A traffic signal adaptive control method and system

Abstract

The application discloses a traffic signal adaptive control method and system, and relates to the technical field of intelligent transportation, which comprises the following steps: acquiring multi-source traffic detection data and performing time consistency preprocessing according to lane groups to form a unified observation set and delay information; determining a risk tolerance parameter and constructing a probability constraint based on traffic state estimation, uncertainty representation and parameter update statistics; converting the probability constraint into a convex constraint which can be solved in real time to perform rolling timing solution; outputting a signal control instruction and a feasible margin; issuing the signal control instruction according to the feasible margin and the comprehensive reliability of the intersection, and adaptively switching between the rolling timing control and the degraded control. The application constructs a probability constraint of traffic signal timing and convexes it into a convex constraint domain which can be solved in real time, so that the rolling optimization output has a calculable confidence boundary and a feasible margin, and provides a verifiable constraint margin representation for the issuance of the traffic signal control instruction and the mode switching judgment.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and in particular to a traffic signal adaptive control method and system. Background Technology

[0002] Traffic signal control technology has gradually evolved from fixed timing to detection-driven adaptive and predictive control, widely incorporating multi-source detection such as coils, video, radar, floating cars, and V2X. Using lane groups as modeling units, it performs traffic state estimation, arrival process inference, and rolling timing optimization, forming a closed-loop control link of perception-estimation-optimization-execution, thereby supporting online timing and coordinated release under complex requirements.

[0003] In scenarios involving parallel multi-source detection and communication link-introduced latency, the main shortcomings of existing methods are: the lack of a unified quantification and online update mechanism for the credibility and uncertainty of each detection data, the difficulty in transforming observation latency, abnormal jitter and historical consistency into reliable metrics that can be used for filtering and constraint construction, the difficulty in consistently determining the confidence boundaries of state estimation and parameter statistics, the lack of verifiable risk tolerance basis for the constraint boundaries of risk constraint timing, and the impact on the feasibility and stability of rolling solution. Summary of the Invention

[0004] In view of the above problems, the present invention is proposed.

[0005] Therefore, this invention provides a traffic signal adaptive control method to solve the problem of stable and feasible rolling optimization of intersection signal timing risk constraints under the fluctuation of multi-source traffic detection delay and reliability.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a traffic signal adaptive control method, which includes acquiring multi-source traffic detection data and performing time consistency preprocessing according to lane groups to form a unified observation set and delay information;

[0008] The credibility weights of each traffic detection data are calculated based on a unified observation set, delay information, and historical reference observations, and then aggregated to obtain the overall credibility of lane groups and the overall credibility of intersections.

[0009] Based on the comprehensive confidence of lane groups, the fusion weights and observation noise are adaptively set and fusion filtering is performed to output traffic state estimates and corresponding uncertainty representations.

[0010] By estimating traffic conditions and using a unified observation set, external arrival flow and turning ratio are inferred and corresponding statistics are established. The corresponding statistics are then updated online based on the overall credibility of the intersection, resulting in parameter update statistics.

[0011] Based on traffic state estimation, uncertainty characterization and parameter update statistics, risk tolerance parameters are determined and probabilistic constraints are constructed. The probabilistic constraints are transformed into convex constraints that can be solved in real time for rolling timing solution, and the signal control command and feasibility margin are output.

[0012] Based on the feasibility margin and the overall credibility of the intersection, the signal control command is issued and executed, and adaptive switching is performed between rolling timing control and degraded control.

[0013] In a preferred embodiment of the traffic signal adaptive control method of the present invention, the specific steps for forming a unified observation set and delay information are as follows:

[0014] Acquire multi-source traffic detection data and extract time stamps, lane attribution information, and data source identifiers to form labeled multi-source traffic detection data;

[0015] Based on lane affiliation information, lanes with the same release direction and corresponding signal phase are identified as lane groups. Marked multi-source traffic detection data are merged by lane groups and subjected to unified time axis alignment and resampling to form aligned multi-source observation sequences of lane groups.

[0016] In the aligned lane group multi-source observation sequence, arrival delay and transmission delay are identified and missing data are repaired and anomaly labeling is performed, and a unified observation set and delay information are output.

[0017] As a preferred embodiment of the traffic signal adaptive control method of the present invention, the specific steps for calculating the confidence weight of each traffic detection data based on a unified observation set, delay information, and historical reference observations, and aggregating them to obtain the comprehensive confidence of lane groups and the comprehensive confidence of intersections, are as follows.

[0018] Consistency verification of the unified observation set at the same time slice is performed to form consistency information, which is then correlated with the delay information to form credibility assessment information;

[0019] Historical time window observation records corresponding to the current time slice are selected from the unified observation set, and after time backtracking and alignment based on delay information, they are summarized by lane group to form historical reference observation values.

[0020] Deviation assessment is performed based on credibility assessment information combined with historical reference observations, and credibility weights for each traffic detection data are calculated.

[0021] The credibility weights of each traffic detection data are aggregated by lane group to obtain the overall credibility of the lane group, and then the overall credibility of the lane group is aggregated at the intersection dimension to obtain the overall credibility of the intersection.

[0022] As a preferred embodiment of the traffic signal adaptive control method of the present invention, the traffic state estimation and corresponding uncertainty characterization are implemented through the following specific steps.

[0023] Based on the overall credibility of lane groups, the fusion weights in the lane group observation inputs are set, and the correspondence between the lane group observation inputs and the fusion weights is established to obtain the fusion weights;

[0024] Based on the overall reliability of the lane group, the set observation noise in the lane group observation input is set, and the correspondence between the lane group observation input and the observation noise is established to obtain the observation noise;

[0025] The fusion filtering of lane group observation inputs is performed using fusion weights and observation noise, and the output is a traffic state estimate and the corresponding uncertainty representation.

[0026] As a preferred embodiment of the traffic signal adaptive control method of the present invention, the steps of inferring external arrival flow and turning ratio through traffic state estimation and a unified observation set, establishing corresponding statistics, and updating the corresponding statistics online based on the intersection's comprehensive reliability to obtain parameter update statistics are as follows.

[0027] Using traffic state estimation and a unified observation set, the observation records were grouped by lane group and aligned along a unified time axis, and then summarized into entrance direction observation records and exit direction observation records respectively.

[0028] Based on the observation records of the entrance direction and the observation records of the exit direction, the net arrival change at the intersection boundary is identified and the non-external arrival components generated inside the intersection are removed to form an external arrival flow inference record;

[0029] Establish a correspondence between external arrival flow inference records and traffic state estimates according to the direction of release, summarize them to form a turning allocation record, and calculate the allocation ratio according to the direction of release to form a turning ratio record;

[0030] Based on the external arrival traffic inference records and turning ratio records, external arrival traffic statistics and turning ratio statistics are established respectively. The external arrival traffic statistics and turning ratio statistics are updated online by the intersection comprehensive confidence level, and the updated parameters are output.

[0031] In a preferred embodiment of the traffic signal adaptive control method of the present invention, the specific steps for determining the risk tolerance parameter and constructing probabilistic constraints are as follows:

[0032] Traffic state estimates, uncertainty representations, and parameter update statistics are aligned with the same time slice and then standardized to form risk assessment information.

[0033] Based on risk assessment information, the degree of dispersion of uncertainty characteristics and the update changes of parameter update statistics are jointly summarized to form risk-sensitive information;

[0034] Risk tolerance parameters are determined based on risk-sensitive information, and probabilistic constraints are constructed by combining traffic state estimation and uncertainty characterization.

[0035] As a preferred embodiment of the traffic signal adaptive control method of the present invention, the specific steps of transforming the probabilistic constraints into real-time solvable convex constraints for rolling timing solution, and outputting signal control commands and feasibility margins are as follows.

[0036] The probability constraints and uncertainty representations are aligned and organized in the same time slice to obtain the aligned probability constraints.

[0037] The aligned probabilistic constraints are transformed into convex constraints that can be solved in real time, and the constraint terms are organized to form a set of convex constraints.

[0038] Under the constraints of the convex constraint set, the rolling timing solution is executed in conjunction with traffic state estimation to output signal control commands, and the feasible margin is summarized based on the constraint redundancy of the signal control commands under the convex constraint set.

[0039] As a preferred embodiment of the traffic signal adaptive control method of the present invention, the specific steps of issuing and executing signal control commands based on feasibility margin and intersection comprehensive reliability, and adaptively switching between rolling timing control and degraded control are as follows.

[0040] The signal control command, feasibility margin, and intersection comprehensive credibility are matched and aligned in the same time slice. The matched signal control command and the corresponding feasibility margin and intersection comprehensive credibility are summarized to form a judgment item, and the execution judgment information is obtained.

[0041] When the feasibility margin has constraint surplus under the convex constraint set, the intersection comprehensive credibility satisfies the continuous time slice stability, and there are no new anomaly labels in the unified observation set, the execution signal control command is issued and rolling timing control is maintained.

[0042] When the feasible margin does not have constraint redundancy under the convex constraint set, or the intersection comprehensive confidence does not meet the continuous time slice stability, or new anomaly labels appear in the unified observation set, the issuance of execution signal control commands is stopped and the system switches to degraded control.

[0043] As a preferred embodiment of the traffic signal adaptive control method of the present invention, the degraded control refers to the conservative signal control command generated and output based on a unified observation set, traffic state estimation and historical reference observations when the execution judgment information does not meet the execution conditions.

[0044] Secondly, the present invention provides a traffic signal adaptive system, including the multi-source alignment module, which is used to acquire multi-source traffic detection data and perform time consistency preprocessing according to lane groups to form a unified observation set and delay information;

[0045] The credibility assessment module is used to calculate the credibility weight of each traffic detection data based on a unified observation set, delay information and historical reference observation values, and aggregate them to obtain the comprehensive credibility of lane groups and the comprehensive credibility of intersections.

[0046] The fusion filtering module is used to adaptively set the fusion weight and observation noise based on the comprehensive confidence of the lane group and perform fusion filtering to output traffic state estimation and corresponding uncertainty characterization.

[0047] The parameter inference module is used to infer external arrival flow and turning ratio through traffic state estimation and unified observation set and establish corresponding statistics. Based on the comprehensive confidence of the intersection, the corresponding statistics are updated online to obtain parameter update statistics.

[0048] The risk optimization module is used to determine the risk tolerance parameter and construct the probability constraint based on traffic state estimation, uncertainty characterization and parameter update statistics. It then transforms the probability constraint into a convex constraint that can be solved in real time for rolling timing solution and outputs signal control commands and feasibility margin.

[0049] The execution switching module is used to issue execution signal control commands based on feasibility margin and intersection comprehensive credibility, and adaptively switch between rolling timing control and degraded control.

[0050] The beneficial effects of this invention are as follows: Based on traffic state estimation and its uncertainty characterization, risk tolerance parameters are generated online using arrival flow and turning ratio update statistics. Probabilistic constraints for traffic signal timing are constructed and convexized into a convex constraint domain that can be solved in real time. This enables the rolling optimization output to have calculable confidence boundaries and feasibility margins, and provides verifiable constraint margin characterization for traffic signal control command issuance and mode switching determination. The feasibility margin quantifies the adjustment margin for phase switching and green light ratio allocation, limits the safe execution range of commands, enhances decision continuity and solution stability, suppresses command jumps, and reduces the risk of execution interruption. Attached Figure Description

[0051] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a flowchart of a traffic signal adaptive control method.

[0053] Figure 2 This is a schematic diagram of a traffic signal adaptive control system.

[0054] Figure 3 This is a flowchart for calculating the credibility weights of each traffic detection data point.

[0055] Figure 4 A flowchart for inferring external arrival flow. Detailed Implementation

[0056] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0057] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0058] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0059] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a traffic signal adaptive control method, comprising the following steps:

[0060] S1: Acquire multi-source traffic detection data and perform time consistency preprocessing according to lane groups to form a unified observation set and delay information.

[0061] S1.1: Acquire multi-source traffic detection data and extract time stamps, lane attribution information and data source identifiers to form labeled multi-source traffic detection data.

[0062] Receive raw traffic detection data records from video detection, geomagnetic detection, and microwave radar, write the receiving time to each record, and retain the original fields.

[0063] Read the observation time, detection area number or lane number, original data source identifier, and target observation field value from the original record.

[0064] If the observation time is missing or cannot be converted into a UTC epoch millisecond timestamp, the write error is marked as a clock error and the record is removed, and it does not enter the marked record generation process.

[0065] Perform a missing check on each of the detection area number or lane number, the original identifier of the data source, and the target observation field value; when any field is missing, set the missing field to null and write the missing reason enumeration (e.g., not reported, parsing failure, or mapping failure).

[0066] The observation time is used as the time stamp input. When recording time zone offset, it is first converted to UTC. When recording clock correction offset, it is corrected based on UTC. If the UTC epoch millisecond timestamp cannot be obtained at any stage, it is marked as a clock error and the record is removed.

[0067] Prioritize using the lane number that comes with the record; if there is no lane number but there is a detection area number, then map the lane number according to the detection area and lane correspondence configuration table; if the mapping result is not unique, mark the mapping as abnormal and remove the record.

[0068] Based on the lane number, query the intersection phase configuration information table and write the direction of passage and the signal phase; if the configuration cannot be matched, mark the configuration as missing and remove the record.

[0069] Map the original data source identifier to a data source identifier enumeration, which includes video detection, geomagnetic detection, and microwave radar. Sources not in the enumeration are mapped to other sources while retaining their original identifier.

[0070] Information such as time stamp, reception time, lane number, release direction, signal phase, data source identifier and its original identifier, target observation field value and missing reason are summarized and written into the same record to form labeled multi-source traffic detection data.

[0071] It should be noted that the configuration table of the correspondence between detection area and lane is generated by offline matching of the geometric range of the detection area of ​​each detection device and the geometric range of the lane at the intersection in the same coordinate system and combined with the validity period. The intersection phase configuration information table is generated by offline sorting of the lane number-release direction-signal phase correspondence and the shortest green light time in the signal control scheme and combined with the validity period. It is read and used according to the validity period during operation.

[0072] S1.2: Based on the lane affiliation information, lanes with the same release direction and corresponding signal phase are identified as lane groups. Marked multi-source traffic detection data are merged by lane groups and uniform time axis alignment and resampling are performed to form aligned multi-source observation sequences of lane groups.

[0073] Lanes with the same release direction and corresponding to the same signal phase are identified as lane groups. Lane group identifiers are generated for lane groups, and the correspondence between lane numbers and lane group identifiers is fixed.

[0074] The labeled multi-source traffic detection data are merged according to lane group identifier and data source identifier to form a merged input sequence.

[0075] Read the timestamps of all valid tagged records, and take the earliest and latest timestamps as the coverage endpoints.

[0076] The shortest green light time at the intersection is used as the upper limit of the time slice length, and the largest of the minimum reporting cycles from each data source is used as the lower priority limit to determine the time slice length.

[0077] If the shortest green light time at the intersection is less than the priority lower limit, the time slice length is taken as the shortest green light time at the intersection; the final time slice length is an integer millisecond.

[0078] Align the start point of the coverage area forward to the time slice boundary and the end point backward to the time slice boundary, and divide the time slices according to left-closed and right-open, and establish a time slice index.

[0079] The time stamp of each record in the merged input sequence is mapped to the time slice index, and time slice slots are established in the dimensions of lane group identifier, data source identifier and time slice index; when there is no record in the corresponding time slice slot, it is reserved as an empty slot.

[0080] For the target observation field values ​​in each time slot, resampling is performed field by field according to the field type table, so that each time slot outputs only one resampling result for each field.

[0081] Specifically, when the field type is count, the output is the summary value within the time slice; when the field type is instantaneous, the output is the arithmetic mean within the time slice; when the field type is status, the output is the value of the latest record of the timestamp within the time slice; the resampling result of the field corresponding to the empty slot is an empty value, and the value of the resampled target observation field is obtained.

[0082] The values ​​of the resampled target observation field corresponding to each time slice index are sorted and output in ascending order of time slice index to form an aligned lane group multi-source observation sequence. The lane group multi-source observation sequence includes at least the lane group identifier, data source identifier, time slice index and resampled target observation field value.

[0083] S1.3: Identify arrival delay and transmission delay in the aligned lane group multi-source observation sequence, perform missing data repair and anomaly labeling, and output a unified observation set and delay information.

[0084] For each tagged multi-source traffic detection data, the difference between the received time and the time stamp is calculated as the transmission delay. If the difference is negative, an anomaly is written as a clock anomaly, and the transmission delay is set to zero.

[0085] The transmission delays are aggregated by lane group identifier, data source identifier, and time slice index, and the median is taken to form a transmission delay sequence.

[0086] The percentage of missing values ​​for each data source is calculated for each lane group, and the source with the lowest percentage of missing values ​​is used as the reference data source; when the percentages are tied, the source is determined in the order of microwave radar, geomagnetic detection, video detection and others.

[0087] Within the current lane group, the field with the largest number of valid samples for offset calculation is selected as the arrival delay estimation field based on the priority of cumulative traffic flow, vehicle speed, and occupancy rate.

[0088] If none of the three are available, then use the alternative field specified in the field type table and write the estimated field missing flag.

[0089] Read the upper limit of device processing delay from the device delay parameter table and combine it with the maximum value of the transmission delay sequence to form the upper limit of total delay; and convert it into the upper limit of candidate offset slices according to the time slice length, with the candidate offset taking positive or negative integers within the upper limit range.

[0090] For each non-reference data source, calculate the average absolute deviation from the reference data source within the candidate offset range, and take the offset with the smallest average absolute deviation as the optimal offset;

[0091] If there are insufficient available alignment samples, the positive offset corresponding to the upper limit of the candidate offset slice count is taken as the conservative offset and written with the insufficient sample flag; the optimal offset is converted into arrival delay according to the time slice length and written into the record.

[0092] The offset with the smallest average absolute deviation is taken as the optimal offset, expressed as:

[0093] ;

[0094] In the formula, This is the optimal time offset. This is the upper bound for the number of candidate offset slices. The range of values ​​for candidate time offsets. The number of effective time slices used in the calculation. For time slice index, For the set of valid time slice indexes, Lane group identification, Lane group markings In time slice index The reference data source is the resampled target observation field value. Lane group markings In time slice index The values ​​of the resampled target observation fields from the data source to be corrected.

[0095] Arrival delay and transmission delay are aligned and summarized according to lane group identifier, data source identifier and time slice index to form delay information. The output includes at least lane group identifier, data source identifier, time slice index, arrival delay, transmission delay and necessary identifiers.

[0096] The alignment sequence performs missing value repair field by field, identifying consecutive null value segments; when the length of the null value segment does not exceed two time slices, count fields are filled with zeros, and instantaneous and status fields are filled with the most recent non-null value before the start of the null value segment, and are uniformly marked as repaired;

[0097] When the length of a null segment exceeds two time slices, retain the null value and mark it as a missing anomaly.

[0098] Based on each record in the aligned lane group multi-source observation sequence, the upper limit value and unit of the field corresponding to the data source identifier and field name are read from the equipment calibration parameter table field by field, and the range verification is performed.

[0099] When a field value is negative or exceeds the field's upper limit, the record will be marked as an observed anomaly; records already marked as observed anomalies will retain the anomaly labeling results.

[0100] Set a minimum threshold consistent with its unit for each target observation field, and use the field resolution in the equipment calibration parameter table. If the resolution is missing, use the minimum threshold calculated from the upper limit value of the field.

[0101] Under the same lane group identifier and the same time slice index, valid field values ​​from various data sources are aggregated, and records with null values, missing anomalies, and anomalies that have been marked are removed to form cross-source samples;

[0102] When there are fewer than two valid sources, cross-source anomaly labeling is not performed. When there are at least two valid sources, the cross-source median and cross-source interquartile range are calculated, and the larger of the interquartile range and the minimum threshold is used as the anomaly threshold.

[0103] The absolute deviation of each record from the cross-source median is compared with the anomaly threshold. Records exceeding the anomaly threshold are marked as observed anomalies, while records not exceeding the threshold retain their original anomaly labels. Subsequently, the alignment results of missing data repair and anomaly labeling are compiled into a unified observation set.

[0104] The fields include lane group identifier, data source identifier, time slice index, resampled target observation field value, anomaly annotation and missing data repair identifier;

[0105] The unified observation set is matched and aligned with the delay information based on lane group identifier, data source identifier, and time slice index before being output.

[0106] S2: Calculate the credibility weight of each traffic detection data based on a unified observation set, delay information and historical reference observations, and aggregate them to obtain the overall credibility of lane groups and the overall credibility of intersections.

[0107] S2.1: Perform time-slice consistency verification on the unified observation set to form consistency information, and associate it with the delay information to form credibility assessment information.

[0108] Based on a unified set of observation records, the resampled target observation field values ​​of the same field are aggregated by the lane group identifier and time slice index. Records with null values, abnormal annotations that are not normal, and missing repair annotations that are missing are removed. The cross-source median and cross-source interquartile range are calculated.

[0109] When there are fewer than two valid data sources, the cross-source median is set to null and the cross-source interquartile range is set to zero.

[0110] For each record, calculate the consistency deviation within the same time slice, taking the deviation of the resampled target observation field value in the record relative to the cross-source median.

[0111] The larger of the interquartile range across sources and the minimum threshold is taken as the normalization benchmark. The consistency deviation is then normalized according to the normalization benchmark to obtain the consistency normalization deviation.

[0112] When the cross-source median is empty or there are fewer than two valid data sources, the consistency deviation and consistency normalization deviation of the same time slice are both set to zero.

[0113] Read the delay information records, associate them with the unified observation set according to the lane group identifier, data source identifier and time slice index to form credibility assessment information.

[0114] The credibility assessment information includes a unified observation set field, time-slice consistency deviation, cross-source interquartile range, arrival delay, and transmission delay.

[0115] S2.2: Select historical time window observation records corresponding to the current time slice from the unified observation set, and after time backtracking and alignment based on delay information, summarize them by lane group to form historical reference observation values.

[0116] The length of the historical time window is determined by the number of time slices. The length of the historical time window is the larger of the number of time slices corresponding to one signal period and the upper bound of the number of candidate offset slices plus one.

[0117] The number of time slices corresponding to a signal cycle is calculated by converting the signal cycle duration and the time slice length. The signal cycle duration is preferably obtained from the intersection phase configuration information table. If it is not available, it is calculated by summing the shortest passage time for each phase and the phase switching clearing time.

[0118] The arrival delay, transmission delay, and time slice length are read one by one from the credibility assessment information. The absolute values ​​of the arrival delay and transmission delay are summed to obtain the total delay in seconds.

[0119] The total delay is rounded up to the nearest integer based on the time slice length to convert it into the number of backtrack slices. The number of backtrack slices is the smallest integer that ensures the coverage time after backtracking is not less than the total delay.

[0120] The backtracking time slice index is obtained by using the current time slice index as the starting point and backtracking along a unified time axis to count the number of slices in the historical direction.

[0121] Using lane group identifiers and backtracking time slice indexes as dimensions, resampled target observation field values ​​are aggregated within the historical time window coverage area. After removing null values, records with abnormal annotations, and records with missing repair annotations as missing anomalies, the cross-source median is calculated as the historical reference observation value, and the cross-source interquartile range is calculated as the historical reference dispersion.

[0122] When there are no valid records within the historical time window, the historical reference observation value is set to empty, and the historical reference dispersion is set to zero.

[0123] S2.3: Based on the credibility assessment information and historical reference observations, deviation assessment is performed, and the credibility weight of each traffic detection data is calculated.

[0124] The credibility assessment information is associated with historical reference observations by lane group identifier and time slice index to obtain the historical reference observation and historical reference dispersion for each record.

[0125] When the historical reference observation is empty, the historical reference deviation is set to zero, and the historical reference dispersion is set to zero; when the historical reference observation is not empty, the historical reference deviation is the magnitude of the deviation of the recorded value from the historical reference observation.

[0126] The larger of the interquartile range across sources and the minimum threshold is taken as the consistency normalization benchmark. The consistency deviation of the same time slice is normalized to obtain the consistency normalization deviation.

[0127] The historical reference dispersion and the minimum threshold are taken as the larger of the two and used as the historical normalization benchmark. The historical reference deviation is then normalized to obtain the historical normalized deviation.

[0128] The delay deviation is taken as the maximum of the absolute values ​​of the arrival delay and the transmission delay, and the delay deviation is normalized using the time slice length as the normalization reference to obtain the normalized delay deviation.

[0129] The largest of the consistency normalization deviation, historical normalization deviation, and delayed normalization deviation is taken as the deviation assessment metric.

[0130] The deviation from the assessment is mapped to an unnormalized confidence score, with the score ranging from zero to one, and the larger the deviation from the assessment, the smaller the score.

[0131] When a record is empty, the anomaly label is not normal, or the missing repair label is missing anomaly, the unnormalized confidence score is set to zero; for the remaining valid records, the unnormalized confidence score is calculated based on the deviation from the evaluation value, and a preset minimum score lower limit is set (e.g., 0.001). When the calculated score is lower than the minimum score lower limit, the score is adjusted to the minimum score lower limit.

[0132] Within the same lane group and the same time slice, the unnormalized credibility scores from various data sources are aggregated and normalized. The normalized credibility weights satisfy the condition that the sum of the credibility weights of all data sources is one, and the credibility weights change in the same direction as the unnormalized credibility scores.

[0133] When there are no available records or the unnormalized confidence score sum is zero, all confidence weights are set to zero and output.

[0134] S2.4: Aggregate the credibility weights of each traffic detection data by lane group to obtain the overall credibility of the lane group, and then aggregate the overall credibility of the lane group at the intersection dimension to obtain the overall credibility of the intersection.

[0135] Unnormalized weights are aggregated in the lane group identifier and time slice index dimensions. After removing records with unnormalized weights of zero, the median is taken as the overall credibility of the lane group. If there are no valid records after removal, the overall credibility of the lane group is set to zero.

[0136] Under the same time slice index, the overall confidence scores of all lane groups are aggregated and the median is taken as the overall confidence score of the intersection. When the lane group identifier is empty, the overall confidence score of the intersection is set to zero. The overall confidence scores of the lane groups and the overall confidence scores of the intersection are output.

[0137] S3: Adaptively set fusion weights and observation noise based on the overall credibility of lane groups and perform fusion filtering to output traffic state estimates and corresponding uncertainty representations.

[0138] S3.1: Merge the unified observation set by lane group, align and organize it according to a unified time axis, and associate delay information to form lane group observation input.

[0139] Based on a unified observation set, lane group observation records are formed by merging lane group identifiers, data source identifiers, and time slice indexes.

[0140] If duplicate records exist for the same lane group identifier, the same data source identifier, and the same time slice index, then the duplicate write identifier is written, and the records are merged into a single record based on the median of the same field.

[0141] The delay information is associated with the lane group observation records according to the lane group identifier, data source identifier, and time slice index to obtain the arrival delay and transmission delay.

[0142] When delay information is missing, the most recent valid delay for the same lane group identifier and the same data source identifier is retrieved within the delay backtracking window length; the delay backtracking window length is the upper bound of the candidate offset slice number plus one; if there is still no valid delay within the window, the upper limit of the device processing delay in the device delay parameter table is read and combined with the maximum value of the available transmission delay to form a conservative delay upper bound.

[0143] The upper bound of the conservative delay is used as the arrival delay and written as a delay missing marker to indicate its existence; otherwise, the delay missing marker is indicated as its non-existence.

[0144] For each associated record, summarize the absolute values ​​of arrival delay and transmission delay to obtain the total delay in seconds. Then, round up to the nearest integer based on the time slice length to convert it into the number of backslices, ensuring that the total duration of the time slices corresponding to the number of backslices is not less than the total delay in seconds. Starting from the time slice index of the associated record, backslice the number of backslices along the unified time axis in the historical direction to obtain the backslice index. If it is earlier than the starting index of the unified time axis, take the starting index.

[0145] Write the backtracking time slice index into the observation record and organize it to obtain the lane group observation input.

[0146] Lane group observation input includes lane group identifier, data source identifier, time slice index, backtracking time slice index, resampled target observation field value, anomaly label, missing data repair label, arrival delay and transmission delay, and necessary labels.

[0147] S3.2: Based on the overall confidence level of the lane group, set the fusion weight in the lane group observation input, and establish the correspondence between the lane group observation input and the fusion weight to obtain the fusion weight.

[0148] Lane group observation inputs are aggregated along the dimensions of lane group identifier and backtracking time slice index. Records that are marked as normal and whose missing repair identifiers are not missing anomalies are filtered to form a usable set.

[0149] When the available set is empty, all fusion weights are set to zero.

[0150] When the available set is not empty, take the overall confidence of the lane group and truncate it to zero or one to obtain the mixing coefficient.

[0151] For each record in the available set, read the arrival delay, transmission delay, and time slice length. Take the largest of the two absolute values ​​as the maximum delay, and round it up according to the time slice length to convert it into a delay level, so that the total time slice duration corresponding to the delay level is not less than the maximum delay.

[0152] The latency rating table is fixed in the configuration, and it is required that the latency rating does not increase monotonically when the latency level increases, and the latency rating is 1 when the latency level is zero. During calculation, the latency rating is read from the latency rating table according to the latency level. When the latency level exceeds the coverage range of the table, the latency rating corresponding to the highest level in the table is taken. The lower limit of the latency rating is taken as the minimum latency rating value in the latency rating table. When the read latency rating is lower than the lower limit of the latency rating, the latency rating is adjusted to the lower limit of the latency rating.

[0153] Within the available set, the delay score of each record is converted into a delay weight, and then normalized so that the sum of the delay weights is one and the larger the delay score, the larger the corresponding delay weight; when the total number of scores required for normalization is zero or unavailable, uniform weighting is used.

[0154] Using the mixing coefficient as an adjustment parameter, the delayed weight and the uniform weight are weighted and fused to obtain the unnormalized fused weight. The unnormalized fused weight is then normalized so that the sum of the fused weights is one.

[0155] For records whose anomaly label is not normal or whose missing repair label is missing anomaly, the fusion weight is reset to zero and they are removed during normalization.

[0156] S3.3: Based on the overall reliability of the lane group, set the set observation noise in the lane group observation input, and establish the correspondence between the lane group observation input and the observation noise to obtain the observation noise.

[0157] The lane group observation input is matched with the lane group identifier and the backtrack time slice index to determine the overall confidence level of the lane group. If the overall confidence level of the lane group is missing, it is set to zero.

[0158] For each lane group observation input, the delay level is determined and the delay score is read. If the delay score is lower than the lower limit of the delay score, the lower limit of the delay score is taken.

[0159] The length of the recent time slice interval is determined by the number of time slices. The length of the recent time slice interval is the maximum value of the number of time slices corresponding to one signal period and the upper bound of the number of candidate offset slices plus one.

[0160] The number of time slices corresponding to a signal cycle is calculated by converting the signal cycle duration and the time slice length. The signal cycle duration is preferably obtained from the intersection phase configuration information table. If it is not available, it is calculated by summing the shortest clearance time for each phase and the phase switching clearing time.

[0161] In the lane group identification dimension, select available records within the recent time slice interval. Available records are those that are marked as normal for anomalies and whose missing repair indicators are not marked as missing for anomalies.

[0162] The overall credibility and delay score of lane groups within the recent time slice interval are sorted separately, and the values ​​at the upper quartile positions of each are taken as the credibility threshold and the score threshold. Records with an overall credibility of lane groups not lower than the credibility threshold and a delay score not lower than the score threshold are selected to form an optimal set.

[0163] For records within the preferred set, calculate the squared deviation of the resampled target observation field value relative to the median across sources in the same time slice, and take the median as the observation noise baseline variance.

[0164] When the preferred set is empty, the baseline variance of the observation noise is taken as the median of the squared deviations of the available records within the recent time slice interval; when the available records are still empty, the baseline variance of the observation noise is taken as the lower limit of the observation noise variance.

[0165] Read the upper limit value and field resolution of the target observation field from the equipment calibration parameter table, and calculate the upper limit and lower limit of the observation noise variance respectively.

[0166] The observation noise variance is set based on the overall reliability of the lane group: when the overall reliability of the lane group is zero, the observation noise variance takes the upper limit of the observation noise variance; when the overall reliability of the lane group is greater than zero, the variance of the observation noise increases as the overall reliability of the lane group decreases, and also increases as the delay score decreases, based on the baseline variance of the observation noise.

[0167] For records where the anomaly label is not normal or the missing repair label is missing anomaly, the observation noise variance is taken as the upper limit of the observation noise variance.

[0168] For the remaining records, the observation noise variance is constrained between the lower and upper limits of the observation noise variance, and a correspondence is established between the lane group observation inputs and the observation noise variance to obtain the observation noise variance, expressed as:

[0169] ;

[0170] In the formula, Lane group markings Data source identification No. The backtracking time slice index sets the observation noise variance. Lane group markings Data source identification No. Backtracking time slice index observation noise baseline variance Lane group markings In the The overall credibility of the backtracking time slice index lane group (value range is 0-1). Lane group markings Data source identification No. The delay score of the backtracking time slice index (the value ranges from 0 to 1). The lower bound of the variance of the observed noise. This is the upper limit of the variance of the observation noise. For backtracking time slice index, This serves as an identifier for the data source.

[0171] S3.4: Perform fusion filtering on the lane group observation input using fusion weights and observation noise, and output traffic state estimates and corresponding uncertainty representations.

[0172] The lane group observation inputs are summarized according to the lane group identifier and the retrospective time slice index. Records that are marked as normal and whose missing repair identifiers are not missing anomalies are filtered to form a filter input set. The records in the filter input set are matched with fusion weights and the observation noise variance is set.

[0173] When the filter input set is not empty, the resampled target observation field values ​​in the filter input set are weighted and summarized according to the fusion weight to obtain the fused observation value, and the fused observation noise variance is obtained by weighting and summing the set observation noise variance with the square of the fusion weight as the coefficient.

[0174] When the filter input set is empty, the fused observations are set to empty, and the fused observation noise variance is set to empty.

[0175] A scalar random walk prediction model is used for state prediction. The predicted traffic state estimate is taken from the traffic state estimate corresponding to the previous backtracking time slice index. When the traffic state estimate corresponding to the previous backtracking time slice index does not exist and the filtered input set is not empty, the predicted traffic state estimate is taken from the fused observation value.

[0176] When the traffic state estimate corresponding to the previous backtracking time slice index does not exist and the filter input set is empty, the predicted traffic state estimate is set to empty.

[0177] Within the lane group identification dimension, the process noise variance is calculated based on the median of the squared differences in traffic state estimates between adjacent time slices within the recent time slice interval;

[0178] The prediction uncertainty characterization is obtained by superimposing the process noise variance on the uncertainty characterization of the previous backtracking time slice index.

[0179] When the traffic state estimate within the recent time slice interval is insufficient for calculation, the process noise variance is taken as the median of the observed noise variance within the recent time slice interval; if it is still not available, the process noise variance is taken as the lower limit of the observed noise variance.

[0180] When the filter input set is empty, the traffic state estimate is the predicted traffic state estimate, and the uncertainty representation is the predicted uncertainty representation and output.

[0181] When the filtered input set is not empty, the Kalman gain is calculated based on the prediction uncertainty characterization and the fused observation noise variance, and the predicted traffic state estimate and prediction uncertainty characterization are updated in combination with the fused observation values, and the traffic state estimate and uncertainty characterization are output.

[0182] Specifically, the Kalman gain is calculated based on the prediction uncertainty characterization and the fused observation noise variance, and the predicted traffic state estimate and prediction uncertainty characterization are updated using the fused observation values, as expressed below:

[0183] ;

[0184] In the formula, Lane group markings No. Kalman gain of backtracking time slice index, Lane group markings No. Characterization of predictive uncertainty in backtracking time slice indexes Lane group markings No. Traffic state estimation after updating the backtracking time slice index. Lane group markings No. Traffic state estimation based on backtracking time slice index. Lane group markings No. Fusion observations from backtracking time slice indexes, Lane group markings No. Uncertainty representation after updating the backtracking time slice index.

[0185] S4: By estimating traffic conditions and using a unified observation set, external arrival flow and turning ratio are inferred and corresponding statistics are established. The corresponding statistics are updated online based on the comprehensive confidence level of the intersection, and the parameter update statistics are obtained.

[0186] S4.1: Using traffic state estimation and a unified observation set, the observation records are grouped by lane group and aligned along a unified time axis, and summarized into entrance direction observation records and exit direction observation records respectively.

[0187] The unified observation set and delay information are associated with lane group identifier, data source identifier and time slice index to obtain arrival delay, transmission delay and time slice length, and the backtracking time slice index already written in the unified observation set record is used.

[0188] When the backtracking time slice index is missing, the number of backtracking slices is calculated based on the arrival delay, transmission delay and time slice length, and the backtracking time slice index is obtained by backtracking the time slice index and then written into the unified observation set record.

[0189] By aligning the traffic state estimation and uncertainty characterization records with the unified observation set records using lane group identifiers and backtracking time slice indexes, lane group aligned observation records are obtained.

[0190] Based on the configuration table of the correspondence between the detection area and the lane, the geometric range of the lane and the geometric range of the intersection boundary and the direction identifier are obtained; the geometric range of the lane and the geometric range of the intersection boundary are represented by polygons under the same projection coordinate system, and the distance unit is meters; the degree of overlap and the geometric distance between the geometric range of the lane and the geometric range of the intersection boundary are calculated, and the degree of overlap is determined according to the ratio of the overlap length to the length of the lane geometric range.

[0191] The entrance and exit directions are determined by the degree of overlap. When the degree of overlap is equal, the direction is determined by the geometric distance from smallest to largest. When the degree of overlap is equal and the difference between the smallest and second smallest geometric distance does not meet the preset distance discrimination threshold, a direction missing identifier is written and the corresponding lane number is removed.

[0192] Based on the correspondence between lane number and lane group identifier, the correspondence between lane group identifier and entrance direction and the correspondence between lane group identifier and exit direction are generated.

[0193] Lane group aligned observation records are aggregated along the dimensions of entrance direction and backtrack time slice index. Traffic state estimation is used as the entrance direction observation value. The output entrance direction observation record contains the fields of entrance direction, backtrack time slice index and entrance direction observation value.

[0194] Lane group aligned observation records are aggregated in the dimensions of exit direction and backtrack time slice index. Traffic state estimation is used as the exit direction observation value. The output exit direction observation record contains the fields of exit direction, backtrack time slice index and exit direction observation value.

[0195] It should be noted that the topological relationship between lanes and intersection boundaries and lane orientation are used to generate benchmark matching. The minimum boundary value of the overlap and distance difference distribution between benchmarks and non-benchmarks is selected as the overlap threshold and distance discrimination threshold and then fixed in the configuration table.

[0196] S4.2: Based on the observation records of the entrance direction and the observation records of the exit direction, identify the net arrival change at the intersection boundary and remove the non-external arrival components generated inside the intersection to form an external arrival flow inference record.

[0197] By collecting the observation records in the ingress direction and the observation records in the egress direction along the time slice index dimension, the total number of observations in the ingress direction and the total number of observations in the egress direction are obtained.

[0198] In the backtracking time slice index dimension, the non-external arrival component is determined as follows: when the total number of observations in the exit direction is higher than the total number of observations in the inlet direction, the difference between the two is taken; when the total number of observations in the exit direction is not higher than the total number of observations in the inlet direction, the value is zero.

[0199] In the entrance direction dimension, the non-external arrival components are allocated according to the proportion of the entrance direction observations in the total entrance direction observations; when the total entrance direction observations are zero, the non-external arrival components corresponding to each entrance direction are zero.

[0200] For each inlet direction, the inferred external arrival flow rate is obtained by subtracting the corresponding non-external arrival component from the inlet direction observation. When the subtraction result is negative, the inferred external arrival flow rate is set to zero.

[0201] The ingress direction, backtracking time slice index, and external arrival traffic inference value are combined to form an external arrival traffic inference record.

[0202] S4.3: Establish a correspondence between the external arrival flow inference records and traffic state estimates according to the release direction, summarize them to form a turning allocation record, and calculate the allocation ratio according to the release direction to form a turning ratio record.

[0203] Based on the correspondence between lane group identifiers and entrance directions, determine the set of lane group identifiers corresponding to the entrance direction; based on the intersection phase configuration information, determine the correspondence between lane group identifiers and the permitted direction.

[0204] Traffic state estimates for each lane group within the corresponding lane group identifier set in the entrance direction are read and aggregated along the dimensions of the entrance direction and the backtracking time slice index to obtain the total traffic state estimate. When the total traffic state estimate is zero, the allocation value of each lane group identifier is set to zero. When the total traffic state estimate is not zero, the proportion is determined according to the relative share of each lane group's traffic state estimate in the total traffic state estimate.

[0205] The estimated external arrival flow value corresponding to the entrance direction is allocated to each lane group according to the proportion, and the lane group identification allocation value is obtained.

[0206] By aggregating the lane group identifier allocation values ​​along the dimensions of the release direction and the retrospective time slice index, the release direction allocation value is obtained, forming a steering allocation record.

[0207] The allocation summary value is obtained by aggregating all release direction allocation values ​​in the same backtracking time slice index dimension. When the allocation summary value is zero, the turning ratio of each release direction is zero. When the allocation summary value is not zero, the turning ratio is determined according to the relative share of each release direction allocation value in the allocation summary value, and a turning ratio record is formed.

[0208] S4.4: Based on the external arrival flow inference records and turning ratio records, establish external arrival flow statistics and turning ratio statistics respectively. Perform online updates on external arrival flow statistics and turning ratio statistics through intersection comprehensive confidence, and output parameter update statistics.

[0209] Extract the overall credibility of the intersection from the backtracking time slice index dimension; if it is missing, fill in the most recent available value in the credibility backtracking window, and if it is still missing, take zero and write the credibility missing flag. The length of the credibility backtracking window is the upper bound of the candidate offset slice number plus one.

[0210] The overall confidence level of the intersection is truncated to zero to one to obtain the online update coefficient. External arrival traffic statistics are established in the entrance direction dimension, and turning ratio statistics are established in the exit direction dimension. The initial value of the statistics is the first occurrence of the external arrival traffic inference value or turning ratio.

[0211] Read external arrival traffic inference records in the ingress direction and backtrack time slice index dimensions; when a record is missing, the external arrival traffic statistics retain the value of the previous backtrack time slice index; when the previous value is missing, the external arrival traffic statistics are set to zero and a statistics missing flag is written.

[0212] Read the steering ratio record in the dimensions of release direction and backtrack time slice index; when the record is missing, the steering ratio statistic retains the value of the previous backtrack time slice index; when the previous value is missing, the steering ratio statistic is set to zero and a statistic missing flag is written.

[0213] The statistics on external arrival flow and turning ratio are updated online. The updated statistics are the weighted sum of the current inferred value and the previous statistics. The weight of the current inferred value is the online update coefficient, and the weight of the previous statistics is one minus the online update coefficient. When the current inferred value is missing, the updated statistics retain the value of the previous statistics.

[0214] Output parameter update statistics. Fields should include at least the backtracking time slice index, inbound direction, external arrival traffic statistics, release direction, turning ratio statistics, online update coefficient, and necessary identifiers.

[0215] S5: Based on traffic state estimation, uncertainty characterization and parameter update statistics, determine the risk tolerance parameter and construct probabilistic constraints. Transform the probabilistic constraints into convex constraints that can be solved in real time for rolling timing solution, and output signal control commands and feasibility margins.

[0216] S5.1: Align traffic state estimates, uncertainty representations, and parameter update statistics with the same time slice and perform consistent processing to form risk assessment information.

[0217] The backtracking time slice index is used as a unified time slice marker. The current backtracking time slice index is taken from the backtracking time slice index corresponding to the control time. The control time is determined by the starting point of the signal cycle.

[0218] Under the current backtracking time slice index, read the traffic state estimation and uncertainty characterization records, and read the parameter update statistics records; when the parameter update statistics are missing, search for the most recent available record in the statistics backtracking window according to the backtracking time slice index from near to far. If it is still missing, set the parameter update statistics to zero and write the statistics missing flag.

[0219] The length of the statistical backtracking window is the greater of the number of time slices corresponding to the fixed period length and the upper bound of the number of candidate offset slices plus one. The upper bound of the number of candidate offset slices is first read from the configuration table. If the configuration is missing, the length of the statistical backtracking window is reduced by one, and an upper bound missing flag is written.

[0220] Based on the correspondence between lane number and lane group identifier and the intersection phase configuration information, the traffic state estimation and uncertainty characterization records are written into the release direction and signal phase.

[0221] The entry direction is written based on the configuration table of the correspondence between detection area and lane and the correspondence between lane number and lane group identifier.

[0222] In the parameter update statistics, the external arrival flow statistics are matched according to the inlet direction, and the turning ratio statistics and online update coefficients are matched according to the release direction.

[0223] Traffic condition estimates, uncertainty characteristics, signal phases, release directions, entrance directions, external arrival flow statistics, turning ratio statistics, and online update coefficients are summarized and organized according to the current backtracking time slice index to form risk assessment information.

[0224] S5.2: Based on the risk assessment information, the degree of dispersion of uncertainty characteristics and the update changes of parameter update statistics are jointly summarized to form risk-sensitive information.

[0225] In the signal phase dimension, the uncertainty representation values ​​corresponding to the same signal phase in the risk assessment information are collected and the median is taken to obtain the signal phase uncertainty representation summary value. The signal phase uncertainty representation summary value is used as the signal phase dispersion summary value, wherein the uncertainty representation is a variance representation.

[0226] In the signal phase dimension, the traffic state estimates corresponding to the same signal phase in the risk assessment information are aggregated and the median is taken to obtain the summative traffic state estimate for the signal phase.

[0227] In the inbound direction dimension, calculate the absolute change of the external arrival traffic statistics relative to the previous backtracking time slice index at the current backtracking time slice index. If the previous value is missing, set the change to zero and write a missing change flag. Similarly, in the outbound direction dimension, calculate the absolute change of the turning ratio statistics and write a missing change flag.

[0228] The median change in external arrival statistics is obtained by taking the median of the change in external arrival statistics in the signal phase dimension, and the median change in steering ratio is obtained by taking the median of the change in steering ratio statistics in the signal phase dimension.

[0229] The median of the external arrival update change and the median of the steering ratio update change are taken to obtain the sum of the signal phase update changes.

[0230] The online update coefficient is truncated to zero to one. The risk-sensitive summary value is obtained by weighting the summary value of dispersion and the summary value of update change in the signal phase dimension. When weighting the summary value, the weight of the summary value of dispersion is the online update coefficient, and the weight of the summary value of update change is one minus the online update coefficient.

[0231] Based on the turning ratio statistics, the external arrival traffic statistics are converted to obtain the release direction demand statistics. Then, according to the fixed correspondence between the release direction and the signal phase in the intersection phase configuration information, the statistics are mapped to the signal phase dimension to obtain the signal phase demand statistics. The statistics are then organized to form risk-sensitive information and retain the change and missing indicators.

[0232] S5.3: Determine the risk tolerance parameters based on risk-sensitive information, and construct probabilistic constraints by combining traffic state estimation and uncertainty characterization.

[0233] The range of risk tolerance parameters is determined in the signal phase dimension, and the lower and upper limits of the risk tolerance parameters are verified for feasibility based on the fixed cycle length and the shortest clearance time of each phase in the intersection phase configuration information and then fixed into the configuration table.

[0234] The feasibility check requires that there still be an allocatable release time under the condition that the minimum release time constraint for each phase and the fixed cycle length constraint are met simultaneously, so as to ensure that the rolling timing problem retains the feasible solution space.

[0235] Read the feasible margin output from the previous backtracking time slice index, convert the feasible margin into a margin percentage based on a fixed period length, and limit it to zero to one;

[0236] When the feasible margin corresponding to the previous backtracking time slice index is missing, the margin percentage is set to zero and a feasible margin missing flag is written.

[0237] Read the risk-sensitive summary value in the signal phase dimension, convert the risk-sensitive summary value into a risk percentage according to a fixed period length, and limit the risk percentage to zero to one; when the risk-sensitive summary value is missing, take the risk percentage as one and write it into the risk missing flag.

[0238] The margin ratio and the risk ratio together determine the risk tolerance parameter:

[0239] The larger the margin ratio, the closer the risk tolerance parameter is to the upper limit; the larger the risk ratio, the closer the risk tolerance parameter is to the lower limit.

[0240] When the adjustment direction given by the margin ratio and the risk ratio is inconsistent, the risk tolerance parameter takes a conservative result that is closer to the lower limit, and the risk tolerance parameter is restricted between the lower limit and the upper limit of the risk tolerance parameter.

[0241] The fixed cycle length and the minimum clearance time for each phase are determined based on the intersection phase configuration information. When the fixed cycle length is missing, it is obtained by summing the minimum clearance time for each phase and the phase switching clearing time and writing it into the cycle missing identifier.

[0242] The remaining allocable clearance time is calculated in the signal phase dimension. The remaining allocable clearance time is then allocated according to the proportion of signal phase demand statistics and summed with the shortest clearance time to obtain the average target clearance time.

[0243] When the total value of the signal phase demand statistics is zero, the remaining release time can be allocated evenly according to the signal phase and written into the demand missing flag.

[0244] Read the summary value of signal phase uncertainty characterization in the signal phase dimension, calculate the square root of the summary value of signal phase uncertainty characterization to obtain the square root value, and use the square root value as the standard deviation of target release time.

[0245] When the summary value representing the signal phase uncertainty is missing, the most recent available summary value is retrieved from the nearest to the furthest within the uncertainty backtracking window according to the backtracking time slice index.

[0246] The square root value is obtained by performing a square root operation on the supplemented summary value. The square root value is used as the standard deviation of the target release time. If it is still missing, the standard deviation of the target release time is taken from the default value in the configuration table and an uncertainty missing flag is written.

[0247] The backtracking time slice index, signal phase, risk tolerance parameters, mean target release duration, standard deviation of target release duration, fixed cycle length and minimum release duration are summarized and organized to form a probability constraint record and output.

[0248] S5.4: Align and organize the probability constraints and uncertainty representations in the same time slice to obtain the aligned probability constraints.

[0249] Using the backtracking time slice index and the signal phase as keys, the probability constraint record is matched and aligned with the summary value of the signal phase uncertainty characterization. The summary value of the signal phase uncertainty characterization is square-rooted and the square root value is written into the aligned probability constraint as the standard deviation of the target release time.

[0250] When the summary value representing the signal phase uncertainty is missing, the most recent available summary value is retrieved from the nearest to the furthest within the uncertainty backtracking window according to the backtracking time slice index.

[0251] The square root value is obtained by performing a square root operation on the supplemented summary value. The square root value is used as the standard deviation of the target release time. If it is still missing, the standard deviation of the target release time is taken from the default value in the configuration table and an uncertainty missing flag is written.

[0252] The uncertainty backtracking window length is taken as the number of time slices corresponding to the fixed period length; when the fixed period length is missing, the default window length in the configuration table is taken and a window missing flag is written. Output the aligned probability constraints.

[0253] S5.5: Transform the aligned probabilistic constraints into convex constraints that can be solved in real time, and organize the constraint terms to form a set of convex constraints.

[0254] A one-sided probability constraint is constructed in the signal phase dimension. The one-sided probability constraint requires that the probability that the random variable of target release time does not exceed the signal phase release time decision variable is not lower than a preset probability threshold.

[0255] Given that the distribution of target release time is unknown and only the mean and variance are required to exist, a one-sided Chebyshev bound is used to transform the one-sided probabilistic constraint into a deterministic convex constraint.

[0256] Here, it is transformed into a deterministic convex constraint, expressed as:

[0257] ;

[0258] In the formula, For signal phase identification, This represents the average target release time for the signal phase at the corresponding backtracking time slice index. The standard deviation of the target release duration is the backtracking time slice index corresponding to the signal phase. The safety margin coefficient determined for the risk tolerance parameter. This is the risk tolerance parameter (its value is strictly between 0 and 1).

[0259] In deterministic convex constraints, the mean target release time is taken as the mean target release time in the aligned probability constraint record, the standard deviation of target release time is taken as the standard deviation of target release time in the aligned probability constraint record, the signal phase release time is used as the decision variable, and the safety margin coefficient is determined by the risk tolerance parameter.

[0260] The deterministic convex constraints obtained from the transformation of the one-sided Chebyshev boundary are combined with the shortest signal phase release time constraint, the fixed period length constraint, and the non-negative signal phase release time constraint to form a set of convex constraints.

[0261] It should be noted that the preset probability threshold is the complement of the risk tolerance parameter, and the range of the preset probability threshold is determined by the lower and upper limits of the risk tolerance parameter. For example, if the risk tolerance parameter is preset to 0.01–0.20, the corresponding probability threshold is preset to 0.80–0.99 (the smaller the risk tolerance parameter, the larger the probability threshold).

[0262] S5.6: Under the constraints of the convex constraint set, combine traffic state estimation to perform rolling timing solution and output signal control commands, and summarize the feasible margin based on the constraint redundancy of the signal control commands under the convex constraint set.

[0263] Set the scrolling window length. The scrolling window length should be no less than the number of time slices corresponding to the fixed period length and no less than the minimum window length given in the configuration table.

[0264] The current backtracking time slice index is used as the starting point of the scrolling window, and the scrolling window covers the length of the consecutive scrolling window backtracking time slice indices after the starting point.

[0265] Within the rolling window, a convex optimization problem is constructed using the release duration of each signal phase as the decision variable.

[0266] The fixed period length is denoted as the period length, and the proportion of release time for each signal phase is calculated.

[0267] Simultaneously, based on the sum of the signal phase demand statistics and the signal phase traffic state estimation, the signal phase priority ratio is obtained, and the priority ratio is normalized.

[0268] When the normalized denominator is zero, the phase priority percentage of each signal is set to zero and a normalized missing flag is written.

[0269] The squared deviations of the percentage of release time and the percentage of priority for each signal phase are calculated and weighted to form a convex target. The weights are taken from the sum of the estimated traffic state values ​​of the signal phases. When the sum of the estimated traffic state values ​​of the signal phases is zero, the weights are uniformly distributed and a weight missing flag is written.

[0270] Under the constraints of the convex constraint set, the optimal solution within the rolling window is obtained by using the convex optimization method; the release duration of each signal phase corresponding to the current backtracking time slice index is read from the optimal solution, and signal control commands are generated and output.

[0271] For each constraint in the convex constraint set, calculate the difference between the value on the right side and the value on the left side of the constraint under the optimal solution as the constraint redundancy; determine the minimum value among the constraint redundancy as the feasibility margin and output it.

[0272] S6: Issues and executes signal control commands based on feasibility margin and overall intersection credibility, and adaptively switches between rolling timing control and degraded control.

[0273] S6.1: Match and align the signal control command, feasibility margin, and intersection comprehensive credibility in the same time slice. Summarize the matched signal control command with the corresponding feasibility margin and intersection comprehensive credibility to form a judgment item and obtain the execution judgment information.

[0274] The backtracking time slice index is used as the same time slice marker. Under the current backtracking time slice index, the signal control command, feasibility margin and intersection comprehensive confidence are obtained and matched. When the feasibility margin is missing, it is set to zero and a feasibility margin missing flag is written.

[0275] When the overall credibility of an intersection is missing, the most recent available value is retrieved from the nearest to the furthest in the credibility backtracking window according to the backtracking time slice index. If it is still missing, zero is retrieved and a credibility missing flag is written. The length of the credibility backtracking window is given by the configuration table and is not less than the number of time slices corresponding to the fixed period length.

[0276] Extracting anomalous states from a unified set of observations along the time slice index dimension:

[0277] If the current backtracking time slice index contains an anomaly marked as an observation anomaly or a missing item repair marked as a missing item anomaly, and the previous backtracking time slice index did not contain the same anomaly, then the newly added anomaly label should be set to "existing"; if the previous backtracking time slice index is missing, the newly added anomaly label should be set to "existing"; otherwise, it should be set to "not existing".

[0278] The signal control instructions, feasibility margin, overall intersection reliability, and newly added anomaly markers are aggregated according to the signal phase to form execution decision information.

[0279] S6.2: When the feasibility margin has constraint redundancy under the convex constraint set, the intersection comprehensive confidence satisfies the stability of continuous time slices, and there are no new anomaly labels in the unified observation set, issue the execution signal control command and maintain rolling timing control.

[0280] The feasibility margin determined based on the execution decision information has a constraint surplus:

[0281] A feasible margin greater than zero is considered to have a constrained surplus, while a feasible margin not greater than zero is considered to have no constrained surplus.

[0282] Set the stability determination window length, and the window covers consecutive backtracking time slice indices with the current backtracking time slice index as the endpoint and the length of the window as the length;

[0283] When the available consecutive backtracking time slice indices before the current backtracking time slice index are insufficient to form a stability determination window of length, the overall confidence level of the intersection does not meet the consecutive time slice stability requirement.

[0284] Within the window, if the overall credibility of the intersection is not lower than the preset stable threshold for overall credibility of the intersection, and the difference between the maximum and minimum values ​​of the overall credibility of the intersection within the window does not exceed the preset fluctuation threshold for overall credibility of the intersection, then the overall credibility of the intersection is determined to meet the stability of continuous time slice; otherwise, it is determined not to meet the stability.

[0285] If the newly added anomaly label is marked as non-existent, it is determined that there are no newly added anomaly labels in the unified observation set; if the newly added anomaly label is marked as existent, it is determined that there are newly added anomaly labels in the unified observation set.

[0286] When all three conditions are met simultaneously—feasibility margin with sufficient constraints, intersection comprehensive credibility satisfying continuous time slice stability, and no new anomaly annotations in the unified observation set—the signal control command corresponding to the current backtracking time slice index is issued and the rolling timing control state is maintained.

[0287] It should be noted that the stability threshold and fluctuation threshold are determined by using time slices within a window that have no new anomalies, no missing anomalies, and whose delay does not exceed the upper limit as stable samples. The stability threshold is the lowest value of the intersection comprehensive confidence within the window in the stable samples (the example value is 0.50–0.80), and the intersection comprehensive confidence fluctuation threshold (the example value is 0.05–0.20) is the maximum and minimum difference of the intersection comprehensive confidence within the window in the stable samples, and the configuration is fixed.

[0288] S6.3: When the feasibility margin does not have constraint redundancy under the convex constraint set, or the intersection comprehensive confidence does not meet the continuous time slice stability, or new anomaly labels appear in the unified observation set, stop issuing execution signal control commands and switch to degraded control.

[0289] Under the current retrospective time slice index, if the feasibility margin is not greater than zero, or the overall confidence level of the intersection does not meet the stability of continuous time slices, or a new anomaly marker is added, then the issuance of signal control commands will be stopped and the system will switch to degraded control.

[0290] Read traffic state estimates and historical reference observations based on lane group identifiers to generate conservative lane group observations: if both are available, take the minimum value; if only one is available, take the available value; if neither is available, take zero.

[0291] The lane groups are mapped to the signal phases according to the intersection phase configuration information. The conservative demand value of the signal phase is obtained by taking the median of the conservative lane group observations in the signal phase dimension.

[0292] The remaining release time can be allocated based on the fixed cycle length and the shortest release time for each phase, and then allocated according to the proportion of conservative demand value for each signal phase; when the conservative demand value is summed up to zero, it is evenly allocated.

[0293] The minimum release time and the allocated time are combined to obtain the conservative signal phase release time, which is then used to form and issue conservative signal control commands to complete the downgrade control.

[0294] This embodiment also provides a traffic signal adaptive system, including:

[0295] The multi-source alignment module is used to acquire multi-source traffic detection data and perform time consistency preprocessing according to lane groups to form a unified observation set and delay information;

[0296] The credibility assessment module is used to calculate the credibility weight of each traffic detection data based on a unified observation set, delay information and historical reference observation values, and aggregate them to obtain the comprehensive credibility of lane groups and the comprehensive credibility of intersections.

[0297] The fusion filtering module is used to adaptively set the fusion weight and observation noise based on the comprehensive confidence of the lane group and perform fusion filtering to output traffic state estimation and corresponding uncertainty characterization.

[0298] The parameter inference module is used to infer external arrival flow and turning ratio through traffic state estimation and unified observation set and establish corresponding statistics. Based on the comprehensive confidence of the intersection, the corresponding statistics are updated online to obtain parameter update statistics.

[0299] The risk optimization module is used to determine the risk tolerance parameter and construct the probability constraint based on traffic state estimation, uncertainty characterization and parameter update statistics. It then transforms the probability constraint into a convex constraint that can be solved in real time for rolling timing solution and outputs signal control commands and feasibility margin.

[0300] The execution switching module is used to issue execution signal control commands based on feasibility margin and intersection comprehensive credibility, and adaptively switch between rolling timing control and degraded control.

[0301] This embodiment also provides a computer device applicable to the traffic signal adaptive control method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the traffic signal adaptive control method proposed in the above embodiment.

[0302] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0303] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the traffic signal adaptive control method as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0304] In summary, this invention achieves the following through:

[0305] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A traffic signal adaptive control method, characterized in that: include, Acquire multi-source traffic detection data and perform time consistency preprocessing according to lane groups to form a unified observation set and delay information; The credibility weights of each traffic detection data are calculated based on a unified observation set, delay information, and historical reference observations, and then aggregated to obtain the overall credibility of lane groups and the overall credibility of intersections. Based on the comprehensive confidence of lane groups, the fusion weights and observation noise are adaptively set and fusion filtering is performed to output traffic state estimates and corresponding uncertainty representations. By estimating traffic conditions and using a unified observation set, external arrival flow and turning ratio are inferred and corresponding statistics are established. The corresponding statistics are then updated online based on the overall credibility of the intersection, resulting in parameter update statistics. Based on traffic state estimation, uncertainty characterization and parameter update statistics, risk tolerance parameters are determined and probabilistic constraints are constructed. The probabilistic constraints are transformed into convex constraints that can be solved in real time for rolling timing solution, and the signal control command and feasibility margin are output. Based on the feasibility margin and the overall credibility of the intersection, the signal control command is issued and executed, and adaptive switching is performed between rolling timing control and degraded control.

2. The traffic signal adaptive control method as described in claim 1, characterized in that: The specific steps for forming a unified observation set and delay information are as follows. Acquire multi-source traffic detection data and extract time stamps, lane attribution information, and data source identifiers to form labeled multi-source traffic detection data; Based on lane affiliation information, lanes with the same release direction and corresponding signal phase are identified as lane groups. Marked multi-source traffic detection data are merged by lane groups and subjected to unified time axis alignment and resampling to form aligned multi-source observation sequences of lane groups. In the aligned lane group multi-source observation sequence, arrival delay and transmission delay are identified and missing data are repaired and anomaly labeling is performed, and a unified observation set and delay information are output.

3. The traffic signal adaptive control method as described in claim 1, characterized in that: The process involves calculating the credibility weights of each traffic detection data point based on a unified observation set, delay information, and historical reference observations, and then aggregating them to obtain the overall credibility of lane groups and intersections. The specific steps are as follows: Consistency verification of the unified observation set at the same time slice is performed to form consistency information, which is then correlated with the delay information to form credibility assessment information; Historical time window observation records corresponding to the current time slice are selected from the unified observation set, and after time backtracking and alignment based on delay information, they are summarized by lane group to form historical reference observation values. Deviation assessment is performed based on credibility assessment information combined with historical reference observations, and credibility weights for each traffic detection data are calculated. The credibility weights of each traffic detection data are aggregated by lane group to obtain the overall credibility of the lane group, and then the overall credibility of the lane group is aggregated at the intersection dimension to obtain the overall credibility of the intersection.

4. The traffic signal adaptive control method as described in claim 1, characterized in that: The traffic state estimation and corresponding uncertainty characterization are described in the following steps. Based on the overall credibility of lane groups, the fusion weights in the lane group observation inputs are set, and the correspondence between the lane group observation inputs and the fusion weights is established to obtain the fusion weights; Based on the overall reliability of the lane group, the set observation noise in the lane group observation input is set, and the correspondence between the lane group observation input and the observation noise is established to obtain the observation noise; The fusion filtering of lane group observation inputs is performed using fusion weights and observation noise, and the output is a traffic state estimate and the corresponding uncertainty representation.

5. The traffic signal adaptive control method as described in claim 1, characterized in that: The process involves estimating external arrival flow and turning ratios using traffic state estimation and a unified observation set, establishing corresponding statistics, and updating these statistics online based on the intersection's overall reliability to obtain updated parameter statistics. The specific steps are as follows: Using traffic state estimation and a unified observation set, the observation records were grouped by lane group and aligned along a unified time axis, and then summarized into entrance direction observation records and exit direction observation records respectively. Based on the observation records of the entrance direction and the observation records of the exit direction, the net arrival change at the intersection boundary is identified and the non-external arrival components generated inside the intersection are removed to form an external arrival flow inference record; Establish a correspondence between external arrival flow inference records and traffic state estimates according to the direction of release, summarize them to form a turning allocation record, and calculate the allocation ratio according to the direction of release to form a turning ratio record; Based on the external arrival traffic inference records and turning ratio records, external arrival traffic statistics and turning ratio statistics are established respectively. The external arrival traffic statistics and turning ratio statistics are updated online by the intersection comprehensive confidence level, and the updated parameters are output.

6. The traffic signal adaptive control method as described in claim 1, characterized in that: The specific steps for determining the risk tolerance parameter and constructing probabilistic constraints are as follows: Traffic state estimates, uncertainty representations, and parameter update statistics are aligned with the same time slice and then standardized to form risk assessment information. Based on risk assessment information, the degree of dispersion of uncertainty characteristics and the update changes of parameter update statistics are jointly summarized to form risk-sensitive information; Risk tolerance parameters are determined based on risk-sensitive information, and probabilistic constraints are constructed by combining traffic state estimation and uncertainty characterization.

7. The traffic signal adaptive control method as described in claim 1, characterized in that: The process of transforming probabilistic constraints into real-time solvable convex constraints for rolling time-matching solution, and outputting control commands and feasibility margins, involves the following specific steps: The probability constraints and uncertainty representations are aligned and organized in the same time slice to obtain the aligned probability constraints. The aligned probabilistic constraints are transformed into convex constraints that can be solved in real time, and the constraint terms are organized to form a set of convex constraints. Under the constraints of the convex constraint set, the rolling timing solution is executed in conjunction with traffic state estimation to output signal control commands, and the feasible margin is summarized based on the constraint redundancy of the signal control commands under the convex constraint set.

8. The traffic signal adaptive control method as described in claim 1, characterized in that: The specific steps for issuing and executing signal control commands based on feasibility margin and overall intersection reliability, and adaptively switching between rolling timing control and degraded control, are as follows. The signal control command, feasibility margin, and intersection comprehensive credibility are matched and aligned in the same time slice. The matched signal control command and the corresponding feasibility margin and intersection comprehensive credibility are summarized to form a judgment item, and the execution judgment information is obtained. When the feasibility margin has constraint surplus under the convex constraint set, the intersection comprehensive credibility satisfies the continuous time slice stability, and there are no new anomaly labels in the unified observation set, the execution signal control command is issued and rolling timing control is maintained. When the feasible margin does not have constraint redundancy under the convex constraint set, or the intersection comprehensive confidence does not meet the continuous time slice stability, or new anomaly labels appear in the unified observation set, the issuance of execution signal control commands is stopped and the system switches to degraded control.

9. The traffic signal adaptive control method as described in claim 1, characterized in that: The aforementioned degradation control refers to the conservative signal control command generated and output based on a unified observation set, traffic state estimation, and historical reference observations when the execution judgment information does not meet the execution conditions.

10. A traffic signal adaptive system, based on the traffic signal adaptive control method according to any one of claims 1 to 9, characterized in that: include, The multi-source alignment module is used to acquire multi-source traffic detection data and perform time consistency preprocessing according to lane groups to form a unified observation set and delay information; The credibility assessment module is used to calculate the credibility weight of each traffic detection data based on a unified observation set, delay information and historical reference observation values, and aggregate them to obtain the comprehensive credibility of lane groups and the comprehensive credibility of intersections. The fusion filtering module is used to adaptively set the fusion weight and observation noise based on the comprehensive confidence of the lane group and perform fusion filtering to output traffic state estimation and corresponding uncertainty characterization. The parameter inference module is used to infer external arrival flow and turning ratio through traffic state estimation and unified observation set and establish corresponding statistics. Based on the comprehensive confidence of the intersection, the corresponding statistics are updated online to obtain parameter update statistics. The risk optimization module is used to determine the risk tolerance parameter and construct the probability constraint based on traffic state estimation, uncertainty characterization and parameter update statistics. It then transforms the probability constraint into a convex constraint that can be solved in real time for rolling timing solution and outputs signal control commands and feasibility margin. The execution switching module is used to issue execution signal control commands based on feasibility margin and intersection comprehensive credibility, and adaptively switch between rolling timing control and degraded control.

Images