Intelligent monitoring-based urban traffic linkage early warning method and system

Abstract

The present application belongs to the field of urban traffic early warning technology, and provides a city traffic linkage early warning method and system based on intelligent monitoring, comprising: acquiring multi-source traffic monitoring data, performing multi-modal fusion processing on the multi-source traffic monitoring data to form traffic state data sequences of a target road network; constructing a traffic change curve based on the traffic state data sequences, separating the change characteristics of the traffic change curve, classifying the change characteristics into rapid change components and trend change components, and quantitatively processing the separated change characteristics. Through principal component analysis and wavelet transform, the components representing rapid disturbance and the components representing slow change accumulation are separated from the mixed traffic data, and differentiated clustering models and data-driven triggering rules are constructed for the two essentially different modes, so that sudden accidents and slow-moving congestion can be distinguished from the source of time evolution, and the problems of instantaneous noise false alarm and trend event missed alarm are optimized.

Description

Technical Field

[0001] This invention belongs to the field of urban traffic early warning technology, specifically to a method and system for coordinated urban traffic early warning based on intelligent monitoring. Background Technology

[0002] With the continuous expansion of urban road traffic and the increasing number of vehicles, urban traffic operation is characterized by frequent fluctuations in traffic flow and a variety of event types. In order to ensure road traffic efficiency and improve the response capabilities of traffic management departments, intelligent traffic monitoring systems based on multi-source information such as video monitoring, radar detection, sensing terminals, and floating car data are gradually becoming the core infrastructure of traffic management.

[0003] Existing traffic linkage early warning technologies typically rely on single monitoring data or judgment mechanisms based on simple threshold rules, which have limited ability to describe changes in traffic conditions. In complex traffic environments, changes in traffic conditions may present sudden jumps or slowly accumulate to form congestion or delays, making it difficult to distinguish the evolutionary characteristics of different events.

[0004] Due to the lack of effective differentiation between rapid-burst traffic events and non-rapid-burst trend events, misjudgments and omissions often occur in practical applications. On the one hand, instantaneous noise may trigger false warnings, causing the traffic control system to over-respond and affecting the overall operational efficiency. On the other hand, slowly evolving trend congestion is difficult to identify in a timely manner, causing traffic management departments to miss the best intervention opportunity. In addition, the event triggering logic lacks specificity, resulting in unstable triggering of warning instructions and insufficient matching between linkage strategies and actual road conditions, which affects the overall effectiveness of urban traffic control.

[0005] To this end, the present invention provides a method and system for coordinated early warning of urban traffic based on intelligent monitoring. Summary of the Invention

[0006] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0007] The technical solution adopted by this invention to solve its technical problem is: a city traffic linkage early warning method based on intelligent monitoring, comprising:

[0008] Acquire multi-source traffic monitoring data, perform multi-modal fusion processing on the multi-source traffic monitoring data, and form a traffic status data sequence for the target road network;

[0009] Traffic change curves are constructed based on traffic state data sequences. The change features of the traffic change curves are separated and classified into rapid change components and trend change components. The separated change features are then quantified.

[0010] By combining the statistical analysis results of historical traffic status data with the change characteristics, feature clustering and event triggering rule extraction are performed on the rapidly changing components and the trend changing components to determine the traffic event triggering type.

[0011] Traffic incident triggering types include rapid-onset incidents and non-rapid-onset incidents;

[0012] Based on the determined traffic event trigger type, a traffic linkage early warning instruction is generated, and an early warning or linkage dispatch strategy is sent to the traffic control system.

[0013] A smart monitoring-based urban traffic linkage early warning system, which includes:

[0014] Traffic data acquisition module: Acquires multi-source traffic monitoring data, performs multi-modal fusion processing on the multi-source traffic monitoring data, and forms a traffic status data sequence of the target road network;

[0015] Traffic change analysis module: Constructs traffic change curves based on traffic state data sequences, separates the change features of the traffic change curves, classifies the change features into rapid change components and trend change components, and quantifies the separated change features;

[0016] Traffic incident determination module: By combining the statistical analysis results of historical traffic status data with the characteristics of change, feature clustering and event triggering rule extraction are performed on the rapidly changing components and the trend changing components respectively to determine the traffic incident triggering type;

[0017] Traffic incident triggering types include rapid-onset incidents and non-rapid-onset incidents;

[0018] Linked early warning module: Based on the determined traffic event trigger type, it generates traffic linked early warning instructions and issues early warning or linked dispatch strategies to the traffic control system.

[0019] The beneficial effects of this invention are as follows:

[0020] This invention quantifies the characteristics of trend-changing components and rapidly changing components, thereby accurately distinguishing between fast and slow events. This provides precise and reliable input for subsequent steps of targeted rule extraction and event type determination, thus optimizing the problems of false alarms due to instantaneous noise and missed alarms due to slowly changing events.

[0021] This invention automatically discovers and summarizes the characteristic patterns of different types of traffic events by performing cluster analysis on historical event samples, transforming implicit and scattered empirical data into explicit and structured event pattern clusters, thereby realizing the mechanization and standardization of traffic management.

[0022] Based on the statistical characteristics of each cluster, this invention automatically generates corresponding triggering rules, which are dynamically determined by the distribution of historical data and can be adjusted by the sensitivity coefficient (k value). This makes the warning rules adaptive and targeted, and can match the unique evolution patterns of different road segments, different time periods, and different types of events, thus fundamentally optimizing the false alarms and false alarms caused by the fixed threshold method.

[0023] In real-time operation, this invention precisely matches the features extracted in real time with all pre-generated rules, thereby not only determining whether an event has occurred, but also identifying the specific type of the event, providing a key decision-making basis for subsequent differentiated responses.

[0024] This invention uses principal component analysis and wavelet transform to separate components representing rapid disturbances and components representing slow-moving accumulations from mixed traffic data. For these two fundamentally different patterns, differentiated clustering models and data-driven triggering rules are constructed respectively, thereby distinguishing between sudden accidents and slow-moving congestion from the source of temporal evolution and optimizing the problems of false alarms of instantaneous noise and missed reports of trend events. Attached Figure Description

[0025] The invention will now be further described with reference to the accompanying drawings.

[0026] Figure 1 This is a flowchart illustrating the steps of the urban traffic linkage early warning method based on intelligent monitoring according to the present invention.

[0027] Figure 2 This is a module architecture diagram of the urban traffic linkage early warning system based on intelligent monitoring, which is based on the present invention. Detailed Implementation

[0028] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0029] Example 1

[0030] Please see Figure 1 As shown in the embodiments of the present invention, the urban traffic linkage early warning method and system based on intelligent monitoring addresses the problems of inaccurate identification of sudden events, false alarms caused by instantaneous noise, and missed alarms of slowly accumulating events in urban traffic linkage early warning. It constructs traffic change curves and separates rapidly changing components from trend-changing components. Combined with historical traffic state data, it performs feature clustering and event triggering rule extraction, and distinguishes between rapid-burst and non-rapid-burst traffic events. This enables the early warning system to match the evolutionary characteristics of different events, reducing false alarm and missed alarm rates, and improving the timeliness, stability, and accuracy of linkage scheduling. The method includes the following steps:

[0031] Step S10: Acquire multi-source traffic monitoring data, perform multi-modal fusion processing on the multi-source traffic monitoring data, and form a traffic status data sequence of the target road network;

[0032] Specifically, in step S10, multi-source traffic monitoring data is collected through various traffic monitoring devices deployed in the target road network, including but not limited to:

[0033] Road traffic images or video streams captured by video cameras;

[0034] Traffic flow data collected by geomagnetic sensors;

[0035] Vehicle speed data collected by microwave radar;

[0036] Weather data collected by meteorological sensors (such as visibility, precipitation, temperature, etc.);

[0037] The collected multi-source traffic monitoring data is preprocessed, including timestamp alignment, handling of outliers and missing values, and data normalization.

[0038] Feature parameters for characterizing traffic conditions are extracted from the preprocessed multi-source traffic monitoring data, including: traffic flow, average vehicle speed, vehicle location distribution, vehicle dwell time, road occupancy, and weather visibility.

[0039] Specifically, traffic flow is the number of vehicles passing through the detection section per unit time.

[0040] Average vehicle speed is the average speed at which a vehicle travels through the detection area.

[0041] Vehicle location distribution is obtained by acquiring the spatial distribution of vehicles in the road network based on image recognition or radar point cloud data;

[0042] The duration of vehicle stagnation is the duration during which the vehicle speed is below a threshold (e.g., 5 km / h).

[0043] Road occupancy rate is the ratio of the road area occupied by vehicles within the detection area;

[0044] Weather visibility is the visibility index in meteorological data;

[0045] The extracted feature parameters are concatenated into a unified multidimensional feature vector according to time segments, with each dimension corresponding to a feature parameter. The high-dimensional features are then reduced in dimensionality by an autoencoder to remove redundant information.

[0046] The fused multidimensional feature vectors are sampled at fixed time intervals (e.g., 1 minute) and arranged in chronological order to form a traffic state data sequence for the target road network.

[0047] Step S20: Construct a traffic change curve based on the traffic state data sequence, separate the change features of the traffic change curve, and classify the change features into rapid change components and trend change components;

[0048] Specifically, in step S20, principal component analysis is used to extract characteristic parameters reflecting traffic conditions from the traffic state data sequence, including: traffic flow, average vehicle speed, and road occupancy.

[0049] The score sequence of the first principal component is selected as the comprehensive traffic state index. The first principal component is the direction with the largest contribution rate of the original data variance, and its linear combination coefficient is determined by the covariance structure of the data itself.

[0050] With time as the horizontal axis and the comprehensive traffic state index as the vertical axis, the traffic change curve of the target road network is plotted.

[0051] This allows the constructed curves to better reflect the main driving forces of traffic state evolution, providing a foundation for subsequent feature separation;

[0052] Signal decomposition is performed on the traffic change curve;

[0053] As a preferred approach in this embodiment, wavelet transform is used for analysis;

[0054] By selecting wavelet basis functions (such as Daubechies wavelets) and the number of decomposition levels, a discrete wavelet transform is performed on the traffic change curve to obtain a set of approximate coefficients and detail coefficient sequences.

[0055] It should be noted that the decomposition layer can be determined based on the periodic characteristics of traffic state changes and the sampling frequency, for example, to ensure that the periodic coverage of the lowest frequency component covers the typical congestion formation duration (e.g., more than 30 minutes).

[0056] The trend change components are reconstructed from the deepest approximation coefficients, which filter out short-term fluctuations and retain the long-term slow change trend of traffic conditions, corresponding to the evolutionary basis of non-rapid outbreak events caused by the accumulation of traffic demand, large-scale events, and the continuous impact of severe weather.

[0057] The rapidly changing components are obtained by reconstructing and superimposing the detailed coefficients of each layer, which fully reflects the rapid fluctuations of traffic conditions on the time scale of minutes or even seconds, corresponding to the impact signals of rapidly breaking events such as traffic accidents, vehicle malfunctions, and sudden failures of traffic lights.

[0058] Calculate the first derivative of the trend change component to quantify the rate of change, and extract the cumulative change and the duration of the trend within the sliding window;

[0059] Local extreme points are detected for rapidly changing components, and the abrupt change amplitude and rate of change between adjacent extreme points are calculated;

[0060] Count the frequency of sudden changes exceeding a preset noise threshold per unit time.

[0061] For example, suppose a 3-kilometer-long target road network of a city ring road is equipped with monitoring equipment such as video, geomagnetic and microwave radar. The system collects data every minute. After processing in step S10, a standardized traffic status data sequence of 2 consecutive hours (120 time points) is obtained, which includes three key characteristic parameters: traffic flow Q(t) (vehicles / minute), average vehicle speed V(t) (km / h), and road occupancy O(t) (percentage).

[0062] Using historical data from the first 30 minutes (t=1 to 30) as the training set, we calculate the covariance matrix and perform principal component analysis (PCA). We assume the linear combination of the first principal component PC1 is as follows:

[0063] ;

[0064] The coefficients are objectively determined by the data covariance structure. PC1's variance contribution rate is 85%, indicating that it can explain most of the traffic condition changes;

[0065] This linear combination was applied to the standardized data at all 120 time points to obtain the comprehensive traffic state index sequence.

[0066] Plot the traffic change curve with time as the horizontal axis. It begins to show a slow downward trend at the 40th minute and shows a sharp negative peak at the 65th minute.

[0067] Select Daubechies 4 wavelets, decompose the number of layers J=5, and perform discrete wavelet transform on the traffic change curve;

[0068] The trend change component was obtained by extracting the approximate coefficients of the 5th layer and reconstructing it. It started to decline slowly from the 40th minute and gradually rebounded at the 90th minute. The whole process was smooth and without drastic fluctuations.

[0069] The detailed coefficients from layers 1 to 3 were reconstructed and superimposed to obtain the rapidly changing components. The components remained relatively stable throughout the time period, but a significant and brief (lasting about 3 minutes) negative spike appeared at the 65th minute.

[0070] For the trend change component: the first derivative was calculated and it was found that its value was consistently negative in the 40-70 minute interval, with an average rate of change of -0.15.

[0071] Using a 60-minute sliding window, the cumulative change from the 40th minute to the 100th minute is calculated to be -8.2.

[0072] The trend lasted for approximately 50 minutes (from the 40th minute to the 90th minute).

[0073] For rapidly changing components: A significant local minimum was detected at the 65th minute. The abrupt change magnitude between this point and the previous minimum (the 63rd minute) was calculated to be -3.5; the rate of change was calculated to be -1.75.

[0074] Within a 5-minute time window (minutes 63-68), the frequency of mutations exceeding the noise threshold (e.g., 0.5 units) was 2 times.

[0075] Step S20 quantifies the characteristics of trend-changing components and rapidly changing components to accurately distinguish between "fast" and "slow" events, providing precise and reliable input for targeted rule extraction and event type determination in subsequent step S30, thereby optimizing the problems of false alarms due to instantaneous noise and missed alarms due to slowly changing events.

[0076] Step S30: By combining the statistical analysis results of historical traffic status data with the change characteristics, perform feature clustering and event triggering rule extraction on the rapidly changing components and trend changing components respectively to determine the traffic event triggering type;

[0077] Traffic incident triggering types include rapid-onset incidents and non-rapid-onset incidents;

[0078] Specifically, in step S30, a historical event sample feature library is constructed based on historical traffic state data;

[0079] The historical traffic status data (including the time periods and type labels of various traffic events confirmed by manual or system processes) are processed through the aforementioned steps S10 and S20 to generate a corresponding historical feature vector for each historical event sample.

[0080] Specifically, for each historical event period, the trend change component features (including change rate, cumulative change amount, and trend duration) are extracted to construct a historical trend feature vector, and the rapid change component features (including mutation magnitude, change rate, and mutation frequency exceeding the noise threshold) are extracted to construct a historical rapid feature vector.

[0081] The trend feature vectors and fast feature vectors of all historical events are collected separately to form a historical trend event sample set and a historical fast event sample set, which serve as a historical event sample feature library.

[0082] Feature clustering is performed on the historical trend event sample set and the historical fast event sample set respectively. The process is as follows:

[0083] For historical trend event sample sets, density-based clustering algorithms (such as the DBSCAN algorithm) are used for cluster analysis.

[0084] The historical trend event sample set is used as the input data for cluster analysis. Each sample is a historical trend feature vector, representing the evolutionary pattern characteristics of a historical non-rapid outbreak event (such as various slow traffic congestion).

[0085] The preset neighborhood radius and the minimum number of samples required to form the core point are determined as follows:

[0086] The neighborhood radius is calculated by analyzing the pairwise Euclidean distances between all sample points in the historical trend event sample set, and plotting the sorted K-distance map (usually the K value is the initial trial value with the minimum number of samples). The distance value corresponding to the inflection point (the obvious turning point where the curve changes from steep to gentle) in the map is selected as the reference value of the neighborhood radius, which can also help determine the density distribution characteristics of the samples in the dataset.

[0087] For example, when analyzing the feature vectors of historical slow-moving and congestion events (including rate of change, cumulative change, and duration of trend), after plotting the K-distance graph, a clear inflection point may appear at position 850 on the horizontal axis (assuming a total of 1000 samples). The corresponding K-distance value here is 1.5. Therefore, setting the neighborhood radius to 1.5 means that the algorithm will tend to classify sample points in the feature space that are within 1.5 units of each other as belonging to the same density region, i.e., the same congestion pattern cluster.

[0088] The minimum sample size takes into account the number of dimensions (i.e. the number of features, such as rate of change, cumulative change, and duration of trend) of historical trend event samples and the tolerance for noise points in the clustering results. Usually, the minimum sample size is set to 2 to 3 times the number of dimensions as an initial value, and can be fine-tuned according to the silhouette coefficient of the clustering results.

[0089] Traverse each sample point in the historical trend event sample set, draw a neighborhood in a high-dimensional space with the sample point as the center and the neighborhood radius as the radius, and count the number of sample points falling into this neighborhood (including the center point itself). If the number is greater than or equal to the set minimum number of samples, the sample point is marked as the core point.

[0090] Starting from a core point, all other sample points within the neighborhood radius (regardless of whether they are core points themselves) are considered to be density points directly reachable from the core point.

[0091] Starting from an unvisited core point, find all sample points that can be reached from that core point through a series of density reachability relationships (i.e., density-reachable sample points). These density-reachable sample points form a cluster.

[0092] During cluster expansion, newly incorporated core points will also incorporate points in their neighborhood, thus enabling the cluster to grow continuously.

[0093] Sample points that are not within the neighborhood radius of any core point and are not themselves core points will be marked as noise points or boundary points (belonging to a cluster but not themselves core points). After traversing all sample points, the set of all core points and their density-reachable samples will form the final cluster.

[0094] The historical trend event sample set is divided into several clusters, and each cluster corresponds to a specific type of non-rapid burst event;

[0095] Optional, non-rapid outbreak events include: weekday morning rush hour traffic congestion clusters heading into the city, persistent congestion clusters after large events, or traffic efficiency reduction clusters associated with rainy weather.

[0096] For example, assuming the historical trend event sample set contains feature vectors of 1000 historical slow traffic congestion events, it may form 5 main clusters through the DBSCAN clustering process described above;

[0097] Among them, the samples in cluster 1 generally showed "slow and negative rate of change, moderate cumulative change, and duration of more than 60 minutes", which can be summarized as "normal commuting congestion during evening rush hour" when combined with the event label.

[0098] The samples in cluster 2 showed "slightly negative rate of change, large cumulative change, and duration of more than 120 minutes", which may correspond to "congestion at highway entrance ramps during holidays".

[0099] The samples in cluster 3 exhibited "rapid negative rate of change, small cumulative change, and a duration of about 30-45 minutes," which may correspond to "sudden short-term congestion, such as the recovery period after a traffic accident is cleared."

[0100] The samples in cluster 4 exhibited "extremely slow and negative rate of change, extremely large cumulative change, and a duration of more than 180 minutes," which may correspond to "regional long-term congestion caused by the end of large-scale events (such as sports events or concerts)."

[0101] The samples in cluster 5 exhibited a "slowly decreasing rate of change with fluctuations, moderate cumulative change, and a duration of approximately 90 minutes," which may correspond to "traffic congestion caused by severe weather (such as rain or snow) leading to reduced traffic efficiency."

[0102] For historical fast event sample sets, hierarchical clustering algorithms (such as agglomerative hierarchical clustering) are used for cluster analysis;

[0103] The historical rapid event sample set is used as input data for hierarchical clustering analysis. Each sample is a historical rapid feature vector, representing the impact pattern characteristics of a historical rapid outbreak event (such as various traffic accidents, vehicle failures, etc.).

[0104] Calculate the distance between every two sample points (i.e., every two historical fast feature vectors) in the historical fast event sample set. Euclidean distance is usually used as the distance metric to obtain a distance matrix.

[0105] Initially, each sample point is treated as an independent cluster. In each step, the two closest clusters among all current clusters are found. The method for calculating the distance between clusters is to use the maximum distance between any two points in the two clusters.

[0106] Merge the two closest clusters into a new cluster, and update the distances between the merged new cluster and the remaining clusters.

[0107] Repeat the steps of finding the nearest cluster and merging clusters until all sample points are finally merged into one large cluster. This will generate a dendrogram. By analyzing the dendrogram, the final cluster division is determined by cutting at the distance threshold.

[0108] It should be noted that the distance threshold can be determined based on business knowledge (the desired number of clusters) or by evaluating the clustering profile coefficients corresponding to different cutting schemes. The closer the profile coefficient is to 1, the denser the samples within the cluster and the better the separation between clusters.

[0109] The clustering process ultimately divides the historical fast event sample set into several clusters, each cluster corresponding to a specific type of fast-burst event;

[0110] Optional, the types of rapid outbreak events can be: vehicle rear-end collision clusters, vehicle breakdown clusters, or pedestrian / non-motorized vehicle intrusion clusters;

[0111] For example, suppose the historical rapid event sample set contains feature vectors of 500 historical sudden events. Through the hierarchical clustering process described above, it may form 4 main clusters.

[0112] Among them, the samples in cluster A generally exhibited "extremely large mutation amplitude, extremely fast change rate, and a mutation frequency of 1 time within a very short time window", which, combined with the event label, can be summarized as "serious vehicle rear-end collision accident".

[0113] The samples in cluster B exhibited "moderate mutation magnitude, moderate rate of change, and mutations may occur 2-3 times," which may correspond to "vehicle breakdown or cargo scattering."

[0114] The samples in cluster C exhibited "small mutation amplitude but rapid change rate and possibly slightly higher mutation frequency", which may correspond to "pedestrians or non-motorized vehicles suddenly entering the motor vehicle lane";

[0115] The samples in cluster D exhibit "large mutation amplitude and rapid change rate, but their spatial location characteristics are significantly different from other clusters," which may correspond to "sudden malfunction of traffic lights."

[0116] It should be noted that the reason for using different clustering methods (DBSCAN and hierarchical clustering) for the historical trend event sample set and the historical rapid event sample set is that it is a targeted selection based on the data distribution characteristics and clustering objectives of the two types of events. Historical trend events usually have a large sample size, continuous evolution patterns, and are prone to forming clusters with uneven density in the feature space. DBSCAN can adaptively discover clusters of arbitrary shapes and effectively distinguish noise, making it suitable for mining typical patterns of such events. On the other hand, historical rapid event samples are relatively sparse and have significant feature differences. Hierarchical clustering can clearly show the phylogenetic relationship of event types through dendrograms, making it easier to determine reasonable cluster divisions based on domain knowledge or statistical indicators, and achieve fine-grained classification. The combination of the two methods matches the density distribution characteristics of slowly changing events and the phylogenetic structure characteristics of sudden events, respectively.

[0117] After clustering the historical trend event sample set and the historical rapid event sample set separately, event triggering rules are extracted for each event category based on the clustering results. The process is as follows:

[0118] For each non-rapidly breaking event cluster (such as clusters 1 to 5 above) obtained by clustering historical trend event samples, the statistical distribution of all historical trend feature vectors within the cluster is analyzed. For each feature (such as rate of change, cumulative change, and trend duration), the mean and standard deviation within the cluster are calculated. The event triggering rule is set as a combination of threshold conditions.

[0119] For each cluster of rapid outbreak events (e.g., clusters A to D above) obtained by clustering historical rapid event samples, the statistical distribution of each feature (mutation amplitude, rate of change, and frequency of mutations exceeding the noise threshold) of the historical rapid feature vector within the cluster is analyzed. For each feature (e.g., mutation amplitude, rate of change, and frequency of mutations exceeding the noise threshold), the mean and standard deviation within the cluster are calculated. The event triggering rule is set as a combination of threshold conditions.

[0120] It should be noted that the threshold condition combination is the sum of the mean and k times the standard deviation, where k is in the range of [-1.5, 1.5]. The sign of the threshold condition is a pre-set proportional coefficient based on the sensitivity requirements of event detection. The sign of the threshold condition depends on the correlation between the feature and the event (for example, congestion is usually accompanied by a negative increase in the "rate of change", so it is negative; accidents are accompanied by a positive increase in the "amplitude of change", so it is positive).

[0121] For example, for the "evening rush hour normal commuting congestion" cluster obtained by clustering historical trend event samples, the mean value of the change rate characteristic of the samples within the cluster is -0.2 and the standard deviation is 0.05. If k is set to -1.0 according to the detection sensitivity, the trigger threshold is calculated to be -0.25. The generated rule condition can be expressed as: if the real-time trend change rate is below -0.25 for 30 minutes, then this condition is met.

[0122] Real-time traffic status data is collected and processed to obtain real-time trend change component characteristics and rapid change component characteristics.

[0123] The real-time rapid component features are compared one by one with the triggering rules of all extracted rapid outbreak event clusters;

[0124] If it meets the rule conditions of a certain rapid outbreak event cluster, then it is immediately determined that a rapid outbreak event has occurred, and the event type is the cluster label corresponding to the rule (such as a serious vehicle rear-end collision).

[0125] If the real-time fast component features do not meet any fast burst event triggering rules, then the real-time trend component features will be compared one by one with the triggering rules of all extracted non-fast burst event clusters.

[0126] If it meets the rule conditions of a certain trend event cluster, it is determined that a non-rapid outbreak event has occurred, and the event type is the cluster label corresponding to the rule (such as normal commuting congestion during evening rush hour).

[0127] If the real-time characteristics do not meet any rapid event rules or trend event rules, the current traffic status is determined to be a normal fluctuation and no event warning is triggered.

[0128] Step S30 involves performing cluster analysis on historical event samples to automatically discover and summarize the characteristic patterns of different types of traffic events, transforming implicit and scattered empirical data into explicit and structured event pattern clusters, thereby realizing the mechanization and standardization of traffic management.

[0129] Based on the statistical characteristics of each cluster, corresponding triggering rules are automatically generated. These rules are dynamically determined by the distribution of historical data and can be adjusted through a sensitivity coefficient (k value). This makes the warning rules adaptive and targeted, capable of matching the unique evolution patterns of different road segments, time periods, and types of events, thus fundamentally optimizing the false alarms and missed alarms caused by the fixed threshold method.

[0130] In real-time operation, the extracted features are precisely matched with all pre-generated rules, thereby not only determining whether an event has occurred, but also identifying the specific type of the event, providing a key decision-making basis for subsequent differentiated responses.

[0131] Step S40: Based on the determined traffic event triggering type, generate a traffic linkage early warning instruction and send an early warning or linkage dispatch strategy to the traffic control system;

[0132] Specifically, in step S40, based on the determined specific traffic event trigger type, a preset early warning instruction and linkage dispatch strategy are matched, and different types of events correspond to different emergency response and control strategies;

[0133] For example, for rapidly erupting incidents, the linkage strategy can provide immediate response and local control. The instructions may include: immediately notifying the traffic signal control system near the accident site to forcibly switch the directional lights to red or flashing yellow; pushing the alarm information containing precise coordinates, incident type, and estimated impact range to the traffic police command platform and emergency rescue departments; and issuing guidance information such as "Accident ahead, please detour" to vehicles about to enter the affected area through roadside variable message signs or linkage with vehicle navigation platforms.

[0134] For non-rapid outbreak events, the coordinated strategy is traffic control and regional guidance. The instructions may include: notifying the signal control system of relevant upstream intersections to activate the targeted "congestion coordination" timing scheme, extending the green light time for outbound directions and shortening the green light time for inbound directions, and issuing a prompt message such as "XX road section is congested, it is recommended to choose XX alternative route" through traffic guidance screens, radio, and navigation apps, and pushing the early warning information and suggested dispatching strategies to the traffic management command center.

[0135] The core of this embodiment lies in separating and identifying pattern components with different propagation / evolution speeds, thus forming a method for temporal pattern separation and independent identification based on differences in state evolution speed. Specifically, through principal component analysis and wavelet transform, components representing rapid disturbances and components representing slow-moving accumulations are separated from mixed traffic data. Differentiated clustering models and data-driven triggering rules are constructed for these two fundamentally different patterns, thereby distinguishing between sudden accidents and slow-moving congestion from the source of temporal evolution. This systematically optimizes the problems of false alarms due to instantaneous noise and missed alarms for trend events, achieving a qualitative change in early warning from perceiving phenomena to analyzing mechanisms.

[0136] Example 2

[0137] Based on the same inventive concept as the smart monitoring-based urban traffic linkage early warning method in the foregoing embodiments, such as Figure 2 As shown, this application provides an urban traffic linkage early warning system based on intelligent monitoring, wherein the system specifically includes:

[0138] Traffic data acquisition module: Acquires multi-source traffic monitoring data, performs multi-modal fusion processing on the multi-source traffic monitoring data, and forms a traffic status data sequence of the target road network;

[0139] Traffic change analysis module: Constructs traffic change curves based on traffic state data sequences, separates the change features of the traffic change curves, and classifies the change features into rapid change components and trend change components;

[0140] Traffic incident determination module: By combining the statistical analysis results of historical traffic status data with the characteristics of change, feature clustering and event triggering rule extraction are performed on the rapidly changing components and the trend changing components respectively to determine the traffic incident triggering type;

[0141] Traffic incident triggering types include rapid-onset incidents and non-rapid-onset incidents;

[0142] Linked early warning module: Based on the determined traffic event trigger type, it generates traffic linked early warning instructions and issues early warning or linked dispatch strategies to the traffic control system.

[0143] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A city traffic linkage early warning method based on intelligent monitoring, characterized in that: include: Acquire multi-source traffic monitoring data, perform multi-modal fusion processing on the multi-source traffic monitoring data, and form a traffic status data sequence for the target road network; Traffic change curves are constructed based on traffic state data sequences. The change features of the traffic change curves are separated and classified into rapid change components and trend change components. The separated change features are then quantified. By combining the statistical analysis results of historical traffic status data with the change characteristics, feature clustering and event triggering rule extraction are performed on the rapidly changing components and the trend changing components to determine the traffic event triggering type. Traffic incident triggering types include rapid-onset incidents and non-rapid-onset incidents; The process of performing feature clustering on rapidly changing components and trend-changing components is as follows: Based on a historical time sample feature library containing historical trend event sample sets and historical fast event sample sets; For the historical trend event sample set, a density-based clustering algorithm is used for cluster analysis, dividing it into several non-fast burst clusters, with each cluster corresponding to a non-fast burst event; For the historical fast event sample set, a hierarchical clustering algorithm is used for cluster analysis, which divides it into several fast outbreak clusters, with each cluster corresponding to a fast outbreak event; The process of obtaining the historical time sample feature library: Generate a corresponding historical feature vector for each historical event sample from historical traffic status data; For each historical event period, extract the trend change component features to construct a historical trend feature vector, and extract the rapid change component features to construct a historical rapid feature vector; The trend feature vectors and fast feature vectors of all historical events are collected separately to form a historical trend event sample set and a historical fast event sample set, which serve as a historical event sample feature library. The process of obtaining event triggering rules: For each non-rapidly breaking event cluster obtained by clustering historical trend event samples, the statistical distribution of all historical trend feature vectors within the cluster is analyzed. For each feature, the mean and standard deviation within the cluster are calculated, and the event triggering rule is set as a combination of threshold conditions. For each cluster of rapid outbreak events obtained by clustering historical rapid event samples, the statistical distribution of each feature of the historical rapid feature vector within the cluster is analyzed. For each feature, the mean and standard deviation within the cluster are calculated, and the event triggering rule is set as a combination of threshold conditions. The threshold condition combination is the sum of the mean and k times the standard deviation, and the sign depends on the direction of the correlation between the feature and the event; Based on the determined traffic event trigger type, a traffic linkage early warning instruction is generated, and an early warning or linkage dispatch strategy is sent to the traffic control system.

2. The urban traffic linkage early warning method based on intelligent monitoring according to claim 1, characterized in that: The process of obtaining traffic status data sequences is as follows: The collected multi-source traffic monitoring data is preprocessed, and feature parameters representing traffic status are extracted from the preprocessed multi-source traffic monitoring data. The characteristic parameters include traffic flow, average vehicle speed, vehicle location distribution, vehicle dwell time, road occupancy, and weather visibility. The extracted feature parameters are concatenated into a unified multidimensional feature vector according to time segments. The high-dimensional features are then reduced by an autoencoder. The fused multidimensional feature vector is sampled at fixed time intervals and arranged in chronological order to form a traffic state data sequence of the target road network.

3. The urban traffic linkage early warning method based on intelligent monitoring according to claim 1, characterized in that: The process of constructing traffic curves is as follows: Principal component analysis was used to extract traffic flow, average speed and road occupancy from traffic state data series; The score sequence of the first principal component is selected as the comprehensive traffic state index. The traffic change curve of the target road network is plotted with time as the horizontal axis and the comprehensive traffic state index as the vertical axis.

4. The urban traffic linkage early warning method based on intelligent monitoring according to claim 1, characterized in that: The process of separating change features includes: Wavelet transform is used to decompose the traffic change curve into a signal. By selecting the wavelet basis function and the number of decomposition levels, discrete wavelet transform is performed on the traffic change curve to obtain a set of approximate coefficients and detail coefficient sequences. The trend change component is obtained by reconstructing the approximate coefficients of the deepest layer, and the rapid change component is obtained by superimposing the detailed coefficients of each layer.

5. The urban traffic linkage early warning method based on intelligent monitoring according to claim 1, characterized in that: The process of quantifying the changes after separation is as follows: Calculate the first derivative of the trend change component to quantify the rate of change, and extract the cumulative change and the duration of the trend within the sliding window; For rapidly changing components, detect local extreme points, calculate the abrupt change amplitude and rate of change between adjacent extreme points, and count the frequency of abrupt changes exceeding a preset noise threshold per unit time.

6. The urban traffic linkage early warning method based on intelligent monitoring according to claim 1, characterized in that: The process for determining the trigger type of a traffic incident is as follows: Real-time traffic status data is collected and processed to obtain real-time trend change component characteristics and rapid change component characteristics. The real-time rapid component features are compared one by one with the traffic event triggering rules of all extracted rapid outbreak event clusters; If the rule conditions of a certain fast-burst event cluster are met, then a fast-burst event is determined to have occurred, and the event type is the cluster label corresponding to that rule; If the real-time fast component features do not meet any fast event triggering rules, then the real-time trend component features will be compared one by one with the triggering rules of all extracted non-fast burst event clusters. If it meets the rule conditions of a certain trend event cluster, it is determined that a non-rapid burst event has occurred, and the event type is the cluster label corresponding to the rule.

7. A smart monitoring-based urban traffic linkage early warning system, characterized in that, The system is used to perform the method according to any one of claims 1-6, the system comprising: Traffic data acquisition module: Acquires multi-source traffic monitoring data, performs multi-modal fusion processing on the multi-source traffic monitoring data, and forms a traffic status data sequence of the target road network; Traffic change analysis module: Constructs traffic change curves based on traffic state data sequences, separates the change features of the traffic change curves, classifies the change features into rapid change components and trend change components, and quantifies the separated change features; Traffic incident determination module: By combining the statistical analysis results of historical traffic status data with the characteristics of change, feature clustering and event triggering rule extraction are performed on the rapidly changing components and the trend changing components respectively to determine the traffic incident triggering type; Traffic incident triggering types include rapid-onset incidents and non-rapid-onset incidents; Linked early warning module: Based on the determined traffic event trigger type, it generates traffic linked early warning instructions and issues early warning or linked dispatch strategies to the traffic control system.

Images