Situation awareness traffic control method and system based on time sequence large model and medium

Abstract

This application provides a situational awareness traffic control method, system, and medium based on a time-series large model, belonging to the field of traffic control technology. The method includes processing road monitoring video streams using a time-series large model to determine the situational awareness dataset corresponding to a first preset time period. Based on historical situational awareness time-series data analysis, the method analyzes the changes in the perceived data of the situational awareness dataset, matching each associated road segment corresponding to the monitored road segment of the first monitoring device to construct a road network space of corresponding association types. According to the similar road association characteristics corresponding to the road network space, the method matches relevant weight sets corresponding to multi-source data types to perform weighted fusion of multi-source data from the monitoring device group in the road network space, obtaining a road control level determination index corresponding to the road network space. Based on each road control level determination index and the preset situational awareness control dataset, the method determines corresponding traffic control strategies and sends each traffic control strategy to the traffic control management terminal.

Description

Technical Field

[0001] This application relates to the field of traffic control technology, and in particular to a situational awareness traffic control method, system and medium based on a time-series large model. Background Technology

[0002] With the continuous expansion of urban transportation and the increasing demand for intelligent development, traffic monitoring and management systems are facing unprecedented challenges. Currently, although infrastructure such as video surveillance equipment has been widely deployed in the road traffic sector, significant shortcomings remain in traffic situation awareness and control strategy formulation under conventional traffic scenarios.

[0003] Existing traffic flow analysis methods mostly rely on traditional statistical models or manual analysis, which makes it difficult to effectively process the complex spatiotemporal information contained in massive video surveillance data. They are prone to overlooking the different hidden factors that may affect traffic on each road, resulting in a lag in the perception of traffic conditions and failing to meet real-time requirements.

[0004] Meanwhile, existing technologies typically analyze traffic data from individual road segments in isolation, failing to predict the propagation path and scope of traffic congestion in advance. Furthermore, current decisions, often based on fixed rules or human experience, cannot dynamically generate optimal control strategies based on real-time traffic conditions. This approach is not only labor-intensive but also prone to decision-making biases due to subjective factors, making it difficult to adapt to complex and ever-changing traffic scenarios. It also hinders the efficient allocation of traffic resources and the scientific management of urban traffic in conventional scenarios, severely restricting the healthy operation and development efficiency of urban transportation systems. Summary of the Invention

[0005] To address the aforementioned issues, this application provides a situational awareness traffic control method, system, and medium based on a time-series large model. This method enables timely traffic situational awareness, intelligent situational awareness, and traffic control, thereby scientifically managing urban traffic in conventional scenarios while reducing labor costs.

[0006] On the one hand, embodiments of this application provide a situational awareness traffic control method based on a time-series large model, the method comprising:

[0007] The road monitoring video stream collected by the first monitoring device is input into a pre-trained time series large model to determine the situational awareness dataset corresponding to the first preset time period based on the model output results.

[0008] Based on the analysis of the changes in the perception data of the situational awareness dataset according to the historical situational awareness time series data, the associated road segments corresponding to the monitoring road segments of the first monitoring device are matched to construct the road network space of the corresponding association type.

[0009] Based on the similar road association features corresponding to the road network space, a relevant weight set corresponding to the multi-source data type is matched, and the multi-source data of the monitoring equipment group in the road network space is weighted and fused according to the relevant weight set to obtain the road control level determination index corresponding to the road network space.

[0010] Based on the road control level determination index and the preset situational awareness control dataset, the corresponding traffic control strategies are determined and sent to the traffic control management terminal.

[0011] In one implementation of this application, the road monitoring video stream collected by the first monitoring device is input into a pre-trained time-series large model to determine the situational awareness dataset corresponding to the first preset time period based on the model output, specifically including:

[0012] Using the aforementioned time-series large model, various traffic entities in the road monitoring video stream during the second preset time period are identified, and the traffic flow curves corresponding to each type of traffic entity are determined; there are no sudden events in the road monitoring video stream.

[0013] According to the preset time interval nesting, determine the traffic flow coding matrix corresponding to the traffic flow curve of each traffic entity, and output multiple initial situational awareness datasets corresponding to the traffic flow coding matrix; wherein, each initial situational awareness dataset has a dependency relationship with the sub-interval in the preset time interval nesting.

[0014] Based on each of the initial situational awareness datasets, a sequence of perception data difference coefficients corresponding to each of the sub-intervals is determined; wherein, the sequence of perception data difference coefficients contains perception data difference coefficients equal to the number of sub-intervals in the nested preset time intervals; the perception data difference coefficients are determined based on the historical same-period difference sub-coefficients and the parent interval structure difference sub-coefficients.

[0015] The traffic distribution data of each sub-interval in the first preset time period is predicted by the time series large model based on the difference coefficient sequence of the perceived data, and the situational awareness dataset corresponding to the first preset time period is output based on the traffic distribution data after weighted calculation.

[0016] In one implementation of this application, the type of traffic entity includes at least one or more of the following: motor vehicles, non-motor vehicles, pedestrians, and buses;

[0017] Determine the traffic flow curves corresponding to each type of traffic entity, specifically including:

[0018] At preset time intervals, the number of traffic entities passing through the monitored road segment is statistically sequenced within the second preset time period, and the traffic flow sequence of each type of traffic entity is calculated per unit time based on the number of traffic entities passing through each entity.

[0019] Each traffic entity's flow sequence is subjected to curve fitting to generate a flow curve for each traffic entity.

[0020] In one implementation of this application, based on historical situational awareness time-series data analysis, the perception data change results of the situational awareness dataset are matched with each associated road segment corresponding to the monitoring road segment of the first monitoring device to construct a road network space of corresponding association type, specifically including:

[0021] Based on the historical situational awareness time-series data, a first situational awareness feature sequence corresponding to each road segment is determined; the first situational awareness feature sequence includes at least multiple traffic flow features of the corresponding road segment; the traffic flow feature is at least one of the following: mean traffic flow, standard deviation of traffic flow, and peak traffic flow.

[0022] Based on the first situational awareness feature sequences and the situational awareness dataset of the same historical period of the first preset time period, the corresponding perception data association multivariate coefficient set is calculated; wherein, the perception data association multivariate coefficient set includes the flow similarity calculated from the flow distribution data of various traffic entities.

[0023] Based on the various multivariate coefficient groups associated with the sensing data and the preset filtering conditions, the corresponding multivariate coefficient groups associated with the sensing data are filtered and the corresponding monitoring road segments are determined as the associated road segments; wherein, the preset filtering conditions are used to compare the multivariate coefficient threshold group with the multivariate coefficient groups associated with the sensing data, and filter a predetermined number of multivariate coefficient groups associated with the sensing data whose traffic similarity is greater than the corresponding multivariate coefficient threshold.

[0024] Based on the traffic entity types that meet the preset screening conditions and the preset association type comparison table, the corresponding association type is determined, and the monitored road segment of the first monitoring device and its corresponding associated road segments are added to the road network space of the corresponding association type.

[0025] In one implementation of this application, matching a relevant weight set corresponding to the multi-source data type is performed based on the similar road association features corresponding to the road network space, specifically including:

[0026] According to a preset association type lookup table, the similar road association features corresponding to the association type in the road network space are determined; wherein, the preset association type lookup table contains several association types and the corresponding relationships of each similar road association feature obtained by pre-clustering;

[0027] Based on the similar road association features and the preset association feature weight mapping relationship, the relevant weights of each of the multi-source data types are matched and the relevant weight set is constructed.

[0028] In one implementation of this application, multi-source data from the monitoring device group in the road network space are weighted and fused according to the relevant weight set to obtain a road control level determination index corresponding to the road network space, specifically including:

[0029] Acquire multi-source data collected by the monitoring equipment group in the road network space during the first preset time period, and perform weighted processing on the data of corresponding multi-source data types in the multi-source data through the relevant weight set to obtain a weighted multi-source data vector;

[0030] The weighted multi-source data vector is input into a multi-dimensional scoring model to determine the road control level judgment index based on the model output; wherein, the multi-dimensional scoring model is a pre-trained machine learning model used to calculate preset multi-dimensional sub-scorings.

[0031] In one implementation of this application, corresponding traffic control strategies are determined based on the road control level determination index and a preset situational awareness control dataset, and each traffic control strategy is sent to a traffic control management terminal, specifically including:

[0032] The road control level determination index corresponding to each of the road network spaces is matched with the preset situational awareness control dataset to map the initial traffic control strategy corresponding to each road control level, and then sent to the expert system for verification.

[0033] The initial traffic control strategy verified from the expert system is used as the traffic control strategy, and each of the traffic control strategies is sent to the traffic control management terminal.

[0034] In one implementation of this application, the method further includes:

[0035] Based on the continuous road control level determination index of each road network space within the first preset time period, generate change curves for each determination index respectively;

[0036] Based on the change curves of each of the judgment indices and the location information of each of the road network spaces, it is determined whether the correlation between adjacent road network spaces has changed; the correlation is that the change curves of each of the judgment indices are at least negatively correlated and have no significant correlation.

[0037] If so, starting from the current time, the road monitoring video stream collected by the first monitoring device is acquired in real time to update the constructed road network space.

[0038] On the other hand, embodiments of this application also provide a situational awareness traffic control system based on a time-series large model, the system comprising:

[0039] The input module is used to input the road monitoring video stream collected by the first monitoring device into a pre-trained time series large model, so as to determine the situational awareness dataset corresponding to the first preset time period based on the model output results; wherein, the situational awareness dataset includes at least traffic flow distribution data of multiple types of traffic entities;

[0040] The first matching module is used to analyze the changes in the perception data of the situational awareness dataset based on historical situational awareness time-series data, and match each associated road segment corresponding to the monitoring road segment of the first monitoring device to construct a road network space of the corresponding association type.

[0041] The second matching module is used to match the relevant weight set corresponding to the multi-source data type according to the similar road association features corresponding to the road network space, so as to perform weighted fusion of the multi-source data of the monitoring equipment group in the road network space according to the relevant weight set to obtain the road control level judgment index corresponding to the road network space.

[0042] The sending module is used to determine the corresponding traffic control strategy based on the road control level judgment index and the preset situational awareness control dataset, and send the traffic control strategy to the traffic control management terminal.

[0043] Furthermore, embodiments of this application also provide a non-volatile computer storage medium storing computer-executable instructions, which are capable of executing a situational awareness traffic control method based on a time-series large model as described above.

[0044] Compared with the prior art, the significant advantages of this application are as follows:

[0045] This application, through the aforementioned technical solution, firstly utilizes a large-scale temporal model to efficiently process video data and accurately identify the traffic flow distribution data of various traffic entities. Compared to traditional analysis methods relying on manual or statistical models, it can shorten the response time of traffic situation awareness from minutes to seconds, effectively solving the problem of perception lag in existing technologies. Secondly, it uses historical situation awareness time-series data to analyze changes in traffic perception data, constructs a road network space, and clarifies the association types of each road segment, breaking the limitations of existing technologies that analyze road segment data in isolation.

[0046] Furthermore, by combining the road control level determination index with a pre-set situational awareness control dataset to determine traffic control strategies, this approach changes the existing model of decision-making based on fixed rules or human experience. This application achieves full automation from traffic data collection and analysis to control strategy generation, significantly reducing human intervention. It makes traffic control strategies in routine scenarios more objective and scientific, avoiding the arbitrariness and uncertainty of human decision-making, significantly improving the level of scientific management of urban traffic, and promoting the development of urban traffic systems towards intelligence and efficiency. Attached Figure Description

[0047] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0048] Figure 1 This is a flowchart illustrating a situational awareness traffic control method based on a large temporal model, as described in an embodiment of this application.

[0049] Figure 2 This is a schematic diagram of the structure of a situational awareness traffic control system based on a time-series large model in an embodiment of this application. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0051] This application provides a situational awareness traffic control method, system, and medium based on a time-series large model to address the technical problems of untimely traffic situational awareness, insufficient intelligence in traffic control based on situational awareness, high manpower costs, and difficulty in scientifically managing urban traffic in conventional traffic scenarios.

[0052] This application aims to provide traffic management departments with a general traffic control strategy that requires no manpower in routine scenarios, rather than to address sudden traffic incidents. In practical scenarios, multi-source sensing devices such as cameras and geomagnetic sensors can be deployed to monitor sudden events such as traffic accidents in real time. For locations where sudden events occur, traffic management departments can then implement traffic control. That is, this application does not directly control traffic lights such as those at intersections, but rather provides a preferred control strategy for reference by traffic management departments and related road departments.

[0053] The various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0054] This application provides a situational awareness-based traffic control method based on a large temporal model, such as... Figure 1 As shown, the method may include steps S101-S104:

[0055] S101, the server inputs the road monitoring video stream collected by the first monitoring device into the pre-trained time series large model, so as to determine the situational awareness dataset corresponding to the first preset time period based on the model output results.

[0056] The situational awareness dataset includes at least traffic flow distribution data, speed data, traffic incident data, and environmental data for multiple types of traffic entities. In addition to processing traffic flow distribution data, this application embodiment can also perform fusion analysis of speed data, traffic incident data, and environmental data according to user needs, thereby enabling situational awareness of traffic; however, no specific limitations are specified here.

[0057] It should be noted that the server, as the executing entity of the situational awareness traffic control method based on the time series large model, exists only as an example, and the executing entity is not limited to the server. This application does not make any specific limitation in this regard.

[0058] The first monitoring device can be understood as a camera deployed on the main road, which can collect road monitoring video streams in real time or according to user-defined working hours. The server connects to the first monitoring device via wired or wireless connection and can acquire the road monitoring video streams collected by the first monitoring device. The pre-trained time-series large model can be a pre-trained large model based on the Transformer architecture. At the same time, this application can use real-time perception data such as vehicle-to-X (V2X) wireless communication technology, UAV inspection data, and meteorological API data as model inputs to the time-series large model, dynamically adjusting the model weights to avoid model update lag.

[0059] In this embodiment of the application, the above-mentioned inputting the road monitoring video stream collected by the first monitoring device into a pre-trained time-series large model to determine the situational awareness dataset corresponding to the first preset time period based on the model output results specifically includes:

[0060] Using a time-series large-scale model, various traffic entities in the road monitoring video stream during the second preset time period are identified, and the traffic flow curves corresponding to each traffic entity are determined. No sudden events occur in the road monitoring video stream. A traffic flow coding matrix corresponding to the traffic flow curves of each traffic entity is determined according to a preset time interval nesting, and multiple initial situational awareness datasets corresponding to the traffic flow coding matrices are output. Each initial situational awareness dataset has a dependency relationship with a sub-interval in the preset time interval nesting. Based on each initial situational awareness dataset, a sequence of perception data difference coefficients corresponding to each sub-interval is determined. This sequence contains perception data difference coefficients equal to the number of sub-intervals in the preset time interval nesting. The perception data difference coefficients are determined based on historical contemporaneous difference sub-coefficients and parent interval structural difference sub-coefficients. The traffic flow distribution data of each sub-interval in the first preset time period predicted by the time-series large-scale model is weighted according to the perception data difference coefficient sequence, and the situational awareness dataset corresponding to the first preset time period is output based on the weighted traffic flow distribution data.

[0061] The types of traffic entities mentioned above include at least one or more of the following: motor vehicles, non-motor vehicles, pedestrians, and buses. In other words, the time-series large-scale model can analyze road surveillance video streams and identify traffic entities such as motor vehicles, non-motor vehicles, pedestrians, and buses in the road surveillance video stream frame by frame within a second preset time period. Motor vehicles, non-motor vehicles, pedestrians, and buses are only examples; users can add more traffic entities in actual scenarios, and this application does not impose specific limitations. The second preset time period can be set by the user according to the actual road usage scenario, and this application does not impose specific limitations on it. Simultaneously, the server will also count the traffic flow of each type of traffic entity and establish a traffic flow curve in a Cartesian coordinate system.

[0062] Subsequently, the server will further analyze the traffic flow curves of each traffic entity using a preset time interval nesting method, extract the traffic flow corresponding to different sub-intervals within the preset time interval nesting method, encode it, generate traffic flow codes, and then construct a traffic flow coding matrix with traffic entity type as rows and different sub-intervals as columns. The preset time interval nesting method can be understood as […]. ]Include[ ], [ ]Include[ ... and This can be understood as the start and end times of the second preset time period. The server also uses the traffic flow coding matrix as an intermediate input to the time series model to generate multiple initial situational awareness datasets based on different sub-intervals. This initial situational awareness dataset can be understood as a dataset consisting of a column of the traffic flow coding matrix.

[0063] Next, the server uses the initial situational awareness datasets obtained above to calculate the sensing data difference coefficients corresponding to each sub-interval, and constructs a sequence of sensing data difference coefficients. The order in this sequence is sorted from smallest to largest according to the time range of the corresponding sensing data difference coefficient sub-interval. For example, a sub-interval of a sensing data difference coefficient might be [...]. Another sub-interval of the perceived data difference coefficient is []. ], [ ]for[ A parent interval of ], [ The time range of ] is greater than [ The time range of ], then in the sequence of perceived data difference coefficients [ The corresponding perceptual data difference coefficient Should be in [ The corresponding perceptual data difference coefficient Then, the server uses a time-series large model to predict the traffic distribution data of each sub-interval in the first preset time period. It normalizes the difference coefficients of the perceived data in the above-mentioned perceived data difference coefficient sequence, ensuring that each coefficient falls within the (0, 1) range, and that the sum of the coefficients in the sequence is 1. Then, it combines the normalized perceived data difference coefficient sequence with the weighted summation of the traffic distribution data of each sub-interval for different traffic entities, obtaining the weighted traffic distribution data for each traffic entity. This allows the construction of a situational awareness dataset containing the traffic distribution data of each traffic entity within the first preset time period.

[0064] In one embodiment of this application, the determination of traffic flow curves corresponding to various traffic entities specifically includes:

[0065] At preset time intervals, the number of traffic entities passing through the monitored road segment is statistically analyzed within a second preset time period. Based on the number of traffic entities passing through each entity, the traffic flow sequence of each traffic entity per unit time is calculated. The traffic flow sequence of each traffic entity is then curve-fitted to generate the traffic flow curve for each traffic entity.

[0066] In other words, the server has a preset time interval. This preset time interval can be selected from the smallest sub-interval of the aforementioned time range, or it can be set by the user according to the scenario; no specific limitation is made here. The server counts the number of entities passing through each preset time interval within the second preset time period, and forms a sequence of entity passing through time. Subsequently, the server calculates the traffic entity flow per unit time and performs curve fitting on the sequence to establish a traffic entity flow curve in a Cartesian coordinate system with traffic entity flow as the vertical axis and time as the horizontal axis.

[0067] In another embodiment of this application, the aforementioned historical time interval difference sub-coefficient is calculated based on the first traffic flow feature vector corresponding to the sub-interval and the second traffic flow feature vector of the historical time interval sub-interval; the parent interval structure difference sub-coefficient is calculated based on the first traffic flow feature vector corresponding to the sub-interval and the third traffic flow feature vector corresponding to its parent interval in the same preset time interval set.

[0068] Specifically, based on the initial situational awareness dataset corresponding to the sub-interval, the first traffic flow feature vector corresponding to the sub-interval is obtained as follows: , It can be understood as the first The first sub-interval Traffic flow characteristic values ​​corresponding to each traffic entity. The second traffic flow characteristic vector for the historical same period sub-interval is: The historical period difference sub-coefficient is calculated using the following formula. :

[0069]

[0070] in, This is a first minimum constant to avoid a denominator of 0, set by the user according to the actual usage scenario, and is not specifically limited here. Parent interval structure difference coefficients The calculation formula is as follows:

[0071]

[0072] in, express The parent interval, in As described above [ When the interval is within a preset time interval nesting, that is, when it has no parent interval, its The third traffic flow feature vector, pre-set by the user on the server, is used; no specific limitations are specified here. The coefficient of difference in perceived data can be expressed by the formula... The calculation yielded that, The preset balance weights for the user, these balance weights The specific value is an empirical value, set by the user based on the actual usage scenario, and is not specifically limited here. The formula for calculating the perceived data difference coefficient is: , The coefficient of difference in the perceived data. For the preset balance weights, The coefficient represents the difference between historical periods. The coefficients of the parent interval structure differences.

[0073] S102, the server analyzes the changes in the perception data of the situational awareness dataset based on historical situational awareness time-series data, and matches each associated road segment with the monitoring road segment of the first monitoring device to construct a road network space of the corresponding association type.

[0074] In this embodiment of the application, based on the analysis of the changes in the sensing data of the situational awareness dataset using historical situational awareness time-series data, each associated road segment corresponding to the monitoring road segment of the first monitoring device is matched to construct a road network space of the corresponding association type, specifically including:

[0075] Based on historical situational awareness time-series data, the first situational awareness feature sequence corresponding to each road segment is determined. The first situational awareness feature sequence includes at least multiple traffic flow characteristics of the corresponding road segment. The traffic flow characteristics are at least one of the following: mean traffic flow, standard deviation of traffic flow, and peak traffic flow. Based on the first situational awareness feature sequences and situational awareness datasets from the same historical period of the first preset time period, the corresponding sensing data association multivariate coefficient sets are calculated. These sets include traffic flow similarity calculated from the traffic flow distribution data of various traffic entities. Based on each sensing data association multivariate coefficient set and preset filtering conditions, the corresponding sensing data association multivariate coefficient sets are filtered, and their corresponding monitored road segments are determined as associated road segments. The preset filtering conditions are used to compare the multivariate coefficient threshold set with the sensing data association multivariate coefficient sets, and to filter a predetermined number of sensing data association multivariate coefficient sets whose traffic flow similarity is greater than the corresponding multivariate coefficient threshold. Based on the traffic entity types that meet the preset filtering conditions and a preset association type comparison table, the corresponding association types are determined, and the monitored road segments of the first monitoring equipment and their corresponding associated road segments are added to the road network space of the corresponding association type.

[0076] In other words, this application can analyze historical situational awareness time-series data, extract traffic flow characteristics, and calculate the similarity between the first situational awareness feature sequence corresponding to the first preset time period and the aforementioned situational awareness dataset. For example, if the first preset time period is from 8:00 AM to 10:00 AM on Sunday, the historical similarity could be from 8:00 AM to 10:00 AM the previous Sunday. The traffic flow similarity can be calculated by performing a traffic flow similarity calculation on the first situational awareness feature sequence and the traffic flow distribution data of the corresponding traffic entity type in the situational awareness dataset, using cosine similarity or the reciprocal of Euclidean distance as the traffic flow similarity.

[0077] Then, the calculated traffic flow similarities for different traffic entity types are added to the multivariate coefficient group associated with the sensing data as a result of the sensing data change. The values ​​of each element in the multivariate coefficient group are compared with the corresponding multivariate coefficient thresholds in the multivariate coefficient threshold group, which contains multiple multivariate coefficient thresholds corresponding to different traffic entity types. If a predetermined number of traffic flow similarities in a multivariate coefficient group are greater than the corresponding multivariate coefficient threshold, the road segment corresponding to that multivariate coefficient group in the historical situational awareness time-series data is designated as the associated road segment of the currently monitored road segment. Simultaneously, the traffic entity types corresponding to the traffic flow similarities greater than the corresponding multivariate coefficient thresholds are extracted, and the extracted predetermined number of traffic entity types are compared with a preset association type lookup table to obtain the association types recorded in the table. Subsequently, a road network space with association type labels and containing the monitored road segment and its associated road segments is established. Furthermore, when determining the first situational awareness feature sequence for the same historical period, if the current road segment has dynamic factors that did not occur in the same historical period, such as road repairs, map openings, or traffic accidents, it may be impossible to match the corresponding first situational awareness feature sequence for the same historical period using the aforementioned traffic similarity. For example, if the traffic similarity of all traffic segments with the first situational awareness feature sequences for each historical period is less than the corresponding multivariate coefficient threshold, and it is determined that there is no relevant historical situational awareness time series data in the historical data, the server can obtain the default historical situational awareness time series data with different dynamic factors added by the experts from the expert system, and extract the first situational awareness feature sequence for that road segment.

[0078] The aforementioned association type can be understood as follows: the traffic entity types corresponding to motor vehicles and non-motor vehicles in the first and second road segments have a high similarity in traffic flow, exceeding the corresponding multivariate coefficient threshold. However, the similarity in pedestrian traffic flow is lower than the corresponding multivariate coefficient threshold. If the predetermined quantity is 2, then the second road segment is an associated road segment of the first road segment, and its association type can be either motor vehicle or non-motor vehicle association. The specific values ​​of each multivariate coefficient threshold in the aforementioned multivariate coefficient threshold group can be set by the user according to the actual usage scenario, and are not specifically limited here.

[0079] It should be noted that the situational awareness dataset corresponding to the first preset time period is determined by processing the road monitoring video stream through a pre-trained time series large model. If the road section is closed due to construction or an emergency, the historical traffic data and the current actual traffic may not be completely matched. In this case, the associated road section cannot be obtained through the historical situational awareness time series data, or a mismatch may occur. Therefore, this application can also deploy an emergency scene recognition model on the edge device (road edge device) to identify whether there is an emergency in the road monitoring video stream before inputting the road monitoring video stream collected by the first monitoring device into the pre-trained time series large model. If there is an emergency, the edge device sends an emergency event prompt message to the server. At this time, the server will not perform associated road segment matching for the monitoring road segment corresponding to the road monitoring video stream, but will generate an emergency event prompt message and send it to the traffic control management terminal, allowing the user to select an associated road segment for the monitoring road segment, or the user to resolve the emergency. In addition, emergency association rules can be deployed on the edge device to automatically associate the upstream and downstream road segments of the monitoring road segment where the emergency occurred to construct a road network space, and call the emergency strategy in the emergency association rules. The emergency strategy can include traffic light control strategies, dispatching nearby traffic police, etc., and the specific settings are set according to the actual use scenario, which are not specifically limited here. The aforementioned emergency scene recognition model can be a neural network model trained by using historical road segments and corresponding emergency event data samples that cannot be accurately matched by a time series large model under several different weather conditions and time periods (time series large model failure), or it can be a machine learning model or a deep learning model. This application does not make any specific limitations on this.

[0080] The above scheme can cluster and associate monitored road segments, and construct a road network space according to the association type, which facilitates more detailed road division and ensures accurate situational awareness of the roads.

[0081] S103, the server matches the relevant weight set corresponding to the multi-source data type based on the similar road association features corresponding to the road network space, and performs weighted fusion of the multi-source data of the monitoring equipment group in the road network space according to the relevant weight set to obtain the road control level judgment index corresponding to the road network space.

[0082] In this embodiment of the application, based on the similar road association features corresponding to the road network space, a relevant weight set corresponding to the multi-source data type is matched, specifically including:

[0083] Based on a pre-defined association type lookup table, the similar road association features corresponding to the association types in the road network space are determined. This table contains several association types and their corresponding similar road association features obtained through pre-clustering. Based on the similar road association features and the pre-defined association feature weight mapping relationship, the relevant weights of each multi-source data type are matched, and a relevant weight set is constructed.

[0084] In other words, the aforementioned preset association type lookup table not only includes the correspondence between traffic entity types and association types, but also the correspondence between different association types and the association features of similar roads obtained through pre-clustering. Similar road association features can include road segment level, number of lanes, adjacent intersection types, traffic flow patterns, etc., which can be adjusted by the user according to the actual scenario, and are not specifically limited here. Subsequently, the server inputs the similar road association features into the mapping function corresponding to the preset association feature weight mapping relationship, calculates the relevant weights for different multi-source data types, and adds the relevant weights to the relevant weight set. The mapping function corresponding to this preset association feature weight mapping relationship can be constructed using a random forest, and can output different relevant weights after inputting different similar road association features. The preset association feature weight mapping relationship can also be established in other ways, such as linear functions, preset mapping relationship lookup tables, etc. The preset association feature weight mapping relationship can be set by experts and updated periodically, and is not specifically limited here. In addition, the update time for the relevant weight set can be set, such as updating once per hour. The relevant weight set can also be directly set by the expert system and updated periodically, and this application does not specifically limit this.

[0085] In one embodiment of this application, the above-mentioned weighted fusion of multi-source data from monitoring device groups in the road network space according to relevant weight sets to obtain the road control level determination index corresponding to the road network space specifically includes:

[0086] Multi-source data collected by monitoring equipment groups in the road network space within a first preset time period is acquired. Data of corresponding data types from these multi-source sources are weighted using relevant weight sets to obtain a weighted multi-source data vector. This weighted multi-source data vector is then input into a multi-dimensional scoring model to determine the road control level assessment index based on the model's output. The multi-dimensional scoring model is a pre-trained machine learning model used to calculate preset multi-dimensional sub-scores.

[0087] In other words, this application can use the aforementioned relevant weight set to weight the extracted multi-source data types, eliminating the need for manual analysis of multi-source data. It effectively analyzes the multi-source data types in the constructed road network space using the relevant weight set, obtaining a weighted multi-source data vector. For example, the relevant weight set might include a weight of 0.3 for video surveillance data, 0.2 for radar data, 0.1 for microphone array data, and 0.4 for geomagnetic data. This weight set is merely an example and is not specifically limited in this application. The calculated weighted multi-source data vector is input into a multi-dimensional scoring model, which can be a pre-trained machine learning model capable of calculating preset multi-dimensional sub-scores for different preset dimensions. The model then performs a weighted summation of these preset multi-dimensional sub-scores to calculate the road control level determination index. The specific weights for the weighted summation can be obtained from training data and are not specifically limited here. The preset multi-dimensional dimensions can include congestion, safety, efficiency, etc., and are not specifically limited here.

[0088] S104, the server determines the corresponding traffic control strategies based on the road control level judgment index and the preset situational awareness control dataset, and sends each traffic control strategy to the traffic control management terminal.

[0089] In this embodiment, the above-mentioned determination of corresponding traffic control strategies based on the road control level judgment index and the preset situational awareness control dataset, and the sending of each traffic control strategy to the traffic control management terminal, specifically includes:

[0090] The control level determination index corresponding to each road network space is matched with a preset situational awareness control dataset to map the initial traffic control strategy corresponding to each road control level, and then sent to an expert system for verification. The verified initial traffic control strategy from the expert system is then used as the traffic control strategy, and each traffic control strategy is sent to the traffic control management terminal.

[0091] This application can pre-define a classification standard, dividing road control levels into N levels. Each level corresponds to a different range of values ​​for a road control level judgment index. This range can be set by the user or an expert, and this application does not impose any specific limitations on it. A pre-defined situational awareness control dataset contains the correlation between different road control levels and different traffic control strategies. Through the matching operation between these two, an initial traffic control strategy can be obtained. Further verification is performed by an expert system, which can modify or confirm the initial traffic control strategy. The server then receives the verified initial traffic control strategy from the expert system and sends it to the traffic control management terminal. The traffic control management terminal can be understood as the terminal device of participants such as traffic management departments or traffic police, including but not limited to mobile phones, computers, and tablets, and this application does not impose any specific limitations on it. The traffic control strategy can include signal control strategies for various traffic entities, lane management strategies (such as tidal lanes, closed lanes, and flow restrictions), and route guidance strategies (such as recommended routes guided by navigation), etc. These can be specifically set by the user according to the actual usage scenario, and this application does not impose any specific limitations on them. In actual traffic control scenarios, management personnel can switch traffic control strategies or set traffic control strategies themselves based on the occurrence of traffic accidents, or actively enable or disable the execution function of the situational awareness traffic control method based on the time-series large model proposed in this application.

[0092] In one embodiment of this application, frequent updates to the road network space may result in situations where updates are unnecessary, wasting computational resources. Therefore, this application further includes:

[0093] Based on the continuous road control level determination indices for each road network space within a first preset time period, change curves for each determination index are generated. Based on these change curves and the location information of each road network space, it is determined whether the correlation between adjacent road network spaces has changed. The correlation is defined as at least a negative correlation, but no significant correlation, between the change curves of each determination index. If a change in the correlation between adjacent road network spaces is determined, starting from the current time, road monitoring video streams collected by the first monitoring device are acquired in real time to update the constructed road network spaces. Otherwise, the road network spaces are not updated.

[0094] In other words, this application can statistically analyze the continuous road control level determination indices of the same road network space within a first preset time period and generate the corresponding determination index change curves for the same road network space. Subsequently, the server compares the determination index change curves between adjacent road network spaces and determines whether the curves maintain their original negative correlation or no significant correlation. If the correlation changes, it indicates that a road control level relationship has occurred between the road network spaces. At this point, the server will re-execute steps S101-S102 of the road network space construction process, starting from the current time. If the determination index change curves between adjacent road network spaces maintain their original negative correlation or no significant correlation, then rebuilding the road network space is unnecessary. This technical solution eliminates the need to determine whether to rebuild the road network space after executing S101-S102; instead, it uses the aforementioned computationally efficient operation to determine whether to update and construct each road network space.

[0095] This application, through the aforementioned technical solution, firstly utilizes a large-scale temporal model to efficiently process video data and accurately identify the traffic flow distribution data of various traffic entities. Compared to traditional analysis methods relying on manual or statistical models, it can shorten the response time of traffic situation awareness from minutes to seconds, effectively solving the problem of perception lag in existing technologies. Secondly, it uses historical situation awareness time-series data to analyze changes in traffic perception data, constructs a road network space, and clarifies the association types of each road segment, breaking the limitations of existing technologies that analyze road segment data in isolation.

[0096] Furthermore, by combining the road control level determination index with a pre-set situational awareness control dataset to determine traffic control strategies, this approach changes the existing model of fixed rules or human experience-based decision-making. This application achieves full automation from traffic data collection and analysis to control strategy generation, significantly reducing human intervention. It makes traffic control strategies more objective and scientific, avoiding the arbitrariness and uncertainty of human decision-making, significantly improving the level of scientific management of urban traffic, and promoting the development of urban traffic systems towards intelligence and efficiency.

[0097] Figure 2 This application provides a schematic diagram of a situational awareness traffic control system based on a time-series large model. The situational awareness traffic control system 200 based on the time-series large model includes:

[0098] Input module 201 is used to input the road monitoring video stream collected by the first monitoring device into a pre-trained time-series large model to determine the situational awareness dataset corresponding to the first preset time period based on the model output results. The situational awareness dataset includes at least traffic flow distribution data for multiple types of traffic entities. First matching module 202 is used to analyze the changes in the perception data of the situational awareness dataset based on historical situational awareness time-series data, and match each associated road segment corresponding to the monitored road segment of the first monitoring device to construct a road network space of corresponding association types. Second matching module 203 is used to match the relevant weight sets corresponding to the multi-source data types based on the similar road association features corresponding to the road network space, and to perform weighted fusion of the multi-source data of the monitoring device group in the road network space according to the relevant weight sets to obtain the road control level determination index corresponding to the road network space. Determine and send module 204 is used to determine the corresponding traffic control strategies based on each road control level determination index and the preset situational awareness control dataset, and send each traffic control strategy to the traffic control management terminal.

[0099] This application also provides a non-volatile computer storage medium storing computer-executable instructions that can be executed:

[0100] The road monitoring video stream collected by the first monitoring device is input into a pre-trained time-series large-scale model to determine the situational awareness dataset corresponding to the first preset time period based on the model output. This situational awareness dataset includes traffic flow distribution data for at least multiple types of traffic entities. Based on historical situational awareness time-series data analysis, the changes in the perceived data in the situational awareness dataset are matched with the associated road segments corresponding to the monitored road segments of the first monitoring device to construct a road network space of corresponding association types. According to the similar road association features corresponding to the road network space, relevant weight sets corresponding to multi-source data types are matched. The multi-source data of the monitoring device group in the road network space are then weighted and fused according to the relevant weight sets to obtain the road control level determination index corresponding to the road network space. Based on each road control level determination index and the preset situational awareness control dataset, corresponding traffic control strategies are determined and sent to the traffic control management terminal.

[0101] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0102] The systems, media, and methods provided in this application are one-to-one correspondences. Therefore, the systems and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and media will not be repeated here.

[0103] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0104] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A situational awareness traffic control method based on a time-series large model, characterized in that, The method includes: The road monitoring video stream collected by the first monitoring device is input into a pre-trained time series large model to determine the situational awareness dataset corresponding to the first preset time period based on the model output results. Based on the analysis of the changes in the perception data of the situational awareness dataset according to the historical situational awareness time series data, the associated road segments corresponding to the monitoring road segments of the first monitoring device are matched to construct the road network space of the corresponding association type. Based on the similar road association features corresponding to the road network space, a relevant weight set corresponding to the multi-source data type is matched, and the multi-source data of the monitoring equipment group in the road network space is weighted and fused according to the relevant weight set to obtain the road control level determination index corresponding to the road network space. Based on the road control level determination index and the preset situational awareness control dataset, the corresponding traffic control strategies are determined and sent to the traffic control management terminal. Specifically, the road monitoring video stream collected by the first monitoring device is input into a pre-trained time-series large model to determine the situational awareness dataset corresponding to the first preset time period based on the model output, which includes: Using the aforementioned time-series large model, various traffic entities in the road monitoring video stream during the second preset time period are identified, and the traffic flow curves corresponding to each type of traffic entity are determined; there are no sudden events in the road monitoring video stream. According to the preset time interval nesting, determine the traffic flow coding matrix corresponding to the traffic flow curve of each traffic entity, and output multiple initial situational awareness datasets corresponding to the traffic flow coding matrix; wherein, each initial situational awareness dataset has a dependency relationship with the sub-interval in the preset time interval nesting. Based on each of the initial situational awareness datasets, a sequence of perception data difference coefficients corresponding to each of the sub-intervals is determined; wherein, the sequence of perception data difference coefficients contains perception data difference coefficients equal to the number of sub-intervals in the nested preset time intervals; the perception data difference coefficients are determined based on the historical same-period difference sub-coefficients and the parent interval structure difference sub-coefficients. The traffic distribution data of each sub-interval in the first preset time period is predicted by the time series large model based on the difference coefficient sequence of the perceived data, and the situational awareness dataset corresponding to the first preset time period is output based on the traffic distribution data after weighted calculation. Specifically, the multi-source data of the monitoring equipment group in the road network space are weighted and fused according to the relevant weight set to obtain the road control level determination index corresponding to the road network space, including: Acquire multi-source data collected by the monitoring equipment group in the road network space during the first preset time period, and perform weighted processing on the data of corresponding multi-source data types in the multi-source data through the relevant weight set to obtain a weighted multi-source data vector; The weighted multi-source data vector is input into a multi-dimensional scoring model to determine the road control level judgment index based on the model output; wherein, the multi-dimensional scoring model is a pre-trained machine learning model used to calculate preset multi-dimensional sub-scorings.

2. The situational awareness traffic control method based on a large temporal model according to claim 1, characterized in that, The types of traffic entities include at least one or more of the following: motor vehicles, non-motor vehicles, and pedestrians; Determine the traffic flow curves corresponding to each type of traffic entity, specifically including: At preset time intervals, the number of traffic entities passing through the monitored road segment is statistically sequenced within the second preset time period, and the traffic flow sequence of each type of traffic entity is calculated per unit time based on the number of traffic entities passing through each entity. Each traffic entity's flow sequence is subjected to curve fitting to generate a flow curve for each traffic entity.

3. The situational awareness traffic control method based on a large temporal model according to claim 1, characterized in that, Based on historical situational awareness time-series data analysis, the changes in the perceived data of the situational awareness dataset are analyzed, and each associated road segment corresponding to the monitored road segment of the first monitoring device is matched to construct a road network space of corresponding association types, specifically including: Based on the historical situational awareness time-series data, a first situational awareness feature sequence corresponding to each road segment is determined; the first situational awareness feature sequence includes at least multiple traffic flow features of the corresponding road segment; the traffic flow feature is at least one of the following: mean traffic flow, standard deviation of traffic flow, and peak traffic flow. Based on the first situational awareness feature sequences and the situational awareness dataset of the same historical period of the first preset time period, the corresponding perception data association multivariate coefficient set is calculated; wherein, the perception data association multivariate coefficient set includes the flow similarity calculated from the flow distribution data of various traffic entities. Based on the various multivariate coefficient groups associated with the sensing data and the preset filtering conditions, the corresponding multivariate coefficient groups associated with the sensing data are filtered and the corresponding monitoring road segments are determined as the associated road segments; wherein, the preset filtering conditions are used to compare the multivariate coefficient threshold group with the multivariate coefficient groups associated with the sensing data, and filter a predetermined number of multivariate coefficient groups associated with the sensing data whose traffic similarity is greater than the corresponding multivariate coefficient threshold. Based on the traffic entity types that meet the preset screening conditions and the preset association type comparison table, the corresponding association type is determined, and the monitored road segment of the first monitoring device and its corresponding associated road segments are added to the road network space of the corresponding association type.

4. The situational awareness traffic control method based on a large temporal model according to claim 1, characterized in that, Based on the similar road association features corresponding to the road network space, a relevant weight set corresponding to the multi-source data type is matched, specifically including: According to a preset association type lookup table, the similar road association features corresponding to the association type in the road network space are determined; wherein, the preset association type lookup table contains several association types and the corresponding relationships of each similar road association feature obtained by pre-clustering; Based on the similar road association features and the preset association feature weight mapping relationship, the relevant weights of each of the multi-source data types are matched and the relevant weight set is constructed.

5. The situational awareness traffic control method based on a large temporal model according to claim 1, characterized in that, Based on the road control level determination index and the preset situational awareness control dataset, corresponding traffic control strategies are determined, and each traffic control strategy is sent to the traffic control management terminal, specifically including: The road control level determination index corresponding to each of the road network spaces is matched with the preset situational awareness control dataset to map the initial traffic control strategy corresponding to each road control level, and then sent to the expert system for verification. The initial traffic control strategy verified from the expert system is used as the traffic control strategy, and each of the traffic control strategies is sent to the traffic control management terminal.

6. The situational awareness traffic control method based on a large temporal model according to claim 1, characterized in that, The method further includes: Based on the continuous road control level determination index of each road network space within the first preset time period, generate change curves for each determination index respectively; Based on the change curves of each of the judgment indices and the location information of each of the road network spaces, it is determined whether the correlation between adjacent road network spaces has changed; the correlation is that the change curves of each of the judgment indices are negatively correlated or have no significant correlation. If so, starting from the current time, the road monitoring video stream collected by the first monitoring device is acquired in real time to update the constructed road network space.

7. A situational awareness traffic control system based on a time-series large model, characterized in that, The system is capable of executing a situational awareness traffic control method based on a large temporal model as described in any one of claims 1-6; the system includes: The input module is used to input the road monitoring video stream collected by the first monitoring device into a pre-trained time series large model, so as to determine the situational awareness dataset corresponding to the first preset time period based on the model output results; wherein, the situational awareness dataset includes at least traffic flow distribution data of multiple types of traffic entities; The first matching module is used to analyze the changes in the perception data of the situational awareness dataset based on historical situational awareness time-series data, and match each associated road segment corresponding to the monitoring road segment of the first monitoring device to construct a road network space of the corresponding association type. The second matching module is used to match the relevant weight set corresponding to the multi-source data type according to the similar road association features corresponding to the road network space, so as to perform weighted fusion of the multi-source data of the monitoring equipment group in the road network space according to the relevant weight set to obtain the road control level judgment index corresponding to the road network space. The sending module is used to determine the corresponding traffic control strategy based on the road control level judgment index and the preset situational awareness control dataset, and send the traffic control strategy to the traffic control management terminal.

8. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are capable of executing a situational awareness traffic control method based on a large temporal model as described in any one of claims 1-6.

Images