Tobacco new energy terminal distribution vehicle safety monitoring and early warning device based on Internet of Things

Abstract

The invention relates to the technical field of safety monitoring, in particular to a tobacco new energy terminal distribution vehicle safety monitoring and early warning device based on the Internet of Things, and the device comprises a multi-source data acquisition module which collects data and forms a data frame set with a time label; the abnormal feature extraction module is used for extracting multi-dimensional abnormal feature vectors including an environment disturbance mode, a structure state abnormal mode and a position track offset mode; the risk level evaluation module inputs the abnormal feature vector into a fuzzy decision model based on multi-factor fusion and generates a structured risk event chain, and the risk event chain is used for traceability and front-and-back correlation analysis; the linkage early warning response module is used for automatically matching a preset response rule set based on the risk level label and the risk event chain, intelligent fusion judgment of multi-source data is achieved, and a traceable basis is provided for follow-up event traceability and responsibility affirmation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of safety monitoring, and particularly relates to a tobacco new energy terminal distribution vehicle safety monitoring and early warning device based on Internet of Things. BACKGROUND

[0002] With the continuous improvement of product quality and transportation safety requirements of the tobacco industry, the traditional tobacco new energy terminal distribution vehicle management mode has been difficult to meet the current needs of fine and real-time supervision. Since tobacco products are highly sensitive to environmental factors such as temperature and humidity, gas volatility, and vibration, if there is a lack of effective monitoring and abnormal early warning mechanism during long-distance transportation, it is easy to cause product quality deterioration, and even induce safety incidents such as damage, theft, etc. In addition, most current transportation monitoring systems still rely on single sensors or fixed-point camera monitoring, which has problems such as large data coverage blind area, untimely response, low risk identification precision, etc.

[0003] In recent years, Internet of Things, edge computing and artificial intelligence technologies have been gradually applied in the field of logistics safety. Some research attempts to monitor the process of tobacco new energy terminal distribution vehicles through temperature and humidity sensors and GPS trajectory monitoring means. However, most systems only support static data collection and simple threshold alarm, and are difficult to identify multi-dimensional coupled abnormal states, lack of deep feature extraction and dynamic risk assessment capabilities for abnormal events. At the same time, in the risk early warning system driven by multiple sources of data, the response strategy is mostly static configuration, which cannot adapt to the changing transportation environment and complex abnormal scenarios, and lacks a closed-loop optimization mechanism and response efficiency self-improvement capability. SUMMARY

[0004] The present application provides a tobacco new energy terminal distribution vehicle safety monitoring and early warning device based on Internet of Things.

[0005] The tobacco new energy terminal distribution vehicle safety monitoring and early warning device based on Internet of Things comprises: A multi-source data acquisition module: through the deployment of multi-modal sensing nodes on the new energy terminal distribution vehicle, the temperature, humidity, volatile organic compound concentration, particulate matter concentration, micro-vibration amplitude, unlocking state, power voltage fluctuation, and vehicle positioning trajectory data in the cargo compartment are collected to form a data frame set with time labels; An abnormal feature extraction module: the data frame set is input into an edge computing node, and a behavior recognition algorithm based on time sequence sliding window is used to extract a multi-dimensional abnormal feature vector including environmental disturbance patterns, structural state abnormal patterns, and position trajectory deviation patterns. The abnormal feature vector includes context information when each type of data anomaly occurs; Risk level assessment module: input the abnormal feature vector into a fuzzy decision model based on multi-factor fusion, output the corresponding risk level label according to the abnormal trigger frequency, duration, interference intensity and geographical distribution law, and generate a structured risk event chain, which is used for traceability and cause-effect correlation analysis; Linkage early warning response module: based on the risk level label and risk event chain, automatically match the preset response rule set, trigger response measures including remote lock control, voice warning, path re-planning, alarm video backhaul, and at the same time, synchronize all data and response records to the supervision cloud platform to form a traceable safety record.

[0006] Optionally, the multi-modal sensing node comprises: Sensor group: composed of a net topology sensor array deployed on the side wall and ceiling of the cargo compartment, including infrared temperature sensor, capacitive humidity sensor, electrochemical VOC gas sensor, laser particulate matter counter, three-axis accelerometer, electromagnetic lock state sensor, power voltage monitoring chip and Beidou / GPS dual-mode positioning module; Data frame structure: each sensor synchronously collects data and encapsulates it into a unified data frame through a wireless protocol, the data frame contains a 16-byte header and a dynamic length payload field, and the vibration data is compressed and stored in the form of Fourier transform frequency energy spectrum.

[0007] Optionally, the multi-source data acquisition module further comprises a redundancy check mechanism, specifically comprising: When the sensor data of any area in the cargo compartment exceeds the preset threshold value for 3 times in succession, the adjacent node is automatically activated for cross verification, and is marked as a high confidence abnormal event.

[0008] Optionally, the behavior recognition algorithm of the time sequence sliding window comprises: Dynamic window length adjustment: according to the real-time vehicle speed of the on-board positioning trajectory, the sliding window length is dynamically adjusted by using linear interpolation method, the window length is 60 seconds when driving at low speed, and the window length is shortened to 20 seconds when the vehicle speed exceeds 80km / h; Space-time feature fusion: after normalizing the temperature, humidity and vibration data in the same time window, inputting the light-weight convolutional network, and outputting the space-time coupled feature vector; Context association modeling: the attention mechanism is used to calculate the association weight of the current feature vector and the similar mode in the historical data, when the attention score is lower than the score threshold, it is determined as a new type of abnormal mode and triggers the model incremental learning.

[0009] Optionally, the light-weight convolutional network comprises 3 layers of 1D convolution kernel and maximum pooling layer, the space-time coupled feature vector is spliced with the Haversine distance change rate of the positioning trajectory to form a multi-dimensional abnormal feature vector.

[0010] Optionally, the execution logic of the multi-factor fusion fuzzy decision model comprises: Fuzzification processing layer: map the abnormal trigger frequency, duration, and interference intensity to Gaussian membership functions respectively, and output the fuzzy membership values of each factor; Dynamic weight allocation: calculate the information entropy of each factor based on the entropy weight method, and dynamically adjust the weight proportion, wherein the geographic distribution weight is positively correlated with the degree of remoteness of the route; Risk event chain generation: after associating the risk level label with the original data, an association matrix including timestamp, latitude and longitude coordinates, abnormal type code, and associated sensor ID is constructed, and the event causal chain is analyzed to mark the key trigger nodes and their confidence.

[0011] Optionally, the linkage early warning response module comprises: Response rule adaptive optimization: according to the manual confirmation results in the history alarm, update the rule priority by using Q-learning algorithm, so that the trigger weight of high-frequency effective rules is improved; Multi-level response mechanism: Primary response: trigger the sound and light warning in the cargo compartment, and upload the compressed sensor data segment to the cloud; Secondary response: forcibly start the vehicle-mounted camera to record video for a predetermined time, and return the key frames through the 5G slice network; Tertiary response: start the electronic lock device of the cargo compartment, and send the SOS signal including encrypted positioning information to the supervision platform at the same time; Emergency power switching: when it is detected that the main power voltage fluctuation exceeds ±15%, automatically switch to the super capacitor backup power source to ensure at least 30 minutes of key data acquisition and transmission capability.

[0012] Optionally, the normalized data of temperature, humidity, and vibration data in the same time window are arranged in time sequence to form an input matrix , wherein is the number of sampling points in the time window, and the 3 columns correspond to temperature, humidity, and vibration data respectively.

[0013] Optionally, the network structure of the lightweight convolutional network comprises 3 layers of 1D convolution kernels and maximum pooling layers, specifically as follows: First layer of convolution: Input: normalized data matrix , the convolution kernel is 16 1D convolution kernels, the kernel size is 5, the step is 1, and the activation function is ReLU; Output feature map ; First layer of maximum pooling: The pooling window size is 2, and the step is 2. output the pooled feature map ; second layer convolution: The convolution kernel is 32 1D convolution kernels, the kernel size is 3, the step is 1, and the activation function is ReLU; output the feature map ; second layer maximum pooling: The size of the pooling window is 2, and the step is 2; output the pooled feature map ; third layer convolution: The convolution kernel is 64 1D convolution kernels, the kernel size is 3, the step is 1, and the activation function is ReLU; output the feature map ; global maximum pooling: The pooling method includes global maximum pooling on the time dimension, and the maximum value of each channel is extracted; output: spatio-temporal coupling feature vector .

[0014] The beneficial effects of the present application are: The present application integrates infrared temperature, electrochemical gas, laser particulate matter, micro-vibration, electromagnetic lock state, power voltage, dual-mode positioning and other multi-modal sensors to construct a mesh topology sensing array covering the entire area of the cargo compartment. It can realize real-time sensing of potential risk factors such as environmental disturbance, structural anomaly and position deviation in complex transportation environment, and support adjusting the feature extraction rhythm according to the real-time running state of the vehicle by combining the time sequence sliding window behavior recognition algorithm with dynamic window length adjustment, to ensure the consistency of anomaly recognition in low-speed parking and high-speed driving.

[0015] The present application proposes a multi-factor fusion risk assessment model based on fuzzy logic and entropy weight dynamic allocation, which can comprehensively quantify and evaluate various anomalies by combining abnormal trigger frequency, interference intensity, duration and geographical distribution law, output risk level labels and generate structured risk event chains, clearly describe the event occurrence time, spatial coordinates, anomaly type and source node, and through multi-source data intelligent fusion and structured risk event chain generation, not only realize real-time risk assessment, but also provide full-process traceable basis for event traceability, responsibility identification and management optimization, significantly improve the intelligentization of supervision and the transparency of processing. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to make the technical solutions in the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced. Obviously, the accompanying drawings in the following description only aim at the present application, and for those skilled in the art, other drawings can also be obtained without creative effort.

[0017] Figure 1 The schematic diagram of each module of the device of the embodiment of the present application. DETAILED DESCRIPTION

[0018] The present application will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that in order to make the embodiments more detailed, the following embodiments are the best, preferred embodiments, and other alternative ways can also be implemented by those skilled in the art for some known technologies; and the drawings are only used to describe the embodiments in more detail, and are not intended to specifically limit the present application.

[0019] It should be noted that in the specification, "one embodiment", "embodiment", "exemplary embodiment", "some embodiments" and the like indicate that the described embodiments can include a specific feature, structure or property, but not necessarily every embodiment includes the specific feature, structure or property. In addition, when a specific feature, structure or property is described in combination with an embodiment, it should be within the knowledge of those skilled in the art to realize this feature, structure or property in combination with other embodiments (whether or not explicitly described).

[0020] Generally, the terms can be understood at least in part from the use in context. For example, depending at least in part on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular or can be used to describe combinations of features, structures, or characteristics that are combinable into one or more instances. In addition, the term "based on" can be understood as not necessarily intended to convey a set of exclusive factors, but can instead, depending at least in part on the context, allow for the existence of other factors that are not necessarily explicitly described.

[0021] As shown in Figure 1 The Internet of Things-based tobacco new energy terminal distribution vehicle safety monitoring and early warning device comprises: A multi-source data acquisition module acquires data frames with time labels by deploying multi-modal sensing nodes on the tobacco new energy terminal distribution vehicle, including temperature, humidity, volatile organic compound concentration, particulate matter concentration, micro-vibration amplitude, unlocking state, power voltage fluctuation, and vehicle positioning trajectory data in the cargo compartment; Anomaly feature extraction module: input data frame set into edge computing node, adopt behavior recognition algorithm based on time sequence sliding window, extract multi-dimensional anomaly feature vector including environmental disturbance pattern, structural state anomaly pattern and position trajectory deviation pattern, anomaly feature vector includes context information when various data anomalies occur; Risk level assessment module: input anomaly feature vector into fuzzy decision model based on multi-factor fusion, output corresponding risk level label according to anomaly trigger frequency, duration, interference intensity and geographical distribution law, and generate structured risk event chain, risk event chain is used for traceability and cause-effect correlation analysis; Linkage early warning response module: based on risk level label and risk event chain, automatically match preset response rule set, trigger response measures including remote lock control, voice warning, path re-planning, alarm video return, and synchronize all data and response records to the supervision cloud platform to form a traceable security record file; Among them, the data frame set is the only input source of anomaly feature extraction; the anomaly feature vector is the only feature input of risk level assessment; the risk level label and the risk event chain are the trigger basis of linkage response, which ensures that the logical dependency between steps is clear and the output is effectively applied.

[0022] Multi-modal sensing node includes: Sensor group: composed of a network topology sensor array deployed on the side wall and ceiling of the cargo compartment, including infrared temperature sensor, capacitive humidity sensor, electrochemical VOC gas sensor, laser particulate matter counter, three-axis accelerometer, electromagnetic lock state sensor, power voltage monitoring chip and Beidou / GPS dual-mode positioning module; Data frame structure: each sensor synchronously collects data and encapsulates it into a unified data frame through a wireless protocol, the data frame contains a 16-byte header (including timestamp, device ID, CRC check code) and a dynamic length load field, where vibration data is compressed and stored in the form of Fourier transform frequency energy spectrum.

[0023] The multi-source data acquisition module also includes a redundancy checking mechanism, specifically including: When the sensor data of any area in the cargo compartment exceeds the preset threshold for 3 consecutive times, automatically activate the adjacent node for cross verification, and mark it as a high confidence anomaly event.

[0024] The time sequence sliding window behavior recognition algorithm specifically includes: Dynamic window length adjustment: according to the real-time vehicle speed of the vehicle positioning trajectory, the sliding window length is dynamically adjusted using linear interpolation method, the window length is 60 seconds when driving at low speed, and is shortened to 20 seconds when the vehicle speed exceeds 80km / h; Space-time feature fusion: After normalizing the temperature, humidity, and vibration data within the same time window, input the lightweight convolutional network to output the space-time coupled feature vector. Context association modeling: Use the attention mechanism to calculate the association weight of the current feature vector and similar patterns in historical data. When the attention score is lower than the score threshold, it is determined as a new type of abnormal pattern and triggers model incremental learning.

[0025] The lightweight convolutional network includes 3 layers of 1D convolution kernel and max pooling layer. The space-time coupled feature vector is concatenated with the Haversine distance change rate of the positioning track to form a multi-dimensional abnormal feature vector.

[0026] The execution logic of the multi-factor fusion fuzzy decision model includes: Fuzzy processing layer: Map the abnormal trigger frequency, duration, and interference intensity to the Gaussian membership function respectively to output the fuzzy membership value of each factor; Dynamic weight allocation: Calculate the information entropy of each factor based on the entropy weight method, and dynamically adjust its weight proportion. The geographical distribution weight is positively correlated with the remoteness of the route; Risk event chain generation: After associating the risk level label with the original data, construct an association matrix including timestamp, latitude and longitude coordinates, abnormal type code, and associated sensor ID, and analyze the event causal chain to mark the key trigger node and its confidence.

[0027] The linkage early warning response module specifically includes: Response rule adaptive optimization: According to the historical alarm manual confirmation results (such as false alarm rate and response delay time), update the rule priority using Q-learning algorithm to increase the trigger weight of high-frequency effective rules; Multi-level response mechanism: First-level response (corresponding to low-risk events): Trigger the sound and light warning in the cargo compartment, and upload the compressed sensor data segment to the cloud; Second-level response (corresponding to medium-risk events): Force the vehicle-mounted camera to record video for a predetermined time, and return the key frames through the 5G slice network; Third-level response (corresponding to high-risk events): Start the cargo compartment electronic lock device, and send the SOS signal including encrypted positioning information to the supervision platform; Emergency power switching: When the main power voltage fluctuation exceeds ±15%, automatically switch to the super capacitor backup power source to ensure at least 30 minutes of key data acquisition and transmission capability.

[0028] The normalized data of temperature, humidity, and vibration within the same time window are arranged in time sequence to form an input matrix , where The number of sampling points in the time window, 3 columns correspond to temperature, humidity and vibration data respectively.

[0029] The network structure of the lightweight convolutional network comprises 3 layers of 1D convolution kernels and maximum pooling layers, and specifically as follows: First layer of convolution: Input: normalized data matrix The convolution kernel is 16 1D convolution kernels, the kernel size is 5, the step is 1, and the activation function is ReLU; Output feature map ; First layer of maximum pooling: The pooling window size is 2, and the step is 2; Output pooled feature map ; Second layer of convolution: The convolution kernel is 32 1D convolution kernels, the kernel size is 3, the step is 1, and the activation function is ReLU; Output feature map ; Second layer of maximum pooling: The pooling window size is 2, and the step is 2; Output pooled feature map ; Third layer of convolution: The convolution kernel is 64 1D convolution kernels, the kernel size is 3, the step is 1, and the activation function is ReLU; Output feature map ; Global maximum pooling: The pooling mode comprises global maximum pooling on the time dimension, and the maximum value of each channel is extracted; Output: spatiotemporal coupling feature vector .

[0030] The present application encompasses any substitutions, modifications, equivalent methods and solutions made on the essence and scope of the present application. In order for the public to have a thorough understanding of the present application, specific details are described in the following preferred embodiments of the present application, and the present application can also be fully understood without the description of these details to those skilled in the art. In addition, in order to avoid unnecessary confusion to the essence of the present application, well-known methods, processes, procedures, elements and circuits, etc. are not described in detail.

[0031] The above is only the preferred embodiment of the present application, and it should be pointed out that for ordinary skilled in the art, without departing from the principle of the present application, a number of improvements and refinements can be made, and these improvements and refinements should be regarded as the protection scope of the present application.

Claims

1. A safety monitoring and early warning device for tobacco new energy terminal distribution vehicles based on the Internet of Things, characterized in that, The application relates to a multi-source data acquisition module, an abnormal feature extraction module, a risk level evaluation module and a linkage early warning response module. The multi-source data acquisition module comprises: a multi-modal sensor node arranged on a tobacco new energy terminal distribution vehicle, which collects data including temperature, humidity, volatile organic compound concentration, particulate matter concentration, micro-vibration amplitude, unlocking state, power voltage fluctuation and vehicle positioning track data in a cargo compartment, and forms a time-labeled data frame set; The abnormal feature extraction module inputs the data frame set into an edge computing node, adopts a behavior recognition algorithm based on a time sequence sliding window, extracts a multi-dimensional abnormal feature vector including an environmental disturbance mode, a structural state abnormal mode and a position track deviation mode, and the abnormal feature vector comprises context information when various data abnormities occur; The risk level evaluation module inputs the abnormal feature vector into a fuzzy decision model based on multi-factor fusion, outputs a corresponding risk level label according to abnormal trigger frequency, duration, interference intensity and geographical distribution law, and generates a structured risk event chain, which is used for traceability and cause-effect correlation analysis; The linkage early warning response module automatically matches a preset response rule set based on the risk level label and the risk event chain, triggers response measures including remote lock control, voice warning, path re-planning and alarm video return, and synchronizes all data and response records to a supervision cloud platform to form a traceable safety record file. 2.The IoT-based safety monitoring and early warning device for tobacco new energy terminal distribution vehicles according to claim 1, characterized in that, The multi-modal sensor node comprises: a sensor group: a net topology sensor array arranged on the side wall and the ceiling of the cargo compartment, comprising an infrared temperature sensor, a capacitive humidity sensor, an electrochemical VOC gas sensor, a laser particulate matter counter, a three-axis accelerometer, an electromagnetic lock state sensor, a power voltage monitoring chip and a Beidou / GPS dual-mode positioning module; a data frame structure: each sensor synchronously collects data and encapsulates the data into a unified data frame through a wireless protocol, the data frame comprises a 16-byte header and a dynamic length load field, and vibration data is compressed and stored in the form of Fourier transform frequency energy spectrum. 3.The IoT-based safety monitoring and early warning device for tobacco new energy terminal distribution vehicles according to claim 2, characterized in that, The multi-source data acquisition module further comprises a redundancy check mechanism, specifically comprising: When sensor data in any area in the cargo compartment exceeds a preset threshold value for three times in succession, adjacent nodes are automatically activated for cross verification, and a high-confidence abnormal event is marked. 4.The IoT-based safety monitoring and early warning device for tobacco new energy terminal distribution vehicles according to claim 1, characterized in that, The behavior recognition algorithm based on the time sequence sliding window specifically comprises: Dynamic window length adjustment: according to the real-time vehicle speed of the vehicle positioning track, a linear interpolation method is adopted to dynamically adjust the sliding window length, the window length is 60 seconds when the vehicle is running at low speed, and the window length is shortened to 20 seconds when the vehicle speed exceeds 80 km / h; spatiotemporal feature fusion: after the temperature, humidity and vibration data in the same time window are normalized, the data are input into a lightweight convolution network, and a spatiotemporal coupling feature vector is output; context correlation modeling: an attention mechanism is adopted to calculate the correlation weight of the current feature vector and a similar mode in historical data, when the attention score is lower than a score threshold value, a new abnormal mode is determined and model incremental learning is triggered. 5.The IoT-based safety monitoring and early warning device for tobacco new energy terminal distribution vehicles according to claim 4, characterized in that, The lightweight convolution network comprises three 1D convolution kernels and a maximum pooling layer, the spatiotemporal coupling feature vector is spliced with the Haversine distance change rate of the positioning track to form a multi-dimensional abnormal feature vector. 6.The IoT-based safety monitoring and early warning device for tobacco new energy terminal distribution vehicles according to claim 1, characterized in that, The execution logic of the multi-factor fused fuzzy decision model comprises: Fuzzification processing layer: map the abnormal trigger frequency, duration, and interference intensity to Gaussian membership functions respectively, and output the fuzzy membership values of each factor; Dynamic weight allocation: calculate the information entropy of each factor based on the entropy weight method, and dynamically adjust the weight proportion, where the geographic distribution weight is positively correlated with the remoteness of the route; Risk event chain generation: after associating the risk level label with the original data, construct an association matrix including timestamp, latitude and longitude coordinates, abnormal type code, and associated sensor ID, and analyze the event causal chain to mark the key trigger nodes and their confidence. 7.The IoT-based safety monitoring and early warning device for tobacco new energy terminal distribution vehicles according to claim 1, characterized in that, The linkage early warning response module specifically comprises: Response rule adaptive optimization: according to the manual confirmation results in the history alarm, update the rule priority using the Q-learning algorithm to improve the trigger weight of high-frequency effective rules; Multi-level response mechanism: First-level response: trigger the sound and light warning in the cargo compartment, and upload the compressed sensor data segment to the cloud; Second-level response: forcibly start the vehicle-mounted camera to record video for a predetermined time, and return the key frames through the 5G slice network; Third-level response: start the electronic lock device of the cargo compartment, and send the SOS signal including encrypted positioning information to the supervision platform; Emergency power switching: when the main power voltage fluctuation is detected to be more than ±15%, automatically switch to the super capacitor backup power source to ensure at least 30 minutes of key data acquisition and transmission capability. 8.The IoT-based safety monitoring and early warning device for tobacco new energy terminal distribution vehicles according to claim 5, characterized in that, The normalized data of temperature, humidity and vibration data in the same time window are arranged in time sequence to form an input matrix , wherein is the number of sampling points in the time window, and the 3 columns correspond to temperature, humidity and vibration data, respectively. 9.The IoT-based safety monitoring and early warning device for tobacco new energy terminal distribution vehicles according to claim 8, characterized in that, The network structure of the lightweight convolutional network comprises 3 layers of 1D convolution kernels and maximum pooling layers, specifically as follows: First layer convolution: Input: normalized data matrix Convolution kernel is 16 1D convolution kernels, kernel size is 5, step is 1, and activation function is ReLU; Output feature map ; First layer maximum pooling: The pooling window size is 2, and the step is 2; output the pooled feature map ; Second layer convolution: The convolution kernel is 32 1D convolution kernels, the kernel size is 3, the step is 1, and the activation function is ReLU; Output feature map ; Second layer maximum pooling: The pooling window size is 2, and the step is 2; output the pooled feature map ; Third layer convolution: The convolution kernel is 64 1D convolution kernels, the kernel size is 3, the step is 1, and the activation function is ReLU; Output feature map ; Global maximum pooling: The pooling method includes global maximum pooling on the time dimension to extract the maximum value of each channel; Output: spatiotemporal coupling feature vector .

Images