CN-121977754-A - Unattended intelligent detection method and detection system for dangerous chemical gas leakage

CN121977754ACN 121977754 ACN121977754 ACN 121977754ACN-121977754-A

Abstract

The invention discloses an unmanned intelligent detection method and system for dangerous chemical gas leakage, and belongs to the technical field of artificial intelligence and safety monitoring. The method comprises the steps of constructing an environmental digital twin body fused with multi-mode sensing data and facility three-dimensional information, achieving dynamic reproduction of panoramic environmental states of an operation area, utilizing a space-time diagram neural network model to learn and reconstruct twin body data, achieving sensitive detection of early leakage anomalies based on reconstruction errors, inverting the position and strength of a leakage source by adopting a Bayesian inference framework once leakage is confirmed, driving a computational fluid dynamics model coupled with a real-time wind field to predict gas diffusion situation, and finally automatically generating and cooperatively executing emergency response instructions of equipment shutdown, personnel evacuation and area isolation. The invention realizes the intelligent closed loop of the whole flow from sensing, diagnosis and prediction to disposal, and remarkably improves the timeliness and accuracy of monitoring and early warning of the gas leakage of the hazardous chemicals and the initiative and scientificity of emergency disposal.

Inventors

Dong Rongtu
ZHANG XUEFENG
FENG LEI
CAI XIYANG
ZHOU XUAN
LIU DEFA
CHEN DONGBIAO

Assignees

原树科创(宁波)科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260122

Claims (9)

1. The unattended intelligent detection method for the gas leakage of the dangerous chemicals is characterized by comprising the following steps of: constructing an environmental digital twin body covering a dangerous chemical operation area, synchronously collecting multi-mode sensing data in the operation area, fusing the multi-mode sensing data with three-dimensional geometric information of facilities, and generating and dynamically updating a continuous four-dimensional space-time data field representing the global environmental state; Performing leakage anomaly detection based on the environmental digital twin body, reconstructing the four-dimensional space-time data field obtained in real time by using a trained space-time diagram neural network model, and continuously judging early anomalies of gas leakage by calculating reconstruction errors and comparing the reconstruction errors with a dynamically adjusted judgment threshold value; performing inversion positioning of a leakage source based on an abnormal detection result, responding to continuously judging a leakage event, and solving the spatial position and the leakage intensity of the leakage source by reverse calculation based on the concentration field and the wind field data in the four-dimensional space-time data field at the current moment; The leakage source parameters and the real-time environment wind field obtained by reverse inversion are used as inputs, a computational fluid dynamics model is driven to carry out simulation calculation, the diffusion track and concentration distribution of leakage gas in a period of time in the future are predicted, and a dynamic risk map is generated; and fusing the dynamic risk map, the static equipment asset information and the real-time personnel positioning information, and generating and issuing control instructions for process equipment, safety evacuation guidance for personnel and isolation regulation and control instructions for physical environment through space analysis and logic reasoning.
2. The unattended intelligent detection method of the hazardous chemical substance gas leakage according to claim 1, wherein the construction of the environmental digital twin body covering the hazardous chemical substance operation area specifically comprises the following steps: Generating a three-dimensional spatial grid covering the work area based on the facility three-dimensional model; Mapping multi-modal sensing data to corresponding nodes of the three-dimensional space grid, wherein the multi-modal sensing data at least comprises target gas concentration, three-dimensional wind speed vector, ambient temperature and humidity; For nodes without direct coverage of sensing data in the three-dimensional space grid, estimating the environmental parameter values of the nodes and quantitatively estimating uncertainty; The four-dimensional spatiotemporal data field is dynamically generated and updated by repeatedly performing data mapping and spatial interpolation on successive time slices.
3. The unattended intelligent detection method for the leakage of the hazardous chemical gas according to claim 1 or 2, wherein the detection of the leakage abnormality based on the environmental digital twin body specifically comprises the following steps: Extracting space-time tensors from the four-dimensional space-time data field as input, and reconstructing by utilizing a pre-trained space-time diagram convolution self-encoder model, wherein the diagram structure of the space-time diagram convolution self-encoder is constructed based on the geographic distance of space nodes and the historical relativity of environmental parameters; calculating a reconstruction error of the input space-time tensor and the model reconstruction output; Determining a reference threshold value based on the historical normal working condition data, and calculating a dynamic judgment threshold value by combining a real-time environment background noise factor; and when the reconstruction error continuously exceeds the dynamic judgment threshold value for a preset time, judging that a gas leakage event occurs.
4. The unattended intelligent detection method for the leakage of the hazardous chemical substance gas according to claim 1 is characterized by carrying out inversion positioning of a leakage source based on an abnormal detection result, and specifically comprising the following steps: Defining an unknown parameter vector containing leak source coordinates and leak rates; establishing a likelihood function which takes a Gaussian smoke mass model as a forward operator and fuses the measurement error of a sensor and the uncertainty of spatial interpolation; Setting prior uniform distribution of leakage source coordinates based on a space range of potential leakage equipment, and setting log-normal prior distribution of leakage rate based on process medium characteristics; And carrying out sampling solution from the posterior distribution by adopting a Markov chain Monte Carlo sampling algorithm, and calculating the optimal estimated value of the leakage source parameter and the confidence interval thereof based on a sampling result.
5. The unattended intelligent detection method for the gas leakage of the dangerous chemicals according to claim 1, wherein the gas diffusion prediction is performed based on a leakage source inversion result, specifically comprising: Setting a leakage source coordinate and leakage intensity obtained by inversion as a simulation quality source item by taking a three-dimensional calculation grid corresponding to the environmental digital twin body as a calculation domain; Taking real-time three-dimensional wind field data provided by the four-dimensional space-time data field as a simulated transient inlet boundary condition, and taking the current full-field environment state as a simulated initial field; carrying out numerical simulation on the convection diffusion process of the leaked gas under the action of the real-time wind field; and extracting a time sequence concentration field output by simulation, and generating a dynamic dangerous area evolution diagram and a gas cloud cluster movement track in a future period by combining a preset hazard concentration threshold.
6. The unattended intelligent detection method of the hazardous chemical gas leakage according to claim 1, wherein the emergency response strategy is generated and executed based on a diffusion prediction result, and specifically comprises the following steps: Performing three-dimensional space superposition analysis on the predicted dangerous area in the dynamic risk map and a factory equipment asset geographic information system, identifying the key equipment which is threatened, and generating an ordered equipment turn-off instruction sequence according to a process safety logic knowledge base; The predicted dangerous area is used as a dynamic obstacle to be merged into a three-dimensional personnel navigation network, and an optimal evacuation path for avoiding risks is planned and pushed for each personnel based on the positions of the personnel in real time; And generating and issuing an entrance and exit isolation instruction for the dangerous area and a directional regulation instruction for the ventilation system according to the property of the leaked gas and the real-time wind field.
7. The unattended intelligent detection method for dangerous chemical gas leakage according to claim 2, wherein the multi-mode sensing data is collected through a heterogeneous sensor network deployed in an operation area, the heterogeneous sensor network comprises: the system comprises a tunable diode laser absorption spectrum sensor for fixed-point concentration monitoring, a mobile inspection sensor carried on an autonomous mobile platform, a three-dimensional ultrasonic anemometer array for constructing a refined wind field model, and hyperspectral imaging equipment and infrared thermal imaging equipment for identifying the form of a gas cloud cluster.
8. The unattended intelligent detection method for the dangerous chemical gas leakage according to claim 3, wherein the training process of the space-time diagram convolution self-encoder model comprises the following steps: constructing a training sample by using the four-dimensional space-time data field under the historical normal working condition; taking a graph structure constructed based on the correlation of the spatial distance and the parameter as an input; extracting and compressing space-time characteristics through an encoder to obtain potential vectors, and reconstructing through a decoder; Unsupervised training is aimed at minimizing the mean square error between the input and the reconstructed output.
9. An unattended intelligent detection system for dangerous chemical gas leakage, which is used for realizing the unattended intelligent detection method for dangerous chemical gas leakage according to any one of claims 1 to 8, and is characterized by comprising the following steps: the multi-mode sensing data acquisition module is used for synchronously acquiring multi-mode sensing data in the operation area; the environment digital twin construction module is used for fusing the multi-mode sensing data with the three-dimensional geometrical information of the facility, and constructing and dynamically updating an environment digital twin body, namely a four-dimensional space-time data field; The leakage anomaly detection module is loaded with a trained space-time diagram neural network model and is used for carrying out real-time anomaly detection and leakage event judgment based on the environmental digital twin; the leakage source inversion positioning module is integrated with a Bayesian inference engine and a sampler and is used for carrying out inversion positioning on the leakage source after judging a leakage event; The diffusion prediction and risk assessment module is integrated with a computational fluid dynamics solver and is used for carrying out gas diffusion simulation based on inversion results and generating a dynamic risk map; the emergency response strategy generation module is used for fusing the dynamic risk map, the static asset information and the dynamic personnel position, generating and cooperatively issuing an emergency response instruction; all modules are connected through communication interfaces to form a closed-loop intelligent detection and response system.

Description

Unattended intelligent detection method and detection system for dangerous chemical gas leakage Technical Field The invention belongs to the field of crossing of artificial intelligence and safety monitoring technology, and particularly relates to an unmanned intelligent detection method and system for dangerous chemical gas leakage. Background In the fields of chemical industry, energy and the like, which relate to dangerous chemicals, the real-time monitoring and early warning of gas leakage are key links for guaranteeing safety. The current mainstream technology relies on a fixed sensor to alarm based on a concentration threshold value or adopts a manual inspection mode. The method has obvious defects that the fixed sensor is easy to be interfered by the environment to generate false alarm and report missing, the distributed points are fixed, the dynamic diffusion area is difficult to be covered on the whole surface, and the manual inspection is low in efficiency, high in risk and incapable of continuous monitoring. The existing improvement scheme is tried to integrate multiple sensors, but has the limitation that a static data fusion strategy is adopted by the system, early weak leakage signals are difficult to accurately capture, the algorithm usually ignores the physical rule of gas diffusion, so that the prediction reliability is insufficient in a complex scene, and in addition, monitoring, early warning and treatment links are usually isolated from each other and cannot form an intelligent closed loop from detecting response. Therefore, there is an urgent need for an intelligent system and method that can achieve early accurate detection, leak source localization, diffusion prediction, and automatically trigger targeted emergency responses. Disclosure of Invention The invention aims to solve the technical problems of single detection mode, dependence on a fixed threshold, monitoring blind area, response lag and incapability of situation deduction and risk prejudgment of leakage events in the prior art. In order to solve the technical problems, the invention provides an unattended intelligent detection method and a detection system for dangerous chemical gas leakage, which are used for constructing a high-fidelity environment digital twin body fused with multi-mode sensing data and three-dimensional geometrical information of facilities, utilizing a space-time diagram neural network to deeply mine complex space-time correlation of environment parameters under normal working conditions, and realizing accurate capture of early weak abnormal signals of leakage based on reconstruction errors. And once the abnormality is detected, immediately starting a reverse calculation model based on Bayesian inference, carrying out accurate positioning and leakage intensity inversion on a leakage source, driving a real-time calculation fluid dynamic model to carry out high-precision prediction on a future diffusion track and an influence range of leakage gas, and finally generating a dynamic and visual risk assessment and emergency response instruction, thereby realizing the fundamental transition from passive alarm to active early warning, from single-point monitoring to global perception, from phenomenon alarm to root positioning and situation deduction. According to one aspect of the invention, an unattended intelligent detection method for dangerous chemical gas leakage is provided, which comprises the following steps: constructing an environmental digital twin body covering a dangerous chemical operation area, synchronously collecting multi-mode sensing data in the operation area, fusing the multi-mode sensing data with three-dimensional geometric information of facilities, and generating and dynamically updating a continuous four-dimensional space-time data field representing the global environmental state; Performing leakage anomaly detection based on the environmental digital twin body, reconstructing the four-dimensional space-time data field obtained in real time by using a trained space-time diagram neural network model, and continuously judging early anomalies of gas leakage by calculating reconstruction errors and comparing the reconstruction errors with a dynamically adjusted judgment threshold value; performing inversion positioning of a leakage source based on an abnormal detection result, responding to continuously judging a leakage event, and solving the spatial position and the leakage intensity of the leakage source by reverse calculation based on the concentration field and the wind field data in the four-dimensional space-time data field at the current moment; The leakage source parameters and the real-time environment wind field obtained by reverse inversion are used as inputs, a computational fluid dynamics model is driven to carry out simulation calculation, the diffusion track and concentration distribution of leakage gas in a period of time in the future are predicted, and a dynamic risk ma