A random forest-based spatiotemporal inference method for sudden traffic incidents

Abstract

The application relates to the technical field of traffic big data, in particular to a sudden traffic event space-time deduction method based on a random forest, which comprises the following steps: S1, feature information extraction and data fusion, wherein different-view traffic multi-source data information is fused, the extracted information is fused, and the fused information is screened and classified; S2, sudden traffic event space-time deduction processing based on a random forest, wherein a sudden event structured scene model based on a random forest is established, and a sudden event evolution model based on a random forest is established; and S3, business event analysis matching, which comprises the following steps: sudden traffic event classification according to highway management business; and matching traffic events according to known event information of a highway and by using the sudden event evolution model based on a random forest in the step S2. The application can provide a sudden traffic event space-time deduction method based on a random forest, which has low false alarm rate of sudden traffic events and high matching degree of management application.

Description

Technical Field

[0001] This invention belongs to the field of traffic big data technology, specifically involving a spatiotemporal extrapolation method for sudden traffic events based on random forest. Background Technology

[0002] Highways are characterized by being fully enclosed and operating at high speeds. When sudden traffic incidents such as collisions occur, clearing and rescue operations are extremely difficult. If management is not timely, secondary incidents often occur, and traffic congestion is easily triggered. Currently, a large number of video cameras (such as bullet cameras, PTZ cameras, and panoramic cameras) are deployed along the sides of highways. On the one hand, these sensing facilities offer high resolution and can capture sudden traffic incidents; on the other hand, video algorithm vendors have built-in numerous video recognition algorithms that identify features / behaviors such as vehicles, pedestrians, and the environment, outputting "alarm event" results.

[0003] The existing technology for providing "alarm events" has the following shortcomings:

[0004] (1) Inconsistent algorithm capabilities: Different models of equipment are deployed in different road sections, resulting in varying algorithm capabilities. The identification results for the same traffic event may differ; for example, equipment 1's algorithm identifies "vehicle stopped," while equipment 2 does not.

[0005] (2) Poor business matching: Due to the serious homogeneity of video manufacturers' products, they cannot meet specific business needs, that is, they cannot bridge the last mile of accurate identification and matching from "alarm events" to "business events". For example, for a traffic accident, "alarm events" broadcast information such as "vehicle stopped, spilled objects, driving against traffic, and no entry", while "business events" broadcast "rear-end collision" information all at once. Summary of the Invention

[0006] Most current traffic incident identification technologies focus on extracting "key alarm information" from video images, without addressing the matching and analysis problem of "alarm information" to "business information." The purpose of this invention is to provide a spatiotemporal extrapolation method for sudden traffic incidents based on random forests, which boasts a low false alarm rate and high compatibility with management applications.

[0007] For ease of description and connection of this technical solution, the explanations of some technical terms and parameter definitions are as follows.

[0008] The parameter set for custom symbols is as follows:

[0009]

[0010]

[0011]

[0012] Random forest refers to a classifier in machine learning that consists of multiple decision trees, and the class of its output is determined by the mode of the classes output by the individual trees.

[0013] Data fusion refers to an information processing technology that uses computers to automatically synthesize and analyze several observational information obtained in a time sequence in order to complete the required decision-making and evaluation tasks.

[0014] Fuzzy rules refer to the binary fuzzy relationship defined between X and Y. The expression of a fuzzy rule is: if Xi is A, then Y is B.

[0015] Information gain refers to the amount of benefit gained from using a certain feature to partition a dataset. In the training process of decision tree algorithms, the best feature branched out from the current node is usually selected recursively, starting from the root node, to construct a subtree.

[0016] Definition of "alarm event": The result of identifying a sudden event directly given by the built-in algorithm of the video camera, such as "discharge of objects", "abnormal parking", "pedestrian intrusion", etc.

[0017] Definition of "Business Event": A sudden traffic incident on the road, defined by highway managers based on business management needs. Different management departments (such as clearing and dispatching, construction and maintenance) often have different definitions and classifications of sudden traffic incidents.

[0018] To achieve the above objectives, the present invention provides the following technical solution: a spatiotemporal extrapolation method for sudden traffic events based on random forests, characterized by comprising the following steps:

[0019] Step S1: Feature information extraction and data fusion. Through network technology, traffic multi-source data information from different perspectives of various collection terminal devices on the highway is extracted and fused. The fused information is then filtered and classified according to "vehicles, pedestrians, roads, and environment".

[0020] Step S2: Spatiotemporal extrapolation and processing of sudden traffic events based on random forest. The specific extrapolation and processing are as follows:

[0021] First, a structured scenario model of emergencies based on random forest is established to describe the events themselves, including constructing a time-space description network for emergency traffic events, using random forest to handle the hierarchical processing of emergencies, and constructing and implementing a structured scenario system.

[0022] Then, a sudden event evolution model based on random forest is established to describe the event development process, including the overall path evolution mechanism analysis of sudden events and the scenario evolution path analysis of sudden traffic events.

[0023] Step S3: Business event analysis and matching, including classifying sudden traffic events according to highway management business; matching traffic events based on known highway event information and using the random forest-based sudden event evolution model in step S2; and feeding back the information to the traffic management business department through the network for handling traffic events according to the established business process.

[0024] As a further improvement of the present invention, the construction of the time-space description network for sudden traffic events specifically includes the following:

[0025] If the spatial entity [highway] that causes the event's consequences is the scene object of the sudden traffic event, define K. m The scene object describing a sudden traffic incident;

[0026] Use S n The dimensions of a sudden traffic incident scenario are represented by the adjacency matrix M, which shows the relationships between the elements in the sudden traffic incident scenario.

[0027] As a further improvement of the present invention, the specific processing of the hierarchical classification of sudden events using random forest includes the following:

[0028] Step m1: Collect relevant data on emergencies as a sample set for random forest hierarchical processing. Use multiple decision trees of the random forest model to classify the data sample set of historical emergency traffic events.

[0029] Step m2: Use a random forest model to classify emergencies. Using a random sampling method with replacement, extract a subset of event samples from the historical emergency sample set and form a classification tree. The unextracted event samples form the remaining dataset. In the extracted classification tree, each classification tree node corresponds to the scene elements of the emergency.

[0030] Step m3: Different emergency event scene elements will correspond to attributes. The scene elements of the event are converted into vectors and used as feature vectors for the classification tree. The feature vector with the greatest classification ability of each classification tree is selected as the splitting direction of the classification tree. No pruning is performed on the classification trees that are not selected.

[0031] Step m4: Repeat the above process to train the parameters of the random forest model using a sample set of historical emergencies; after determining the model parameters, classify and process the sudden traffic events.

[0032] As a further improvement of the present invention, the construction and implementation of the structured scene system specifically includes the following:

[0033] Step n1: If it is the same type of sudden traffic incident, then record it as E. i and E jThe similarity between two emergencies is calculated using the following steps:

[0034] Based on the scenario chain of the emergency, the similarity of each level of the emergency is first calculated;

[0035] Step n2: Calculate the similarity of scene nodes in the sudden traffic incident scene chain;

[0036] Step n3: Calculate the overall similarity of the attribute set of the sudden traffic incident scenario; the greater the similarity, the higher the similarity of the attribute set; after calculating the similarity of the sudden traffic incident, analyze the evolution process of the event and construct a structured scenario system;

[0037] Step n4: Based on the probability of event development, the scene system is expressed in a structured way; when a sudden traffic incident occurs, scene elements and the various attributes between elements are extracted, the scene evolution process of the sudden traffic incident is analyzed, and the structured scene of sudden traffic incident based on random forest is completed.

[0038] As a further improvement to the present invention, the analysis of the overall path evolution mechanism of the sudden event specifically includes the following:

[0039] Step p1: Divide the dynamic time state of the emergency into occurrence, development, evolution, and disappearance;

[0040] Step p2: Based on the evolution process of emergencies, the evolution mechanism is divided into occurrence mechanism, development mechanism, evolution mechanism and extinction mechanism.

[0041] As a further improvement of the present invention, the analysis of the evolution path of the sudden traffic incident scenario specifically includes the following:

[0042] Step t1: Calculate the relationship matrix between variables during the evolution of the sudden traffic event; when TS ij When ∈ [0,1], this indicates that the driving factor of the sudden event evolves the outcome of event i into the outcome of event j, and is therefore considered a positive factor; when TS ij When ∈[-1,0), this indicates that the driving factor of the sudden event promotes the evolution of the outcome of event j into the outcome of event i, and is therefore considered a negative factor; when TS ij When = 0, it is generally considered that there are no affected elements, which is a special case;

[0043] Step t2: Fuzzy rules for the direction of event development during the emergency response process, and calculation of the evolution direction under the influence of each factor during the emergency response process;

[0044] Step t3: The ID3 decision tree algorithm based on random forest predicts the evolution direction of traffic events. The ID3 decision tree algorithm classifies the sample set of emergencies and constructs a decision tree for the evolution direction of emergencies. The ID3 decision tree algorithm splits the event by using the entropy value of each sub-event sample in the sample set of each type of emergency.

[0045] Step t4: Calculate the information gain of the evolution direction of event t based on the evolution direction of the sub-events; divide the sudden events according to the calculated information gain and construct a decision tree; if the information gain value of a certain variable is the largest, take this variable as the vertex of the decision tree, and then continue to divide the vertex variables of the decision tree to complete the prediction of the evolution direction of the sudden events.

[0046] Step t5: After processing the sample event set, reconstruct the evolution process of the sudden event and calculate the similarity of the model prediction results; finally, predict the evolution direction of the sudden traffic event and complete the construction of the evolution model of the sudden traffic event based on random forest.

[0047] Compared with existing technologies, the beneficial effects of this invention are as follows: This technical solution addresses the problem of inconsistencies and mismatches between alarm events detected by video algorithm vendors and business events defined by business management. It provides a spatiotemporal extrapolation method for sudden traffic events based on a time-space network of the entire process of a sudden traffic event, catering to the personalized and differentiated business needs of highway managers. This method can significantly improve the efficiency of business event detection, reduce false alarm rates, and enhance traffic management. Furthermore, when encountering sudden traffic events such as collisions, this technical solution can quickly predict and extrapolate the traffic situation, enabling rapid clearance and rescue operations. This avoids secondary events caused by untimely management responses, which can lead to traffic congestion. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the overall process of the present invention;

[0049] Figure 2 A logic diagram is established for the random forest-based structured scenario model for sudden events in this invention;

[0050] Figure 3 This is a flowchart of the random forest model of the present invention;

[0051] Figure 4 This is a structural representation diagram of the emergency traffic incident scenario system of the present invention;

[0052] Figure 5 This is a diagram showing the correspondence between the evolution mechanism of this invention and traffic incident management.

[0053] Figure 6 This is a flowchart illustrating the random forest-based prediction of the evolution direction of sudden traffic events according to the present invention. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] As attached Figure 1 As shown, this invention provides a technical solution: a spatiotemporal extrapolation method for sudden traffic events based on random forests, comprising:

[0056] Feature information extraction and data fusion: Extract multi-source data information from different perspectives and fuse the data; after fusion, classify and summarize the information on "vehicles, pedestrians, roads, and environment";

[0057] The vehicle information is categorized into vehicle status and corresponding data for various vehicle types (e.g., buses, trucks, work vehicles, and special vehicles). Vehicle status includes normal driving, breakdown, collision, and maintenance. Vehicle data includes vehicle type, axle data, license plate data, and load data. Personnel information is categorized into workers and drivers / passengers. Road network data is divided into micro-traffic flow, meso-traffic flow, and macro-traffic flow. Road facility data includes road alignment and road markings. Environmental information is divided into road surface features and road area features. Road surface features include obstacles and potholes, while road area features include meteorological data and illumination data.

[0058] Spatiotemporal simulation of sudden traffic incidents; including establishing a structured scenario model of sudden incidents based on random forest to describe the incident itself; constructing a time-space description network for sudden traffic incidents; using random forest to handle the hierarchical classification of sudden incidents; construction and implementation of the structured scenario system; establishing an evolution model of sudden incidents based on random forest to describe the development process of the incident; overall path evolution of sudden incidents; analysis of possible evolution paths of sudden traffic incident scenarios;

[0059] Business event analysis is based on the specific definitions of sudden traffic events in highway management operations; it clarifies the prediction of traffic events based on known event information, including the simulation of one or more traffic times such as traffic accidents, traffic control, construction and maintenance, vehicle breakdowns, severe weather, obstacles, and high traffic volume, in order to execute specific processing procedures for traffic management operations.

[0060] Example 1: A Structured Scenario Model for Sudden Events Based on Random Forest [Model A]

[0061] 1.1 Constructing a temporal-spatial description network for sudden traffic events

[0062] The evolutionary logic of sudden traffic incidents is complex. For example, traffic accidents involve alarm events such as vehicle stopping, unauthorized entry into restricted areas, littering, driving against traffic, slow passage of a single vehicle, reversing, and traffic congestion. To improve the accuracy of structured scenario descriptions of sudden traffic incidents, it is first necessary to establish a structured description space that includes the [scenario attributes] and [consequence vectors] of the sudden traffic incident.

[0063] As attached Figure 2 As shown, if the spatial entity [highway] that causes the event is the scene object of the sudden traffic incident, the scene object of the sudden traffic incident is abstractly described by formula (1):

[0064] K m =(N m A m ,R m (1)

[0065] K m The scene object of a sudden traffic incident [highway]

[0066] N m : A collection of attributes of objects in a sudden traffic incident scenario [people, vehicles, environment];

[0067] A m A set of attribute states that provides a quantitative or qualitative description of a sudden traffic incident scenario. This set of attribute states includes the input attributes, output attributes, and state attributes of the structured scenario of the sudden incident.

[0068] R m : A vector representing the relationships between different attributes in a scene object.

[0069] The scenario dimension of the sudden traffic incident is represented by information according to formula (2):

[0070] S n =(C,H w (2)

[0071] S n Scene clips of sudden traffic incidents;

[0072] C: The time dimension of sudden traffic incidents;

[0073] H w : Spatial dimension representing a spatial entity;

[0074] D: Business management dimension of sudden traffic incidents.

[0075] The relationships between the various elements in a sudden traffic incident scenario can be represented by the adjacency matrix of equation (3):

[0076]

[0077] In formula (3), the adjacency matrix element m ij The value of k is determined by the relationship between various scene elements. If the event scene element k i For event scene element k j When there is an influence, then the element m in the adjacency matrix ij A value of 1 indicates an interaction between two scene elements; otherwise, the element m in the adjacency matrix... ij The value is 0. Additionally, when the element m in the adjacency matrix... ij When the value of is 1, there is a causal relationship between the two attributes in the emergency scenario; otherwise, there is no causal relationship.

[0078] 1.2 Using Random Forest for Event Classification

[0079] Collect relevant data on emergencies as a sample set for random forest hierarchical processing, and utilize multiple decision trees {h(x,θ)} of the random forest model. n The dataset of historical traffic emergencies is classified. x is the input vector of the decision tree, representing the consequence vector of the emergency in this study; n is the classification tree; θ n r represents a random vector when multiple decision trees take values ​​that are independent and identically distributed; its specific vector values ​​are determined after training the random forest model. i This represents the number of decision trees in the model.

[0080] The random forest model is used to handle the classification of emergencies. r samples are drawn from the historical emergency sample set using a random sampling method with replacement. i A subset of event samples is used to form a classification tree, and the unsampled event samples form the remaining dataset. In the extracted r... i In each classification tree, each classification tree node corresponds to a scene element of the emergency.

[0081] Different event scenario elements will correspond to several attributes. The event scenario elements are converted into vectors, which serve as feature vectors for the classification tree. The feature vector with the strongest classification ability for each classification tree is selected as the splitting direction of the classification tree. Classification trees that are not selected are not pruned.

[0082] Repeat the above process, using a sample set of historical emergencies to train the parameters of the random forest model. After determining the model parameters, classify and process sudden traffic events to construct a structured scenario system; as shown in the attached figure. Figure 3 This is a flowchart of the random forest model.

[0083] 1.3 Construction and Implementation of Structured Scene System

[0084] The similarity of sudden traffic incidents can improve the accuracy of the event scenario representation system. If the sudden traffic incidents are of the same type, they are denoted as E. i and E j The similarity between two emergencies is calculated by following these steps.

[0085] Based on the scenario chain of the emergency, the similarity of each level of the emergency is first calculated as shown in formula (4):

[0086]

[0087] In formula (4), level i To analyze sudden events E based on the random forest model i Processed classification level; level j To analyze sudden events E based on the random forest model j The processed classification level. The level similarity between two similar traffic emergencies. The value range is [0,1]. When two emergencies have the same level, the event level similarity value is 1, and the level similarity of the two emergencies is the highest.

[0088] The formula for calculating the similarity of scene nodes in the sudden traffic incident scene chain is shown in (5):

[0089]

[0090] In formula (5), N i and N j , representing the number of scene nodes in two emergency event scenario chains, are both natural integers greater than 1. The similarity of scene nodes in an emergency event scenario chain ranges from [0,1]. The similarity between similar emergency event scenario chains can be represented by the differences between scene nodes, thus highlighting the differences in the evolution and development of events.

[0091] The formula for calculating the overall similarity of the attribute set of sudden traffic incident scenarios is shown in (6):

[0092]

[0093] In formula (6), Pro i and Pro j It is a set of scene attribute vectors for two sudden traffic events. The greater the similarity, the higher the similarity of the attribute sets. After calculating the similarity of sudden traffic events, the evolution process of the events is analyzed to construct a structured scene system.

[0094] When a sudden traffic incident occurs, the preceding event guides its development. The initial diffusion force generated by the preceding event, acting on the scene space, has a unidirectional transmission effect on subsequent events. There is a coupling relationship between preceding and subsequent events. The occurrence of a particular sudden event may be caused by different inducing factors; when a single inducing factor is reached, it can trigger the event independently. Therefore, based on the probability of the event's development, according to the attached... Figure 4 The scene system is expressed in a structured way.

[0095] When a sudden traffic incident occurs, extract scene elements and the various attributes between these elements, and then follow the attached... Figure 4 The structured representation process is analyzed to understand the scenario evolution of sudden traffic events, and a structured scenario of sudden traffic events based on random forest is constructed.

[0096] Example 2:

[0097] 2. Random Forest-Based Burst Event Evolution Model [Model B]

[0098] 2.1 Analysis of the Evolution Mechanism of Sudden Traffic Incidents

[0099] In the process of the evolution of a traffic emergency, various factors such as the nature, type, level, material, scope, and area of ​​the emergency may change to varying degrees.

[0100] Studying the dynamic evolution patterns of emergencies helps decision-makers analyze the causes, development, and evolutionary paths of events in a timely manner. Based on the evolutionary process of emergencies, the evolutionary mechanisms are divided into occurrence mechanisms, development mechanisms, evolutionary mechanisms, and extinction mechanisms, specifically as follows: Figure 5 As shown.

[0101] 2.2 Fuzzy Rule Reasoning for Sudden Traffic Incidents

[0102] If the set of causative events leading to a sudden traffic incident is: IS = {IS1, IS2, IS3, ..., IS...} m The time result set of the outbreak is OS = {OS1, OS2, OS3, ..., OS}. n}, then the relationship between variables in the evolution of a sudden traffic incident can be represented by the correlation matrix equation (7):

[0103]

[0104] In formula (7), the variable TS ij The value of describes the direction of change among various factors. Here, TS can be considered as the driving factor of the sudden event.

[0105] TS ijWhen ∈(0,1], this indicates that the driving factors of the sudden event will evolve the outcome of event i into the outcome of event j, which is the so-called positive factor;

[0106] When TS ij When ∈[-1,0), this indicates that the driving factors of the sudden event promote the evolution of the outcome of event j into the outcome of event i, which is generally considered to be a negative factor;

[0107] When TS ij When = 0, it is generally considered that there are no affected elements, which is a special case.

[0108] Each element in the matrix represents a fuzzy rule for the direction of the event's development during the emergency evolution process. Therefore, under the influence of each factor during the emergency evolution process, the fuzzy rule for the direction of time evolution can be expressed by equation (8):

[0109]

[0110] Then y is OS B

[0111] Fact:IS→consequence:OS

[0112] In equation (8), and OS B It is the set of causes and results of the evolution of emergency events containing fuzzy attribute values. X = (x1, x2, ..., x...) n ) T It is a description of the evolutionary driving force of sudden events, that is, X is related to the causal set. Correspondingly, y represents the evolutionary result OS. B The description uses the superscript B to indicate the number of fuzzy rules. Based on these fuzzy rules, the evolution process of the sudden event was reconstructed.

[0113] 2.3 Prediction of the Evolution Trend of Sudden Traffic Events Based on Random Forest

[0114] The ID3 decision tree algorithm based on random forest predicts the evolution direction of traffic events. The ID3 decision tree algorithm constructs a decision tree for the evolution direction of sudden events by classifying the sample set of sudden events. The ID3 decision tree algorithm uses the entropy value of each sub-event sample in each type of sudden event sample set for splitting. The entropy value is calculated by formula (9):

[0115]

[0116] Equation (9) is the formula for calculating the total information entropy value.

[0117] P iIt is the frequency of occurrence of emergency response evolution direction i.

[0118] c is the total number of evolution directions for each sub-event sample in the sudden event sample set.

[0119] Calculate the information gain of the evolution direction of event t based on the evolution direction of the sub-events. Then, classify the sudden events based on the calculated information gain and construct a decision tree. Typically, if a variable has the largest information gain, this variable is chosen as the vertex of the decision tree. The vertex variables of the decision tree are then further divided to predict the evolution direction of the sudden events.

[0120]

[0121] Equation (10) is mainly used to calculate the characteristic value in the direction of a certain sudden event prediction. Then, the obtained H is used... j The difference between the value of H0 and the value of H0 is then used for further prediction.

[0122]

[0123] In equation (11), G(t) represents the information acquired in the direction of event evolution, |N j | refers to the number of sample sets for the development direction of the j-th event, and |N| is the total number of sample sets for the development directions of events.

[0124] After processing the sample event set, such as Figure 6 The process shown reconstructs the evolution of an emergency situation:

[0125]

[0126] In the similarity calculation formula (12), N t G represents the actual direction of the evolution of a sudden event. i (t) represents the similarity of the model prediction results under the influence of i parameters.

[0127] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A spatiotemporal extrapolation method for sudden traffic events based on random forest, characterized in that, Specifically, the steps include the following: Step S1: Feature information extraction and data fusion. Through network technology, traffic multi-source data information from different perspectives of various collection terminal devices on the highway is extracted and fused. The fused information is then filtered and classified according to "vehicles, pedestrians, roads, and environment". Step S2: Spatiotemporal extrapolation and processing of sudden traffic events based on random forest. The specific extrapolation and processing are as follows: First, a structured scenario model of emergencies based on random forest is established to describe the events themselves, including constructing a time-space description network for emergency traffic events, using random forest to handle the hierarchical processing of emergencies, and constructing and implementing a structured scenario system. Then, a sudden event evolution model based on random forest is established to describe the event development process, including the overall path evolution mechanism analysis of sudden events and the scenario evolution path analysis of sudden traffic events. Step S3: Business event analysis and matching, including classifying sudden traffic events according to highway management business; matching traffic events based on known highway event information and using the random forest-based sudden event evolution model in step S2; and feeding back the information to the traffic management business department for handling traffic events according to the established business process. The construction of the time-space description network for sudden traffic events specifically includes the following: If the spatial entity [highway] that causes the event is the scene object of the sudden traffic incident, the scene object of the sudden traffic incident is abstractly described by equation (1): ； The scene object of a sudden traffic incident [highway]; : A collection of attributes of objects in a sudden traffic incident scenario [people, vehicles, environment]; A set of attribute states that provides a quantitative or qualitative description of a sudden traffic incident scenario. This set of attribute states includes the input attributes, output attributes, and state attributes of the structured scenario of the sudden incident. : A vector representing the relationships between different attributes in a scene object; The scenario dimension of the sudden traffic incident is represented by information according to formula (2): ； Scene clips of sudden traffic incidents; The time dimension of sudden traffic incidents; w : Spatial dimension representing a spatial entity; : Business management dimension of sudden traffic incidents; The relationships between the elements in a sudden traffic incident scenario are represented by the adjacency matrix of equation (3): ； In equation (3), the adjacency matrix elements The value of is determined by the relationship between various scene elements. If the event scene elements Event scene elements When there is an influence, the elements in the adjacency matrix... A value of 1 indicates an interaction between two scene elements; otherwise, the elements in the adjacency matrix... The value is 0; additionally, when the element in the adjacency matrix... When the value of is 1, there is a causal relationship between the two attributes in the emergency scenario; otherwise, there is no causal relationship.

2. The spatiotemporal extrapolation method for sudden traffic events based on random forest according to claim 1, characterized in that: The process of using random forests to handle the classification of sudden events includes the following specific steps: Step m1: Collect relevant data on sudden events as a sample set for random forest hierarchical processing, and utilize multiple decision trees of the random forest model. Classify the data sample set of historical traffic emergencies; in The input vector for the decision tree serves as the consequence vector for the unexpected event; For classification trees; This is a random vector when multiple decision trees take values ​​that are independent and identically distributed. The specific vector values ​​are determined after the random forest model is trained. Step m2: Use This indicates the number of decision trees in the model. A random forest model is used to handle the classification of emergencies, drawing trees from a historical sample set of emergencies using a random sampling method with replacement. A subset of event samples is used to form a classification tree, and the unextracted event samples form the remaining dataset; in the extracted... In each classification tree, each classification tree node corresponds to a scene element of the emergency; Step m3: Different emergency event scene elements will correspond to attributes. The scene elements of the event are converted into vectors and used as feature vectors for the classification tree. The feature vector with the greatest classification ability of each classification tree is selected as the splitting direction of the classification tree. No pruning is performed on the classification trees that are not selected. Step m4: Repeat the above process to train the parameters of the random forest model using a sample set of historical emergencies; after determining the model parameters, classify and process the sudden traffic events.

3. The spatiotemporal extrapolation method for sudden traffic events based on random forest according to claim 1, characterized in that: The construction and implementation of the structured scene system specifically includes the following: Step n1: If the sudden traffic incidents are of the same type, they are recorded as follows: and Calculate the similarity between two emergencies using the following steps: Based on the scenario chain of the emergency, the similarity of each level of the emergency is first calculated as shown in equation (4): ； In equation (4), To analyze sudden events using the random forest model The processed classification level; To analyze sudden events using the random forest model The processed classification level; the level similarity between two similar traffic emergencies. The value range is [0,1]. When two emergencies have the same level, the event level similarity value is 1, and the level similarity between the two emergencies is the highest. Step n2: The similarity calculation formula for scene nodes in the sudden traffic incident scene chain is shown in (5): ； In equation (5), and These represent the number of scene nodes in two emergency event scenario chains, both being natural integers greater than 1; the similarity of scene nodes in the emergency event scenario chain ranges from [0,1]; the similarity between emergency event scenario chains of the same type is represented by the difference between scene nodes, thus highlighting the differences in the evolution and development of events; Step n3: The formula for calculating the overall similarity of the set of attributes of sudden traffic incident scenarios is shown in (6): ； In equation (6), and It is a set of scene attribute vectors of two sudden traffic events. The greater the similarity, the higher the similarity of the attribute sets. After calculating the similarity of sudden traffic events, the evolution process of the events is analyzed to construct a structured scene system. Step n4: Based on the possibilities of event development, express the scene system in a structured way; When a sudden traffic incident occurs, the scene elements and their attributes are extracted, and the scene evolution process of the sudden traffic incident is analyzed according to the structured expression process to complete the construction of a structured scene of the sudden traffic incident based on random forest.

4. The spatiotemporal extrapolation method for sudden traffic events based on random forest according to claim 1, characterized in that: The analysis of the overall path evolution mechanism of the emergency includes the following: Step p1: Divide the dynamic time state of the emergency into occurrence, development, evolution, and disappearance; Step p2: Based on the evolution process of emergencies, the evolution mechanism is divided into occurrence mechanism, development mechanism, evolution mechanism and extinction mechanism.

5. The spatiotemporal extrapolation method for sudden traffic events based on random forest according to claim 1, characterized in that: The analysis of the evolution path of the sudden traffic incident scenario specifically includes the following: Step t1: If the set of causal events leading to the sudden traffic incident is: The time result set of the outbreak is Then, the relationship between variables in the evolution of a sudden traffic incident can be represented by the correlation matrix formula (7): ； In equation (7), the variable The value of describes the direction of change among various factors; here it is considered that... It is a driving factor for sudden events; when This indicates that the driving factors of a sudden event will affect the event itself. The result evolved into an event The result was therefore considered a positive factor; when This indicates that the driving factors of the sudden event facilitated the event. The result evolved into an event The result was therefore considered a negative factor; when At that time, it is considered that there are no influencing elements, which is a special case; Step t2: Each element in the correlation matrix of equation (7) represents a fuzzy rule for the direction of the event development outcome during the emergency evolution process. Then, under the influence of each factor during the emergency evolution process, the fuzzy rule for the direction of time evolution is expressed by equation (8): ； In equation (8), and It is the set of causes and results of the evolution of emergency events containing fuzzy attribute values; It is a description of the evolutionary driving force of sudden events, that is... Is the set of causes Correspondingly; Regarding the evolutionary results Description; superscript The number of fuzzy rules is represented; based on the above fuzzy rules, the evolution process of the sudden event is reconstructed; Step t3: The ID3 decision tree algorithm based on random forest predicts the evolution direction of traffic events. The ID3 decision tree algorithm classifies the sample set of sudden events and constructs a decision tree for the evolution direction of sudden events. The ID3 decision tree algorithm uses the entropy value of each sub-event sample in each type of sudden event sample set for splitting. The entropy value is calculated by equation (9): ； Equation (9) is the formula for calculating the total information entropy; where It is the direction of emergency response evolution. The frequency of occurrence, It is the total number of evolution directions for each sub-event sample in the sudden event sample set; Step t4: Calculate the event based on the evolution direction of the sub-events. Information gain in the direction of evolution; classify sudden events based on the calculated information gain and construct a decision tree; If a variable has the highest information gain, it is selected as the vertex of the decision tree. The vertex variables are then further partitioned to predict the direction of the event's evolution, and the eigenvalues ​​for that predicted direction are calculated. : ； Then use the results Value and The values ​​are subtracted separately and then further predicted. ； In equation (11), It is information obtained from the direction of event evolution. Refers to the first The number of sample sets representing the development direction of each event. It is the total number of samples representing the direction of event development; Step t5: After processing the sample event set, reconstruct the evolution process of the sudden events: ； Equation (12) is the formula for calculating similarity. Indicates the actual direction of the evolution of the emergency. Indicates in The similarity of the prediction results of the model is calculated under the influence of several parameters; finally, the evolution direction of the sudden traffic event is predicted, and the evolution model of the sudden traffic event based on random forest is finally completed.

Images