Dangerous chemical substance road transportation accident tracing method based on big data

Abstract

The invention discloses a dangerous chemical substance road transportation accident tracing method based on big data, and the method comprises the steps: carrying out the preprocessing of a historical accident report, building an accident scene reconstruction model based on the preprocessed data, taking the output as a bayesian network child node, building a bayesian network through considering the interaction between different reasons, and carrying out the clustering according to a vehicle track, the method comprises the following steps: acquiring regional risk adjustment Bayesian network parameters, performing time sequence analysis on a Bayesian network to obtain a chain reaction feature chain, iteratively analyzing a plurality of traceability end points according to the chain reaction feature chain, calculating cause intensity of each traceability end point, and finally determining an accident cause path based on the traceability end point with the highest cause intensity. According to the method, the accident data value can be deeply mined, the key cause factors can be accurately positioned, the tracing accuracy and reliability can be obviously improved, effective tracing in a complex accident scene can be ensured, the formulation of an accident prevention strategy can be powerfully supported, and meanwhile, the method has relatively good interpretability.

Description

Technical Field

[0001] The present invention relates to the field of intelligent analysis of transportation risks, and in particular to a method for tracing the source of hazardous chemical road transportation accidents based on big data. Background Art

[0002] With the rapid development of the chemical industry, the demand for road transportation of hazardous chemicals continues to grow. The road transportation of hazardous chemicals involves many links, each of which is interrelated and information is scattered. Traditional management and supervision models make it difficult to fully, real-timely and accurately control the entire transportation process.

[0003] Traditional accident tracing methods rely heavily on on-site investigations and limited monitoring data. However, in the era of big data, a vast amount of data related to hazardous chemical road transport, such as daily operational data from transport companies, detailed physical and chemical properties of hazardous chemicals, historical traffic flow data from road networks, and long-term monitoring of meteorological data along routes, has not been effectively integrated and deeply mined. This data contains a wealth of information and has enormous potential for comprehensively and accurately tracing the source of accidents.

[0004] The big data-based hazardous chemical road transport accident tracing method can model the complex nonlinear relationships in the hazardous chemical transportation process, thereby more accurately predicting the possibility of accidents, accurately tracing the root causes of accidents, and determining the key links and main factors in the development of accidents, providing a scientific basis for accident prevention and emergency response, and improving the safety and management level of the entire hazardous chemical road transport industry. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for tracing the source of hazardous chemicals road transportation accidents based on big data.

[0006] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0007] The first aspect of the present invention provides a method for tracing the source of hazardous chemical road transport accidents based on big data, comprising:

[0008] Pre-process historical accident reports, including data cleaning, text segmentation, and word frequency statistics;

[0009] The system uses the real-time status of the vehicle and road environment characteristics as input, and the status of the vehicle and hazardous chemicals at the time of the accident as output. It then constructs an accident scene reconstruction model based on the pre-processed data.

[0010] The output of the scenario accident reconstruction model is used as a child node of the Bayesian network, and the Bayesian network is constructed by considering the mutual influence between different causes; the vehicle trajectories are clustered to obtain the vehicle distribution in different areas, and the Bayesian network parameters are adjusted based on regional risks;

[0011] Performing time series analysis on the Bayesian network to obtain a chain reaction feature chain;

[0012] Performing iterative tracing analysis based on the chain reaction feature chain to obtain multiple tracing endpoints, and calculating the causal strength of each tracing endpoint using the Bayesian network;

[0013] Analyze the traceability endpoint with the highest causal intensity to determine the most likely accident cause path.

[0014] As a further method, the method of constructing an accident scene reconstruction model using a neural network based on the preprocessed data includes:

[0015] Real-time vehicle status features include speed, acceleration, braking system status, steering angle, vehicle stability, and tire status; road environment features include road type, road surface status, weather conditions, traffic flow, and road geometry; vehicle status at the time of the accident includes collision speed, collision angle, component integrity, and vehicle body posture; hazardous chemical status at the time of the accident includes hazardous chemical temperature, hazardous chemical pressure, hazardous chemical sealing status, and hazardous chemical container appearance integrity;

[0016] A neural network model is constructed using the vehicle's real-time status and road environment characteristics as input and the vehicle and hazardous chemicals' status at the time of the accident as output;

[0017] To evaluate the severity of historical accidents, the expression is:

[0018]

[0019] Among them, S i is the severity assessment value of the i-th historical accident, α, β and γ are the weights of the hazardous chemicals, human economic losses and environmental losses respectively, m i is the number of hazardous chemicals involved in the i-th historical accident, q ij 、r ij 、s ij , t ij and u ij are the leakage volume, toxicity level, flammability level, corrosiveness level and leakage range of the jth hazardous chemical in the i-th historical accident, v i and w i are the personal injury and economic loss of the i-th historical accident, respectively, i is the environmental impact level of the i-th historical accident;

[0020] The preprocessed data is divided into training set, validation set and test set in a ratio of 7:2:1 to train the neural network model. During the training process, the learning rate is dynamically adjusted based on the severity of historical accidents. The expression is:

[0021]

[0022] Among them, η i is the learning rate for training using the i-th historical accident, η0 is the initial learning rate, S max is the maximum severity evaluation value of all historical accidents, p is the coefficient that controls the influence of severity on learning rate, t is the current number of training steps, and T is the total number of training steps.

[0023] As a further method, the method of using the output of the scenario accident reconstruction model as a sub-node of the Bayesian network and considering the mutual influence between different causes to construct the Bayesian network includes:

[0024] The status of the vehicle and hazardous chemicals at the time of the accident is used as the child node in the Bayesian network, and the cause variable is used as the parent node in the Bayesian network;

[0025] Determine the mutual influence function between the cause variables as the dependency relationship of the edges connecting the parent nodes, and construct a Bayesian network based on the hierarchical relationship of the cause variables. The expression is:

[0026]

[0027] Among them, p(θ|Y) is the posterior probability, which represents the probability distribution of the model parameter θ after observing the data Y, n is the number of parameters in the model parameter vector, and θ i is the i-th parameter in the model parameter vector, p(θ i ) is θ i The prior probability of σ is the standard deviation of the noise, y r (θ) is the rth observation value predicted based on the model parameters θ, y r is the tth data point actually observed;

[0028] The output data of the scenario accident reconstruction model is used to adjust the conditional probability representation of the corresponding sub-nodes in the Bayesian network.

[0029] As a further method, clustering the vehicle trajectories to obtain the vehicle distribution in different areas and adjusting the Bayesian network parameters based on regional risks include:

[0030] Divide the area where the vehicle operates into multiple area units;

[0031] Cluster the vehicle trajectories according to a preset threshold. For each cluster, count the number of vehicles in each area, their frequency, and their residence time in the area to obtain the regional vehicle distribution feature vector.

[0032] Conduct risk assessment for each area based on the regional vehicle distribution feature vector;

[0033] The regional risk assessment value is input into the Bayesian network as evidence to update the posterior probability of the parent node in the network. The expression is:

[0034]

[0035] Among them, p j is the jth parent node, R is the regional risk assessment value, P(p j |R) is the parent node p when the regional risk assessment value R is known. j The posterior probability of m is the number of child nodes, n is i is the i-th child node, P(R|n i ) is the value of the given child node n i Under the state, the probability of regional risk assessment value is R, (n i |p j ) is the given parent node p j In the state, child node n i The conditional probability, P(p j ) is the prior probability of the parent node, and k is the number of parent nodes.

[0036] As a further method, the method of performing temporal analysis on the Bayesian network to obtain a chain reaction feature chain includes:

[0037] Construct a linear Gaussian state space model and use Bayesian networks to set the prior distribution of the parameters in the linear Gaussian state space model, including the prior distribution of the transformation parameters and observation parameters;

[0038] Infer the posterior probability of the model parameters of the linear Gaussian state space model based on the prior distribution to obtain the posterior samples of the parameters;

[0039] Based on the posterior samples, the temporal changes of the parent node states in the Bayesian network are analyzed, and the state transition probabilities of the parent nodes in adjacent time steps and the joint state transition probabilities of different parent node pairs in adjacent time steps are calculated;

[0040] If the state transition probability of a parent node in adjacent time steps exceeds a preset first threshold, the parent node is regarded as a key node. If the joint state transition probability of different parent node pairs in adjacent time steps exceeds a preset second threshold and both can point to the same child node, the parent node pair is regarded as a joint key node.

[0041] Starting from key nodes and joint key nodes, a chain reaction feature chain is constructed based on the state transition relationship.

[0042] As a further method, the method of performing iterative tracing analysis based on the chain reaction feature chain to obtain multiple tracing end points includes: starting from the child node, using a depth-first search algorithm, and performing a reverse recursive search in the Bayesian network based on the chain reaction feature chain. If the in-degree of the terminating node is 0 and it is a parent node, then marking the parent node as the tracing end point.

[0043] As a further method, the method of calculating the causal strength of each traceability endpoint using the Bayesian network includes:

[0044] Calculate the causal strength of each traceability endpoint using the expression:

[0045]

[0046] Where ξ is the number of traceability endpoints, X i is the ith tracing end point, P(X i ) is X i The marginal probability, Pa(X i ) is X i The set of adjacent nodes, P(X i =1|X j =1) means that in X j If true, X i The conditional probability of being true, P(X j =1) is X j The prior probability of being true, P(X i =1|X j =0) means that in X j If it is false, X i The conditional probability of being true, P(X j =0) is X j The prior probability of being false.

[0047] As a further method, the method of analyzing the traceability endpoint with the highest causal intensity to determine the most likely accident causal path includes:

[0048] Construct a hidden Markov chain model, where the hidden state is the tracing endpoint and the observed state is the accident feature. Based on the transition probability between nodes in the Bayesian network, determine the transition probability matrix of the hidden Markov chain.

[0049] Determine the initial state probability distribution of the hidden Markov chain based on the tracing endpoint with the highest causal intensity;

[0050] Based on the initial state probability distribution, the state transition probability of each time step is calculated by the Viterbi algorithm in the hidden Markov chain. The expression is:

[0051]

[0052] Among them, V q (h) is the probability of the most likely causal path in state h at time step q, N is the total number of states, δ q-1 (i) is the initial probability distribution of being in state i at time step q-1, a ih is the probability of transitioning from state i to state h, γ q-1 is the time decay factor at time step q-1, b h (o q ) is observed in state h q The probability of q is the observation value at time step q;

[0053] The path with the highest probability is determined as the most likely accident causal path.

[0054] In a second aspect, an embodiment of the present application further provides an electronic device, comprising: a processor; and a memory arranged to store computer-executable instructions, wherein the executable instructions, when executed, cause the processor to perform the method steps described in the first aspect.

[0055] In a third aspect, an embodiment of the present application further provides a computer-readable storage medium, which stores one or more programs. When the one or more programs are executed by an electronic device including multiple applications, the electronic device executes the method steps described in the first aspect.

[0056] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0057] (1) The present invention combines the accident scene reconstruction model constructed by the neural network with the Bayesian network and the hidden Markov chain model, which can learn complex mapping relationships and provide rich sub-node information for the subsequent construction of the Bayesian network. This multi-model fusion method makes the entire traceability analysis more systematic and comprehensive, and improves the accuracy and reliability of the traceability results.

[0058] (2) The present invention obtains the vehicle distribution in different regions by clustering analysis of vehicle driving trajectories and adjusts the Bayesian network parameters based on regional risks. This can take into account the special risk factors in different regions and make the traceability results more in line with the actual situation. It also helps to formulate personalized risk prevention and control strategies for different regions and optimize the overall safety management of hazardous chemicals road transportation.

[0059] (3) The present invention uses the Bayesian network to calculate the causal strength of each tracing end point, and constructs a hidden Markov chain model with the tracing end point with the highest causal strength as the core to determine the accident causal path. It can accurately determine the main cause and development process of the accident, further improve the accuracy of the accident causal path determination, and provide a strong basis for accident responsibility determination, preventive measures formulation, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a flowchart of the steps of a method for tracing the source of hazardous chemical road transport accidents based on big data in an embodiment of the present invention.

[0061] Figure 2 This is a schematic diagram of the structure of an electronic device in an embodiment of this specification. DETAILED DESCRIPTION

[0062] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0063] Reference Figure 1 As shown, the present invention provides a method for tracing the source of hazardous chemicals road transportation accidents based on big data, including:

[0064] S100 pre-processes historical accident reports, including data cleaning, text segmentation, and word frequency statistics;

[0065] In the actual assessment, detailed reports on 100 hazardous chemical road transport accidents in a certain area in the past three years were collected, and the accidents were pre-processed. After text segmentation and word frequency statistics, it was found that words such as "speeding", "road construction", "fatigue driving", and "traffic congestion" appeared frequently.

[0066] S200 takes the real-time status of the vehicle and road environment characteristics as input, and the status of the vehicle and hazardous chemicals at the time of the accident as output, and builds an accident scene reconstruction model based on the pre-processed data;

[0067] In the actual assessment, real-time vehicle status data, road environment characteristic data, vehicle status data at the time of the accident, and hazardous chemical status data at the time of the accident were collected. The real-time vehicle status data showed that 31 of the accident vehicles were traveling at speeds exceeding 80 km / h before the accident; acceleration data showed that 14 of the accident vehicles experienced sudden acceleration or deceleration, with the maximum acceleration reaching 3 m / s. 2In terms of brake system status, 21 of the accident vehicles had varying degrees of potential faults in their brake systems. Steering angle data showed that 8 of the accident vehicles had abnormal steering angles before the accident, with the maximum steering angle reaching 60 degrees. Vehicle stability data revealed that 8 of the accident vehicles had poor stability during driving. Tire status data showed that 13 of the accidents involved abnormal tire pressure or severe wear.

[0068] In the actual evaluation, this data was divided into training, validation, and test sets in a ratio of 7:2:1 to construct a neural network model, and the severity of historical accidents was evaluated. For an accident involving a large amount of gasoline leakage and serious pollution to the surrounding environment, its evaluation value was calculated according to the severity assessment formula. During the training process, the learning rate was dynamically adjusted according to the severity of each accident. The initial learning rate was set to 0.008, and the learning rate was adjusted according to the formula.

[0069] S300 uses the output of the scenario accident reconstruction model as a child node of the Bayesian network and constructs the Bayesian network considering the mutual influence between different causes; clusters the driving trajectories of vehicles to obtain the distribution of vehicles in different areas, and adjusts the Bayesian network parameters based on regional risks;

[0070] In the actual evaluation, a Bayesian network was constructed using the vehicle and hazardous chemicals status at the time of the accident as child nodes, and the driver status, vehicle condition, and road environment factors as parent nodes. The interaction function between the causal variables was determined, and it was found that fatigue driving increased the probability of driver inattention by 27%. This was used to construct the dependency relationship between the parent nodes. The output data of the scenario accident reconstruction model was used to adjust the conditional probability of the corresponding child nodes in the Bayesian network. When the vehicle's braking system failed, the conditional probability of the vehicle loss of control child node was adjusted based on the data.

[0071] In the actual assessment, the vehicle operation area was divided into eight regional units. Cluster analysis revealed that the average daily number of vehicles in Region 1 was 426, with a frequency of 8-40 vehicles per hour and an average dwell time of 15 minutes. The average daily number of vehicles in Region 3 was 748, with a frequency of 10-50 vehicles per hour and an average dwell time of 20 minutes. Risk was assessed based on the regional vehicle distribution feature vectors. Region 5, due to its narrow roads and high traffic volume, received a risk assessment of 0.72. Region 8, due to its proximity to mountainous areas and complex road conditions, received a risk assessment of 0.60. These regional risk assessments were input into the Bayesian network to update the posterior probabilities of the parent nodes. A high risk value for Region 5 would increase the posterior probability of the "narrow road" variable in the road environment factor parent node by 20%.

[0072] S400 performs time series analysis on the Bayesian network to obtain a chain reaction feature chain;

[0073] In the actual evaluation, a linear Gaussian state space model was constructed, and prior distributions were set for its parameters. The prior distribution of the conversion parameters was set to a normal distribution with a mean of 0.5 and a standard deviation of 0.1; the prior distribution of the observation parameters was set to a normal distribution with a mean of 1.2 and a standard deviation of 0.2; the posterior probability of the model parameters was inferred based on the prior distribution, and the posterior samples were obtained. The time series changes of the parent node states were analyzed, and the state transition probabilities of the parent nodes in adjacent time steps and the joint state transition probabilities of different parent node pairs were calculated. It was found that the state transition probability of the "fatigue driving" parent node in adjacent time steps exceeded the preset threshold of 0.6, and it was marked as a critical node; the joint state transition probability of the "slippery road" and "speeding" parent node pairs exceeded the preset threshold of 0.5 and pointed to the "vehicle out of control" child node, which was marked as a joint critical node, and a chain reaction feature chain was constructed from these critical nodes and joint critical nodes.

[0074] S500 performs iterative tracing analysis based on the chain reaction feature chain to obtain multiple tracing endpoints, and calculates the causal strength of each tracing endpoint using the Bayesian network;

[0075] In the actual evaluation, the depth-first search algorithm was used to perform a reverse recursive search based on the chain reaction feature chain in the Bayesian network. In one accident, "fatigue driving" and "vehicle brake failure" were finally determined to be the tracing end points. After calculation, "fatigue driving" was found to have the highest causal intensity.

[0076] S600 analyzes the traceability endpoint with the highest causal intensity to determine the most likely accident cause path.

[0077] In the actual assessment, a hidden Markov chain model was constructed, with the tracing end point as the hidden state and the traces at the accident scene (brake mark length, hazardous chemical leakage diffusion range) as the observed state. The hidden Markov chain transition probability matrix was determined based on the transition probability between Bayesian network nodes. The Bayesian network transition probability of "fatigue driving" leading to "vehicle loss of control" was 0.45, and the corresponding transition probability in the hidden Markov chain was also 0.45. According to the highest causal intensity of "fatigue driving", the initial state probability distribution of the hidden Markov chain was determined to be 0.6. The Viterbi algorithm was used to calculate the state transition probability of each time step, and the most likely accident cause path was determined to be "fatigue driving-vehicle operation error-vehicle loss of control-hazardous chemical container damage-hazardous chemical leakage".

[0078] In this embodiment, the method of constructing an accident scene reconstruction model using a neural network based on preprocessed data includes:

[0079] Real-time vehicle status features include speed, acceleration, braking system status, steering angle, vehicle stability, and tire status; road environment features include road type, road surface status, weather conditions, traffic flow, and road geometry; vehicle status at the time of the accident includes collision speed, collision angle, component integrity, and vehicle body posture; hazardous chemical status at the time of the accident includes hazardous chemical temperature, hazardous chemical pressure, hazardous chemical sealing status, and hazardous chemical container appearance integrity;

[0080] A neural network model is constructed using the vehicle's real-time status and road environment characteristics as input and the vehicle and hazardous chemicals' status at the time of the accident as output;

[0081] To evaluate the severity of historical accidents, the expression is:

[0082]

[0083] Among them, S i is the severity assessment value of the i-th historical accident, α, β and γ are the weights of the hazardous chemicals, human economic losses and environmental losses respectively, m i is the number of hazardous chemicals involved in the i-th historical accident, q ij 、r ij 、s ij , t ij and u ij are the leakage volume, toxicity level, flammability level, corrosiveness level and leakage range of the jth hazardous chemical in the i-th historical accident, v i and w i are the personal injury and economic loss of the i-th historical accident, respectively, i is the environmental impact level of the i-th historical accident;

[0084] The preprocessed data is divided into training set, validation set and test set in a ratio of 7:2:1 to train the neural network model. During the training process, the learning rate is dynamically adjusted based on the severity of historical accidents. The expression is:

[0085]

[0086] Among them, η i is the learning rate for training using the i-th historical accident, η0 is the initial learning rate, S max is the maximum severity evaluation value of all historical accidents, p is the coefficient that controls the influence of severity on learning rate, t is the current number of training steps, and T is the total number of training steps.

[0087] In this embodiment, the method of using the output of the scenario accident reconstruction model as a sub-node of the Bayesian network and constructing the Bayesian network considering the mutual influence between different causes includes:

[0088] The status of the vehicle and hazardous chemicals at the time of the accident is used as the child node in the Bayesian network, and the cause variable is used as the parent node in the Bayesian network;

[0089] Determine the mutual influence function between the cause variables as the dependency relationship of the edges connecting the parent nodes, and construct a Bayesian network based on the hierarchical relationship of the cause variables. The expression is:

[0090]

[0091] Among them, p(θ|Y) is the posterior probability, which represents the probability distribution of the model parameter θ after observing the data Y, n is the number of parameters in the model parameter vector, and θ i is the i-th parameter in the model parameter vector, p(θ i ) is θ i The prior probability of σ is the standard deviation of the noise, y r (θ) is the rth observation value predicted based on the model parameters θ, y r is the tth data point actually observed;

[0092] The output data of the scenario accident reconstruction model is used to adjust the conditional probability representation of the corresponding sub-nodes in the Bayesian network.

[0093] In this embodiment, the method of clustering vehicle trajectories, obtaining vehicle distribution in different areas, and adjusting Bayesian network parameters based on regional risks includes:

[0094] Divide the area where the vehicle operates into multiple area units;

[0095] Cluster the vehicle trajectories according to a preset threshold. For each cluster, count the number of vehicles in each area, their frequency, and their residence time in the area to obtain the regional vehicle distribution feature vector.

[0096] Conduct risk assessment for each area based on the regional vehicle distribution feature vector;

[0097] The regional risk assessment value is input into the Bayesian network as evidence to update the posterior probability of the parent node in the network. The expression is:

[0098]

[0099] Among them, p j is the jth parent node, R is the regional risk assessment value, P(p j |R) is the parent node p when the regional risk assessment value R is known. j The posterior probability of m is the number of child nodes, n is i is the i-th child node, P(R|ni ) is the value of the given child node n i Under the state, the probability of regional risk assessment value is R, (n i |p j ) is the given parent node p j In the state, child node n i The conditional probability, P(p j ) is the prior probability of the parent node, and k is the number of parent nodes.

[0100] In this embodiment, the method of performing time series analysis on the Bayesian network to obtain a chain reaction feature chain includes:

[0101] Construct a linear Gaussian state space model and use Bayesian networks to set the prior distribution of the parameters in the linear Gaussian state space model, including the prior distribution of the transformation parameters and observation parameters;

[0102] Infer the posterior probability of the model parameters of the linear Gaussian state space model based on the prior distribution to obtain the posterior samples of the parameters;

[0103] Based on the posterior samples, the temporal changes of the parent node states in the Bayesian network are analyzed, and the state transition probabilities of the parent nodes in adjacent time steps and the joint state transition probabilities of different parent node pairs in adjacent time steps are calculated;

[0104] If the state transition probability of a parent node in adjacent time steps exceeds a preset first threshold, the parent node is regarded as a key node. If the joint state transition probability of different parent node pairs in adjacent time steps exceeds a preset second threshold and both can point to the same child node, the parent node pair is regarded as a joint key node.

[0105] Starting from key nodes and joint key nodes, a chain reaction feature chain is constructed based on the state transition relationship.

[0106] In this embodiment, the method of performing iterative tracing analysis based on the chain reaction feature chain to obtain multiple tracing end points includes: starting from the child node, using a depth-first search algorithm, and performing a reverse recursive search in the Bayesian network based on the chain reaction feature chain. If the in-degree of the terminating node is 0 and is a parent node, the parent node is marked as the tracing end point.

[0107] In this embodiment, the method of calculating the causal strength of each traceability endpoint using the Bayesian network includes:

[0108] Calculate the causal strength of each traceability endpoint using the expression:

[0109]

[0110] Where ξ is the number of traceability endpoints, X iis the ith tracing end point, P(X i ) is X i The marginal probability, Pa(X i ) is X i The set of adjacent nodes, P(X i =1|X j =1) means that in X j If true, X i The conditional probability of being true, P(X j =1) is X j The prior probability of being true, P(X i =1|X j =0) means that in X j If it is false, X i The conditional probability of being true, P(X j =0) is X j The prior probability of being false.

[0111] In this embodiment, the method of analyzing the traceability endpoint with the highest causal intensity to determine the most likely accident causal path includes:

[0112] Construct a hidden Markov chain model, where the hidden state is the tracing endpoint and the observed state is the accident feature. Based on the transition probability between nodes in the Bayesian network, determine the transition probability matrix of the hidden Markov chain.

[0113] Determine the initial state probability distribution of the hidden Markov chain based on the tracing endpoint with the highest causal intensity;

[0114] Based on the initial state probability distribution, the state transition probability of each time step is calculated by the Viterbi algorithm in the hidden Markov chain. The expression is:

[0115]

[0116] Among them, V q (h) is the probability of the most likely causal path in state h at time step q, N is the total number of states, δ q-1 (i) is the initial probability distribution of being in state i at time step q-1, a ih is the probability of transitioning from state i to state h, γ q-1 is the time decay factor at time step q-1, b h (o q ) is observed in state h q The probability of q is the observation value at time step q;

[0117] The path with the highest probability is determined as the most likely accident causal path.

[0118] Figure 2 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. Figure 2 At the hardware level, the electronic device includes a processor and, optionally, an internal bus, a network interface, and memory. The memory may include internal memory, such as high-speed random-access memory (RAM), or non-volatile memory, such as at least one disk drive. Of course, the electronic device may also include other hardware required for its services.

[0119] The processor, network interface, and memory can be interconnected via an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc.

[0120] The memory is used to store programs. Specifically, the program may include program code, which includes computer operating instructions. The memory may include internal memory and non-volatile memory, and provides instructions and data to the processor.

[0121] The processor reads the corresponding computer program from the non-volatile memory into the internal memory and then runs it, forming a method for tracing the source of hazardous chemical road transport accidents based on big data at the logical level. The processor executes the program stored in the memory and is specifically configured to execute any of the aforementioned methods for tracing the source of hazardous chemical road transport accidents based on big data.

[0122] The present invention can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by hardware integrated logic circuits in the processor or by software instructions. The above processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The various methods, steps, and logic block diagrams disclosed in the embodiments of this application can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of this application can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium mature in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0123] An embodiment of the present application also proposes a computer-readable storage medium, which stores one or more programs, and the one or more programs include instructions. When the instructions are executed by an electronic device including multiple applications, they execute any of the aforementioned hazardous chemicals road transport accident tracing methods based on big data.

[0124] The above content is merely an example and explanation of the structure of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the scope of protection of the present invention.

Claims

1. A method for tracing the source of hazardous chemicals road transportation accidents based on big data, characterized in that: The following steps are involved: Pre-process historical accident reports, including data cleaning, text segmentation, and word frequency statistics; Taking the real-time status of the vehicle and the characteristics of the road environment as input, and the status of the vehicle and hazardous chemicals at the time of the accident as output, the accident scene reconstruction model is constructed based on the preprocessed data; The output of the scenario accident reconstruction model is used as a sub-node of the Bayesian network, and the Bayesian network is constructed by considering the mutual influence between different causes; the driving trajectories of the vehicles are clustered to obtain the distribution of vehicles in different areas, and the Bayesian network parameters are adjusted based on regional risks; Performing time series analysis on the Bayesian network to obtain a chain reaction feature chain; Performing iterative tracing analysis according to the chain reaction feature chain to obtain multiple tracing endpoints, and using the Bayesian network to calculate the causal strength of each tracing endpoint; Analyze the traceability endpoint with the highest causal intensity to determine the most likely accident cause path.

2. The method for tracing the source of hazardous chemicals road transport accidents based on big data according to claim 1 is characterized in that: The method of constructing an accident scene reconstruction model using a neural network based on preprocessed data includes: The real-time status characteristics of the vehicle include vehicle speed, acceleration, braking system status, steering angle, vehicle stability and tire status; the road environment characteristics include road type, road surface status, weather conditions, traffic flow and road geometry characteristics; the vehicle status at the time of the accident includes collision speed, collision angle, component integrity and vehicle body posture; the status of hazardous chemicals at the time of the accident includes hazardous chemical temperature, hazardous chemical pressure, hazardous chemical sealing status and hazardous chemical container appearance integrity; The neural network model is constructed by taking the real-time status of the vehicle and the characteristics of the road environment as input and the status of the vehicle and hazardous chemicals at the time of the accident as output; To evaluate the severity of historical accidents, the expression is: Among them, S i is the severity assessment value of the i-th historical accident, α, β and γ are the weights of the hazard of hazardous chemicals, economic losses of personnel and environmental losses, respectively, and m i is the number of hazardous chemicals involved in the i-th historical accident, q ij 、r ij 、s ij ,t ij and u ij are the leakage volume, toxicity level, flammability level, corrosiveness level and leakage range of the jth hazardous chemical in the i-th historical accident, v i and w i are the personal injury and economic loss of the i-th historical accident, respectively, i is the environmental impact level of the ith historical accident; The preprocessed data is divided into training set, validation set and test set in a ratio of 7:2:1 to train the neural network model. During the training process, the learning rate is dynamically adjusted based on the severity of historical accidents. The expression is: Among them, η i is the learning rate for training using the i-th historical accident, η0 is the initial learning rate, S max is the maximum severity evaluation value of all historical accidents, p is the influence coefficient of control severity on learning rate, t is the current training step number, and T is the total training step number.

3. The method for tracing the source of hazardous chemicals road transportation accidents based on big data according to claim 1 is characterized in that: The method of using the output of the scenario accident reconstruction model as a sub-node of the Bayesian network and constructing the Bayesian network considering the mutual influence between different causes includes: The status of the vehicle and hazardous chemicals at the time of the accident is taken as the child node in the Bayesian network, and the cause variable is taken as the parent node in the Bayesian network; Determine the mutual influence function between the cause variables as the dependency relationship of the edges connecting the parent nodes, and construct a Bayesian network based on the hierarchical relationship of the cause variables. The expression is: Among them, p(θ|Y) is the posterior probability, which represents the probability distribution of the model parameter θ after observing the data Y, n is the number of parameters in the model parameter vector, and θ i is the i-th parameter in the model parameter vector, p(θ i ) is θ i The prior probability of , σ is the standard deviation of the noise, y r (θ) is the rth observation value predicted based on the model parameters θ, y r is the tth data point actually observed; The output data of the scenario accident reconstruction model is used to adjust the conditional probability representation of the corresponding sub-nodes in the Bayesian network.

4. The method for tracing the source of hazardous chemicals road transportation accidents based on big data according to claim 1 is characterized in that: The method of clustering the driving trajectories of vehicles, obtaining the distribution of vehicles in different areas, and adjusting the Bayesian network parameters based on regional risks includes: Divide the area where the vehicle operates into a plurality of area units; Cluster the vehicle's driving trajectories according to a preset threshold. For each cluster, count the number of vehicles in each area, the frequency, and the residence time in the area to obtain the regional vehicle distribution feature vector; Conduct risk assessment for each area based on the regional vehicle distribution feature vector; The regional risk assessment value is input into the Bayesian network as evidence to update the posterior probability of the parent node in the network. The expression is: Among them, p j is the jth parent node, R is the regional risk assessment value, P(p j |R) is the parent node p when the regional risk assessment value R is known. j The posterior probability of m is the number of child nodes, n is i is the i-th child node, P(R|n i ) is the value of the given child node n i Under this condition, the probability that the regional risk assessment value is R, (n i |p j ) is the value of the given parent node p j In this state, child node n i The conditional probability, P(p j ) is the prior probability of the parent node, and k is the number of parent nodes.

5. The method for tracing the source of hazardous chemicals road transportation accidents based on big data according to claim 1 is characterized in that: The method of performing temporal analysis on the Bayesian network to obtain a chain reaction feature chain comprises: Construct a linear Gaussian state space model and use the Bayesian network to set the prior distribution of the parameters in the linear Gaussian state space model, including the prior distribution of the transformation parameters and the observation parameters; Based on the prior distribution, the posterior probability of the model parameters of the linear Gaussian state space model is inferred to obtain the posterior samples of the parameters; Based on the posterior samples, the temporal changes of the state of the parent nodes in the Bayesian network are analyzed, and the state transition probabilities of the parent nodes in adjacent time steps and the joint state transition probabilities of different parent node pairs in adjacent time steps are calculated; If the state transition probability of a parent node in adjacent time steps exceeds a preset first threshold, the parent node is used as a key node. If the joint state transition probability of different parent node pairs in adjacent time steps exceeds a preset second threshold and both can point to the same child node, the parent node pair is used as a joint key node. Starting from key nodes and joint key nodes, a chain reaction feature chain is constructed based on the state transfer relationship.

6. The method for tracing the source of hazardous chemicals road transportation accidents based on big data according to claim 1 is characterized in that: The method of performing iterative tracing analysis based on the chain reaction feature chain to obtain multiple tracing endpoints includes: starting from the child node, using a depth-first search algorithm, and performing a reverse recursive search in the Bayesian network based on the chain reaction feature chain. If the in-degree of the terminating node is 0 and it is a parent node, then marking the parent node as the tracing endpoint.

7. The method for tracing the source of hazardous chemicals road transportation accidents based on big data according to claim 1 is characterized in that: The method of calculating the causal strength of each traceability endpoint using the Bayesian network includes: Calculate the causal strength of each traceability endpoint using the expression: Among them, ξ is the number of traceability endpoints, X i is the ith tracing end point, P(X i ) is X i The marginal probability, Pa(X i ) is X i The set of adjacent nodes, P(X i =1|X j =1) indicates that in X j If true, X i The conditional probability of being true, P(X j =1) is X j The prior probability of being true, P(X i =1|X j =0) indicates that in X j If it is false, X i The conditional probability of being true, P(X j =0) is X j is the prior probability of being false.

8. The method for tracing the source of hazardous chemicals road transportation accidents based on big data according to claim 1 is characterized in that: The method of analyzing the traceability endpoint with the highest cause intensity to determine the most likely accident cause path includes: Construct a hidden Markov chain model, where the hidden state is the tracing end point and the observed state is the accident feature. Based on the transition probability between nodes in the Bayesian network, determine the transition probability matrix of the hidden Markov chain. Determine the initial state probability distribution of the hidden Markov chain according to the tracing end point with the highest causal intensity; Based on the initial state probability distribution, the state transition probability of each time step is calculated through the Viterbi algorithm in the hidden Markov chain. The expression is: Among them, V q (h) is the probability of the most likely causal path in state h at time step q, N is the total number of states, δ q-1 (i) is the initial probability distribution of being in state i at time step q-1, a ih is the probability of transitioning from state i to state h, γ q-1 is the time decay factor at time step q-1, b h (o q ) is the observation o in state h q The probability of q is the observed value at time step q; The path with the highest probability is determined as the most likely accident causative path.

9. An electronic device, comprising: processor; as well as A memory arranged to store computer executable instructions, which when executed cause the processor to perform the method according to any one of claims 1 to 8.

10. A computer-readable storage medium storing one or more programs, which, when executed by an electronic device including a plurality of application programs, enables the electronic device to execute the method according to any one of claims 1 to 8.

Images