A method for intelligent correlation analysis and prediction of highway events

Abstract

The present application provides a method for intelligent correlation analysis and prediction of highway events, comprising: constructing an event spatiotemporal correlation network based on the spatiotemporal distribution characteristics of events, wherein nodes represent events and edges represent correlations between events; training the event spatiotemporal correlation network using a graph neural network model to obtain the correlation evolution laws of events in time and space; if an event that may occur in the future is predicted, then inferring other related events that may be triggered by the event based on the event spatiotemporal correlation network to form an event causal logic chain; using a reinforcement learning algorithm based on the event causal logic chain to evaluate the effectiveness of different response strategies and obtain an event handling plan; during the occurrence and evolution of events, continuously acquiring vehicle trajectory data and road condition monitoring data, and dynamically updating the event spatiotemporal correlation network and prediction model.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method for intelligent correlation analysis and prediction of highway events. Background Art

[0002] Intelligent correlation analysis and prediction of highway events face the dual technical challenges of complex spatiotemporal data and diverse event correlation patterns. Massive amounts of vehicle trajectory data and road condition monitoring data are distributed across continuous spatiotemporal dimensions, and different types of events intersect and influence each other in both time and space. Accurately characterizing the spatiotemporal characteristics of events from this spatiotemporal data and exploring the evolutionary patterns of correlations between different events are key to intelligent event correlation analysis and prediction. Traditional spatiotemporal data analysis methods, primarily based on discretization using fixed time intervals and spatial grids, struggle to adapt to the continuous and dynamic nature of event spatiotemporal characteristics. Simply increasing spatiotemporal resolution leads to data sparsity and an explosion in computational complexity. Therefore, a new spatiotemporal data representation and computational paradigm is urgently needed that can adaptively capture the occurrence and evolution characteristics of events across spatiotemporal dimensions and enable event correlation analysis and prediction with reduced spatiotemporal complexity. Furthermore, due to differences in spatiotemporal scale, correlation strength, and causal logic among different event types, their correlation patterns are diverse and dynamic. Another pressing technical challenge is how to automatically learn and extract typical correlation patterns from complex spatiotemporal event networks, and use these patterns to implement cross-event correlation reasoning and prediction. Existing correlation pattern mining methods are mostly based on static network structures and lack the ability to model the temporal dependencies and dynamic evolution of events. This makes it difficult to capture the dynamic changes in the causal logic chain and correlation strength of events. Summary of the Invention

[0003] The present invention provides a method for intelligent correlation analysis and prediction of highway events, which mainly includes:

[0004] Obtain vehicle trajectory data and road condition monitoring data, cluster the vehicle trajectory data and road condition monitoring data using an adaptive spatiotemporal clustering algorithm to obtain the spatiotemporal distribution characteristics of events; construct an event spatiotemporal association network based on the spatiotemporal distribution characteristics of events, where nodes represent events and edges represent associations between events; use a graph neural network model to train the event spatiotemporal association network to obtain the correlation evolution laws of events in time and space; based on the spatiotemporal distribution characteristics and correlation evolution laws of events, construct a sequence prediction model using an attention mechanism to predict events that may occur in the future and their spatiotemporal distribution probabilities; if an event that may occur in the future is predicted, then based on the event spatiotemporal association network, infer other related events that may be triggered by the event to form an event causal logic chain; based on the event causal logic chain, use a reinforcement learning algorithm to evaluate the effects of different response strategies to obtain an event handling plan; during the occurrence and evolution of events, continuously obtain vehicle trajectory data and road condition monitoring data, and dynamically update the event spatiotemporal association network and prediction model. The technical solution provided by the embodiment of the present invention may include the following beneficial effects:

[0005] The present invention discloses a method for intelligent correlation analysis and prediction of highway events. The method obtains vehicle trajectory and road condition monitoring data, uses an adaptive spatiotemporal clustering algorithm to capture event features, and constructs an event spatiotemporal correlation network. A graph neural network model is used to mine event correlation patterns, and combined with a sequence prediction model of an attention mechanism, it predicts possible future events and their distribution. For predicted events, the present invention infers the associated events that may be triggered, forms a causal logic chain, and uses a reinforcement learning algorithm to evaluate response strategies to derive the optimal disposal plan. By continuously updating data and models, the present invention realizes real-time correlation analysis, accurate prediction, and intelligent disposal of highway events, effectively improving the efficiency and safety of highway traffic management. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 This is a flow chart of the intelligent correlation analysis and prediction of highway events of the present invention.

[0007] Figure 2 Schematic diagram of the intelligent correlation analysis and prediction of highway events according to the present invention.

[0008] Figure 3 This is another schematic diagram of the intelligent correlation analysis and prediction of highway events according to the present invention.

[0009] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0010] like Figure 1-3 In this embodiment, the intelligent correlation analysis and prediction of highway events may specifically include:

[0011] Step S101 : Acquire vehicle trajectory data and highway road condition monitoring data, and use an adaptive spatiotemporal clustering algorithm to adaptively capture the occurrence and evolution characteristics of events in the spatiotemporal dimension to obtain the spatiotemporal distribution characteristics of the events.

[0012] Acquire vehicle trajectory data and highway road condition monitoring data, pre-process the vehicle trajectory data and highway road condition monitoring data, remove noise through Kalman filtering, and eliminate outliers through box plot method to improve data quality; construct the feature vector required by the adaptive spatiotemporal clustering algorithm based on the timestamp and longitude and latitude information of vehicle trajectory data, as well as the timestamp and spatiotemporal characteristics of road condition monitoring data and road congestion index; use DBSCAN algorithm to cluster the feature vectors, and dynamically adjust the eps and minPts parameters to enable DBSCAN algorithm to adaptively capture the characteristics and evolution characteristics of event occurrence; post-process the DBSCAN clustering results, and The density and distance characteristics of events are used to eliminate noise clusters, merge clusters with density greater than a preset threshold and close distance, and obtain the temporal and spatial distribution characteristics of events; according to the temporal and spatial distribution characteristics of events, the time series analysis method is used to analyze the temporal pattern of event occurrence, and the spatial interpolation method is used to analyze the spatial distribution pattern of events, and the time and place of event occurrence are predicted; the temporal and spatial distribution characteristics of events are combined with map information to generate an event distribution heat map to display high-incidence areas of events; according to the event prediction results and the event distribution heat map, the duration of traffic lights is adjusted, traffic guidance information and traffic control measures are issued to guide vehicles to detour congested sections, relieve traffic pressure, and improve road traffic efficiency.

[0013] For example, obtaining vehicle trajectory data and highway road condition monitoring data is the cornerstone of intelligent transportation systems. Vehicle trajectory data typically includes vehicle ID, timestamp, latitude and longitude information, while highway road condition monitoring data covers road section ID, timestamp, congestion index, etc. Taking a city highway as an example, suppose that one day's vehicle trajectory data and road condition monitoring data are obtained. First, the data is preprocessed. Preprocessing includes data cleaning and formatting. For example, there may be missing values ​​or outliers in the vehicle trajectory data, such as a vehicle traveling an unreasonably long distance in a short period of time. The Kalman filter algorithm can effectively remove these noises. Kalman filtering uses a recursive method to combine predicted values ​​and observed values ​​to continuously optimize the estimated values, thereby smoothing the data. Suppose a vehicle's latitude and longitude data is (34.05, 118.25), (34.10, 118.30), and (34.20, 118.40). After Kalman filtering, the resulting smoother trajectory might be (34.05, 118.25), (34.08, 118.28), and (34.15, 118.35). Next, a boxplot is used to remove outliers. Boxplots identify outliers using quartiles and the boxplot rule. Suppose the congestion index data for a road section is [1.2, 1.5, 1.3, 5.0, 1.4]. A boxplot analysis might identify 5.0 as an outlier and remove it. Constructing a feature vector is key to the adaptive spatiotemporal clustering algorithm. The feature vector should contain spatiotemporal features such as timestamp, latitude and longitude, and congestion index. For example, a vehicle at timestamp 1609459200 (2021-01-01 00:00:00) was at latitude and longitude (34.05, 118.25), and the corresponding road congestion index was 1.5. The feature vector is [1609459200, 34.05, 118.25, 1.5]. The DBSCAN algorithm is used for clustering. DBSCAN dynamically adjusts clustering parameters by setting the eps (neighborhood radius) and minPts (minimum number of samples). Assuming eps is set to 0.1 and minPts is set to 5, DBSCAN will cluster based on the distance and density of the feature vectors, identifying high-density areas where events occurred. Post-processing of the DBSCAN clustering results is performed. Based on the density and distance characteristics of the clusters, noisy clusters are removed and clusters with high density and close proximity are merged. For example, clusters A and B, which have high density and are close together, are merged into a new cluster C, reflecting the temporal and spatial distribution of events. Time series analysis methods, such as the ARIMA model, can be used to analyze the temporal patterns of events. For example, if the time series data for congestion events on a certain road section is [1.2, 1.3, 1.4, 1.5, 1.6], using the ARIMA model to predict the congestion index for future time periods reveals that the congestion index is higher during peak hours in the morning and evening. Spatial interpolation methods, such as Kriging, can be used to analyze the spatial distribution of events.Suppose the congestion index data for multiple road sections in a certain area is (34.05, 118.25, 1.5) and (34.10, 118.30, 1.6). Kriging interpolation is used to predict the congestion index for unmonitored sections, revealing a high congestion index in a certain area. Combining the spatiotemporal distribution of events with map information, a heat map is generated using the folium library. Suppose the heat map shows severe congestion on a city's ring road, with a deep red color, visually highlighting areas with high incident rates. Based on the event prediction results and heat map analysis, traffic light durations are adjusted and traffic guidance information is issued. For example, if the congestion index for a certain road section is predicted to reach 1.8 within the next hour, traffic light durations can be adjusted in advance to redirect vehicles, alleviating congestion and improving traffic efficiency. These steps not only improve data quality but also enable accurate prediction and visualization of traffic events, providing strong support for traffic control, ultimately alleviating traffic pressure and improving road efficiency.

[0014] Step S102: construct an event spatiotemporal association network based on the spatiotemporal distribution characteristics of the events. The nodes in the event spatiotemporal association network represent events, and the edges represent the associations between events.

[0015] Obtain the time attribute data and space attribute data of multiple events, where the time attribute data is the timestamp of the event, and the space attribute data is the geographic coordinates of the event. The time attribute data and the space attribute data constitute the original event data; for each event, extract the time characteristics and space characteristics of the event from the original event data, where the time characteristics include the time stamp of the event, and the space characteristics include the geographic coordinates of the location where the event occurs; combine the time characteristics and space characteristics of each event to construct the time-space feature vector of the event, where each event corresponds to a time-space feature vector; calculate the distance between the time-space feature vectors of multiple events, and construct an event time-space association matrix, where the element values ​​of the event time-space association matrix are the distances between events; if the distance value in the event time-space association matrix is ​​less than or equal to a preset time distance threshold and less than or equal to a preset space distance threshold, then set the distance value of the position in the event time-space association matrix to 1, otherwise set it to 0, and obtain the event time-space association adjacency matrix; calculate the connected components of the event time-space association adjacency matrix to obtain a set of several events, where a set of events represents an event community. ; For each event community, obtain the spatiotemporal feature vectors of all events in the event community, and calculate the average spatiotemporal feature vector of the event community based on the spatiotemporal feature vectors of all events in the event community; for each event community, construct a time series of the spatiotemporal feature vectors of the event community based on the average spatiotemporal feature vector of the event community; perform time series analysis on the time series of the spatiotemporal feature vectors of the event community, calculate the trend of the time series, and obtain the evolution trend of the event community; for the evolution trend of the event community, calculate the predicted value at the current moment, and if the absolute value of the difference between the actual value of the time series of the spatiotemporal feature vector of the event community at the current moment and the predicted value at the current moment is greater than a preset threshold, then the event community is determined to be an abnormal event community; obtain the attribute data of the events contained in the abnormal event community, and the attribute data of the events include event type and event level; determine the geographical distribution range of the abnormal event community based on the spatial characteristics of the events in the abnormal event community; determine the impact degree of the abnormal event based on the attribute data of the event and the weight value of the event community; generate abnormal event warning information based on the geographical distribution range and impact degree of the abnormal event.

[0016] For example, first, obtaining the event's temporal and spatial attribute data is the basis for constructing the event's spatiotemporal feature vector. For example, in an intelligent transportation system, the event's temporal attribute data may be the timestamp of a vehicle accident, such as "2023-10-01 08:30:00," while the spatial attribute data may be the geographic coordinates of the accident, such as the latitude and longitude (34.05, 118.25). Together, these data constitute the raw event data. Temporal and spatial features are extracted from the raw event data. Temporal features include the timestamp of the event, while spatial features include the geographic coordinates of the event. For example, a traffic accident has a timestamp of "2023-10-01 08:30:00" and geographic coordinates of (34.05, 118.25). These features are combined to construct the event's spatiotemporal feature vector, such as [1664587400, 34.05, 118.25], where 1664587400 is the timestamp in Unix format. Euclidean distance is typically used to calculate the distance between the spatiotemporal feature vectors of two events. Suppose there are two events, A and B, whose spatiotemporal feature vectors are [1664587400, 34.05, 118.25] and [1664588000, 34.10, 118.30], respectively. Calculating their Euclidean distance can help understand the spatiotemporal proximity between the events. Based on the calculated distances, an event spatiotemporal association matrix is ​​constructed. The matrix elements represent the distances between the events. For example, if the distance between events A and B is 0.15, 0.15 is entered in the corresponding position in the matrix. Preset temporal and spatial distance thresholds are set, such as a temporal threshold of 300 seconds (5 minutes) and a spatial threshold of 0.1 degrees. If the distance between events is less than or equal to these thresholds, the distance value at that position in the matrix is ​​set to 1; otherwise, it is set to 0. This processed matrix is ​​called the event spatiotemporal association adjacency matrix. By calculating the connected components of the event spatiotemporal association adjacency matrix, event communities can be identified. For example, calculations reveal that events A, B, and C form a connected component, or event community, indicating a high degree of spatiotemporal correlation between these events. The number of events within the event community is calculated as the weight of the event community. For example, event community ABC contains three events and has a weight of 3. For each event community, the spatiotemporal feature vectors of all events within it are obtained, and the average spatiotemporal feature vector is calculated using the arithmetic mean method. Assume that the average spatiotemporal feature vector of event community ABC is [1664587700, 34.08, 118.28]. Based on the average spatiotemporal feature vector, a time series of event community spatiotemporal feature vectors is constructed. For example, the average spatiotemporal feature vectors of event community ABC at three consecutive time points are [1664587700, 34.08, 118.28], [1664588600, 34.09, 118.29], and [1664589500, 34.10, 118.30], forming a time series.Perform linear regression analysis on the time series to calculate its trend. For example, linear regression reveals that the average longitude and latitude of event community ABC gradually shift northward over time, indicating that events in this community may be spreading northward. Calculate the predicted value for the current moment. If the absolute difference between the actual and predicted values ​​exceeds a preset threshold, the event community is identified as an anomalous event community. For example, the actual values ​​at the current moment are [1664590400, 34.12, 118.32] and the predicted values ​​are [1664590400, 34.10, 118.30]. The absolute difference is 0.02, exceeding the preset threshold of 0.01, and is therefore identified as an anomaly. Attribute data, such as event type and severity, is obtained based on the events included in the anomalous event community. For example, event community ABC contains three traffic accidents with severity levels of intermediate, high, and high. Based on the spatial characteristics of the events, the geographic distribution range of the anomalous event community is determined, for example, from (34.05, 118.25) to (34.12, 118.32). The impact of abnormal events is determined based on the event attribute data and community weights. For example, event community ABC has a weight of 3 and contains two high-level accidents, which indicates a high impact. Based on the potential impact range and impact, abnormal event warning information is generated. For example, an alert might be issued: "Multiple traffic accidents have occurred in the area from (34.05, 118.25) to (34.12, 118.32). The impact is high, and detours are recommended." These steps not only effectively identify and warn abnormal event communities, but also provide strong support for traffic management and emergency response, improving the safety and efficiency of the transportation system.

[0017] In step S103, a graph neural network model is used to train the event spatiotemporal correlation network to obtain the correlation evolution rules of events in time and space, and to mine the correlation patterns between different events.

[0018] Acquire multiple event data, which include the time and space attributes of events; construct an event spatiotemporal association network based on the event data, in which nodes represent events and edges represent the association relationship between events; use graph convolutional neural network to perform representation learning on the event spatiotemporal association network, and obtain low-dimensional vector representations of event nodes, which capture the association patterns of events in time and space; in the training process of graph convolutional neural network, use the timestamp information of event nodes to learn the hidden state representation of events at different time steps, and explore the evolution law of events in the time dimension; in the training process of graph convolutional neural network, use the geographic location information of event nodes, through spatial graph convolution operation, learn the association pattern of events in the spatial dimension, and obtain different The correlation between events in geographical locations; based on the event node representation learned by the graph convolutional neural network, the cosine similarity is used to calculate the similarity between events and discover highly correlated event sets; cluster analysis is performed on highly correlated event sets to obtain different types of event correlation patterns; event correlation patterns are applied to event prediction, and for current events, based on their spatiotemporal correlation, historical events that are highly correlated with them are found in the event spatiotemporal correlation network, and the subsequent evolution of the correlated events is used to predict the future development trend of the current event; event correlation patterns are applied to anomaly detection, and by comparing the similarity between the current event and the known correlation pattern, abnormal events that do not conform to the known pattern are identified, thereby realizing real-time detection and early warning of abnormal events.

[0019] For example, based on the temporal and spatial attributes of events, an event spatiotemporal association network is constructed, where nodes represent events and edges represent the associations between events. For example, in an intelligent transportation system, each traffic accident is a node, and the edges between nodes represent the temporal and spatial proximity of these accidents. Suppose there are three traffic accidents, A, B, and C. A occurs at 08:30 on October 1, 2023, at (34.05, 118.25); B occurs at 08:35 on October 1, 2023, at (34.06, 118.26); and C occurs at 09:00 on October 1, 2023, at (34.10, 118.30). Based on this information, a spatiotemporal association network containing three nodes can be constructed. Due to their temporal and spatial proximity, the edge weight between nodes A and B is high, while the edge weight between nodes A and C is relatively low. A graph convolutional neural network (GCN) is used to learn representations for event spatiotemporal association networks. It learns low-dimensional vector representations of event nodes and captures the temporal and spatial correlation patterns of events. Through graph convolution operations, GCN effectively aggregates information about node neighbors to learn node embeddings. For example, for node A, GCN considers information about its connected node B. Through convolution, the representation vector of node A incorporates the feature information of node B, thereby capturing the temporal and spatial correlation between A and B. During GCN training, the timestamp information of event nodes is used to learn hidden state representations of events at different time steps, thereby exploring the temporal evolution of events. For example, the timestamps of nodes A and B are 1664587400 and 1664587700, respectively. The GCN model uses time encoding techniques to convert these timestamps into time embedding vectors. During training, it learns the correlation between these vectors, thereby capturing the temporal evolution of events. Furthermore, by leveraging the geographic location information of event nodes and performing spatial graph convolution operations, it learns spatial correlation patterns, specifically the relationships between events at different locations. For example, the geographic coordinates of nodes A and C are (34.05, 118.25) and (34.10, 118.30), respectively. The GCN model uses spatial encoding techniques to convert these coordinates into spatial embedding vectors and considers the spatial relationship between these vectors in convolution operations, thereby capturing the spatial correlation patterns of events. Based on the event node representations learned by the GCN model, cosine similarity is used to calculate the similarity between events and discover highly correlated event sets. For example, the representation vectors of nodes A and B are [0.1, 0.2, 0.3] and [0.11, 0.21, 0.31], respectively. By calculating cosine similarity, it is found that their similarity is high, indicating that A and B are highly correlated in space and time. K-means clustering analysis is performed on highly correlated event sets to automatically discover different types of event correlation patterns.For example, K-means clustering can be used to classify highly correlated events into two categories: traffic accidents concentrated in a specific time period and geographic location, and traffic accidents dispersed across different time periods and geographic locations. This reveals different types of event correlation patterns. The mined event correlation patterns are then applied to event prediction and anomaly detection. For a current event, based on its spatiotemporal correlation, highly correlated historical events are identified within the event correlation network. The subsequent evolution of these correlated events is then used to predict the current event's likely future development trends. For example, if the spatiotemporal characteristics of current event D are highly correlated with those of historical event A, the subsequent evolution of A (such as incident handling time and traffic restoration) can be used to predict D's future development trends. Furthermore, by comparing the similarity between the current event and known correlation patterns, anomalous events that do not conform to these patterns can be identified, enabling real-time detection and early warning of these events. For example, if the spatiotemporal characteristics of current event E differ significantly from known correlation patterns, similarity calculations reveal that it does not conform to any known patterns, thus determining it as an anomalous event and issuing an immediate warning. The above steps not only effectively capture the temporal and spatial correlation patterns of events, but also enable the prediction of event trends and real-time detection of anomalies, providing strong support for traffic management and emergency response. For example, in intelligent transportation systems, through GCN model learning and cluster analysis, frequent traffic accidents in a certain area can be promptly detected and their development trends can be predicted. Traffic diversion measures can be implemented in advance to reduce the impact of accidents on traffic flow and improve the safety and efficiency of the transportation system. The implementation of this method not only relies on the GCN model's effective capture of spatiotemporal correlations, but also requires the combined application of techniques such as cosine similarity and K-means clustering to achieve in-depth exploration of event correlation patterns and accurate identification of anomalies. The organic combination of these techniques can uncover hidden event correlation patterns in complex spatiotemporal data, providing strong technical support for event prediction and anomaly detection.

[0020] In step S104, a sequence prediction model using an attention mechanism is constructed based on the spatiotemporal distribution characteristics and association patterns of events to predict events that may occur in the future and their spatiotemporal distribution probabilities.

[0021] Obtain historical highway event data, extract features of the time, space, and type attributes of the events, and obtain event feature vectors; use clustering algorithms to cluster event feature vectors and classify similar events into the same category; based on the clustering results, analyze the association patterns between events in the same category and construct an event association graph, in which nodes represent events and edges represent the association relationships between events; use association rule mining algorithms to discover frequent association patterns and causal relationships between events; select LSTM as the basic sequence prediction model, and introduce an attention mechanism to dynamically adjust the weights of different event attributes; use event feature vectors and event association graphs as inputs to the LSTM model to train the prediction model; in the attention In the implementation of the mechanism, the Softmax function is used to normalize the attention weights and perform weighted summation with the event feature vector to obtain the attention-weighted event representation; the attention weights are learned through the backpropagation algorithm, so that the model can adaptively focus on different event attributes; the trained LSTM prediction model is used to predict events that will occur in the future based on the current time and space range; the predicted events are post-processed, and the frequency of events under different conditions is counted based on historical data, and the probability of events under given conditions is calculated using the Bayesian formula; data visualization tools are used to present the prediction results and probability distribution in an intuitive way, providing a reference for decision-making and risk prevention and control.

[0022] For example, suppose traffic accident data from the past year is obtained from highway management. This data includes information such as the time, location, accident type (e.g., rear-end collision, rollover), weather conditions, and traffic volume. Feature extraction is performed on this data, converting each event into a multidimensional feature vector. For example, the feature vector for event A might include a timestamp, latitude and longitude coordinates, an accident type code, a weather condition code, and traffic volume. Clustering these event feature vectors using a clustering algorithm allows similar events to be grouped into the same category. Suppose, using the K-means clustering algorithm, the events are classified into three categories: rear-end collisions concentrated during the morning rush hour; rollovers occurring at night; and multi-vehicle pileups on rainy days. This approach allows for preliminary discovery of correlations between events, identifying patterns in accident occurrence under certain times and weather conditions. Based on the clustering results, the correlation patterns between events within the same category are further analyzed to construct an event correlation graph. For example, in the first category of rear-end collisions, multiple accidents were found to occur at different times on the same road section, indicating a higher risk during rush hour on this road section. The nodes in the association graph represent these rear-end collisions, and the edges indicate their temporal and spatial proximity. Association rule mining algorithms, such as the Apriori algorithm, are used to discover frequent association patterns. For example, the probability of a rear-end collision is higher under the following conditions: "morning rush hour, road section X, and heavy traffic." An LSTM is chosen as the underlying sequence prediction model, and an attention mechanism is introduced to dynamically adjust the weights of different event attributes. Assuming the event feature vector consists of 10 dimensions, the attention mechanism assigns different weights to each dimension based on its importance. For example, when predicting rear-end collisions, traffic volume and time period may be given higher weights. The event feature vector and the event association graph serve as input to the LSTM model. Through training, the model learns the temporal and spatial association patterns between events. In the implementation of the attention mechanism, the attention weights are normalized using the Softmax function and then weighted summed with the event feature vector to obtain the attention-weighted event representation. For example, assuming the weight vector calculated by the Softmax function is [0.2, 0.3, 0.5] and the event feature vector is [10, 20, 30], the weighted event representation is [6, 9, 15]. These weights are learned through the backpropagation algorithm, enabling the model to adaptively focus on different event attributes. Using the trained LSTM prediction model, we predict events that are likely to occur within a certain timeframe, based on the current temporal and spatial scope. For example, if the current time is 8:00 AM, we predict the likelihood of a rear-end collision on a certain road section within the next hour. The predicted events are post-processed, using probabilistic statistical methods to calculate the probability of each predicted event occurring at different temporal and spatial locations.Suppose historical data shows that the frequency of rear-end collisions on road section X between 8:00 and 9:00 a.m. is 0.05. Using the Bayesian formula, combined with current traffic volume and weather conditions, the probability of a rear-end collision under these conditions is calculated to be 0.08. Use data visualization tools such as Tableau or PowerBI to intuitively present the prediction results and probability distribution. For example, a heat map can display the probability of accidents on different road sections at different time periods, with darker colors indicating higher probabilities. A time series graph can also be used to illustrate the trend of accidents on a particular road section. This visualization interface allows traffic management departments to intuitively analyze the spatiotemporal distribution and occurrence trends of future events and implement proactive traffic diversion measures, such as increasing police presence or adjusting traffic light hours, to reduce the impact of accidents on traffic flow and improve the safety and efficiency of the transportation system. The above steps not only effectively capture the temporal and spatial correlation patterns of events, but also enable prediction of event trends and real-time detection of abnormal events. This approach relies on the LSTM model's effective capture of spatiotemporal correlations. Combined with the attention mechanism and probabilistic statistical methods, it can deeply explore event correlation patterns, providing powerful support for traffic management and emergency response. Through the application of data visualization tools, complex data analysis results become intuitive and easy to understand, making it easier for relevant departments to make scientific and reasonable decisions and improve the overall operational efficiency of the transportation system.

[0023] Step S105: If an event that may occur in the future is predicted, other related events that may be triggered by the event are inferred based on the event spatiotemporal association network to form an event causal logic chain.

[0024] A1. Based on a pre-established event spatiotemporal association network, obtain a predicted target event that may occur in the future. A2. For the target event, search the event spatiotemporal association network for related events directly associated with the target event. A3. If a directly related related event exists, obtain the temporal and spatial attribute information of the related event. A4. Using a spatiotemporal association analysis algorithm, determine whether the related event is likely to be triggered by the target event based on the spatiotemporal attributes of the target and related events. B1. If the related event is likely to be triggered by the target event, add the related event to the causal chain list and use it as the new target event. Repeat steps A2-A3 to continue inferring the next level of related events that may be triggered. B2. If the related event is unlikely to be triggered by the target event, or if there are no related events directly associated with the current target event, determine whether the causal chain list is empty. C1. If the causal chain list is not empty, output the resulting event causal logic chain. C2. If the causal chain list is empty, determine that the initial target event will not trigger other related events. Through the above steps, we can automatically form an event causal logic chain based on the predicted possible future events, combined with the event spatiotemporal correlation network, and use the spatiotemporal correlation analysis algorithm to predict the related events and causal logic that may be triggered by the event.

[0025] For example, a pre-established event spatiotemporal association network is used to obtain predicted future target events. For example, in a highway intelligent transportation system, a rear-end collision is predicted to occur within the next hour on a certain highway section X. This is the target event. This prediction is based on historical data analysis and sequence prediction models, combined with factors such as current traffic volume and weather conditions. For the target event, the event spatiotemporal association network is searched for related events directly associated with the target event. The event spatiotemporal association network is a complex graph structure, with nodes representing events and edges representing the spatiotemporal associations between events. Suppose the association network reveals that rear-end collisions on section X often cause traffic congestion on the adjacent section Y. Then, the traffic congestion event on section Y is a related event directly associated with the target event. If a directly related related event exists, the temporal and spatial attribute information of the related event is obtained. For example, traffic congestion events on section Y typically occur within 10 to 20 minutes after a rear-end collision and are primarily concentrated at the entrance to section Y. Using a spatiotemporal correlation analysis algorithm, based on the spatiotemporal properties of the target event and associated events, it is determined whether the associated event is likely triggered by the target event. This algorithm can make this determination by calculating the spatiotemporal distance and correlation strength between events. For example, if a rear-end collision on road section X is closely followed by a traffic congestion event on road section Y in both time and space, and historical data shows frequent co-occurrence of the two events, it can be determined that the traffic congestion on road section Y is likely triggered by the rear-end collision on road section X. If the associated event is likely triggered by the target event, the associated event is added to the causal chain list and used as the new target event. The search and judgment process is repeated to infer the next level of associated events that may be triggered. For example, if the traffic congestion on road section Y is likely to cause slow traffic on the more distant road section Z, the slow traffic event on road section Z is also added to the causal chain list and further analyzed as a new target event. If the associated event is unlikely to be triggered by the target event, or if there are no associated events directly linked to the current target event, the causal chain list is determined to be empty. For example, if a rear-end collision on road section X does not trigger any other related events, or if no directly related events are found, the causal chain list is empty. If the causal chain list is not empty, the resulting causal logic chain of events is output. For example, the final causal logic chain output might be: "Rear-end collision on road section X → Traffic congestion on road section Y → Slow traffic on road section Z." This chain clearly demonstrates the causal relationship between events, helping traffic management departments to take proactive countermeasures, such as deploying additional traffic police and adjusting traffic lights, to reduce the impact of the accident on traffic flow. If the causal chain list is empty, it is determined that the initial target event will not trigger any other related events. For example, if the rear-end collision on road section X does not trigger any other related events, it indicates that the impact of the event is limited, and traffic management departments can take targeted measures without overreacting.Through the above steps, based on predicted future events and combined with the event spatiotemporal correlation network, a spatiotemporal correlation analysis algorithm can be used to automatically form a causal logic chain, predicting the related events and causal logic that may be triggered by an event. The implementation of this method not only improves the predictability and proactiveness of traffic management, but also effectively reduces the chain reaction of traffic accidents and improves the overall operational efficiency of the transportation system. For example, in practical applications, traffic management departments can use the system to predict an impending rear-end collision on a certain road section and quickly identify the chain reactions that this accident may trigger, such as congestion on adjacent roads and slow traffic on more distant roads. Based on this information, management departments can deploy police forces in advance, issue traffic warnings, and redirect vehicles, effectively alleviating traffic pressure and reducing the incidence of accidents.

[0026] Step S106: Based on the event causal logic chain, a reinforcement learning algorithm is used to evaluate the effectiveness of different response strategies and obtain the optimal event handling plan.

[0027] Obtain the event causal logic chain, obtain data related to the current event and historical handling strategy data from the historical event database; use relevant data and handling strategy data to build an event evolution prediction model through data mining and machine learning algorithms; predict the development trend and results of the event based on the event evolution prediction model; combine the rules and strategies in the event handling domain knowledge base to generate multiple event response strategies; for each response strategy, conduct multiple rounds of simulation in the event evolution prediction model to evaluate the effects of each strategy in different scenarios; compare the simulation evaluation results of each strategy, and select the strategy with the best comprehensive effect as the candidate event handling plan; conduct a rationality test on the candidate handling plan, and optimize and improve it based on expert experience and actual conditions to form a The final incident handling execution plan; break down the incident handling execution plan into specific operational steps, clarify the execution subject, time node and required resources of each step; continuously monitor the development of the incident and dynamically adjust the execution plan according to the actual situation; in the process of incident handling, continuously collect incident-related data, including incident development data, handling action data and effect feedback data; use the collected relevant data to optimize and improve the event evolution prediction model and strategy generation model; through continuous analysis of the event causal logic chain, dynamically identify the key factors affecting the direction of the event and predict new situations that may arise; for predicted new situations, formulate response plans and handling strategies in advance to minimize the negative impact of the incident and strive to guide the incident in a favorable direction.

[0028] For example, based on the event causal logic chain, data related to the current event and historical handling strategy data are retrieved from a historical event database. For example, if a highway intelligent transportation system predicts a possible rear-end collision on highway section X, the system will extract data from the historical event database on similar rear-end collisions, including the time, location, traffic volume, weather conditions, and the handling strategies at the time, such as traffic police dispatch and traffic control measures. Using this acquired data, an event evolution prediction model is constructed using data mining and machine learning algorithms. Data mining techniques can extract valuable information from massive amounts of historical data, while machine learning algorithms such as neural networks and decision trees are used to build models. For example, by analyzing historical rear-end collision data, traffic congestion patterns after the accident can be identified. These patterns can be used to train a model to predict the degree and scope of congestion that may be caused by future accidents. Based on the prediction results of the event evolution prediction model and combined with the rules and strategies in the event handling domain knowledge base, multiple possible event response strategies are generated. The knowledge base contains various traffic management rules and emergency response strategies, such as vehicle diversion and temporary speed limits. The model predicts that a rear-end collision on road section X may lead to congestion on road section Y. The system generates multiple response strategies, such as pre-installing warning signs at the entrance to road section Y or adjusting traffic light cycles. For each response strategy, multiple simulations are conducted within the event evolution prediction model to evaluate the effectiveness of each strategy under different scenarios. Simulations can simulate how traffic flow changes after implementing different strategies. For example, the degree to which traffic congestion is alleviated after installing a warning sign at the entrance to road section Y is simulated. Through multiple simulations, the effectiveness of different strategies is compared and evaluated under varying traffic volume and weather conditions. By comparing the simulation evaluation results of each strategy, the strategy with the best overall effect is selected as a candidate for the incident response. For example, simulations show that installing a warning sign combined with temporary speed limits is the most effective way to alleviate congestion on road section Y, and this strategy is selected as a candidate. The candidate solutions are then tested for rationality and optimized based on expert experience and actual conditions to form a final incident response plan. Experts may adjust the plan based on actual road conditions and experience, such as adding on-site traffic police guidance, to ensure the plan's feasibility and effectiveness. The incident handling execution plan is broken down into specific operational steps, clarifying the execution entities, time nodes, and required resources for each step. For example, the plan can be broken down into: dispatching traffic police to the scene within 10 minutes, setting up warning signs within 15 minutes, adjusting the traffic light cycle within 20 minutes, etc., with each step clearly defining the responsible person, required equipment, and time nodes. During the incident handling process, the development of the incident is continuously monitored, and the implementation plan is dynamically adjusted according to the actual situation. For example, through the real-time traffic monitoring system, if it is found that the traffic congestion is more serious than expected, the police force will be immediately increased, the diversion range will be expanded, and the effectiveness of the plan will be ensured. Throughout the incident handling process, relevant data on the incident is continuously collected, including incident development data, handling action data, and effect feedback data.This data is used to optimize and improve event evolution prediction models and strategy generation models. For example, if the collected data indicates that temporary speed limits are effective, the system will increase the weight of this strategy in the model to improve the accuracy of future predictions. By continuously analyzing the causal chain of events, key factors influencing the course of events are dynamically identified, and new scenarios can be predicted. For example, if analysis reveals that rear-end collisions on road section X are often accompanied by inclement weather, the system will predict new traffic risks that may arise under similar weather conditions in the future, such as chain reactions caused by slippery roads. Based on these predicted scenarios, response plans and handling strategies are developed in advance to minimize the negative impact of the incident and strive to steer the event in a favorable direction. For example, for rear-end collisions in inclement weather, a pre-emptive contingency plan is developed, including increased road patrols and the preparation of emergency supplies, to ensure a rapid response in the event of an incident and minimize its impact on the traffic system. Through these steps, the system not only predicts the development trend of an event but also generates effective response strategies, which are continuously optimized during actual handling, enhancing the intelligence and effectiveness of event handling. The implementation of this method not only improves the foresight and initiative of traffic management, but also effectively reduces the chain reaction of traffic accidents and improves the overall operational efficiency of the transportation system.

[0029] Step S107: During the occurrence and evolution of an event, the latest vehicle trajectory data and road condition monitoring data are continuously acquired, and the event spatiotemporal correlation network and prediction model are dynamically updated to achieve real-time and continuous event correlation analysis and prediction.

[0030] Obtain vehicle GPS trajectory data and road traffic flow monitoring data in the area where the incident occurred; construct an event spatiotemporal association network based on vehicle GPS trajectory data and road traffic flow monitoring data. The nodes in the event spatiotemporal association network represent vehicles or road sections. The attributes of vehicle nodes include vehicle ID, timestamp, latitude and longitude coordinates. The attributes of road section nodes include road section ID, road grade, and traffic flow. Edges represent vehicle trajectories or road condition associations. The weight of vehicle trajectory edges is the time interval between trajectory points, and the weight of road condition association edges is the traffic flow difference between road sections. By statistically analyzing the node attributes and edge weights in the event spatiotemporal association network, the association scope and impact degree of the current event are determined. The spatial association scope of the event is calculated based on the spatiotemporal distribution of vehicle nodes, and the impact degree of the event is calculated based on the traffic flow changes of road section nodes. The graph attention network model is used to model the event spatiotemporal association network. By aggregating node features and updating edge weights, Learn the embedding representation of nodes and predict the event association strength and propagation trend. The node embedding representation is obtained by aggregating the features of neighboring nodes through the attention mechanism, and the edge weight is updated by calculating the similarity of the node embedding representation; if the event association range predicted by the graph attention network model exceeds the preset association range threshold or the impact degree exceeds the preset impact degree threshold, the early warning mechanism is triggered, and event warning information is sent to the traffic management department, and corresponding traffic diversion measures are recommended; based on the latest acquired vehicle GPS trajectory data and road traffic flow monitoring data, the node attributes and edge weights of the event spatiotemporal association network are dynamically updated. For newly added vehicle trajectory data, the corresponding vehicle nodes and trajectory edges are added to the event spatiotemporal association network. For new traffic flow monitoring data, the traffic flow attributes and association edge weights of the corresponding road section nodes are updated; through continuous data acquisition, network update and model prediction, real-time and continuity of event association analysis and prediction are achieved.

[0031] For example, in a highway intelligent transportation system, to understand the occurrence and evolution of a specific event, it is first necessary to obtain vehicle GPS trajectory data and road traffic flow monitoring data within the event area. For example, suppose a rear-end collision occurs on Section A of a certain city highway. The system immediately extracts relevant data from the vehicle's GPS device and road monitoring equipment. Vehicle GPS trajectory data may include information such as each vehicle's ID, timestamp, and longitude and latitude coordinates, while road traffic flow monitoring data includes information such as the ID of each road section, road grade, and real-time traffic flow. Based on this data, a spatiotemporal correlation network of events is constructed. In this network, nodes are divided into two categories: vehicle nodes and road section nodes. Vehicle node attributes include vehicle ID, timestamp, and longitude and latitude coordinates. For example, the longitude and latitude of a node with vehicle ID "V123" at the timestamp "2023-10-01 14:30:00" are "(34.0522, -118.2437)". Attributes of a road segment node include segment ID, road grade, and traffic flow. For example, a node with segment ID "R456" has a road grade of "First-Class Highway" and a current traffic flow of "3,000 vehicles / hour." Edges represent vehicle trajectories or road condition associations. The weight of a vehicle trajectory edge is the time interval between trajectory points. For example, the edge weight between the trajectory points of vehicle V123 at 14:30 and 14:35 is "5 minutes." The weight of a road condition-related edge is the difference in traffic flow between segments. For example, the difference in traffic flow between segments R456 and R789 is "500 vehicles / hour." By statistically analyzing the node attributes and edge weights in the event spatiotemporal association network, the scope and impact of the current event can be determined. Specifically, the spatial scope of an event is calculated based on the spatiotemporal distribution of vehicle nodes. For example, within 30 minutes after an accident, affected vehicles were primarily concentrated on segment A and its upstream and downstream segments B and C. The impact of an event is calculated based on changes in traffic flow at road segment nodes. For example, a decrease in traffic flow on road segment A from a normal rate of 4,000 vehicles / hour to 1,500 vehicles / hour indicates a significant impact. The Graph Attention Network (GAT) model is used to model the spatiotemporal association network of events. The GAT model learns node embeddings by aggregating node features and updating edge weights, and predicts the strength of event associations and propagation trends. Node embeddings are derived by aggregating features of neighboring nodes using an attention mechanism. For example, the embedding for vehicle V123 incorporates not only its own attributes but also the features of its neighboring vehicles V124 and V125. Edge weights are updated based on the similarity between node embeddings. For example, the edge weights for the trajectories of vehicles V123 and V124 are adjusted based on the similarity between their embeddings. If the GAT model predicts an event's association range exceeds 30% of the urban road network or its impact exceeds 50% of the average traffic flow, an early warning mechanism is triggered.For example, if the model predicts that the accident has spread to 35% of the city's main road network and traffic flow has dropped to 60% of the average level, the system immediately sends an event warning to the traffic management department and recommends appropriate traffic diversion measures, such as "setting up a diversion point upstream of road section A to redirect vehicles to route via road section D." The node attributes and edge weights of the event spatiotemporal correlation network are dynamically updated based on the latest vehicle GPS trajectory data and road traffic flow monitoring data. For example, if newly acquired vehicle trajectory data shows that vehicle V126 entered the affected area at 2:40 PM, the system adds the corresponding vehicle node and trajectory edge to the network. If new traffic flow monitoring data shows that traffic flow on road section B has increased from 2,500 vehicles / hour to 3,500 vehicles / hour, the system updates the traffic flow attributes and associated edge weights of the corresponding road section nodes. Through continuous data acquisition, network updates, and model predictions, real-time and continuous event correlation analysis and prediction are achieved. For example, the system updates data every five minutes, recalculates the attributes of network nodes and edges, and uses the GAT model to generate new predictions. This ensures that traffic management departments can understand the development of events in real time and adjust traffic diversion measures promptly. This approach not only improves the predictability and proactiveness of traffic management, but also effectively reduces the chain reaction of traffic accidents and improves the overall operational efficiency of the transportation system.

[0032] The above specific implementation methods further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above are only specific implementation methods of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for intelligent correlation analysis and prediction of highway events, characterized in that: include: Obtain vehicle trajectory data and road condition monitoring data, cluster them using an adaptive spatiotemporal clustering algorithm to obtain the spatiotemporal distribution characteristics of events; According to the spatiotemporal distribution characteristics of events, an event spatiotemporal association network is constructed, where nodes represent events and edges represent the associations between events. The graph neural network model is used to train the event spatiotemporal correlation network to obtain the correlation evolution law of events in time and space; Based on the spatiotemporal distribution characteristics and associated evolution laws of events, a sequence prediction model using an attention mechanism is constructed to predict possible events and their spatiotemporal distribution probabilities in the future. If a possible event in the future is predicted, then based on the event spatiotemporal correlation network, other related events that may be triggered by the event are inferred to form a causal logic chain of events; Based on the causal logic chain of events, reinforcement learning algorithms are used to evaluate the effectiveness of different response strategies and obtain event handling plans; During the occurrence and evolution of events, vehicle trajectory data and road condition monitoring data are continuously acquired, and the event spatiotemporal correlation network and prediction model are dynamically updated; The process of constructing a spatiotemporal association network of events based on the spatiotemporal distribution characteristics of the events, wherein nodes represent events and edges represent associations between events, further includes: Obtaining time attribute data and spatial attribute data of multiple events; For each event, extract the temporal and spatial features of the event; Combine the temporal and spatial features of each event to construct the spatiotemporal feature vector of the event; Calculate the distances between the spatiotemporal feature vectors of multiple events and construct the spatiotemporal correlation matrix of events; If the distance value of the matrix element in the event spatiotemporal association matrix is ​​less than or equal to the preset time distance threshold and less than or equal to the preset spatial distance threshold, then the distance value of the matrix element in the event spatiotemporal association matrix is ​​set to 1, otherwise it is set to 0, and the event spatiotemporal association adjacency matrix is ​​obtained; The connected components of the event spatiotemporal adjacency matrix are calculated to obtain a set of several events. A set of events represents an event community. Calculate the average spatiotemporal feature vector of the event community based on the spatiotemporal feature vectors of all events in the event community, and construct a time series of spatiotemporal feature vectors of the event community; Perform time series analysis on the time series of the spatiotemporal feature vectors of the event community to obtain the evolution trend of the event community; Based on the evolution trend of the event community, the predicted value at the current moment is calculated. If the absolute value of the difference between the actual value of the event community's spatiotemporal feature vector time series at the current moment and the predicted value at the current moment is greater than the preset threshold, the event community is determined to be an abnormal event community; Obtaining attribute data of events contained in the abnormal event community, the attribute data of events including event type and event level; Determine the geographical distribution of abnormal event communities based on the spatial characteristics of events in the abnormal event community; The impact of abnormal events is determined based on the attribute data of the event and the weight value of the event community; abnormal event warning information is generated based on the geographical distribution range and impact of the abnormal events.

2. The method according to claim 1, wherein Obtain vehicle trajectory data and road condition monitoring data, and use an adaptive spatiotemporal clustering algorithm to cluster the vehicle trajectory data and road condition monitoring data to obtain the spatiotemporal distribution characteristics of the event, including: Preprocess vehicle trajectory data and road condition monitoring data to remove noise and outliers; Construct feature vectors, cluster them using the DBSCAN algorithm, and dynamically adjust parameters; Post-process the clustering results, remove noise clusters, and merge clusters; According to the density and distance characteristics of clusters, the distribution characteristics of events in time and space are obtained; Use time series analysis methods to analyze the temporal patterns of events, use spatial interpolation methods to analyze the spatial distribution patterns of events, and predict the time and place of event occurrence; Combine the spatiotemporal distribution characteristics of events with map information to generate event distribution heat maps; Based on event prediction results and event distribution heat maps, traffic control measures are taken to improve road traffic efficiency.

3. The method according to claim 1, wherein The graph neural network model is used to train the event spatiotemporal correlation network to obtain the correlation evolution rules of events in time and space, including: A graph convolutional neural network is used to learn the representation of event spatiotemporal association networks and obtain low-dimensional vector representations of event nodes. Using the timestamp information of event nodes, we learn the hidden state representation of events at different time steps and explore the evolution of events in the time dimension. By using the geographic location information of event nodes and performing spatial graph convolution operations, we can learn the spatial association patterns of events and obtain the association relationships between events in different geographic locations. Based on the learned event node representation, cosine similarity is used to calculate the similarity between events and discover highly correlated event sets; Perform cluster analysis on highly correlated event sets to obtain different types of event correlation patterns; Apply event correlation model to event prediction. For current events, based on their spatiotemporal correlation, find historical events that are highly correlated with them in the event spatiotemporal correlation network. Then, use the subsequent evolution of the correlated events to predict the future development trend of the current event. Apply event correlation patterns to anomaly detection. By comparing the similarity between the current event and the known correlation pattern, abnormal events that do not conform to the known pattern can be identified, thus achieving real-time detection and early warning of abnormal events.

4. The method according to claim 1, wherein Based on the spatiotemporal distribution characteristics and correlation patterns of events, a sequence prediction model using an attention mechanism is constructed to predict possible events and their spatiotemporal distribution probabilities in the future, including: Extract the time, space and type attributes of the event to obtain the event feature vector; Clustering algorithm is used to cluster event feature vectors; Based on the clustering results, an event association graph is constructed; Use association rule mining algorithms to discover frequent association patterns and causal relationships between events; We choose LSTM as the basic sequence prediction model and introduce an attention mechanism to dynamically adjust the weights of different event attributes. The event feature vector and event association graph are used as inputs to the LSTM model to train the prediction model. In the implementation of the attention mechanism, the Softmax function is used to normalize the attention weights and perform weighted summation with the event feature vector to obtain the attention-weighted event representation; The attention weights are learned through the back-propagation algorithm, enabling the model to adaptively focus on different event attributes; Use the trained LSTM prediction model to predict events that will occur in the future based on the current time and space scope; The predicted events are post-processed, the frequency of events under different conditions is statistically analyzed based on historical data, and the probability of events under given conditions is calculated using the Bayesian formula.

5. The method according to claim 1, wherein If a possible future event is predicted, then based on the event spatiotemporal association network, other related events that may be triggered by the event are inferred to form a causal logic chain of events, including: A1. Obtain predicted target events that may occur in the future based on the pre-established event spatiotemporal association network A2. Search for related events directly related to the target event in the event spatiotemporal association network; A3. Obtaining the time and space attribute information of the associated event; A4. Using a spatiotemporal correlation analysis algorithm, based on the spatiotemporal attributes of the target event and the associated events, determine whether the associated event is likely to be caused by the target event; B1. If the associated event may be caused by the target event, add the associated event to the causal chain list and use it as the new target event. Repeat steps A2-A3 to continue reasoning about the next level of associated events that may be caused. B2. If the associated event cannot be triggered by the target event, or there is no associated event directly associated with the current target event, determine whether the causal chain list is empty; C1. If the causal chain list is not empty, then output the formed event causal logic chain; C2. If the causal chain list is empty, it is determined that the initial target event will not trigger other related events.

6. The method according to claim 1, wherein Based on the causal logic chain of the event, a reinforcement learning algorithm is used to evaluate the effectiveness of different response strategies and obtain an event handling plan, including: Obtain the event causal logic chain and obtain data related to the current event and historical handling strategy data from the historical event database; Using relevant data and disposal strategy data, we build an event evolution prediction model through data mining and machine learning algorithms; Predict the development trend and outcome of events based on the event evolution prediction model; Combine the rules and strategies in the incident handling domain knowledge base to generate multiple incident response strategies; For each response strategy, multiple rounds of simulations are conducted in the event evolution prediction model to evaluate the effectiveness of each strategy under different scenarios; Compare the simulation evaluation results of each strategy and select the strategy with the best overall effect as the candidate event handling plan; Conduct rationality checks on candidate disposal plans, optimize and improve them based on expert experience and actual conditions, and form the final incident disposal execution plan; Continuously monitor the development of the incident and dynamically adjust the implementation plan based on actual conditions; During the incident handling process, continuously collect incident-related data, including incident development data, handling action data, and effect feedback data; Utilize the collected relevant data to optimize and improve event evolution prediction models and strategy generation models; Through continuous analysis of the causal logic chain of events, we can dynamically identify key factors that influence the direction of events and predict new situations that may arise. In response to predicted new situations, response plans and disposal strategies should be formulated in advance to minimize the negative impact of the incident and strive to guide the incident in a favorable direction.

7. The method according to claim 1, wherein During the occurrence and evolution of events, vehicle trajectory data and road condition monitoring data are continuously acquired, and the event spatiotemporal correlation network and prediction model are dynamically updated, including: Construct a spatiotemporal event correlation network based on vehicle trajectory data and road traffic flow monitoring data; By statistically analyzing the node attributes and edge weights in the event spatiotemporal correlation network, the spatial correlation scope and impact of the current event can be determined; The graph attention network model is used to model the spatiotemporal correlation network of events and predict the event correlation strength and propagation trend; Based on the acquired vehicle trajectory data and road traffic flow monitoring data, the node attributes and edge weights of the event spatiotemporal association network are dynamically updated.

Images