A Highway Traffic Operation Status Determination Method Based on Ensemble Learning

Abstract

This invention discloses a highway traffic operation status discrimination method based on ensemble learning, belonging to the field of intelligent transportation technology. This method collects vehicle traffic data through ETC antennas deployed at ramp entrances and calculates the vehicle density difference ΔQin using statistical methods and a time-series difference function to assess periods of high vehicle density. It predicts vehicle speeds based on a vehicle speed prediction model and calculates a false traffic flow judgment index Fin using the vehicle speed false traffic flow residual to assess road traffic flow. When a false traffic flow state is identified, a quadratic model adjustment training and a sliding trend index Hres are used to assess the misjudgment trend. The vehicle speed prediction model is optimized through a sliding window mechanism and statistical analysis to dynamically adjust the judgment logic for false traffic flow states. This invention improves the real-time performance and accuracy of traffic status assessment and provides a flexible solution under varying traffic flow conditions.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, specifically to a method for determining the traffic operation status of highways based on ensemble learning. Background Technology

[0002] Transportation systems are a core component of modern urban development, significantly impacting sustainable economic, social, and environmental progress. With the continuous increase in vehicle numbers, traditional traffic monitoring and management methods are showing signs of inefficiency and slow response, especially on highways where high traffic volume and speed necessitate real-time monitoring and assessment of traffic conditions. To promptly identify and address potential traffic congestion and accidents on highways, highway traffic status assessment has emerged. Traditional assessment methods often rely on simple rules and manual analysis, which are ill-suited to reflecting the complex traffic dynamics on highways. The highway traffic status assessment method based on ensemble learning leverages the advantages of multi-model ensembles, achieving high-precision real-time assessment of traffic conditions through comprehensive analysis of multi-dimensional data such as vehicle flow and speed, providing decision-making support for traffic management departments.

[0003] Traditional highway traffic condition monitoring methods are often limited to single data sources and simple rule-based judgments, making them unsuitable for comprehensively analyzing the complexity and dynamic changes of traffic flow. Especially under conditions of high-density traffic flow, adverse weather, or sudden events, traditional methods may struggle to accurately and in real-time assess traffic conditions, leading to delayed traffic management decisions and hindering timely intervention. Many current traffic condition methods rely on fixed thresholds for judgment, neglecting the nonlinear characteristics and complex behavioral patterns of highway traffic. Furthermore, traditional methods often lack flexible adaptive capabilities during model training and tuning, making it difficult to effectively respond to changes under varying traffic conditions. Ensemble learning methods, through joint judgments from multiple models, can fully leverage the advantages of different algorithms, overcoming the limitations of single algorithms, thereby achieving more accurate highway traffic condition assessment and effectively addressing the data bias and slow response issues of traditional methods. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a highway traffic operation status discrimination method based on ensemble learning, which solves the problems mentioned in the background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for determining the traffic operation status of highways based on ensemble learning, comprising the following steps: S1. Collect vehicle traffic data, process the data to obtain a traffic record set, and calculate the vehicle density difference ΔQin based on statistical methods and time series difference functions to evaluate the dense vehicle period. Trigger mainline traffic analysis based on the evaluation results. S2. After obtaining the mainline traffic analysis command, collect the vehicle traffic speed data, perform preprocessing to obtain the measured vehicle speed Vm on the mainline, and build a vehicle speed prediction model. Output the predicted speed Vy on the mainline through the vehicle speed prediction model. S3. Calculate the false smooth traffic residual Δres by comparing the measured speed Vm on the main line with the predicted speed Vy on the main line, and construct the false smooth traffic judgment index Fin to assess road smoothness. S4. When the road traffic flow assessment indicates that the current traffic condition is a false smooth flow condition, the vehicle speed prediction model is trained by a secondary model adjustment. The false smooth flow judgment index Fin of the false smooth flow condition is input into the trained vehicle speed prediction model, and the slip trend index Hres is constructed to assess the misjudgment trend.

[0006] Preferably, S1 includes S11; S11. Collect vehicle passage data within the time window by deploying ETC antennas at the ramp entrance, and record the timestamp, passage device number, passage status identifier and unique vehicle identifier. Perform data preprocessing on vehicle traffic data to obtain a traffic record set; The data preprocessing includes data cleaning and reordering, as well as standardization. The data cleaning and reordering process is used to remove duplicate passage data recorded within the same time window, correct time format errors, missing fields, and abnormal vehicle IDs, and reorder the records according to the timestamp. The standardization process is used to merge continuous traffic data into a standard time period using a sliding window mechanism.

[0007] Preferably, S1 further includes S12; S12. Calculate the number of vehicles passing through the time window in the traffic record set according to the statistical method, and compare it with the collection duration of the time window to obtain the vehicle traffic density Qin. Use the time difference function to calculate the vehicle density difference ΔQin between adjacent time windows. When the vehicle density difference ΔQin ≤ the standard vehicle difference threshold Qt, the current time interval is marked as a normal driving period. No intervention is performed at this time, and normal monitoring is maintained. When the vehicle density difference ΔQin > the standard vehicle difference threshold Qt, the current time interval is marked as a dense vehicle period, and the mainline traffic analysis is triggered at this time. The standard vehicle differential threshold Qt acquisition process is as follows: extract the vehicle traffic density Qin at the ramp entrances within the past 30 days, smooth it using an exponentially weighted moving average method, and then use the percentile method to set the 95th percentile of the smoothed vehicle traffic density Qin as the standard vehicle differential threshold Qt.

[0008] Preferably, S2 includes S21; S21. After obtaining the mainline traffic analysis instruction, collect vehicle speed data based on the radar speed measurement equipment deployed on the mainline lanes, and mark the collection timestamp and spatial location. Divide the time window into morning peak, off-peak, evening peak and nighttime periods according to the time interval to which the timestamp belongs, and use the corresponding category as time period data. Based on the GIS vector topology layer of the road structure, the vehicle merging points of the ramps are mapped to the locations of the main line speed measurement points according to the location numbering rules, forming a correspondence between the entrance and the main line. The traffic speed data of each speed measurement point within the time window is extracted, and the traffic speed data is preprocessed to obtain the measured vehicle speed Vm on the main line. The preprocessing includes missing value imputation and deviation value removal; The missing value imputation is used to fill in the traffic speed data with missing values ​​according to the linear interpolation method, and the deviation value removal is used to remove the traffic speed data that deviates from the mean by more than three times the standard deviation.

[0009] Preferably, S2 further includes S22; S22. Extract the vehicle traffic density Qin and the measured speed Vm of each speed measurement point in each time window within the past month. Then, use timestamp alignment technology to align the vehicle traffic density Qin, the measured speed Vm of the main line and the time feature data to generate a training sample set. The training sample set includes an input sample set and a speed sample set; the input sample set includes vehicle traffic density Qin and time feature data; the speed sample set includes measured vehicle speed Vm on the main line; Using gradient boosting decision tree as the regression algorithm, the input sample set is used as the model input feature and the speed sample set is used as the supervision label to construct the vehicle speed prediction model. The input sample set is divided into training subset and validation subset and input into the vehicle speed prediction model. The model is trained using five-fold cross-validation. After training, the real-time vehicle traffic density Qin and time feature data are input into the vehicle speed prediction model. The vehicle speed prediction model outputs the main line predicted speed Vy.

[0010] Preferably, S3 includes S31; S31. Based on the sliding window alignment algorithm, the measured speed Vm and the predicted speed Vy of the main line are uniformly mapped to the same time sampling period, and then converted to the same time axis through time resampling. The difference between the measured speed Vm and the predicted speed Vy of the main line is calculated to obtain the false smooth flow residual Δres of the speed, which reflects the degree of deviation between the actual running state of the main line and the theoretical predicted state under the current dense traffic conditions of the ramps.

[0011] Preferably, S3 further includes S32; S32. Based on the fitting results of vehicle density difference ΔQin, standard vehicle difference threshold Qt, and vehicle speed false smooth flow residual Δres, the smooth flow status of highway vehicles is analyzed and evaluated, specifically including S321 and S322. S321. Compare the vehicle density difference ΔQin with the standard vehicle difference threshold Qt, and extract the minimum value Δres from the set of false smooth traffic residuals Δres during historical normal driving periods. min A comprehensive calculation is performed to construct a false smooth traffic index Fin, which reflects the coupling relationship between the traffic density of the entrance ramps and the mainline operation status, as follows; ; Wherein, Fin(t) represents the false accessibility index Fin at time t.

[0012] Preferably, in step S322, the false smooth traffic judgment index Fin is extracted from the past month under normal vehicle traffic conditions, and the mean is calculated according to the statistical method. The mean is set as the traffic discrimination threshold Rx, and then compared with the false smooth traffic judgment index Fin obtained in real time to assess the smooth traffic of the current highway ramp. When the false smooth traffic index Fin ≤ the traffic discrimination threshold Rx, it means that the current traffic state is a load-balanced state with no abnormal traffic bottlenecks or congestion risks. At this time, the current ramp is marked as a normal smooth traffic state and is monitored normally. When the false traffic flow judgment index Fin > the traffic discrimination threshold Rx, it indicates that the current traffic state is a false traffic flow state. At this time, the current ramp is marked as a false traffic flow state and error correction is performed.

[0013] Preferably, S4 includes S41 and S42; S41. When the road traffic flow assessment indicates that the current traffic state is a false smooth state, record the current vehicle traffic density Qin, the measured vehicle speed Vm on the main line, time feature data, and the false smooth speed residual Δres as misjudged samples and input them into the training sample pool of the vehicle speed prediction model. Use the false smooth speed residual Δres in the false smooth state as a reward or penalty signal in reinforcement learning, and introduce a learning rate adjustment mechanism to gradually adjust the learning weight for the false smooth state during the training process, and perform secondary model adjustment training on the vehicle speed prediction model. S42. After the vehicle speed prediction model is trained, the false smooth flow judgment index Fin for false smooth flow conditions is input into the vehicle speed prediction model to construct the slip trend index Hres, which is used to measure the cumulative error trend of false smooth flow conditions over multiple time periods, as follows. ; Where Hres(t) represents the sliding trend index at time t, w represents the number of time periods, i represents the first i time periods within the time window, and Fin(ti) represents the false smooth flow judgment index at time ti.

[0014] Preferably, S4 further includes S43; S43. Calculate the mean and standard deviation of the historical sliding trend index Hres according to the statistical method, and set the sum of the mean and standard deviation as the critical threshold Wy for misjudged trend. Then, evaluate the misjudged trend with the real-time obtained sliding trend index Hres, as follows. When the sliding trend index Hres ≤ the critical threshold Wy for misjudgment trend, it means that the misjudgment is within the normal fluctuation range. At this time, the current state is maintained and the traffic status is assessed according to the original judgment logic. When the slip trend index Hres is greater than the critical threshold Wy for misjudgment trend, it indicates that there is a misjudgment of the false smooth flow state. At this time, the slip trend index Hres at the time of the misjudgment is fed back to the vehicle speed prediction model, and the vehicle speed prediction model is corrected by weighted training, and iterative analysis is performed through S1.

[0015] This invention provides a method for determining the traffic operation status of highways based on ensemble learning. It has the following beneficial effects: (1) This method utilizes ETC antennas deployed at ramp entrances to efficiently collect large amounts of vehicle traffic data. Data preprocessing techniques are used for cleaning, sorting, and standardization, resolving the misjudgment problem caused by unstable data quality in traditional methods. Next, the method calculates the vehicle density difference ΔQin using statistical methods and time-series difference functions, accurately assessing vehicle traffic conditions and successfully identifying peak traffic periods. Subsequent speed prediction models analyze mainline vehicle speed, achieving joint monitoring of traffic flow density and vehicle speed, avoiding bias from a single indicator. This method not only improves the real-time performance and accuracy of the data but also automatically identifies and triggers relevant traffic analyses, optimizing the efficiency of highway traffic management.

[0016] (2) This method overcomes the limitations of traditional traffic flow monitoring by introducing the innovative false smoothness judgment index Fin and the sliding trend index Hres. False smoothness is a common problem in highway traffic flow management. Traditional methods often rely on static data of vehicle speed or traffic density to judge traffic status, easily overlooking potential traffic pressure or congestion risks. This invention uses the difference between predicted and measured vehicle speed (i.e., the false smoothness residual Δres) as the key indicator for judging false smoothness, combined with the false smoothness index Fin to evaluate real-time traffic status, effectively distinguishing between false smoothness and true smoothness. In addition, the sliding trend index Hres dynamically adjusts the error accumulation trend over multiple time periods, solving the problem of lagging traffic flow trend identification in traditional traffic assessment methods, and significantly improving the early warning capability and response speed of traffic flow. This series of innovative technologies enables this method to accurately identify and handle abnormal traffic conditions in high-density and high-fluctuation traffic scenarios.

[0017] (3) This method combines the vehicle speed prediction model with the identification of false traffic flow conditions, achieving higher accuracy in traffic prediction and intelligent decision support. Compared with existing technologies, this method introduces a secondary model adjustment training and sliding trend evaluation mechanism. Through real-time feedback and weighted training, the model can quickly adjust when false traffic flow conditions occur, avoiding judgment errors caused by model rigidity in traditional methods. In addition, this method uses verification methods such as five-fold cross-validation to ensure the generalization ability of the vehicle speed prediction model, making the performance of the vehicle speed prediction model more stable and reliable under different traffic conditions. Especially in the process of road traffic flow assessment, by comparing and analyzing the false traffic flow judgment index Fin with the vehicle speed residual set during historical normal driving periods, the identification of traffic conditions is further strengthened, and the level of intelligent traffic management is improved. Compared with traditional single vehicle speed or density monitoring methods, the comprehensive evaluation mechanism of this invention can identify potential risks of traffic congestion in a timely manner, and avoid misjudgments caused by rapid traffic changes through dynamic adjustment of the sliding trend index Hres, thereby improving the reaction speed and adaptability of the traffic management system. Attached Figure Description

[0018] Figure 1 This is a schematic diagram illustrating the steps of a highway traffic operation status discrimination method based on ensemble learning according to the present invention. Figure 2 This is a block diagram illustrating the logical principle of a highway traffic operation status discrimination method based on ensemble learning according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1 Please see Figure 1 This invention provides a method for determining the traffic operation status of highways based on ensemble learning. To achieve the above objectives, this invention is implemented through the following technical solution, including the following steps: S1. Collect vehicle traffic data, process the data to obtain a traffic record set, and calculate the vehicle density difference ΔQin based on statistical methods and time series difference functions to evaluate the dense vehicle period. Trigger mainline traffic analysis based on the evaluation results. S2. After obtaining the mainline traffic analysis command, collect the vehicle traffic speed data, perform preprocessing to obtain the measured vehicle speed Vm on the mainline, and build a vehicle speed prediction model. Output the predicted speed Vy on the mainline through the vehicle speed prediction model. S3. Calculate the false smooth traffic residual Δres by comparing the measured speed Vm on the main line with the predicted speed Vy on the main line, and construct the false smooth traffic judgment index Fin to assess road smoothness. S4. When the road traffic flow assessment indicates that the current traffic condition is a false smooth flow condition, the vehicle speed prediction model is trained by a secondary model adjustment. The false smooth flow judgment index Fin of the false smooth flow condition is input into the trained vehicle speed prediction model, and the slip trend index Hres is constructed to assess the misjudgment trend.

[0021] In this embodiment, by introducing vehicle density difference ΔQin in S1, the changes in traffic flow at different times are quantitatively characterized. This allows the system to no longer rely solely on instantaneous flow or average speed, but to identify the evolution trend of traffic conditions, distinguishing between normal traffic and potential congestion from the source. This process enables automatic identification of periods of high vehicle density and triggers the mainline analysis mechanism when necessary, providing clear triggering conditions for subsequent calculations and avoiding the resource waste and judgment noise problems caused by indiscriminate, all-time calculations in traditional methods. S2 elevates traffic operation status from apparent vehicle speed judgment to a multi-layered discrimination mode of prediction, deviation, and verification. It not only obtains the measured vehicle speed Vm on the mainline but also obtains the predicted vehicle speed Vy through training with historical features. S3 constructs a false smooth flow residual Δres based on the deviation between the measured vehicle speed Vm and the predicted speed Vy on the mainline, identifying traffic conditions that appear smooth but are internally strained. Furthermore, it combines the vehicle density difference ΔQin to construct a false smooth flow judgment index Fin, so that traffic condition assessment no longer relies on a single indicator but is judged through the coupling relationship between density changes and speed deviations. Compared to traditional methods relying on fixed thresholds or single vehicle speed judgments, this method can identify potential congestion earlier, avoiding misjudgments of traffic conditions before vehicle speeds have significantly decreased. In S4, a secondary training and sliding trend evaluation mechanism based on misjudgment feedback is introduced, enabling the vehicle speed prediction model to have adaptive correction capabilities. By back-inputting the judgment results under falsely smooth traffic conditions into the model and constructing a sliding trend index Hres, misjudgment behavior is continuously tracked and accumulated, allowing the model to dynamically adjust with changes in traffic characteristics, rather than relying on fixed parameters in the long term. This mechanism compensates for the lack of stability in fluctuating traffic conditions, enabling the vehicle speed prediction model to maintain continuous and evolvable judgment capabilities in complex traffic scenarios. Overall, this method completes the transformation from static rule-based judgment to dynamic learning and trend perception, achieving significant improvements in the accuracy, foresight, and stability of traffic state recognition, providing more practical technical support for highway operation and management.

[0022] Example 2 Please refer to Figure 2 Specifically: S1 includes S11; S11. Collect vehicle passage data within the time window by deploying ETC antennas at the ramp entrance, and record the timestamp, passage device number, passage status identifier (entry / exit) and unique vehicle identifier. Perform data preprocessing on vehicle traffic data to obtain a traffic record set; The data preprocessing includes data cleaning and reordering, as well as standardization. The data cleaning and reordering process is used to remove duplicate passage data recorded within the same time window, correct time format errors, missing fields, and abnormal vehicle IDs, and reorder the records according to the timestamp. The standardization process is used to merge continuous traffic data into a standard time period using a sliding window mechanism.

[0023] S1 further includes S12; S12. Calculate the number of vehicles passing through the time window in the traffic record set according to the statistical method, and compare it with the collection duration of the time window to obtain the vehicle traffic density Qin. Use the time difference function to calculate the vehicle density difference ΔQin between adjacent time windows. When the vehicle density difference ΔQin ≤ the standard vehicle difference threshold Qt, the current time interval is marked as a normal driving period. No intervention is performed at this time, and normal monitoring is maintained. When the vehicle density difference ΔQin > the standard vehicle difference threshold Qt, the current time interval is marked as a dense vehicle period, and the mainline traffic analysis is triggered at this time. The standard vehicle differential threshold Qt acquisition process is as follows: extract the vehicle traffic density Qin at the ramp entrances within the past 30 days, smooth it using an exponentially weighted moving average method, and then use the percentile method to set the 95th percentile of the smoothed vehicle traffic density Qin as the standard vehicle differential threshold Qt.

[0024] In this embodiment, ETC antennas are deployed at ramp entrances to continuously collect vehicle traffic data. Data cleaning, reordering, and standardization are performed by combining timestamps, traffic direction, and vehicle identification, transforming the original traffic data into a structurally consistent and temporally continuous traffic record set. Based on this, the number of vehicles passing through the time window in the traffic record set is calculated using statistical methods, and the vehicle density difference ΔQin is calculated using a time-series differential method to reflect changes in traffic load within adjacent time periods. This is then combined with a standard vehicle difference threshold Qt constructed based on historical data to achieve segmented identification of traffic status. When the vehicle density difference ΔQin is less than or equal to the standard vehicle difference threshold Qt, the monitoring status is maintained; when the vehicle density difference ΔQin is greater than the standard vehicle difference threshold Qt, the subsequent mainline traffic analysis process is automatically triggered, forming a continuous processing link consisting of data normalization, density characterization, and status determination. This implementation method eliminates the reliance on traffic volume at a single moment for traffic status assessment. Instead, it analyzes traffic evolution characteristics to ensure continuity and traceability in traffic status identification. It also reduces interference from irrelevant time periods in subsequent calculations, providing reliable input conditions for subsequent mainline speed analysis and false traffic flow detection. This results in a more reasonable processing effect in terms of temporal integrity and logical coherence of the overall traffic operation assessment process.

[0025] Example 3 Please refer to Figure 2 Specifically: S2 includes S21; S21. After obtaining the mainline traffic analysis instruction, collect vehicle speed data based on the radar speed measurement equipment deployed on the mainline lanes, and mark the collection timestamp and spatial location. Divide the time window into morning peak, off-peak, evening peak and nighttime periods according to the time interval to which the timestamp belongs, and use the corresponding category as time period data. Based on the GIS vector topology layer of the road structure, the vehicle merging points of the ramps are mapped to the locations of the main line speed measurement points according to the location numbering rules, forming a correspondence between the entrance and the main line. The traffic speed data of each speed measurement point within the time window is extracted, and the traffic speed data is preprocessed to obtain the measured vehicle speed Vm on the main line. The preprocessing includes missing value imputation and deviation value removal; The missing value imputation is used to fill in the traffic speed data with missing values ​​according to the linear interpolation method, and the deviation value removal is used to remove the traffic speed data that deviates from the mean by more than three times the standard deviation.

[0026] S2 further includes S22; S22. Extract the vehicle traffic density Qin and the measured speed Vm of each speed measurement point in each time window within the past month. Then, use timestamp alignment technology to align the vehicle traffic density Qin, the measured speed Vm of the main line and the time feature data to generate a training sample set. The training sample set includes an input sample set and a speed sample set; the input sample set includes vehicle traffic density Qin and time feature data; the speed sample set includes measured vehicle speed Vm on the main line; Using gradient boosting decision tree as the regression algorithm, the input sample set is used as the model input feature and the speed sample set is used as the supervision label to construct the vehicle speed prediction model. The input sample set is divided into training subset and validation subset and input into the vehicle speed prediction model. The model is trained using five-fold cross-validation. After training, the real-time vehicle traffic density Qin and time feature data are input into the vehicle speed prediction model. The vehicle speed prediction model outputs the main line predicted speed Vy.

[0027] In this embodiment, vehicle speeds are collected by radar speed measurement devices deployed on the main line. Time periods and road structure mapping are then performed by combining timestamps and spatial locations, establishing a clear correspondence between ramp entrances and main line speed measurement points, thus constructing a spatiotemporally consistent vehicle speed data foundation. Based on this, missing data is repaired and anomalies are removed from the original speed data to ensure the continuity and availability of the input data. Furthermore, a training sample set is formed by combining the vehicle traffic density Qin (over one month) and the measured main line speed Vm through time alignment. A gradient boosting decision tree is used to establish a vehicle speed prediction model, reflecting the operational characteristics under different time periods and traffic flow conditions. During real-time operation, the current vehicle traffic density Qin and time feature data are input into the vehicle speed prediction model to obtain the predicted main line speed Vy, which is compared with the measured main line speed Vm, providing a foundation for subsequent false traffic flow identification. This process achieves a complete closed loop from data acquisition, spatiotemporal mapping, sample construction to model output, enabling traffic state analysis to no longer rely on a single vehicle speed or experience threshold, but to make joint judgments based on historical patterns and real-time characteristics, forming analysis results that conform to actual operational characteristics in terms of the continuity, interpretability, and adaptability of traffic state identification.

[0028] Example 4 Please refer to Figure 2 Specifically: S3 includes S31; S31. Based on the sliding window alignment algorithm, the measured speed Vm and the predicted speed Vy of the main line are uniformly mapped to the same time sampling period, and then converted to the same time axis through time resampling. The difference between the measured speed Vm and the predicted speed Vy of the main line is calculated to obtain the false smooth flow residual Δres of the speed, which reflects the degree of deviation between the actual running state of the main line and the theoretical predicted state under the current dense traffic conditions of the ramps.

[0029] S3 further includes S32; S32. Based on the fitting results of vehicle density difference ΔQin, standard vehicle difference threshold Qt, and vehicle speed false smooth flow residual Δres, the smooth flow status of highway vehicles is analyzed and evaluated, specifically including S321 and S322. S321. Compare the vehicle density difference ΔQin with the standard vehicle difference threshold Qt to reflect the degree of similarity between the current ramp traffic status and the historical high-density status, and extract the minimum value Δres from the set of false smooth traffic residuals Δres from the historical normal driving period. min A comprehensive calculation is performed to construct a false smooth traffic index Fin, which reflects the coupling relationship between the traffic density of the entrance ramps and the mainline operation status, as follows; ; Wherein, Fin(t) represents the false accessibility index Fin at time t.

[0030] S322. Extract the false smooth traffic judgment index Fin from the past month's vehicle traffic under normal conditions, calculate the mean according to the statistical method, set the mean as the traffic discrimination threshold Rx, and then compare it with the false smooth traffic judgment index Fin obtained in real time to conduct a road smoothness assessment of the current highway ramp. When the false smooth traffic index Fin ≤ the traffic discrimination threshold Rx, it means that the current traffic state is a load-balanced state with no abnormal traffic bottlenecks or congestion risks. At this time, the current ramp is marked as a normal smooth traffic state and is monitored normally. When the false smooth traffic index Fin > the traffic discrimination threshold Rx, it indicates that the current traffic state is a false smooth traffic state, that is, traffic pressure has formed at the entrance, but it has not yet been reflected in the main line speed. At this time, the current ramp is marked as a false smooth traffic state and error correction is performed.

[0031] In this embodiment, by using sliding window alignment and time resampling, the measured speed Vm and the predicted speed Vy on the main line are unified to the same time scale, eliminating interference caused by inconsistent sampling frequencies. This ensures that the difference between the measured speed Vm and the predicted speed Vy accurately reflects the deviation between the main line's operating state and the theoretically predicted state. Based on this, a false traffic flow residual Δres is constructed. The vehicle density difference ΔQin, the standard vehicle difference threshold Qt, and the false traffic flow residual Δres are combined to construct the false traffic flow judgment index Fin, which characterizes the correlation between ramp traffic pressure and the main line's operating state. This allows traffic state judgment to no longer rely solely on a single speed indicator but simultaneously reflect the intrinsic relationship between flow rate changes and speed response. In this formula... The derivation is based on the normalized expression of the relative rate of change. In this formula, ΔQin represents the change in vehicle density of the entrance ramp within an adjacent time window, reflecting the rate of change in traffic pressure. Qt is derived from the statistical upper bound of historical high-density states and is used to characterize the acceptable range of density change of the system under historical conditions. The ratio of the two represents the relative position of the current ramp density change in the historical high-density state scale, corresponding to the normalized expression of the relative rate of change. The derivation is based on the residual consistency metric function. Where e represents the deviation between the actual observed value and the model prediction, e ref The minimum speed residual benchmark value under historical normal traffic conditions is used. In this formula, Δres is the residual between the measured speed and the predicted speed at the current moment, representing the actual deviation of the speed response. min The minimum residual under normal historical conditions represents the minimum deviation from the baseline that the main line can achieve under stable conditions. The "1 - ratio" structure maps the "degree of deviation" to the "degree of consistency", which facilitates coupling with the density term and corresponds to the residual consistency measurement function. and A false traffic flow index, Fin, is constructed through multiplication. Furthermore, by introducing the false traffic flow index Fin from historical normal conditions as a reference benchmark and calculating its mean using statistical methods, this mean is set as the traffic discrimination threshold Rx. The real-time calculation results are compared with the traffic discrimination threshold Rx to identify superficially smooth traffic but internally stressed conditions, thus distinguishing between normal traffic and false traffic flow. This allows the scheme to identify potential traffic risks before vehicle speeds significantly decrease. Through these steps, the overall process forms a complete judgment chain from data alignment and state characterization to threshold discrimination, shifting the analysis of traffic operation status from static description to a judgment method based on evolutionary trends. Without relying on human experience, it achieves continuous characterization of the relationship between ramp pressure and mainline operation, providing a clear basis for subsequent regulation and model correction.

[0032] Example 5 Please refer to Figure 2 Specifically: S4 includes S41 and S42; S41. When the road traffic flow assessment indicates that the current traffic state is a false smooth state, record the current vehicle traffic density Qin, the measured vehicle speed Vm on the main line, time feature data, and the false smooth speed residual Δres as misjudged samples and input them into the training sample pool of the vehicle speed prediction model. Use the false smooth speed residual Δres in the false smooth state as a reward or penalty signal in reinforcement learning, and introduce a learning rate adjustment mechanism to gradually adjust the learning weight for the false smooth state during the training process, and perform secondary model adjustment training on the vehicle speed prediction model. S42. After the vehicle speed prediction model is trained, the false smooth flow judgment index Fin for false smooth flow conditions is input into the vehicle speed prediction model to construct the slip trend index Hres, which is used to measure the cumulative error trend of false smooth flow conditions over multiple time periods, as follows. ; Where Hres(t) represents the sliding trend index at time t, w represents the number of time periods, i represents the first i time periods within the time window, and Fin(ti) represents the false smooth flow judgment index at time ti.

[0033] S4 also includes S43; S43. Calculate the mean and standard deviation of the historical sliding trend index Hres according to the statistical method, and set the sum of the mean and standard deviation as the critical threshold Wy for misjudged trend. Then, evaluate the misjudged trend with the real-time obtained sliding trend index Hres, as follows. When the sliding trend index Hres ≤ the critical threshold Wy for misjudgment trend, it means that the misjudgment is within the normal fluctuation range. At this time, the current state is maintained and the traffic status is assessed according to the original judgment logic. When the slip trend index Hres is greater than the critical threshold Wy for misjudgment trend, it indicates that there is a misjudgment of the false smooth flow state. At this time, the slip trend index Hres at the time of the misjudgment is fed back to the vehicle speed prediction model, and the vehicle speed prediction model is corrected by weighted training, and iterative analysis is performed through S1.

[0034] In this embodiment, when it is determined that the current state is a false smooth flow, the vehicle traffic density Qin, the measured speed Vm on the main line, the time feature data, and the false smooth flow residual Δres corresponding to the current moment are all included in the training sample pool. The false smooth flow residual Δres is used as a feedback signal to participate in the retraining of the vehicle speed prediction model, thus introducing response constraints to abnormal states into the original prediction structure. Simultaneously, by accumulating the false smooth flow judgment index Fin over time, a sliding trend index Hres reflecting the evolution of misjudgments is constructed to characterize the changing pattern of false smooth flow over continuous time periods. The derivation of this formula is based on the moving average operator in discrete time series analysis. In time series theory, a moving average of length w is defined for a discrete sequence x(t), and its standard form is: The aim is to characterize the cumulative level of a sequence over time. Its core mathematical meaning is to describe the overall trend of the current moment using historical samples within a finite time window, rather than depicting single-point fluctuations. Based on this, a critical threshold Wy for misjudgment trends is formed using historical statistical results as the criterion. When the sliding trend index Hres exceeds the critical threshold Wy, the current deviation is fed back to the model, triggering a weighted correction and iterative analysis process, allowing the model parameters to dynamically adjust with changes in traffic conditions. Through this approach, an operational structure centered on false traffic flow identification, cumulative misjudgment analysis, and adaptive model correction is formed. This ensures that traffic condition judgment no longer relies on single calculations but continuously corrects the prediction logic based on temporal changes, thereby supporting continuous discrimination and dynamic analysis of operational states in complex traffic scenarios and meeting the application requirements of highway operation management for long-term stable discrimination and trend identification.

[0035] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended technical solutions and their equivalents.

Claims

1. A method for determining the traffic operation status of highways based on ensemble learning, characterized in that: Includes the following steps: S1. Collect vehicle traffic data, process the data to obtain a traffic record set, and calculate the vehicle density difference ΔQin based on statistical methods and time series difference functions to evaluate the dense vehicle period. Trigger mainline traffic analysis based on the evaluation results. S2. After obtaining the mainline traffic analysis command, collect the vehicle traffic speed data, perform preprocessing to obtain the measured vehicle speed Vm on the mainline, and build a vehicle speed prediction model. Output the predicted speed Vy on the mainline through the vehicle speed prediction model. S3. Calculate the false smooth traffic residual Δres by comparing the measured speed Vm on the main line with the predicted speed Vy on the main line, and construct the false smooth traffic judgment index Fin to assess road smoothness. S4. When the road traffic flow assessment indicates that the current traffic condition is a false smooth flow condition, the vehicle speed prediction model is trained by a secondary model adjustment. The false smooth flow judgment index Fin of the false smooth flow condition is input into the trained vehicle speed prediction model, and the slip trend index Hres is constructed to assess the misjudgment trend.

2. The method for determining highway traffic operation status based on ensemble learning according to claim 1, characterized in that: S1 includes S11; S11. Collect vehicle passage data within the time window by deploying ETC antennas at the ramp entrance, and record the timestamp, passage device number, passage status identifier and unique vehicle identifier. Perform data preprocessing on vehicle traffic data to obtain a traffic record set; The data preprocessing includes data cleaning and reordering, as well as standardization. The data cleaning and reordering process is used to remove duplicate passage data recorded within the same time window, correct time format errors, missing fields, and abnormal vehicle IDs, and reorder the records according to the timestamp. The standardization process is used to merge continuous traffic data into a standard time period using a sliding window mechanism.

3. The method for determining highway traffic operation status based on ensemble learning according to claim 2, characterized in that: S1 further includes S12; S12. Calculate the number of vehicles passing through the time window in the traffic record set according to the statistical method, and compare it with the collection duration of the time window to obtain the vehicle traffic density Qin. Use the time difference function to calculate the vehicle density difference ΔQin between adjacent time windows. When the vehicle density difference ΔQin ≤ the standard vehicle difference threshold Qt, the current time interval is marked as a normal driving period. No intervention is performed at this time, and normal monitoring is maintained. When the vehicle density difference ΔQin > the standard vehicle difference threshold Qt, the current time interval is marked as a dense vehicle period, and the mainline traffic analysis is triggered at this time. The standard vehicle differential threshold Qt acquisition process is as follows: extract the vehicle traffic density Qin at the ramp entrances within the past 30 days, smooth it using an exponentially weighted moving average method, and then use the percentile method to set the 95th percentile of the smoothed vehicle traffic density Qin as the standard vehicle differential threshold Qt.

4. The method for determining highway traffic operation status based on ensemble learning according to claim 3, characterized in that: S2 includes S21; S21. After obtaining the mainline traffic analysis instruction, collect vehicle speed data based on the radar speed measurement equipment deployed on the mainline lanes, and mark the collection timestamp and spatial location. Divide the time window into morning peak, off-peak, evening peak and nighttime periods according to the time interval to which the timestamp belongs, and use the corresponding category as time period data. Based on the GIS vector topology layer of the road structure, the vehicle merging points of the ramps are mapped to the locations of the main line speed measurement points according to the location numbering rules, forming a correspondence between the entrance and the main line. The traffic speed data of each speed measurement point within the time window is extracted, and the traffic speed data is preprocessed to obtain the measured vehicle speed Vm on the main line. The preprocessing includes missing value imputation and deviation value removal; The missing value imputation is used to fill in the traffic speed data with missing values ​​according to the linear interpolation method, and the deviation value removal is used to remove the traffic speed data that deviates from the mean by more than three times the standard deviation.

5. The method for determining highway traffic operation status based on ensemble learning according to claim 4, characterized in that: S2 further includes S22; S22. Extract the vehicle traffic density Qin and the measured speed Vm of each speed measurement point in each time window within the past month. Then, use timestamp alignment technology to align the vehicle traffic density Qin, the measured speed Vm of the main line and the time feature data to generate a training sample set. The training sample set includes an input sample set and a speed sample set; the input sample set includes vehicle traffic density Qin and time feature data; the speed sample set includes measured vehicle speed Vm on the main line; Using gradient boosting decision tree as the regression algorithm, the input sample set is used as the model input feature and the speed sample set is used as the supervision label to construct the vehicle speed prediction model. The input sample set is divided into training subset and validation subset and input into the vehicle speed prediction model. The model is trained using five-fold cross-validation. After training, the real-time vehicle traffic density Qin and time feature data are input into the vehicle speed prediction model. The vehicle speed prediction model outputs the main line predicted speed Vy.

6. The method for determining highway traffic operation status based on ensemble learning according to claim 5, characterized in that: S3 includes S31; S31. Based on the sliding window alignment algorithm, the measured speed Vm and the predicted speed Vy of the main line are uniformly mapped to the same time sampling period, and then converted to the same time axis through time resampling. The difference between the measured speed Vm and the predicted speed Vy of the main line is calculated to obtain the false smooth flow residual Δres of the speed, which reflects the degree of deviation between the actual running state of the main line and the theoretical predicted state under the current dense traffic conditions of the ramps.

7. The method for determining highway traffic operation status based on ensemble learning according to claim 6, characterized in that: S3 further includes S32; S32. Based on the fitting results of vehicle density difference ΔQin, standard vehicle difference threshold Qt, and vehicle speed false smooth flow residual Δres, the smooth flow status of highway vehicles is analyzed and evaluated, specifically including S321 and S322. S321. Compare the vehicle density difference ΔQin with the standard vehicle difference threshold Qt, and extract the minimum value Δres from the set of false smooth traffic residuals Δres during historical normal driving periods. min A comprehensive calculation is performed to construct a false smooth traffic index Fin, which reflects the coupling relationship between the traffic density of the entrance ramps and the mainline operation status, as follows; ； Wherein, Fin(t) represents the false accessibility index Fin at time t.

8. The method for determining highway traffic operation status based on ensemble learning according to claim 7, characterized in that: S322. Extract the false smooth traffic judgment index Fin from the past month's vehicle traffic under normal conditions, calculate the mean according to the statistical method, set the mean as the traffic discrimination threshold Rx, and then compare it with the false smooth traffic judgment index Fin obtained in real time to conduct a road smoothness assessment of the current highway ramp. When the false smooth traffic index Fin ≤ the traffic discrimination threshold Rx, it means that the current traffic state is a load-balanced state with no abnormal traffic bottlenecks or congestion risks. At this time, the current ramp is marked as a normal smooth traffic state and is monitored normally. When the false traffic flow judgment index Fin > the traffic discrimination threshold Rx, it indicates that the current traffic state is a false traffic flow state. At this time, the current ramp is marked as a false traffic flow state and error correction is performed.

9. The method for determining highway traffic operation status based on ensemble learning according to claim 8, characterized in that: S4 includes S41 and S42; S41. When the road traffic flow assessment indicates that the current traffic state is a false smooth state, record the current vehicle traffic density Qin, the measured vehicle speed Vm on the main line, the time feature data, and the false smooth speed residual Δres as misjudged samples and input them into the training sample pool of the vehicle speed prediction model. Use the false smooth speed residual Δres in the false smooth state as a reward or penalty signal in reinforcement learning, and introduce a learning rate adjustment mechanism to gradually adjust the learning weight for the false smooth state during the training process, and perform secondary model adjustment training on the vehicle speed prediction model. S42. After the vehicle speed prediction model is trained, the false smooth flow judgment index Fin for false smooth flow conditions is input into the vehicle speed prediction model to construct the slip trend index Hres, which is used to measure the cumulative error trend of false smooth flow conditions over multiple time periods, as follows. ； Where Hres(t) represents the sliding trend index at time t, w represents the number of time periods, i represents the first i time periods within the time window, and Fin(ti) represents the false smooth flow judgment index at time ti.

10. The method for determining highway traffic operation status based on ensemble learning according to claim 9, characterized in that: S4 also includes S43; S43. Calculate the mean and standard deviation of the historical sliding trend index Hres according to the statistical method, and set the sum of the mean and standard deviation as the critical threshold Wy for misjudged trend. Then, evaluate the misjudged trend with the real-time obtained sliding trend index Hres, as follows. When the sliding trend index Hres ≤ the critical threshold Wy for misjudgment trend, it means that the misjudgment is within the normal fluctuation range. At this time, the current state is maintained and the traffic status is assessed according to the original judgment logic. When the slip trend index Hres is greater than the critical threshold Wy for misjudgment trend, it indicates that there is a misjudgment of the false smooth flow state. At this time, the slip trend index Hres at the time of the misjudgment is fed back to the vehicle speed prediction model, and the vehicle speed prediction model is corrected by weighted training, and iterative analysis is performed through S1.

Images