A multi-source traffic data processing method based on confidence evaluation

Abstract

This invention discloses a multi-source traffic data processing method based on confidence assessment. It is used to fuse data collected at the same spatiotemporal granularity for the same traffic characteristic parameter and aggregate data for different traffic characteristic parameters to obtain various traffic flow indicators for a spatial research unit. The method first calibrates the confidence levels of traffic characteristic parameters from different detection systems. For directly obtainable traffic flow indicators, the traffic characteristic value provided by the detector with the highest confidence level is selected as the traffic flow indicator for the spatial research unit. For traffic flow indicators that cannot be directly obtained, the traffic flow indicator for the spatial research unit is derived from the traffic characteristic value. This method solves the problems of isolated operation and data fusion application of different types of traffic detection systems in urban intelligent transportation systems, thereby improving data utilization and availability.

Description

Technical Field

[0001] The present invention relates to the field of urban intelligent traffic data collection and application, and in particular to a multi-source traffic data processing method based on confidence evaluation. Background Art

[0002] After more than a decade of continuous construction and development of urban intelligent transportation systems, the coexistence of detection systems such as electric police, checkpoints, video, and microwave systems is a common phenomenon in various cities. Making full use of various detection data to achieve complementary integration is an effective means to reduce costs and increase efficiency of urban intelligent transportation systems and obtain reliable traffic detection data. The current research directions of data fusion technology for multi-sensor systems are concentrated in the following two aspects:

[0003] First, data-level fusion of raw information can yield more accurate traffic detection results and be applied to deep data mining to improve detection performance. However, due to large data volumes or the instability of raw sensor data, computer processing time is long, real-time performance and anti-interference capabilities are weak, and accuracy is relatively reduced.

[0004] Second, feature-level fusion of output information can improve the real-time performance of the fusion process and reduce the interference of errors in the original data on the fusion results. Existing research on feature-level fusion technology is limited to the fusion of specific features between specific sensors. Its micro-processing characteristics are not suitable for macro-level system-level data fusion applications. Summary of the Invention

[0005] The purpose of the present invention is to provide a multi-source traffic data processing method based on confidence evaluation in order to solve the above problems, which takes the applicability of the detection areas of various detection systems as a benchmark, proposes the extraction confidence of different detection equipment and different detection areas on different traffic characteristic parameters as the basis for data fusion.

[0006] In order to achieve the above object, the present invention is implemented through the following technical solutions:

[0007] A multi-source traffic data processing method based on confidence assessment is used to fuse data of the same traffic characteristic parameter collected at the same spatiotemporal granularity and aggregate data of different traffic characteristic parameters to obtain various traffic flow indicators under a spatial research unit. First, the confidence of the traffic characteristic parameters under different detection systems is calibrated; for traffic flow indicators that can be directly obtained, the confidence of the corresponding traffic flow indicator under the spatial research unit is calculated by confidence fusion, and the traffic characteristic value provided by the detector with the largest confidence is selected as the traffic flow indicator of the spatial research unit; for traffic flow indicators that cannot be directly obtained, the traffic flow indicator of the corresponding time range of the spatial research unit is derived through other traffic characteristic values, and then the confidence of the corresponding traffic flow indicator of the spatial research unit is calculated through confidence aggregation; it is worth mentioning that the traffic flow indicator only considers spatial factors in the process of deriving other traffic characteristics, but the synchronization of indicator time should be considered when obtaining the traffic flow indicator of the corresponding time range of the spatial research unit.

[0008] Granularity refers to the scale of a meaningful unit within a domain. The spatial granularity in the spatiotemporal granularity in this invention refers to the smallest spatial unit for traffic processing, namely, the branch roads in all directions centered at the intersection. The smallest spatial granularity varies depending on the characteristics of the traffic flow. For example, the smallest spatial granularity for traffic volume is the lane, the smallest spatial granularity for travel speed is the turn or branch road, and the smallest spatial granularity for traffic density is the branch road. Temporal granularity refers to the detector acquisition period. Since different detectors may have different acquisition periods, traffic data from different acquisition intervals need to be converted into traffic data from the same period. This is done by taking the common multiples of the different acquisition intervals and then mapping the collected traffic data one-to-one.

[0009] There are many data fusion methods for directly obtainable traffic flow indicators. However, due to the large amount of data or the influence of the instability of the original sensor data, simply taking the average and median cannot well reflect the data characteristics. The present invention proposes a multi-source traffic data processing method based on confidence assessment, which provides data support and confidence reference for reflecting data characteristics. It includes the following steps: First, traffic feature analysis and detection system calibration based on the expert system as the basic support and relationship topology for confidence assessment and data fusion; Second, data spatiotemporal matching and missing data compensation to ensure that the basic granularity traffic feature parameters output by the system meet the fusion conditions; Third, data fusion and model calculation based on the multi-source confidence of traffic features, and output the data confidence after fusion calculation; Fourth, system aggregates traffic feature confidence calculation, and realizes system data confidence level classification and risk warning through the setting of multi-level confidence thresholds, specifically including:

[0010] 1. Traffic Characteristics Analysis and Equipment Calibration

[0011] 1. Based on the installation of fixed-point poles, the detection area inside the intersection can be divided into the oncoming direction detection area, the outgoing direction detection area, and the exit direction detection area, which are defined as w1, w2, and w3 respectively. The device type in the urban detection system is defined as t, type 1 is t1, type 2 is t2, ..., and so on. The traffic characteristic parameter that can be extracted in the detection system is defined as p, parameter 1 is p1, parameter 2 is p2, ..., and so on. It is worth mentioning that the detection system of the present invention refers to different areas and different devices to detect traffic characteristics, thereby generating characteristic values of traffic characteristic parameters. The characteristic values here refer to the actual values of traffic characteristic parameters, rather than the confidence levels of traffic characteristic parameters.

[0012] 2. Define the availability of traffic characteristic parameters in different detection areas as α, and the availability of characteristic parameter p in detection area w as α p-w , the value of α is 0-unavailable, 1-available, 2-available but with low confidence in spatial matching, the availability of traffic characteristic parameters of different equipment types is defined as β, and the availability of characteristic parameter p on equipment type t is β p-t ,β takes the value of 0-unavailable, 1-available, 2-model calculation. For model calculation, different data processing algorithms can be used to conduct supervised training on the availability of feature parameter p on device type t to obtain a long-term available model to classify the data availability;

[0013] 3. Referring to the availability index, the expert system scores the confidence of traffic features available in different detection areas and equipment types (0-100%, the confidence of unavailable features is 0), using Φ p-w represents the extraction confidence of the feature parameter p in the detection area w, Φ p-t represents the extraction confidence of feature parameter p on device type t, replacing availability α and β. The expert system here can also be an algorithm, a scoring system, etc., which realizes the classification of availability data and assigns different weights;

[0014] 4. After calibrating the detection device s in the system and determining the detection area w and device type t of the traffic characteristic parameter p, the data confidence Φ of the detection device is determined based on the volatility and rationality of the historical detection data of the detection device s. s (0-100%);

[0015] The above completes the analysis of the original data of the processing method. The specific collection includes the following according to the different types of detection equipment t:

[0016] Checkpoint t1 collection: The checkpoint device collects traffic data on the current road to obtain traffic characteristic parameters p at different spatial granularities;

[0017] Video t2 acquisition: The traffic flow data collected through the current road is analyzed through video to obtain traffic characteristic parameters p at different spatial granularities;

[0018] Radar microwave t3 acquisition: The radar sensor collects traffic data on the current road to obtain traffic characteristic parameters p at different spatial granularities;

[0019] The traffic characteristic parameters p include traffic rate p1, occupancy rate p2, queue length p3, headway p4, exit overflow p5 and vehicle passing data p6;

[0020] At different spatial granularities, according to the availability of traffic characteristic parameters p in different detection areas w, they are divided into the current road's oncoming direction detection area w1, the outgoing direction detection area w2, and the exit direction detection area w3;

[0021] 2. Data Spatiotemporal Matching and Missing Data Compensation

[0022] In order to obtain various traffic flow indicators within a spatial research unit, data fusion is premised on time matching and spatial matching. Time matching mainly includes three aspects: first, based on the approximate time range of data fusion, the detection data of multiple detectors within the time range are screened; second, because the acquisition cycles of different detectors may be different, traffic data with different acquisition intervals need to be converted into traffic data with the same cycle; third, the specific time range of data fusion is determined; spatial matching means that multi-source traffic detection data must reflect the traffic flow information of the same spatial research unit, such as an import lane, an import turn, etc.

[0023] In terms of data compensation, the advancement of road traffic detection technology has made urban road traffic data more comprehensive and accurate, but it is inevitable that detectors will be damaged and relevant data and traffic characteristic parameters will be missing. The traffic characteristic data of urban roads have temporal correlation, which is mainly divided into longitudinal time series correlation and transverse time series correlation. The transverse time series mainly reflects the trend of traffic flow changes within a day. Even if the days are different, the daily traffic flow change trends are roughly the same. The longitudinal time series is the correlation between traffic flows on different dates, which mainly reflects the long-term change trend of traffic flow data;

[0024] Based on the study of the temporal correlation of traffic data, missing data can be supplemented using the historical mean method, that is, using the detection data of a specific day in proportion, or taking the mean of historical data of multiple days corresponding to the time to replace the missing data. Let y(k,t) represent the missing traffic data at time t on day k, and the calculation is as follows:

[0025] y(k,t)=[y(k-1,t)+y(k-2,t)+...+y(kn,t)] / n

[0026] Where y(ki,t) represents the historical data at time t on day ki, and n is the number of days taken.

[0027] 3. Data Fusion and Confidence Analysis

[0028] 1. Directly accessible traffic flow indicators include traffic characteristic parameters p under different spatial research units, where the spatial scale of the spatial research unit is larger than the spatial granularity. The various traffic characteristics of a spatial research unit can be provided by multiple detectors. For the same characteristic parameter, the characteristic value provided by the detector with the highest confidence is selected as the characteristic value of the spatial research unit.

[0029] For each traffic characteristic of each spatial research unit, the confidence level is independent and is expressed as Φ. The calculation method is as follows:

[0030] Φ l-p-s =Φ p-w ×Φ p-t ×Φ s ×f l-p-s

[0031] Among them, l represents the spatial research unit, p represents the traffic characteristics, s represents the detection equipment, w and t represent the detection location and detection type of the detection equipment respectively, and f l-p-s is the influence coefficient of missing data compensation on source data. Select the maximum Φ l-p-s The parameter value provided by the corresponding detector is used as the value of the traffic characteristic p of the spatial research unit l, and Φ l-p is Φ l-p-s ;

[0032] 2. Traffic flow indicators that cannot be directly obtained include characteristic parameters at the granularity level, intersection level, and regional level. They can all be derived from existing traffic characteristic values based on the corresponding relationship model or calculation model. The characteristic parameters at the granularity level include traffic characteristics such as travel speed that describes the smoothness of the road, the number of stops at the intersection that describes the smoothness of the road, and traffic density that describes the bearing pressure of the road section. They are important and intuitive in system applications but cannot be directly detected and obtained. They can be obtained through other direct features and parameter calculations. Their confidence is indirectly achieved by aggregating the confidence of the detection source data, that is, the minimum value of the confidence of the traffic characteristic values involved in the derivation is selected as the confidence of the granularity-level characteristic parameters.

[0033] Φ l-p-s =min[Φ l-p-s,1 ,Φ l-p-s,2 ,...,Φ l-p-s,n ]

[0034] (1) Travel speed:

[0035] (1-1) Match the upstream and downstream vehicle passing data according to the identity information to obtain the time difference t between the upstream and downstream starting and ending points i , i=1,2,…,N, N is the number of matched vehicles;

[0036] (1-2) Take the effective time difference within the specified confidence distribution range and calculate the mean

[0037] Where T is the set of valid time differences, and M is the number of elements in T;

[0038] (1-3) The physical length L of the downstream is divided by Get the travel speed of upstream and downstream sections

[0039] (1-4) Calculate the confidence level of the output travel speed based on the confidence level of the traffic characteristic values involved in the calculation.

[0040] (2) Number of stops:

[0041] (2-1) Match the upstream and downstream vehicle passing data according to the identity information to obtain the time difference t between the upstream and downstream starting and ending points i , i=1,2,…,N, N is the number of matched vehicles;

[0042] (2-2) Take the effective time difference within the specified confidence distribution range and calculate the mean

[0043] Where T is the set of valid time differences, and M is the number of elements in T;

[0044] (2-3) Obtain the time t for free flow, i.e., passing through the upstream and downstream intersections without stopping f = / v f , number of stops

[0045] (2-4) Calculate the confidence level of the output parking times based on the confidence levels of the traffic characteristic values involved in the calculation.

[0046] (3) Traffic density

[0047] (3-1) Given that the basic relationship among traffic volume q, travel speed v, and traffic density k is q = v·k, we can obtain k = q / v, where q and v are the direct and indirect characteristics, respectively.

[0048] (3-2) Calculate the confidence level of the output traffic density based on the confidence levels of the traffic characteristic values involved in the calculation.

[0049] 4. Data Application and Confidence Assessment

[0050] 1. The traffic feature data obtained through the above-mentioned compensation, fusion, and deduction are all detection source data, which need to be further aggregated and processed into high-dimensional traffic features required for system display and analysis, such as intersection-level and regional-level feature parameters.

[0051] (1) Common intersection-level feature parameters are as follows:

[0052] (1-1) Traffic volume: The intersection traffic volume is obtained by summing the traffic volumes of all channel granularities at the intersection. The subscript C represents the intersection, F represents the channel, and n represents the total number of channels included in the intersection.

[0053]

[0054] (1-2) Number of stops: The number of stops at an intersection is equal to the weighted average of all channel traffic volumes and the number of stops at the channel.

[0055]

[0056] (1-3) Queue length: The queue length at an intersection is equal to the maximum queue length of all lanes.

[0057]

[0058] (2) Common regional characteristic parameters are as follows:

[0059] (2-1) Average vehicle speed: The average vehicle speed in a region is equal to the mean of the vehicle speeds of all valid detection sections within the statistical range. The subscript R represents the region, S represents the section, and n represents the total number of valid detection sections included in the region.

[0060]

[0061] (2-2) Traffic density: Regional traffic density is equal to the average traffic density of all valid detected road sections within the statistical range.

[0062]

[0063] (2-3) Congested mileage / ratio: A road section is considered congested when the ratio of the travel speed to the free-flow speed is greater than the congestion threshold σ. The regional congested mileage is equal to the sum of the lengths of all valid congested detection sections within the statistical range.

[0064]

[0065] 2. For the spatiotemporal aggregated traffic characteristic parameters required for system applications, their confidence is calculated by the confidence of the basic spatiotemporal granularity traffic characteristic parameters involved in the aggregation process using the aggregated statistical method. The confidence calculation method of the relevant statistical method is as follows:

[0066] (1) During the aggregation process, feature parameters are added and subtracted, and confidence indicators are averaged:

[0067]

[0068] (2) During the aggregation process, the feature parameters are multiplied and divided, and the confidence index takes the minimum confidence value of the participating operations:

[0069] Φ p =in[Φ p1 ,Φ p2 ,...,Φ pn ]

[0070] (3) During the aggregation process, the characteristic parameters take extreme values, and the confidence index takes the confidence of the extreme value element itself:

[0071] Φ p =Φ p,m

[0072] Among them, Φ p,m is the characteristic parameter p m The confidence level,

[0073] p m =ax / min[Φ p1 ,Φ p2 ,...,Φ pn ].

[0074] 3. According to actual application requirements, two confidence thresholds Φ1 and Φ2 are defined to divide the system traffic characteristics into three intervals, namely the low level interval [0, Φ1), the general level interval [Φ1, Φ2), and the complete level interval [Φ2, 100%].

[0075] 4. Generally speaking, we believe that analysis data with a confidence level between moderate and complete can serve as the basis for traffic situation analysis and control optimization within the intelligent transportation system. Low confidence levels indicate data risk. Based on this basic understanding, the confidence analysis results of this system can serve as the basis for system data risk warnings and dynamic optimization control.

[0076] Compared with existing technologies, the present invention has the following beneficial effects: it solves the problem of isolated operation and data fusion application of different types of traffic detection systems in urban intelligent transportation systems, improves data utilization and availability, and uses the confidence levels extracted from different traffic characteristic parameters of different detection equipment and different detection areas as the basis for data fusion; during the data fusion process, the confidence levels are synchronously updated as system data confidence levels, which serve as reference indicators for system data applications; provides calculation and confidence update methods for system aggregated application data, and sets multi-level confidence thresholds to conduct risk assessments on system application data. The present invention is suitable for the integrated application of urban overall traffic detection systems, providing data support and confidence references for regional-level situation assessments. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 Flow chart of the method of the present invention;

[0078] Figure 2 This is a schematic diagram of the intersection detection area classification;

[0079] Figure 3 This is the flow data graph of the same detector for four consecutive Fridays. DETAILED DESCRIPTION

[0080] The present invention will be further described in detail below with reference to the embodiments in the accompanying drawings, but this does not constitute any limitation to the present invention.

[0081] The present invention aims to solve the problem that the existing traffic detection system operates in isolation and the traffic data fusion and feature fusion cannot meet the requirements. Figure 1 、 2 3. This invention proposes a traffic data feature fusion scheme, which aims at the data fusion of the intersection traffic detection system, specifically including: 1. Traffic feature analysis and equipment calibration

[0082] 1. Considering the universality and applicability of technical research in implementation, we use common intersection detection methods such as electric police checkpoints, video detection and microwave detection as the main detection methods in the system. The detection application modes are divided into oncoming vehicle direction detection, outgoing vehicle direction detection and exit direction detection, such as Figure 2 shown.

[0083] 2. Using common data indicators such as traffic volume, occupancy rate, queue length, headway, and exit overflow as intersection traffic characteristic parameters, the availability of traffic characteristic parameters α and β for different detection areas and different detection equipment are shown in Table 1 and Table 2 respectively.

[0084] Table 1 Detection capabilities of different detection areas for traffic parameters

[0085]

[0086]

[0087] Note: 0 - Unavailable, 1 - Available, 2 - Available but with low confidence in spatial matching. For example, in traffic volume detection, the detection accuracy of oncoming traffic is lower than that of outgoing traffic due to vehicles changing lanes in the detection area far from the stop line.

[0088] Table 2 Detection capabilities of different equipment types for traffic parameters

[0089]

[0090] Note: 0-undetectable, 1-directly detectable, 2-model calculated.

[0091] 3. Referring to the availability index, the expert system scores the confidence of traffic features available in different detection areas and equipment types, where the confidence of unavailable features is 0. p-w and Φ p-t See Table 3 and Table 4 respectively.

[0092] Table 3. Confidence of traffic parameters extracted in different detection areas

[0093]

[0094]

[0095] Table 4. Confidence of traffic parameter extraction on different equipment types

[0096]

[0097] For different intersection channelization organization conditions, compensation coefficients can be set to correct the baseline confidence of specific intersections and turns to meet the needs of actual application scenarios.

[0098] 4. Calibrate the detection device s in the system, determine its detection area w and device type t; at the same time, determine the data confidence Φ of the detection device based on the volatility and rationality of the historical detection data of the detection device s s The calibration of the detection equipment included in the intersection detection system given in the example is shown in Table 5.

[0099] Table 5 Calibration of intersection detection system equipment

[0100]

[0101]

[0102] 2. Data Spatiotemporal Matching and Missing Data Compensation

[0103] 1. For configurable detectors and system aggregation methods, determine a unified detection reporting cycle and superimposable aggregation granularity to ensure the time granularity consistency of the source data. For example, set the calculation cycle of the electric alarm and checkpoint model and the statistical reporting cycle of the detector to 1 minute, and use the common multiple of the collection intervals of multiple groups of detectors for different calculation cycles;

[0104] 2. Based on the binding relationship between the channel and road network unit in the system of the computable road network specification detector, the spatial unit consistency of the source data is guaranteed. Figure 2 Taking the intersection channelization structure in as an example, the channel-road network unit binding relationship is shown in Table 6.

[0105] Table 6 Channel-Road Network Unit Binding Relationship

[0106]

[0107] 3. To compensate for traffic flow, we analyze traffic flow data for four consecutive weeks. The traffic volume trend of the same detector for four consecutive Fridays is as follows: Figure 3 As shown, from Figure 3 It can be seen that traffic flow exhibits strong longitudinal time series correlation. That is, the traffic flow data patterns within the same time period on Fridays across the four weeks are similar, and the time series changes exhibit essentially the same patterns. Further quantitative descriptions of the correlations between longitudinal time series data show that the correlation coefficients between any two longitudinal time series data exceed 0.96, indicating strong longitudinal time series correlation, as shown in Table 6.

[0108] Table 6 Correlation coefficients of Friday traffic data for four consecutive weeks

[0109]

[0110] Furthermore, based on the study of the time correlation of traffic data, the historical mean method can be used to fill in the missing data.

[0111] 3. Data Fusion and Confidence Analysis

[0112] 1. North branch lane 1 is used as the spatial research unit in the example analysis. Its associated traffic characteristic parameters and data sources are shown in Table 7.

[0113] Table 7 Matching relationship between spatial research units and data sources

[0114]

[0115] 2. The traffic volume of lane 1 of the north branch has four data sources providing original feature parameters. The confidence of the traffic volume parameters of the four data sources is calculated respectively. The calculation formula is Φ l-p-s =Φ p-w ×Φ p-t×Φ s ×f l-p-s , the following assumes that there is no missing data to compensate for the loss, that is, f l-p-s =1.

[0116] (1)Φ 1-p1-s4 =Φ p1-w1 ×Φ p1-t1 ×Φ s4 =0.6×1×0.85=0.51

[0117] (2)Φ 1-p1-s8 =Φ p1-w2 ×Φ p1-t1 ×Φ s8 =1×1×1=1

[0118] (3)Φ 1-p1-s9 =Φ p1-w2 ×Φ p1-t2 ×Φ s9 =1×1×1=1

[0119] (4)Φ 1-p1-s14 =Φ p1-w2 ×Φ p1-t3 ×Φ s14 =1×1×0.9=0.9

[0120] 3. After calculation, Φ 1-p1-s8 =Φ 1-p1-s9 >Φ 1-p1-s14 >Φ 1-p1-s4 , take the mean of the original traffic volume parameters provided by data sources s8-1 and s9-1 as the traffic volume characteristic value of north branch lane 1, and set Φ 1-p1 =Φ 1-p1-s8 .

[0121] 4. For traffic flow indicators that cannot be obtained by the detector, the derivation and calculation of indirect traffic characteristic values and the confidence value determination method can be introduced based on the corresponding relationship model or calculation model, taking the travel speed that describes the smoothness of the road as an example.

[0122] (1) The granularity unit of the travel speed is the road section. The north branch is taken as the research object, and the checkpoint equipment with identity information collection is used as the data source. The vehicle passing data of the north branch checkpoint equipment s8 and the upstream checkpoint equipment s15 are collected and sorted every 5 minutes; the distance L between the upstream and downstream equipment is calibrated.

[0123] (2) Match the passing vehicle data of the same identity in the upstream and downstream groups within the same time range, and calculate the passing time difference data t1, t2, ..., t n .

[0124] (3) At t1, t2, …, t nEliminate data that is twice the speed limit and is too small and within the maximum 10% range (with a confidence level of 90%), and calculate the average of the remaining data. Where T is the set of valid time differences, and M is the number of elements in T.

[0125] (4) Divide the calibrated upstream and downstream equipment distance L by Get the travel speed of upstream and downstream sections

[0126] (5) The eigenvalue data involved in the calculation is the vehicle passing data of devices s8 and s15, and the confidence calculation result is

[0127] Φ l-p-s8 / s15 =min[Φ l-p6-s8 ,Φ l-p6-s15 ]=min[Φ p6-w2 ×Φ p6-t1 ×Φ s8 ,Φ p6-w3 ×Φ p6-t1 ×Φ s15 ]=0.9.

[0128] 4. Data Application and Confidence Assessment

[0129] When the derivation formulas of high-dimensional traffic features, such as intersection-level and regional-level feature parameters, are clear, the intersection-level and regional-level feature parameters are calculated in sequence in a similar way to calculating travel speed. It is worth noting that the confidence calculation method is as follows:

[0130] (1) During the aggregation process, feature parameters are added and subtracted, and confidence indicators are averaged:

[0131]

[0132] (2) During the aggregation process, the feature parameters are multiplied and divided, and the confidence index takes the minimum confidence value of the participating operations:

[0133] Φ p =in[Φ p1 ,Φ p2 ,...,Φ pn ]

[0134] (3) During the aggregation process, the characteristic parameters take extreme values, and the confidence index takes the confidence of the extreme value element itself:

[0135] Φ p =Φ p,m

[0136] Among them, Φ p,m is the characteristic parameter p m The confidence level,

[0137] p m =ax / min[Φ p1 ,Φ p2 ,...,Φ pn ].

[0138] 3. According to actual application requirements, two confidence thresholds Φ1 and Φ2 are defined to divide the system traffic characteristics into three intervals, namely the low level interval [0, Φ1), the general level interval [Φ1, Φ2), and the complete level interval [Φ2, 100%].

[0139] (1) For intersection-level characteristic parameters, Φ1 = 50% and Φ2 = 75% can be taken; for regional-level characteristic parameters, the threshold level can be reduced accordingly due to the larger statistical scale, such as Φ1 = 40% and Φ2 = 60%.

[0140] (2) According to the settings in step (1), the confidence level intervals of the intersection-level aggregate feature parameters are set to [0, 50%), [50%, 75%), and [75%, 100%), and the confidence level intervals of the area-level aggregate feature parameters are set to [0, 40%), [40%, 60%), and [60%, 100%).

[0141] 4. Analysis data with confidence levels between general and complete levels can be used as the basis for traffic situation analysis and control optimization by intelligent transportation systems. If the confidence level is in the low level range, there is a data risk and it is not recommended to be used as input for other system applications.

[0142] In summary, the confidence-based multi-source data fusion technology method proposed in the present invention can be applied to urban intelligent traffic detection systems that combine different types of detection systems. It is not restricted by the specific detection equipment category and installation conditions. At the same time, it has low computing power requirements, strong real-time and anti-interference capabilities, and can effectively realize the fusion application and data quality assessment of regional-level traffic detection data.

[0143] The above embodiments are preferred implementation modes of the present invention and are only used to facilitate the explanation of the present invention. They are not intended to limit the present invention in any form. Any person with ordinary knowledge in the technical field can, without departing from the scope of the technical features of the present invention, make partial changes or modifications to the technical contents disclosed in the present invention and make equivalent embodiments without departing from the technical features of the present invention. Such modifications still fall within the scope of the technical features of the present invention.

Claims

1. A multi-source traffic data processing method based on confidence assessment is used to fuse data of the same traffic characteristic parameters collected at the same spatiotemporal granularity and aggregate data of different traffic characteristic parameters to obtain various traffic flow indicators within a spatial research unit. The method is characterized by: First, the confidence levels of traffic characteristic parameters under different detection systems are calibrated. The detection system is a multi-source detection system composed of at least two detection devices: an electric police checkpoint, video detection, and radar microwave. For traffic flow indicators that can be directly obtained, the confidence levels of the corresponding traffic flow indicators under the spatial research unit are calculated by confidence fusion, and the traffic characteristic values provided by the detector with the highest confidence level are selected as the traffic flow indicators for the spatial research unit. For traffic flow indicators that cannot be directly obtained, the traffic flow indicators for the spatial research unit are derived from the traffic characteristic values, and the confidence levels of the corresponding traffic flow indicators under the spatial research unit are calculated by confidence aggregation. The confidence aggregation is the calculation of the comprehensive confidence level after fusion of multi-source data using statistical methods. Obtain traffic characteristic parameters p at different spatial granularities based on traffic flow data collected by at least two detection devices passing through the current road, and collect the same traffic characteristic parameters and different traffic characteristic parameters at the same spatial and temporal granularities; Methods for applying traffic data of different spatiotemporal granularities to the same spatiotemporal granularity include: time matching and space matching, where time matching includes first screening out detection data of multiple detection equipment types t within the time range according to the time range of data fusion, then converting traffic data with different collection intervals into traffic data of the same period, and finally determining the time range of spatiotemporal granularity; space matching means that multi-source traffic detection data must reflect the traffic flow information of the same spatial research unit; calibrating the confidence of traffic characteristic parameters under different detection systems includes: scoring the confidence of traffic characteristic parameters p that can be obtained for different detection areas w and equipment types t; where the availability of traffic characteristic parameter p in detection area w is , The value of is 0-not available, 1-available, 2-available but with low confidence in spatial matching; the availability of feature parameter p on device type t is , The value is 0-unavailable, 1-available, 2-model calculation; according to 、 The confidence level of the traffic characteristic parameter p is scored based on the value of represents the extraction confidence of traffic feature parameter p in detection area w, Represents the extraction confidence of feature parameter p on device type t.

2. The method for processing multi-source traffic data based on confidence evaluation according to claim 1, characterized in that: Methods for collecting the same traffic characteristic parameters and different traffic characteristic parameters at the same spatiotemporal granularity include the following depending on the type of detection equipment t: Checkpoint t1 collection: The checkpoint device collects traffic data on the current road to obtain traffic characteristic parameters p at different spatial granularities; Video t2 acquisition: The traffic flow data collected through the current road is analyzed through video to obtain traffic characteristic parameters p at different spatial granularities; Radar microwave t3 acquisition: The radar sensor collects traffic data on the current road to obtain traffic characteristic parameters p at different spatial granularities; The traffic characteristic parameters p include traffic rate p1, occupancy rate p2, queue length p3, headway p4, exit overflow p5 and vehicle passing data p6; At different spatiotemporal granularities, according to the different availability of traffic characteristic parameters p in different detection areas w, they are divided into the current road's incoming vehicle direction detection area w1, the outgoing vehicle direction detection area w2, and the exit direction detection area w3.

3. The method for processing multi-source traffic data based on confidence evaluation according to claim 1, characterized in that: When traffic data of different spatiotemporal granularities are applied to the same spatiotemporal granularity, the missing time unit data of multi-source traffic data should be compensated. The compensation methods include replacing the longitudinal time series correlation data and the horizontal time series correlation data. The influence coefficient of the source data after compensating the data is defined as , where l represents the spatial research unit, p represents the traffic characteristics, and s represents the detection equipment.

4. The method for processing multi-source traffic data based on confidence evaluation according to claim 1, characterized in that: The confidence fusion method is based on the assumption that the confidence of each traffic characteristic parameter p at each spatiotemporal granularity is independent, and the calculation method for obtaining it is as follows: in To determine the data confidence of the testing device s based on the volatility and rationality of the historical testing data of the testing device s, and then select the maximum The parameter value provided by the corresponding detector is used as the value of the traffic characteristic p of the spatial research unit l, and is set for .

5. The method for processing multi-source traffic data based on confidence evaluation according to claim 1, characterized in that: The traffic flow indicators that can be directly obtained include the traffic characteristic parameters p under different spatial research units, where the spatial scale of the spatial research unit is larger than the spatial granularity; the traffic flow indicators that cannot be directly obtained include characteristic parameters at the granularity level, intersection level, and regional level; the characteristic parameters at the granularity level include travel speed, number of stops, and traffic density; the characteristic parameters at the intersection level include traffic volume, number of stops at the intersection, and queue length; the characteristic parameters at the regional level include average speed, regional traffic density, and congestion mileage.

6. The method for processing multi-source traffic data based on confidence evaluation according to claim 5, characterized in that: The confidence aggregation method is divided into detection source data confidence aggregation and high-dimensional application data confidence aggregation based on the dimensions of various traffic flow indicators. Among them, detection source data confidence aggregation is applicable to granular feature parameters, and high-dimensional application data confidence aggregation is applicable to intersection-level and regional-level feature parameters.

7. The method for processing multi-source traffic data based on confidence evaluation according to claim 6, characterized in that: The aggregation method of the detection source data confidence is to select the minimum value of the confidence of the traffic characteristic values involved in the derivation as the confidence of the characteristic parameter at the granularity level.

8. The method for processing multi-source traffic data based on confidence evaluation according to claim 6, characterized in that: The aggregation method of the high-dimensional application data confidence is calculated according to the aggregation statistical method, and the calculation method is as follows: When the feature parameters are added or subtracted during the aggregation process, the confidence indicators are averaged: ； When the feature parameters are multiplied or divided during the aggregation process, the confidence index takes the minimum confidence value of the participating operations: When the characteristic parameter takes an extreme value during the aggregation process, the confidence index takes the confidence of the extreme value element itself: in is the characteristic parameter The confidence level, .

9. The method for processing multi-source traffic data based on confidence evaluation according to any one of claims 1 to 8, used for risk warning of traffic flow indicator parameters, characterized in that: Divide the confidence interval according to actual needs and divide the confidence of traffic flow indicators into three intervals: low level interval , general level range and complete level intervals ,in is the confidence threshold, and then according to the confidence of the traffic flow indicator parameters, an early warning is issued to determine whether there is a data risk.

Images