rate for each type of leak hole.

[0276] 4. Determine the actual leakage mass of the nth type of leakage hole. n

[0277] The actual leakage rate of the nth type of leak is rate n :

[0278] rate n =W n (1-fact di );

[0279] Where: fact di - Leakage rate reduction coefficient, selected according to Table 46 below;

[0280] Table 46

[0281]

[0282] The maximum leakage duration of the nth type of leakage hole is ld max·n Select from Table 47 below:

[0283] Table 47

[0284]

[0285] The actual leakage duration of the nth type of leak is ld n :

[0286]

[0287] The actual leakage amount of the nth type of leakage hole is mass n :

[0288] mass n =min[(rate n ×ld n ), massavail·n ];

[0289] Based on the mitigation system, the effect area reduction coefficient is selected. mit (See Table 48 below);

[0290] Table 48

[0291] Slow down system type <![CDATA[Consequence area reduction coefficient (fact min )]]> Venting system, and isolation system of level B or higher. 0.25 Fire sprinkler system and fire monitor 0.20 foam sprayer 0.15 Fire monitor 0.05

[0292] Determine the energy efficiency adjustment coefficient (eneff) for the nth type of leakage hole. n :

[0293] eneff n =4lg(2.2223×mass) n -15;

[0294] continuous leaks n =1.

[0295] 5. Determine the area of ​​equipment damage and personnel injury caused by continuous leakage.

[0296] The constants a and b are determined by the medium (see Table 49 below, selection of constants in the formula for calculating the area of ​​equipment damage caused by combustion);

[0297]

[0298] A. Non-self-igniting medium:

[0299]

[0300] B. Potentially flammable medium:

[0301]

[0302] Where: fact mit - Consequence area reduction coefficient; a, b - Medium constants; rate n - Actual leakage rate;

[0303] -The area of ​​equipment damage caused by continuous leakage of the spontaneously combustible medium through the nth type of leakage hole; -The area of ​​equipment damage caused by continuous leakage of spontaneously combustible media from the nth type of leakage hole;

[0304] Determine the area of ​​personal injury

[0305] Constants a and b are obtained from Table 50 below (Selection of constants in the formula for calculating the area of ​​injury to persons due to combustion consequences):

[0306] Table 50

[0307]

[0308] A. Non-self-igniting medium:

[0309]

[0310] B. Potentially flammable medium:

[0311]

[0312] In the formula: -The nth type of leak hole cannot cause continuous leakage of spontaneously combustible media and the area of ​​personnel injury; -The area of ​​potential injury to personnel from continuous leakage of spontaneously combustible media through the nth type of leak hole.

[0313] 6. Determine the area of ​​equipment damage and personnel injury caused by the instantaneous leak.

[0314] instantaneous leakage equipment damage area

[0315] a. Non-self-igniting medium:

[0316]

[0317] b, Possibly spontaneously combustible medium:

[0318]

[0319] fact mit -Consequence area reduction coefficient;

[0320] a, b - medium constants;

[0321] mass n -Actual amount leaked;

[0322] -The area of ​​equipment damage caused by instantaneous leakage of spontaneously combustible media through the nth type of leakage hole;

[0323] -The area of ​​equipment damage caused by the instantaneous leakage of spontaneously combustible media from the nth type of leakage hole;

[0324] eneff n -Energy efficiency adjustment factor

[0325] Instantaneous leak of personnel injury area

[0326] a. Non-self-igniting medium:

[0327]

[0328] b, Possibly spontaneously combustible medium:

[0329]

[0330] fact mit -Consequence area reduction coefficient;

[0331] a, b - medium constants; where: mass n -Actual amount leaked; -The nth type of leak hole cannot cause instantaneous leakage of flammable media and affect the area of ​​personnel injury; - The area of ​​personnel injury caused by the instantaneous leakage of spontaneously combustible media from the nth type of leak hole; n - Energy efficiency adjustment factor.

[0332] 7. Area balance based on the consequences of leakage

[0333] Determine the balance factor

[0334]

[0335] Instantaneous leakage

[0336] For each type of leak:

[0337]

[0338]

[0339]

[0340]

[0341] In the formula: -The area of ​​equipment damage caused by the leakage of spontaneously combustible media from the nth type of leakage hole; -The nth type of leakage hole cannot leak spontaneously combustible media, thus damaging the equipment area; -The area of ​​personnel injury that may be caused by the leakage of spontaneously combustible media from the nth type of leak hole; -The nth type of leak hole is unlikely to leak flammable media, thus affecting the area of ​​personnel injury; -The area of ​​equipment damage caused by continuous leakage of the spontaneously combustible medium through the nth type of leakage hole; -The area of ​​equipment damage caused by continuous leakage of spontaneously combustible media from the nth type of leakage hole; -The nth type of leak hole cannot cause continuous leakage of spontaneously combustible media and the area of ​​personnel injury; -The area of ​​personnel injury caused by continuous leakage of spontaneously combustible media from the nth type of leak hole; -The area of ​​equipment damage caused by instantaneous leakage of spontaneously combustible media through the nth type of leakage hole; -The area of ​​equipment damage caused by the instantaneous leakage of spontaneously combustible media from the nth type of leakage hole; -The nth type of leak hole cannot cause instantaneous leakage of flammable media and affect the area of ​​personnel injury; -The area of ​​personnel injury caused by the instantaneous leakage of spontaneously combustible media from the nth type of leak hole.

[0342] 8. Consequence area balance based on auto-ignition temperature

[0343] Determine the auto-ignition temperature equilibrium factor fact AIT ;

[0344] fact AIT =0, when T s When +55.6≤AIT;

[0345] When T s +55.6>AIT>T s -55.6 hours;

[0346] fact AIT =1, when T s When -55.6 ≥ AIT;

[0347] In the formula: T s - Normal operating temperature or storage temperature (°C); AIT - Autoignition temperature (°C);

[0348] fact AIT - Spontaneous combustion temperature equilibrium factor; After obtaining the spontaneous combustion temperature equilibrium, the area of ​​equipment damage and the area of ​​personnel injury.

[0349]

[0350]

[0351] In the formula: - The area of ​​equipment damage resulting from combustion of the nth type of leakage hole; -The area of ​​personnel injury resulting from the combustion of the nth type of leak hole; -The area of ​​equipment damage caused by the leakage of spontaneously combustible media from the nth type of leakage hole; -The nth type of leakage hole cannot leak spontaneously combustible media, thus damaging the equipment area; -The area of ​​personnel injury that may be caused by the leakage of spontaneously combustible media from the nth type of leak hole; -The nth type of leak hole is unlikely to leak flammable media, thus affecting the area of ​​personnel injury;

[0352] fact AIT -Self-ignition temperature balance factor;

[0353] Finally obtained

[0354] 2. Determine the severity level of failure.

[0355] The final consequence area CA is obtained.

[0356] CA = max(CA) cmd CA inf );

[0357] The failure consequence level is obtained from Table 51 below;

[0358] Table 51 is in square meters.

[0359] Area Consequence Level Final Consequence Area (CA) Range A CA≤9.29 B 9.29 < CA ≤ 92.9 C 92.9<CA≤279 D 279<CA≤929 E CA>929

[0360] The risk level is finally determined based on the risk matrix. See [link / reference]. Figure 3 .

[0361] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention.

Claims

1. A model-data dual-driven dynamic risk early warning method for hazardous chemical equipment, characterized in that, Includes the following steps: Data collection steps: Collect historical operational data and / or simulation experiment data of hazardous chemical equipment and perform data preprocessing to obtain a sample training dataset; Pre-trained model building steps: Use the sample training dataset to pre-train the deep learning model that combines residual network and bidirectional long short-term memory network to build a deep learning pre-trained model; The first real-time data acquisition step: collect the current real-time operating data of the hazardous chemical equipment and perform data preprocessing to obtain the first real-time input data that meets the model input requirements; First prediction step: Input the first real-time input data into the deep learning pre-trained model established in the pre-trained model building step, and output the first prediction parameters; The second real-time data acquisition step: Acquire the next real-time operating data of the current hazardous chemical equipment and preprocess it to obtain the second real-time input data; Model reinforcement and fine-tuning steps: The second real-time input data is used as the actual running parameters in the first prediction step of the model, and compared with the first prediction parameters. The comparison deviation is calculated, thereby performing reinforcement learning and fine-tuning on the deep learning pre-trained model, updating the model parameters, and obtaining the deep learning pre-trained-reinforcement model. The second prediction step: Input the second real-time input data into the deep learning pre-trained-reinforcement model and output the second prediction parameters; Failure probability calculation steps: Calculate the failure probability of the current hazardous chemical equipment using the second prediction parameters output from the second prediction step; Failure consequence calculation steps: Calculate the failure consequences of the current hazardous chemical equipment using the second prediction parameters output from the second prediction step; Risk level determination steps: Based on the calculated failure probability and failure consequences, and according to the preset failure probability level and failure consequence level, the current risk level of the equipment is determined using a risk matrix. Risk warning procedure: If the current risk level reaches the preset warning threshold, the warning mechanism is triggered; Iterative steps: Repeatedly execute the second real-time data acquisition step to the risk warning step.

2. The model-data dual-driven dynamic risk early warning method for hazardous chemical equipment according to claim 1, characterized in that, The data collection step also includes: Scope preset step: Based on the type of hazardous chemical equipment, preset the range of values ​​for failure probability and the range of values ​​for consequence characterization indicators, assign a corresponding failure probability level to different ranges of the failure probability, and assign a corresponding failure consequence level to different ranges of values ​​for the consequence characterization indicators. Risk matrix construction steps: Map the failure probability level and failure consequence level to the risk matrix level by level, and match them with the corresponding risk level in the risk matrix; Parameter determination steps: Determine the parameters that need to be collected that match the type of hazardous chemical equipment, and select the appropriate sensors; The first real-time data acquisition step is as follows: using the sensor selected in the parameter determination step, the current real-time operating data of the hazardous chemical equipment is collected and preprocessed to obtain the first real-time input data that meets the model input requirements. The second real-time data acquisition step is as follows: using the sensor selected in the parameter determination step, the next real-time operating data of the current hazardous chemical equipment is acquired in real time and preprocessed to obtain the second real-time input data. The failure consequence calculation steps are as follows: using the second prediction parameters output from the second prediction step, calculate the consequence characterization index of the current hazardous chemical equipment; The steps for determining the risk level are as follows: based on the calculated failure probability and consequence characterization index, match the preset value range in the search range preset step, determine the failure probability level value and failure consequence level value, and determine the corresponding risk level in the risk matrix to obtain the current risk level.

3. The model-data dual-driven dynamic risk early warning method for hazardous chemical equipment according to claim 1, characterized in that, In the data collection step and / or the first real-time data acquisition step and / or the second real-time data acquisition step, the operating data includes at least one of wall thickness, crack depth, temperature and pressure, and the data preprocessing includes at least one of data cleaning, normalization and feature extraction. The risk warning process also includes: if the rate of change of the risk level is detected to reach a preset warning value, the warning rule is triggered.

4. The model-data dual-driven dynamic risk early warning method for hazardous chemical equipment according to claim 1, characterized in that, The failure probability calculation steps are as follows: using the second prediction parameters output from the second prediction step, and calculating the failure probability of the hazardous chemical equipment according to Formula 1; Formula 1: F=F G ×F M ×F L ×F E ; In Formula 1, F G Indicates the average failure probability of similar equipment, F M Indicates the evaluation coefficient of the management system, F L Indicates the influence coefficient of defects exceeding the standard, F E Indicates the equipment correction factor; The failure consequence calculation steps are as follows: using the second prediction parameters output by the second prediction step, and calculating the final consequence area according to Formula 2, the failure consequence of the current hazardous chemical equipment is obtained based on the final consequence area. Formula 2: ; ; ; ; In Formula 2, Indicates the final area of ​​equipment damage; Indicates the final area of ​​injury to the person; Indicates the area of ​​equipment damaged by the fire; Indicates the area of ​​personal injury; Indicates the area of ​​equipment damaged by the fire; This represents the probability of equipment failure of the same type corresponding to the nth type of leakage hole; This represents the sum of the average failure probabilities of the same type of equipment corresponding to all leakage holes; This indicates leakage in the nth type of hole.

5. The model-data dual-driven dynamic risk early warning method for hazardous chemical equipment according to claim 4, characterized in that, In the failure probability calculation step, the equipment correction factor F E It is determined by technical module factors, general condition factors, mechanical factors, and process factors. The technical module factors include the thinning sub-factor. If the type of equipment involving hazardous chemicals is container equipment, the thinning factor is obtained from Formula 3; if the type of equipment involving hazardous chemicals is pipeline equipment, the thinning factor is obtained from Formula 4; where: Formula 3: ; In Formula 3, An index indicating the severity of corrosion and thinning. This indicates the minimum wall thickness measured in the most recent inspection and testing. Indicates service time. Indicates the corrosion rate. Indicates the calculated wall thickness. This represents the corrosion allowance; where r is obtained based on the medium conditions. , ,r, , The parameters are obtained from the second prediction parameters output in the second prediction step or the inherent parameters of the hazardous chemical equipment. Formula 4: ; In Formula 4, An index indicating the severity of corrosion and thinning. The value represents the minimum wall thickness measured in the most recent inspection and testing, and t represents the nominal thickness. The value represents the service time, and r represents the corrosion rate. This represents the corrosion allowance; where r is obtained based on the medium conditions. ,t, ,r, It is obtained from the second prediction parameters output by the second prediction step or the inherent parameters of the hazardous chemical equipment.

6. The model-data dual-driven dynamic risk early warning method for hazardous chemical equipment according to claim 5, characterized in that, The technical module factors also include stress corrosion cracking sub-factor, high temperature hydrogen corrosion sub-factor, mechanical fatigue sub-factor, external damage sub-factor, and brittle fracture sub-factor; General condition factors include engineering condition sub-factors, cold climate operation sub-factors, and seismic activity sub-factors; Mechanical factors include sub-factors of structural complexity, construction specifications, life cycle, safety factor, and vibration monitoring; among which, the sub-factors of structural complexity include sub-factors of equipment complexity, pipeline complexity, and structural rationality; and the sub-factors of safety factor include sub-factors of operating pressure and operating temperature. Process factors include process continuity sub-factors, process stability sub-factors, and safety valve status sub-factors. The process continuity sub-factor includes planned shutdown sub-factor and unplanned shutdown sub-factor; the safety valve status sub-factor includes maintenance procedure sub-factor, fouling status sub-factor, corrosion status sub-factor, and cleanliness status sub-factor.

7. A model-data dual-driven dynamic risk early warning system for hazardous chemical equipment, characterized in that, include: The data collection module is used to perform the data acquisition steps: collect historical operational data and / or simulation experimental data of hazardous chemical equipment and perform data preprocessing to obtain a sample training dataset; The pre-trained model building module is used to perform the pre-trained model building steps: pre-training a deep learning model that combines residual networks and bidirectional long short-term memory networks using a sample training dataset to build a deep learning pre-trained model; The first real-time data acquisition module is used to perform the first real-time data acquisition step: real-time acquisition of the current real-time operating data of the hazardous chemical equipment and data preprocessing to obtain the first real-time input data that meets the model input requirements; The first prediction module is used to perform the first prediction step: inputting the first real-time input data into the deep learning pre-trained model established in the pre-trained model building step, and outputting the first prediction parameters; The second real-time data acquisition module is used to perform the second real-time data acquisition step: real-time acquisition of the next real-time operating data of the current hazardous chemical equipment and preprocessing it to obtain the second real-time input data; The model enhancement and fine-tuning module is used to perform the model enhancement and fine-tuning steps: the second real-time input data is used as the actual running parameters in the first prediction step of the model, and compared with the first prediction parameters. The comparison deviation is calculated, thereby performing reinforcement learning and fine-tuning on the deep learning pre-trained model, updating the model parameters, and obtaining the deep learning pre-trained-enhanced model. The second prediction module is used to perform the second prediction step: inputting the second real-time input data into the deep learning pre-trained-reinforcement model and outputting the second prediction parameters; The failure probability calculation module is used to perform the failure probability calculation step: using the second prediction parameters output from the second prediction step, the failure probability of the current hazardous chemical equipment is calculated. The failure consequence calculation module is used to perform the failure consequence calculation steps: using the second prediction parameters output by the second prediction step, the failure consequences of the current hazardous chemical equipment are calculated. The risk level determination module is used to perform the risk level determination steps: based on the calculated failure probability and failure consequences, and according to the preset failure probability level and failure consequence level, the current risk level of the equipment is determined using a risk matrix. The risk warning module is used to execute risk warning steps: if the current risk level reaches the preset warning threshold, the warning mechanism is triggered; The loop iteration module is used to execute loop iteration steps: looping through the second real-time data acquisition step to the risk warning step.

8. The model-data dual-drive dynamic risk early warning system for hazardous chemical equipment according to claim 7, characterized in that, Also includes: The range preset module is used to perform the range preset step: according to the type of hazardous chemical equipment, preset the value range of failure probability and the value range of consequence characterization index, assign a matching failure probability level to different value ranges of the failure probability, and assign a matching failure consequence level to different value ranges of the consequence characterization index. The risk matrix construction module is used to perform the risk matrix construction steps: mapping the failure probability level and failure consequence level to the risk matrix level by level, and matching them with the corresponding risk level in the risk matrix. The parameter determination module is used to perform the parameter determination steps: determine the parameters that need to be collected that match the type of hazardous chemical equipment, and select the appropriate sensors.

9. An electronic device, characterized in that, include: One or more processors; Memory used to store executable instructions; The one or more processors are configured to invoke executable instructions stored in the memory to execute the method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, It stores computer program instructions, which, when executed by a processor, implement the method described in any one of claims 1 to 6.

Images