Highway congestion prediction method and system based on big data

Abstract

The present application provides a highway congestion prediction method and system based on big data. By accurately extracting highway road condition state vector data and combining it with prior pattern knowledge data for intelligent integration analysis, the accuracy of highway congestion prediction is significantly improved. It can not only generate the potential congestion trend characteristics of the target section in real time, but also further identify specific congestion risk patterns and automatically match the causes of similar congestion cases in historical data, so as to quickly locate the potential causes of current congestion. Compared with traditional prediction methods, the present invention not only has strong prediction timeliness, but also more detailed and in-depth prediction results, providing more scientific and accurate decision support for traffic management departments. By identifying and warning of potential traffic congestion in advance, it is helpful to take effective measures in a timely manner to alleviate traffic pressure and improve road traffic efficiency.

Description

Technical Field

[0001] The present application relates to the field of big data technology, and in particular to a highway congestion prediction method and system based on big data. Background Art

[0002] With the acceleration of urbanization and the continuous growth of car ownership, highways, as an important part of the modern transportation system, have a direct impact on the overall operation efficiency of the city and the quality of life of residents. However, highway congestion is becoming increasingly serious and has become one of the key factors restricting traffic flow. Traditional traffic management methods often rely on manual inspections and empirical judgments, which makes it difficult to achieve real-time and accurate predictions of highway congestion, resulting in a passive state when dealing with congestion.

[0003] In recent years, with the rapid development of big data technology, its application in the field of traffic management has become increasingly widespread. By collecting and analyzing massive amounts of traffic monitoring data, the inherent laws of traffic flow changes can be revealed, providing a scientific basis for congestion prediction. However, most of the existing highway congestion prediction methods have problems such as low prediction accuracy, limited prediction range, and poor prediction timeliness. This is mainly due to the shortcomings of these methods in data processing, feature extraction, pattern recognition, etc., and they are unable to fully utilize the advantages of big data and deeply explore the potential value of traffic flow data. Specifically, existing prediction methods often rely on a single traffic parameter (such as traffic flow, average speed, etc.) for prediction, ignoring the importance of diversified information such as road condition labels and historical congestion cases. Summary of the invention

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present application, an embodiment of the present application provides a highway congestion prediction method based on big data, the method comprising:

[0005] Extracting target road condition state vector data with set highway road condition labels from target traffic monitoring big data containing multiple highway road condition labels;

[0006] According to the priori pattern knowledge data of the set highway road condition label, the target road condition state vector data is integrated to generate target potential congestion trend characteristics, and the priori pattern knowledge data is used to reflect: the priori road condition state vector data extracted from the set highway road condition label by multiple priori traffic monitoring big data and the corresponding candidate congestion cause knowledge points;

[0007] Performing congestion risk pattern feature identification on the target potential congestion trend feature to generate a target congestion risk pattern feature, and selecting one or more candidate congestion risk pattern features that meet the target matching requirements from the priori congestion risk pattern features respectively corresponding to the multiple priori traffic monitoring big data for the target congestion risk pattern feature;

[0008] The candidate congestion cause knowledge points corresponding to the one or more candidate congestion risk pattern features are used as predicted congestion cause knowledge points of the target traffic monitoring big data.

[0009] On the other hand, an embodiment of the present application also provides a traffic monitoring system, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to run the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0010] Based on the above aspects, the embodiment of the present application significantly improves the accuracy of highway congestion prediction by accurately extracting highway road condition state vector data and combining it with prior pattern knowledge data for intelligent integration analysis. It can not only generate the potential congestion trend characteristics of the target section in real time, but also further identify specific congestion risk patterns and automatically match the causes of similar congestion cases in historical data, so as to quickly locate the potential causes of the current congestion. Compared with traditional prediction methods, the present invention not only has strong prediction timeliness, but also more detailed and in-depth prediction results, providing more scientific and accurate decision support for traffic management departments. By identifying and warning of potential traffic congestion in advance, it is helpful to take effective measures in a timely manner to alleviate traffic pressure and improve road traffic efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a schematic diagram of the execution flow of the highway congestion prediction method based on big data provided in an embodiment of the present application.

[0012] Figure 2 It is a schematic diagram of the hardware architecture of the traffic monitoring system provided in the embodiment of the present application. DETAILED DESCRIPTION

[0013] The present application will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of a highway congestion prediction method based on big data provided by an embodiment of the present application. The highway congestion prediction method based on big data is introduced in detail below.

[0014] Step S110 , extracting target road condition state vector data with set highway road condition labels from target traffic monitoring big data containing multiple highway road condition labels.

[0015] In detail, the target traffic monitoring big data refers to a collection of real-time or historical data from a highway traffic monitoring system, including but not limited to vehicle speed, flow, road occupancy, accident information, weather conditions, etc., which are usually collected by various sensors along the highway (such as cameras, speed detectors, weather stations, etc.) and transmitted to a data center through the network for storage and analysis.

[0016] The highway traffic label is used to describe the real-time or historical traffic status of a specific section of the highway, such as "partial congestion caused by an accident", "adverse effects on nighttime visibility", "lane change restrictions caused by road maintenance", etc. Specifically, each piece of traffic monitoring data can be assigned a corresponding traffic label by manual labeling or automatic identification by a machine learning algorithm.

[0017] The target road condition state vector data may be a multi-dimensional feature data set extracted from the target traffic monitoring big data for a set highway road condition label (such as "partial congestion caused by an accident"), reflecting the road condition state of the set highway road condition label. For example, firstly, the data stream containing the set road condition label is screened out, then the data stream is divided according to the time trajectory window, the road condition parameters (such as vehicle flow, vehicle speed, road occupancy, etc.) in each window are extracted, and finally the road condition state vector data is generated according to these parameters.

[0018] That is, in this embodiment, it is assumed that the server receives target traffic monitoring big data from a certain highway section within a week, and these target traffic monitoring big data contain multiple complex highway road condition labels such as "local congestion caused by accidents", "adverse effects on nighttime visibility", and "lane change restrictions caused by road maintenance".

[0019] First, the server obtains the target traffic monitoring data stream containing the set highway road condition label (such as "partial congestion caused by an accident") from this massive target traffic monitoring big data. These data streams may include vehicle speed, flow, road occupancy, location information of the accident site, accident handling progress and other information.

[0020] Next, the server splits the target traffic monitoring data stream into multiple time trajectory windows. For example, a week's data stream is divided into a time trajectory window of 30 minutes. For each time trajectory window, the server obtains its corresponding road condition parameter set. For example, in a 30-minute time trajectory window, the average vehicle speed is 20 kilometers per hour, the traffic volume is 80 vehicles per minute, the road occupancy rate is 70%, the accident site is located at 5 kilometers of the road section, and the accident handling progress is 30%. These data constitute a road condition parameter set. Each road condition parameter reflects a road condition indicator in the corresponding time trajectory window.

[0021] Then, based on the sequential relationship of these time trajectory windows, the server processes multiple road condition parameter sets in sequence. Assuming that the road condition parameter set of the first time trajectory window is A and the second is B, the server will comprehensively consider the relationship between A and B, such as comparing the vehicle speed changes, traffic changes, and the progress of accident handling in the two windows, so as to generate the basic road condition parameter set of the target traffic monitoring data stream.

[0022] Finally, the server extracts the road condition state vector from the basic road condition parameter set based on the degree of association of each road condition parameter with the set highway road condition label ("partial congestion caused by an accident"). For example, if the vehicle speed is lower than a certain threshold (such as 40 km / h), the road occupancy rate is higher than a certain value (such as 50%), and the accident handling progress has not reached 100%, it is considered that the road condition parameter has a high degree of association with the "partial congestion caused by an accident" state. In this way, the server generates the target road condition state vector data of the set highway road condition label ("partial congestion caused by an accident") in the target traffic monitoring big data.

[0023] Step S120, based on the prior pattern knowledge data of the set highway road condition label, the target road condition state vector data is integrated to generate target potential congestion trend characteristics, and the prior pattern knowledge data is used to reflect: the prior road condition state vector data extracted from multiple prior traffic monitoring big data for the set highway road condition label and the corresponding candidate congestion cause knowledge points.

[0024] In detail, the prior pattern knowledge data refers to the prior road condition state vector data about specific road condition labels (such as "partial congestion caused by an accident") extracted from historical traffic monitoring big data, and the mapping relationship between these data and the corresponding congestion cause knowledge points. For example, an association model between road condition state vector data and congestion causes can be established by analyzing historical data. These models can be based on statistics, rules, or machine learning.

[0025] Fusion can refer to the process of comprehensively analyzing and matching the target traffic state vector data with the prior pattern knowledge data to generate the target potential congestion trend characteristics. For example, the mapping relationship in the prior pattern knowledge data can be used to perform feature mapping and interaction on the target traffic state vector data to predict the potential congestion trend.

[0026] The target potential congestion trend feature is a feature set generated by integrating the target road condition state vector data and the priori pattern knowledge data, and reflects the potential congestion trend of the expressway in the future.

[0027] In this embodiment, the server has already acquired the target road condition state vector data, and then processes it according to the prior pattern knowledge data.

[0028] Assume that the server stores multiple prior traffic monitoring big data, extracted from the prior road condition state vector data for the set highway road condition label ("partial congestion caused by the accident"), as well as the mapping relationship between these prior data and the corresponding candidate congestion cause knowledge points. For example, some prior road condition state vector data show that the traffic volume of a specific road section suddenly increases, the speed drops sharply, and the accident handling time is long in a specific time period. The corresponding candidate congestion cause knowledge point may be that the accident vehicle occupies multiple lanes and rescue is difficult; while other prior data show that the traffic volume does not change much at the beginning of the accident, but the speed gradually decreases. The corresponding candidate congestion cause knowledge point may be that the warning signs at the accident site are not set up in time.

[0029] The server obtains the prior pattern knowledge data with the set knowledge dimension space, and then performs feature mapping interaction on the knowledge dimension space of the target road condition state vector data based on the prior pattern knowledge data. For example, the features such as vehicle flow, vehicle speed, and accident handling progress in the target road condition state vector data are compared and integrated with similar features in the prior pattern knowledge data to generate target potential congestion trend features that are adapted to the set knowledge dimension space. This feature can reflect the possible development trend of the target road condition state when the accident is not completely handled, such as whether it will gradually ease as the accident is handled or further deteriorate due to improper handling.

[0030] Step S130, performing congestion risk pattern feature identification on the target potential congestion trend feature, generating a target congestion risk pattern feature, and selecting one or more candidate congestion risk pattern features that meet the target matching requirements from the prior congestion risk pattern features corresponding to the multiple prior traffic monitoring big data for the target congestion risk pattern feature.

[0031] The congestion risk pattern feature recognition further analyzes the target potential congestion trend features to identify the congestion risk pattern contained therein (such as the probability of congestion, duration, impact range, etc.). The candidate congestion cause knowledge points are a set of cause knowledge points associated with a specific congestion risk pattern that may cause congestion, such as "accident vehicles occupy multiple lanes and rescue is difficult", "road construction causes lanes to narrow", etc.

[0032] In this embodiment, the server first performs congestion risk pattern feature recognition on the generated target potential congestion trend feature.

[0033] Assume that the server obtains a priori potential congestion trend features corresponding to multiple prior traffic monitoring big data, and a priori congestion risk pattern features corresponding to each priori potential congestion trend feature. For the target potential congestion trend feature, the server determines the priori congestion risk pattern features corresponding to the priori potential congestion trend features that meet the set risk trend matching requirements from these prior potential congestion trend features as the target congestion risk pattern features.

[0034] For example, the risk trend matching requirement is that the traffic flow growth rate, speed reduction rate and accident handling progress reach a certain threshold within a specific time period after the accident. If the target potential congestion trend feature shows that the traffic flow increases by 40% within 1 hour, the speed decreases by 50% and the accident handling progress is only 20%, and there are multiple similar situations in the prior potential congestion trend feature, and the corresponding prior congestion risk pattern feature is "long-term severe congestion caused by major accidents", then the server will select these prior congestion risk pattern features as candidate congestion risk pattern features that meet the target matching requirements.

[0035] Step S140: taking the candidate congestion cause knowledge points corresponding to the one or more candidate congestion risk pattern features as the predicted congestion cause knowledge points of the target traffic monitoring big data.

[0036] In this embodiment, the server has selected one or more candidate congestion risk pattern features.

[0037] Assume that the candidate congestion cause knowledge points corresponding to these candidate congestion risk pattern features are "major accident and slow rescue", "difficult to clean up the accident site", "accident triggers a chain reaction", etc. The server uses these candidate congestion cause knowledge points as the predicted congestion cause knowledge points for the current target traffic monitoring big data.

[0038] For example, if the selected candidate congestion risk pattern characteristics are mainly related to major accidents and slow rescue, the server will predict that the cause of congestion reflected by the current target traffic monitoring big data is likely to be a major accident and slow rescue, thereby providing decision-making support for the traffic management department so that it can take corresponding guidance and handling measures in a timely manner, such as increasing rescue forces, adjusting traffic lights, and issuing road conditions information to guide vehicles to detour.

[0039] In this way, valuable information can be extracted from complex target traffic monitoring big data, possible causes of congestion can be predicted, and strong support can be provided for improving traffic conditions on highways.

[0040] Based on the above steps, the embodiment of the present application significantly improves the accuracy of highway congestion prediction by accurately extracting highway road condition state vector data and combining it with prior pattern knowledge data for intelligent integration analysis. It can not only generate the potential congestion trend characteristics of the target section in real time, but also further identify specific congestion risk patterns and automatically match the causes of similar congestion cases in historical data, so as to quickly locate the potential causes of the current congestion. Compared with traditional prediction methods, the present invention not only has strong prediction timeliness, but also more detailed and in-depth prediction results, providing more scientific and accurate decision support for traffic management departments. By identifying and warning of potential traffic congestion in advance, it is helpful to take effective measures in a timely manner to alleviate traffic pressure and improve road traffic efficiency.

[0041] In a possible implementation, step S110 includes:

[0042] Step S111, obtaining a target traffic monitoring data stream including a set highway road condition tag from target traffic monitoring big data including multiple highway road condition tags.

[0043] Step S112, splitting the target traffic monitoring data stream into multiple time trajectory windows, and obtaining road condition parameter sets corresponding to the multiple time trajectory windows respectively, wherein a road condition parameter in each road condition parameter set is used to reflect: a road condition indicator in the corresponding time trajectory window.

[0044] Step S113: Based on the temporal relationship of the multiple time trajectory windows, multiple road condition parameter sets are processed in sequence to generate a basic road condition parameter set of the target traffic monitoring data stream.

[0045] Step S114, extracting the road condition state vector of the basic road condition parameter set according to the association degree of each road condition parameter to the set highway road condition label, and generating target road condition state vector data of the set highway road condition label in the target traffic monitoring big data.

[0046] In this embodiment, the server receives target traffic monitoring big data of a certain highway section within one month. These target traffic monitoring big data cover rich, diverse and complex highway road condition labels such as "local congestion caused by accidents", "adverse impact on night visibility", "lane change restrictions caused by road maintenance", and "slow traffic caused by bad weather".

[0047] First, the server obtains the target traffic monitoring data stream containing the set highway road condition labels (such as "partial congestion caused by accidents") from this huge and complex target traffic monitoring big data. In this process, the server will accurately extract the data based on the pre-set screening rules and keywords. The obtained target traffic monitoring data stream contains multi-dimensional and detailed data such as the real-time speed, flow, road occupancy rate of the vehicle, the specific location and time of the accident, the real-time progress of the accident handling, and the surrounding environment information (such as weather conditions).

[0048] Next, the server splits the target traffic monitoring data stream into multiple time trajectory windows. Assume that the server divides the data stream for one month into one time trajectory window of 1 hour. For each 1-hour time trajectory window, the server will carefully obtain its corresponding set of road condition parameters. For example, in a specific 1-hour time trajectory window, the average vehicle speed is 15 kilometers per hour, the traffic volume is 50 vehicles per minute, the road occupancy rate is 65%, the accident site is located at 10 kilometers of the road section, the accident handling progress is 20%, and the weather at the time is light rain. These rich data together constitute the road condition parameter set of this time trajectory window. Each road condition parameter accurately reflects a key road condition indicator in the corresponding time trajectory window.

[0049] Then, based on the sequential relationship of these time trajectory windows, the server processes multiple road condition parameter sets in sequence. Assume that the road condition parameter set of the first time trajectory window is A, the second is B, the third is C, and so on. The server will comprehensively consider the internal connections and change trends between these sets. For example, the server will compare the changes in vehicle speeds in windows A and B. If the average vehicle speed in window A is 20 km / h, and it drops to 15 km / h in window B, the server will further analyze the changes in other parameters such as traffic flow and road occupancy to determine the degree of deterioration and development trend of road conditions. At the same time, the server will also pay attention to the progress of accident handling, such as from 10% in window A to 20% in window B. Through such comprehensive and in-depth sequential processing, the server can generate a basic road condition parameter set for the target traffic monitoring data stream. This basic road condition parameter set not only contains the independent information of each time trajectory window, but also integrates the dynamic changes and relationships between them.

[0050] Finally, the server extracts the road condition state vector of the basic road condition parameter set based on the correlation of each road condition parameter to the set highway road condition label ("partial congestion caused by accidents"). For example, if the vehicle speed is continuously lower than 30 km / h, the road occupancy rate is higher than 50% for a long time, and the accident handling progress does not reach more than 80%, the server will determine that these road condition parameters have a high correlation with the "partial congestion caused by accidents" state. By using complex algorithms and models, the server conducts in-depth mining and detailed analysis of the basic road condition parameter set, thereby successfully generating the target road condition state vector data of the set highway road condition label ("partial congestion caused by accidents") in the target traffic monitoring big data. These vector data represent the key road condition characteristics and trends related to "partial congestion caused by accidents" in a highly concentrated and precise form, providing a solid data foundation for subsequent in-depth analysis and prediction.

[0051] In another specific scenario, the target traffic monitoring big data processed by the server comes from highways during holidays. The highway traffic label set is "lane change restriction caused by road maintenance."

[0052] The server first accurately obtains the target traffic monitoring data stream containing this specific road condition label from the massive data. This data stream covers detailed information such as the instantaneous speed of vehicles, traffic distribution in different lanes, the specific location and scope of road maintenance areas, the status of lane change indication signals, and construction progress.

[0053] Next, the server divides the data stream into time trajectory windows of 45 minutes each. In each time trajectory window, for example, one window shows that the average vehicle speed is 40 kilometers per hour, the flow rate of a specific lane (the lane where the maintenance area is located) is 30 vehicles per minute, the road occupancy rate is 40%, the maintenance area is located at 15 kilometers of the road section, and the construction progress is 50%. These detailed data constitute the road condition parameter set of the window.

[0054] Based on the order of these time trajectory windows, the server processes them one by one. The server finds that over time, although the average vehicle speed does not change much, the flow of a specific lane gradually decreases, while the flow of adjacent lanes increases, and the construction progress is steadily advancing.

[0055] Finally, based on the correlation between each road condition parameter and the label "lane change restrictions due to road maintenance", the server extracts the road condition state vector. If the flow rate of a specific lane is lower than a certain percentage of the normal level, the road occupancy rate fluctuates within a specific range, and the construction progress is not completed, the server determines that these parameters are highly correlated with the set road condition label. Through such detailed processing and analysis, the server successfully extracted the target road condition state vector data that accurately reflects the "lane change restrictions due to road maintenance" situation.

[0056] To give another example, suppose the server is facing a target traffic monitoring big data of a highway that is often affected by bad weather in winter, and the set road condition label is "slow traffic caused by bad weather."

[0057] The server first obtains the target traffic monitoring data stream containing this road condition label from the complex data, which includes key information such as real-time vehicle speed, flow, road occupancy, snowfall, visibility, road icing conditions, etc.

[0058] Then, the data stream is divided into a time trajectory window of every 2 hours. In a certain time trajectory window, the average vehicle speed may be only 25 kilometers per hour, the traffic volume may be 60 vehicles per minute, the road occupancy rate may be 55%, the snowfall may reach 5 centimeters per hour, the visibility may be less than 200 meters, and some roads may have thin ice, etc., which constitute the road condition parameter set of the window.

[0059] Based on the chronological order of the time trajectory windows, the server found that as the snowfall increased and visibility decreased, vehicle speed and flow were gradually decreasing.

[0060] Finally, based on the correlation between each road condition parameter and the label "slow traffic caused by bad weather", for example, when the vehicle speed is lower than 50% of the normal speed, the traffic volume decreases by more than 30%, the snowfall continues to increase and the visibility is lower than a certain value, the server extracts the target road condition state vector data that can accurately depict the slow traffic conditions under the influence of this bad weather.

[0061] In this way, the target road condition state vector data with set highway road condition labels can be effectively extracted from the complex target traffic monitoring big data containing multiple highway road condition labels, providing key and accurate basic information for subsequent traffic condition analysis and prediction.

[0062] In a possible implementation manner, the step of acquiring the priori pattern knowledge data includes:

[0063] Step A110, obtaining the priori road state vector data and the corresponding candidate congestion cause knowledge points extracted from the plurality of priori traffic monitoring big data for the road state vector data, and obtaining the preliminarily constructed basic priori pattern knowledge data.

[0064] Step A120, performing optimization adjustment of the basic priori pattern knowledge data in multiple cycles according to each priori road condition state vector data and the corresponding candidate congestion cause knowledge point, generating optimized and adjusted priori pattern knowledge data, in one cycle:

[0065] Step A130, based on the basic priori pattern knowledge data corresponding to this cycle process, a priori road condition state vector data is integrated to generate a priori potential congestion trend characteristics.

[0066] Step A140, based on the deviation between the prior potential congestion trend characteristics and the reference congestion trend characteristics of the corresponding candidate congestion cause knowledge points, optimize and adjust the basic prior pattern knowledge data corresponding to the current cycle process, generate optimized and adjusted prior pattern knowledge data, and use it as the basic prior pattern knowledge data corresponding to the next cycle process.

[0067] In this embodiment, the server stores a large amount of prior traffic monitoring big data, which comes from highways in different time periods and different sections.

[0068] First, the server obtains the prior road state vector data and the corresponding candidate congestion cause knowledge points extracted from multiple prior traffic monitoring big data for the road state vector data. For example, for a set of prior data, its prior road state vector data shows that in a specific road section, the traffic volume suddenly increases and the speed drops sharply and lasts for a long time in a certain period of time. The corresponding candidate congestion cause knowledge point is that a multi-vehicle rear-end collision occurred in the road section, and the accident vehicles occupied multiple lanes. At the same time, the server also obtains the preliminary constructed basic prior pattern knowledge data, which contains some initial general association rules about road conditions and congestion causes.

[0069] Then, the server starts to optimize and adjust multiple cycles. In the first cycle, the server blends a priori road condition state vector data based on the basic prior pattern knowledge data corresponding to this cycle. Assuming that this priori road condition state vector data indicates that on a certain road section, the vehicle speed drops from 80 km / h to 20 km / h in a short period of time, and the traffic volume increases from 120 vehicles / minute to 180 vehicles / minute, the server comprehensively analyzes and processes this information based on the current basic prior pattern knowledge data to generate a priori potential congestion trend characteristics, which may show that congestion has a tendency to further intensify.

[0070] Next, the server calculates the deviation between the prior potential congestion trend feature and the reference congestion trend feature of the corresponding candidate congestion cause knowledge point. Assume that the corresponding reference congestion trend feature indicates that this situation is usually caused by a large truck failure, which leads to road congestion, and the currently generated prior potential congestion trend feature has a certain deviation from it. Based on this deviation, the server optimizes and adjusts the basic prior pattern knowledge data corresponding to this cycle. For example, increase the weight of the speed reduction amplitude and the traffic volume increase amplitude in judging the cause of congestion, generate optimized and adjusted prior pattern knowledge data, and use it as the basic prior pattern knowledge data corresponding to the next cycle.

[0071] In the second cycle, the server again blends another prior road condition state vector data based on the new basic prior pattern knowledge data. For example, if this data shows that on another road section, the traffic volume increases slowly and the speed gradually decreases, but the rate of decrease is relatively slow, the server generates the corresponding prior potential congestion trend characteristics. Then, the deviation from the reference congestion trend characteristics of the corresponding candidate congestion cause knowledge point is calculated again to further optimize and adjust the basic prior pattern knowledge data.

[0072] This cycle is carried out continuously, and the server continuously processes various different prior road state vector data, compares and adjusts them with the reference congestion trend characteristics of the corresponding candidate congestion cause knowledge points, and gradually optimizes the prior pattern knowledge data so that it can more accurately reflect the relationship between the road state vector data and the congestion cause.

[0073] For example, in a certain cycle, the prior road condition state vector data shows that on a road section during a holiday, the traffic volume has been high since the morning, and the vehicle speed is relatively low but has not dropped sharply. The server generates a priori potential congestion trend characteristics based on the basic prior pattern knowledge data at that time, and finds that it is close to the reference congestion trend characteristics of the candidate congestion cause knowledge point (that is, the road section is congested due to the peak of tourist travel), but there are still some slight deviations. Based on this, the server fine-tunes the basic prior pattern knowledge data, such as adjusting the weight of traffic volume and speed during holidays in judging the cause of congestion.

[0074] Through multiple such cycles and optimizations, the priori pattern knowledge data is continuously improved and precise, and can better provide strong support for subsequent road condition analysis and congestion cause prediction.

[0075] In a possible implementation, step A130 includes:

[0076] Step A131, performs pattern correlation analysis on the basic prior pattern knowledge data corresponding to the current cycle process and the basic prior pattern knowledge data corresponding to each previous cycle process, and performs state correlation analysis on the prior pattern knowledge data corresponding to the current cycle process and a prior road condition state vector data to generate candidate prior pattern knowledge data.

[0077] Step A132: blending the one priori road condition state vector data according to the candidate priori pattern knowledge data to generate a priori potential congestion trend feature of the one priori road condition state vector data.

[0078] In this embodiment, the server has performed multiple cycles of optimization when processing the prior road condition data. For example, in the current cycle, the server first obtains the basic prior pattern knowledge data corresponding to this cycle. It is assumed that this part of the data contains some rules and weights about the relationship between factors such as traffic flow changes, vehicle speed reduction speed, and road construction progress and potential congestion.

[0079] At the same time, the server reviews the basic prior pattern knowledge data corresponding to each past cycle process. For example, in the previous cycle, the focus was on the impact of accident handling time on congestion, and the cycle before that emphasized the relationship between weather conditions and congestion. The server performs pattern correlation analysis on these past data to find out the common patterns and difference patterns. For example, it was found that the rapid growth of traffic volume was strongly correlated with severe congestion in multiple cycles.

[0080] Next, the server performs state correlation analysis on the prior pattern knowledge data corresponding to this cycle process and a specific prior road condition state vector data. Assume that this prior road condition state vector data shows that on a certain road section, the traffic volume increased by 50% in a short period of time, the vehicle speed decreased by 30%, and there was road construction nearby but the progress was slow. Through analysis and processing, the server generates candidate prior pattern knowledge data, which may further clarify the importance of traffic volume and speed changes to congestion in this case, as well as the superimposed impact of the slow progress of road construction.

[0081] Then, based on the generated candidate prior pattern knowledge data, the server blends the prior road state vector data. In the blending process, the weights and relationships of various factors are comprehensively considered to generate the prior potential congestion trend characteristics of the prior road state vector data. For example, it is concluded that congestion may continue to worsen in the next period of time, and the impact range may expand.

[0082] For example, in another cycle, the basic prior pattern knowledge data of this cycle focuses on analyzing the characteristics of holiday travel peaks. In the past, some cycles focused on special weather conditions, while others focused on traffic changes caused by large-scale events. For a new prior road condition state vector data, it shows that during the holidays, the traffic volume on a certain road section has steadily increased, the speed has gradually decreased, and there are temporary traffic controls around it. The server performs the above analysis and processing process to generate a priori potential congestion trend characteristics that reflect the possibility of moderate congestion for a long time.

[0083] In this way, new data and past experience can be continuously utilized to more accurately generate prior characteristics of potential congestion trends, providing valuable information for subsequent traffic condition analysis and prediction.

[0084] In a possible implementation, step A140 includes:

[0085] Step A141, calculating a first deviation between the a priori potential congestion trend feature and a reference congestion trend feature of a candidate congestion cause knowledge point of corresponding a priori road condition state vector data.

[0086] Step A142, acquiring other priori road condition state vector data except the corresponding priori road condition state vector data from the acquired multiple priori road condition state vector data.

[0087] Step A143, respectively calculating the second deviation between the a priori potential congestion trend feature and the reference congestion trend feature of the candidate congestion cause knowledge point corresponding to each other a priori road condition state vector data.

[0088] Step A144, based on the excitation adjustment effect of the first deviation on the basic prior pattern knowledge data corresponding to the current cycle process, and based on the inhibitory adjustment effect of each second deviation on the basic prior pattern knowledge data corresponding to the current cycle process, optimize and adjust the basic prior pattern knowledge data corresponding to the current cycle process to generate optimized and adjusted prior pattern knowledge data.

[0089] In this embodiment, the server has generated a priori potential congestion trend characteristics and is ready to optimize and adjust the basic a priori pattern knowledge data corresponding to this cycle process.

[0090] For example, for a specific prior road state vector data, its prior potential congestion trend characteristics indicate that the traffic volume on a certain road section increased by 80% within an hour and the vehicle speed decreased by 40%. It is preliminarily judged that congestion may rapidly increase. The reference congestion trend characteristics of the candidate congestion cause knowledge points corresponding to the prior road state vector data indicate that this situation is usually caused by a major traffic accident.

[0091] The server first calculates the first deviation between the prior potential congestion trend feature and the reference congestion trend feature of the candidate congestion cause knowledge point of the corresponding prior road condition state vector data. It is assumed that there is a certain deviation between the two calculated by a specific algorithm and indicator, such as a difference in the expected speed and degree of congestion aggravation.

[0092] Then, the server obtains other priori road condition state vector data except the corresponding priori road condition state vector data from the obtained multiple priori road condition state vector data, for example, obtaining several other priori road condition state vector data, respectively reflecting different congestion conditions caused by road construction, bad weather, etc.

[0093] Next, the server calculates the second deviation between the current a priori potential congestion trend feature and the reference congestion trend features of the candidate congestion cause knowledge points corresponding to other prior road condition state vector data. For example, it is found that compared with a reference congestion trend feature of congestion caused by road construction, the current a priori potential congestion trend feature has a large difference in traffic flow change and speed reduction mode.

[0094] Based on the incentive adjustment effect of the first deviation on the basic prior pattern knowledge data corresponding to this cycle, the server realizes that the current judgment may need to be further optimized to be closer to the actual candidate congestion cause knowledge point. At the same time, based on the inhibitory adjustment effect of each second deviation on the basic prior pattern knowledge data corresponding to this cycle, it avoids overfitting the current data and deviating from other common congestion patterns.

[0095] For example, if the first deviation is large, the server may increase the weight of traffic flow and speed changes in determining the cause of congestion. If a second deviation is large, the server may appropriately reduce the weight of certain specific factors to make the basic prior model knowledge data more accurate and universal.

[0096] Through such comprehensive adjustments, the server optimizes and adjusts the basic prior pattern knowledge data corresponding to this cycle process, and generates optimized and adjusted prior pattern knowledge data, so that the causes and trends of congestion can be more accurately determined in subsequent analysis.

[0097] For example, in another scenario, the prior potential congestion trend characteristics show that the traffic volume on a certain road section is slowly increasing and the speed is slowly decreasing. The reference congestion trend characteristics of the candidate congestion cause knowledge point believe that it is caused by a minor traffic accident. After calculating the first deviation, it is found that the two have different expectations for congestion relief time. Obtain other prior road condition state vector data, such as congestion caused by holiday travel. Calculate the second deviation and find that the traffic volume growth law is different from the reference congestion trend characteristics of holiday travel. Based on these deviations, the server performs optimization operations on the basic prior pattern knowledge data, such as adjusting the weight of the accident impact time and balancing the impact of holiday travel factors, to generate more reasonable prior pattern knowledge data.

[0098] By repeating this process, the server can continuously improve the prior pattern knowledge data and enhance its ability to analyze and predict highway congestion.

[0099] In a possible implementation, step S120 includes:

[0100] Step S121, obtaining prior pattern knowledge data having a set knowledge dimension space.

[0101] Step S122, based on the priori pattern knowledge data, feature mapping interaction is performed on the knowledge dimension space of the target road condition state vector data to generate target potential congestion trend features adapted to the set knowledge dimension space.

[0102] In this embodiment, when processing highway traffic data, the server first needs to blend the target traffic state vector data based on the priori pattern knowledge data for setting highway traffic labels to generate target potential congestion trend characteristics.

[0103] For example, the server receives the target road condition state vector data of a section of highway in a specific time period. These data cover detailed information in multiple dimensions such as vehicle speed, traffic volume, road occupancy, weather conditions, accident information, etc.

[0104] The server first obtains the prior knowledge data of the knowledge dimension space. It is assumed that this knowledge dimension space includes time dimension (such as traffic flow patterns in different time periods), space dimension (such as frequent congestion points on a specific road section), vehicle type dimension (such as the impact of large trucks on traffic), weather dimension (such as the impact of rainy and snowy weather on vehicle speed), and other aspects.

[0105] After obtaining the prior pattern knowledge data, the server performs feature mapping interaction on the knowledge dimension space of the target road condition state vector data based on this data. For example, the prior pattern knowledge data shows that during the morning and evening rush hours on weekdays, the traffic flow on a certain road section usually increases sharply due to the presence of large office buildings and residential areas nearby, which can easily lead to congestion. The current target road condition state vector data shows that it is the morning rush hour on weekdays, and vehicles are traveling on this specific road section, with traffic volume gradually increasing and vehicle speeds starting to decrease. The server maps and interacts the traffic rules for this time period and road section in the prior pattern knowledge with the current target road condition state vector data.

[0106] For example, the prior knowledge data of the model records that on rainy days, the friction coefficient of the road will decrease and the speed of vehicles will generally slow down. The current target road condition state vector data contains the weather information of raining and the situation that the speed of vehicles has decreased. The server further analyzes the traffic conditions through the correspondence and interaction of such features.

[0107] For another example, the prior pattern knowledge data indicates that when a traffic accident occurs and involves a large vehicle, the degree and duration of road congestion are often long. However, the target road condition state vector data indicates that an accident involving a large truck occurred on the current road section, and traffic volume accumulated rapidly. The server correlates and analyzes the characteristics of the two.

[0108] Through such comprehensive and detailed feature mapping interaction, the server generates target potential congestion trend features that are adapted to the set knowledge dimension space.

[0109] Assume that in another scenario, the target road condition state vector data obtained by the server shows that it is currently a holiday period, the traffic volume on a certain section of the highway continues to be high, the speed of vehicles is slowly decreasing, and road maintenance construction is being carried out on the section.

[0110] The prior knowledge data of patterns acquired by the server includes information such as the travel patterns of people during holidays and the impact of road maintenance and construction on traffic. For example, prior knowledge shows that during holidays, this section of road is usually a popular route for tourists and has a large traffic volume; at the same time, road maintenance and construction will occupy some lanes, further reducing the road's traffic capacity.

[0111] The server maps and interacts the features of the target road condition state vector data, such as traffic volume, speed changes, and construction information, with the features of holiday travel patterns and construction impact in the prior pattern knowledge data. It finds that the current traffic growth rate and construction occupation match the characteristics of severe congestion in the prior knowledge.

[0112] Furthermore, the server also takes weather factors into consideration. If the prior pattern knowledge data mentions that the probability of vehicle failure increases in hot weather, and the current target road condition state vector data contains high temperature weather information, the server will also include this feature in the interactive analysis.

[0113] Finally, through the comprehensive mapping interaction of these multi-dimensional features, the server generates features that can accurately reflect the potential congestion trend of the current road conditions. This may indicate that congestion will continue or even intensify in the next period of time, and the scope of impact may expand, requiring timely traffic diversion measures.

[0114] Let's look at another example. The target road condition state vector data received by the server shows that on a section of highway at night, the traffic volume suddenly increased, the vehicle speed dropped significantly, and there was a large logistics park near the section.

[0115] The prior pattern knowledge data records the time periods and patterns of concentrated travel of logistics vehicles at night, as well as the traffic characteristics of roads around large logistics parks. The server maps and interacts the traffic flow and speed change characteristics in the target road condition state vector data with the travel patterns of logistics vehicles and the characteristics of roads around the park in the prior knowledge.

[0116] The server found that the time when the current traffic volume suddenly increased coincided with the time when logistics vehicles traveled in large numbers, and the extent of the speed reduction also matched the impact of heavy-loaded logistics vehicles on road traffic. Through this interactive analysis, a target potential congestion trend feature reflecting the possible occurrence of periodic congestion was generated.

[0117] Therefore, the embodiment of the present application performs feature mapping interaction on the knowledge dimension space of the target road condition state vector data based on the prior pattern knowledge data, thereby generating features that can accurately reflect the potential congestion trend of the road condition, providing strong support for subsequent traffic management and decision-making.

[0118] In a possible implementation, step S130 includes:

[0119] Step S131, obtaining the a priori potential congestion trend features corresponding to the multiple a priori traffic monitoring big data respectively, each a priori potential congestion trend feature is generated by blending the a priori road condition state vector data of the corresponding a priori traffic monitoring big data based on the a priori pattern knowledge data.

[0120] Step S132, obtaining a priori congestion risk pattern features corresponding to each a priori potential congestion trend feature, wherein each a priori congestion risk pattern feature is generated after performing congestion risk pattern feature recognition on the corresponding a priori potential congestion trend feature.

[0121] Step S133, for the target potential congestion trend feature, determine the priori congestion risk pattern feature corresponding to the priori potential congestion trend feature that meets the set risk trend matching requirement from the multiple acquired priori potential congestion trend features, as the target congestion risk pattern feature.

[0122] In this embodiment, when processing traffic monitoring data, the server needs to perform congestion risk pattern feature recognition on target potential congestion trend features to generate target congestion risk pattern features.

[0123] For example, the server has acquired multiple prior traffic monitoring big data. For one of the prior traffic monitoring big data, the server integrates its prior road condition state vector data based on the prior pattern knowledge data to generate a prior potential congestion trend feature. Assume that this prior potential congestion trend feature indicates that on a certain road section, due to the holiday travel peak, the traffic volume increases significantly in a short period of time and the vehicle speed decreases significantly.

[0124] Then, the server identifies the congestion risk pattern characteristics of this prior potential congestion trend feature and generates the corresponding prior congestion risk pattern characteristics. For example, it determines that long-term severe congestion may occur and the impact will affect multiple surrounding road sections.

[0125] The server processes other multiple prior traffic monitoring big data in the same way and obtains a series of prior potential congestion trend characteristics and their corresponding prior congestion risk pattern characteristics.

[0126] Next, the server obtains the current target potential congestion trend feature. Assume that this target potential congestion trend feature is displayed on another road section, and due to holidays, the traffic volume increases rapidly and the speed begins to decrease.

[0127] The server compares the target potential congestion trend feature with multiple prior potential congestion trend features obtained previously. The risk trend matching requirements set include the similarity of factors such as the growth rate of traffic flow, the speed reduction range, and holidays.

[0128] After detailed comparison and analysis, the server found that one of the prior potential congestion trend features is very similar to the current target potential congestion trend feature. The prior congestion risk pattern feature corresponding to this prior potential congestion trend feature is "moderate congestion caused by holidays, expected to last for 3 hours".

[0129] Therefore, the server determines this prior congestion risk pattern feature as the target congestion risk pattern feature as the congestion risk assessment result for the current target traffic monitoring data.

[0130] For example, in another scenario, one of the prior potential congestion trend features obtained by the server shows that on a road section where road construction is often carried out, the vehicle speed slowly decreases due to unfinished construction and heavy traffic. After the congestion risk pattern feature recognition, the generated prior congestion risk pattern feature is "mild congestion during construction, expected to return to normal after the construction is completed."

[0131] When the server obtains a new target potential congestion trend feature, it indicates that the traffic volume and speed changes on another road section under construction are similar to the above-mentioned prior potential congestion trend feature. After comparing according to the set matching requirements, the server determines that the prior congestion risk pattern feature is the target congestion risk pattern feature.

[0132] In this way, it is possible to find congestion risk pattern characteristics that match the current target potential congestion trend characteristics from a large number of prior data, providing an accurate basis for traffic management and decision-making.

[0133] In a possible implementation manner, the step of generating the priori congestion risk pattern feature corresponding to each priori potential congestion trend feature includes:

[0134] Step B110 , tracking risk state nodes for a priori potential congestion trend characteristics, generating a risk state node set, each risk state node in the risk state node set is used to reflect: a state vector distribution of a risk characterization attribute.

[0135] Step B120, convert each risk state node in the risk state node set into the risk dimension domain corresponding to the corresponding risk representation attribute, generate a converted risk state node corresponding to each risk state node, and each converted risk state node is used to reflect: the state vector conversion parameter of the corresponding risk representation attribute.

[0136] Step B130: generating a priori congestion risk pattern feature of the one priori potential congestion trend feature based on each generated conversion risk state node.

[0137] In this embodiment, when the server processes the a priori potential congestion trend feature to generate the a priori congestion risk pattern feature, the specific operation process is as follows.

[0138] For example, the server obtains a priori potential congestion trend characteristics, indicating that on a certain highway section, due to sudden bad weather, the traffic volume gradually increases and the speed continues to decrease.

[0139] First, the server tracks the risk state nodes based on this prior potential congestion trend feature. Assume that the server has determined the following risk state nodes: the growth rate node of traffic flow, the decrease amplitude node of vehicle speed, the duration node of bad weather, etc. These risk state nodes together constitute a risk state node set, and each node reflects a state vector distribution of a risk characterization attribute. For example, the growth rate node of traffic flow may reflect the specific quantity distribution of traffic flow increase per hour.

[0140] Next, the server converts each risk state node in the risk state node set to the risk dimension domain corresponding to the corresponding risk characterization attribute. Taking the traffic growth rate node as an example, the server converts it to the dimension domain of traffic growth risk and generates the corresponding conversion risk state node. This converted node is used to reflect the state vector conversion parameter of traffic growth in risk assessment, such as a value indicating the degree of risk. Similarly, the speed reduction node is converted to the dimension domain of speed reduction risk, and the duration node of severe weather is converted to the dimension domain of weather impact risk, and each corresponding conversion risk state node is generated.

[0141] Finally, based on these generated conversion risk state nodes, the server generates the prior congestion risk pattern characteristics of this prior potential congestion trend characteristic. Assuming that the converted traffic volume growth risk node is represented as high risk, the speed reduction risk node is represented as medium risk, and the bad weather duration risk node is represented as highly dependent, combining this information, the prior congestion risk pattern characteristics generated by the server may be "high risk congestion caused by bad weather, medium risk caused by speed reduction, and congestion conditions are highly dependent on weather changes."

[0142] For example, another a priori potential congestion trend feature shows that due to the dual factors of road construction and holidays, the traffic volume on a certain road section fluctuates greatly and the vehicle speed is unstable.

[0143] The server tracks risk status nodes and determines risk status nodes such as construction progress nodes, traffic flow fluctuation amplitude nodes, and vehicle speed change frequency nodes. These nodes are then converted to corresponding risk dimension domains to generate corresponding converted risk status nodes. For example, the construction progress node is converted to severe impact in the construction impact risk dimension domain, the traffic flow fluctuation amplitude node is converted to moderate instability in the traffic instability risk dimension domain, and the vehicle speed change frequency node is converted to high risk in the driving safety risk dimension domain.

[0144] Based on these transition risk state nodes, the server-generated prior congestion risk pattern characteristics may be "severe congestion risk due to the combined effects of road construction and holidays, moderately unstable traffic flow, and high driving safety risks brought about by vehicle speed changes."

[0145] In this way, accurate and valuable prior congestion risk pattern features can be generated for each prior potential congestion trend feature, providing strong support for subsequent traffic condition analysis and decision-making.

[0146] In a possible implementation, the method further includes:

[0147] Step C110, obtaining the candidate congestion cause knowledge points corresponding to each prior congestion risk pattern feature, wherein each time a candidate congestion cause knowledge point corresponding to a priori congestion risk pattern feature is obtained, reference congestion trend features corresponding to the candidate congestion cause knowledge points of the multiple prior traffic monitoring big data are obtained.

[0148] Step C120 , clustering the obtained multiple reference congestion trend features to generate one or more reference congestion trend feature clusters.

[0149] Step C130 : selecting, from the cluster representation vector data corresponding to the one or more reference congestion trend feature clusters, cluster representation vector data that meets a set matching requirement with the one priori congestion risk pattern feature.

[0150] Step C140 , based on the selected cluster representation vector data, the candidate congestion cause knowledge points reflected by each reference congestion trend feature in the corresponding reference congestion trend feature cluster are all used as candidate congestion cause knowledge points corresponding to the one priori congestion risk pattern feature.

[0151] In this embodiment, the server performs the following operations in the process of processing traffic monitoring data:

[0152] First, the server obtains the candidate congestion cause knowledge points corresponding to each prior congestion risk pattern feature. For example, the server obtains a prior congestion risk pattern feature, which indicates that on a certain highway section, due to excessive traffic flow during the morning rush hour and a minor traffic accident, the vehicle speed has dropped significantly, forming a moderate congestion risk. For this prior congestion risk pattern feature, the server obtains reference congestion trend features corresponding to candidate congestion cause knowledge points from multiple prior traffic monitoring big data. These reference congestion trend features may include the changing pattern of traffic flow, accident handling time, diversion of surrounding roads, etc.

[0153] Next, the server clusters the multiple reference congestion trend features obtained. Assume that the reference congestion trend features obtained by the server include: Feature 1, the morning peak traffic volume increases from the normal level to 80% exceeding the road carrying capacity within half an hour; Feature 2, the handling time of minor traffic accidents is about 30 minutes; Feature 3, the diversion capacity of surrounding roads is limited and cannot effectively alleviate the pressure on the main road. The server clusters these features based on the similarity and correlation to generate one or more reference congestion trend feature clusters. For example, feature 1 related to traffic volume is grouped into one cluster, feature 2 related to accident handling is grouped into another cluster, and feature 3 related to diversion of surrounding roads is grouped into another cluster.

[0154] Then, from the cluster representation vector data corresponding to the generated one or more reference congestion trend feature clusters, the server selects the cluster representation vector data that meets the set matching requirements with the current a priori congestion risk pattern feature. For example, the set matching requirement is that the similarity between the traffic flow change and the accident handling time and the current a priori congestion risk pattern feature is more than 80%. Through calculation and comparison, the server finds that one of the cluster representation vector data has a high match between the traffic flow change and the accident handling time and the current a priori congestion risk pattern feature.

[0155] Finally, based on the selected cluster representation vector data, the server uses the candidate congestion cause knowledge points used to reflect the reference congestion trend features in the corresponding reference congestion trend feature cluster as the candidate congestion cause knowledge points corresponding to the current prior congestion risk pattern feature. For example, the candidate congestion cause knowledge points in this cluster include "concentrated travel during morning peak hours", "accident handling efficiency needs to be improved", "unreasonable surrounding road planning", etc. The server identifies these knowledge points as candidate congestion cause knowledge points for the current prior congestion risk pattern feature, providing a more comprehensive and accurate basis for further analysis and resolution of traffic congestion problems.

[0156] To give another example, another prior congestion risk pattern feature obtained by the server shows that on a certain highway section, due to the simultaneous occurrence of holiday tourism peaks and road maintenance, serious congestion occurs. The server obtains relevant reference congestion trend features, such as the growth rate of traffic volume during holidays, the progress and scope of road maintenance, etc., and performs clustering. Then, the matching ones are selected from the cluster representation vector data, and the candidate congestion cause knowledge points in the corresponding cluster, such as "insufficient carrying capacity of roads around tourist attractions" and "unreasonable road maintenance arrangements", are determined as candidate congestion cause knowledge points of the prior congestion risk pattern feature.

[0157] Through such processing, the server can more effectively integrate and utilize the information in the traffic monitoring data, providing strong support for traffic management and planning.

[0158] In a possible implementation, the target road condition state vector data is blended based on the priori pattern knowledge data for setting the highway road condition label to generate target potential congestion trend features, which are performed by a pre-trained highway congestion prediction network, wherein the training step of the highway congestion prediction network includes:

[0159] Step S101, obtain a template traffic monitoring big data sequence and the various template highway road condition labels covered therein, and obtain a sample learning data sequence for each template highway road condition label, wherein each sample learning data in the sample learning data sequence includes: a template road condition state vector data and corresponding template annotation feature data, wherein the template road condition state vector data is generated by vector extraction for a template highway road condition label in a template traffic monitoring big data, and the template annotation feature data is a feature of a candidate congestion cause knowledge point of the template traffic monitoring big data.

[0160] Step S102, based on the sample learning data sequence, respectively, the template prior pattern knowledge data preset for the template highway road condition label in the initialized highway congestion prediction network is trained with training parameters, wherein, in a cycle process, after obtaining the template potential congestion trend features generated by blending the extracted template road condition state vector data based on the template prior pattern knowledge data, the template potential congestion trend features are decoded to generate decoded traffic monitoring data.

[0161] Step S103, optimizing the template priori pattern knowledge data according to the feature distance between each decoded traffic monitoring data and the corresponding template traffic monitoring big data and in combination with the template annotation feature data.

[0162] In the highway congestion prediction work, the server performs a series of operations to train and optimize the highway congestion prediction network.

[0163] First, the server obtains a series of template traffic monitoring big data sequences, which cover a variety of different situations. For example, traffic monitoring data of a section of highway in different seasons, time periods, and weather conditions are obtained. At the same time, the road condition labels of each template highway covered by these data are also diverse, such as "holiday peak congestion", "slow traffic caused by bad weather", "partial congestion caused by accidents", etc.

[0164] For each template highway road condition label, the server obtains the corresponding sample learning data sequence. Taking the label "Holiday peak congestion" as an example, each sample learning data in the sample learning data sequence includes a template road condition state vector data and corresponding template annotation feature data. The template road condition state vector data is generated by vector extraction for the label "Holiday peak congestion" from the specific template traffic monitoring big data, which may include features such as a substantial increase in traffic volume, a significant decrease in vehicle speed, and concentrated travel on specific sections of the road. The template annotation feature data is the feature of the candidate congestion cause knowledge point of this template traffic monitoring big data, such as "the concentration of tourist attractions has led to a surge in travel demand."

[0165] Next, the server learns the training parameters of the template prior pattern knowledge data preset for the template highway road condition label in the initialized highway congestion prediction network based on the acquired sample learning data sequence. In a cycle, the server obtains the template potential congestion trend features generated by blending the extracted template road condition state vector data based on the template prior pattern knowledge data. For example, for the template road condition state vector data of "holiday peak congestion", the generated template potential congestion trend features may indicate that the congestion level will continue to rise in the next period of time.

[0166] Then, the server decodes the potential congestion trend characteristics of the template and generates decoded traffic monitoring data. For example, it decodes specific traffic monitoring data such as expected traffic volume, vehicle speed, and road occupancy.

[0167] Afterwards, the server optimizes the template prior pattern knowledge data based on the feature distance between each decoded traffic monitoring data and the corresponding template traffic monitoring big data, combined with the template annotation feature data. Assuming that there is a certain gap between the traffic volume predicted by the decoded traffic monitoring data and the actual traffic volume in the template traffic monitoring big data, and the template annotation feature data indicates that the cause of congestion is unreasonable road planning around a specific scenic spot, the server will adjust the relevant parameters and weights in the template prior pattern knowledge data accordingly to make the prediction result more accurate.

[0168] For example, for the road condition label "slow traffic caused by bad weather", the sample learning data obtained by the server includes the changes in traffic volume and the reduction in vehicle speed on the road in the rain, as well as the corresponding annotated feature data such as "slippery road surface affects vehicle speed". During the training process, the generated template potential congestion trend features and decoded traffic monitoring data are compared and optimized with the actual data, and the template prior pattern knowledge data is continuously improved to improve the accuracy of traffic congestion prediction in bad weather.

[0169] Through such continuous cycles and optimization, the highway congestion prediction network can be made more accurate and effective, providing reliable support for actual traffic management and decision-making.

[0170] In a possible implementation, step S102 includes:

[0171] Step S1021 , performing congestion risk pattern feature recognition on the template potential congestion trend feature to generate a template congestion risk pattern feature.

[0172] Step S1022, based on the decoding priori pattern knowledge data preset for the one template highway road condition label, the template congestion risk pattern features are blended to generate template output features.

[0173] Step S1023: Decode the template output features to generate decoded traffic monitoring data.

[0174] When processing the potential congestion trend features of the template, the server generates decoded traffic monitoring data according to the following steps.

[0175] For example, the server obtains a template potential congestion trend feature for a certain highway section, which indicates that during a specific time period, due to a large-scale event held nearby, the traffic volume begins to increase and the vehicle speed tends to gradually decrease.

[0176] First, the server identifies the congestion risk pattern characteristics of the potential congestion trend characteristics of this template. By analyzing factors such as the growth rate of traffic flow, the magnitude of vehicle speed reduction, and the duration of the activity, the template congestion risk pattern characteristics are generated. For example, it is determined that the situation may lead to a moderate congestion risk, and the congestion may gradually ease after the activity ends.

[0177] Next, based on the preset decoded prior pattern knowledge data for the highway road condition label of this template (such as "traffic changes caused by large-scale events"), the server blends the template congestion risk pattern features. It is assumed that the decoded prior pattern knowledge data contains information such as the impact range of this type of event on surrounding roads, the changing patterns of traffic volume and speed in different time periods, etc. After blending, the template output features are generated, which may include the expected length of the congested section, the specific range of traffic volume values, and the expected low value of the speed, etc.

[0178] Finally, the server decodes the template output features. For example, the traffic flow range is decoded into a specific traffic flow value, the length of the congested road section is decoded into an accurate number of kilometers, and the expected low value of the vehicle speed is decoded into an accurate speed value. This generates decoded traffic monitoring data, such as a traffic flow of 80 vehicles per minute, a congested road section length of 5 kilometers, and a low value of 30 kilometers per hour.

[0179] For example, another template's potential congestion trend feature shows that during holidays, traffic volume on a certain section of highway increased significantly and vehicle speeds dropped significantly due to the combination of road construction and tourist travel.

[0180] The server identifies the congestion risk pattern characteristics, generates a template congestion risk pattern characteristic, and determines that there is a high congestion risk that may last for a long time.

[0181] Then, based on the preset decoding prior pattern knowledge data, the template congestion risk pattern features are blended to generate template output features containing information such as the expected impact of construction progress on traffic and the duration of the tourist travel peak.

[0182] Finally, decoding is performed to obtain decoded traffic monitoring data such as the specific estimated completion time of the construction, the specific end time of the peak tourist travel period, the precise value of the traffic volume and the accurate range of the vehicle speed.

[0183] As a result, accurate and detailed decoded traffic monitoring data can be generated from the potential congestion trend characteristics of the template, providing strong support for traffic management and planning.

[0184] In a possible implementation, step S103 includes:

[0185] Step S1031, obtaining the decoded traffic monitoring data corresponding to each template road condition state vector data and the distinction information between the decoded traffic monitoring data and the corresponding template traffic monitoring big data.

[0186] Step S1032, optimizing the template priori pattern knowledge data and the template decoding prior pattern knowledge data according to the suppression adjustment effect of each discrimination information on the template priori pattern knowledge data and the template decoding priori pattern knowledge data.

[0187] Step S1033, obtaining the template potential congestion trend features corresponding to each template road condition state vector data and the congestion feature deviation information between the corresponding template annotation feature data.

[0188] Step S1034, optimizing the template priori pattern knowledge data according to the suppression adjustment effect of each congestion feature deviation information on the template priori pattern knowledge data.

[0189] Step S1035 , obtaining the template output features corresponding to each template road state vector data and the output deviation information between the template output features and the corresponding template annotation feature data.

[0190] Step S1036, optimizing the template decoding prior pattern knowledge data according to the suppression adjustment effect of each output deviation information on the template decoding prior pattern knowledge data.

[0191] In this embodiment, when optimizing the template priori pattern knowledge data and the template decoding priori pattern knowledge data, the server performs the following specific operations:

[0192] First, the server obtains the decoded traffic monitoring data corresponding to each template road condition state vector data. For example, for a template road condition of a specific highway section during the morning rush hour, the decoded traffic monitoring data may include a predicted traffic volume of 800 vehicles per hour and an average speed of 40 kilometers per hour. The corresponding template traffic monitoring big data shows that the actual traffic volume is 900 vehicles per hour and the average speed is 35 kilometers per hour. The server obtains the difference between the two through detailed comparison and calculation, such as the difference in traffic volume is 100 vehicles per hour and the difference in speed is 5 kilometers per hour.

[0193] Based on this discrimination information, the server plays a suppressive adjustment role on the template prior pattern knowledge data and the template decoding prior pattern knowledge data. If the predicted value of traffic flow is always lower than the actual value, the server may reduce the weight of the factors affecting the morning peak traffic flow in the template prior pattern knowledge data, or adjust the calculation method of the traffic flow decoding in the template decoding prior pattern knowledge data to optimize these two data.

[0194] Next, the server obtains the template potential congestion trend features corresponding to each template road condition state vector data. Assume that for the same template road condition during the morning rush hour, its template potential congestion trend features show that congestion will gradually increase, but the corresponding template annotation feature data shows that congestion should quickly reach a peak and then gradually ease. The server obtains the congestion feature deviation information between the two through precise analysis.

[0195] The template prior pattern knowledge data is suppressed and adjusted based on the deviation information of each congestion feature. For example, if the template potential congestion trend feature always over-conservatively estimates the speed of congestion aggravation, the server will adjust the parameters of the congestion development speed in the template prior pattern knowledge data to make it more consistent with the actual labeled feature data, thereby optimizing the template prior pattern knowledge data.

[0196] Then, the server obtains the template output features corresponding to each template road condition state vector data. For example, for the above morning rush hour road condition, the template output feature may be that the estimated length of the congested road section is 5 kilometers, while the corresponding template annotation feature data is that the actual length of the congested road section is 7 kilometers.

[0197] The server calculates the output deviation information between the two, and uses this output deviation information to suppress and adjust the template decoding prior pattern knowledge data. For example, if the template output features often underestimate the length of congested sections, the server will optimize the part of the template decoding prior pattern knowledge data related to the calculation of the length of congested sections to improve the accuracy of the prediction.

[0198] Through such comprehensive and detailed analysis and adjustment, the server can continuously optimize the template prior pattern knowledge data and template decoding prior pattern knowledge data, making highway congestion prediction more accurate and reliable.

[0199] For another example, for another highway section during holidays, the traffic flow predicted by the decoded traffic monitoring data is 1,200 vehicles per hour, while the traffic flow in the actual template traffic monitoring big data is 1,500 vehicles per hour; the template potential congestion trend feature predicts that congestion will increase slowly, while the template annotated feature data shows that congestion should increase rapidly; the template output feature estimates that the congestion duration is 2 hours, while the actual annotated feature data is 3 hours.

[0200] The server obtains and analyzes the differences and deviations between these data, and suppresses, adjusts and optimizes the template prior pattern knowledge data and template decoding prior pattern knowledge data. For example, the weight of factors affecting traffic flow during holidays is increased, and the calculation method of the speed and duration of congestion aggravation is adjusted to make the prediction results closer to the actual situation.

[0201] By continuously repeating this optimization process, the performance of highway congestion prediction can be continuously improved, providing more effective support for traffic management and planning.

[0202] For example, in a possible implementation, the step of determining the highway traffic label includes:

[0203] Step D110, obtaining a first highway road condition learning data sequence, wherein the highway road condition learning data in the first highway road condition learning data sequence includes sample multi-dimensional acquisition fusion data and road condition feature annotation data of the sample multi-dimensional acquisition fusion data, wherein the road condition feature annotation data includes a road condition feature label and road condition severity of the sample multi-dimensional acquisition fusion data. The road condition feature annotation data of at least one of the highway road condition learning data in the first highway road condition learning data sequence includes a plurality of road condition feature labels.

[0204] Step D120, encode and represent each of the highway road condition learning data, and generate a coding vector sequence for each of the highway road condition learning data, wherein the coding vector sequence of the highway road condition learning data includes a sample road condition state trend coding vector of the sample multi-dimensional acquisition fusion data, a road condition feature label coding vector of the sample multi-dimensional acquisition fusion data, and a road condition severity coding vector of the sample multi-dimensional acquisition fusion data.

[0205] Step D130, based on the sample road condition state trend coding vector and the road condition feature label coding vector of the sample multi-dimensional acquisition and fusion data, execute the first network learning task of the highway road condition prediction network, wherein the first network learning task is used to perform parameter learning on the highway road condition prediction network so that the highway road condition prediction network predicts the road condition feature label corresponding to the sample multi-dimensional acquisition and fusion data.

[0206] Step D140, based on the sample road condition state trend coding vector and the road condition severity coding vector of the sample multi-dimensional acquisition and fusion data, runs the second network learning task of the highway road condition prediction network, the second network learning task is used to perform parameter learning on the highway road condition prediction network so that the highway road condition prediction network predicts the road condition severity of the sample multi-dimensional acquisition and fusion data, so that the highway road condition prediction network that has completed parameter learning can predict the highway road condition feature label and road condition severity of any target multi-dimensional acquisition and fusion data in the target traffic monitoring big data.

[0207] In this embodiment, the server first obtains a first highway road condition learning data sequence. The data in the first highway road condition learning data sequence comes from multiple channels and multiple monitoring devices, covering a wealth of highway road condition information.

[0208] For example, one of the highway road condition learning data contains sample multi-dimensional acquisition fusion data, including information such as traffic volume, speed, and vehicle type collected by cameras, road temperature, humidity, slope, and other data obtained by sensors, and vehicle location information obtained by satellite positioning systems. At the same time, the road condition feature annotation data of the sample multi-dimensional acquisition fusion data includes road condition feature labels, such as "congestion caused by accidents", and road condition severity, such as "serious". In the first highway road condition learning data sequence, at least part of the data's road condition feature annotation data contains multiple road condition feature labels, such as "congestion caused by accidents and accompanied by bad weather."

[0209] Next, the server encodes and represents each highway road condition learning data. Taking a specific learning data as an example, the server converts information such as traffic flow and speed into digital form, and performs standardization processing to generate a sample road condition state trend coding vector. For the road condition feature label "congestion caused by accidents", the server assigns it a specific code to form a road condition feature label coding vector. Similarly, for the road condition severity "serious", the corresponding road condition severity coding vector is also generated. In this way, this highway road condition learning data has a coding vector sequence including a sample road condition state trend coding vector, a road condition feature label coding vector, and a road condition severity coding vector.

[0210] Then, the server performs the first network learning task of the highway road condition prediction network based on the sample road condition state trend coding vector and road condition feature label coding vector of the sample multi-dimensional acquisition fusion data. Assume that the sample road condition state trend coding vector shows that the traffic volume suddenly decreases and the speed drops sharply, and the corresponding road condition feature label coding vector is "congestion caused by accidents". The server learns the parameters of the highway road condition prediction network through a large amount of such data, so that the network can gradually understand and grasp the inherent relationship and pattern between the state trends of traffic volume, speed, etc. and the road condition feature label "congestion caused by accidents".

[0211] At the same time, the server runs the second network learning task of the highway road condition prediction network based on the sample road condition state trend coding vector and road condition severity coding vector of the sample multi-dimensional collection fusion data. For example, when the sample road condition state trend coding vector indicates that the traffic flow is stagnant for a long time and the speed is almost zero, the corresponding road condition severity coding vector is "severe". The server uses these data to train the network so that the network can learn to accurately judge the severity of the road condition based on the road condition state trend.

[0212] After many such learning and training, the highway traffic prediction network has completed parameter learning. At this point, when the server receives new target traffic monitoring big data, such as multi-dimensional collection and fusion data including traffic volume, vehicle speed, road environment, etc. of a certain section of highway in a specific time period, the network can predict its corresponding highway traffic feature label, such as "slow traffic caused by road construction", and the severity of the road condition, such as "moderate".

[0213] For another example, in another highway road condition learning data obtained by the server, the sample multi-dimensional acquisition fusion data shows that during the holidays, the traffic volume on a certain highway section increases steadily, the speed gradually decreases, and the weather is hot. The road condition feature annotation data includes the road condition feature label "holiday travel peak" and the road condition severity "moderate". After the server encodes it, the first network learning task allows the network to learn the relationship between this specific traffic volume and speed change and the "holiday travel peak" label, and the second network learning task allows the network to learn to judge that the road condition severity corresponding to this situation is "moderate".

[0214] When faced with new targets that collect and fuse data in multiple dimensions, such as a highway with a slowly increasing traffic volume, a decreasing speed, and fine weather on a weekend afternoon, the trained highway traffic prediction network can accurately predict that its traffic feature label may be "a small weekend travel peak" and the severity of the traffic condition may be "mild."

[0215] By continuously acquiring and processing a variety of rich and diverse highway road condition learning data, the server continuously optimizes the highway road condition prediction network, enabling it to predict the highway road condition feature labels and road condition severity more and more accurately, providing strong support and guarantee for traffic management and travel planning.

[0216] For example, the server obtains a set of complex highway traffic learning data, in which a sample of multi-dimensional acquisition and fusion data shows that the traffic volume on a certain road section increases abnormally at night, the speed fluctuates frequently, and there are road maintenance and foggy weather. The road condition feature annotation data contains multiple road condition feature labels, such as "special period at night", "impact of road maintenance", "interference from bad weather", and the severity of the road condition "serious".

[0217] The server performs fine encoding processing on these complex data to generate accurate encoding vector sequences. In the first network learning task, the network gradually understands the relationship between the changing patterns of traffic flow and speed under the combined effects of multiple factors such as nighttime, road maintenance, and bad weather, and the complex road condition feature labels by learning such encoding vector sequences. In the second network learning task, the network learns to accurately judge the severity of the road condition as "serious" based on these complex factors.

[0218] When new target multi-dimensional collection and fusion data is subsequently received, such as a sudden increase in traffic volume on a certain road section in the early morning, unstable vehicle speed, road construction and light rain, the network can accurately predict based on previous learning and training that the road condition feature label may be "road construction and weather impact during special periods" and the severity of the road condition may be "moderate".

[0219] By continuously processing and learning various complex and diverse highway traffic data, the server can make the highway traffic prediction network more adaptable and accurate, and better cope with the complex and changeable highway traffic conditions in reality.

[0220] Figure 2 The hardware structure of the traffic monitoring system 100 provided in the embodiment of the present application for implementing the above-mentioned highway congestion prediction method based on big data is shown. Figure 2 As shown, the traffic monitoring system 100 may include a processor 110 , a machine-readable storage medium 120 , a bus 130 , and a communication unit 140 .

[0221] In one possible design, the traffic monitoring system 100 can be a single server or a server group. The server group can be centralized or distributed (for example, the traffic monitoring system 100 can be a distributed system). In some embodiments, the traffic monitoring system 100 can be local or remote. For example, the traffic monitoring system 100 can access information and / or data stored in a machine-readable storage medium 120 via a network. For another example, the traffic monitoring system 100 can be directly connected to the machine-readable storage medium 120 to access the stored information and / or data. In some embodiments, the traffic monitoring system 100 can be implemented on a traffic monitoring system. By way of example only, the traffic monitoring system can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc., or any aggregation thereof.

[0222] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data acquired from an external terminal. In some embodiments, the machine-readable storage medium 120 may store data and / or instructions used by the traffic monitoring system 100 to execute or use to complete the exemplary methods described in this application.

[0223] During the specific implementation process, one or more processors 110 execute computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the highway congestion prediction method based on big data in the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0224] The specific implementation process of the processor 110 can refer to the various method embodiments executed by the above-mentioned traffic monitoring system 100. The implementation principles and technical effects are similar and will not be repeated in this embodiment.

[0225] In addition, an embodiment of the present application also provides a readable storage medium, in which computer executable instructions are set. When a processor runs the computer executable instructions, the above-mentioned highway congestion prediction method based on big data is implemented.

[0226] It should be noted that in order to simplify the description disclosed in this application and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, multiple features are sometimes combined into one embodiment, drawings, or descriptions thereof. Similarly, it should be noted that in order to simplify the description disclosed in this application and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, multiple features are sometimes combined into one embodiment, drawings, or descriptions thereof.

Claims

1. A highway congestion prediction method based on big data, characterized in that: The method comprises: Extracting target road condition state vector data with set highway road condition labels from target traffic monitoring big data containing multiple highway road condition labels; According to the priori pattern knowledge data of the set highway road condition label, the target road condition state vector data is integrated to generate target potential congestion trend characteristics, and the priori pattern knowledge data is used to reflect: the priori road condition state vector data extracted from the set highway road condition label by multiple priori traffic monitoring big data and the corresponding candidate congestion cause knowledge points; Performing congestion risk pattern feature identification on the target potential congestion trend feature to generate a target congestion risk pattern feature, and selecting one or more candidate congestion risk pattern features that meet the target matching requirements from the priori congestion risk pattern features respectively corresponding to the multiple priori traffic monitoring big data for the target congestion risk pattern feature; Using the candidate congestion cause knowledge points corresponding to the one or more candidate congestion risk pattern features as the predicted congestion cause knowledge points of the target traffic monitoring big data; The step of blending the target road condition state vector data based on the priori pattern knowledge data for setting the highway road condition label to generate a target potential congestion trend feature includes: Acquire prior pattern knowledge data with set knowledge dimension space; According to the priori pattern knowledge data, feature mapping interaction is performed on the knowledge dimension space of the target road condition state vector data to generate target potential congestion trend features adapted to the set knowledge dimension space; The step of identifying the congestion risk pattern characteristics of the target potential congestion trend characteristics to generate the target congestion risk pattern characteristics includes: Acquire a priori potential congestion trend features respectively corresponding to the plurality of a priori traffic monitoring big data, each a priori potential congestion trend feature is generated by blending the a priori road condition state vector data of the corresponding a priori traffic monitoring big data based on the a priori pattern knowledge data; Obtaining a priori congestion risk pattern features corresponding to each a priori potential congestion trend feature, wherein each a priori congestion risk pattern feature is generated after congestion risk pattern feature recognition of the corresponding a priori potential congestion trend feature; For the target potential congestion trend feature, determine, from the multiple acquired prior potential congestion trend features, a priori congestion risk pattern feature corresponding to the prior potential congestion trend feature that meets the set risk trend matching requirement as the target congestion risk pattern feature; The steps of generating the priori congestion risk pattern features corresponding to each priori potential congestion trend feature include: Tracking risk state nodes for a priori potential congestion trend characteristics to generate a risk state node set, wherein each risk state node in the risk state node set is used to reflect: a state vector distribution of a risk characterization attribute; Convert each risk state node in the risk state node set to the risk dimension domain corresponding to the corresponding risk representation attribute, and generate a converted risk state node corresponding to each risk state node, each converted risk state node is used to reflect: a state vector conversion parameter of the corresponding risk representation attribute; Based on each generated conversion risk state node, a priori congestion risk pattern feature of the one priori potential congestion trend feature is generated.

2. The highway congestion prediction method based on big data according to claim 1 is characterized in that: The step of extracting target traffic state vector data with a highway traffic label set from target traffic monitoring big data containing a plurality of highway traffic labels comprises: Obtaining a target traffic monitoring data stream including a set highway road condition label from target traffic monitoring big data including multiple highway road condition labels; Splitting the target traffic monitoring data stream into multiple time trajectory windows, and obtaining road condition parameter sets corresponding to the multiple time trajectory windows, wherein a road condition parameter in each road condition parameter set is used to reflect: a road condition indicator in the corresponding time trajectory window; Based on the temporal relationship of the multiple time trajectory windows, multiple road condition parameter sets are processed in sequence to generate a basic road condition parameter set of the target traffic monitoring data stream; According to the association degree of each road condition parameter to the set highway road condition label, the road condition state vector is extracted from the basic road condition parameter set to generate target road condition state vector data of the set highway road condition label in the target traffic monitoring big data.

3. The highway congestion prediction method based on big data according to claim 1 is characterized in that: The step of acquiring the priori pattern knowledge data comprises: Acquire the priori road condition state vector data and the corresponding candidate congestion cause knowledge points extracted from the plurality of priori traffic monitoring big data respectively for the road condition state vector data; and acquire the preliminarily constructed basic priori pattern knowledge data; According to each priori road condition state vector data and the corresponding candidate congestion cause knowledge point, the basic priori pattern knowledge data is optimized and adjusted in multiple cycles to generate the optimized and adjusted priori pattern knowledge data. In one cycle: According to the basic prior pattern knowledge data corresponding to this cycle process, a priori road condition state vector data is integrated to generate a priori potential congestion trend characteristics; Based on the deviation between the prior potential congestion trend characteristics and the reference congestion trend characteristics of the corresponding candidate congestion cause knowledge points, the basic prior pattern knowledge data corresponding to the current cycle process is optimized and adjusted, and the optimized and adjusted prior pattern knowledge data is generated and used as the basic prior pattern knowledge data corresponding to the next cycle process.

4. The highway congestion prediction method based on big data according to claim 3 is characterized in that: According to the basic priori pattern knowledge data corresponding to this cycle process, a priori road condition state vector data is integrated to generate a priori potential congestion trend characteristics, including: Performing pattern correlation analysis on the basic priori pattern knowledge data corresponding to the current cycle process and the basic priori pattern knowledge data corresponding to each previous cycle process, and performing state correlation analysis on the priori pattern knowledge data corresponding to the current cycle process and a priori road condition state vector data to generate candidate priori pattern knowledge data; The one priori road condition state vector data is fused according to the candidate priori pattern knowledge data to generate a priori potential congestion trend feature of the one priori road condition state vector data.

5. The highway congestion prediction method based on big data according to claim 3 is characterized in that: The step of optimizing and adjusting the basic priori pattern knowledge data corresponding to the current cycle based on the deviation between the priori potential congestion trend characteristics and the reference congestion trend characteristics of the corresponding candidate congestion cause knowledge points to generate the optimized and adjusted priori pattern knowledge data includes: Calculating a first degree of deviation between the a priori potential congestion trend feature and a reference congestion trend feature of a candidate congestion cause knowledge point of corresponding a priori road condition state vector data; Acquire other priori road condition state vector data except the corresponding priori road condition state vector data from the acquired multiple priori road condition state vector data; respectively calculating the second deviation between the a priori potential congestion trend feature and the reference congestion trend features of the candidate congestion cause knowledge points corresponding to other a priori road condition state vector data; Based on the stimulating adjustment effect of the first deviation on the basic prior pattern knowledge data corresponding to the current cycle process, and based on the inhibitory adjustment effect of each second deviation on the basic prior pattern knowledge data corresponding to the current cycle process, the basic prior pattern knowledge data corresponding to the current cycle process is optimized and adjusted to generate optimized and adjusted prior pattern knowledge data.

6. The highway congestion prediction method based on big data according to any one of claims 1 to 5, characterized in that: The method further comprises: Obtaining candidate congestion cause knowledge points corresponding to each priori congestion risk pattern feature, wherein each time a candidate congestion cause knowledge point corresponding to a priori congestion risk pattern feature is obtained, obtaining reference congestion trend features corresponding to the candidate congestion cause knowledge points of the plurality of priori traffic monitoring big data respectively; Clustering the obtained multiple reference congestion trend features to generate one or more reference congestion trend feature clusters; Selecting, from the cluster representation vector data corresponding to the one or more reference congestion trend feature clusters, cluster representation vector data that meets a set matching requirement with the one priori congestion risk pattern feature; According to the selected cluster representation vector data, the candidate congestion cause knowledge points reflected by each reference congestion trend feature in the corresponding reference congestion trend feature cluster are all used as the candidate congestion cause knowledge points corresponding to the one priori congestion risk pattern feature.

7. The highway congestion prediction method based on big data according to any one of claims 1 to 5, characterized in that: The target road condition state vector data is blended according to the priori pattern knowledge data of the highway road condition label, and the target potential congestion trend characteristics are generated, which are performed by a pre-trained highway congestion prediction network, wherein the training step of the highway congestion prediction network includes: Obtain a template traffic monitoring big data sequence and each template highway road condition label covered therein, and for each template highway road condition label, obtain a sample learning data sequence, wherein each sample learning data in the sample learning data sequence includes: a template road condition state vector data and corresponding template annotation feature data, wherein the template road condition state vector data is generated by vector extraction for a template highway road condition label in a template traffic monitoring big data, and the template annotation feature data is a feature of a candidate congestion cause knowledge point of the template traffic monitoring big data; According to the sample learning data sequence, training parameter learning is performed on the template prior pattern knowledge data preset for the template highway road condition label in the initialized highway congestion prediction network, wherein in a cycle process, after obtaining the template potential congestion trend features generated by blending the extracted template road condition state vector data according to the template prior pattern knowledge data, the template potential congestion trend features are decoded to generate decoded traffic monitoring data; According to the characteristic distance between each decoded traffic monitoring data and the corresponding template traffic monitoring big data, and in combination with the template annotation characteristic data, the template prior pattern knowledge data is optimized; The step of decoding the potential congestion trend characteristics of the template to generate decoded traffic monitoring data includes: Performing congestion risk pattern feature recognition on the potential congestion trend features of the template to generate template congestion risk pattern features; According to the decoding priori pattern knowledge data preset for the one template highway road condition label, the template congestion risk pattern features are blended to generate template output features; Decoding the template output features to generate decoded traffic monitoring data; The optimizing of the template priori pattern knowledge data based on the characteristic distance between each decoded traffic monitoring data and the corresponding template traffic monitoring big data and combining the template annotation characteristic data includes: Obtaining the decoded traffic monitoring data corresponding to each template road condition state vector data and the distinction information between the decoded traffic monitoring data and the corresponding template traffic monitoring big data; According to each discrimination information, the template a priori pattern knowledge data and the template decoding a priori pattern knowledge data are respectively inhibited and adjusted to optimize the template a priori pattern knowledge data and the template decoding a priori pattern knowledge data; Obtaining the template potential congestion trend features corresponding to each template road condition state vector data and the congestion feature deviation information between the corresponding template annotated feature data; Optimizing the template priori pattern knowledge data according to the suppression adjustment effect of each congestion feature deviation information on the template priori pattern knowledge data; Obtaining the template output features corresponding to each template road condition state vector data and the output deviation information between the template output features and the corresponding template annotation feature data; The template decoding prior pattern knowledge data is optimized according to the suppression and adjustment effect of each output deviation information on the template decoding prior pattern knowledge data.

8. A traffic monitoring system, characterized in that: The traffic monitoring system includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to run the programs, instructions or codes in the memory to implement the highway congestion prediction method based on big data as described in any one of claims 1 to 7 above.

Images