A liquefied gas cylinder multi-dimensional perception early warning disposal terminal and internet of things system

Abstract

The application discloses a liquefied gas cylinder multi-dimensional perception early warning disposal terminal and an Internet of Things system, the terminal comprising an explosion-proof shell, an electromagnetic lock buckle mechanism is arranged on the explosion-proof shell, the electromagnetic lock buckle mechanism is detachably installed at a handle of a neck of the cylinder, a battery, a main control module, a positioning module, a communication module and an environment perception module are arranged in the explosion-proof shell, the electromagnetic lock buckle mechanism, the battery, the positioning module, the communication module and the environment perception module are connected with the main control module respectively, the main control module is integrated with an edge AI model, is used for dynamically constructing a behavior baseline and an elastic geographic fence according to a deployment scene and a use mode of the cylinder, and unsupervised anomaly detection, hierarchical response and disposal strategy generation are completed through the communication module receiving multi-source perception data. The application has the advantages of active prediction, early warning and disposal capacity for the liquefied gas cylinder, low false alarm probability, fast response speed and high emergency response efficiency.

Description

Technical Field

[0001] This invention relates to the field of gas safety equipment technology, and more specifically, to a multi-dimensional sensing, early warning and handling terminal and Internet of Things system for liquefied petroleum gas cylinders. Background Technology

[0002] As widely used mobile pressure vessels, liquefied petroleum gas (LPG) cylinders face complex distribution channels and variable operating environments, making traditional management models inadequate for meeting actual safety management needs. Existing technologies still have several significant shortcomings in cylinder supervision scenarios: Current cylinder management largely relies on QR codes or RFID tags for identification and traceability, requiring manual scanning for information entry. This fails to achieve real-time cylinder location and dynamic monitoring of operational status, resulting in a passive and lagging regulatory approach. Furthermore, existing positioning devices often employ fixed-threshold geofencing technology, which is ill-suited to the complex and ever-changing urban environment. This leads to high false alarm rates and a lack of early warning capabilities for potential risks, resulting in a simplistic and impractical early warning mechanism. In addition, traditional monitoring equipment only possesses basic positioning functions, failing to identify core safety risks such as gas leaks and illegal modifications in their early stages, and lacking proactive intervention methods, resulting in insufficient comprehensiveness and foresight in safety protection. Moreover, individual monitoring terminals often operate in isolation, hindering the construction of a networked intelligent collaborative system. This leads to a lack of efficient linkage mechanisms in emergency response, further restricting the overall effectiveness of cylinder safety management. Summary of the Invention

[0003] In view of the above problems, embodiments of the present invention provide a multi-dimensional sensing, early warning and disposal terminal and Internet of Things system for liquefied gas cylinders to solve the existing technical problems.

[0004] To address the aforementioned technical problems, this invention provides a multi-dimensional sensing, early warning, and response terminal for liquefied petroleum gas (LPG) cylinders. The terminal includes an explosion-proof housing with an electromagnetic locking mechanism detachably mounted on the handle of the cylinder neck. Inside the explosion-proof housing are a battery, a main control module, a positioning module, a communication module, and an environmental sensing module. The electromagnetic locking mechanism, battery, positioning module, communication module, and environmental sensing module are all connected to the main control module. The main control module integrates an edge AI model to dynamically construct behavioral baselines and flexible geofences based on the cylinder's deployment scenario and usage pattern. Through the communication module, it receives multi-source sensing data to perform unsupervised anomaly detection, tiered response, and response strategy generation.

[0005] Furthermore, the present invention provides a multi-dimensional sensing, early warning, and handling terminal for liquefied petroleum gas cylinders. The main control module downloads a basic model from the cloud based on a federated learning framework. It then uses the typical activity trajectory of the cylinder, statistical gas consumption patterns, and environmental parameter changes collected by the environmental sensing module to fine-tune the basic model using the FedAvg algorithm, thereby constructing a behavioral baseline that conforms to reality.

[0006] Furthermore, the present invention provides a multi-dimensional sensing, early warning, and handling terminal for liquefied petroleum gas cylinders. The main control module uses the GraphSAGE algorithm to perform feature fusion based on real-time location information collected by the positioning module, as well as scenario-based dynamic data and historical risk data, to calculate the risk propagation probability within the area and output a dynamically adjusted safety distance threshold, thereby constructing an elastic geofence.

[0007] Furthermore, the present invention provides a multi-dimensional sensing, early warning, and disposal terminal for liquefied petroleum gas cylinders. The environmental sensing module includes a temperature sensor, a gas concentration sensor, a microphone, and an inertial measurement module. The main control module performs parallel processing and feature extraction on multi-source heterogeneous sensor data based on a temporal convolutional network to identify cylinder risks, including micro-leakage, overturning, abnormal handling, and violent damage.

[0008] Furthermore, the present invention provides a multi-dimensional sensing, early warning, and handling terminal for liquefied petroleum gas cylinders, wherein a camera is installed on the outer surface of the explosion-proof shell, and the main control module uses a temporal convolutional network based on an attention mechanism to perform end-side fusion sensing of the video stream collected by the camera, the acoustic scene information collected by the microphone, and the spatial motion data of the inertial measurement module to complete unsupervised anomaly detection.

[0009] Furthermore, the present invention provides a multi-dimensional sensing, early warning, and handling terminal for liquefied petroleum gas cylinders, wherein the environmental sensing module further includes an accelerometer sensor, which is used to sense whether the cylinder is moving, and the main control module switches between working mode and sleep mode based on the sensing data from the accelerometer sensor.

[0010] Furthermore, the present invention provides a multi-dimensional sensing, early warning, and response terminal for liquefied petroleum gas (LPG) cylinders, which further includes an expansion module. The expansion module includes an expansion module housing, which is fixedly connected to the bottom of the cylinder. The expansion module housing integrates an ultrasonic gas leak detection unit, a weighing sensor, an emergency response unit, and an expansion device circuit. The ultrasonic gas leak detection unit is used to detect LPG leaks in the cylinder, the weighing sensor is used to detect the weight of the cylinder, the emergency response unit is used for emergency fire extinguishing, and the expansion device circuit is connected to the ultrasonic gas leak detection unit, the weighing sensor, and the emergency response unit, and is communicatively connected to the main control module.

[0011] Furthermore, the present invention provides a multi-dimensional sensing, early warning, and handling terminal for liquefied petroleum gas cylinders. The main control module, based on the time-series data of the weighing sensor uploaded by the expansion device circuit, uses a Bayesian change point detection algorithm to identify abnormal consumption patterns of the cylinder's remaining volume and combines it with positioning information to construct a digital profile of the entire chain of the cylinder from filling to transportation to use.

[0012] Furthermore, the present invention provides a multi-dimensional sensing, early warning, and handling terminal for liquefied petroleum gas cylinders, wherein the hierarchical response and handling strategy of the main control module includes:

[0013] Low-risk response: Mark potential risk patterns in geofencing records and recommend corresponding safety education content through knowledge graphs;

[0014] Medium-risk response: Push out early warnings, and the decision engine generates a response plan that includes augmented reality guidance, and simultaneously notifies the regional safety officer;

[0015] High-risk response: Activate the local deterrence function of the terminal, and upload the multi-source sensing data stream from the scene to the cloud for in-depth analysis;

[0016] Emergency Response: Perform all the above operations and automatically activate the emergency self-handling unit of the extended module. At the same time, synchronize the event to the regional emergency response digital platform for collaborative scheduling and simulation.

[0017] This invention also provides an Internet of Things (IoT) system for gas safety sensing, comprising:

[0018] Multiple terminals described in any of the above embodiments are interconnected to form a terminal sensing network, which captures multi-source sensing data in real time.

[0019] By utilizing edge computing nodes to receive data from the terminal sensing network, the information uploaded by the terminals within the region is aggregated and preliminarily analyzed, and the structured processing results are synchronized to the cloud security platform.

[0020] By leveraging the cloud security platform and its built-in large-scale graph computing engine, the aggregated data transmitted by edge nodes is deeply mined to identify risk propagation patterns and group anomalies in the terminal network. At the same time, the edge AI model deployed in the main control module of each terminal is continuously optimized through joint learning to improve the data analysis accuracy and response speed of the terminal perception network.

[0021] Based on MAC address signals collected by the terminal sensing network, combined with a cloud-optimized swarm intelligence algorithm, the distribution map of people indoors and outdoors is reconstructed in real time, and dynamic, escape-optimal path planning is quickly generated in emergency situations.

[0022] The beneficial effects of this invention are as follows: By utilizing a positioning and communication module mounted on an explosion-proof shell, combined with an electromagnetic locking mechanism detachably fixed at the cylinder neck handle, this invention enables real-time positioning and dynamic monitoring of the cylinder's operational status during circulation. The linkage design between the electromagnetic locking mechanism and the main control module effectively prevents cylinders from escaping the liquefied gas regulatory scope, promoting a shift in regulatory methods from passive traceability to proactive prevention. The edge AI model integrated into the main control module dynamically constructs behavioral baselines and flexible geofencing based on the actual deployment scenarios and usage patterns of the cylinders. Compared to traditional fixed-threshold geofencing technology, this is more accurately adapted to complex and changing urban environments, significantly reducing false alarm rates. Simultaneously, intelligent analysis of multi-source data enables early prediction of potential risks, upgrading the early warning mechanism from single-threshold triggering to multi-dimensional risk prediction. The synergistic effect of the environmental perception module and the edge AI model enables early and accurate identification of core safety risks such as gas leaks and illegal operations. Combined with the tiered response and handling strategies generated by the main control module, a full-chain safety protection system is formed, significantly improving the comprehensiveness and foresight of safety protection. Through real-time sharing and collaborative analysis of multi-source sensing data, the efficiency of emergency response is significantly improved, and the overall collaborative effectiveness of cylinder safety management is comprehensively optimized. At the same time, the explosion-proof shell design ensures the stable operation of the equipment in complex operating environments, and the integrated design of each module and the main control module improves the reliability and ease of installation of the terminal. Attached Figure Description

[0023] Figure 1 This is a front view structural diagram of a multi-dimensional sensing, early warning, and handling terminal for liquefied gas cylinders according to the present invention;

[0024] Figure 2 This is a rear view structural diagram of a multi-dimensional sensing, early warning, and handling terminal for liquefied gas cylinders according to the present invention.

[0025] Figure 3 This is a cross-sectional structural diagram of the electronic module in a multi-dimensional sensing, early warning, and handling terminal for liquefied gas cylinders according to the present invention.

[0026] Figure 4 This is a schematic diagram of the structure of the expansion module in the multi-dimensional sensing, early warning and handling terminal for liquefied gas cylinders of the present invention;

[0027] Figure 5 This is a schematic diagram of the overall architecture of an IoT system for gas safety sensing according to the present invention. Detailed Implementation

[0028] The following is a detailed explanation and description of a multi-dimensional sensing, early warning, and handling terminal for liquefied gas cylinders according to the present invention, with reference to the accompanying drawings.

[0029] like Figure 1 and combined Figure 2As shown, this invention discloses a multi-dimensional sensing, early warning, and handling terminal for liquefied petroleum gas (LPG) cylinders, including an explosion-proof housing 1. The explosion-proof housing 1 is injection molded from UL94V-0 grade flame-retardant engineering plastic, and the back of the explosion-proof housing 1 is arc-shaped to conform to the curvature of the cylinder's protective handle. An electromagnetic locking mechanism 2 is provided on the back of the explosion-proof housing 1. The electromagnetic locking mechanism 2 is detachably installed at the handle of the cylinder neck. The electromagnetic locking mechanism 2 is fixed to the cylinder by four rectangularly distributed N52 grade neodymium iron boron permanent magnets 3 on the back of the explosion-proof housing 1. The mechanical locking of the electromagnetic locking mechanism 2 adopts a spring-loaded wedge-shaped locking pin structure, and the locking pin is made of 304 stainless steel. When the terminal is installed at the cylinder handle, it automatically engages with the preset holes in the handle. The explosion-proof housing 1 contains a battery, a main control module, a positioning module, a communication module, and an environmental sensing module. The main control module uses an RK1808AIoT chip with an integrated edge AI model. The positioning module is a multi-mode positioning module, using the Hi2115 chip, supporting LBS / GPS / BeiDou / Wi-Fi fingerprint fusion positioning. The communication module is a self-organizing network communication module, using the SX1302 LoRa chip. The electromagnetic locking mechanism 2, battery, positioning module, communication module, and environmental perception module are all connected to the main control module. The main control module is used to dynamically construct behavioral baselines and flexible geofences based on the deployment scenario and usage mode of the gas cylinders. It receives multi-source perception data through the communication module to complete unsupervised anomaly detection, hierarchical response, and disposal strategy generation.

[0030] This embodiment utilizes the positioning and communication modules mounted on the explosion-proof shell 1, combined with the detachable and fixed installation of the electromagnetic locking mechanism 2 at the cylinder neck handle, to achieve real-time positioning and dynamic monitoring of the cylinder's operating status during circulation. The linkage design between the electromagnetic locking mechanism 2 and the main control module effectively prevents the cylinder from escaping the scope of liquefied gas supervision, promoting a shift in supervision from passive traceability to proactive prevention. The edge AI model integrated into the main control module can dynamically construct behavioral baselines and flexible geofencing based on the actual deployment scenarios and usage patterns of the cylinder. Compared to traditional fixed-threshold geofencing technology, it can more accurately adapt to complex and changing urban environments, significantly reducing the false alarm rate. Simultaneously, intelligent analysis of multi-source data enables early prediction of potential risks, upgrading the early warning mechanism from single-threshold triggering to multi-dimensional risk prediction. The synergistic effect of the environmental perception module and the edge AI model enables early and accurate identification of core safety risks such as gas leaks and illegal operations. Combined with the graded response and handling strategies generated by the main control module, a full-chain safety protection system is formed, significantly improving the comprehensiveness and foresight of safety protection. Through real-time sharing and collaborative analysis of multi-source sensing data, the linkage efficiency of emergency response is significantly improved, and the overall collaborative efficiency of cylinder safety management is comprehensively optimized. At the same time, the design of the explosion-proof housing 1 ensures the stable operation of the equipment in complex operating environments, and the integrated design of each module and the main control module improves the reliability and ease of installation of the terminal.

[0031] To continuously monitor the stationary state of the terminal, an AH49E linear Hall sensor is installed 32mm away from the permanent magnet to detect changes in the magnetic field. When unauthorized disassembly causes a change in magnetic field strength exceeding a set threshold of 30%, the linear Hall sensor sends an interrupt signal to the main control module within 100ms. Figure 3 As shown, all components inside the explosion-proof enclosure 1 are encapsulated with Henkel Loctite EA 9466 epoxy resin 9 to form an integral protective structure.

[0032] like Figure 1 As shown, in one embodiment of the present invention, the environmental perception module includes a temperature sensor 4, a gas concentration sensor 5, a microphone 6, and an inertial measurement module. The main control module performs parallel processing and feature extraction on multi-source heterogeneous sensor data based on a temporal convolutional network to complete the identification of gas cylinder risks, including micro-leakage, overturning, abnormal handling, and violent damage of the gas cylinder.

[0033] Among them, temperature sensor 4 is an MLX90614ESF-DCI non-contact infrared temperature sensor with a measurement accuracy of ±0.5℃; gas concentration sensor 5 is an SS-138 laser scattering gas concentration sensor with a detection range of 1-10000ppm; microphone 6 is an array of four SPH0641LM4H-BMEMS microphones; and inertial measurement module is a BMI270 nine-axis inertial measurement module. The identification of gas cylinder risks can refer to the following rules: when gas concentration sensor 5 detects a continuous and slow increase in liquefied gas concentration within the set safety threshold range, and this concentration change trend, after analysis using a temporal convolutional network, excludes interference factors such as ambient temperature fluctuations and changes in ventilation conditions, and the duration reaches a preset threshold, it is determined to be a micro-leakage; when the inertial measurement module detects that the cylinder tilt angle exceeds 45° and lasts for more than 3 seconds, and the vertical acceleration value deviates from the normal stationary / handling range for a long period, it is considered to be overturned. This can also be combined with the positioning module to confirm that the cylinder is not in a normal tilted state within the designated storage area, ruling out intentional placement; when the inertial measurement module... If the peak acceleration exceeds the normal handling threshold, the attitude angle changes frequently and drastically, or the positioning module shows that the handling path deviates from the preset legal circulation range, the handling period is outside the specified working period, and the microphone 6 does not detect sound signals with violent damage characteristics, and the movement trajectory is irregular and illogical in terms of timing, it is considered abnormal handling; if the microphone 6 detects high-frequency, high-intensity impact sound, cutting sound, or cracking sound, the inertial measurement module simultaneously detects extreme impact force or sudden violent attitude change, which may be accompanied by instantaneous fluctuations of the gas concentration sensor 5, and the electromagnetic locking mechanism 2 detects force signals of illegal external prying, it is considered violent damage.

[0034] In one embodiment of the present invention, the main control module downloads a basic model from the cloud based on a federated learning framework, and fine-tunes the basic model using the FedAvg algorithm based on the typical activity trajectory of the gas cylinder, statistical gas consumption patterns, and environmental parameter changes collected by the environmental perception module, thereby constructing a behavior baseline that conforms to reality.

[0035] In this embodiment, the main control module first connects to the federated learning node of the cloud security platform and downloads a pre-trained basic model from the cloud (this model has been trained based on massive amounts of general gas cylinder scenario data and has preliminary behavioral feature recognition capabilities), avoiding the inefficiency caused by training from scratch and providing basic parameter support for the construction of personalized behavior baselines. Subsequently, the module cleans, aligns temporally, and standardizes the three types of data: typical gas cylinder activity trajectories recorded by the positioning module, gas consumption patterns statistically analyzed by the gas concentration sensor 5, and environmental parameter changes collected by the environmental perception module, forming a structured local training dataset. Then, based on the FedAvg algorithm, the main control module fine-tunes the parameters of the downloaded basic model using the locally preprocessed dataset (more specifically, it takes the location sequence of the activity trajectory, the statistical characteristics of gas consumption, and the changing trends of environmental parameters as input, calculates the error between the model's predicted output and the actual behavior of the local gas cylinder through forward propagation, and then iteratively updates the model's key parameters such as convolutional kernel weights and fully connected layer parameters through backpropagation, enabling the model to gradually learn the personalized behavioral characteristics of the gas cylinder). After local fine-tuning, the main control module uploads the updated model parameters to the cloud-based federated learning server. The cloud server collects the uploaded model parameters, calculates the weighted average using the FedAvg algorithm, and integrates the common behavioral characteristics and individual differences of different gas cylinders to generate globally optimized model parameters. The cloud then distributes the aggregated parameters to the main control module. Finally, after receiving the globally optimized parameters from the cloud, the main control module performs a second round of fine-tuning based on the latest locally collected real-time data to verify the model's adaptability to the actual scenario of the gas cylinder. If the model's accuracy in recognizing normal behavior reaches a preset threshold and can stably distinguish the gas cylinder's regular activity trajectory, reasonable consumption range, and normal environmental adaptation state, this is considered a fixed behavioral baseline. If the threshold is not reached, the above process is repeated iteratively until the model fully adapts to the actual usage scenario of the gas cylinder, ultimately forming a behavioral baseline that accurately reflects the individual behavioral patterns of the gas cylinder and conforms to the actual application scenario. This ensures that the constructed behavioral baseline has both general scenario adaptability and precise matching of the specific usage mode of a single gas cylinder, providing a reliable reference standard for subsequent anomaly detection.

[0036] In one embodiment of the present invention, the main control module uses the GraphSAGE algorithm to perform feature fusion based on the real-time location information collected by the positioning module, as well as the scenario-based dynamic data and historical risk data, to calculate the risk propagation probability within the area and output a dynamically adjusted safe distance threshold, thereby constructing an elastic geofence.

[0037] In this embodiment, the main control module first completes the collection and preprocessing of multi-dimensional input data. The input data includes: real-time location information of the gas cylinders continuously captured by the multi-mode positioning module, reflecting the cylinder's current spatial coordinates and movement trajectory; scenario-based dynamic data, covering real-time changing scenario information such as the distribution density of gas cylinders within the area, surrounding environmental characteristics (e.g., pedestrian and vehicle traffic density, population density distribution, building layout, ventilation conditions), and current meteorological parameters; and historical risk data, including records of past gas cylinder leaks, overturning, and abnormal clustering events in the area, as well as historical risk propagation paths and impact ranges. The main control module cleans and standardizes this multi-source heterogeneous data. Subsequently, the main control module constructs a regionalized graph structure model based on the GraphSAGE algorithm, using a single gas cylinder as a core node in the graph. Connection edges between nodes are constructed based on the spatial distance between cylinders and potential risk propagation relationships, while each node is assigned multi-dimensional feature attributes. Next, the GraphSAGE algorithm achieves deep feature fusion of multi-source data by aggregating the features of individual nodes with those of their neighbors. This preserves the individual location and state information of each cylinder while fully incorporating spatial relationships between cylinders within the region, dynamic impacts of the scene, and historical risk patterns, effectively uncovering hidden risk propagation correlation features within the data. Based on this, the main control module combines the fused comprehensive features with patterns such as risk propagation rate and impact range in similar scenarios from historical risk data to quantify the risk propagation probability within the current region. Finally, the main control module dynamically outputs a safe distance threshold adapted to the current situation based on the calculated regional risk propagation probability. Simultaneously, the multi-mode positioning module continuously uses the real-time location information of the cylinders as dynamic input to the GraphSAGE algorithm. The main control module iteratively adjusts the safe distance threshold based on real-time changes in location updates and risk propagation probabilities, ultimately constructing a flexible geofence capable of responding to regional risk situations in real time and adapting to the dynamic location and scene changes of cylinders, achieving intelligent and dynamic control over the activity range of cylinders.

[0038] In one embodiment of the present invention, a camera 7 is installed on the outer surface of the explosion-proof housing 1. The main control module uses a temporal convolutional network based on the attention mechanism to perform end-side fusion sensing of the video stream collected by the camera 7, the acoustic scene information collected by the microphone 6, and the spatial motion data of the inertial measurement module to complete unsupervised anomaly detection.

[0039] In this embodiment, complex scenarios such as illegal filling (e.g., unauthorized filling, overfilling) and non-standard operating procedures (e.g., rough handling, improper disassembly, failure to use according to regulations) are addressed. Time series analysis is performed on video streams, acoustic scene information, and spatial motion data respectively. Based on the time series analysis results, feature vectors for each modality are extracted. For example, for spatial motion data, motion feature vectors such as peak value, root mean square, and frequency spectral density are extracted. Feature-level fusion algorithms (e.g., splicing fusion, weighted fusion) are used to integrate the feature vectors of the three modalities into a unified multi-dimensional comprehensive feature matrix, eliminating interference from single-modal data. Finally, the VAE model is trained using multi-modal time series data from legal scenarios such as normal use and compliant filling of gas cylinders, enabling the model to learn and model the multi-modal feature distribution patterns corresponding to normal behavior. During the detection phase, the multi-dimensional comprehensive feature matrix, which is collected in real time and processed as described above, is input into the trained VAE model. The model reconstructs the input features and calculates the reconstruction error between the real-time features and the reconstructed features. If the gas cylinder is in a legal scenario, the real-time features conform to the normal distribution learned by the model, and the reconstruction error is less than the preset threshold. If there is illegal filling (such as the concentration, vibration, and sound features deviating from the normal distribution during filling) or non-standard operation (such as the acceleration peak of violent handling and the sound spectrum exceeding the normal range), the real-time features will deviate from the normal distribution, causing the reconstruction error to exceed the preset threshold. The model then judges it as an anomaly, thereby achieving unsupervised anomaly detection for the above complex scenarios, and model training and detection can be completed without relying on the labeling of abnormal samples.

[0040] In one embodiment of the present invention, the environmental perception module further includes an acceleration sensor, which is used to sense whether the gas cylinder is moving, and the main control module switches between working mode and sleep mode based on the sensing data of the acceleration sensor.

[0041] In this embodiment, the accelerometer senses both stationary and moving states. When the accelerometer detects the cylinder as stationary, the main control module switches to sleep mode; when the accelerometer detects the cylinder as moving, the main control module switches to operating mode. This mode switching improves the main control module's battery life. To further enhance the battery life, a solar panel 8 is installed on the explosion-proof housing 1. The solar panel 8 converts solar energy into electrical energy to provide emergency power to the main control module. Power supply from the solar panel 8 is achieved through a power management circuit using the BQ25895 chip. Furthermore, by learning user behavior patterns and network coverage quality, the main control module can dynamically adjust the positioning frequency, communication interval, and sensing power consumption, achieving ultra-long battery life while ensuring safe monitoring.

[0042] like Figure 4As shown, in one embodiment of the present invention, an expansion module is also included. The expansion module includes an expansion module housing 10, which is fixedly connected to the bottom of the cylinder. The expansion module housing 10 integrates an ultrasonic gas leak detection unit, a weighing sensor 12, an emergency response unit, and an expansion device circuit. The ultrasonic gas leak detection unit is used to detect liquefied gas leaks from the cylinder, the weighing sensor 12 is used to detect the weight of the cylinder, the emergency response unit is used for emergency fire extinguishing, and the expansion device circuit is connected to the ultrasonic gas leak detection unit, the weighing sensor 12, and the emergency response unit, and is also connected to the main control module for communication.

[0043] In this embodiment, the expansion module housing 10 and the explosion-proof housing 1 are made of the same flame-retardant material. The expansion module housing 10 has an M8 standard threaded hole at its bottom center, and is fixedly connected to the bottom of the gas cylinder via a stainless steel screw. Four N35 grade neodymium iron boron auxiliary positioning magnets are embedded along the edge of the expansion module housing 10. To detect the connection status between the expansion module and the cylinder bottom, a clamping switch 11 is also provided on the expansion module housing 10. The clamping switch 11 detects the contact status between the expansion module and the cylinder bottom, triggering a locking signal when the clamping force reaches a set threshold. The ultrasonic gas leak detection unit uses a UESystems UE100 sensor with a detection frequency range of 20-100kHz and a sensitivity of -20dB. The weighing sensor 12 uses a JLBS-VI strain gauge sensor with a range of 0-50kg and an accuracy of ±10g, and is connected to the expansion device circuitry via an HX711 analog-to-digital converter chip. The emergency response unit includes a 50ml 304 stainless steel pressure tank filled with 3kg of perfluorohexanone liquid extinguishing agent. The tank opening is controlled by an EBV-05 type electric explosion valve and can be used for emergency fire suppression. The expansion device circuit is responsible for sensor data acquisition, communication control, and power management. It establishes a secure communication channel with the main terminal via LoRa communication based on AES-256 encryption. The communication interval is configurable, and data is synchronized once every 1 second by default.

[0044] In one embodiment of the present invention, the main control module uses the time-series data of the weighing sensor 12 uploaded by the expansion device circuit to identify abnormal consumption patterns of the remaining amount of the gas cylinder using the Bayesian change point detection algorithm and combines it with the positioning information to construct a digital profile of the entire chain of the gas cylinder from filling to transportation to use.

[0045] In this embodiment, after the main control module connects to the weighing sensor 12 of the expansion device circuit, it acquires real-time weight time-series data (including real-time balance, rate of change, etc., recorded according to fixed timestamps) of the entire process of filling, transporting, and using the gas cylinder. At the same time, it synchronously collects the latitude and longitude, movement trajectory, and other location information of the positioning module and aligns them precisely according to the timestamps. Then, it cleans and smooths the weight data, extracts key time-series features such as the rate of change per unit time, clarifies the normal weight change patterns under different scenarios, and then inputs the preprocessed data into the Bayesian change point detection algorithm. The algorithm constructs a prior probability model based on historical data of compliant scenarios. By calculating the likelihood function and posterior probability, it locates the abnormal consumption patterns of weight mutations in each stage of filling, transporting, and using. Then, it matches the abnormal change points with the location information through timestamps, locks the specific location of the abnormality, and judges the scenario attributes, realizing a three-dimensional association of time, space, and abnormality type. Finally, the main control module integrates the data of the entire process to construct a digital portrait of the entire life cycle of the gas cylinder, which includes the characteristics of each stage of filling, transporting, and using and a summary of the status of the entire chain, achieving the purpose of full-process traceability and monitoring.

[0046] In one embodiment of the present invention, the main control module has a built-in security self-destruct protocol. When continuous illegal cracking or specific emergency commands are detected, the hardware can be self-locked and key security data can be cleared. Furthermore, after the end of the life cycle of the terminal of the present invention, its core module supports convenient disassembly and classified recycling, which is in line with the green and environmentally friendly design concept.

[0047] In one embodiment of the present invention, the hierarchical response and handling strategy of the main control module includes:

[0048] Low-risk response: Mark potential risk patterns in geofencing records and recommend corresponding safety education content through knowledge graphs;

[0049] Medium-risk response: Push out early warnings, and the decision engine generates a response plan that includes augmented reality guidance, and simultaneously notifies the regional safety officer;

[0050] High-risk response: Activate the local deterrence function of the terminal, and upload the multi-source sensing data stream from the scene to the cloud for in-depth analysis;

[0051] Emergency Response: Perform all the above operations and automatically activate the emergency self-handling unit of the extended module. At the same time, synchronize the event to the regional emergency response digital platform for collaborative scheduling and simulation.

[0052] In this embodiment, tiered prevention and control measures are used to avoid over-intervention and under-handling, minimizing emergency response time and reducing accident losses. Seamless connection between different levels of response forms a full-chain management system from prevention and early warning to handling and emergency response, ensuring timely intervention for routine risks and coordinating scheduling in extreme scenarios, thus improving the comprehensiveness and foresight of safety management. At the same time, multi-source data uploads and simulations under high-risk and critical conditions provide support for subsequent strategy optimization, enabling efficient resource scheduling through regional collaboration. Furthermore, differentiated handling strategies are adapted to the complex circulation and usage scenarios of gas cylinders, balancing safety control and normal usage needs, and avoiding misoperation or resource waste caused by a single mode.

[0053] like Figure 5 As shown, this embodiment of the invention also provides an Internet of Things (IoT) system for gas safety sensing, comprising:

[0054] Multiple terminals of any of the above embodiments are interconnected to form a terminal sensing network, which captures multi-source sensing data in real time.

[0055] By utilizing edge computing nodes to receive data from the terminal sensing network, the information uploaded by terminals within the area is aggregated and preliminarily analyzed, and the structured processing results are synchronized to the cloud security platform.

[0056] By leveraging the cloud security platform and its built-in large-scale graph computing engine, the aggregated data transmitted by edge nodes is deeply mined to identify risk propagation patterns and group anomalies in the terminal network. At the same time, the edge AI model deployed in the main control module of each terminal is continuously optimized through joint learning to improve the data analysis accuracy and response speed of the terminal perception network.

[0057] Based on MAC address signals collected by the terminal sensing network, combined with a cloud-optimized swarm intelligence algorithm, the distribution map of people indoors and outdoors is reconstructed in real time, and dynamic, escape-optimal path planning is quickly generated in emergency situations.

[0058] In this embodiment, the edge computing node uses Huawei Atlas 500 intelligent small stations, deployed in the regional central computer room, responsible for aggregating terminal data and running lightweight graph neural network models for preliminary risk prediction. The terminal transmits data through an electronic module within its explosion-proof casing 1 and utilizes solar panels 8 for sustainable power supply. The cloud-based security brain platform is deployed on Alibaba Cloud, incorporating the Spark GraphX ​​large-scale graph computing engine to mine risk propagation patterns and group anomalies in the terminal network. This platform continuously optimizes the edge AI models deployed on each terminal through a federated learning approach. During this process, it provides real-time data to the terminal's environmental perception array (including cameras 7, temperature sensors 4, gas concentration sensors 5, and microphones 6). By using MAC address signals in the terminal's perception network, it reconstructs the indoor and outdoor personnel distribution map in real time using an ant colony algorithm, providing emergency departments with dynamic and optimal escape route planning in critical situations. Specifically, this includes analyzing the MAC address signal density of all mobile devices within a 1-kilometer radius, calculating safe evacuation routes with signal densities below 0.1 devices / square meter, and pushing this information to emergency personnel's handheld terminals in real time. This embodiment forms a cross-level emergency response system through collaboration between the cloud, terminals, and edge nodes, significantly improving the ability to control accident losses; it covers the entire process of cylinder management, and through risk prediction and adaptive optimization, it achieves a transformation from passive response to proactive prevention and control, reduces the cost of manual intervention, and ultimately achieves comprehensive, real-time, intelligent, and efficient cylinder safety management.

[0059] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0060] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and simple improvements made on the substantive content of the present invention should be included within the protection scope of the present invention.

Claims

1. A liquefied gas cylinder multi-dimensional perception early warning disposal terminal, characterized in that: The system includes an explosion-proof housing with an electromagnetic locking mechanism detachably mounted on the handle of the cylinder neck. Inside the housing are a battery, a main control module, a positioning module, a communication module, and an environmental perception module. These modules are connected to the main control module, which integrates an edge AI model to dynamically construct behavioral baselines and flexible geofences based on the cylinder's deployment scenario and usage patterns. The communication module receives multi-source sensing data to perform unsupervised anomaly detection, tiered response, and response strategy generation. The environmental perception module includes a temperature sensor, a gas concentration sensor, a microphone, and an inertial measurement unit (IMU). The main control module uses a temporal convolutional network to perform parallel processing and feature extraction on multi-source heterogeneous sensor data to identify cylinder risks, including micro-leakage, tipping, abnormal handling, and violent damage. A camera is mounted on the outer surface of the explosion-proof housing. The main control module uses an attention-based temporal convolutional network to integrate the video stream from the camera, the acoustic scene information from the microphone, and the IMU data. The spatial motion data of the cylinder is fused and perceived at the edge to complete unsupervised anomaly detection. The environmental perception module also includes an accelerometer to sense whether the cylinder is moving. The main control module switches between working mode and sleep mode based on the sensing data from the accelerometer. The module also includes an expansion module, which includes an expansion module shell that is fixedly connected to the bottom of the cylinder. The expansion module shell integrates an ultrasonic gas leak detection unit, a weighing sensor, an emergency response unit, and an expansion device circuit. The ultrasonic gas leak detection unit is used to detect liquefied gas leaks in the cylinder. The weighing sensor is used to detect the weight of the cylinder. The emergency response unit is used for emergency fire extinguishing. The expansion device circuit is connected to the ultrasonic gas leak detection unit, the weighing sensor, and the emergency response unit, and communicates with the main control module. Based on the time-series data of the weighing sensor uploaded by the expansion device circuit, the main control module uses a Bayesian change point detection algorithm to identify abnormal consumption patterns of the cylinder's remaining volume and combines it with positioning information to construct a digital profile of the entire chain of the cylinder from filling to transportation to use. The hierarchical response and handling strategy of the main control module includes: Low-risk response: Mark potential risk patterns in geofencing records and recommend corresponding safety education content through knowledge graphs; Medium-risk response: Push out early warnings, and the decision engine generates a response plan that includes augmented reality guidance, and simultaneously notifies the regional safety officer; High-risk response: Activate the local deterrence function of the terminal, and upload the multi-source sensing data stream from the scene to the cloud for in-depth analysis; Emergency Response: Perform all the above operations and automatically activate the emergency response unit of the extended module. At the same time, synchronize the event to the regional emergency response digital platform for collaborative scheduling and simulation.

2. The multi-dimensional perception early warning and disposal terminal for liquefied gas cylinder according to claim 1, characterized in that: The main control module downloads a basic model from the cloud based on a federated learning framework. It then uses the typical activity trajectory of the gas cylinder, statistical gas consumption patterns, and environmental parameter changes collected by the environmental perception module to fine-tune the basic model using the FedAvg algorithm, thus constructing a behavioral baseline that conforms to reality. 3.The liquefied gas cylinder multi-dimensional perception early warning and disposal terminal according to claim 1, characterized in that: The main control module uses the GraphSAGE algorithm to perform feature fusion based on the real-time location information collected by the positioning module, as well as scenario-based dynamic data and historical risk data, to calculate the risk propagation probability within the area and output a dynamically adjusted safe distance threshold, thereby constructing an elastic geofence.

4. A gas safety aware IoT system, characterized in that, include: Multiple terminals as described in any one of claims 1-3 are interconnected to form a terminal sensing network, which captures multi-source sensing data in real time. By utilizing edge computing nodes to receive data from the terminal sensing network, the information uploaded by the terminals within the region is aggregated and preliminarily analyzed, and the structured processing results are synchronized to the cloud security platform. By leveraging the cloud security platform and its built-in large-scale graph computing engine, the aggregated data transmitted by edge nodes is deeply mined to identify risk propagation patterns and group anomalies in the terminal network. At the same time, the edge AI model deployed in the main control module of each terminal is continuously optimized through joint learning to improve the data analysis accuracy and response speed of the terminal perception network. Based on MAC address signals collected by the terminal sensing network, combined with a cloud-optimized swarm intelligence algorithm, the distribution map of people indoors and outdoors is reconstructed in real time, and dynamic, escape-optimal path planning is quickly generated in emergency situations.

Images