A Quantile Robust Standardization and Adaptive Entropy Weight Traffic Index Construction Method and System

Abstract

This invention relates to a method and system for constructing a traffic index using quantile-robust standardization and adaptive entropy weighting, belonging to the field of intelligent transportation technology. The method includes: real-time acquisition of multi-source traffic data; after data validity verification, performing robust standardization and nonlinear mapping correction based on quantiles; dynamically calculating the information entropy of each indicator and updating the weights using a sliding window adaptive entropy weighting method; and synthesizing a traffic operation status index through hierarchical nested aggregation and threshold filtering strategies. The system includes a data preprocessing module, a dynamic weight calculation module, an index synthesis module, and a data validity verification module. The advantages of this invention are: through quantile-robust standardization and the sliding window adaptive entropy weighting mechanism, it significantly improves the index's resistance to noise data, achieves real-time adaptive weight updates, ensures the temporal smoothness and state response sensitivity of the index output, and is suitable for real-time monitoring and evaluation of urban road traffic operation status.

Description

Technical Field

[0001] This invention relates to a method and system for constructing a traffic index based on quantile robust standardization and adaptive entropy weight, belonging to the field of intelligent transportation technology. Background Technology

[0002] Traffic operation status index is a core indicator for measuring the degree of urban road traffic congestion and guiding traffic management and travel decisions. Traditional traffic index construction methods typically use fixed weights or static weights based on historical data to weight and aggregate multi-source traffic data (such as average road speed, traffic flow, time occupancy, headway, queue length, and travel time ratio). However, existing technologies have significant shortcomings in the following aspects:

[0003] 1. Actual traffic data often contains extreme values ​​(outliers) caused by equipment failures, communication anomalies, or unforeseen events. Traditional standardization methods (such as maximum-minimum normalization and Z-score standardization) are easily affected by these extreme values, causing the standardized data to deviate from the true distribution and thus distorting the final index. Especially in low-traffic or sudden congestion scenarios, individual outliers may cause drastic fluctuations in the index, affecting the stability of the assessment results.

[0004] 2. Most existing methods use fixed indicator weights or static weights that are updated in batches periodically, which cannot respond to the dynamic evolution of traffic flow characteristics in real time. For example, during peak and off-peak hours, the contribution of various indicators (such as speed and flow) to traffic condition assessment differs significantly, and static weights cannot capture this change, resulting in a decrease in the accuracy of the index assessment in different time periods and scenarios.

[0005] 3. Traditional index synthesis often directly sums the indicators at the current moment, ignoring the continuity and smooth transition characteristics of traffic conditions over time. This can easily cause unreasonable jumps in the index between adjacent moments, reducing the interpretability and practicality of the index.

[0006] Therefore, there is an urgent need for a traffic operation status index construction method that can suppress extreme data interference, dynamically update weights, and ensure the stability of the index time series, so as to solve the problem of state assessment distortion in the existing technology. Summary of the Invention

[0007] To overcome the shortcomings of existing technologies, this invention provides a method and system for constructing a traffic index based on quantile robust standardization and adaptive entropy weight. The technical solution of this invention is as follows:

[0008] A method for constructing a quantile-robust standardization and adaptive entropy-weighted traffic index includes the following steps:

[0009] Step S1, Data Preprocessing Step: Collect multi-source traffic data in real time, and perform robust standardization and nonlinear mapping correction processing based on quantiles on the multi-source traffic data to suppress the influence of extreme values ​​in the data and obtain standardized index data.

[0010] Step S2, Dynamic Weight Calculation Step: Using the sliding window adaptive entropy weight method, based on the standardized index data in the current sliding window, the information entropy of each evaluation index is dynamically calculated and the weight is updated to obtain the dynamic weight of each index.

[0011] Step S3, Index Synthesis Step: Based on the dynamic weights, the standardized index data is synthesized through hierarchical nested aggregation and threshold filtering strategies to obtain the final traffic operation status index.

[0012] In step S1, the data preprocessing step is described as follows:

[0013] The robust standardization based on quantiles includes: for the original indicator data sequence ,in Given the total number of data points, calculate the first number of data points in the sequence. quantiles and the quantiles ,in Using formula Standardize to obtain the standardized value. and for those exceeding the interval of Perform truncation, that is, let , Indicates the first One original data point;

[0014] The nonlinear mapping correction includes: using the formula or formula The standardized data is then corrected to obtain the final corrected indicator data. ,in, For the power-corrected parameter, and , For logarithmic correction parameters, and .

[0015] The dynamic weight calculation step in step S2 specifically includes:

[0016] Set the length of the sliding window to Where L is a positive integer, for time t, standardized index data of L time samples within the current sliding window are taken to form an evaluation matrix. ,in, This represents the standardized value of the j-th indicator for the i-th sample within the window, where m is the total number of evaluation indicators.

[0017] Calculate the weight of the j-th indicator in the i-th sample. ;

[0018] Calculate the information entropy of the j-th indicator. ;

[0019] Calculate the dynamic weight of the j-th indicator Where k is the index; as the sliding window moves forward, the above process is repeated to achieve dynamic updating of the weights; where the length L of the sliding window is an adaptive length, determined according to the sampling frequency f of the traffic data and the preset time window T, i.e. ,in This represents the floor function. The value ranges from 5 minutes to 30 minutes.

[0020] In the exponential synthesis step of step S3:

[0021] The hierarchical nested aggregation includes: dividing the lower-level indicators into several categories according to a preset traffic state impact dimension, and first calculating the sub-index for each category. ,in Indicates the first Sub-indices of each category, For the first The dynamic weights of each indicator For the first The data for each indicator has been corrected. Indicates belonging to the first The system comprises a set of indicators for each category; then, the sub-indices are weighted and aggregated a second time to obtain the total index. Where I is the initial total index and C is the total number of categories. The preset weight of the c-th sub-index or the weight dynamically calculated by the entropy weight method;

[0022] The threshold filtering strategy includes: setting a set of incremental threshold values. and the corresponding state level, where K is the number of thresholds, and the initial total index is calculated for the current time t. Compared with the final index of the previous moment If a comparison is made, If the state is unchanged, then the state remains unchanged; otherwise, the state is divided according to the threshold. The threshold filtering strategy uses a preset stability tolerance parameter to achieve exponential smoothing through the following first-order hysteresis filtering formula: ,in, The final traffic operation status index at time t. This is a smoothing coefficient, with a value range of (0, 1], which is dynamically adjusted when traffic conditions change abruptly. The value of .

[0023] The multi-source traffic data includes at least two of the following: average speed of road segment, traffic flow, time occupancy rate, headway, queue length, and travel time ratio; and before the data preprocessing step, a data validity verification step is also included: performing integrity checks on the real-time collected multi-source traffic data, removing outlier data that exceeds the physical range or has an abnormal rate of change, and filling missing values ​​with linear interpolation or historical data from the same period.

[0024] The traffic operation status index construction system implementing the method includes a data preprocessing module, which is used to collect multi-source traffic data in real time and perform robust standardization and nonlinear mapping correction processing based on quantiles on the multi-source traffic data to suppress the influence of extreme values ​​in the data and obtain standardized index data.

[0025] The dynamic weight calculation module, connected to the data preprocessing module, is used to dynamically calculate the information entropy of each evaluation indicator and update the weight based on the standardized indicator data in the current sliding window using the sliding window adaptive entropy weight method, so as to obtain the dynamic weight of each indicator.

[0026] The index synthesis module is connected to the data preprocessing module and the dynamic weight calculation module, respectively. It is used to synthesize the standardized index data based on the dynamic weight through hierarchical nested aggregation and threshold filtering strategies to obtain the final traffic operation status index.

[0027] The data preprocessing module includes a quantile normalization unit, used to process the original indicator data sequence. Let n be the total number of data points, and calculate the p-th quantile of the sequence. and the qth quantile ,in , This indicates that the i-th original data point uses the formula Standardize to obtain the standardized value. and for those exceeding the interval of Perform truncation, that is, let ;

[0028] Nonlinear mapping correction unit, used to employ formula or formula The standardized data is then corrected to obtain the final corrected indicator data. ,in For power-corrected parameters and , For logarithmic correction parameters and .

[0029] The dynamic weight calculation module includes a sliding window construction unit, used to set the length of the sliding window to L, where L is a positive integer, and to determine the adaptive length based on the sampling frequency f of the traffic data and the preset time window T. ,in The function represents the floor function, where T ranges from 5 minutes to 30 minutes. For time t, the evaluation matrix is ​​constructed from standardized index data of L time samples within the current sliding window. ,in This represents the standardized value of the j-th indicator for the i-th sample within the window, where m is the total number of evaluation indicators.

[0030] The information entropy calculation unit is used to calculate the weight of the j-th indicator in the i-th sample. And calculate the information entropy of the j-th index. ;

[0031] The weight update unit is used to calculate the dynamic weight of the j-th indicator. , where k is the index of the indicator, and the above calculation is repeated as the sliding window moves forward to achieve dynamic updating of the weights.

[0032] The index synthesis module includes a hierarchical nested aggregation unit, used to divide the underlying indicators into multiple categories according to a preset traffic state impact dimension, and first calculate the sub-index for each category. ,in This represents the sub-index of the c-th category. The dynamic weight of the j-th indicator. For the j-th indicator, the data has been corrected. This represents the set of indicators belonging to the c-th category; then, the sub-indices are weighted and aggregated twice to obtain the total index. Where I is the initial total index and C is the total number of categories. The preset weight of the c-th sub-index or the weight dynamically calculated by the entropy weight method;

[0033] Threshold filtering unit, used to filter by first-order hysteresis formula Exponential smoothing is performed to obtain the final traffic operation status index. ,in Let be the initial total exponent at the current time t. The final index at the previous moment. Smoothing coefficient and And set an incremental threshold. And the corresponding state level, where K is the number of thresholds.

[0034] It also includes a data validity verification module, connected between the data acquisition end and the data preprocessing module, used to perform integrity checks on multi-source traffic data, remove outlier data that exceeds the physical range or has an abnormal rate of change, and fill missing values ​​using linear interpolation or historical data from the same period; and the multi-source traffic data includes at least two of the following: average speed of road segment, traffic flow, time occupancy rate, headway, queue length, and travel time ratio.

[0035] The advantages of this invention are:

[0036] 1. A robust standardization method based on quantiles is introduced. By selecting the p-th and q-th quantiles instead of the traditional maximum and minimum values ​​for normalization, the impact of extreme values ​​on the standardization process is effectively suppressed. Combined with nonlinear mapping correction (power function or logarithmic function), the sensitivity distribution of the data in different intervals is further optimized, making the standardized data more realistically reflect the traffic conditions.

[0037] 2. An adaptive entropy weighting method using a sliding window is employed. This method dynamically calculates the information entropy of each indicator and updates its weights based on data within a real-time sliding window. The window length can be adaptively adjusted according to sampling frequency and time requirements, enabling the weights to quickly respond to changes in traffic flow characteristics. For example, when a road segment changes from smooth to congested, the weight of the speed indicator automatically increases, and the weight of the flow rate indicator is adjusted accordingly, thus ensuring that the index's assessment of different traffic conditions is more sensitive and accurate.

[0038] 3. A threshold filtering strategy and a first-order lag filtering formula are introduced into the index synthesis step to smooth the initial total index. By setting an incremental threshold and a stability tolerance parameter δ, unreasonable jumps in the index between adjacent time points are effectively suppressed. Simultaneously, the smoothing coefficient λ can be dynamically adjusted according to the degree of traffic state abrupt changes, ensuring stability during stable periods while preserving the response speed during abrupt changes. This results in the final traffic operation state index having good temporal continuity and interpretability.

[0039] 4. A hierarchical nested aggregation strategy is adopted. First, the underlying indicators are divided into multiple categories according to the dimensions of traffic condition impact (such as "efficiency dimension", "density dimension", etc.), and sub-indices are calculated for each category. Then, the sub-indices are weighted and aggregated a second time. This hierarchical structure can not only output a comprehensive total index, but also provide sub-indices for each dimension, which makes it easier for traffic managers to analyze the causes of congestion from different levels, thereby improving the interpretability and application value of the index.

[0040] 5. A data validity verification step was added before data preprocessing. This step performed integrity checks on the real-time multi-source traffic data, removed outlier data, and filled in missing values ​​using linear interpolation or historical data from the same period. This mechanism effectively ensured the data quality for subsequent standardization and weight calculation, further improving the robustness of the index construction.

[0041] In summary, this invention significantly outperforms existing technologies in terms of anti-interference capability, adaptive weight update, temporal stability, multi-dimensional evaluation, and system deployability, and can provide a more reliable and sensitive traffic operation status index for intelligent traffic management and public travel services. Attached Figure Description

[0042] Figure 1 This is a flowchart illustrating the method of the present invention.

[0043] Figure 2 This is a structural block diagram of the system of the present invention.

[0044] Figure 3 yes Figure 2 The structural block diagram of the data preprocessing module.

[0045] Figure 4 yes Figure 2 The structural block diagram of the dynamic weight calculation module.

[0046] Figure 5 yes Figure 2 Block diagram of the index synthesis module. Detailed Implementation

[0047] The present invention will be further described below with reference to specific embodiments, and the advantages and features of the present invention will become clearer as a result. However, these embodiments are merely exemplary and do not constitute any limitation on the scope of the present invention. Those skilled in the art should understand that modifications or substitutions can be made to the details and form of the technical solutions of the present invention without departing from the spirit and scope of the present invention, but all such modifications and substitutions fall within the protection scope of the present invention.

[0048] See Figures 1 to 5 This invention relates to a method for constructing a quantile-robust standardization and adaptive entropy weight traffic index, comprising the following steps:

[0049] Step S1, Data Preprocessing Step: Collect multi-source traffic data in real time, and perform robust standardization and nonlinear mapping correction processing based on quantiles on the multi-source traffic data to suppress the influence of extreme values ​​in the data and obtain standardized index data.

[0050] Step S2, Dynamic Weight Calculation Step: The sliding window adaptive entropy weight method is adopted to dynamically calculate the information entropy of each evaluation index and update the weights based on the standardized index data within the current sliding window, so as to obtain the dynamic weights of each index.

[0051] Step S3, Index Synthesis Step: Based on the dynamic weights, the standardized index data is synthesized through a hierarchical nested aggregation and threshold filtering strategy to obtain the final traffic operation state index.

[0052] The advantage of Step S1 is that by adopting quantile-based robust standardization (for example, selecting the p-th quantile and the q-th quantile , where 0 < p < q < 1) to replace the traditional normalization method based on the maximum and minimum values, it can effectively weaken the distortion of the data distribution caused by outliers or extreme anomalies, and avoid the drastic fluctuation of the index caused by a single abnormal data. Combined with non-linear mapping correction (power function or logarithmic function), it further optimizes the sensitivity of the data in different index intervals, making the standardized data more truly and stably reflect the traffic operation state, and significantly improving the anti-interference ability of the subsequent index calculation.

[0053] The advantage of Step S2 is that by introducing a sliding window adaptive entropy weight mechanism, it can capture the dynamic evolution of traffic flow characteristics in real time. The window length can be adaptively adjusted according to the sampling frequency and actual needs (such as 5 - 30 minutes), so that the weights no longer depend on static historical data, but are continuously updated with the latest data within the window. When the traffic state changes (such as from unobstructed to congested), the information entropy of each index (such as speed, flow, time occupancy) automatically changes, and the weights are adjusted accordingly, thus ensuring that the index evaluation has both sensitivity to the current state and avoids the distortion problem caused by the lag of weight update.

[0054] The advantage of Step S3 is that on the one hand, hierarchical nested aggregation is adopted to divide the underlying indicators into multiple categories according to the traffic state impact dimension (such as efficiency dimension, density dimension, etc.), calculate the sub-index first, and then obtain the total index by secondary weighting. It not only outputs the comprehensive index, but also provides the sub-index of each dimension, which is convenient for multi-angle analysis of the causes of congestion. On the other hand, the threshold filtering strategy combined with the first-order lag filtering formula smooths the initial total index, effectively suppressing the unreasonable jumps between adjacent moments, making the index output have excellent temporal continuity and stability, while retaining the fast response ability to sudden congestion.

[0055] [[ID=!9]]In the data preprocessing step of Step S1:

[0056] The quantile-based robust standardization includes: for the original index data sequence where [[ID=2!5]] Given the total number of data points, calculate the first number of data points in the sequence. quantiles and the quantiles ,in Using formula Standardize to obtain the standardized value. and for those exceeding the interval of Perform truncation, that is, let , Indicates the first One original data point;

[0057] The nonlinear mapping correction includes: using the formula or formula The standardized data is then corrected to obtain the final corrected indicator data. ,in, For the power-corrected parameter, and , For logarithmic correction parameters, and .

[0058] In this invention, the robust standardization and nonlinear mapping correction algorithm based on quantiles in step S1 has the following input, output, and correlation with the transportation field:

[0059] 1. Input Data Settings

[0060] The original index data sequence is the input to the algorithm. Multi-source traffic data collected in real time from traffic detectors (such as geomagnetic coils, radar, video surveillance, etc.). Taking average road segment speed as an example:

[0061] Sampling frequency: 1 time per minute (f = 1 time / minute);

[0062] Sequence length : Take n=100 historical data points before the current time (approximately 1.5 hours);

[0063] Data unit: km / h.

[0064] Typical input sequence example (unit: km / h):

[0065] ;

[0066] Where x=0 may be a device malfunction or a vehicle coming to a complete stop (outlier), and x=120 may be a speeding vehicle or detection noise (extreme value).

[0067] 2. Algorithm Processing

[0068] (1) Quantile standardization

[0069] Choose p=0.05 and Calculate the 5th quantile. and the 95th percentile :

[0070] After sorting the sequence in the example above, (This indicates that 5% of the data is below 52);

[0071] (This indicates that 95% of the data is below 65);

[0072] Using formula Standardize it.

[0073] Intrinsic Relationship: Quantiles reflect the robust position of traffic data in the statistical distribution, unaffected by extreme outliers (such as 0 or 120). Compared to the traditional max-min method (using 0 and 120 as the denominator), the denominator... It reflects the speed fluctuation range under normal traffic conditions, and its physical meaning is the speed change amplitude within a typical unobstructed section. (Standardized) This indicates the current speed's relative position within this typical range: The response is smooth. Corresponding to congestion.

[0074] Truncating: Values ​​exceeding [0,1] after calculation are truncated. For example, outliers. Calculated The value after truncation is 0; extreme values Calculated The value is truncated to 1. This process ensures that all standardized values ​​fall within a meaningful range, and outliers are automatically suppressed to the most extreme states (congestion or unobstructed flow) without generating abnormal intermediate values.

[0075] (2) Nonlinear mapping correction

[0076] Using power function form ,Pick (Square root). Input Output .

[0077] Intrinsic correlation: In traffic flow characteristics, people are more sensitive to changes in congestion in low-speed ranges. For example, a speed decrease from 60 km / h to 40 km / h (a 33% reduction) is perceived as slight congestion, but a decrease from 20 km / h to 0 km / h (a 100% reduction) is perceived as severe congestion. Power function Make low The output value for the region (corresponding to low speed) is stretched, thereby amplifying the sensitivity of congestion identification.

[0078] For example: when (Severe congestion corresponds to a standardized speed of approximately 53.5 km / h; actual calculation: if...) ,but The original speed is 52 + 0.113 = 53.3 km / h, which is unreasonable. It should be noted that the lower value after standardization corresponds to a lower speed. Correction: The corresponding speed is ≤52 km / h (actual congestion). The corresponding speed is approximately 53.3 km / h (generally unobstructed).

[0079] A more reasonable parameter: If the typical congested area has a speed range of 20~50 km / h, take... ,but Corresponding to ≤25 km / h (congestion). Corresponding to speeds ≥45 km / h (unobstructed traffic). Assume... (Corresponding speed approximately 29 km / h, light congestion), via Revised The output value is increased, which means that the congestion state will contribute a greater weight in the exponential synthesis, which is in line with people's sensitivity to congestion.

[0080] 3. The output is the final revised indicator data. This value is fed into steps S2 and S3 as input for subsequent dynamic weight calculation and index synthesis. Its physical meaning is: a normalized traffic state value after robust standardization and sensitivity correction. The closer the value is to 1, the smoother the traffic flow; the closer it is to 0, the more congested the traffic flow. It also has good distinguishability and stability under different traffic conditions.

[0081] The dynamic weight calculation step in step S2 specifically includes:

[0082] Set the length of the sliding window to Where L is a positive integer, for time t, standardized index data of L time samples within the current sliding window are taken to form an evaluation matrix. ,in, This represents the standardized value of the j-th indicator for the i-th sample within the window, where m is the total number of evaluation indicators.

[0083] Calculate the weight of the j-th indicator in the i-th sample. ;

[0084] Calculate the information entropy of the j-th indicator. ;

[0085] Calculate the dynamic weight of the j-th indicator Where k is the index; as the sliding window moves forward, the above process is repeated to achieve dynamic updating of the weights; where the length L of the sliding window is an adaptive length, determined according to the sampling frequency f of the traffic data and the preset time window T, i.e. ,in This represents the floor function. The value ranges from 5 minutes to 30 minutes.

[0086] In this invention, the dynamic weight calculation step S2 adopts the sliding window adaptive entropy weight method, and its input, output, and correlation with the transportation field are set as follows:

[0087] 1. Input Data Settings

[0088] The algorithm's input is the standardized index data obtained after processing in step S1. Let the current time be t, and there be m evaluation indicators (e.g., m=3, indicator 1 is the average speed of the road segment). Indicator 2 is traffic flow. Indicator 3 is time occupancy. The values ​​of each indicator have been normalized to the [0,1] range.

[0089] The sliding window length L is adaptively determined based on the sampling frequency f and the preset time window T: For example, if the traffic data sampling frequency f = 1 time / minute and the preset time window T = 15 minutes, then L = 15. This means that the algorithm takes the most recent 15 time samples (i.e., data from the past 15 minutes) to form the evaluation matrix each time.

[0090] Example of input evaluation matrix (time t=08:30:00):

[0091] ;

[0092] Among them, column 1 The speed indicator (high value indicates smooth flow), column 2. Column 3 shows the flow rate indicator (high value for moderate flow, which may decrease in oversaturated flow). This is a occupancy rate indicator (high values ​​indicate congestion). Each row corresponds to one time sample (minutes).

[0093] 2. Algorithm Processing

[0094] (1) Calculate the specific gravity

[0095] For the j-th indicator, calculate the weight of the i-th sample in the sum of all samples for that indicator:

[0096] ; This reflects the relative size of the current sample on the j-th indicator. In a traffic scenario, if the speed indicator at a certain moment... If the speed is significantly lower than at other times within the window (i.e., when congestion occurs), the corresponding speed will be smaller; conversely, if the speed is higher, then... The proportion is relatively large. This proportion provides relative distribution information for subsequent information entropy calculations.

[0097] (2) Calculate information entropy

[0098] Information entropy This measures the dispersion or variability of the j-th indicator within the sliding window. A larger entropy value (closer to 1) indicates a more uniform distribution and smaller fluctuations in the indicator's values; a smaller entropy value indicates greater differences and more drastic fluctuations in the indicator's values.

[0099] During traffic operation:

[0100] Speed ​​index: When traffic conditions are stable (continuously smooth or continuously congested), speed fluctuations are small and entropy values ​​are high, which means that this index provides less information in the current window (because the change is not significant), and its weight should be reduced.

[0101] Traffic flow or occupancy indicators: When traffic events (such as accidents or construction) cause a sudden drop in traffic flow or a sudden increase in occupancy, these indicators fluctuate significantly within the window, and their entropy values ​​decrease. This means that they carry richer information about changes, and their weight should be increased.

[0102] (3) Calculate the dynamic weights

[0103] ;

[0104] Intrinsic connection: This is called the information utility value; the smaller the entropy value, the greater the fluctuation. The larger the value, the higher the weight. This formula ensures that the sum of all indicator weights is 1. The output is a dynamic weight. This directly reflects the importance of each indicator under the current traffic conditions: the greater the fluctuation of an indicator, the more significant its contribution to the change in conditions, and the higher its weight should be given in the index synthesis.

[0105] For example: Assume that within the window, the speed index remains stable at around 0.85 (unimpeded flow), and the traffic and occupancy are also relatively stable. The calculation yields... , , Then 1-e are 0.02, 0.03, and 0.04 respectively, and the normalized weights are approximately , , At this point, market share has the highest weight because it is more sensitive to even small fluctuations.

[0106] If a sudden congestion occurs in the latter half of the window: the speed drops sharply from 0.85 to 0.30, the flow rate drops from 0.70 to 0.40, and the occupancy rate rises from 0.30 to 0.80. At this point, the speed index fluctuates drastically, and its information entropy... A significant decrease (e.g., down to 0.75). The fluctuations in traffic and market share are relatively small (1-e = 0.10 and 0.15 respectively). After normalization, the new weights become... , , The weight of the speed indicator is automatically increased because it is a key variable reflecting the occurrence of congestion. Speed ​​changes are amplified during index synthesis, thus quickly and accurately capturing congestion events.

[0107] 3. Adaptive sliding window mechanism

[0108] Window length The time window T can be adjusted within 5 to 30 minutes. The inherent relationship is that a shorter time window (e.g., 5 minutes) makes the weights more sensitive to short-term fluctuations in traffic conditions, suitable for rapid response to emergencies; a longer time window (e.g., 30 minutes) provides a smoother transition, suitable for assessing macro trends under stable conditions. Users can dynamically adjust T according to the actual application scenario (e.g., expressways, urban arterial roads).

[0109] Sliding update: Every time a new time sample is collected (e.g., every minute), the window slides forward, discarding the oldest sample and adding the latest sample, recalculating the evaluation matrix, information entropy, and weights. This ensures that the weights can track the dynamic evolution of traffic flow in real time, solving the problem of lagging updates in traditional static weights.

[0110] 4. Output data and its uses

[0111] The algorithm outputs a dynamic weight vector for each indicator. And satisfy The weight vector is fed into step S3 for calculating the sub-indices and the total index in the hierarchical nested aggregation. Its physical meaning is: the degree to which each evaluation index contributes to the overall operational status assessment under the recent traffic conditions represented by the current sliding window. The weights adaptively change with traffic conditions, making the final traffic index more sensitive and reliable.

[0112] In the exponential synthesis step of step S3:

[0113] The hierarchical nested aggregation includes: dividing the lower-level indicators into several categories according to a preset traffic state impact dimension, and first calculating the sub-index for each category. ,in Indicates the first Sub-indices of each category, For the first The dynamic weights of each indicator For the first The data for each indicator has been corrected. Indicates belonging to the first The system comprises a set of indicators for each category; then, the sub-indices are weighted and aggregated a second time to obtain the total index. Where I is the initial total index and C is the total number of categories. The preset weight of the c-th sub-index or the weight dynamically calculated by the entropy weight method;

[0114] The threshold filtering strategy includes: setting a set of incremental threshold values. and the corresponding state level, where K is the number of thresholds, and the initial total index is calculated for the current time t. Compared with the final index of the previous moment If a comparison is made, If the state is unchanged, then the state remains unchanged; otherwise, the state is divided according to the threshold. The threshold filtering strategy uses a preset stability tolerance parameter to achieve exponential smoothing through the following first-order hysteresis filtering formula: ,in, The final traffic operation status index at time t. This is a smoothing coefficient, with a value range of (0, 1], which is dynamically adjusted when traffic conditions change abruptly. The value of .

[0115] In this invention, the exponential synthesis step S3 includes hierarchical nested aggregation and a threshold filtering strategy. Its inputs, outputs, and associations with the transportation domain are set as follows.

[0116] I. Hierarchical Nested Aggregation

[0117] 1. Input Data Settings

[0118] The inputs to hierarchical nested aggregations include:

[0119] The dynamic weights of each indicator output in step S2 ;

[0120] The corrected index data output in step S1 (Normalized to the [0,1] interval).

[0121] Suppose there are m=4 underlying indicators: average speed Traffic flow Time occupancy Queue length Based on the dimensions of impact from traffic conditions, they are divided into two categories:

[0122] Efficiency dimension c=1: Includes speed indicators (the higher the speed, the smoother the flow);

[0123] Load dimension c=2: includes traffic, occupancy, and queue length (high traffic, high occupancy, and long queues indicate congestion).

[0124] Example of input data at current time t:

[0125] ;

[0126] ;

[0127] ;

[0128] 2. Processing procedure

[0129] (1) Calculate the sub-indices of each dimension

[0130] formula: ;

[0131] For the efficiency dimension (c=1):

[0132] For the load dimension (c=2): ;

[0133] sub-index This reflects the contribution of a specific traffic impact dimension (such as traffic efficiency or road load) to the overall operational status. A lower efficiency sub-index indicates slower traffic flow, while a higher load sub-index indicates greater road congestion. These two sub-indices describe traffic conditions from different physical perspectives, facilitating managers in pinpointing the causes of congestion (for example, if both the efficiency and load sub-indices are low, it may be due to insufficient traffic flow; however, the actual situation requires consideration of the physical meaning: a high load sub-index indicates congestion, and a low efficiency sub-index also indicates congestion; when both are consistent, it reflects typical congestion; if the efficiency sub-index is low but the load sub-index is moderate, it may be due to an accident causing localized congestion while overall traffic flow remains unsaturated). Through multi-dimensional calculations, this invention provides richer diagnostic information than a single total index.

[0134] (2) The total index is obtained by weighted aggregation twice.

[0135] formula: ;

[0136] Set the preset sub-index weights , (This can be adjusted according to actual needs; for example, urban expressways prioritize efficiency, while main roads prioritize load capacity.)

[0137] The initial total index is: ;

[0138] Internal correlation: I integrates information from different dimensions, and its value range is [0,1]. The smaller the value, the worse the traffic operation (congestion), and the larger the value, the smoother the traffic. 0.239 corresponds to a severe congestion state.

[0139] 3. Output data

[0140] The initial total exponent I(t) is output as 0.239, which will be fed into the threshold filtering strategy for time-series smoothing.

[0141] II. Threshold Filtering Strategy

[0142] 1. Input Data Settings

[0143] The inputs to the threshold filtering strategy include:

[0144] The initial total exponent I(t) at the current moment (obtained by hierarchical nested aggregation);

[0145] Final index at the previous moment (Stored in the system);

[0146] Preset threshold set and the corresponding status level;

[0147] Stability tolerance parameters Smoothing coefficient .

[0148] Typical parameter setting example:

[0149] The corresponding status levels are: [0,0.2] Severe congestion, (0.2,0.4] Moderate congestion, (0.4,0.6] Mild congestion, (0.6,0.8] Basically smooth, (0.8,1.0] Smooth traffic.

[0150] Stability tolerance parameters (This means that when the exponential change is less than 0.05, it is considered stable and the state does not switch).

[0151] Smoothing coefficient The default value is 0.3 (emphasizing historical smoothness), and it is dynamically increased to 0.8 when a mutation is detected.

[0152] Assuming the final index at the previous moment (Moderate congestion).

[0153] 2. Algorithm Processing

[0154] (1) State stability judgment

[0155] calculate .

[0156] because If the stability condition is met, the system determines that the traffic state has not changed substantially and maintains the original state level (moderate congestion). This judgment avoids frequent state jumps caused by minor data fluctuations and improves the interpretability of the index.

[0157] Traffic conditions exhibit temporal continuity; the level of congestion at adjacent moments should not exhibit abrupt changes that defy physical laws (e.g., a sudden jump from severe congestion to smooth traffic). The physical meaning of Δ is the threshold of exponential change that is tolerable to the human eye or management; changes smaller than this threshold are considered noise or measurement error.

[0158] (2) First-order hysteresis filtering smoothing

[0159] Even though the state remains unchanged, the exponent values ​​are still smoothed to reduce random fluctuations:

[0160] ;

[0161] Substitution :

[0162] ;

[0163] The smoothed index is 0.247, which is still in the moderate congestion range, but it is closer to the historical trend than the original value of 0.239, thus avoiding drastic fluctuations in the index.

[0164] (3) Dynamic adjustment under sudden changes

[0165] Suppose a sudden accident occurs at a certain moment, and the initial total index I(t) drops sharply from 0.25 in the previous moment to 0.08 (severe congestion). The difference is... At this point, the system determines a sudden state change and dynamically increases... The value is reduced to 0.8, making the filtering formula more dependent on the current measured values:

[0166] ;

[0167] The index eventually dropped rapidly to 0.114, accurately reflecting sudden changes in congestion while retaining a small amount of historical smoothing to prevent extreme jumps.

[0168] Smoothing coefficient The physical meaning of represents the degree of confidence the index has in the current data. The smaller value is taken during periods of stability. It can filter out noise and take the largest value during the mutation period. It ensures a fast response time. The dynamic adjustment mechanism simulates the traffic manager's response logic to abnormal events: cautious in normal times, and rapid response in case of emergencies.

[0169] 3. Output data and its uses

[0170] Final output This value is the traffic operation status index constructed in this invention. According to... It can be mapped to intuitive levels such as "severe congestion / moderate congestion / light congestion / basically smooth / smooth traffic" for use in real-time monitoring screens in traffic command centers; route recommendations in public travel apps; and traffic congestion warnings and emergency dispatch.

[0171] The multi-source traffic data includes at least two of the following: average speed of road segment, traffic flow, time occupancy rate, headway, queue length, and travel time ratio; and before the data preprocessing step, a data validity verification step is also included: performing integrity checks on the real-time collected multi-source traffic data, removing outlier data that exceeds the physical range or has an abnormal rate of change, and filling missing values ​​with linear interpolation or historical data from the same period.

[0172] Before the data preprocessing in step S1, a data validity verification step is also included. This step performs integrity checks and cleaning on the real-time collected multi-source traffic data to ensure the quality of the input data.

[0173] 1. Input Data Settings

[0174] The input is raw multi-source traffic data, including at least the following two: average speed of road segment (unit: km / h), traffic flow (unit: vehicles / hour / lane), time occupancy rate (unit: %), headway (unit: s), queue length (unit: m), and travel time ratio (dimensionless).

[0175] Data is sourced from geomagnetic coils, microwave radar, video detectors, or floating car GPS tracks, with sampling frequencies... times per minute.

[0176] Input example (raw data collected at a certain minute for a certain road segment):

[0177] Average speed of the road segment: v = 85 km / h; Traffic flow: q = 1200 vehicles / hour / lane;

[0178] Time occupancy: occ = 75%; Headway: h = 1.2s;

[0179] Queue length: (No queue); Trip Time Ratio: TTR = 1.0.

[0180] 2. Algorithm Processing

[0181] (1) Integrity check

[0182] Check if any data fields are missing. For example, if the speed data is empty (NULL) or the flow data is NaN at a certain moment, it is considered a missing value.

[0183] Intrinsic Correlation: During traffic data collection, packet loss and momentary equipment malfunctions may occur, leading to the loss of some data. If left unaddressed, subsequent standardization and weight calculations will be incorrect.

[0184] (2) Physical range test

[0185] Remove data that is clearly beyond the physical possibilities. For example:

[0186] Speed: Normal range is 0 ~ 120 km / h (urban roads). If v < 0 or v > 150, it is considered an outlier.

[0187] Traffic flow: The normal range is 0 ~ 2000 vehicles / hour / lane. If q > 2500, it is considered an outlier.

[0188] Occupancy rate: The normal range is 0% ~ 100%. If occ > 100% or occ < 0, it is considered an outlier.

[0189] Headway: The normal range is 0.5 s ~ 10 s. If h < 0.5 or h > 10, it is considered an outlier.

[0190] Queue length: Normal range is 0 ~ 1000 m, if A negative value is considered a wild point;

[0191] Trip time ratio: The normal range is 0.5 to 5.0. If it exceeds this range, it is considered a wild point.

[0192] For example: In the above input, a speed of 85 km / h is normal; an occupancy rate of 75% is normal; and a headway of 1.2 s is normal. If at a certain moment a speed v = 0 km / h (equipment malfunction) or an occupancy rate occ = 105% (measurement error) is detected, it is determined to be an outlier and is directly removed.

[0193] (3) Anomaly test of rate of change

[0194] Check whether the rate of change of the data over adjacent time points exceeds a reasonable range. For example:

[0195] Rate of change of velocity: (That is, a speed change of more than 40 km / h within 1 minute) is considered abnormal;

[0196] Rate of change in flow: Vehicles / hour / lane are considered abnormal;

[0197] Market share change rate: This is considered abnormal.

[0198] Traffic flow is continuous and inertial; under normal circumstances, the speed change between adjacent minutes will not be too large. If a sudden change occurs (such as the speed jumping from 60 km / h to 0 km / h and then recovering), it is usually caused by detector noise or transmission errors and should be rejected.

[0199] For example: If the speed in the previous minute is v(t-1) = 60 km / h and the speed in the current minute is v(t) = 15 km / h, the rate of change is 45 km / h > 40, which is determined to be an outlier.

[0200] (4) Filling missing values

[0201] For outliers that have been removed or original missing values, the following strategy is used to fill them:

[0202] Linear interpolation: Suitable for short-term missing values ​​(e.g., consecutive 1-3 minutes). For example, if data at t=5 minutes and t=8 minutes are valid, but data at t=6 minutes and 7 minutes are missing, then linear interpolation can be performed based on the two endpoints.

[0203] Historical data from the same period: suitable for longer periods of missing data (e.g., more than 10 consecutive minutes). Fill with the median or mean of the same time over the past 7 days (e.g., every Tuesday at 8:30 AM).

[0204] For example: If speed data for a certain road segment is missing at t=10 minutes, and the data before and after are continuous, the average of the two minutes before and after the missing data is used to fill the gap. If all five consecutive minutes are missing, the median speed of the same time period on the same day of the previous week is used to fill the gap.

[0205] 3. Output data and its uses

[0206] The output is a complete and reasonable multi-source traffic data sequence after validity verification and cleaning. This sequence serves as the input for step S1 and is used for subsequent quantile robust standardization processing.

[0207] Through the above settings, the data validity verification step is closely integrated with the traffic data collection scenario, effectively ensuring the quality of the input data and improving the robustness and reliability of the entire index construction system.

[0208] The present invention also relates to a traffic operation status index construction system for implementing the method, including a data preprocessing module 1, which is used to collect multi-source traffic data in real time and perform robust standardization and nonlinear mapping correction processing based on quantiles on the multi-source traffic data to suppress the influence of extreme values ​​in the data and obtain standardized index data.

[0209] The dynamic weight calculation module 2, connected to the data preprocessing module, is used to dynamically calculate the information entropy of each evaluation indicator and update the weight based on the standardized indicator data in the current sliding window using the sliding window adaptive entropy weight method, so as to obtain the dynamic weight of each indicator.

[0210] The index synthesis module 3 is connected to the data preprocessing module and the dynamic weight calculation module, respectively. It is used to synthesize the standardized index data based on the dynamic weight through hierarchical nested aggregation and threshold filtering strategies to obtain the final traffic operation status index.

[0211] The data preprocessing module is used to collect multi-source traffic data in real time (including but not limited to at least two of the following: average speed of road segment, traffic flow, time occupancy rate, headway, queue length, and travel time ratio), and to perform robust standardization and nonlinear mapping correction processing based on quantiles on the multi-source traffic data to suppress the influence of extreme values ​​in the data and obtain standardized index data.

[0212] By incorporating embedded quantile normalization and nonlinear mapping correction units, the data preprocessing module effectively filters out extreme values ​​(outliers) caused by equipment malfunctions, communication anomalies, or unforeseen events at the source, avoiding the bias caused by outliers in traditional normalization methods (such as maximum-minimum normalization). Simultaneously, the nonlinear mapping correction (power function or logarithmic function) automatically adjusts the sensitivity of different intervals based on the perceived characteristics of traffic flow (e.g., amplifying differences in low-speed congestion areas). Therefore, this module significantly improves the robustness and accuracy of the input data, providing high-quality foundational data for subsequent weight calculations and index synthesis.

[0213] The dynamic weight calculation module is connected to the data preprocessing module and is used to dynamically calculate the information entropy of each evaluation indicator and update the weight based on the standardized indicator data in the current sliding window using the sliding window adaptive entropy weight method, so as to obtain the dynamic weight of each indicator.

[0214] This module introduces a sliding window mechanism, where the window length can be adaptively adjusted based on the sampling frequency and a preset time window (5-30 minutes). This allows the weights to no longer rely on static historical data, but rather track the dynamic evolution of traffic flow characteristics in real time. When traffic conditions change (e.g., from smooth flow to congestion), the information entropy changes automatically, and the weights adjust accordingly—indicators with drastic fluctuations (e.g., sudden speed drops) receive higher weights, thus more sensitively reflecting abrupt changes in state. Compared to traditional fixed-weight methods, this module completely solves the problem of lagging weight updates, ensuring the accuracy and real-time performance of index evaluation.

[0215] The index synthesis module is connected to the data preprocessing module and the dynamic weight calculation module, respectively, and is used to synthesize the standardized index data based on the dynamic weight through hierarchical nested aggregation and threshold filtering strategies to obtain the final traffic operation status index.

[0216] This module employs a hierarchical nested aggregation strategy, first calculating sub-indices based on traffic state impact dimensions (such as efficiency and load dimensions), and then weighting them a second time to obtain the total index. This structure not only outputs a comprehensive index but also provides sub-indices for each dimension, facilitating traffic managers to analyze congestion causes from multiple perspectives and improving the interpretability of the index. Simultaneously, a threshold filtering strategy combined with a first-order lag filtering formula smooths the initial total index, effectively suppressing unreasonable jumps between adjacent time points, resulting in good temporal continuity and stability in the index output. When a sudden change in real traffic conditions is detected, the system can dynamically adjust the smoothing coefficient, balancing noise resistance during stable periods and response speed during abrupt changes.

[0217] Through the modular design described above, the functions of each part of the system are clearly defined and the connections between them are explicit. It can be deployed independently as a complete system or embedded into existing traffic management platforms, exhibiting good compatibility and scalability. The advantages of each module synergize to achieve the technical effects of resisting extreme data interference, adaptive weight updates, exponential time-series smoothing, and multi-dimensional comprehensive evaluation.

[0218] The data preprocessing module 1 includes a quantile normalization unit 11, used to normalize the original indicator data sequence. Let n be the total number of data points, and calculate the p-th quantile of the sequence. and the qth quantile ,in , This indicates that the i-th original data point uses the formula Standardize to obtain the standardized value. and for those exceeding the interval of Perform truncation, that is, let ;

[0219] Nonlinear mapping correction unit 12, used to apply formula or formula The standardized data is then corrected to obtain the final corrected indicator data. ,in For power-corrected parameters and , For logarithmic correction parameters and .

[0220] The dynamic weight calculation module 2 includes a sliding window construction unit 21, which is used to set the length of the sliding window to L, where L is a positive integer, and to determine the adaptive length based on the sampling frequency f of the traffic data and the preset time window T. ,in The function represents the floor function, where T ranges from 5 minutes to 30 minutes. For time t, the evaluation matrix is ​​constructed from standardized index data of L time samples within the current sliding window. ,in This represents the standardized value of the j-th indicator for the i-th sample within the window, where m is the total number of evaluation indicators.

[0221] Information entropy calculation unit 22 is used to calculate the weight of the j-th indicator in the i-th sample. And calculate the information entropy of the j-th index. ;

[0222] Weight update unit 23 is used to calculate the dynamic weight of the j-th indicator. , where k is the index of the indicator, and the above calculation is repeated as the sliding window moves forward to achieve dynamic updating of the weights.

[0223] The index synthesis module 3 includes a hierarchical nested aggregation unit 31, which is used to divide the underlying indicators into multiple categories according to a preset traffic state influence dimension, and first calculate the sub-index of each category. ,in This represents the sub-index of the c-th category. The dynamic weight of the j-th indicator. For the j-th indicator, the data has been corrected. This represents the set of indicators belonging to the c-th category; then, the sub-indices are weighted and aggregated twice to obtain the total index. Where I is the initial total index and C is the total number of categories. The preset weight of the c-th sub-index or the weight dynamically calculated by the entropy weight method;

[0224] Threshold filtering unit 32 is used to filter by a first-order hysteresis filter formula. Exponential smoothing is performed to obtain the final traffic operation status index. ,in Let be the initial total exponent at the current time t. The final index at the previous moment. Smoothing coefficient and And set an incremental threshold. And the corresponding state level, where K is the number of thresholds.

[0225] It also includes a data validity verification module, connected between the data acquisition end and the data preprocessing module. This module performs integrity checks on multi-source traffic data, removes outlier data that exceeds physical limits or has abnormal rates of change, and fills in missing values ​​using linear interpolation or historical data from the same period. The multi-source traffic data includes at least two of the following: average road segment speed, traffic flow, time occupancy rate, headway, queue length, and travel time ratio. This data validity verification module performs integrity checks, outlier removal, and missing value filling on the raw multi-source traffic data, ensuring the quality of the input data from the source and avoiding exponential distortion caused by equipment failure or transmission anomalies.

[0226] The quantile-based robust standardization unit replaces traditional extreme value normalization, effectively suppressing the interference of extreme values ​​(outliers) on the data distribution, and making the standardization results more realistically reflect normal traffic conditions.

[0227] The nonlinear mapping correction unit performs nonlinear correction on standardized data using power or logarithmic functions. It can adjust the sensitivity of different intervals according to traffic flow perception characteristics (such as amplifying the differences in low-speed congestion areas) and improve the index's ability to identify congestion status.

[0228] The sliding window construction unit adaptively determines the window length (5~30 minutes), enabling subsequent weight calculations to flexibly match the dynamic characteristics of traffic flow under different scenarios, taking into account both short-term fluctuation capture and long-term trend smoothing.

[0229] The information entropy calculation unit quantitatively calculates the information entropy of each indicator within the sliding window, accurately measures the dispersion or fluctuation range of the indicators, and provides an objective basis for dynamic weight allocation.

[0230] The weight update unit updates the indicator weights in real time based on the information utility value, so that the indicators with greater fluctuations get higher weights in the index synthesis, thus solving the problem of lagging updates in traditional static weights.

[0231] The hierarchical nested aggregation unit first calculates sub-indices according to traffic impact dimensions (such as efficiency and load), and then performs a second weighting to obtain the total index. This can output a comprehensive index and provide multi-dimensional diagnostic information, enhancing the interpretability of the index.

[0232] The threshold filtering unit uses a first-order lag filter formula to smooth the initial total exponent and dynamically adjusts the smoothing coefficient to effectively suppress unreasonable jumps in the exponent, ensuring the continuity of the time sequence while taking into account the speed of sudden change response.

[0233] The aforementioned modules and units together achieve the technical effects of resisting extreme data interference, adaptive weight updates, exponential time-series smoothing, and multi-dimensional comprehensive evaluation.

[0234] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for constructing a quantile-robust standardization and adaptive entropy-weighted traffic index, characterized in that, Includes the following steps: Step S1, Data Preprocessing Step: Collect multi-source traffic data in real time, and perform robust standardization and nonlinear mapping correction processing based on quantiles on the multi-source traffic data to suppress the influence of extreme values ​​in the data and obtain standardized index data. Step S2, Dynamic Weight Calculation Step: Using the sliding window adaptive entropy weight method, based on the standardized index data in the current sliding window, the information entropy of each evaluation index is dynamically calculated and the weight is updated to obtain the dynamic weight of each index. Step S3, Index Synthesis Step: Based on the dynamic weights, the standardized index data is synthesized through hierarchical nested aggregation and threshold filtering strategies to obtain the final traffic operation status index.

2. The method according to claim 1, characterized in that, In step S1, the data preprocessing step is described as follows: The robust standardization based on quantiles includes: for the original indicator data sequence ,in Given the total number of data points, calculate the first number of data points in the sequence. quantiles and the quantiles ,in Using formula Standardize to obtain the standardized value. and for those exceeding the interval of Perform truncation, that is, let , Indicates the first One original data point; The nonlinear mapping correction includes: using the formula or formula The standardized data is then corrected to obtain the final corrected indicator data. ,in, For the power-corrected parameter, and , For logarithmic correction parameters, and .

3. The method according to claim 1, characterized in that, The dynamic weight calculation step in step S2 specifically includes: Set the length of the sliding window to Where L is a positive integer, for time t, standardized index data of L time samples within the current sliding window are taken to form an evaluation matrix. ,in, This represents the standardized value of the j-th indicator for the i-th sample within the window, where m is the total number of evaluation indicators. Calculate the weight of the j-th indicator in the i-th sample. ; Calculate the information entropy of the j-th indicator. ; Calculate the dynamic weight of the j-th indicator , where k is the index; As the sliding window moves forward, the above process is repeated to achieve dynamic updates of the weights; The length L of the sliding window is an adaptive length, determined based on the sampling frequency f of the traffic data and the preset time window T. ,in This represents the floor function. The value ranges from 5 minutes to 30 minutes.

4. The method according to claim 1, characterized in that, In the exponential synthesis step of step S3: The hierarchical nested aggregation includes: dividing the lower-level indicators into several categories according to a preset traffic state impact dimension, and first calculating the sub-index for each category. ,in Indicates the first Sub-indices of each category, For the first The dynamic weights of each indicator For the first The data for each indicator has been corrected. Indicates belonging to the first The system comprises a set of indicators for each category; then, the sub-indices are weighted and aggregated a second time to obtain the total index. Where I is the initial total index and C is the total number of categories. The preset weight of the c-th sub-index or the weight dynamically calculated by the entropy weight method; The threshold filtering strategy includes: setting a set of incremental threshold values. and the corresponding state level, where K is the number of thresholds, and the initial total index is calculated for the current time t. Compared with the final index of the previous moment If a comparison is made, If the state is unchanged, then the state remains unchanged; otherwise, the state is divided according to the threshold. The threshold filtering strategy uses a preset stability tolerance parameter to achieve exponential smoothing through the following first-order hysteresis filtering formula: ,in, The final traffic operation status index at time t. This is a smoothing coefficient, with a value range of (0, 1], which is dynamically adjusted when traffic conditions change abruptly. The value of .

5. The method according to claim 1, characterized in that, The multi-source traffic data includes at least two of the following: average speed of road segment, traffic flow, time occupancy rate, headway, queue length, and travel time ratio; and before the data preprocessing step, a data validity verification step is also included: performing integrity checks on the real-time collected multi-source traffic data, removing outlier data that exceeds the physical range or has an abnormal rate of change, and filling missing values ​​with linear interpolation or historical data from the same period.

6. A traffic operation status index construction system for implementing the methods of claims 1 to 5, characterized in that, include: The data preprocessing module is used to collect multi-source traffic data in real time and perform robust standardization and nonlinear mapping correction processing based on quantiles on the multi-source traffic data to suppress the influence of extreme values ​​in the data and obtain standardized index data. The dynamic weight calculation module, connected to the data preprocessing module, is used to dynamically calculate the information entropy of each evaluation indicator and update the weight based on the standardized indicator data in the current sliding window using the sliding window adaptive entropy weight method, so as to obtain the dynamic weight of each indicator. The index synthesis module is connected to the data preprocessing module and the dynamic weight calculation module, respectively. It is used to synthesize the standardized index data based on the dynamic weight through hierarchical nested aggregation and threshold filtering strategies to obtain the final traffic operation status index.

7. The system according to claim 6, characterized in that, The data preprocessing module includes a quantile normalization unit, used to process the original indicator data sequence. Let n be the total number of data points, and calculate the p-th quantile of the sequence. and the qth quantile ,in , This indicates that the i-th original data point uses the formula Standardize to obtain the standardized value. and for those exceeding the interval of Perform truncation, that is, let ; Nonlinear mapping correction unit, used to employ formula or formula The standardized data is then corrected to obtain the final corrected indicator data. ,in For power-corrected parameters and , For logarithmic correction parameters and .

8. The system according to claim 6, characterized in that, The dynamic weight calculation module includes a sliding window construction unit, used to set the length of the sliding window to L, where L is a positive integer, and to determine the adaptive length based on the sampling frequency f of the traffic data and the preset time window T. ,in The function represents the floor function, where T ranges from 5 minutes to 30 minutes. For time t, the evaluation matrix is ​​constructed from standardized index data of L time samples within the current sliding window. ,in This represents the standardized value of the j-th indicator for the i-th sample within the window, where m is the total number of evaluation indicators. The information entropy calculation unit is used to calculate the weight of the j-th indicator in the i-th sample. And calculate the information entropy of the j-th index. ; The weight update unit is used to calculate the dynamic weight of the j-th indicator. , where k is the index of the indicator, and the above calculation is repeated as the sliding window moves forward to achieve dynamic updating of the weights.

9. The system according to claim 6, characterized in that, The index synthesis module includes: The hierarchical nested aggregation unit is used to divide the underlying indicators into multiple categories according to the preset traffic state impact dimension. First, the sub-index of each category is calculated. ,in This represents the sub-index of the c-th category. The dynamic weight of the j-th indicator. For the j-th indicator, the data has been corrected. This represents the set of indicators belonging to the c-th category; then, the sub-indices are weighted and aggregated twice to obtain the total index. Where I is the initial total index and C is the total number of categories. The preset weight of the c-th sub-index or the weight dynamically calculated by the entropy weight method; Threshold filtering unit, used to filter by first-order hysteresis formula Exponential smoothing is performed to obtain the final traffic operation status index. ,in Let be the initial total exponent at the current time t. The final index at the previous moment. Smoothing coefficient and And set an incremental threshold. And the corresponding state level, where K is the number of thresholds.

10. The system according to claim 6, characterized in that, It also includes a data validity verification module, connected between the data acquisition end and the data preprocessing module, used to perform integrity checks on multi-source traffic data, remove outlier data that exceeds the physical range or has an abnormal rate of change, and fill missing values ​​using linear interpolation or historical data from the same period; and the multi-source traffic data includes at least two of the following: average speed of road segment, traffic flow, time occupancy rate, headway, queue length, and travel time ratio.

Images