A method for adaptive prediction of freeway congestion levels

Abstract

The application discloses a highway congestion level adaptive prediction method, and relates to the technical field of intelligent traffic control. Traffic environment perception parameters are constructed by collecting multi-dimensional traffic state characteristics, environmental sensing quality parameters and historical statistical characteristics, and adaptive weighting coefficients are dynamically generated based on the parameters to obtain a basic congestion index through fusion, which can dynamically respond to traffic environment changes and improve congestion determination accuracy. A sliding time window is introduced to extract congestion state evolution characteristics, and the basic characteristics and the congestion state evolution characteristics are fused to predict the final congestion index. Finally, a K-means clustering algorithm is used to divide the optimal decision boundary of the final congestion index and predict the congestion, realize congestion level prediction, improve prediction robustness through environmental sensing perception quality dynamic adaptation and time series modeling, and is suitable for active traffic control and decision-making in complex traffic environments.

Description

Technical Field

[0001] This application relates to the field of intelligent traffic control technology, specifically to an adaptive prediction method for highway congestion levels. Background Technology

[0002] Traffic congestion prediction is a core technical issue in intelligent transportation systems. Accurate and timely congestion level prediction has significant application value for highway operation and management, dynamic route guidance, toll station control, and traffic accident prevention. Existing technologies have proposed various congestion identification methods based on fixed detectors (loop coils, cameras), floating car GPS data, or multi-source information fusion. However, analysis reveals the following technical shortcomings: Existing technologies determine the weights of various traffic state characteristics through offline analytic hierarchy process (AHP) and perform weighted summation, failing to respond to the dynamic impact of road environmental conditions (such as visibility, rain / snow, and light intensity) on the reliability of various sensor data. Existing technologies primarily consider the impact of traffic state characteristics on congestion; the temporal characteristics of the congestion evolution process are not explicitly modeled and used for prediction, resulting in poor prediction robustness under adverse environmental conditions and failing to provide a basis for proactive traffic management decisions. Summary of the Invention

[0003] The purpose of this application is to provide an adaptive prediction method for highway congestion levels, which solves the problems of existing technologies where congestion discrimination weights are statically fixed, unable to respond to dynamic changes in environmental perception quality, and the temporal characteristics of the congestion evolution process are not explicitly modeled and used for prediction, resulting in poor prediction robustness under severe environmental conditions and inability to provide decision-making basis for proactive traffic management.

[0004] This application is achieved through the following technical solution:

[0005] An adaptive prediction method for highway congestion levels includes:

[0006] Collect multi-dimensional traffic state characteristics, environmental sensing quality parameters, and historical traffic state statistical characteristics corresponding to the preset road sensing area on the highway to obtain traffic environment sensing parameters.

[0007] The adaptive weighting coefficients corresponding to the multi-dimensional traffic state features are obtained based on the traffic environment perception parameters, and the multi-dimensional traffic state features are weighted and fused based on the adaptive weighting coefficients to obtain the basic congestion index.

[0008] A sliding time window is introduced to model and analyze the changes of the basic congestion index over multiple consecutive detection periods, thereby obtaining the temporal evolution characteristics of congestion status.

[0009] Based on the congestion state evolution characteristics and the basic congestion index, congestion prediction is performed on the road sensing area to obtain the final congestion index.

[0010] Based on the final congestion index, the K-means clustering algorithm is used to divide the optimal decision boundary and predict congestion levels of highways.

[0011] In one possible implementation, after predicting the highway congestion level, the method further includes:

[0012] Based on the highway congestion level, find the corresponding early warning strategy and traffic control strategy, and implement the early warning strategy and traffic control strategy.

[0013] In one possible implementation, multi-dimensional traffic state features, environmental sensing quality parameters, and historical traffic state statistical features corresponding to a pre-defined road sensing area on the highway are collected to obtain traffic environment sensing parameters, including:

[0014] The system collects the total number of vehicles in the preset road perception area on the highway, the proportion of stationary and driving vehicles with speeds below the preset speed threshold, the average headway or average inter-vehicle distance, the proportion of the total projected area of ​​vehicles to the area of ​​the road perception area, and the dispersion characteristics of multi-lane traffic flow to obtain multi-dimensional traffic state characteristics.

[0015] Collect weather conditions, lighting conditions, road conditions, and environmental sensor data corresponding to the road sensing area to obtain environmental sensing quality parameters;

[0016] Collect typical operating parameters, historical congestion patterns, and characteristic parameters from previous data corresponding to the road sensing area to obtain historical traffic status statistical characteristics;

[0017] Multi-dimensional traffic condition characteristics, environmental sensing quality parameters, and historical traffic condition statistical characteristics are used together as traffic environment sensing parameters.

[0018] In one possible implementation, the adaptive weighting coefficients corresponding to the multi-dimensional traffic state features obtained based on the traffic environment perception parameters are as follows:

[0019] ;

[0020] in, This represents the adaptive weighted coefficient vector corresponding to the multi-dimensional traffic state at time t, which contains the adaptive weighted coefficients corresponding to each traffic state feature. This represents the weight generation function corresponding to the adaptive weighting coefficients. This represents the multi-dimensional traffic state characteristics at time t. This represents the environmental sensing quality parameter at time t. This represents the statistical characteristics of historical traffic conditions at time t.

[0021] In one possible implementation, the multi-dimensional traffic state features are weighted and fused according to the adaptive weighting coefficients to obtain a basic congestion index, including:

[0022] The adaptive weighting coefficients are normalized to obtain the normalized adaptive weighting coefficients.

[0023] The traffic state features in the multi-dimensional traffic state features are normalized to obtain the normalized traffic state features.

[0024] The normalized traffic state features are weighted and fused using adaptive weighting coefficients, resulting in the basic congestion index:

[0025] ;

[0026] in, This represents the basic congestion index at time t. This represents the adaptive weighting coefficient after the i-th normalization process. This represents the traffic state characteristics after the i-th normalization process at time t.

[0027] In one possible implementation, a sliding time window is introduced to model and analyze the changes of the basic congestion index over multiple consecutive detection periods, obtaining the temporal evolution characteristics of congestion states, including:

[0028] Build length is The sliding time window, that is, the sliding time window allows for the inclusion of... The basic congestion index corresponding to each detection cycle;

[0029] The rate of change of the basic congestion index within a sliding time window is obtained to determine the growth rate characteristic.

[0030] Obtain the cumulative change of the basic congestion index within the sliding time window to obtain the cumulative trend characteristics;

[0031] Obtain the short-time difference index of the basic congestion index within the sliding time window to obtain the abrupt change characteristics;

[0032] The growth rate feature, cumulative trend feature, and abrupt change feature are collectively used as the evolutionary features of congestion state that change over time.

[0033] In one possible implementation, the rate of change of the baseline congestion index within a sliding time window is obtained, resulting in a growth rate characteristic:

[0034] ;

[0035] in, This represents the growth rate characteristic at time t. This represents the basic congestion index at time t. Indicates time The basic congestion index on the internet, Indicates the preset time window length.

[0036] In one possible implementation, the cumulative change of the basic congestion index within a sliding time window is obtained, yielding the cumulative trend characteristic as follows:

[0037]

[0038] in, This represents the cumulative trend characteristics at time t. Indicates time The basic congestion index on the internet, Represents the integral variable, representing any historical moment within the sliding time window; Represents the time variable The derivative of is used to represent the continuous integral over the time interval. Indicates the preset time window length and integration interval. This represents a sliding time window ending at the current time t.

[0039] In one possible implementation, the short-time difference index of the baseline congestion index within a sliding time window is obtained to yield the abrupt change characteristic:

[0040] ;

[0041] in, This represents the abrupt change characteristic at time t. This represents the basic congestion index at time t. Indicates time The basic congestion index on the internet, This represents the short-time differential time interval, which is a preset time step used to characterize the magnitude of changes in traffic conditions within a short time scale.

[0042] In one possible implementation, based on the congestion state evolution characteristics and the basic congestion index, congestion prediction is performed on the road sensing area to obtain the final congestion index:

[0043] ;

[0044] in, This represents the final congestion index at time t. This represents the basic congestion index at time t. This represents the cumulative trend characteristics at time t. This represents the growth rate characteristic at time t. This represents the abrupt change characteristic at time t. This represents the weighting coefficient corresponding to the basic congestion index. This represents the weighting coefficient corresponding to the growth rate feature. This represents the weighting coefficient corresponding to the cumulative trend feature. This represents the weighting coefficient corresponding to the mutation feature.

[0045] Compared with the prior art, this application has the following advantages and beneficial effects:

[0046] This application provides an adaptive prediction method for highway congestion levels. It constructs traffic environment perception parameters by collecting multi-dimensional traffic state features, environmental sensor quality parameters, and historical statistical features. Based on these parameters, it dynamically generates adaptive weighting coefficients and fuses them to obtain a basic congestion index. This method can dynamically respond to changes in the traffic environment and improve the accuracy of congestion judgment. A sliding time window is introduced to extract congestion state evolution features, and the basic features and congestion state evolution features are fused to predict the final congestion index. Finally, the K-means clustering algorithm is used to perform optimal decision boundary division and congestion prediction on the final congestion index, achieving congestion level prediction. Dynamic adaptation of environmental sensor perception quality and time series modeling improve prediction robustness, making it suitable for proactive traffic control decisions in complex traffic environments. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the exemplary embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0048] Figure 1 A flowchart illustrating a highway congestion level prediction method based on multi-dimensional traffic state characteristics, provided for embodiments of this application;

[0049] Figure 2 This is a schematic diagram illustrating an application scenario provided in the embodiments of this application;

[0050] Figure 3 A schematic diagram of the field equipment deployment provided in the embodiments of this application. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the embodiments and accompanying drawings. The illustrative embodiments and descriptions of this application are only for explaining this application and are not intended to limit this application.

[0052] like Figure 1 As shown in the figure, this application provides an adaptive prediction method for highway congestion levels, including:

[0053] S101. Collect multi-dimensional traffic state characteristics, environmental sensing quality parameters, and historical traffic state statistical characteristics corresponding to the preset road sensing area on the highway to obtain traffic environment sensing parameters.

[0054] S102. Obtain the adaptive weighting coefficients corresponding to the multi-dimensional traffic state features based on the traffic environment perception parameters, and perform weighted fusion of the multi-dimensional traffic state features based on the adaptive weighting coefficients to obtain the basic congestion index.

[0055] S103. Introduce a sliding time window to model and analyze the changes of the basic congestion index over multiple consecutive detection periods, and obtain the temporal evolution characteristics of the congestion state.

[0056] S104. Based on the congestion state evolution characteristics and the basic congestion index, perform congestion prediction on the road perception area to obtain the final congestion index.

[0057] S105. Based on the final congestion index, the K-means clustering algorithm is used to divide the optimal decision boundary and predict congestion levels of highways.

[0058] This application provides an adaptive prediction method for highway congestion levels. It constructs traffic environment perception parameters by collecting multi-dimensional traffic state features, environmental sensor quality parameters, and historical statistical features. Based on these parameters, it dynamically generates adaptive weighting coefficients and fuses them to obtain a basic congestion index. This method can dynamically respond to changes in the traffic environment and improve the accuracy of congestion judgment. A sliding time window is introduced to extract congestion state evolution features, and the basic features and congestion state evolution features are fused to predict the final congestion index. Finally, the K-means clustering algorithm is used to perform optimal decision boundary division and congestion prediction on the final congestion index, achieving congestion level prediction. Dynamic adaptation of environmental sensor perception quality and time series modeling improve prediction robustness, making it suitable for proactive traffic control decisions in complex traffic environments.

[0059] In one possible implementation, multi-dimensional traffic state features, environmental sensing quality parameters, and historical traffic state statistical features corresponding to a pre-defined road sensing area on the highway are collected to obtain traffic environment sensing parameters, including:

[0060] The system collects the total number of vehicles in the preset road perception area on the highway, the proportion of stationary and driving vehicles with speeds below the preset speed threshold, the average headway or average inter-vehicle distance, the proportion of the total projected area of ​​vehicles to the area of ​​the road perception area, and the dispersion characteristics of multi-lane traffic flow to obtain multi-dimensional traffic state characteristics.

[0061] Collect weather conditions, lighting conditions, road conditions, and environmental sensor data corresponding to the road sensing area to obtain environmental sensing quality parameters;

[0062] Collect typical operating parameters, historical congestion patterns, and characteristic parameters from previous data corresponding to the road sensing area to obtain historical traffic status statistical characteristics;

[0063] Multi-dimensional traffic condition characteristics, environmental sensing quality parameters, and historical traffic condition statistical characteristics are used together as traffic environment sensing parameters.

[0064] The system collects weather conditions, lighting conditions, road conditions, and environmental sensor data corresponding to the road sensing area, and quantifies the data to obtain environmental sensing quality parameters, denoted as:

[0065] ;

[0066] Collect typical operating parameters, historical congestion patterns, and characteristic parameters from previous data corresponding to the road sensing area to obtain historical traffic status statistical characteristics, m=5;

[0067] The environmental sensing quality parameters include: visibility index. Precipitation intensity index: Obtained from meteorological station data or based on image sharpness inversion. Light intensity index: Acquired from rain gauges or meteorological interface data. Road surface condition indicators collected by light sensors The reliability index of the sensor is determined based on temperature, humidity, and icing detection data. The calculation method is as follows:

[0068] ;

[0069] in: For data missing rate, For short-term volatility, The deviation rate from the historical mean is represented by α, β, and γ, which are weighting coefficients.

[0070] Typical operating parameters and historical traffic data corresponding to the road sensing area are collected, and statistical analysis and pattern recognition methods are used to obtain the statistical characteristics of historical traffic conditions, denoted as:

[0071] ;

[0072] Where n=4, then the historical average vehicle speed characteristics for:

[0073] ;

[0074] This represents the average historical traffic flow across different time periods. Historical market share characteristics;

[0075] Historical congestion patterns are identified, and these patterns are obtained through a clustering algorithm:

[0076] ;

[0077] in, Indicates time Historical congestion patterns and characteristics; The set of historical traffic features is constructed as a multi-dimensional feature vector: .in: The proportion of slow-moving vehicles is used to characterize the degree of traffic congestion. This refers to the average headway or distance between vehicles, used to reflect the closeness of vehicle following. Traffic density is used to describe the number of vehicles per unit length of road. The rate of change of traffic flow is used to characterize the fluctuations in traffic flow over time; superscript The transpose operation represents a vector, used to convert a row vector into a column vector; This represents a clustering analysis function used for pattern recognition and classification of historical traffic feature sets; the function performs clustering analysis on the input feature vector. Unsupervised learning is used to classify traffic operation status into several typical pattern categories, thereby extracting historical congestion pattern features;

[0078] The historical congestion pattern features further include: congestion occurrence feature vector, congestion duration distribution, and congestion dissipation rate.

[0079] Multidimensional traffic state characteristics Environmental sensing quality parameters and historical traffic status statistical characteristics By integrating the data, the traffic environment perception parameters are constructed as follows: .

[0080] For example, within each detection cycle, multi-dimensional traffic state features can be extracted from a preset road perception area (i.e., a pre-defined ROI area) to obtain a traffic state feature vector:

[0081] ;

[0082] in, The traffic state features are multi-dimensional and are presented in the form of traffic state feature vectors. The total number of vehicles corresponding to the preset road sensing area reflects the traffic flow or vehicle density of that road sensing area. The proportion of stationary or moving vehicles with speeds below a preset speed threshold is used to measure the degree of traffic congestion or queuing, and can be expressed as... The higher this ratio, the more severe the vehicle stagnation within the road perception area; This represents the number of low-speed vehicles within the road sensing area at time t, i.e., the number of vehicles traveling at speeds below a preset speed threshold. The proportion of the total projected area of ​​vehicles to the area of ​​the road sensing zone can be expressed as: As a representation of traffic density or space occupancy, This represents the total projected area of ​​the vehicle within the road sensing area. The area of ​​the road sensing zone; This refers to the average headway or average inter-vehicle distance, reflecting the longitudinal spacing of vehicles. For example, in terms of time... It can be defined as the average headway of vehicles within the road sensing area, expressed as... , For the first The time interval between a vehicle and the vehicle in front of it. This represents the total number of vehicles within the road perception area at time t. The smaller the average headway or average inter-vehicle distance, the more compact the vehicle movement and the higher the potential risk of congestion. This refers to the dispersion characteristics of multi-lane traffic flow, representing the degree of difference in the distribution of traffic load or speed among lanes. For example, the standard deviation of the average lane speed can be used to quantify this dispersion, expressed as... , For the number of lanes, For the first The average speed of vehicles in each lane. This represents the average vehicle speed across all lanes within the road sensing area. When one lane experiences severe congestion while adjacent lanes remain relatively clear... It will increase significantly, and vice versa. A smaller value indicates that traffic flow is more spatially balanced. (Superscript) The transpose operation represents a vector, used to convert a row vector into a column vector.

[0083] These multidimensional features collectively depict the overall traffic operation status within the road perception area, including flow rate, speed, density, and flow uniformity, providing a comprehensive data foundation for congestion assessment.

[0084] In one possible implementation, the adaptive weighting coefficients corresponding to the multi-dimensional traffic state features obtained based on the traffic environment perception parameters are as follows:

[0085] ;

[0086] in, This represents the adaptive weighted coefficient vector corresponding to the multi-dimensional traffic state at time t, which contains the adaptive weighted coefficients corresponding to each traffic state feature. This represents the weight generation function corresponding to the adaptive weighting coefficients. This represents the multi-dimensional traffic state characteristics at time t. This represents the environmental sensing quality parameters at time t, such as current weather conditions (rain, snow, fog, etc.), lighting conditions, road conditions, and reliability indicators of data from various sensors. It represents the historical traffic status statistics at time t, which may include the typical operating parameters of the road segment at different time periods, historical congestion patterns, and the importance statistics of each characteristic parameter in previous data.

[0087] For example, the adaptive weighted coefficient vector can be represented as:

[0088] (1) Weight vector

[0089] ;

[0090] (2) Definition of a single weight (core formula)

[0091]

[0092] in: Represents the adaptive weighted coefficient vector at time t; superscript The transpose operation represents a vector, used to convert a row vector into a column vector; Indicates time Upper Weight coefficients corresponding to each traffic state feature; Indicates time Upper Multi-dimensional traffic condition characteristics ; Indicates time Upper A quality parameter of environmental sensing. ; Indicates time Above and below Historical traffic state statistical features corresponding to each traffic state feature; Indicates the first The influence coefficients corresponding to the environmental sensing quality parameters; These represent preset adjustment coefficients, which are used to control the degree of influence of traffic state characteristics, environmental perception parameters, and historical statistical characteristics on the weight calculation. Indicates the number of multi-dimensional traffic state features; This indicates the number of environmental sensing quality parameters.

[0093] In the actual operation environment of highways, the measurement error of camera occupancy at night or in rainy or foggy weather can reach more than 30%. Loop detectors are prone to pulse noise when heavy trucks pass through densely, and GPS signals attenuate significantly in tunnels or mountainous sections. When the quality of environmental sensing deteriorates (for example, heavy fog causes inaccurate camera occupancy measurements), the static weighting system of existing technologies will continuously assign fixed high weights to low-quality data, leading to the accumulation of errors in congestion index calculation and ultimately outputting incorrect congestion level judgments. Therefore, the embodiments of this application obtain adaptive weighting coefficients corresponding to the multi-dimensional traffic state features based on the traffic environment sensing parameters, realizing real-time assessment of the traffic environment and dynamic redistribution of feature weights. This ensures that each feature obtains a weight that matches its reliability under the current environmental conditions, effectively adapting to changes in the traffic environment and improving the accuracy of congestion level judgments.

[0094] This dynamic weighting mechanism ensures that the contribution of different features to congestion assessment can be adjusted in a timely manner when traffic conditions or the external environment change. For example, under conditions of heavy rain or dense fog, when the reliability of video detection decreases, the system will automatically reduce the weight of features related to video extraction while increasing the weight of features such as radar traffic flow; when the proportion of low-speed vehicles within the road perception area is detected... When the congestion level rises sharply, the weight of this feature in the congestion index calculation will be increased accordingly to more accurately reflect the formation of congestion. Through the above adaptive adjustment, the congestion determination based on multi-source data can maintain high accuracy and robustness under various complex scenarios.

[0095] In one possible implementation, the multi-dimensional traffic state features are weighted and fused according to the adaptive weighting coefficients to obtain a basic congestion index, including:

[0096] The adaptive weighting coefficients are normalized to obtain the normalized adaptive weighting coefficients.

[0097] The traffic state features in the multi-dimensional traffic state features are normalized to obtain the normalized traffic state features.

[0098] The normalized traffic state features are weighted and fused using adaptive weighting coefficients, resulting in the basic congestion index:

[0099] ;

[0100] in, This represents the basic congestion index at time t, which is a measure of the basic congestion index at time t. A real-time quantitative value of the degree of traffic congestion; This represents the adaptive weighting coefficient after the i-th normalization process. This represents the traffic state characteristics after the i-th normalization process at time t, thereby normalizing features of different dimensions to the same evaluation scale.

[0101] For example, weight generation function Taking into account the currently extracted traffic state feature vector Environmental conditions and sensor-sensed quality parameters set and historical traffic status statistical characteristics The feature weight vector of each feature is obtained by dynamic calculation. To ensure the rationality of the fusion calculation, the calculated weights can be normalized so that the sum of all components is 1, as shown in the following formula: .

[0102] In one possible implementation, a sliding time window is introduced to model and analyze the changes of the basic congestion index over multiple consecutive detection periods, obtaining the temporal evolution characteristics of congestion states, including:

[0103] Build length is The sliding time window, that is, the sliding time window allows for the inclusion of... The basic congestion index corresponding to each detection cycle;

[0104] The rate of change of the basic congestion index within a sliding time window is obtained to determine the growth rate characteristic.

[0105] Obtain the cumulative change of the basic congestion index within the sliding time window to obtain the cumulative trend characteristics;

[0106] Obtain the short-time difference index of the basic congestion index within the sliding time window to obtain the abrupt change characteristics;

[0107] The growth rate feature, cumulative trend feature, and abrupt change feature are collectively used as the evolutionary features of congestion state that change over time.

[0108] In one possible implementation, the rate of change of the baseline congestion index within a sliding time window is obtained, resulting in a growth rate characteristic:

[0109] ;

[0110] in, This represents the growth rate characteristic at time t, reflecting the speed at which congestion worsens or eases. A positive value indicates that the degree of congestion is increasing, and the larger the value, the faster the congestion worsens. A negative value indicates that congestion is easing. This represents the basic congestion index at time t. Indicates time The basic congestion index on the internet, Indicates the preset time window length.

[0111] In one possible implementation, the cumulative change of the basic congestion index within a sliding time window is obtained, yielding the cumulative trend characteristic as follows:

[0112]

[0113] in, Indicates time The cumulative trend characteristics are used to reflect the average level of the basic congestion index within the sliding time window; Indicates time The basic congestion index; Let represent the integration variable, and satisfy: This represents any historical moment within the sliding time window; Represents the time variable The derivative is used to perform integration operations over a continuous time interval; This indicates the preset time window length, used to characterize the time scale of traffic state analysis; the lower limit of integration. Indicates the start time of the sliding time window; upper limit of integration. Indicates the current time; This represents the normalization coefficient for time averaging of the baseline congestion index within the sliding time window. It is used to eliminate the scaling effect of the time window length on the integral results, making the cumulative trend characteristics comparable.

[0114] In one possible implementation, the short-time difference index of the baseline congestion index within a sliding time window is obtained to yield the abrupt change characteristic:

[0115] ;

[0116] in, This represents the abrupt change characteristics at time t, and can detect rapid changes (sudden increases or decreases) in the congestion index within a short period of time. It is used to capture sudden changes in congestion caused by accidents, sudden surges in traffic, etc. When the value is positive and exceeds the preset threshold, it indicates a sudden increase in congestion within a very short period of time; when... It is a positive value and exceeds the preset positive threshold. When, it indicates a sudden increase in congestion within a very short period of time; when Negative values ​​and below the preset negative threshold When, it indicates a sudden easing of traffic congestion or a sharp drop in the index; when When the congestion index changes gradually, it indicates that the traffic situation is relatively stable. This characteristic allows for the timely identification of abrupt changes in congestion. This represents the basic congestion index at time t. Indicates time The basic congestion index on the internet, This represents the short time difference interval, which is a preset time step used to characterize changes in traffic conditions over a short time scale.

[0117] Based on the above evolutionary characteristics, a traffic congestion state evolution model is constructed. This traffic congestion state evolution model uses the baseline congestion index at the current moment. Taking into account the aforementioned characteristics of congestion evolution, a final congestion index is output that reflects the impact of traffic condition evolution. To incorporate congestion evolution trends into the index assessment, the final congestion index... It can be expressed as a functional relationship between the basic index and each evolutionary characteristic.

[0118] In one possible implementation, based on the congestion state evolution characteristics and the basic congestion index, congestion prediction is performed on the road sensing area to obtain the final congestion index:

[0119] ;

[0120] in, This represents the final congestion index at time t. This represents the basic congestion index at time t. This represents the cumulative trend characteristics at time t. This represents the growth rate characteristic at time t. This represents the abrupt change characteristic at time t. This represents the weighting coefficient corresponding to the basic congestion index. This represents the weighting coefficient corresponding to the growth rate feature. This represents the weighting coefficient corresponding to the cumulative trend feature. Indicates mutation characteristics. to The final congestion index is used to adjust the influence of the base index and various evolutionary features on the final index, and can be determined through calibration using historical data. It incorporates information on recent congestion trends, enabling it to characterize the development stage of congestion. For example, even if vehicle speed hasn't decreased significantly at the moment, if the congestion index deteriorates rapidly in the last few minutes, the final congestion index will... This will increase accordingly, thus providing early warning of potential severe congestion.

[0121] This application utilizes the Final Congestion Index. Based on its evolutionary characteristics, the system intelligently determines the level of congestion. It combines historical traffic condition sample data to train a congestion level classification model, dividing congestion into several levels (e.g., smooth traffic, light congestion, moderate congestion, and severe congestion), and automatically generating the optimal decision boundary between each level as the judgment threshold.

[0122] In some possible implementations, unsupervised clustering analysis methods can be employed. The K-means clustering algorithm can be used to analyze a large amount of historical data. The samples are clustered, automatically divided into a preset number of N clusters based on numerical similarity. The clustering algorithm minimizes intra-cluster variance and maximizes inter-cluster variance, ensuring that each cluster corresponds to a typical congestion state. The clustering results are then sorted by cluster center location. The data is sorted by size on the axis, and the midpoint between adjacent cluster centers can be used as the threshold for classifying the congestion levels, thus forming congestion level boundaries based on the natural intervals of data distribution.

[0123] In another implementation, if historical congestion level labeled data is available, a supervised classification model can be used for threshold learning. Models such as Support Vector Machines (SVMs), Decision Trees, or Logistic Regression are used, with the actual congestion levels of historical samples as supervisory signals for training. SVMs form the optimal separating hyperplane in the feature space by maximizing the classification margin between samples of different levels; when only the classification margin of historical samples is used... When used as a discriminant variable, the hyperplane degenerates into several split points in one-dimensional space, i.e., the optimal set of rank thresholds. The decision tree model then iteratively selects the split point that maximizes the reduction in information gain or Gini impurity. By performing recursive partitioning, the segmentation values ​​corresponding to the internal nodes naturally constitute the congestion level threshold.

[0124] Gaussian mixture modeling can also be used to model different levels of congestion. The probability distribution is fitted to the data, and the intersection of adjacent distribution functions is used as the level threshold. While the different algorithms described above differ in their implementation paths, their essence is to uncover the inherent boundaries between low and high congestion levels from historical data, ensuring that threshold determination is based on objective statistical laws. This process replaces the traditional method of relying on manual experience to specify thresholds, giving the determination of thresholds a data-driven, objective basis.

[0125] Assuming the degree of congestion is divided into: Levels (e.g.) Corresponding to the four levels mentioned above, the congestion index threshold set for each level is obtained through training the above classification model. This serves as the dividing point for congestion levels. The congestion level determination module uses... As input, output the current congestion level based on the threshold range. .For example:

[0126] ;

[0127] In the above judgment rules, Level 1 corresponds to the "unimpeded" state. The most severe congestion level corresponds to the lowest level, with intermediate levels representing varying degrees of congestion. When the system is actually running, the congestion level is determined based on real-time calculations. The congestion level is determined by the threshold range into which the congestion falls. .

[0128] Meanwhile, this application establishes a threshold adaptive update mechanism: during long-term system operation, real traffic data and human feedback are continuously collected to continuously update and optimize the classification model online or offline. For example, when operators report that a judgment result does not match the actual situation, the sample can be included in the training to correct the model; with seasonal changes in traffic patterns or adjustments to the road network structure, new monitoring data is introduced to retrain or fine-tune the classification model. Through this adaptive mechanism, the congestion level threshold can be dynamically adjusted according to the evolution of traffic flow characteristics, ensuring that the system maintains stable and reliable discrimination performance throughout long-term operation.

[0129] In one possible implementation, after predicting the highway congestion level, the method further includes:

[0130] Based on the highway congestion level, find the corresponding early warning strategy and traffic control strategy, and implement the early warning strategy and traffic control strategy.

[0131] When the congestion index When the preset congestion level threshold is reached, the system will automatically trigger corresponding early warnings and traffic control strategies, achieving rapid linkage between perception, analysis, and regulation. Specifically:

[0132] For mild congestion (such as just reaching the minimum congestion level threshold), the system can issue speed limit prompts and congestion warnings through the front-end information board, reminding drivers to slow down, drive cautiously, and change lanes or choose alternative routes in advance.

[0133] When moderate to severe congestion is detected, or when congestion shows a continuing worsening trend, the central analysis module can take further measures to guide traffic diversion. For example, it can guide subsequent vehicles to detour via navigation systems or variable message signs along the route, reducing the amount of traffic entering the congested area.

[0134] In the event of severe congestion caused by an accident, the system can coordinate with highway control measures to trigger temporary traffic control or emergency response plans (such as closing relevant ramps or opening hard shoulders) to restore traffic order as quickly as possible.

[0135] Through the above-mentioned linkage control mechanism, this application can closely integrate traffic condition perception, congestion assessment and decision control to achieve closed-loop management from congestion monitoring to response and handling, and minimize the negative impact of traffic congestion on highway operating efficiency and driving safety.

[0136] Compared with the prior art, this application has at least the following beneficial effects:

[0137] (1) By collecting multi-dimensional traffic state features, environmental sensing quality parameters, and historical statistical features, traffic environment perception parameters are constructed, and adaptive weighting coefficients are dynamically generated based on these parameters. The resulting basic congestion index is obtained through fusion, which can dynamically respond to changes in the traffic environment. This solves the problem that existing technologies have statically fixed congestion discrimination weights and cannot respond to dynamic changes in environmental perception quality. It fully leverages the advantages of various sensors such as video and radar while avoiding their disadvantages, ensuring that congestion detection maintains high reliability under various weather and traffic conditions.

[0138] (2) A sliding time window is introduced to model and analyze the changes in the basic congestion index over multiple consecutive detection periods, obtaining the temporal evolution characteristics of the congestion state. The basic characteristics and congestion state evolution characteristics are then integrated to predict the final congestion index. This solves the problem that existing technologies do not explicitly model and use the temporal characteristics of the congestion evolution process for prediction, resulting in poor prediction robustness under adverse environmental conditions. It can capture the dynamic process of congestion from its formation to its aggravation, enabling early perception of congestion trends. The system can issue an early warning when there are signs of worsening congestion, helping to take timely intervention measures and reduce the risk of secondary congestion and accidents.

[0139] (3) The congestion level discrimination model trained based on historical data can automatically determine the threshold boundaries of different congestion levels, reducing the reliance on human experience in setting thresholds. The thresholds will be dynamically optimized with new data and feedback, improving the scientific nature and adaptability of the judgment criteria and ensuring accurate and stable long-term operation.

[0140] (4) The method of this application is designed for complex highway scenarios, taking into account the impact of factors such as severe weather and abnormal events, and has good environmental robustness. At the same time, through continuous learning and model updates, the system can adapt to seasonal changes and trend evolution of traffic flow, and its performance will not degrade during long-term use, making it suitable for various complex highway operating environments.

[0141] like Figure 2 As shown, in order to enable those skilled in the art to better understand the technical solutions described in the embodiments of this application, the following examples are provided.

[0142] The method described in this application can be used to monitor congestion in areas such as highway interchanges, merging areas of main roads and ramps, toll plazas, and accident-prone sections. These locations have complex traffic flows and are prone to congestion, further highlighting the advantages of this application. Specific application scenarios of the method described in this application may include:

[0143] Front-end multi-dimensional perception subsystem: Front-end sensing devices deployed on the roadside or in vehicles are used to collect multi-source heterogeneous traffic data, including video images, millimeter-wave radar detection data, vehicle detector data, and environmental status data such as weather. This module covers a preset road area in real time to obtain comprehensive traffic flow information.

[0144] Roadside edge computing nodes: These are computing units integrated into roadside equipment, performing real-time preprocessing and analysis of collected data. On one hand, they construct traffic state feature vectors, performing dynamic weight calculations and real-time calculations of congestion indices; on the other hand, they analyze short-term traffic data sequences to extract traffic state evolution characteristics. The edge computing module can quickly complete calculations locally, reducing data transmission latency.

[0145] The central cloud management and training platform, deployed at the traffic management center, integrates information uploaded from various edge nodes to determine congestion levels and adaptively update thresholds. Based on the determination results, it also triggers relevant control measures. The platform feeds back the analysis and decision-making results to the multi-dimensional terminal execution subsystem, where front-end guidance devices (such as variable message signs and signal control systems) respond, triggering early warning and diversion control strategies to achieve global optimized scheduling.

[0146] Through the coordinated operation of the above modules, we can achieve all-time and all-round perception of highway traffic conditions, as well as intelligent assessment and coordinated control of congestion conditions.

[0147] Deployment method: such as Figure 3 As shown, roadside edge computing nodes are typically integrated into roadside units (RSUs) or smart sensing device boxes mounted on the roadside. They connect to front-end sensing modules such as high-definition cameras and radar, and communicate with the central cloud management and training platform via wired or wireless networks. The central cloud management and training platform can be located in the server cluster of the highway monitoring center to comprehensively analyze the congestion index and characteristic data sent by each monitoring point.

[0148] Operation: The system continuously performs data collection and processing in a preset short cycle (e.g., every 30 seconds or 1 minute). The roadside edge computing nodes calculate and output the final congestion index for the current road segment in real time. And congestion level. Once the final congestion index of a certain road segment is determined... If the congestion exceeds a threshold and is deemed severe, the roadside edge computing nodes will immediately upload an alert to the central cloud management and training platform. The central cloud management and training platform then aggregates the information from each road segment to generate a global traffic congestion status. Simultaneously, based on the congestion level and trend, it automatically selects and issues appropriate control commands.

[0149] Application Example: For instance, deploy this system at a major highway interchange. During off-peak hours, the system monitors balanced traffic flow on the main line, ultimately adjusting the congestion index. Maintaining smooth traffic flow. During peak hours, due to the surge in traffic, the feature vector of a lane narrowing area... The data shows a sharp increase in the total number of vehicles and a greater proportion of slow-moving vehicles, leading to a decrease in the basic congestion index. The congestion level is gradually rising. Based on the continuous upward trend of the index over the past few minutes, the system determines that the congestion on this ramp is rapidly worsening, and the final congestion index... The congestion threshold has been exceeded. At this point, the roadside edge computing node immediately classifies the road segment as "severely congested" and notifies the central cloud management and training platform.

[0150] Based on this, the central cloud management and training platform triggered an emergency plan: on the one hand, it issued warnings of severe congestion ahead via variable message signs along the route and guided vehicles to take alternative routes; on the other hand, it notified adjacent toll stations to temporarily suspend traffic flow to control traffic volume. Because the system achieved early identification and rapid response to congestion, subsequent traffic was diverted in a timely manner, and the congestion did not worsen further, beginning to ease within a short period. Throughout the process, the system also fed back the actual results to the model, continuously optimizing threshold settings and weight allocation, making the response more accurate when encountering similar situations in the future.

[0151] In summary, the technical solution described in this application can operate stably in the complex and ever-changing highway traffic environment. Through real-time multi-source data fusion, adaptive feature weight adjustment, and modeling and analysis of congestion evolution trends, this application achieves continuous quantitative assessment and graded early warning of highway congestion levels, and can also take targeted control measures in conjunction with these measures, demonstrating high application value. The traffic status data continuously accumulated during system operation can be used for offline analysis and model upgrades, continuously improving the intelligence level of congestion detection and management.

[0152] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0153] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0154] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0155] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0156] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0157] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An adaptive prediction method for highway congestion levels, characterized in that, include: The traffic environment perception parameters are obtained by collecting multi-dimensional traffic state characteristics, environmental sensing quality parameters, and historical traffic state statistical characteristics corresponding to the preset road perception area on the highway; the environmental sensing quality parameters include at least sensor reliability indicators. The adaptive weighting coefficients corresponding to the multi-dimensional traffic state features are obtained based on the traffic environment perception parameters, and the multi-dimensional traffic state features are weighted and fused based on the adaptive weighting coefficients to obtain the basic congestion index. The adaptive weighting coefficients corresponding to the multi-dimensional traffic state features are obtained based on the traffic environment perception parameters as follows: ;in, This represents the adaptive weighted coefficient vector corresponding to the multi-dimensional traffic state at time t, which contains the adaptive weighted coefficients corresponding to each traffic state feature. This represents the weight generation function corresponding to the adaptive weighting coefficients. This represents the multi-dimensional traffic state characteristics at time t. This represents the environmental sensing quality parameter at time t. This represents the statistical characteristics of historical traffic conditions at time t; The adaptive weighted coefficient vector is: ; The i-th weight coefficient in the adaptive weighted coefficient vector is: Among them, superscript The transpose operation represents a vector, used to convert a row vector into a column vector; Indicates time Upper The weight coefficients corresponding to each traffic state feature are i-th weight coefficients in the adaptive weighted coefficient vector. Indicates time Upper Multi-dimensional traffic condition characteristics ; Indicates time Upper A quality parameter of environmental sensing. ; Indicates time Above and below Historical traffic state statistical features corresponding to each traffic state feature; Indicates the first The influence coefficients corresponding to the environmental sensing quality parameters; These represent the preset first adjustment coefficient, second adjustment coefficient, and third adjustment coefficient, respectively. These are used to control the degree of influence of traffic state characteristics, environmental perception parameters, and historical statistical characteristics on the weight calculation, respectively. Indicates the number of multi-dimensional traffic state features; This indicates the number of environmental sensing quality parameters; A sliding time window is introduced to model and analyze the changes of the basic congestion index over multiple consecutive detection periods, obtaining the temporal evolution characteristics of congestion states, including: constructing a length of... The sliding time window, that is, the sliding time window allows for the inclusion of... The basic congestion index corresponding to each detection period is obtained; the rate of change of the basic congestion index within the sliding time window is obtained to obtain the growth rate feature; the cumulative change of the basic congestion index within the sliding time window is obtained to obtain the cumulative trend feature; the short-term difference index of the basic congestion index within the sliding time window is obtained to obtain the abrupt change feature; the growth rate feature, the cumulative trend feature, and the abrupt change feature are used together as the congestion state evolution feature that changes over time. The short-time difference index of the basic congestion index within the sliding time window is obtained, and the abrupt change characteristics are as follows: ;in, This represents the abrupt change characteristic at time t. This represents the basic congestion index at time t. Indicates time The basic congestion index on the internet, This represents the short time difference time interval, which is a preset time step used to characterize the magnitude of changes in traffic conditions within a short time scale. Based on the congestion state evolution characteristics and the basic congestion index, congestion prediction is performed on the road sensing area to obtain the final congestion index: ;in, This represents the final congestion index at time t. This represents the basic congestion index at time t. This represents the cumulative trend characteristics at time t. This represents the growth rate characteristic at time t. This represents the abrupt change characteristic at time t. This represents the weighting coefficient corresponding to the basic congestion index. This represents the weighting coefficient corresponding to the growth rate feature. This represents the weighting coefficient corresponding to the cumulative trend feature. This represents the weighting coefficient corresponding to the mutation feature; Based on the final congestion index, the K-means clustering algorithm is used to divide the optimal decision boundary and predict congestion levels of highways.

2. The adaptive prediction method for highway congestion levels according to claim 1, characterized in that, Following the prediction of highway congestion levels, the following is also included: Based on the highway congestion level, find the corresponding early warning strategy and traffic control strategy, and implement the early warning strategy and traffic control strategy.

3. The adaptive prediction method for highway congestion levels according to claim 1, characterized in that, Multi-dimensional traffic state characteristics, environmental sensing quality parameters, and historical traffic state statistical characteristics are collected from pre-defined road sensing areas on highways to obtain traffic environment sensing parameters, including: The system collects the total number of vehicles in the preset road perception area on the highway, the proportion of stationary and driving vehicles with speeds below the preset speed threshold, the average headway or average inter-vehicle distance, the proportion of the total projected area of ​​vehicles to the area of ​​the road perception area, and the dispersion characteristics of multi-lane traffic flow to obtain multi-dimensional traffic state characteristics. Collect weather conditions, lighting conditions, road conditions, and environmental sensor data corresponding to the road sensing area to obtain environmental sensing quality parameters; Collect typical operating parameters, historical congestion patterns, and characteristic parameters from previous data corresponding to the road sensing area to obtain historical traffic status statistical characteristics; Multi-dimensional traffic condition characteristics, environmental sensing quality parameters, and historical traffic condition statistical characteristics are used together as traffic environment sensing parameters.

4. The adaptive prediction method for highway congestion levels according to claim 1, characterized in that, The multi-dimensional traffic state features are weighted and fused based on the adaptive weighting coefficients to obtain a basic congestion index, including: The adaptive weighting coefficients are normalized to obtain the normalized adaptive weighting coefficients. The traffic state features in the multi-dimensional traffic state features are normalized to obtain the normalized traffic state features. The normalized traffic state features are weighted and fused using adaptive weighting coefficients, resulting in the basic congestion index: ； in, This represents the basic congestion index at time t. This represents the adaptive weighting coefficient after the i-th normalization process. This represents the traffic state characteristics after the i-th normalization process at time t.

5. The adaptive prediction method for highway congestion levels according to claim 1, characterized in that, The rate of change of the basic congestion index within the sliding time window is obtained, and the growth rate characteristic is as follows: ； in, This represents the growth rate characteristic at time t. This represents the basic congestion index at time t. Indicates time The basic congestion index on the internet, Indicates the preset time window length.

6. The adaptive prediction method for highway congestion levels according to claim 5, characterized in that, The cumulative change of the basic congestion index within the sliding time window is obtained, and the cumulative trend characteristics are as follows: ； in, This represents the cumulative trend characteristics at time t. Indicates time The basic congestion index on the internet, Represents the integral variable, representing any historical moment within the sliding time window; Represents the time variable The derivative of is used to represent the continuous integral over the time interval. Indicates the preset time window length and integration interval. This represents a sliding time window ending at the current time t.

Images