IoT-based Gas Leakage Monitoring Method, Device, Equipment and Storage Medium

Abstract

The present invention relates to the technical field of the Internet of Things, and discloses a gas leakage monitoring method, device, equipment and storage medium based on the Internet of Things. Among them, the method includes: collecting data from a gas storage device to obtain an initial multi-dimensional data set; performing filtering and edge detection processing to obtain potential abnormal point data; inputting the potential abnormal point data into a hybrid deep learning model for gas leakage detection to obtain initial leakage source position data and leakage intensity values; performing precise leakage positioning with multi-sensor collaboration to obtain target leakage source position data; performing gas diffusion simulation to obtain gas concentration distribution data and risk indices; performing dynamic reconstruction of the sensor network based on the gas concentration distribution data and risk indices, and outputting an emergency response strategy and sensor network configuration information. This method realizes the full-process management from leakage detection, source location to risk assessment and emergency response.

Description

Technical Field

[0001] The present invention relates to the field of the Internet of Things, and in particular to a gas leakage monitoring method, device, equipment and storage medium based on the Internet of Things. Background Art

[0002] Traditional gas leakage monitoring methods mainly rely on regular inspections and fixed gas detectors. Such methods have problems such as limited monitoring coverage, poor real-time performance, and inability to accurately locate the leakage source.

[0003] In recent years, the rapid development of Internet of Things technology has provided new solutions for gas leakage monitoring. By deploying a large number of low-cost and low-power gas sensors, it is possible to achieve all-round and real-time monitoring of the environment around the gas storage device. However, how to effectively process and analyze massive sensor data and accurately identify potential leakage events is still a huge challenge. In addition, traditional gas diffusion models are often too simplified and cannot accurately reflect the gas diffusion behavior in complex environments, resulting in insufficient accuracy of risk assessment. On the other hand, most existing gas leakage monitoring systems adopt static sensor deployment schemes and lack the ability to dynamically adjust according to the actual situation. This not only leads to waste of resources but also cannot quickly respond to emergencies. Summary of the Invention

[0004] The present invention provides a gas leakage monitoring method, device, equipment and storage medium based on the Internet of Things, which is used to realize the full-process management from leakage detection, source location to risk assessment and emergency response.

[0005] In a first aspect, the present invention provides a gas leakage monitoring method based on the Internet of Things. The gas leakage monitoring method based on the Internet of Things includes:

[0006] Collect data from the gas storage device to obtain an initial multi-dimensional data set;

[0007] Perform filtering and edge detection processing on the initial multi-dimensional data set to obtain potential abnormal point data;

[0008] Input the potential abnormal point data into a hybrid deep learning model for gas leakage detection to obtain initial leakage source location data and leakage intensity values;

[0009] Perform precise leakage location with multi-sensor collaboration according to the initial leakage source location data to obtain target leakage source location data;

[0010] Input the target leakage source location data and the leakage intensity values into a three-dimensional computational fluid dynamics model for gas diffusion simulation to obtain gas concentration distribution data and risk indices;

[0011] Perform dynamic reconstruction of the sensor network based on the gas concentration distribution data and the risk index, and output an emergency response strategy and sensor network configuration information.

[0012] In a second aspect, the present invention provides an Internet of Things-based gas leakage monitoring device, and the Internet of Things-based gas leakage monitoring device includes:

[0013] An acquisition module, configured to collect data from the gas storage device to obtain an initial multi-dimensional data set;

[0014] A processing module, configured to perform filtering and edge detection processing on the initial multi-dimensional data set to obtain potential anomaly point data;

[0015] A detection module, configured to input the potential anomaly point data into a hybrid deep learning model for gas leakage detection to obtain initial leakage source location data and leakage intensity values;

[0016] A positioning module, configured to perform precise leakage positioning with multi-sensor collaboration according to the initial leakage source location data to obtain target leakage source location data;

[0017] A simulation module, configured to input the target leakage source location data and the leakage intensity values into a three-dimensional computational fluid dynamics model for gas diffusion simulation to obtain gas concentration distribution data and a risk index;

[0018] An output module, configured to perform dynamic reconstruction of the sensor network based on the gas concentration distribution data and the risk index, and output an emergency response strategy and sensor network configuration information.

[0019] In a third aspect of the present invention, there is provided an Internet of Things-based gas leakage monitoring device, including: a memory and at least one processor, and instructions are stored in the memory; the at least one processor calls the instructions in the memory so that the Internet of Things-based gas leakage monitoring device executes the above-mentioned Internet of Things-based gas leakage monitoring method.

[0020] In a fourth aspect of the present invention, there is provided a computer-readable storage medium, and instructions are stored in the computer-readable storage medium, and when it runs on a computer, it causes the computer to execute the above-mentioned Internet of Things-based gas leakage monitoring method.

[0021] In the technical solution provided by the present invention, by performing filtering and edge detection processing on the initial multi-dimensional data set and combining a hybrid deep learning model of 3D-CNN and LSTM, potential abnormal points can be effectively identified, the false alarm rate can be reduced, and the accuracy of leakage detection can be improved. The multi-sensor collaborative positioning is carried out by using the triangulation principle and the particle filter algorithm, which overcomes the problem of low positioning accuracy of traditional methods in complex environments and can quickly and accurately determine the location of the leakage source. The three-dimensional computational fluid dynamics model is used for gas diffusion simulation, considering the influence of factors such as terrain, obstacles, and meteorological conditions, and can more accurately predict the gas concentration distribution and risk areas. Based on the gas concentration distribution and risk index, the sensor network is dynamically reconfigured to optimize the sensor layout and working parameters, improving the efficiency and coverage of the monitoring system. By integrating the supplementary sensor deployment plan, sensor working parameter configuration, and emergency response plan, a comprehensive and operable emergency response strategy is provided for decision-makers, improving the handling ability of emergencies. The present invention constructs a closed-loop intelligent gas leakage monitoring system from data collection, abnormal detection, leakage location to risk assessment and emergency response, improving the overall supervision efficiency and safety.

[0022] Other features and advantages of the present invention will be described in the following specification, and in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention are achieved and obtained by the structures specifically pointed out in the specification, claims, and drawings.

[0023] To make the above objectives, features, and advantages of the present invention more obvious and understandable, the following specific preferred embodiments are given, and in conjunction with the accompanying drawings, the detailed description is as follows. Brief Description of the Drawings

[0024] Figure 1 It is a schematic diagram of an embodiment of the gas leakage monitoring method based on the Internet of Things in an embodiment of the present invention;

[0025] Figure 2 It is a schematic diagram of an embodiment of the gas leakage monitoring device based on the Internet of Things in an embodiment of the present invention;

[0026] Figure 3 It is a schematic diagram of an embodiment of the gas leakage monitoring equipment based on the Internet of Things in an embodiment of the present invention. Detailed Embodiments

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0028] As used in the embodiments of the present invention, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes other unlisted steps or units, or optionally further includes other steps or units inherent to these processes, methods, products, or devices.

[0029] To facilitate the understanding of this embodiment, first, a gas leakage monitoring method based on the Internet of Things disclosed in the embodiments of the present invention will be introduced in detail. As Figure 1 shown, this method includes the following steps:

[0030] 101. Collect data from the gas storage device to obtain an initial multi-dimensional data set;

[0031] It can be understood that the execution subject of the present invention can be a gas leakage monitoring device based on the Internet of Things, or a terminal or a server. Specifically, it is not limited here. In the embodiments of the present invention, the server is taken as an example of the execution subject for illustration.

[0032] Specifically, grid division is carried out on key positions around the gas storage device. By spatially dividing the monitoring area, a preset monitoring grid is formed. The grid division takes into account the structural characteristics of the gas storage device, gas dynamics, and potential leakage risk areas. Internet of Things (IoT) gas sensors are strategically deployed within each preset monitoring grid to construct an initial sensor network topology. The deployment of sensors should not only cover all key positions but also be encrypted in potentially high-risk areas to ensure that once a gas leak occurs, the initial signal of the leak can be quickly and accurately captured. Set the data acquisition frequency for each IoT gas sensor. The frequency setting should comprehensively consider factors such as the speed of environmental change, the characteristics of gas diffusion, and the power consumption of the sensor. Through reasonable frequency setting, a sampling time series is obtained, enabling each sensor to collect data at a predetermined time interval. Based on this sampling time series, each sensor collects gas concentration data to obtain the original gas concentration data. At the same time, environmental parameter sensors arranged around the gas storage device are used to synchronously collect environmental parameter data such as temperature, humidity, and air pressure. Align the time stamps of the original gas concentration data and the environmental parameter data to unify the asynchronous data from different sensors on the same time basis, ensuring that the data collected by all sensors can be correlated and analyzed at the same time point. After time stamp alignment, aligned multi-dimensional data containing gas concentration and environmental parameters is obtained. Perform noise removal processing on the aligned multi-dimensional data. For example, noise reduction methods based on wavelet transform, Kalman filtering, local regression, etc. are used to filter out irregular mutations and interference signals in the multi-dimensional data and retain the true signal characteristics to obtain noise-reduced multi-dimensional data. Correlate the noise-reduced multi-dimensional data with the position information of the sensors to obtain an initial multi-dimensional data set containing gas concentration, environmental parameters, and position information.

[0033] 102. Filter and perform edge detection on the initial multi-dimensional data set to obtain potential abnormal point data;

[0034] Specifically, wavelet transform is applied to the initial multi-dimensional dataset for multi-scale decomposition. Wavelet transform is an effective time-frequency analysis method that can decompose the original signal into sub-datasets of different frequency components, and each sub-dataset corresponds to the signal characteristics at a specific scale. In this way, the low-frequency trend component and high-frequency fluctuation component in the gas concentration signal are effectively separated. Based on the sub-datasets of different frequency components, the local variance of each sensor data is calculated. The local variance is an important indicator reflecting the fluctuation characteristics of the data and can reveal the degree of change intensity of the sensor within a specific time window. By calculating the local variance, the local fluctuation characteristics of each sensor at different time periods are obtained. The local fluctuation characteristics are subjected to adaptive threshold analysis to establish a dynamic threshold curve for the local fluctuation of each sensor. The generation of the dynamic threshold curve is based on the historical data and current fluctuation conditions of each sensor. The Gaussian filter is applied to the initial multi-dimensional dataset for smoothing. The Gaussian filter can effectively smooth the data, remove random noise, and retain the main features of the signal. Through smoothing, the smoothed dataset is obtained. Based on the smoothed dataset, edge detection operations are performed. Multiple methods are used for edge detection, such as the Sobel operator, Canny operator, etc., to identify the mutation points in the data, and these mutation points are usually the places where the signal characteristics change significantly. On this basis, an edge feature map is obtained, which reflects the possible abnormal positions in the dataset. The dynamic threshold curve is fused with the edge feature map, comprehensively considering the historical background of sensor fluctuations and the current edge change situation, to generate a preliminary set of abnormal points, including all abnormal spatio-temporal points. The preliminary set of abnormal points is subjected to abnormal region clustering. Clustering algorithms such as K-means and DBSCAN are used to aggregate adjacent abnormal points into individual abnormal regions, effectively removing isolated abnormal points, and obtaining more representative clustered abnormal regions. Each abnormal region represents a region with significant abnormal fluctuations in a period of time and space and has a higher probability of abnormal occurrence. The coordinates of the center point, the detection time, and the numbers of associated sensors are extracted from the clustered abnormal regions. The center point coordinates reflect the geographical location of the abnormal region, the detection time indicates the specific time period when the abnormality occurs, and the associated sensor numbers indicate which sensors contribute to the abnormal region. These information is summarized to obtain potential abnormal point data.

[0035] 103. Input the potential abnormal point data into a hybrid deep learning model for gas leakage detection to obtain the initial leakage source location data and leakage intensity values;

[0036] Specifically, the potential anomaly point data is reconstructed in the spatio-temporal sequence. The potential anomaly point data is arranged in chronological order and combined with its spatial location information to be organized into a three-dimensional spatio-temporal data cube. The three dimensions of the data cube respectively represent characteristic dimensions such as time, space, and gas concentration. Through data reconstruction, the time series information and the spatial distribution information are combined. The three-dimensional spatio-temporal data cube is input into the 3D-CNN module of the hybrid deep learning model. This 3D-CNN module contains four three-dimensional convolutional layers. Each convolutional layer uses the ReLU activation function and batch normalization operation to extract the spatial features in the data layer by layer. Through the 3D convolution operation, the model captures the local patterns in the three-dimensional space, such as the distribution law of gas concentration in space and the trend of abnormal changes. The ReLU activation function introduces non-linear features, enabling the model to learn more complex feature relationships. Batch normalization helps to accelerate the training process of the model and improve the generalization ability of the model. After the convolution operation, a mapping containing spatial features, that is, the spatial feature mapping, is obtained. The 3D max pooling operation is applied to the spatial feature mapping for dimensionality reduction. Max pooling can significantly reduce the dimension of the data while retaining the main feature information, reducing the computational complexity, and preventing the model from overfitting. The dimension-reduced feature map is used as the input for feature extraction and passed to the feature extraction network in the hybrid deep learning model, which contains three 3D convolutional layers. Each convolutional layer of the feature extraction network uses the LeakyReLU activation function. Compared with the ordinary ReLU, LeakyReLU retains a part of the activation value on the negative half-axis, thus solving the "dead neuron" problem caused by ReLU and enhancing the feature extraction ability of the model. After being processed by the deep network, deep spatial features are obtained. The global average pooling operation is performed on the deep spatial features to compress the deep feature mapping into a feature vector with a fixed dimension, eliminating the uncertainty of spatial information. The feature vector obtained through global average pooling is reshaped into a time series and input into the recurrent neural network in the hybrid deep learning model, which contains two layers of LSTM units. Each layer of LSTM units contains 128 hidden states. Through the memory and forgetting mechanisms of LSTM, the temporal features in the gas leakage process can be captured. The temporal feature representation contains the time-dependent relationship of the leakage event and can effectively predict the timing of gas leakage. The temporal feature representation is input into the fully connected layer and the Softmax activation function is used to obtain the leakage probability distribution. The Softmax activation function can convert the temporal features into a probability distribution of multiple categories, characterizing the possibility of gas leakage. Based on the leakage probability distribution, a threshold judgment is performed to identify the time point of leakage, indicating the initial occurrence time of gas leakage. The feature vector corresponding to the time point of leakage occurrence is input into the regression network in the hybrid deep learning model. The regression network contains three fully connected layers, and each layer uses the ReLU activation function. Through the regression operation, the initial leakage source location data is calculated.The ReLU activation function can effectively learn non - linear mapping relationships, enabling the model to more accurately predict the location of the leakage source. Applying a fully - connected layer with residual connections to the initial leakage source location data for processing, the design of the residual connection effectively alleviates the vanishing gradient problem in deep networks and retains more feature information. The Sigmoid activation function is used to obtain the leakage intensity value. The Sigmoid activation function maps the output value between 0 and 1, making the leakage intensity value interpreted as a normalized representation of the leakage degree, and finally obtaining the location data of the initial leakage source and the leakage intensity value.

[0037] 104. Perform precise leakage localization with multi - sensor collaboration based on the initial leakage source location data to obtain the target leakage source location data;

[0038] Specifically, the initial leakage source location data is used to select the N nearest gas sensors around it, forming a candidate sensor set. The selection process of the candidate sensor set is based on the triangulation principle. By calculating the geometric relationships between multiple sensors and the initial leakage source location, the sensors most likely to detect the leakage event are determined. These candidate sensors are located around the leakage source, and their distribution is as wide as possible to cover the entire possible leakage area, so as to better capture the information characteristics of the leakage source. Signal intensity measurements are performed on each sensor in the candidate sensor set to obtain a signal intensity data set, which contains the intensity information of indicators such as gas concentration or pressure monitored by each sensor. The change in signal intensity reflects the characteristics of the gas leakage source. Generally, the closer the leakage source location is to the sensor, the stronger the signal intensity. According to the signal intensity data set and the preset signal propagation model, the distance from each sensor to the leakage source is calculated to obtain a distance measurement set. The signal propagation model describes the process of gas diffusion from the leakage source to the surrounding sensors, usually considering the attenuation law of gas under different environmental conditions. Through this model, the signal intensity is converted into a measured value of distance. The distance measurement set contains the estimated distance from each sensor to the leakage source. Based on the distance measurement set, the nonlinear state equation and the observation equation of the particle filter algorithm are constructed to obtain a particle filter model. Particle filter is a tool for processing the state estimation of nonlinear and non-Gaussian distribution systems, used to accurately locate the leakage source position. The nonlinear state equation describes the dynamic change of the leakage source position, while the observation equation associates the sensor signal intensity with the leakage source position to form an observation model of the leakage source position. Through the combination of these two equations, the particle filter model can gradually converge to the true leakage source position based on the continuously updated sensor data. The particle filter model is initialized to generate M particles, each particle representing a possible leakage source position, obtaining an initial particle set. Each particle in the initial particle set is evenly distributed in space, covering all possible leakage source position areas. The nonlinear state equation is used to perform state prediction on the initial particle set to generate a predicted particle set. The process of state prediction is to update the position of the initial particles according to the model. This process takes into account the movement of the leakage source or the change of environmental conditions, enabling the particle set to better track the dynamic change of the leakage source. The likelihood probability of each particle in the predicted particle set is calculated based on the observation equation. The likelihood probability reflects the possibility of each particle being the true leakage source position. By matching the predicted particles with the sensor observation data, their compliance is evaluated to obtain the likelihood value of each particle. Importance resampling is performed to eliminate the particles with lower likelihood values and increase the number of particles with higher likelihood values, so that the particle set can be more concentrated around the true leakage source. Weighted average processing is performed on the updated particle set. The weight of each particle is proportional to its likelihood probability. Through weighted average, a more accurate leakage source position estimate is obtained, obtaining the target leakage source position data.

[0039] 105. Input the target leakage source location data and leakage intensity value into a three-dimensional computational fluid dynamics model to conduct gas diffusion simulation, and obtain gas concentration distribution data and risk indices;

[0040] Specifically, the target leakage source location data and leakage intensity values are input into a three-dimensional computational fluid dynamics model. The three-dimensional computational domain is constructed using the target leakage source location data. This computational domain is determined based on the specific location of the leakage source in space and covers the entire potential diffusion area around the leakage source. The three-dimensional computational domain is meshed to generate computational grids. Finer grids are used around the leakage source to more accurately describe the gas flow and concentration changes near the leakage source, while relatively coarser grids are adopted in areas far from the leakage source to save computational resources. The leakage intensity values are set as the boundary conditions at the leakage source location in the computational grids, defined as the initial boundary conditions, to ensure that at the beginning of the simulation, the location and leakage rate of the gas leakage source are accurately described in the model. According to the topography and obstacle distribution around the gas storage device, the solid boundary conditions within the computational domain are set. The solid boundary conditions include the undulations of the terrain, the distribution of buildings, the obstruction of vegetation, etc. By defining the solid boundaries, a complete computational boundary is formed to ensure that the gas diffusion within the computational domain can truly reflect the blocking and guiding effects of the actual environment. At the same time, the inlet boundary conditions of the computational domain are set based on the current meteorological data. The meteorological data includes information such as wind speed, wind direction, and temperature distribution. The wind speed and wind direction determine the main propagation direction and speed of the gas, while the temperature affects the density and diffusion rate of the gas. By setting the environmental boundary conditions, the diffusion behavior of the gas under actual meteorological conditions is simulated, improving the accuracy and reliability of the simulation. The SIMPLE algorithm is used to solve the Navier-Stokes equations and the mass transport equation. The Navier-Stokes equations describe the motion laws of the gas in the fluid, including the momentum equation and the continuity equation, while the mass transport equation is used to describe the change in gas concentration. Solving the equations can obtain the velocity field and pressure field within the entire computational domain, which are the key variables for describing the gas diffusion process. Based on the velocity field and pressure field, the finite volume method is used to discretize the diffusion equation, converting the continuous partial differential equation into a discretized system of equations. By dividing the computational domain into several volume elements and applying the conservation law within each element, the discretized form of the mass transport equation is obtained. The TDMA algorithm is used to solve the discretized system of equations. The TDMA algorithm is an effective method for solving linear systems of equations and can quickly solve the discretized system of equations generated by the finite volume method. Through the solution of this algorithm, the gas concentration distribution data within the entire computational domain is obtained. The gas concentration distribution data reflects the distribution of the leaked gas in space, that is, the gas concentration values at different positions and heights. According to the gas concentration distribution data and the preset exposure limit values, the hazard index of each grid point is calculated. The exposure limit values are set according to safety standards and indicate that above a certain concentration value, it may cause harm to humans or the environment. By comparing the concentration data with the exposure limit values, the potential hazard degree of each grid point is quantified. Spatial interpolation operations are performed on the hazard index to generate the risk index for the entire computational domain.The risk index is a comprehensive indicator that takes into account the spatial variation of gas concentration distribution and the impact of exposure limits on health and safety, and represents the potential threat level of a leakage event to the entire area.

[0041] 106. Based on the gas concentration distribution data and the risk index, perform dynamic reconfiguration of the sensor network, and output the emergency response strategy and the sensor network configuration information.

[0042] Specifically, perform gradient analysis on the gas concentration distribution data. By calculating the rate of change of gas concentration in space, identify high-concentration regions and regions where the rate of change of concentration exceeds a preset target value, and mark them as key monitoring regions. Within the key monitoring regions, the drastic change in gas concentration indicates the activity of the leakage source or the further spread of the leakage. Conduct more intensive monitoring and real-time data collection in these regions so that rapid response and effective emergency measures can be taken when a danger occurs. Divide the monitoring regions into risk levels according to the risk index. The risk index is a comprehensive indicator calculated based on the gas concentration distribution data and the exposure limit, which can quantitatively describe the degree of potential threat received by different regions. By classifying the risk index, a classified risk map is obtained. The risk levels from low to high represent different degrees of hazards and the emergency measures to be taken. On the classified risk map, high-risk regions need to be focused on and monitored in real time, while the monitoring frequency for low-risk regions can be appropriately reduced. According to the classified risk map, conduct coverage analysis on the initial sensor network topology structure to evaluate the coverage of the current sensor network for all risk regions. Through analysis, determine the monitoring blind spots and redundant regions. Monitoring blind spots refer to regions that are not covered by the current sensor network or have insufficient monitoring accuracy. These regions are often distributed at the edges of high-risk regions or in complex terrains, and effective data cannot be obtained due to improper sensor deployment. Redundant regions are regions where the sensor density is too high or multiple sensors collect the same data, resulting in waste of resources and data redundancy. For the monitoring blind spots, optimize the sensor layout to calculate the optimal sensor positions and obtain a supplementary sensor deployment plan. The goal of sensor layout optimization is to achieve effective coverage of the blind spots with the least number of sensors while ensuring the lowest deployment cost and system complexity. During the optimization process, consider factors such as the effective coverage radius of the sensors, the stability of signal transmission, and the specific geographical location of the blind spots. Through simulation and optimization algorithms, determine the optimal deployment positions and quantities of each sensor, supplement the coverage deficiencies of the existing network, and enhance the monitoring ability for high-risk regions. At the same time, calculate the information gain of the redundant regions, evaluate the roles and contributions of each sensor in the entire network, and obtain the ranking of sensor importance. Information gain reflects the contribution of each sensor to the amount of information in the entire system. Sensors with higher information gain monitor key data points, while sensors with lower information gain have little effect on the overall network due to data redundancy. According to the ranking of sensor importance, selectively turn off low-importance sensors to obtain a streamlined sensor network. Dynamically adjust the sampling frequency of the streamlined sensor network. The setting of the sampling frequency affects the real-time nature of the data and the energy consumption of the system. For high-risk regions and key monitoring regions, appropriately increase the sampling frequency to obtain more continuous and detailed information on gas concentration changes. For low-risk regions, appropriately reduce the sampling frequency to reduce the data transmission volume and the workload of the sensors.By dynamically adjusting the sampling frequency, a new sensor operating parameter configuration is obtained, enabling the sensor network to be flexibly adjusted according to the actual situation, and improving the response ability and resource utilization efficiency of the entire system. According to the hierarchical risk map and the sensor network configuration information, an emergency response plan is generated. The emergency response plan includes the personnel evacuation route, the emergency disposal process, and the resource allocation plan. Integrate the supplementary sensor deployment plan, the sensor operating parameter configuration, and the emergency response plan, and output the emergency response strategy and the sensor network configuration information.

[0043] In the embodiment of the present invention, by performing filtering and edge detection processing on the initial multi-dimensional data set, combined with a hybrid deep learning model of 3D-CNN and LSTM, potential abnormal points can be effectively identified, the false alarm rate can be reduced, and the accuracy of leakage detection can be improved. Using the triangulation principle and the particle filter algorithm for multi-sensor collaborative positioning overcomes the problem of low positioning accuracy of traditional methods in complex environments and can quickly and accurately determine the leakage source location. A three-dimensional computational fluid dynamics model is used for gas diffusion simulation, considering the influence of factors such as terrain, obstacles, and meteorological conditions, and can more accurately predict the gas concentration distribution and the risk area. Based on the gas concentration distribution and the risk index, the sensor network is dynamically reconfigured, optimizing the sensor layout and operating parameters, and improving the efficiency and coverage of the monitoring system. By integrating the supplementary sensor deployment plan, the sensor operating parameter configuration, and the emergency response plan, a comprehensive and operable emergency response strategy is provided for decision-makers, improving the handling ability of emergencies. The present invention constructs a closed-loop intelligent gas leakage monitoring system from data collection, abnormal detection, leakage location to risk assessment and emergency response, improving the overall supervision efficiency and safety.

[0044] In a specific embodiment, the process of executing step 101 may specifically include the following steps:

[0045] Perform grid division on the key positions around the gas storage device to obtain a preset monitoring grid, and deploy the Internet of Things gas sensors according to the preset monitoring grid to obtain the initial sensor network topology structure;

[0046] Set the data collection frequency of the Internet of Things gas sensors to obtain a sampling time series, and collect gas concentration data for each Internet of Things gas sensor based on the sampling time series to obtain the original gas concentration data;

[0047] Use the environmental parameter sensor to synchronously collect temperature, humidity, and air pressure to obtain environmental parameter data, and align the time stamps of the original gas concentration data and the environmental parameter data to obtain the aligned multi-dimensional data;

[0048] Perform noise removal on the aligned multi-dimensional data to obtain the denoised multi-dimensional data, and associate the denoised multi-dimensional data with the sensor location information to obtain an initial multi-dimensional data set containing gas concentration, environmental parameters, and location information.

[0049] Specifically, perform precise grid division on the monitoring area. The grid division is based on the spatial distribution of the gas storage device, the complexity of the surrounding environment, and the analysis of potential leakage risks. By dividing the entire monitoring area into several regular or irregular grid cells, assign a unique number to each grid cell, and each grid cell represents a specific spatial location. Assume that the monitoring area of the gas storage device is a rectangular area with a width of , and a height of . Divide this area into grid cells, with each grid cell having a width of and a height of . Then the total number of grids is . Cover the entire monitoring area in this way to form a preset monitoring grid. Based on the preset monitoring grid, deploy the Internet of Things gas sensors reasonably. The deployment of the sensors takes into account the following factors: the risk level of each grid cell, the effective coverage of the monitoring range, and the stability of signal transmission. To ensure monitoring accuracy and coverage, increase the density of sensors in high-risk areas and appropriately reduce the number of sensors in areas with lower risks. Set the effective coverage radius of each sensor to be , and assume that the optimal deployment position of each sensor is the center point of the grid cell , where i = 1, 2,..., m, j = 1, 2,..., n. By calculating the coverage range of each sensor, construct an initial sensor network topology. Set the data acquisition frequency for each Internet of Things gas sensor to obtain a sampling time series. The setting of the sampling time series is determined according to the characteristics of gas leakage and changes in environmental conditions. The sampling time interval is expressed as , which determines the time interval for the sensor to collect data. For each sensor, set its sampling time series to be , where represents the time point of the th sampling. Based on the sampling time series, each Internet of Things gas sensor collects gas concentration data to form an original gas concentration data set , where represents the gas concentration value measured by the sensor at time . At the same time, use environmental parameter sensors to synchronously collect data such as temperature, humidity, and air pressure. The environmental parameter data is , where respectively represent the temperature at time ​ 、 Humidity and air pressure . Align the original gas concentration data and the environmental parameter data by timestamp, unify the data from different sources on the same time basis, and ensure the accuracy and consistency during data analysis. Let the dataset after time alignment be , where represents a multi-dimensional data set containing gas concentration and environmental parameters. Perform noise removal processing on the multi-dimensional data after time alignment to eliminate the invalid information in the data caused by factors such as sensor errors and environmental interference, and improve the quality and reliability of the data. Use the wavelet transform method to perform noise reduction processing on the data. Let the time series of the original gas concentration data be , and its wavelet transform is expressed as:

[0050] ;

[0051] Among them, represents the transformation coefficient of the gas concentration data at scale and position , is the scaled and translated form of the mother wavelet function, represents the conjugate complex number of the mother wavelet function. By selecting appropriate scales and positions , effectively remove the noise components and retain the main features of the signal. The gas concentration data obtained after noise reduction is . Associate the multi-dimensional data after noise reduction with the position information of the sensor. The position information of the sensor is , where represents the coordinate position of the th sensor in the monitoring area. Associate the gas concentration data after noise reduction with the corresponding sensor position to obtain an initial multi-dimensional data set containing gas concentration, environmental parameters and position information, and its representation is:

[0052] ;

[0053] Among them, represents the spatial position of the sensor, represents the gas concentration data after noise reduction at time , and represent temperature, humidity and air pressure data respectively.

[0054] In a specific embodiment, the process of executing step 102 may specifically include the following steps:​​

[0055] Apply wavelet transform to the initial multi-dimensional dataset for multi-scale decomposition to obtain sub-datasets of different frequency components;

[0056] Calculate the local variance of each sensor data based on the sub-datasets to obtain local fluctuation features, and perform adaptive threshold analysis on the local fluctuation features to obtain a dynamic threshold curve;

[0057] Apply a Gaussian filter to the initial multi-dimensional dataset for smoothing to obtain a smoothed dataset, and perform edge detection based on the smoothed dataset to obtain an edge feature map;

[0058] Fuse the dynamic threshold curve with the edge feature map to obtain a preliminary set of abnormal points, and perform abnormal region clustering on the preliminary set of abnormal points to obtain the clustered abnormal regions;

[0059] Extract the center point coordinates, detection time, and associated sensor numbers from the clustered abnormal regions to obtain potential abnormal point data.

[0060] Specifically, apply wavelet transform to the initial multi-dimensional dataset for multi-scale decomposition, decompose the signal into frequency components of different scales, and capture the local features of the signal in both the time and frequency domains. Let the initial multi-dimensional dataset be , where represents the gas concentration data collected by the rd sensor at time , is the total number of sensors. Perform continuous wavelet transform on the data sequence of each sensor, which is expressed as:

[0061] ;

[0062] where is the wavelet transform coefficient of the th sensor data at scale and position , is the dilated and translated form of the mother wavelet function at scale and time position , is the complex conjugate of the mother wavelet function. By changing the scale , decompose the signal into sub-datasets of different frequency components, each of which represents the signal characteristics at different scales. Components of different frequencies can reveal the changes in gas concentration signals at different time resolutions. Based on the sub-datasets of different frequency components, calculate the local variance of each sensor data. The local variance reflects the degree of fluctuation of the data within a certain time window and is used to quantify the fluctuation characteristics of the data. Set a window size of , for the A subset of sensor data , which is in the window The local variance calculation formula is:

[0063] ;

[0064] in, Indicates Sensors at time The local variance at time, is the local average value in the time window. The size of the local variance reflects the fluctuation of gas concentration in the time period. The local variance exceeding the preset target value usually indicates that the concentration changes more drastically and there may be abnormal conditions. The local fluctuation characteristics are subjected to adaptive threshold analysis to obtain a dynamic threshold curve. According to the fluctuation characteristics of the data, a suitable threshold is automatically generated to distinguish normal fluctuations from abnormal fluctuations. Set an initial threshold , and dynamically adjust the threshold according to the distribution of local variance to obtain the dynamic threshold curve at each time point:

[0065] ;

[0066] in, For time The dynamic threshold at the moment, is a weight factor used to adjust the sensitivity of the threshold. The dynamic threshold curve is adaptively adjusted as the local fluctuation characteristics change, so that the threshold can better adapt to the signal changes in different situations and improve the accuracy of anomaly detection. In order to eliminate noise and mutation points in the data, a Gaussian filter is applied to the initial multidimensional data set for smoothing. The Gaussian filter is a signal smoothing tool that effectively removes high-frequency noise in the data, making the data smoother and more continuous. The gas concentration data of each sensor is , the data after Gaussian filtering is expressed as:

[0067] ;

[0068] in, represents the smoothed data, is the standard deviation of the Gaussian filter, which determines the smoothness of the filter. Values exceeding the preset target value will result in smoother data but will lose some useful detail information, while smaller will retain more details but the denoising effect is not obvious. By selecting an appropriate value, a balance is achieved between smoothness and feature retention. After obtaining the smoothed dataset , edge detection is performed on it to identify mutation points and abnormal fluctuation regions in the data. Edge detection methods such as the Sobel operator or Canny operator are used. By calculating the gradient change of the data to determine the edge position, an edge feature map is obtained, where represents the edge feature value of the th sensor at time . The dynamic threshold curve and the edge feature map are fused to identify the initial set of abnormal points. The dynamic threshold curve is compared point by point with the edge feature map. If the edge feature value at a certain time point exceeds the corresponding dynamic threshold , then this point is marked as an abnormal point. The initial set of abnormal points is represented as:

[0069] ;

[0070] where, is the initial set of abnormal points, represents that the th sensor is marked as an abnormal point at time . In this way, the spatio-temporal points that may be abnormal are initially screened out. Cluster analysis is performed on the initial set of abnormal points. Abnormal points adjacent in space and time are aggregated into an abnormal region to better reflect the overall situation of gas leakage. The DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm is used for clustering analysis. This algorithm can automatically identify different abnormal regions according to the density of abnormal points without presetting the number of clusters. Set the neighborhood radius of the DBSCAN algorithm to , and the minimum number of sample points to MinPts. Then the set of abnormal regions after clustering is represented as:

[0071]

[0072] where, is the set of abnormal regions after clustering, represents the th abnormal region, is the total number of abnormal regions. Through clustering analysis, isolated noise points are effectively removed, and truly meaningful abnormal regions are identified. The central point coordinates, detection time, and associated sensor numbers are extracted from the clustered abnormal regions to obtain potential abnormal point data. Let the central point of each abnormal region be , then the coordinates of the central point are represented as the mean of all abnormal points within the region:

[0073] ;

[0074] where is the number of points in the th abnormal region. By extracting the central point coordinates, detection time, and associated sensor numbers of each abnormal region, potential abnormal point data is obtained.

[0075] In a specific embodiment, the process of performing step 103 may specifically include the following steps:

[0076] Perform spatio-temporal sequence reconstruction on the potential abnormal point data to obtain a three-dimensional spatio-temporal data cube;

[0077] Input the three-dimensional spatio-temporal data cube into the 3D-CNN module of the hybrid deep learning model. The 3D-CNN module contains 4 3D convolutional layers, and each layer uses the ReLU activation function and batch normalization to obtain a spatial feature map;

[0078] Apply 3D max pooling operation to the spatial feature map to obtain a feature map with reduced dimensions, and input the feature map with reduced dimensions into the feature extraction network containing 3 3D convolutional layers in the hybrid deep learning model. Each layer uses the LeakyReLU activation function to obtain deep spatial features;

[0079] Perform global average pooling on the deep spatial features to obtain a feature vector with a fixed dimension, and reshape the feature vector into a time series. Input it into the recurrent neural network containing 2 layers of LSTM units in the hybrid deep learning model. Each layer of LSTM units contains 128 hidden states to obtain a time series feature representation;

[0080] Input the time series feature representation into the fully connected layer, use the Softmax activation function to obtain a leakage probability distribution, and perform threshold judgment based on the leakage probability distribution to obtain the time point of leakage occurrence;

[0081] Input the feature vector corresponding to the time point of leakage occurrence into the regression network in the hybrid deep learning model. The regression network contains 3 fully connected layers, and each layer uses the ReLU activation function to obtain the initial leakage source location data;

[0082] Apply a fully connected layer with residual connection to the initial leakage source location data, and use the Sigmoid activation function to obtain the leakage intensity value.

[0083] Specifically, perform spatio-temporal sequence reconstruction on the potential anomaly point data and organize it into a three-dimensional spatio-temporal data cube. Assume that the potential anomaly point data contains the gas concentration values at each time step and sensor location . Arrange these data in three dimensions according to time, spatial coordinates, and concentration values to form a three-dimensional spatio-temporal data cube . Input the three-dimensional spatio-temporal data cube into the 3D convolutional neural network module of the hybrid deep learning model. This 3D-CNN module contains four 3D convolutional layers, and each convolutional layer uses the ReLU activation function and batch normalization. The role of the 3D convolutional layer is to extract the spatial features of the data through the sliding operation of the convolutional kernel in three-dimensional space. For the th convolutional layer, its output is expressed as: ;

[0084] ;

[0085] where, represents the output feature map of the th convolutional layer, is the number of convolutional kernels in the th layer, is the weight matrix of the th convolutional kernel in the th layer, * represents the three-dimensional convolution operation, is the output of the previous layer, is the bias term of the th convolutional kernel in the th layer. The ReLU activation function can introduce non-linear features, enabling the model to learn more complex spatial features, while batch normalization helps to stabilize the model training process and accelerate convergence. Through the extraction of multiple layers of 3D convolution, the model captures the spatial features of different scales in the data layer by layer, obtaining a spatial feature map. Apply 3D max pooling operation to the spatial feature map. While maintaining the main feature information, reduce the dimension of the feature map, reduce the computational complexity, and suppress the overfitting problem. Assume that the size of the max pooling window is , then the pooling operation is expressed as:

[0086] ;

[0087] where, is the pooled feature map, Window represents the size of the pooling window, is the feature map value within the pooling window coverage. After the max pooling operation, a feature map with reduced dimensions is obtained and input into the feature extraction network in the hybrid deep learning model. The feature extraction network consists of three 3D convolutional layers, and each layer uses the Leaky ReLU activation function. Compared with the ordinary ReLU, Leaky ReLU retains a certain gradient in the negative value part (usually , where is a very small positive number), which can enhance the model's perception ability of negative value signals while avoiding the "dead neuron" problem. The output of the feature extraction network is expressed as:

[0088] ;

[0089] where, is the output feature map of the th layer of the feature extraction network, is the number of convolutional kernels in the th layer, is the weight matrix of the rd convolutional kernel in the th layer, and is the bias term. After being processed by the three-layer feature extraction network, deep spatial features are obtained. The deep spatial features retain the complex patterns of the gas concentration changing in space and time and can reflect the spatio-temporal distribution characteristics of the leakage event. A global average pooling operation is performed on the deep spatial features to compress them into a feature vector with a fixed dimension. Global average pooling converts the originally high-dimensional feature map into a feature vector with a fixed size by taking the average value of each channel of the entire feature map. Assuming the size of the feature map is , then the feature vector after global average pooling is expressed as:

[0090] ;

[0091] where, is the output of the last layer of the feature extraction network. Through global average pooling, the complex feature map is compressed into a low-dimensional feature vector . The feature vector is reshaped into a time series and input into the recurrent neural network (RNN) module in the hybrid deep learning model. The RNN module consists of two layers of LSTM units, and each layer contains 128 hidden states. LSTM (Long Short-Term Memory network) can effectively capture the long-term and short-term dependencies in the sequence data and has good modeling ability for phenomena with significant time series changes such as gas leakage. Let the input feature vector sequence be , and the hidden state of the th layer of LSTM be , the state update of the LSTM cell is expressed as:

[0092] ;

[0093] where, is the hidden state of the -th layer at the previous time step, is the memory state of the -th layer at the previous time step. Through the recursive calculation of two layers of LSTM, the temporal feature representation is obtained, and these temporal features can reflect the evolution law of gas concentration over time. The temporal feature representation is input into the fully connected layer, and the leakage probability distribution is obtained through the Softmax activation function. The Softmax function maps the output feature vector to a probability value between 0 and 1, representing the probability of leakage occurring at each time step. Assuming the probability of leakage occurring is , the Softmax activation function is expressed as:

[0094] ;

[0095] where, is the probability of leakage occurring at the -th time step. By performing a threshold judgment on the leakage probability distribution, the time point of leakage occurrence is determined, that is, when , it is considered that leakage occurs at time . The feature vector corresponding to the time point of leakage occurrence is input into the regression network in the hybrid deep learning model. The regression network contains three fully connected layers, and each layer uses the ReLU activation function. The role of the regression network is to estimate the location of the leakage source based on the temporal features. Let the output of the regression network be , representing the spatial location coordinates of the leakage source. Each layer of the regression network is expressed as:

[0096] ;

[0097] where, are the weight matrices of the three fully connected layers respectively, is the bias vector. After the calculation of the three-layer regression network, the initial leakage source location data is obtained. Apply a fully connected layer with residual connection to the initial leakage source location data and use the Sigmoid activation function to obtain the leakage intensity value. The residual connection can effectively alleviate the problem of gradient disappearance in the deep network and retain more feature information. The Sigmoid activation function limits the output value between 0 and 1, representing the normalized value of the leakage intensity. Assuming the final leakage intensity is , it is expressed as:

[0098] ;

[0099] Among them, is the Sigmoid activation function, is the weight matrix of the residual connection layer, is the bias term. Finally, the leakage intensity value is obtained.

[0100] In a specific embodiment, the process of executing step 104 may specifically include the following steps:

[0101] Using the triangulation principle and the initial leakage source position data, select the N nearest gas sensors around, obtain the candidate sensor set, and measure the signal strength of each sensor in the candidate sensor set to obtain the signal strength data set;

[0102] According to the signal strength data set and the preset signal propagation model, calculate the distance from each sensor to the leakage source to obtain the distance measurement set, and based on the distance measurement set, construct the nonlinear state equation and the observation equation of the particle filter algorithm to obtain the particle filter model;

[0103] Initialize the particle filter model, generate M particles, each particle represents a possible leakage source position, obtain the initial particle set, and use the nonlinear state equation to predict the state of the initial particle set to obtain the predicted particle set;

[0104] Calculate the likelihood probability of each particle in the predicted particle set based on the observation equation, perform importance resampling to obtain the updated particle set, and perform weighted averaging on the updated particle set to obtain the target leakage source position data.

[0105] Specifically, use the triangulation principle and the initial leakage source position data to select the nearest gas sensors around to form the candidate sensor set. Assume that the initial leakage source position data is , and there are sensors in the system, and the position of each sensor is , where . Select the nearest sensors by calculating the Euclidean distance from each sensor to the initial leakage source position. The Euclidean distance is expressed as:

[0106] ;

[0107] Among them, represents the distance from the th sensor to the initial leakage source position. Sort all sensors according to the distance and select the first The nearest sensors form a candidate sensor set These candidate sensors are used for signal strength measurement to more accurately determine the location of the leakage source. Signal strength measurements are performed on each sensor in the candidate sensor set to obtain a signal strength data set , where represents the signal strength value measured by the th candidate sensor. The signal strength decreases as the distance between the sensor and the leakage source increases. The signal strength is expressed as:

[0108] ;

[0109] where is the initial release intensity of the leakage source, is the distance from the sensor to the leakage source, is a correction term to prevent the denominator from being zero, is the signal attenuation factor. This model shows that the signal strength has a negative power relationship with the distance from the sensor to the leakage source, that is, the closer the distance, the greater the signal strength, and vice versa. According to the signal strength data set and the preset signal propagation model, the estimated distance from each sensor to the leakage source is calculated backward to obtain a distance measurement set . The calculation formula for the estimated distance is:

[0110] ;

[0111] where is the estimated distance from the th sensor to the leakage source. This formula is deduced from the signal propagation model and reflects the relationship between the measured signal strength and the actual distance to the leakage source. Through calculation, the distance measurement value from each candidate sensor to the leakage source is obtained. Based on the distance measurement set , a non-linear state equation and an observation equation of the particle filter algorithm are constructed to estimate the location of the leakage source. Particle filter is a state estimation method for dealing with non-linear and non-Gaussian systems. Let the location state vector of the leakage source be , representing the spatial location of the leakage source at the time step . The non-linear state equation is expressed as:

[0112] ;

[0113] where is the state transition function of the leakage source location, describing the dynamic change law of the leakage source over time, is the system noise, representing the uncertainty of the model. The observation equation is expressed as:

[0114] ;

[0115] where is the observation value, i.e., the distance measurement set , is the observation model, describing the relationship between the leakage source location and the observation value, is the observation noise. Through the state equation and the observation equation, a particle filter model is constructed. The particle filter model is initialized to generate particles, and each particle represents a possible leakage source location. Each particle in the initial particle set is randomly sampled from the state space. The initial particle set is distributed around the initial leakage source location to cover all possible leakage source locations. The state of the initial particle set is predicted using the non - linear state equation to obtain the predicted particle set , and its update formula is:

[0116] ;

[0117] where is the predicted position of the th particle at time step , is the system noise of the th particle. Through state prediction, the changing trend of the leakage source location over time is simulated. The likelihood probability of each particle in the predicted particle set is calculated based on the observation equation. The likelihood probability represents the probability size of the observed data under the given particle state and is usually expressed as:

[0118] ;

[0119] where is the likelihood probability of the th particle, is the observation value at time step , is the th particle's predicted position. According to the likelihood probability, importance resampling is performed on the particles, removing the particles with lower likelihood probabilities and increasing the number of particles with higher likelihood probabilities, so that the particle set can be more concentratedly distributed near the true leakage source location. The resampled particle set better represents the possible location distribution of the leakage source. The updated particle set is weighted averaged to obtain the target leakage source location data. Let the weight of the resampled particle be , then the target leakage source location is expressed as:

[0120] ;

[0121] wherein, is the time step The estimated value of the leakage source position at the moment. By weighted averaging and integrating the information of all particles, a more accurate estimated value of the leakage source position is obtained.

[0122] In a specific embodiment, the process of executing step 105 may specifically include the following steps:

[0123] Input the target leakage source position data and the leakage intensity value into a three-dimensional computational fluid dynamics model, construct a three-dimensional computational domain based on the target leakage source position data, and perform grid division on the three-dimensional computational domain to obtain computational grids;

[0124] Set the leakage intensity value as the boundary condition of the leakage source position in the computational grid to obtain the initial boundary condition, and set the solid boundary within the computational domain according to the terrain and obstacle distribution around the gas storage device to obtain the complete computational boundary;

[0125] Set the inlet boundary condition of the computational domain based on the current meteorological data, including wind speed, wind direction, and temperature distribution, to obtain the environmental boundary condition;

[0126] Use the SIMPLE algorithm to solve the Navier-Stokes equations and the mass transport equation to obtain the velocity field and the pressure field, and discretize the diffusion equation based on the velocity field and the pressure field using the finite volume method to obtain the discretized equations;

[0127] Use the TDMA algorithm to solve the discretized equations to obtain the gas concentration distribution data, calculate the hazard index of each grid point according to the gas concentration distribution data and the preset exposure limit value, and perform spatial interpolation to obtain the risk index.

[0128] Specifically, input the target leakage source position data and the leakage intensity value into the three-dimensional computational fluid dynamics model. Assume the target leakage source position is , and the leakage intensity is . These two parameters are used as the input variables of the CFD model. Construct a three-dimensional computational domain based on the target leakage source position data, and this computational domain covers the leakage source and its surrounding areas that may be affected. Assume the size of the computational domain is , indicating the range of the computational domain in three spatial dimensions. Perform grid division on the computational domain. The density of the grid division directly affects the accuracy of the simulation results and the complexity of the calculation. Use a finer grid near the leakage source and a coarser grid in the areas far from the leakage source to balance the calculation efficiency and accuracy. Assume the total number of grids after grid division is , where are the numbers of grids in the three dimensions of the computational domain, and the volume of each grid cell is expressed as:

[0129] ;

[0130] where, , , respectively represent the grid cell sizes in the , and directions. After the grid division is completed, the computational grid is obtained, and each grid cell is used to solve the gas diffusion equation. The leakage intensity value is set as the boundary condition at the leakage source position in the computational grid, representing the gas release rate at the leakage source position. The leakage source boundary condition is expressed as:

[0131] ;

[0132] where, represents the gas concentration boundary condition at the leakage source position, that is, the mass of gas per unit volume per unit time. The setting of the initial boundary condition is the basis of the whole simulation, which determines the initial state of the gas diffusion from the leakage source. After setting the initial boundary condition, according to the terrain and obstacle distribution around the gas storage device, the solid boundary in the computational domain is set. The solid boundary refers to solid structures such as gas storage tanks, buildings, and the ground that will block the gas diffusion. Let the velocity of the solid boundary be zero, that is, the solid boundary condition is expressed as:

[0133] ;

[0134] where, is the fluid velocity at the solid boundary, is the normal vector of the solid boundary. By setting the solid boundary condition, it is ensured that the gas flow in the computational domain conforms to the actual terrain and obstacle distribution, and the complete computational boundary is obtained. The inlet boundary condition of the computational domain is set based on the current meteorological data. The meteorological data includes information such as wind speed, wind direction, and temperature distribution, and these parameters have a direct impact on the gas diffusion behavior. Assume that the wind speed at the inlet boundary is , the wind direction is , and the temperature is , then the inlet boundary condition is expressed as:

[0135] ;

[0136] where, is the wind speed vector at the inlet, representing the components of the wind speed in the and directions, while The direction is usually set to zero to represent the horizontal wind. The temperature boundary condition is expressed as:

[0137] ;

[0138] The setting of the boundary conditions can reflect the diffusion state of the gas in the computational domain under the current meteorological conditions, and obtain the complete environmental boundary conditions. The SIMPLE algorithm is used to solve the Navier-Stokes equations and the mass transport equation. The Navier-Stokes equations describe the motion laws of the fluid, including the momentum equation and the continuity equation. The momentum equation is expressed as:

[0139] ;

[0140] Among them, is the fluid density, is the velocity field, is the pressure field, is the fluid dynamic viscosity coefficient, is the external force term. The continuity equation is expressed as:

[0141] ;

[0142] These two equations jointly describe the motion state of the fluid. The SIMPLE algorithm alternately iteratively solves the momentum equation and the continuity equation to obtain the distributions of the velocity field and the pressure field. Based on the obtained velocity field and the pressure field , the mass transport equation is solved to obtain the concentration distribution of the gas. The mass transport equation is expressed as:

[0143] ;

[0144] Among them, is the gas concentration, is the diffusion coefficient of the gas. The finite volume method is used to discretize the mass transport equation to obtain a discretized system of equations. The finite volume method integrates each grid cell in the computational domain to transform the partial differential equation into an algebraic equation. The form of the discretized equation is:

[0145] ;

[0146] Among them, represents the gas concentration in the th grid cell at the th time step, represents the flow velocity from the grid cell to , is the unit normal vector in this direction, is the distance between two grid cells. Through the discretization method, the original partial differential equation is transformed into a system of linear algebraic equations for numerical solution. To solve the discretized equations, the TDMA algorithm is used. The TDMA algorithm is an efficient method for solving linear equations and is applicable to the discretized equations obtained by the finite volume method. The gas concentration distribution data for each grid cell is obtained by solving through the TDMA algorithm . These data represent the spatial distribution of the gas within the entire computational domain. Based on the gas concentration distribution data and the preset exposure limit , the hazard index for each grid point is calculated . The hazard index indicates the degree to which the gas concentration exceeds the safety limit and is expressed as:

[0147] ;

[0148] where is the hazard index for the -th grid point, and when , it indicates that the gas concentration exceeds the safety limit. By performing spatial interpolation on the hazard indices of all grid points, the risk index distribution of the entire computational domain is obtained . The risk index reflects the potential impact degree of different regions affected by gas leakage.

[0149] In a specific embodiment, the process of executing step 106 may specifically include the following steps:

[0150] Perform gradient analysis on the gas concentration distribution data to obtain high-concentration regions and regions where the concentration change rate exceeds a preset target value, and mark them as key monitoring regions;

[0151] Based on the risk index, conduct risk level division for the monitoring regions to obtain a graded risk map, and based on the graded risk map, perform coverage analysis on the initial sensor network topology to obtain monitoring blind spots and redundant regions;

[0152] Optimize the sensor layout for the monitoring blind spots, calculate the optimal sensor positions, and obtain a supplementary sensor deployment plan;

[0153] Based on the redundant region information, calculate the information gain of each sensor to obtain the sensor importance ranking, and selectively turn off low-importance sensors according to the ranking results to obtain a streamlined sensor network;

[0154] Dynamically adjust the sampling frequency for the streamlined sensor network to obtain the sensor operating parameter configuration, and based on the graded risk map and the sensor network configuration information, generate an emergency response plan, including evacuation routes for personnel, emergency response procedures, and resource allocation plans;

[0155] Integrate the supplementary sensor deployment plan, sensor working parameter configuration, and emergency response plan, and output the emergency response strategy and sensor network configuration information.

[0156] Specifically, perform gradient analysis on the gas concentration distribution data to identify high-concentration areas and areas where the concentration change rate exceeds a preset target value. Gradient analysis is a method for detecting the data change rate. By calculating the gradient of the concentration distribution in space, it reveals the direction and speed of gas diffusion. Let the gas concentration distribution data be , where represents the gas concentration value at the spatial coordinate . The gradient is expressed as:

[0157] ;

[0158] where, is the gradient vector of the gas concentration, , and respectively represent the change rates of the gas concentration in the , and directions. By calculating the gradient magnitude of each grid point, high-concentration areas and areas where the concentration change rate exceeds the preset target value are obtained. These areas are usually the main diffusion directions of gas leakage, posing a significant threat to the environment and personnel safety, and are marked as key monitoring areas. Based on the risk index, the risk levels of the monitoring areas are divided. The risk index represents the potential impact degree of different areas affected by gas leakage, usually a function of gas concentration and exposure limit. The following formula is used to define the risk index:

[0159] ;

[0160] where, is the gas concentration at the spatial position , is the safe exposure limit of the gas. When , it means that the gas concentration in this area exceeds the safety limit, and there is a health risk exceeding the preset target value. According to the value range of the risk index , the monitoring areas are divided into different risk levels, such as low-risk, medium-risk, and high-risk areas, to obtain a hierarchical risk map. The hierarchical risk map can visually represent the threat degree of gas leakage to different areas. According to the hierarchical risk map, perform coverage analysis on the initial sensor network topology structure to identify monitoring blind spots and redundant areas. The initial sensor network topology structure is represented as a graph , where represents the set of sensor nodes, Represents the connection relationship between sensors. Each sensor node is located at , and its coverage radius is . The monitoring blind area refers to the area with a relatively high risk level but not covered by sensors, while the redundant area is the area with too high sensor density and repeated information collection. To quantify the monitoring blind area, calculate the distance from each high-risk area to the nearest sensor . If the distance , then this area is considered a monitoring blind area. The redundant area is evaluated by calculating the overlap degree of the coverage area of each sensor. For the identified monitoring blind areas, optimize the sensor layout to improve the monitoring accuracy of these areas. The goal of sensor layout optimization is to cover all blind areas with the least number of sensors, while minimizing the layout cost and system complexity. Set the blind area set as , and the optimal sensor position is , so that each blind area is covered by at least one sensor . By solving the optimization problem, obtain the deployment plan of supplementary sensors to ensure that all high-risk areas can be effectively monitored. At the same time, based on the information of the redundant area, calculate the information gain of each sensor to obtain the importance ranking of the sensors. Information gain refers to the size of the contribution of each sensor to the information of the entire network, which is used to measure its importance. Let the information gain of each sensor be , then the information gain is expressed as:

[0161] ;

[0162] where is the entropy of the measurement information of all sensors in the network, representing the overall uncertainty of the system, is the entropy of the system information after removing sensor . The higher the information gain, the greater the contribution of sensor to the system information. According to the information gain ranking result, selectively turn off low-importance sensors to obtain a streamlined sensor network. Reduce data redundancy, lower the energy consumption and maintenance cost of the system. At the same time, based on the information of the redundant area, calculate the information gain of each sensor to obtain the importance ranking of the sensors. Information gain refers to the size of the contribution of each sensor to the information of the entire network, which is used to measure its importance. Let the information gain of each sensor be , then the information gain is expressed as:

[0163] ;

[0164] where is the entropy of the measurement information of all sensors in the network, representing the overall uncertainty of the system. is the entropy of the system information after removing sensor . The higher the information gain, the greater the contribution of the sensor to the system information. According to the sorting result of the information gain, selectively turn off the sensors with low importance to obtain a streamlined sensor network. Dynamically adjust the sampling frequency of the sensors to improve the resource utilization efficiency and response speed of the system. The dynamic adjustment of the sampling frequency is set according to the risk level of the area where the sensor is located. Assume that the streamlined sensor set is , and the sampling frequency of each sensor is , then the dynamically adjusted sampling frequency is expressed as:

[0165] ;

[0166] where and are the minimum and maximum sampling frequencies respectively, is the risk index of the location where the sensor is located is the maximum value of the risk index. Through dynamic adjustment, the sampling frequency of the sensor changes flexibly according to the actual situation, increasing the sampling frequency in high-risk areas to obtain more detailed information on gas concentration changes, and decreasing the sampling frequency in low-risk areas to reduce the data transmission volume and system load. Based on the hierarchical risk map and the sensor network configuration information, generate an emergency response plan. The emergency response plan includes the personnel evacuation route, the emergency disposal process, and the resource allocation plan. The personnel evacuation route needs to combine the direction and speed of gas diffusion to design a safe and efficient escape route to avoid people passing through high-concentration areas and near the leakage source. The emergency disposal process includes all steps from discovering the leakage to controlling the leakage, such as closing the leakage source, starting the ventilation system, isolating the dangerous area, etc. The resource allocation plan involves the allocation and scheduling of emergency supplies, rescue personnel, and equipment to ensure that the accident can be responded to quickly and effectively when it occurs. Integrate the supplementary sensor deployment plan, the sensor working parameter configuration, and the emergency response plan, and output the emergency response strategy and the sensor network configuration information.

[0167] The above describes the gas leakage monitoring method based on the Internet of Things in the embodiments of the present invention. Next, the gas leakage monitoring device based on the Internet of Things in the embodiments of the present invention will be described. Please refer to Figure 2 . An embodiment of the gas leakage monitoring device based on the Internet of Things in the embodiments of the present invention includes:

[0168] The acquisition module 201 is used to collect data from the gas storage device to obtain an initial multi-dimensional data set;

[0169] The processing module 202 is configured to perform filtering and edge detection processing on the initial multi-dimensional data set to obtain potential abnormal point data;

[0170] The detection module 203 is configured to input the potential abnormal point data into a hybrid deep learning model for gas leakage detection to obtain initial leakage source location data and leakage intensity values;

[0171] The positioning module 204 is configured to perform precise leakage positioning with multi-sensor collaboration based on the initial leakage source location data to obtain target leakage source location data;

[0172] The simulation module 205 is configured to input the target leakage source location data and leakage intensity values into a three-dimensional computational fluid dynamics model for gas diffusion simulation to obtain gas concentration distribution data and risk indices;

[0173] The output module 206 is configured to perform dynamic reconstruction of the sensor network based on the gas concentration distribution data and risk indices, and output an emergency response strategy and sensor network configuration information.

[0174] Through the collaborative cooperation of the above-mentioned various components, by performing filtering and edge detection processing on the initial multi-dimensional data set, combined with a hybrid deep learning model of 3D-CNN and LSTM, potential abnormal points can be effectively identified, the false alarm rate can be reduced, and the accuracy of leakage detection can be improved. Using the triangulation principle and particle filtering algorithm for multi-sensor collaborative positioning overcomes the problem of low positioning accuracy of traditional methods in complex environments and can quickly and accurately determine the leakage source location. Using a three-dimensional computational fluid dynamics model for gas diffusion simulation takes into account the effects of factors such as terrain, obstacles, and meteorological conditions, and can more accurately predict the gas concentration distribution and risk areas. Based on the gas concentration distribution and risk indices, the sensor network is dynamically reconstructed, the sensor layout and working parameters are optimized, and the efficiency and coverage of the monitoring system are improved. By integrating and supplementing the sensor deployment plan, sensor working parameter configuration, and emergency response plan, a comprehensive and operable emergency response strategy is provided for decision-makers, and the ability to handle emergencies is improved. The present invention constructs a closed-loop intelligent gas leakage monitoring system from data acquisition, abnormal detection, leakage positioning to risk assessment and emergency response, improving the overall supervision efficiency and safety.

[0175] Above Figure 2 The gas leakage monitoring device based on the Internet of Things in the embodiments of the present invention has been described in detail from the perspective of modular functional entities. Next, the gas leakage monitoring device based on the Internet of Things in the embodiments of the present invention will be described in detail from the perspective of hardware processing.

[0176] Figure 3FIG. 0 is a schematic structural diagram of a gas leakage monitoring device based on the Internet of Things provided by an embodiment of the present invention. The gas leakage monitoring device 300 based on the Internet of Things may have differences exceeding a preset target value due to configuration or performance differences, and may include one or more processors (central processing units, CPUs) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 for storing application programs 333 or data 332 (for example, one or more mass storage device ends). Among them, the memory 320 and the storage media 330 may be transient storage or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the gas leakage monitoring device 300 based on the Internet of Things. Further, the processor 310 may be configured to communicate with the storage media 330 and execute a series of instruction operations in the storage media 330 on the gas leakage monitoring device 300 based on the Internet of Things to implement the steps of the above-mentioned gas leakage monitoring method based on the Internet of Things.

[0177] The gas leakage monitoring device 300 based on the Internet of Things may further include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, and so on. Those skilled in the art can understand that Figure 3 the shown structural diagram of the gas leakage monitoring device based on the Internet of Things does not limit the gas leakage monitoring device based on the Internet of Things provided by the present invention, and may include more or fewer components than shown in the figure, or combine certain components, or arrange different components.

[0178] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are run on a computer, the computer is caused to execute the steps of the gas leakage monitoring method based on the Internet of Things.

[0179] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described systems, systems, and units can refer to the corresponding processes in the foregoing method embodiments and will not be described herein again.

[0180] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0181] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of various embodiments of the present invention.

Claims

1. A gas leakage monitoring method based on the Internet of Things, characterized in that, The method includes: Collecting data from the gas storage device to obtain an initial multi-dimensional data set; Performing filtering and edge detection processing on the initial multi-dimensional data set to obtain potential abnormal point data; specifically including: applying wavelet transform to the initial multi-dimensional data set for multi-scale decomposition to obtain sub-data sets of different frequency components; calculating the local variance of each sensor data based on the sub-data sets to obtain local fluctuation characteristics, and performing adaptive threshold analysis on the local fluctuation characteristics to obtain a dynamic threshold curve; applying a Gaussian filter to the initial multi-dimensional data set for smoothing processing to obtain a smoothed data set, and performing edge detection based on the smoothed data set to obtain an edge feature map; fusing the dynamic threshold curve with the edge feature map to obtain a preliminary abnormal point set, and clustering the abnormal regions of the preliminary abnormal point set to obtain clustered abnormal regions; extracting the center point coordinates, detection time, and associated sensor numbers from the clustered abnormal regions to obtain potential abnormal point data; Inputting the potential abnormal point data into a hybrid deep learning model for gas leakage detection to obtain initial leakage source location data and leakage intensity values; Performing precise leakage localization with multi-sensor collaboration according to the initial leakage source location data to obtain target leakage source location data; Inputting the target leakage source location data and the leakage intensity values into a three-dimensional computational fluid dynamics model for gas diffusion simulation to obtain gas concentration distribution data and risk indices; Performing dynamic reconstruction of the sensor network based on the gas concentration distribution data and the risk indices, and outputting an emergency response strategy and sensor network configuration information.

2. The gas leakage monitoring method based on the Internet of Things according to claim 1, wherein The collecting data from the gas storage device to obtain an initial multi-dimensional data set includes: Dividing the key positions around the gas storage device into grids to obtain preset monitoring grids, and deploying Internet of Things gas sensors according to the preset monitoring grids to obtain an initial sensor network topology structure; Setting the data collection frequency of the Internet of Things gas sensors to obtain a sampling time series, and collecting gas concentration data for each Internet of Things gas sensor based on the sampling time series to obtain original gas concentration data; Synchronously collecting temperature, humidity, and air pressure using environmental parameter sensors to obtain environmental parameter data, and aligning the time stamps of the original gas concentration data and the environmental parameter data to obtain aligned multi-dimensional data; Removing noise from the aligned multi-dimensional data to obtain denoised multi-dimensional data, and associating the denoised multi-dimensional data with sensor location information to obtain an initial multi-dimensional data set including gas concentration, environmental parameters, and location information.

3. The gas leakage monitoring method based on the Internet of Things according to claim 2, characterized in that, The inputting the potential abnormal point data into a hybrid deep learning model for gas leakage detection to obtain initial leakage source location data and leakage intensity values includes: Reconstructing the spatio-temporal sequence of the potential abnormal point data to obtain a three-dimensional spatio-temporal data cube; Input the three-dimensional spatio-temporal data cube into the 3D-CNN module of the hybrid deep learning model. The 3D-CNN module contains 4 3D convolutional layers, and each layer uses the ReLU activation function and batch normalization to obtain a spatial feature map; Apply 3D max pooling operation to the spatial feature map to obtain a feature map with reduced dimensions, and input the feature map with reduced dimensions into the feature extraction network containing 3 3D convolutional layers in the hybrid deep learning model. Each layer uses the LeakyReLU activation function to obtain deep spatial features; Perform global average pooling on the deep spatial features to obtain a feature vector with a fixed dimension, reshape the feature vector into a time series, and input it into the recurrent neural network containing 2 layers of LSTM cells in the hybrid deep learning model. Each layer of LSTM cells contains 128 hidden states to obtain a time series feature representation; Input the time series feature representation into a fully connected layer, use the Softmax activation function to obtain a leakage probability distribution, and perform a threshold judgment based on the leakage probability distribution to obtain the time point of leakage occurrence; Input the feature vector corresponding to the time point of leakage occurrence into the regression network in the hybrid deep learning model. The regression network contains 3 fully connected layers, and each layer uses the ReLU activation function to obtain the initial leakage source location data; Apply a fully connected layer with residual connection to the initial leakage source location data, use the Sigmoid activation function to obtain the leakage intensity value.

4. The method for monitoring gas leakage based on the Internet of Things according to claim 3, characterized in that The performing multi-sensor collaborative precise leakage positioning according to the initial leakage source location data to obtain the target leakage source location data includes: Select N nearest gas sensors around using the triangulation principle and the initial leakage source location data to obtain a candidate sensor set, and perform signal strength measurement on each sensor in the candidate sensor set to obtain a signal strength data set; According to the signal strength data set and a preset signal propagation model, calculate the distance from each sensor to the leakage source to obtain a distance measurement set, and construct a non-linear state equation and an observation equation of the particle filter algorithm based on the distance measurement set to obtain a particle filter model; Initialize the particle filter model to generate M particles, each particle representing a possible leakage source location, to obtain an initial particle set, and use the non-linear state equation to perform state prediction on the initial particle set to obtain a predicted particle set; Calculate the likelihood probability of each particle in the predicted particle set based on the observation equation, perform importance resampling to obtain an updated particle set, and perform weighted averaging on the updated particle set to obtain the target leakage source location data.

5. The method for monitoring gas leakage based on the Internet of Things according to claim 4, wherein The inputting the target leakage source location data and the leakage intensity value into a three-dimensional computational fluid dynamics model to perform gas diffusion simulation to obtain gas concentration distribution data and a risk index includes: Input the target leakage source location data and the leakage intensity value into a three-dimensional computational fluid dynamics model, construct a three-dimensional computational domain based on the target leakage source location data, and perform grid division on the three-dimensional computational domain to obtain a computational grid; Set the leakage intensity value as the boundary condition of the leakage source position in the computational grid to obtain the initial boundary condition, and set the solid boundary in the computational domain according to the terrain and obstacle distribution around the gas storage device to obtain the complete computational boundary; Set the inlet boundary condition of the computational domain based on the current meteorological data, including wind speed, wind direction and temperature distribution, to obtain the environmental boundary condition; Use the SIMPLE algorithm to solve the Navier-Stokes equations and the mass transport equation to obtain the velocity field and pressure field, and discretize the diffusion equation based on the velocity field and pressure field using the finite volume method to obtain the discretized equations; Use the TDMA algorithm to solve the discretized equations to obtain the gas concentration distribution data, calculate the hazard index of each grid point according to the gas concentration distribution data and the preset exposure limit, and perform spatial interpolation to obtain the risk index.

6. The method for monitoring gas leakage based on the Internet of Things according to claim 5, characterized in that, Perform dynamic reconstruction of the sensor network based on the gas concentration distribution data and the risk index, and output the emergency response strategy and sensor network configuration information, including: Perform gradient analysis on the gas concentration distribution data to obtain the high-concentration area and the area where the concentration change rate exceeds the preset target value, and mark them as key monitoring areas; Divide the monitoring area into risk levels based on the risk index to obtain a graded risk map, and perform coverage analysis on the initial sensor network topology according to the graded risk map to obtain the monitoring blind area and redundant area; Optimize the sensor layout in the monitoring blind area, calculate the optimal sensor position, and obtain the supplementary sensor deployment plan; Based on the redundant area, calculate the information gain of each sensor to obtain the sensor importance ranking result, and selectively turn off the low-importance sensors according to the sensor importance ranking result to obtain the streamlined sensor network; Dynamically adjust the sampling frequency of the streamlined sensor network to obtain the sensor working parameter configuration, and generate an emergency response plan according to the graded risk map and the sensor network configuration information, including the personnel evacuation route, emergency response process and resource allocation plan; Integrate the supplementary sensor deployment plan, the sensor working parameter configuration and the emergency response plan, and output the emergency response strategy and sensor network configuration information.

7. An Internet of Things-based gas leakage monitoring device, characterized in that, For implementing the gas leakage monitoring method based on the Internet of Things as described in any one of claims 1-6, the gas leakage monitoring device based on the Internet of Things includes: An acquisition module for acquiring data of the gas storage device to obtain an initial multi-dimensional data set; A processing module for filtering and edge detection processing of the initial multi-dimensional data set to obtain potential anomaly point data; A detection module for inputting the potential anomaly point data into a hybrid deep learning model for gas leakage detection to obtain the initial leakage source position data and leakage intensity value; A positioning module for performing precise leakage positioning with multi-sensor collaboration according to the initial leakage source position data to obtain the target leakage source position data; A simulation module, configured to input the target leak source location data and the leak intensity value into a three-dimensional computational fluid dynamics model, perform gas diffusion simulation, and obtain gas concentration distribution data and a risk index; An output module, configured to perform dynamic reconstruction of a sensor network based on the gas concentration distribution data and the risk index, and output an emergency response strategy and sensor network configuration information.

8. An Internet of Things-based gas leakage monitoring device, characterized in that, The Internet of Things-based gas leakage monitoring device includes: a memory and at least one processor, and instructions are stored in the memory; The at least one processor calls the instructions in the memory, so that the Internet of Things-based gas leakage monitoring device executes the Internet of Things-based gas leakage monitoring method according to any one of claims 1-6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that, When the instructions are executed by the processor, the Internet of Things-based gas leakage monitoring method according to any one of claims 1-6 is implemented.

Images