Carrying safety monitoring method and system based on multiple sensors

Abstract

The invention discloses a carrying safety monitoring method and system based on multiple sensors, and belongs to the technical field of carrying safety monitoring, and the method comprises the steps: constructing three state analysis units, carrying out the data collection of the states of contents, a driver and a vehicle of an aproll logistics vehicle, setting an abnormality judgment standard, and carrying out the monitoring of the state of the aproll logistics vehicle; constructing a content anomaly data set, a driver anomaly data set and a vehicle anomaly data set; three types of abnormal data are extracted, an abnormal feature matrix is constructed after normalization, a heterogeneous risk association graph is constructed by adopting a graph neural network, historical co-occurrence frequency, causal association strength and time correlation among abnormal nodes are dynamically calculated, and edge weights are generated; and calculating a real-time comprehensive risk value based on a dynamic weighted fusion algorithm, and judging a transportation safety risk level according to a preset risk threshold, thereby triggering a hierarchical early warning mechanism.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle safety monitoring, and specifically to a vehicle safety monitoring method and system based on multi-sensors. Background Art

[0002] With the continuous development of the logistics industry and the improvement of market demand, as an important transportation tool, the safety problems faced by bulk water logistics vehicles during transportation have become increasingly prominent. During long-distance transportation, bulk water logistics vehicles are usually involved in the transportation of hazardous chemicals, liquids or other perishable items. Any problem during transportation may lead to serious consequences, including cargo loss, casualties and potential harm to the environment. Although traditional monitoring means can, to a certain extent, deal with some potential safety hazards, they often cannot comprehensively and effectively ensure transportation safety in the face of a diversified and complex transportation environment.

[0003] Traditional safety monitoring systems mainly rely on means such as visual monitoring, single vehicle status alarms or manual inspections. These monitoring means have strong limitations. For example, visual monitoring systems can only monitor the obvious states inside and outside the vehicle and cannot comprehensively reflect various hidden risks during vehicle operation; while simple vehicle status alarm systems can only give an alarm when an abnormality occurs and often do not take into account the dynamic changes during transportation, making it difficult to achieve early warning and real-time monitoring. Therefore, a single monitoring means often cannot comprehensively cover the potential dangers during transportation. Especially during long-distance transportation, problems such as driver fatigue driving, vehicle mechanical failures, and cargo leakage occur frequently, posing great potential hazards to transportation safety. Summary of the Invention

[0004] The purpose of the present invention is to provide a vehicle safety monitoring method and system based on multi-sensors to solve the problems raised in the above background art.

[0005] To solve the above technical problems, the present invention provides the following technical solution: A vehicle safety monitoring method based on multi-sensors, the method comprising: Step S100: Construct three state analysis units to monitor the contents, driver and vehicle status of the bulk water logistics vehicle respectively; The first state analysis unit uses a variety of sensors to collect the liquid level, pressure, temperature and leakage signals of the tank contents, generates a state data set and sets abnormal determination criteria, compares the data in real time to output an abnormal determination result, and constructs a content abnormal data set; Step S200: The second state analysis unit installs sensors in the cab to collect the driver's facial features, steering wheel grip force, grip duration and heart rate data, constructs a driver state data set; By setting multi-condition fatigue level determination rules, combining real-time data with determination conditions, outputs a fatigue level result, and constructs a driver abnormal data set; Step S300: The third - state analysis unit installs sensors on the vehicle to collect data on tire pressure, vibration frequency, braking response time, and engine temperature, generates a vehicle operating - state data set, and sets abnormal indicators; by comparing the data in real - time, obtains an abnormal determination result, and constructs a vehicle abnormal - data set; Step S400: Extract the abnormal data from the abnormal - data sets of the contents, driver, and vehicle, construct an abnormal - feature matrix after normalization processing, use a graph neural network to construct a heterogeneous risk - association graph, dynamically calculate the historical co - occurrence frequency, causal - association strength, and time correlation between abnormal nodes to generate edge weights; calculate the real - time comprehensive risk value based on the dynamic weighted - fusion algorithm, and judge the transportation safety risk level according to the preset risk threshold, thereby triggering a hierarchical early - warning mechanism.

[0006] Further, the step S100 includes: Step S101: The first - state analysis unit installs a liquid - level sensor on the tank body of the watering logistics vehicle to measure the liquid level. By transmitting and receiving ultrasonic signals, calculates the liquid - level height according to the signal round - trip time; installs a pressure sensor to monitor the pressure inside the tank body; installs a thermistor - type temperature sensor to collect the temperature data of the contents in real - time; installs an electrochemical leakage sensor, and once a leakage is detected, immediately records the leakage signal S, where S = 1 indicates leakage and S = 0 indicates no leakage; sets the liquid - level data number as D L , the pressure - data number as D P , the temperature - data number as D T , and the leakage - signal number as D S ; Step S102: Sets the sampling frequency of the liquid - level sensor as f L , and collects the liquid - level data once every 1 / f L time; the sampling frequency of the pressure sensor is f P , the sampling frequency of the temperature sensor is f T , the leakage sensor is in a real - time monitoring state and immediately records once a state change is detected; the collected data is first filtered using the Kalman - filtering algorithm to remove noise interference; indexed by time t, the processed data is stored in chronological order to generate a contents - state data set D1={(D L ,L t ),(D P ,P t ),(D T ,T t ),(D S ,S t )}, where L t , P t , T t , S tThey are the liquid level, pressure, temperature, and leakage signal data corresponding to time t; according to the safety standards for transported goods, the liquid level safety threshold range is set as [Lmin, Lmax], the pressure safety threshold range is [Pmin, Pmax], and the temperature safety threshold range is [Tmin, Tmax]; When it satisfies , , , S t = 1, any one of the conditions, it is determined that the state of the content is abnormal; at the same time, record the abnormal occurrence time t y , as well as the abnormal type and the corresponding data number, and construct the content abnormal data set E1; Furthermore, the step S200 includes: Step S201: The second state analysis unit installs a vision sensor in the cab to collect the driver's facial image, extracts facial expression and eye opening degree features, and forms facial feature data I; embeds a pressure sensor array on the steering wheel to measure the steering wheel grip force F and grip time w; uses a photoelectric heart rate sensor, worn on the driver's wrist, to obtain heart rate data HR by detecting the photoplethysmogram; add numbers to each data, the facial feature data number is D I , the steering wheel grip force data number is D F , the grip time data number is D w , the heart rate data number is D HR ; integrate these data to construct the driver state data set D2 = {(D I , I t ), (D F , F t ), (D w , w t ), (D HR , HR t )}, where I t , F t , w t , HR t are the facial features, steering wheel grip force, grip time, and driver heart rate data corresponding to time t respectively; Step S202: Determine the fatigue driving level using a comprehensive evaluation algorithm; Condition 1: When the eye opening degree is lower than the set threshold E0 and the duration exceeds t1 seconds; Condition 2: The number of times the steering wheel grip force F is lower than the minimum grip force threshold F0 within a period T1 exceeds n times; Condition 3: The driver's heart rate HR exceeds the normal range [HRmin, HRmax]; Based on the above three conditions, set the fatigue driving level: Meeting any one of the above conditions is determined as mild fatigue driving, and record the fatigue level number A1; Meeting two of the above conditions is determined as moderate fatigue driving, and record the fatigue level number A2; Meeting all three of the above conditions is determined as severe fatigue driving, and record the fatigue level number A3; At the same time, record the fatigue level change time t p , construct the driver abnormal data set E2.

[0007] Further, the step S300 includes: Step S301: The third state analysis unit installs a tire pressure sensor inside the vehicle tire to monitor the tire pressure data Pt in real time; installs a vibration sensor on the vehicle chassis to collect vibration frequency data V; installs a pressure sensor and a time sensor in the braking system to measure the braking response time tb; installs a temperature sensor on the engine to measure the engine temperature Te; add numbers to each data, the tire pressure data number is D Pt , the vibration frequency data number is D V , the braking response time data number is D tb , the engine temperature data number is D Te ; Step S302: Set the sampling frequency of each sensor, perform denoising processing on the collected data, and store the processed data in chronological order to generate the vehicle operation state data set D3 = {(D Pt , Pt t ), (D V , V t ), (D Te , Te t ), (D tb , tb t )}; where Pt t , V t , Te t , tb t respectively represent the tire pressure, vibration frequency, braking response time, and engine temperature data corresponding to the t moment; according to the vehicle safety standard, set the safety threshold range, and if the vehicle state data collected in real time exceeds the safety range, it is regarded as the vehicle state being abnormal; record the abnormal occurrence time t c and the abnormal type and the corresponding data number, and construct the vehicle abnormal data set E3.

[0008] Furthermore, the step S400 includes: Step S401: extract abnormal data from abnormal data sets E1, E2 and E3; for the data in E1 and E3, extract the abnormal occurrence time, abnormal type and corresponding data number and data value; similarly extract the fatigue level change time, fatigue level number and related data number and data value from E2; normalize the extracted abnormal data to construct an abnormal feature matrix M in a unified format, where the rows of the matrix represent different abnormal events, and the columns represent the abnormal type, abnormal occurrence time, data number and normalized data value, and fill all abnormal records into the matrix M in chronological order; Step S402: Use graph neural network to construct heterogeneous risk association graph, map the abnormal features represented by each row of data in the abnormal feature matrix M into graph nodes, and dynamically calculate the edge weights using the following formula: ij =(α·C ij +β·H ij ) / (γ·T ij ), where i and j represent abnormal nodes in the knowledge graph, C ij represents the historical co-occurrence frequency of node i and node j, H ij represents the causal relationship strength between node i and node j, T ij represents the temporal correlation between node i and node j; α, β, and γ are the corresponding adaptive weight coefficients; The C ij By analyzing historical transportation data, count the number of times the anomalies represented by node i and node j have appeared simultaneously in the past ij , and the total number of transport times C total , then C ij =count ij / C total ; The H ij Traverse the historical transportation data and extract all the historical transportation fragment information related to the anomalies represented by nodes i and j as the basic sample library; calculate the conditional probability P(j|i) that node j will be abnormal when node i is abnormal, and at the same time, calculate the probability P(j) that node j will appear in the overall sample; introduce the Bayesian network model, and according to the calculated conditional probability P(j|i) and prior probability P(j), combined with the conditional independence assumption in the network structure, through the probability reasoning and propagation mechanism, infer the possibility of causal relationship between nodes i and j, and map it to the [0,1] interval through normalization; The T ij Calculate the absolute value of the time difference between the abnormal occurrence of node i and node j |t i -t j | and through a time correlation function G(|ti -t j Map the result to the interval [0, 1], T ij =G(|t i -t j |); The time correlation function is , where λ represents the weight adjustment coefficient, λ ∈ [0.1, 0.5]; According to the calculated edge weights, connect each node to construct a heterogeneous risk association graph, and display the association strength between different anomalies.

[0009] Furthermore, the step S400 further includes: Step S403: Design a dynamic weighted fusion algorithm to calculate the real-time comprehensive risk value; First, determine the real-time weight w k (t) at time t for each analysis unit, where k = 1, 2, 3, representing the first state analysis unit, the second state analysis unit, and the third state analysis unit respectively; The weight is dynamically adjusted in real time according to the task type of the transportation task, road conditions, and vehicle status; For the m-th anomaly index in the k-th type of analysis unit, calculate the standardized score S km (t); It is realized through a preset scoring function S, S km (t) = S(abnormal data value, abnormal type, normal range); First, set the basic influence weight of this anomaly on transportation safety according to the abnormal type; The setting of the basic influence weight is as follows: By collecting the historical data of vehicle operation, set a basic weight for each discovered abnormal type and store it in the weight setting table. When a new abnormal situation is detected through real-time data, automatically retrieve the weight setting table to find the corresponding abnormal type and its basic weight; For new abnormal types that have never appeared in the historical data, they will be recorded and prompted for relevant personnel to set; Then compare the abnormal data value with the normal range, calculate the degree of deviation of the abnormal data value from the normal range, and adjust the proportion in the scoring according to the degree of deviation. The greater the degree of deviation, the higher the proportion in the scoring; Combine the degree of deviation of the abnormal data value from the normal range with the basic influence weight of the abnormal type, and calculate the preliminary score through a weighted formula; Finally, normalize the preliminary score and map the score to a specific numerical interval [0, 1]; Step S404: According to the formula: R(t)=∑ 3 k=1 (w k (t)·∑ nk m=1 S km (t)); where w k (t) is the real-time weight of each analysis unit at time t; S kmLet \(R(t)\) be the standardized score of the \(m\)-th abnormal index in the \(k\)-th type of analysis unit at time \(t\); \(n_k\) represents the number of abnormal indexes in the \(k\)-th type of analysis unit; according to the preset risk thresholds \(R_1\) and \(R_2\), the transportation risk levels are divided, including low-risk, medium-risk and high-risk states; if \(R(t) \lt R_1\), it is determined as a low-risk state, and a primary warning is given, and the abnormal type and abnormal data are displayed in a pop-up window in the cockpit; if \(R_1\leq R(t) \lt R_2\), it is determined as a medium-risk state, an alarm is issued and the safe stopping point is automatically calculated, and a safe driving path is planned; if \(R(t)\geq R_2\), it is determined as a high-risk state, and at this time, an emergency rescue process is automatically triggered. First, a safe driving path and a stopping point are planned, and relevant rescue departments are notified to report the vehicle position, the state of the tank contents, the abnormal type and the current risk level.

[0010] A multi-sensor-based vehicle safety monitoring system, the system includes a contents state monitoring module, a driver behavior monitoring module, a vehicle operation monitoring module, and a dynamic risk assessment module; The contents state monitoring module uses a variety of sensors to collect the liquid level, pressure, temperature and leakage signals of the tank contents, generates a state data set and sets abnormal determination criteria, compares the data in real time to output an abnormal determination result, and constructs a contents abnormal data set; The driver behavior monitoring module collects the driver's facial features, steering wheel grip force, grip duration and heart rate data, and constructs a driver state data set; by setting multi-condition fatigue level determination rules, combining real-time data with determination conditions, the fatigue level result is output, and a driver abnormal data set is constructed; The vehicle operation monitoring module collects tire pressure, vibration frequency, braking response time and engine temperature data, generates a vehicle operation state data set and sets abnormal indexes; by comparing the data in real time, an abnormal determination result is obtained, and a vehicle abnormal data set is constructed; The dynamic risk assessment module is used to extract abnormal data from the contents, driver and vehicle abnormal data sets, construct an abnormal feature matrix after normalization processing, use a graph neural network to construct a heterogeneous risk association map, dynamically calculate the historical co-occurrence frequency, causal association strength and time correlation between abnormal nodes, and generate edge weights; based on the dynamic weighted fusion algorithm, the real-time comprehensive risk value is calculated, and the transportation safety risk level is judged according to the preset risk threshold, so as to trigger a hierarchical warning mechanism.

[0011] Compared with the prior art, the beneficial effects achieved by the present invention are: By constructing three independent state analysis units, the present invention realizes the full-dimensional real-time monitoring of the contents, driver and vehicle states of the bulk water logistics vehicle. Compared with the traditional single-sensor monitoring scheme, the coverage and accuracy of transportation safety monitoring are significantly improved; The present invention constructs a heterogeneous risk association graph using a graph neural network, breaking through the limitation of traditional monitoring methods that only issue independent anomaly alarms. By dynamically calculating the historical co-occurrence frequency, causal association strength, and temporal correlation between abnormal nodes, it can reveal potential risk propagation paths, achieve the leap of risk prediction from single events to associated evolution, and provide more forward-looking decision-making support for risk early warning; Through a dynamic weighted fusion algorithm combined with a hierarchical early warning mechanism, the present invention dynamically adjusts the weights of each analysis unit according to real-time road conditions, vehicle status, etc., and uses a standardized scoring function to quantify the risk level, realizing the objectivity and dynamics of risk level assessment, and effectively improving the intelligent level of vehicle safety monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, and do not constitute a limitation to the present invention. In the drawings: Figure 1 is a method flow chart of a vehicle safety monitoring method based on multi-sensors. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0013] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0014] Please refer to Figure 1 , the present invention provides a technical solution: a vehicle safety monitoring method based on multi-sensors, the method comprising: Step S100: Construct three state analysis units to monitor the content, driver, and vehicle status of the bulk water logistics vehicle respectively; the first state analysis unit uses a variety of sensors to collect the liquid level, pressure, temperature, and leakage signals of the tank content, generates a state data set and sets an abnormal determination criterion, and compares the data in real time to output an abnormal determination result, constructing a content abnormal data set; Step S200: The second state analysis unit installs sensors in the cab to collect the facial features, steering wheel grip force, grip duration, and heart rate data of the driver, constructs a driver state data set; by setting multi-condition fatigue level determination rules, combining real-time data with the determination conditions, outputs a fatigue level result, and constructs a driver abnormal data set; Step S300: The third - state analysis unit installs sensors on the vehicle to collect data on tire pressure, vibration frequency, braking response time, and engine temperature, generates a vehicle operation - state data set, and sets abnormal indicators; by comparing the data in real - time, obtains an abnormal determination result, and constructs a vehicle abnormal data set. Step S400: Extract the abnormal data from the abnormal data sets of the content, driver, and vehicle, construct an abnormal - feature matrix after normalization processing, use a graph neural network to construct a heterogeneous risk - association graph, dynamically calculate the historical co - occurrence frequency, causal - association strength, and time correlation between abnormal nodes to generate edge weights; calculate the real - time comprehensive risk value based on the dynamic weighted - fusion algorithm, and judge the transportation safety risk level according to the preset risk threshold, thereby triggering a hierarchical early - warning mechanism.

[0015] Further, the step S100 includes: Step S101: The first - state analysis unit installs a liquid - level sensor on the tank body of the water - spraying logistics vehicle to measure the liquid level. By transmitting and receiving ultrasonic signals, calculates the liquid - level height according to the signal round - trip time; installs a pressure sensor to monitor the pressure inside the tank body; installs a thermistor - type temperature sensor to collect the temperature data of the content in real - time; installs an electrochemical leakage sensor, and once a leakage is detected, immediately records the leakage signal S, where S = 1 indicates leakage and S = 0 indicates no leakage; sets the liquid - level data number as D L , the pressure - data number as D P , the temperature - data number as D T , and the leakage - signal number as D S ; Step S102: Sets the sampling frequency of the liquid - level sensor as f L , and collects the liquid - level data once every 1 / f L time; the sampling frequency of the pressure sensor is f P , the sampling frequency of the temperature sensor is f T , the leakage sensor is in a real - time monitoring state and immediately records once a state change is detected; the collected data is first filtered using the Kalman - filtering algorithm to remove noise interference; with time t as the index, stores the processed data in chronological order to generate a content - state data set D1={(D L ,L t ),(D P ,P t ),(D T ,T t ),(D S ,S t )}, where L t ,P t ,T t ,S tThey are the liquid level, pressure, temperature, and leakage signal data corresponding to time t respectively; according to the safety standards of the transported goods, the liquid level safety threshold range is set as [Lmin, Lmax], the pressure safety threshold range is [Pmin, Pmax], and the temperature safety threshold range is [Tmin, Tmax]; When it satisfies , , , S t = 1, any one of the conditions, it is determined that the state of the content is abnormal; at the same time, record the abnormal occurrence time t y , as well as the abnormal type and the corresponding data number, and construct the content abnormal data set E1; Furthermore, the step S200 includes: Step S201: The second state analysis unit installs a vision sensor in the cab to collect the driver's facial image, extracts facial expression and eye opening degree features to form facial feature data I; embeds a pressure sensor array on the steering wheel to measure the steering wheel holding force F and holding time w; uses a photoelectric heart rate sensor, worn on the driver's wrist, to obtain heart rate data HR by detecting the photoplethysmogram; add numbers to each data, the facial feature data number is D I , the steering wheel holding force data number is D F , the holding time data number is D w , the heart rate data number is D HR ; integrate these data to construct the driver state data set D2 = {(D I , I t ), (D F , F t ), (D w , w t ), (D HR , HR t )}, where I t , F t , w t , HR t are the facial features, steering wheel holding force, holding time, and driver heart rate data corresponding to time t respectively; Step S202: Determine the fatigue driving level using a comprehensive evaluation algorithm; Condition 1: When the eye opening degree is lower than the set threshold E0 and the duration exceeds t1 seconds; Condition 2: The number of times the steering wheel grip force F is lower than the minimum grip force threshold F0 within a period T1 exceeds n times; Condition 3: The driver's heart rate HR exceeds the normal range [HRmin, HRmax]; Based on the above three conditions, set the fatigue driving level: Meeting any one of the above conditions is determined as mild fatigue driving, and record the fatigue level number A1; Meeting two of the above conditions is determined as moderate fatigue driving, and record the fatigue level number A2; Meeting all three of the above conditions is determined as severe fatigue driving, and record the fatigue level number A3; At the same time, record the fatigue level change time t p , construct the driver abnormal data set E2.

[0016] Further, the step S300 includes: Step S301: The third state analysis unit installs a tire pressure sensor inside the vehicle tire to monitor the tire pressure data Pt in real time; installs a vibration sensor on the vehicle chassis to collect vibration frequency data V; installs a pressure sensor and a time sensor in the braking system to measure the braking response time tb; installs a temperature sensor on the engine to measure the engine temperature Te; add numbers to each data, the tire pressure data number is D Pt , the vibration frequency data number is D V , the braking response time data number is D tb , the engine temperature data number is D Te ; Step S302: Set the sampling frequency of each sensor, perform denoising processing on the collected data, and store the processed data in chronological order to generate the vehicle operating state data set D3={(D Pt ,Pt t ),(D V ,V t ),(D Te ,Te t ),(D tb ,tb t )}; where Pt t , V t , Te t , tb t respectively represent the tire pressure, vibration frequency, braking response time, and engine temperature data corresponding to the t moment; according to the vehicle safety standard, set the safety threshold range, and if the real-time collected vehicle state data exceeds the safety range, it is regarded as an abnormal vehicle state; record the abnormal occurrence time t c as well as the abnormal type and the corresponding data number, and construct the vehicle abnormal data set E3.

[0017] Furthermore, the step S400 includes: Step S401: extract abnormal data from abnormal data sets E1, E2 and E3; for the data in E1 and E3, extract the abnormal occurrence time, abnormal type and corresponding data number and data value; similarly extract the fatigue level change time, fatigue level number and related data number and data value from E2; normalize the extracted abnormal data to construct an abnormal feature matrix M in a unified format, where the rows of the matrix represent different abnormal events, and the columns represent the abnormal type, abnormal occurrence time, data number and normalized data value, and fill all abnormal records into the matrix M in chronological order; Step S402: Use graph neural network to construct heterogeneous risk association graph, map the abnormal features represented by each row of data in the abnormal feature matrix M into graph nodes, and dynamically calculate the edge weights using the following formula: ij =(α·C ij +β·H ij ) / (γ·T ij ), where i and j represent abnormal nodes in the knowledge graph, C ij represents the historical co-occurrence frequency of node i and node j, H ij represents the causal relationship strength between node i and node j, T ij represents the temporal correlation between node i and node j; α, β, and γ are the corresponding adaptive weight coefficients; The C ij By analyzing historical transportation data, count the number of times the anomalies represented by node i and node j have appeared simultaneously in the past ij , and the total number of transport times C total , then C ij =count ij / C total ; The H ij Traverse the historical transportation data and extract all the historical transportation fragment information related to the anomalies represented by nodes i and j as the basic sample library; calculate the conditional probability P(j|i) that node j will be abnormal when node i is abnormal, and at the same time, calculate the probability P(j) that node j will appear in the overall sample; introduce the Bayesian network model, and according to the calculated conditional probability P(j|i) and prior probability P(j), combined with the conditional independence assumption in the network structure, through the probability reasoning and propagation mechanism, infer the possibility of causal relationship between nodes i and j, and map it to the [0,1] interval through normalization; The T ij Calculate the absolute value of the time difference between the abnormal occurrence of node i and node j |t i -t j | and through a time correlation function G(|ti -t j Map the result to the interval [0, 1], T ij =G(|t i -t j |); The time correlation function is , where λ represents the weight adjustment coefficient, λ ∈ [0.1, 0.5]; According to the calculated edge weights, connect each node to construct a heterogeneous risk association graph, showing the association strength between different anomalies.

[0018] Furthermore, the step S400 further includes: Step S403: Design a dynamic weighted fusion algorithm to calculate the real-time comprehensive risk value; First, determine the real-time weight w k (t) at time t for each analysis unit, where k = 1, 2, 3, representing the first state analysis unit, the second state analysis unit, and the third state analysis unit respectively; The weight is dynamically adjusted in real time according to the task type of the transportation task, road conditions, and vehicle status; For the m-th anomaly index in the k-th type of analysis unit, calculate the standardized score S km (t); It is realized through a pre-set scoring function S, S km (t)=S(abnormal data value, abnormal type, normal range); First, set the basic impact weight of this anomaly on transportation safety according to the abnormal type; The setting of the basic impact weight is as follows: By collecting historical vehicle operation data, set a basic weight for each discovered abnormal type and store it in the weight setting table. When a new abnormal situation is detected through real-time data, automatically retrieve the weight setting table to find the corresponding abnormal type and its basic weight; For new abnormal types that have never appeared in the historical data, they will be recorded and prompted for relevant personnel to set; Then compare the abnormal data value with the normal range, calculate the degree of deviation of the abnormal data value from the normal range, and adjust the proportion in the scoring according to the degree of deviation. The greater the degree of deviation, the higher the proportion in the scoring; Combine the degree of deviation of the abnormal data value from the normal range with the basic impact weight of the abnormal type, and calculate the preliminary score through the weighted formula; Finally, normalize the preliminary score and map the score to a specific numerical interval [0, 1]; Step S404: According to the formula: R(t)=∑ 3 k=1 (w k (t)·∑ nk m=1 S km (t)); where w k (t) is the real-time weight of each analysis unit at time t; S kmLet \(R(t)\) be the standardized score of the \(m\) -th abnormal index in the \(k\) -th analysis unit at time \(t\); \(n_k\) represents the number of abnormal indexes in the \(k\) -th analysis unit. According to the pre - set risk thresholds \(R_1\) and \(R_2\), the transportation risk levels are divided, including low - risk, medium - risk, and high - risk states. If \(R(t)\lt R_1\), it is determined to be in a low - risk state, and a primary warning is given, and the abnormal type and abnormal data are displayed in a pop - up window in the cockpit. If \(R_1\leq R(t)\lt R_2\), it is determined to be in a medium - risk state, an alarm is issued, and the safe stopping point is automatically calculated, and a safe driving path is planned. If \(R(t)\geq R_2\), it is determined to be in a high - risk state. At this time, the emergency rescue process is automatically triggered. First, a safe driving path and a stopping point are planned, and relevant rescue departments are notified to report the vehicle position, the state of the tank content, the abnormal type, and the current risk level.

[0019] A multi - sensor - based vehicle - carrying safety monitoring system, the system includes a content state monitoring module, a driver behavior monitoring module, a vehicle operation monitoring module, and a dynamic risk assessment module; The content state monitoring module uses a variety of sensors to collect the liquid level, pressure, temperature, and leakage signals of the tank content, generates a state data set, sets abnormal determination criteria, compares the data in real - time, outputs an abnormal determination result, and constructs a content abnormal data set; The driver behavior monitoring module collects the driver's facial features, steering wheel grip force, grip duration, and heart rate data, constructs a driver state data set; by setting multi - condition fatigue level determination rules, combining real - time data with determination conditions, outputs a fatigue level result, and constructs a driver abnormal data set; The vehicle operation monitoring module collects tire pressure, vibration frequency, braking response time, and engine temperature data, generates a vehicle operation state data set, and sets abnormal indexes; by comparing the data in real - time, obtains an abnormal determination result, and constructs a vehicle abnormal data set; The dynamic risk assessment module is used to extract abnormal data from the content, driver, and vehicle abnormal data sets, construct an abnormal feature matrix after normalization processing, use a graph neural network to construct a heterogeneous risk association map, dynamically calculate the historical co - occurrence frequency, causal association strength, and time correlation between abnormal nodes, and generate edge weights; based on a dynamic weighted fusion algorithm, calculate the real - time comprehensive risk value, and judge the transportation safety risk level according to the preset risk threshold, thereby triggering a hierarchical warning mechanism.

[0020] Embodiment of the present invention: Step S100: Install liquid level, pressure, temperature, and leakage sensors on the tank body of the bulk water logistics vehicle, set the sampling frequency to collect data, set the safety threshold range of each data, when the data exceeds the range or leakage is detected, record the abnormal time, type, and data number, and construct an abnormal data record set \(E1\); Install a vision sensor in the cab, embed a pressure sensor array in the steering wheel, and the driver wears a photoelectric heart rate sensor; set the eye opening threshold E0 = 0.3, the duration t1 = 60 seconds, the minimum grip force threshold F0 = 5N, the time T1 = 10 minutes, the number of times n = 5 times, and the normal heart rate range [HRmin, HRmax] = [60, 100] beats per minute; determine the fatigue level according to the conditions, record the change time, and construct a fatigue driving record set E2; Install a tire pressure sensor in the vehicle tire, install a vibration sensor on the chassis, install pressure and time sensors on the braking system, install a temperature sensor on the engine, set the safety threshold range of each data, and when the data exceeds the range, record the abnormal time, type, and data number, and construct a vehicle abnormal data record set E3; Extract abnormal data from E1, E2, and E3. A certain record in E1 is (abnormal occurrence time: 2025-03-01 10:00, abnormal type: liquid level anomaly, data number: DL, data value: 550); a certain record in E2 is (fatigue level change time: 2025-03-01 10:30, fatigue level number: A1, related data number: DI, data value: eye opening degree 0.2); a certain record in E3 is (abnormal occurrence time: 2025-03-01 11:00, abnormal type: tire pressure anomaly, data number: D Pt , data value: 2.0). Perform normalization processing on the extracted data, and fill the normalized data into the abnormal feature matrix M in chronological order. The rows of the matrix represent different abnormal events, and the columns represent the abnormal type, abnormal occurrence time, data number, and normalized data value; map the abnormal features represented by each row of data in the abnormal feature matrix M to graph nodes. Taking the above liquid level anomaly, driver mild fatigue, and tire pressure anomaly as examples, they correspond to three nodes in the graph respectively; By analyzing historical transportation data, count the number of times two anomalies occur simultaneously ij and the total number of transports C total ; Taking tire anomaly and tire pressure anomaly as an example, they occurred simultaneously 10 times in the past 100 transports, then C ij = count ij / C total = 0.1; Traverse the historical transportation data, extract all relevant historical transportation segment information as the basic sample library; calculate the conditional probability P(j|i) and the prior probability P(j); introduce the Bayesian network model to infer the possibility of causal association and normalize and map it to the [0, 1] interval; in the case of liquid level anomaly, the conditional probability of tire pressure anomaly P(j|i) = 0.2, and the probability of tire pressure anomaly in the overall sample P(j) = 0.1. After calculation and normalization through the Bayesian network model, Hij = 0.3; The said T ij Calculate the absolute value of the time difference |t i - t j | between the occurrence times of anomalies of computing node i and node j, and map the result to the interval [0, 1] through a time correlation function G(|t i - t j |), T ij = G(|t i - t j |); The said time correlation function is ; According to the above anomaly records, the liquid level anomaly occurrence time t i = 10:00, the tire pressure anomaly occurrence time t j = 11:00, the time difference |t i - t j | = 60 minutes, λ = 0.1, T ij = 1 / (1 + e -6 ) = 0.9975; Set α = 0.3, β = 0.4, γ = 0.3 respectively, and calculate the edge weight value W ij ≈ 0.5; According to the calculated edge weights, connect each node to construct a heterogeneous risk correlation graph to show the correlation strength between different anomalies; According to the transportation task type (whether it is a hazardous chemical substance), road conditions (whether the traffic is congested), and vehicle status, dynamically adjust the weights of each analysis unit in real time, and set the current w 1 (t) = 0.4, w 2 (t) = 0.3, w 3 (t) = 0.3; Standardized score calculation: For each anomaly index, calculate the standardized score S km (t) through a pre-set scoring function S; Taking the liquid level anomaly as an example, set the basic impact weight to 0.6 according to the anomaly type. Since the liquid level value of 550 exceeds the normal range to a large extent, adjust the proportion and make a preliminary score, and then normalize to get S 11 (t) = 0.7; According to the formula R(t) = ∑ 3 k=1 (w k (t) · ∑ nk m=1 S km (t)) calculate the comprehensive risk value, where n1 = 2, n2 = 3, n3 = 1, and calculate R(t) = 0.6; Set the risk thresholds R1 = 0.3, R2 = 0.8. Since R1 <= R(t) < R2, it is determined to be in a medium-risk state, an alarm is issued, and the safe docking point is automatically calculated, and the safe driving path is planned.

[0021] It is obvious to those skilled in the art that the present invention is not limited to the details of the above-described exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, in all respects, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be embraced within the present invention. Any reference signs in the claims should not be construed as limiting the claims involved.

Claims

1. A multi-sensor based transportation safety monitoring method, characterized in that: The method comprises: Step S100: construct three state analysis units to monitor the contents, driver and vehicle status of the bulk water logistics vehicle respectively; the first state analysis unit uses multiple sensors to collect the liquid level, pressure, temperature and leakage signal of the tank contents, generates a state data set and sets an abnormality judgment standard, compares the data in real time to output the abnormality judgment result, and constructs a content abnormality data set; Step S200: The second state analysis unit installs sensors in the cab to collect the driver's facial features, steering wheel grip force, grip duration and heart rate data to construct a driver state data set; by setting a multi-condition fatigue level determination rule, combining real-time data with determination conditions, outputs the fatigue level result, and constructs a driver abnormal data set; Step S300: The third state analysis unit installs sensors on the vehicle to collect tire pressure, vibration frequency, brake response time and engine temperature data, generates a vehicle operation state data set and sets abnormal indicators; obtains abnormality determination results by comparing data in real time, and constructs a vehicle abnormality data set; Step S400: Extract abnormal data from the abnormal data sets of contents, drivers and vehicles, construct an abnormal feature matrix after normalization, use graph neural network to construct a risk association map, dynamically calculate the historical co-occurrence frequency, causal association strength and time correlation between abnormal nodes, and generate edge weights; calculate the real-time comprehensive risk value based on the dynamic weighted fusion algorithm, and judge the transportation safety risk level according to the preset risk threshold, thereby triggering a graded warning mechanism.

2. The method for monitoring transportation safety based on multiple sensors according to claim 1, characterized in that: The step S100 includes: Step S101: The first state analysis unit is to install a liquid level sensor on the tank of the bulk water logistics vehicle to measure the liquid level, and calculate the liquid level height according to the round-trip time of the signal by transmitting and receiving ultrasonic signals; install a pressure sensor to monitor the pressure inside the tank; install a thermistor temperature sensor to collect the temperature data of the contents in real time; install an electrochemical leakage sensor, and once a leak is detected, immediately record the leakage signal S, where S=1 indicates leakage and S=0 indicates no leakage; set the liquid level data number to D L , pressure data number is D P , temperature data number is D T , the leakage signal number is D S ; Step S102: Set the liquid level sensor sampling frequency to f L , every 1 / f L The liquid level data is collected once every 24 hours; the sampling frequency of the pressure sensor is f P , the temperature sensor sampling frequency is f T , the leakage sensor is in real-time monitoring state, and once a state change is detected, it is recorded immediately; the collected data is first filtered using the Kalman filter algorithm to remove noise interference; the processed data is stored in chronological order with time t as the index, generating the content state data set D1={(D L ,L t ),(D P ,P t ),(D T ,T t ),(D S ,S t )}, where L t , P t 、T t , S t are the liquid level, pressure, temperature and leakage signal data corresponding to time t respectively; according to the safety standards for transporting goods, the liquid level safety threshold range is set to [Lmin, Lmax], the pressure safety threshold range is set to [Pmin, Pmax], and the temperature safety threshold range is set to [Tmin, Tmax]; When satisfied , , , S t =1, the content is judged to be abnormal; at the same time, the abnormality time t is recorded. y , as well as the abnormal type and the corresponding data number, to construct the content abnormality data set E1.

3. The method for monitoring transportation safety based on multiple sensors according to claim 1, characterized in that: The step S200 includes: Step S201: The second state analysis unit is installed in the cab with a visual sensor to collect the driver's facial image, extract facial expression and eye opening and closing characteristics, and form facial feature data I; a pressure sensor array is embedded in the steering wheel to measure the steering wheel grip force F and grip time w; a photoelectric heart rate sensor is used, worn on the driver's wrist, and the heart rate data HR is obtained by detecting the photoelectric volume pulse wave; each data is numbered, and the facial feature data is numbered D I , the steering wheel grip data number is D F , the holding time data number is D w , heart rate data number is D HR ; Integrate these data to construct the driver status data set D2 = {(D I ,I t ),(D F ,F t ),(D w ,w t ),(D HR ,HR t )}, where I t 、F t 、w t , HR t They are the facial features, steering wheel grip force, grip time and driver’s heart rate data corresponding to time t; Step S202: using a comprehensive evaluation algorithm to determine the fatigue driving level; condition one: when the eye opening degree is lower than the set threshold E0 and the duration exceeds t1 seconds; condition two: the steering wheel grip force F is lower than the minimum grip force threshold F0 for more than n times within a period of time T1; condition three: the driver's heart rate HR exceeds the normal range [HRmin, HRmax]; based on the above three conditions, set the fatigue driving level: if any of the above conditions is met, it is determined as mild fatigue driving, and the fatigue level number A1 is recorded; if the above two conditions are met, it is determined as moderate fatigue driving, and the fatigue level number A2 is recorded; if the above three conditions are met, it is determined as severe fatigue driving, and the fatigue level number A3 is recorded; at the same time, record the fatigue level change time t p , construct the driver abnormality dataset E2.

4. The method for monitoring transportation safety based on multiple sensors according to claim 1, characterized in that: The step S300 includes: Step S301: The third state analysis unit installs a tire pressure sensor in the vehicle tire to monitor the tire pressure data Pt in real time; installs a vibration sensor on the vehicle chassis to collect vibration frequency data V; installs a pressure sensor and a time sensor in the braking system to measure the braking response time tb; installs a temperature sensor on the engine to measure the engine temperature Te; adds a number to each data, and the tire pressure data is numbered D Pt , the vibration frequency data number is D V , the brake response time data number is D tb , the engine temperature data number is D Te ; Step S302: Set the sampling frequency of each sensor, perform denoising on the collected data, store the processed data in chronological order, and generate a vehicle operation status data set D3 = {(D Pt ,Pt t ),(D V ,V t ),(D Te ,Te t ),(D tb ,tb t )}; where Pt t 、V t 、Te t ,tb t They represent the tire pressure, vibration frequency, brake response time, and engine temperature data corresponding to time t respectively; according to the vehicle safety standards, the safety threshold range is set. If the real-time collected vehicle status data exceeds the safety range, it is considered that the vehicle status is abnormal; the abnormality time t is recorded c As well as the anomaly type and the corresponding data number, the vehicle anomaly dataset E3 is constructed.

5. The method for monitoring transportation safety based on multiple sensors according to claim 1, characterized in that: The step S400 includes: Step S401: extract abnormal data from abnormal data sets E1, E2 and E3; for the data in E1 and E3, extract the abnormal occurrence time, abnormal type and corresponding data number and data value; similarly extract the fatigue level change time, fatigue level number and related data number and data value from E2; normalize the extracted abnormal data to construct an abnormal feature matrix M in a unified format, where the rows of the matrix represent different abnormal events, and the columns represent the abnormal type, abnormal occurrence time, data number and normalized data value, and fill all abnormal records into the matrix M in chronological order; Step S402: Use graph neural network to construct heterogeneous risk association graph, map the abnormal features represented by each row of data in the abnormal feature matrix M into graph nodes, and dynamically calculate the edge weights using the following formula: ij =(α·C ij +β·H ij ) / (γ·T ij ), where i and j represent abnormal nodes in the knowledge graph, C ij represents the historical co-occurrence frequency of node i and node j, H ij represents the causal relationship strength between node i and node j, T ij represents the temporal correlation between node i and node j; α, β, and γ are the corresponding adaptive weight coefficients; The C ij By analyzing historical transportation data, count the number of times the anomalies represented by node i and node j have appeared simultaneously in the past ij , and the total number of transport times C total , then C ij =count ij / C total ; The H ij Traverse the historical transportation data and extract all the historical transportation fragment information related to the anomalies represented by nodes i and j as the basic sample library; calculate the conditional probability P(j|i) that node j will be abnormal when node i is abnormal, and at the same time, calculate the probability P(j) that node j will appear in the overall sample; introduce the Bayesian network model, and according to the calculated conditional probability P(j|i) and prior probability P(j), combined with the conditional independence assumption in the network structure, through the probability reasoning and propagation mechanism, infer the possibility of causal relationship between nodes i and j, and map it to the [0,1] interval through normalization; The T ij Calculate the absolute value of the time difference between the abnormal occurrence of node i and node j |t i -t j | and through a time correlation function G(|t i -t j |) maps the result to the interval [0,1], T ij =G(|t i -t j |); the time correlation function is , where λ represents the weight adjustment coefficient, λ∈[0.1, 0.5]; According to the calculated edge weights, each node is connected to construct a heterogeneous risk association map to show the association strength between different anomalies.

6. The method for monitoring transportation safety based on multiple sensors according to claim 1, characterized in that: The step S400 further includes: Step S403: Design a dynamic weighted fusion algorithm to calculate the real-time comprehensive risk value; first determine the real-time weight w of each analysis unit at time t k (t), where k=1, 2, 3, respectively representing the first state analysis unit, the second state analysis unit, and the third state analysis unit; the weight is adjusted dynamically in real time according to the task type, road conditions, and vehicle status of the transportation task; For the mth abnormal indicator in the kth analysis unit, calculate the standardized score S km (t); This is achieved through a pre-set scoring function S, S km (t)=S(abnormal data value, abnormal type, normal range): First, the basic impact weight of this abnormality on transportation safety is set according to the abnormal type; First, the basic impact weight of this abnormality on transportation safety is set according to the abnormal type; The setting of the basic impact weight is to collect the historical data of vehicle operation, set a basic weight for each abnormal type that has been found, and store it in the weight setting table. When a new abnormal situation is monitored through real-time data, the weight setting table is automatically retrieved to find the corresponding abnormal type and its basic weight; For new abnormal types that have never appeared in historical data, they will be recorded and prompted to be set by relevant personnel; Then, the abnormal data value is compared with the normal range, the degree of deviation of the abnormal data value from the normal range is calculated, and the proportion in the score is adjusted according to the degree of deviation. The greater the degree of deviation, the higher the proportion in the score; The degree of deviation of the abnormal data value from the normal range is combined with the basic impact weight of the abnormal type, and a preliminary score is calculated through a weighted formula; Finally, the preliminary score is normalized and the score is mapped to a specific numerical interval [0,1]; Step S404: According to the formula: R(t)=∑ 3 k=1 (w k (t)·∑ nk m=1 S km (t)); where w k (t) is the real-time weight of each analysis unit at time t; S km (t) is the standardized score of the m-th abnormal index in the k-th type of analysis unit at time t; nk represents the number of abnormal indexes in the k-th type of analysis unit; According to the preset risk thresholds R1 and R2, the transportation risk levels are divided, including low risk, medium risk, and high risk states; If R(t) < R1, it is determined as a low risk state, and a primary warning is given, and the abnormal type and abnormal data are displayed in a pop-up window in the cockpit; If R1 <= R(t) < R2, it is determined as a medium risk state, an alarm is issued and the safe stopping point is automatically calculated, and a safe driving path is planned; If R(t) >= R2, it is determined as a high risk state, and at this time, the emergency rescue process is automatically triggered. First, a safe driving path and a stopping point are planned, and relevant rescue departments are notified to report the vehicle position, the state of the tank contents, the abnormal type, and the current risk level.

7. A multi-sensor based transportation safety monitoring system, characterized in that: The system includes a content status monitoring module, a driver behavior monitoring module, a vehicle operation monitoring module, and a dynamic risk assessment module; The content status monitoring module uses a variety of sensors to collect the liquid level, pressure, temperature and leakage signals of the tank contents, generates a status data set and sets abnormality judgment criteria, compares data in real time to output abnormality judgment results, and constructs a content abnormality data set; The driver behavior monitoring module collects the driver's facial features, steering wheel grip force, grip duration and heart rate data to construct a driver status data set; By setting multi-condition fatigue level judgment rules, combining real-time data with judgment conditions, output fatigue level results and build driver abnormal data sets; The vehicle operation monitoring module collects tire pressure, vibration frequency, brake response time and engine temperature data, generates a vehicle operation status data set and sets abnormal indicators; By comparing data in real time, we can obtain abnormality judgment results and build a vehicle abnormality data set; The dynamic risk assessment module is used to extract abnormal data from the abnormal data sets of content, driver and vehicle, construct an abnormal feature matrix after normalization, use graph neural network to construct a heterogeneous risk association map, dynamically calculate the historical co-occurrence frequency, causal association strength and time correlation between abnormal nodes, and generate edge weights; The real-time comprehensive risk value is calculated based on the dynamic weighted fusion algorithm, and the transportation safety risk level is judged according to the preset risk threshold, thereby triggering a graded early warning mechanism.

8. The multi-sensor based transportation safety monitoring system according to claim 7, characterized in that: The dynamic risk assessment module includes a feature matrix construction unit, a risk association map unit, a dynamic risk assessment unit and a risk warning unit; The feature matrix construction unit generates an abnormal feature matrix by extracting the abnormal time, type and data value in E1, E2 and E3 and normalizing them; The risk association graph unit dynamically calculates the co-occurrence frequency, causal strength and time correlation between nodes through a graph neural network to generate a risk association graph.

9. The multi-sensor based transportation safety monitoring system according to claim 8, characterized in that: The dynamic risk assessment unit dynamically adjusts the weight of the analysis unit according to the task type, road conditions, and vehicle status, and calculates the real-time risk value in combination with the abnormal deviation scoring function; The risk warning unit divides the risk into low risk, medium risk and high risk levels by presetting risk thresholds, and sets up a graded warning mechanism.

Images