A method for assessing the individual curb weight of highway trucks

Abstract

This invention discloses an individual curb weight assessment method for highway freight trucks, relating to the field of highway management. The method involves: extracting passage records for a specified time period from the highway weighing system using the vehicle's unique identifier as the key; cleaning and filtering the data to form an individual vehicle load history dataset; performing multimodality assessment, variability analysis, and gap clarity assessment on each individual dataset to obtain multimodality score, variability score, and gap clarity score, and then weighting them to obtain a weighted TDS (Total Dispersion Scale) comprehensive differentiation score; comparing the TDS comprehensive differentiation score with a preset threshold to select suitable vehicles for clustering; performing adaptive parameter clustering on suitable vehicles to obtain different load state clusters; calibrating the empty load state clusters and using the median of all data points in the empty load state cluster as the vehicle's curb weight assessment value; and combining the curb weight assessment value of each vehicle with the data from all load state clusters to generate an individualized load pattern profile. This invention achieves high-precision identification of vehicle curb weight.

Description

Technical Field

[0001] This invention relates to the field of highway management technology, and in particular to a method for assessing the individual curb weight of highway trucks. Background Technology

[0002] With the comprehensive coverage of ETC gantry systems and the improvement of non-stop weighing systems for trucks, traffic management departments have accumulated massive amounts of vehicle weighing and passage data. However, existing ETC and weighing systems mainly acquire the total mass of the vehicle and cargo, lacking a mechanism to effectively separate cargo weight from vehicle curb weight. This results in a persistent bottleneck of insufficient accuracy in cargo weight estimation during refined management and value analysis of freight transportation.

[0003] Existing highway freight transport technology solutions use fixed values ​​or empirical thresholds for simplified estimation, ignoring the heterogeneity among different brands, models, and ages of trucks. The estimation accuracy is insufficient to meet the needs of modern transportation for accurate statistics on freight transport and the management of overloading.

[0004] The existing load status detection technologies face the following problems: (1) Load identification technology based on vehicle sensors: Load status is estimated by monitoring the deformation of the vehicle suspension system. Although dynamic monitoring can be achieved, the hardware installation cost is high and it is difficult to promote and apply on a large scale. (2) Load identification technology based on GPS trajectory: Load status is identified by the characteristic that the speed of a loaded vehicle slows down when climbing a hill. However, the accuracy is easily affected by factors such as road slope, driver habits and environmental noise. (3) Computer vision technology: It provides a non-contact load estimation approach and analyzes the loading of the cargo box through image analysis. However, it is easily affected by environmental factors such as weather and lighting, and it is difficult to accurately distinguish the actual weight of different goods. (4) Problem of unified clustering strategy: The clustering strategy that processes all vehicles uniformly ignores the heterogeneity of the distribution of individual vehicle load data. The load patterns of different individuals of the same type of vehicle are significantly different. Some vehicle data are continuously distributed and do not show obvious weight boundaries of different states. They do not have the basic conditions for cluster analysis, which makes it impossible for the clustering algorithm to form a clear classification boundary of load status. The clustering results lose practical application value. Summary of the Invention

[0005] The purpose of this invention is to solve the core technical problem of insufficient accuracy in cargo weight estimation in highway freight systems due to the lack of a mechanism to effectively separate cargo weight from vehicle curb weight. This invention provides an individual curb weight assessment method for highway freight trucks, which quantitatively assesses the suitability of clustering based on the historical weighing data of individual vehicles, intelligently filters out vehicles with clear load status differentiation characteristics, and then applies an advanced density clustering algorithm only to these "high-value" data, thereby achieving high-precision identification of vehicle curb weight.

[0006] The technical solution adopted in this invention is:

[0007] A method for assessing the individual curb weight of highway trucks, comprising the following steps:

[0008] Step 1, Individualized Data Aggregation and Preprocessing: Using the vehicle's unique identifier as the key, extract the passage records within a specified time period from the highway weighing system, perform data cleaning and filtering, and form several independent individual vehicle load history datasets grouped by the vehicle's unique identifier.

[0009] Step 2, TDS three-dimensional differentiation assessment: For each individual dataset, multimodality assessment, variability analysis, and gap clarity analysis are performed to obtain the corresponding multimodal scores. Variation score and gap clarity score The three are weighted by preset weights. The TDS comprehensive differentiation score is obtained by weighted calculation. The formula for calculating the TDS comprehensive differentiation score is:

[0010] ;

[0011] in, .

[0012] Step 3, Vehicle screening based on evaluation results: Compare the TDS comprehensive differentiation score with the preset threshold to screen out vehicles suitable for clustering;

[0013] Step 4, Individualized clustering based on DBSCAN: Perform adaptive parameter clustering on vehicles with suitable clustering conditions to obtain clusters with different load states;

[0014] Step 5, Curb weight extraction: The cluster with the smallest average weight value of all data points within the cluster is designated as the unloaded state cluster; the median of all data points in the unloaded state cluster is used as the curb weight assessment value of the corresponding vehicle.

[0015] Step 6, Clustering Result Analysis: Based on the curb weight assessment value of each vehicle and the data of all load status clusters of each vehicle, generate an individualized load pattern profile for each vehicle.

[0016] Furthermore, the key data fields of the passage record in step 1 include vehicle identification, passage timestamp, total vehicle and cargo weight, and toll vehicle type code.

[0017] Furthermore, the data cleaning in step 1 includes removing records with zero weight, negative weight, or weights exceeding physical limits, and filtering vehicles with a pass count exceeding the minimum sample size requirement. Specifically, the minimum sample size requirement is a pass count exceeding 200.

[0018] Specifically, in step 2, the multimodal assessment refers to determining whether the data has obvious load state differentiation by analyzing the number and distribution characteristics of peaks in the load data distribution.

[0019] Furthermore, in step 2, the multimodality assessment detects the peak value of the load distribution through kernel density estimation. The peak value detection conditions include a relative height threshold and a minimum distance threshold. The calculation steps for the multimodality score are as follows:

[0020] Step 2-1-1: Estimate the probability density function by using Gaussian kernel density to calculate the vehicle load data;

[0021] The expression for the probability density function is:

[0022] (1);

[0023] in, It is a probability density function, representing the probability density function of weight. The probability density at that location; It refers to the sample size, which is the total number of passage records in the historical data of individual vehicle loads; It is a bandwidth parameter that controls the smoothness of kernel density estimation. It is a Gaussian kernel function; It is the first The total weight of the vehicle and cargo in each passage record; It is a weight value variable used to assess density.

[0024] Step 2-1-2: Obtain peak point data on the probability density function to form multimodal characteristics.

[0025] The detection conditions for peak points are:

[0026] (2);

[0027] in, It is a relative height threshold. It is the minimum distance threshold; It is a set used to store all data points that pass the filtering criteria and are ultimately identified as "significant peaks". . It is a relative height threshold. It is the minimum distance threshold. It is the absolute difference between two weight values.

[0028] Step 2-1-3: Calculate the multimodality score based on the number of peaks, height ratio, and inter-peak distance.

[0029] Specifically, bimodal distribution ( The score is awarded based on the ratio of peak heights and the distance between them, ranging from 60 to 100 points. The closer the peak heights and the farther the distance between the peaks, the higher the score. (Three-peak distribution) A score is assigned based on peak uniformity (e.g., assessing the dispersion of peak height and peak spacing), such as 60 to 90 points; unimodal ( or an overly complex multimodal distribution ( This indicates poor differentiation in load conditions, and a lower score, i.e., below 60 points, is given.

[0030] Furthermore, in step 2, the dispersion and variation of the load data are evaluated to conduct a variability analysis; that is, the statistical indicators of the coefficient of variation, range ratio and quartile ratio of the load data are calculated and obtained respectively, and the variability score is obtained by weighted calculation.

[0031] coefficient of variation The calculation formula is as follows:

[0032] (3);

[0033] Range Ratio The calculation formula is as follows:

[0034] (4);

[0035] Quartile proportions The calculation formula is as follows:

[0036] (5);

[0037] The formula for calculating the variability score is as follows:

[0038] (6);

[0039] in, It is the standard deviation of the load data. It is the mean. These are the 25th, 50th, and 75th percentiles, respectively. The variability score reflects the richness of load condition variations; the greater the variability, the more complex the load pattern, and the more suitable it is for cluster analysis.

[0040] Furthermore, the calculation steps for the gap sharpness score are as follows:

[0041] Specifically, gap clarity assessment identifies natural gaps in load distribution through histogram analysis.

[0042] Step 2-2-1: Divide the weighing data h into m equal-width intervals, calculate the frequency of each interval, identify the zero-frequency interval as the gap, and calculate the relative width of the maximum continuous gap.

[0043] (7);

[0044] in, It is the relative width of the maximum continuous gap. It is a continuous gap (consisting of one or more consecutive zero-frequency intervals). It is the set of all continuous gaps. It is a gap The end position (weight value). It is a gap The starting position (weight value). It is the maximum value in the load data. It is the minimum value in the load data.

[0045] Step 2-2-2: Calculate the balance coefficient of the data on both sides of the maximum continuous gap. ;

[0046] (8);

[0047] in, and These represent the number of data points on the left and right sides of the gap, respectively.

[0048] Step 2-2-3: Calculate the gap sharpness score. The calculation expression is as follows:

[0049] ;

[0050] in, and The preset weighting coefficients, and .

[0051] Specifically, the gap clarity score takes into account both gap width and balance, providing a quantitative basis for the feasibility of subsequent cluster analysis.

[0052] Furthermore, in step 3, a pre-set suitability threshold is obtained ( ), determine whether the TDS comprehensive differentiation score exceeds the suitability threshold; if so, determine that the load data of the corresponding vehicle has good differentiation and clear clustering characteristics, mark the corresponding vehicle as "suitable for clustering", and proceed to step 4; otherwise, determine that the corresponding vehicle data is not suitable for clustering, mark it as "unsuitable for clustering", and terminate the calculation process for the corresponding vehicle.

[0053] Furthermore, the DBSCAN clustering parameters in step 4 include the neighborhood radius ε and the minimum density threshold minPts, where ε is an optimal value automatically determined based on the k-distance graph, and the minimum density threshold minPts is set according to the data dimension.

[0054] Specifically, the neighborhood radius ( ) can usually be based on the k-distance graph of the data ( This automatically determines an optimal value that reflects the local density of the data points. Minimum density threshold ( This defines the minimum number of points required to form a high-density region, typically set based on the data dimensionality and noise tolerance (e.g., ...). ).

[0055] Furthermore, step 4 includes the following steps:

[0056] Step 4-1: For clustered vehicles, an optimal value determined based on the k-distance graph is used as the neighborhood radius ε; at the same time, a minimum density threshold minPts is set according to the dimension of the load data of clustered vehicles.

[0057] Step 4-2: Check the number of points in the ε-neighborhood of the unvisited data points; determine if the number of points is greater than or equal to the minimum density threshold minPts; if so, create a new cluster for the corresponding data point as the core point and execute step 4-3; otherwise, execute step 4-4.

[0058] Step 4-3: All data points reachable from the corresponding core point (including other core points and boundary points) are assigned to the corresponding cluster based on density reachability.

[0059] Step 4-4: Determine whether all points have been visited; if so, mark the points that cannot be assigned to any cluster as noise points; otherwise, obtain the next unvisited data point and execute step 4-2.

[0060] Furthermore, step 5 specifically includes the following steps:

[0061] Step 5-1: Calculate the average weight value of all data points in each cluster, and sort all clusters from low to high according to the average weight value;

[0062] Step 5-2: Define the state of the sorted clusters; designate the cluster with the lowest average weight value as the "unloaded state cluster" of the corresponding vehicle.

[0063] Step 5-3: Calculate the (arithmetic mean or) median of all data points within the unloaded state cluster, and use it as the curb weight assessment value for the corresponding vehicle.

[0064] Furthermore, other clusters are labeled as different “cargo state clusters” according to the sorting; other cargo states include “light load state”, “half load state”, and “full load state”, which fully depict the typical load mode hierarchy of the vehicle.

[0065] Furthermore, the curb weight calculation in step 5 uses the median of the unloaded state cluster as an estimate.

[0066] Furthermore, the individualized load pattern profile in step 6 includes curb weight, load condition type, and average load weight for each condition.

[0067] This invention employs the above technical solutions and innovatively proposes a personalized adaptive analysis framework of "assessment first, clustering later". This framework abandons the "one-size-fits-all" data processing mode and instead focuses on the unique load behavior pattern of each vehicle. First, it quantitatively assesses the clustering suitability of historical weighing data of individual vehicles, intelligently selecting vehicles with clear load status differentiation characteristics. Then, it applies advanced density clustering algorithms only to these "high-value" data, thereby achieving high-precision identification of vehicle curb weight. Attached Figure Description

[0068] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments;

[0069] Figure 1 This is a flowchart illustrating an individual curb weight assessment method for highway trucks according to the present invention.

[0070] Figure 2 This is a schematic diagram of the calculation process for TDS differentiation analysis in this invention;

[0071] Figure 3 This is a schematic diagram of the clustering algorithm processing of the present invention;

[0072] Figure 4 This is a visualization diagram of the clustering results of different types of trucks in this invention. Detailed Implementation

[0073] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0074] Traditional estimation methods ignore vehicle heterogeneity, existing technologies (such as vehicle sensors, GPS, and computer vision) have limitations such as high cost or susceptibility to environmental interference, and traditional clustering algorithms (such as K-Means and GMM) have difficulty handling common non-convex shapes, noise points, and individual distribution differences in weighing data, resulting in technical bottlenecks that make clustering results lose their practical application value.

[0075] like Figures 1 to 4 As shown in one example, this invention discloses a method for assessing the individual curb weight of highway trucks, which includes the following steps:

[0076] Step 1, Individualized Data Aggregation and Preprocessing: Using the vehicle's unique identifier as the key, extract the passage records within a specified time period from the highway weighing system, perform data cleaning and filtering, and form several independent individual vehicle load history datasets grouped by the vehicle's unique identifier.

[0077] Step 2, TDS three-dimensional differentiation assessment: For each individual dataset, multimodality assessment, variability analysis, and gap clarity assessment are performed to obtain the corresponding multimodal scores. Variation score and gap clarity score The three are weighted by preset weights. The TDS comprehensive differentiation score is obtained by weighted calculation. The formula for calculating the TDS comprehensive differentiation score is:

[0078] ;

[0079] in, .

[0080] Step 3, Vehicle screening based on evaluation results: Compare the TDS comprehensive differentiation score with the preset threshold to screen out vehicles suitable for clustering;

[0081] Step 4, Individualized clustering based on DBSCAN: Perform adaptive parameter clustering on vehicles with suitable clustering conditions to obtain clusters with different load states;

[0082] Step 5, Curb weight extraction: The cluster with the smallest average weight value of all data points within the cluster is designated as the unloaded state cluster; the median of all data points in the unloaded state cluster is used as the curb weight assessment value of the corresponding vehicle.

[0083] Step 6, Clustering Result Analysis: Based on the curb weight assessment value of each vehicle and the data of all load status clusters of each vehicle, generate an individualized load pattern profile for each vehicle.

[0084] Furthermore, the key data fields of the passage record in step 1 include vehicle identification, passage timestamp, total vehicle and cargo weight, and toll vehicle type code.

[0085] Furthermore, the data cleaning in step 1 includes removing records with zero weight, negative weight, or weights exceeding physical limits, and filtering vehicles with a pass count exceeding the minimum sample size requirement. Specifically, the minimum sample size requirement is a pass count exceeding 200.

[0086] Specifically, in step 2, the multimodal assessment refers to determining whether the data has obvious load state differentiation by analyzing the number and distribution characteristics of peaks in the load data distribution.

[0087] Furthermore, in step 2, the multimodality assessment detects the peak value of the load distribution through kernel density estimation. The peak value detection conditions include a relative height threshold and a minimum distance threshold. The calculation steps for the multimodality score are as follows:

[0088] Step 2-1-1: Estimate the probability density function by using Gaussian kernel density to calculate the vehicle load data;

[0089] The expression for the probability density function is:

[0090] (1);

[0091] in, It is a probability density function, representing the probability density function of weight. The probability density at that location; It refers to the sample size, which is the total number of passage records in the historical data of individual vehicle loads; It is a bandwidth parameter that controls the smoothness of kernel density estimation. It is a Gaussian kernel function; It is the first The total weight of the vehicle and cargo in each passage record; It is a weight value variable used to assess density.

[0092] Step 2-1-2: Obtain peak point data on the probability density function to form multimodal characteristics.

[0093] The detection conditions for peak points are:

[0094] (2);

[0095] in, It is a relative height threshold. It is the minimum distance threshold; It is a set used to store all data points that pass the filtering criteria and are ultimately identified as "significant peaks". . It is a relative height threshold. It is the minimum distance threshold. It is the absolute difference between two weight values.

[0096] Step 2-1-3: Calculate the multimodality score based on the number of peaks, height ratio, and inter-peak distance.

[0097] Specifically, bimodal distribution ( The score is awarded based on the ratio of peak heights and the distance between them, ranging from 60 to 100 points. The closer the peak heights and the farther the distance between the peaks, the higher the score. (Three-peak distribution) A score is assigned based on peak uniformity (e.g., assessing the dispersion of peak height and peak spacing), such as 60 to 90 points; unimodal ( or an overly complex multimodal distribution ( This indicates poor differentiation in load conditions, resulting in a lower score, below 60. Further, step 2 involves evaluating the dispersion and magnitude of load data to conduct a variability analysis; specifically, the coefficient of variation, range proportion, and quartile proportion of the load data are calculated separately, and then weighted to obtain the variability score.

[0098] coefficient of variation The calculation formula is as follows:

[0099] (3);

[0100] Range Ratio The calculation formula is as follows:

[0101] (4);

[0102] Quartile proportions The calculation formula is as follows:

[0103] (5);

[0104] The formula for calculating the variability score is as follows:

[0105] (6);

[0106] in, It is the standard deviation of the load data. It is the mean. These are the 25th, 50th, and 75th percentiles, respectively. It is the maximum value in the load data. It is the minimum value in the load data.

[0107] The variability score reflects the richness of load condition changes. The greater the variability, the more complex the load pattern, and the more suitable it is for cluster analysis.

[0108] Furthermore, the calculation steps for the gap sharpness score are as follows:

[0109] Specifically, gap clarity assessment identifies natural gaps in load distribution through histogram analysis.

[0110] Step 2-2-1: Divide the weighing data h into m equal-width intervals, calculate the frequency of each interval, identify the zero-frequency interval as the gap, and calculate the relative width of the maximum continuous gap.

[0111] (7);

[0112] in, It is the relative width of the maximum continuous gap. It is a continuous gap (consisting of one or more consecutive zero-frequency intervals). It is the set of all continuous gaps. It is a gap The end position (weight value). It is a gap The starting position (weight value). It is the maximum value in the load data. It is the minimum value in the load data.

[0113] Step 2-2-2: Calculate the balance coefficient of the data on both sides of the maximum continuous gap. ;

[0114] (8);

[0115] in, and These represent the number of data points on the left and right sides of the gap, respectively.

[0116] Step 2-2-3: Calculate the gap sharpness score. The calculation expression is as follows:

[0117] ;

[0118] in, and The preset weighting coefficients, and .

[0119] Specifically, the gap clarity score takes into account both gap width and balance, providing a quantitative basis for the feasibility of subsequent cluster analysis.

[0120] Furthermore, in step 3, a pre-set suitability threshold is obtained ( ), determine whether the TDS comprehensive differentiation score exceeds the suitability threshold; if so, determine that the load data of the corresponding vehicle has good differentiation and clear clustering characteristics, mark the corresponding vehicle as "suitable for clustering", and proceed to step 4; otherwise, determine that the corresponding vehicle data is not suitable for clustering, mark it as "unsuitable for clustering", and terminate the calculation process for the corresponding vehicle.

[0121] Furthermore, the DBSCAN clustering parameters in step 4 include the neighborhood radius ε and the minimum density threshold minPts, where ε is an optimal value automatically determined based on the k-distance graph, and the minimum density threshold minPts is set according to the data dimension.

[0122] Specifically, the neighborhood radius ( ) can usually be based on the k-distance graph of the data ( This automatically determines an optimal value that reflects the local density of the data points. Minimum density threshold ( This defines the minimum number of points required to form a high-density region, typically set based on the data dimensionality and noise tolerance (e.g., ...). ).

[0123] Furthermore, step 4 includes the following steps:

[0124] Step 4-1: For clustered vehicles, an optimal value determined based on the k-distance graph is used as the neighborhood radius ε; at the same time, a minimum density threshold minPts is set according to the dimension of the load data of clustered vehicles.

[0125] Step 4-2: Check the number of points in the ε-neighborhood of the unvisited data points; determine if the number of points is greater than or equal to the minimum density threshold minPts; if so, create a new cluster for the corresponding data point as the core point and execute step 4-3; otherwise, execute step 4-4.

[0126] Step 4-3: All data points reachable from the corresponding core point (including other core points and boundary points) are assigned to the corresponding cluster based on density reachability.

[0127] Step 4-4: Determine whether all points have been visited; if so, mark the points that cannot be assigned to any cluster as noise points; otherwise, obtain the next unvisited data point and execute step 4-2.

[0128] Furthermore, step 5 specifically includes the following steps:

[0129] Step 5-1: Calculate the average weight value of all data points in each cluster, and sort all clusters from low to high according to the average weight value;

[0130] Step 5-2: Define the state of the sorted clusters; designate the cluster with the lowest average weight value as the "unloaded state cluster" of the corresponding vehicle.

[0131] Step 5-3: Calculate the (arithmetic mean or) median of all data points within the unloaded state cluster, and use it as the curb weight assessment value for the corresponding vehicle.

[0132] Furthermore, other clusters are labeled as different “cargo state clusters” according to the sorting; other cargo states include “light load state”, “half load state”, and “full load state”, which fully depict the typical load mode hierarchy of the vehicle.

[0133] Furthermore, the curb weight calculation in step 5 uses the median of the unloaded state cluster as an estimate.

[0134] Furthermore, the individualized load pattern profile in step 6 includes curb weight, load condition type, and average load weight for each condition.

[0135] The specific principles of this invention will be explained in detail below:

[0136] Individualized data modeling and preprocessing: Using the vehicle's unique identifier (such as license plate number) as the key, extract all passage records for the vehicle within a specified time period (the past 6-12 months) from the backend database of the highway ETC gantry system and non-stop weighing system. Key data fields include vehicle identifier, passage timestamp, total vehicle and cargo weight, and toll vehicle type code. Group all records by vehicle identifier to form N independent individual vehicle load history datasets.

[0137] Data cleaning and filtering: A rigorous cleaning process is performed on each individual dataset. First, obvious outliers are removed, such as weights of zero or negative, or weights far exceeding the physical limits of the vehicle type. Second, to ensure the effectiveness and stability of the statistical analysis, vehicles with a total number of passage records that meet the minimum sample size requirement (e.g., more than 200 passages) are selected; vehicles with insufficient data will not be included in subsequent calculations.

[0138] Constructing a TDS differentiation assessment system for suitability screening: This is one of the core innovations of this invention. This invention constructs a three-dimensional TDS assessment system that includes multimodality assessment (determining the degree of load state differentiation through kernel density estimation), variability analysis (assessing the richness of load variation through coefficient of variation, range ratio, etc.), and gap clarity assessment (identifying the clarity of state boundaries through histogram analysis). This system quantitatively scores the load data of each vehicle to scientifically screen vehicles with obvious load state differentiation characteristics suitable for cluster analysis.

[0139] Multimodality assessment refers to determining whether the load data exhibits significant load state differentiation by analyzing the number and distribution characteristics of peaks in the load data distribution. Kernel density estimation is used to detect the peak characteristics of the vehicle load distribution. For vehicle load data, Gaussian kernel density is used to estimate the probability density function.

[0140] (1);

[0141] in, It is a probability density function, representing the probability density function of weight. The probability density at that location; It refers to the sample size, which is the total number of passage records in the historical data of individual vehicle loads; It is a bandwidth parameter that controls the smoothness of kernel density estimation. It is a Gaussian kernel function; It is the first The total weight of the vehicle and cargo in each passage record; It is a weight value variable used to assess density.

[0142] (2);

[0143] Where α is the relative height threshold. It is the minimum distance threshold. It is a set used to store all data points that pass the filtering criteria and are ultimately identified as "significant peaks". . It is a relative height threshold. It is the minimum distance threshold. It is the absolute difference between two weight values.

[0144] Multimodality scores are calculated based on the number of peaks, height ratio, and interpeak distance. Bimodal distributions are given 60 to 100 points based on peak height ratio and distance, trimodal distributions are given corresponding scores based on peak uniformity, and unimodal or excessively complex multimodal distributions are given lower scores.

[0145] Variance analysis is a crucial step in assessing the dispersion and magnitude of changes in load data. It comprehensively evaluates the dispersion of load data through multiple statistical indicators. Key indicators include:

[0146] Coefficient of variation:

[0147] (3);

[0148] Range Ratio:

[0149] (4);

[0150] Quartile proportions:

[0151] (5)

[0152] in, It is the standard deviation of the load data. It is the mean. These are the 25th, 50th, and 75th percentiles, respectively. The overall variability score is a weighted combination of the three indicators:

[0153] (6);

[0154] The variability score reflects the richness of load condition changes. The greater the variability, the more complex the load pattern, and the more suitable it is for cluster analysis.

[0155] Gap clarity assessment identifies natural gaps in the load distribution through histogram analysis. The weighing data h is divided into m equal-width intervals, the frequency of each interval is calculated, the zero-frequency interval is identified as the gap, and the relative width of the largest continuous gap is calculated.

[0156] (7);

[0157] Simultaneously assess the balance of data on both sides of the gap:

[0158] (8);

[0159] in, and These represent the number of data points on the left and right sides of the gap, respectively. The gap sharpness score comprehensively considers both gap width and balance, providing a quantitative basis for the feasibility of subsequent cluster analysis. The formula for calculating the gap sharpness score is as follows:

[0160] ;

[0161] in, and The preset weighting coefficients, and .

[0162] The scores of the three dimensions of multimodality, variability, and gap clarity are weighted and fused to obtain the final TDS comprehensive differentiation score.

[0163] Vehicle screening based on evaluation results is a decision-making step to ensure that only high-quality data enters the core algorithm. The calculated TDS comprehensive differentiation score is then compared with a pre-set suitability threshold. The data is compared. If the score is higher than the threshold, the vehicle's load data is considered to have good differentiation and clear clustering characteristics. It is then marked as a "suitable clustering vehicle" and proceeds to the next step (individualized clustering analysis based on the DBSCAN algorithm). If the score is lower than the threshold, the vehicle's data is considered unsuitable for clustering. It is then marked as a "unsuitable clustering vehicle," and the calculation process for that vehicle is terminated.

[0164] Individualized Clustering Analysis Based on DBSCAN Algorithm: For vehicles selected through TDS evaluation, the density-based DBSCAN clustering algorithm is used to independently cluster their individual load data. This algorithm can effectively identify clusters of arbitrary shapes and has natural robustness to noise points, perfectly matching the "high density, low variability" regional characteristics of weighing data under no-load conditions in high-dimensional space.

[0165] Specifically, clustering operation parameters are adaptively set through the selected vehicle data: for each "suitable clustering vehicle", two core parameters of the DBSCAN algorithm are set independently: (1) neighborhood radius ( ): Typically, it can be based on a k-distance plot of the data ( (2) Minimum density threshold () to automatically determine an optimal value that reflects the local density of the data points. This parameter defines the minimum number of points required to form a high-density region, and is typically set based on the data dimensionality and noise tolerance (e.g., ).

[0166] The clustering algorithm starts with any unvisited data point and checks the number of points in its ε-neighborhood. If the number of points is greater than or equal to minPts, then that point is a core point, and a new cluster is created. Subsequently, based on density reachability, all data points reachable from that core point (including other core points and boundary points) are assigned to this cluster. This process is repeated until all points have been visited. Points that cannot be assigned to any cluster are marked as noise points.

[0167] Precise extraction of curb weight: After clustering, the algorithm automatically identifies the cluster with the smallest mass value and the highest data point density among all clusters as the unloaded state cluster of the vehicle. By calculating the mean or median of all data points within this cluster, the final estimated curb weight of the vehicle is determined, thus completing the precise separation of the vehicle's tare weight from the total vehicle and cargo weight. The final step of this invention aims to professionally analyze the output results of the clustering algorithm, identify various typical load states of the vehicle, and accurately extract its curb weight based on these. Specifically, this includes the following aspects:

[0168] (1) Cluster Identification and State Labeling: The output of the DBSCAN algorithm is a set (one or more) of high-density data clusters and possible noise points. Each cluster represents a stable load state of the vehicle during operation. First, the average weight value of all data points in each cluster is calculated. Then, all clusters are sorted from low to high according to the average weight value. Based on engineering common sense and physical meaning, the state of these sorted clusters is labeled: Empty state (curb weight): The cluster with the lowest average weight value is labeled as the "empty state cluster" of the vehicle. The weight distribution of this cluster represents the curb weight of the vehicle. Cargo state: The remaining clusters with higher average weight values ​​are labeled as different "cargo state clusters" in turn. According to the vehicle type and the actual clustering results, these states may be further refined into "light load state", "half load state", "full load state", etc., to fully depict the typical load mode hierarchy of the vehicle.

[0169] (2) Precise calculation of curb weight: After successfully calibrating the "unloaded state cluster", the arithmetic mean or median of all data points in the cluster is precisely calculated. Considering the possible slight fluctuations in the weighing data, using the median as the final result can better resist the influence of individual extreme values ​​within the cluster, and the result is more robust. This calculation result is the final output of this invention, the estimated curb weight of the vehicle.

[0170] (3) Generation and Output of Individualized Load Pattern Profile: The output of this invention is not just a single curb weight value. By comprehensively analyzing all state clusters (including empty and various loaded states), an "individualized load pattern profile" can be generated for each vehicle. This profile contains rich information such as the vehicle's precise curb weight, the number of typical loaded states, and the average load weight in each state. Finally, the vehicle's estimated curb weight and its complete load pattern profile data are associated with the vehicle's unique identifier and stored in the results database for subsequent querying, in-depth analysis, and diversified applications.

[0171] Table 1 Weight distribution

[0172]

[0173] Table 2 Performance Comparison of Different Clustering Algorithms

[0174]

[0175] This invention employs the above technical solutions and innovatively proposes a personalized adaptive analysis framework of "assessment first, clustering later". This framework abandons the "one-size-fits-all" data processing mode and instead focuses on the unique load behavior pattern of each vehicle. First, it quantitatively assesses the clustering suitability of historical weighing data of individual vehicles, intelligently selecting vehicles with clear load status differentiation characteristics. Then, it applies advanced density clustering algorithms only to these "high-value" data, thereby achieving high-precision identification of vehicle curb weight.

[0176] Compared with existing technologies, the technical solution of the present invention has the following advantages and beneficial effects:

[0177] 1. Improved estimation accuracy and reliability: This invention ensures the effectiveness of the analysis through a "first evaluation and screening, then precise clustering" model. As shown in Table 2, experiments demonstrate that this invention outperforms traditional algorithms in key indicators such as profile coefficient and CH index, and can more accurately and compactly aggregate the curb weight of trucks. The reliability of the estimation results is superior to traditional fixed-value or simplified models.

[0178] 2. Achieved Individualized Adaptive Analysis: This invention solves the "state confusion" problem caused by the neglect of vehicle heterogeneity in traditional unified clustering strategies. By independently modeling and analyzing each vehicle, it can adapt to different vehicle operating modes and load characteristics, improving the universality and analytical depth of the method.

[0179] 3. Scientific Screening Mechanism Before Clustering: The TDS differentiation assessment system is the core innovation of this invention, providing a scientific prerequisite for the effectiveness of cluster analysis. This mechanism can automatically identify and filter out "low-quality" data with continuous data distribution and no clear state boundaries, ensuring that computing resources are used for valuable analysis objects and guaranteeing the effectiveness of the final results.

[0180] 4. Strong algorithm adaptability and no additional hardware investment required: The DBSCAN algorithm does not require a preset number of clusters, can handle data distributions of arbitrary shapes, is robust to noisy data, and can adapt to complex weighing data patterns in the real world. This invention is entirely based on existing ETC and weighing data, requires no installation of any vehicle-mounted sensors or vision devices, does not increase hardware costs, and has good economic efficiency and feasibility for large-scale promotion and application.

[0181] Obviously, the described embodiments are only a part of the embodiments of this application, not all of them. Without conflict, the embodiments and features in the embodiments of this application can be combined with each other. The components of the embodiments of this application described and illustrated herein can generally be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of this application is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

Claims

1. A method of estimating individual kerb mass of a motorway truck, characterised by: It comprises the following steps: Step 1, individualized data aggregation and preprocessing: extracting the passing record of the specified time period from the highway weighing system with the vehicle unique identification as the key, data cleaning and screening, forming a number of independent individual vehicle load history data sets grouped by vehicle unique identification; Step 2, TDS three-dimensional differentiation assessment: For each individual dataset, multimodality assessment, variability analysis, and gap clarity analysis are performed to obtain the corresponding multimodal scores. Variation score and gap clarity score By presetting weights The weighted calculation yields the TDS comprehensive differentiation score. The formula for calculating the TDS comprehensive differentiation score is: ； wherein ; The calculation steps of the multi-peak score in step 2 are as follows: Step 2-1-1, estimate the probability density function of the vehicle load data by Gaussian kernel density, the expression is: ；（1）； wherein, is a probability density function representing the probability density at the weight value variable ; is the sample size, i.e. the total number of passages in the individual vehicle load history data set; is a bandwidth parameter controlling the degree of smoothing of the kernel density estimate; is a Gaussian kernel function; is the total mass of the vehicle and cargo of the th passage; is the weight value variable used for the density evaluation; Step 2-1-2, obtain the peak point data on the probability density function to form the multi-peak feature, the detection condition of the peak point is: ；（2）； wherein, is a set for storing all data points that pass the filtering criteria and are eventually identified as "significant peaks" ; is a relative height threshold, is a minimum distance threshold, is an absolute difference between two weight values; Step 2-1-3, the multiplet score is calculated according to the peak number, height ratio and distance between peaks ; The calculation steps of the gap clarity score are as follows: Step 2-2-1, divide the weighing data h into m equal width intervals, calculate the frequency of each interval, identify the zero frequency interval as the gap, and calculate the relative width of the maximum continuous gap; （7）； wherein, is the relative width of the maximum continuous gap; is a continuous gap; is the set of all continuous gaps; is the end position of the gap corresponding to the weight value; is the start position of the gap corresponding to the weight value; is the maximum value in the weight data; is the minimum value in the weight data; Step 2-2-2, calculate the balance coefficient of the data on both sides of the maximum continuous gap ; （8）； wherein, and are the number of data points on the left and right side of the gap, respectively. Step 2-2-3, calculate the gap clarity score, the expression is as follows: ； wherein, and are preset weight coefficients, and ; Step 3, vehicle screening based on the evaluation result: comparing the TDS comprehensive differentiation score with the preset threshold, screening out the clustering suitable vehicle; Step 4, individualized clustering based on DBSCAN: performing adaptive parameter clustering on the clustering suitable vehicle, and clustering to obtain different load state clusters; Step 5, tare mass extraction: the cluster with the minimum average weight value of all data points in the cluster is marked as the empty load state cluster; the median of all data points in the empty load state cluster is taken as the tare mass evaluation value of the corresponding vehicle; Step 6, clustering result analysis: according to the tare mass evaluation value of each vehicle and all load state cluster data of each vehicle, an individualized load mode image is generated for each vehicle.

2. The method of estimating the individual kerb mass of a motorway truck according to claim 1, characterised in that: The key data fields of the passing record in step 1 include vehicle identification, passing timestamp, total mass of vehicle and cargo, and toll vehicle type code; data cleaning includes removing records with zero, negative or physical limit weight, and screening vehicles with passing frequency higher than the minimum sample size requirement, which is higher than 200 times.

3. The method of claim 1, wherein: In step 2, the statistical indicators of the coefficient of variation, the range ratio and the quartile ratio of the load data are calculated respectively, and the variation score is calculated by weighting; Coefficient of variation The formula for calculating the coefficient of variation is as follows: （3）； range ratio The formula for calculating the range ratio is as follows: （4）； Quartile Proportions The formula for calculating the quartile proportions is as follows: （5）； The calculation formula of the variation score is as follows: （6）； wherein, is the standard deviation of the load data, is the mean of the load data, are the 25th, 50th, 75th percentiles, respectively, is the maximum value in the load data; is the minimum value in the load data.

4. The method of estimating individual kerb mass of a truck on a highway as claimed in claim 1, wherein: In step 3, the pre-set suitability threshold is obtained, and it is judged whether the TDS comprehensive differentiation score exceeds the suitability threshold; if yes, it is judged that the load data of the corresponding vehicle is well differentiated and has clear clustering characteristics, and the corresponding vehicle is marked as "clustering suitable vehicle" and enters step 4; Otherwise, it is judged that the data of the corresponding vehicle is not suitable for clustering, and it is marked as "clustering unsuitable vehicle" and the calculation process of the corresponding vehicle is terminated.

5. The method of claim 1, wherein: Step 4 comprises the following steps: Step 4-1, determine an optimal value based on k-distance map as the neighborhood radius ε for the clustering suitable vehicle; at the same time, set the minimum density threshold minPts according to the dimension of the load data of the clustering suitable vehicle; Step 4-2, check the number of points in the ε-neighborhood of the unvisited data point; judge whether the number of points is greater than or equal to the minimum density threshold minPts; if yes, the corresponding data point is a core point, a new cluster is created and step 4-3 is executed; otherwise, step 4-4 is executed; Step 4-3, all data points reachable from the corresponding core point are classified into the corresponding cluster by density reachable relationship; Step 4-4, judge whether all points are accessed; if yes, the points that cannot be classified into any cluster are marked as noise points; otherwise, the next unvisited data point is obtained and step 4-2 is executed.

6. The method of estimating the individual kerb mass of a motorway truck according to claim 1, characterized in that: Step 5 specifically comprises the following steps: Step 5-1, calculate the average weight value of all data points in each cluster, and sort all clusters from low to high according to the average weight value; Step 5-2, state calibration is performed on the sorted clusters; the cluster with the lowest average weight value is calibrated as the "empty state cluster" of the corresponding vehicle; Step 5-3, calculate the arithmetic mean or median of all data points in the empty state cluster as the estimated value of the gross weight of the corresponding vehicle.

7. The method of claim 6, wherein: In step 5, the median of the empty state cluster is taken as the estimated value, and other clusters are calibrated as other different "loaded state clusters" according to the sorting; other loaded states include "light load state", "half load state" and "full load state".

8. The method of estimating the individual kerb mass of a motorway truck according to claim 1, characterized in that: In step 6, the individualized load mode image includes the gross weight, the load state category and the average load weight of each state.

Images