Intersection timing signal control multi-period automatic division method

Abstract

The invention discloses an intersection timing signal control multi-period automatic division method. The method comprises the following steps: processing intersection traffic flow data; calculating the traffic flow of the key lane in the lane group; discretizing the key lane total flow sequence into a character coding sequence; carrying out sampling clustering on the character coding sequence; and performing signal control time interval segmentation and adjustment based on the clustering tag sequence. According to the method, the full-day lane-level five-minute flow of the intersection is checked and repaired, the key lane total flow of the intersection is calculated according to channelization information and selectable constraints, control period division is performed based on the key lane total flow, and a signal timing scheme is more reasonable.

Description

Technical Field

[0001] The invention belongs to the field of road traffic signal control, and in particular relates to a method for automatically dividing multiple time periods of intersection timing signal control. Background Art

[0002] Intersection timed signal control multi-period division is a technology that divides the 24 hours of the day into multiple time intervals by analyzing the changing characteristics of traffic flow, and sets different traffic signal timing plans according to the traffic flow demand of each interval.

[0003] The existing methods for dividing multiple time periods of intersection time signal control mainly include: Fisher optimal segmentation algorithm, rule-based and manually set threshold method, Kmeans and other general clustering algorithms; but most of them have the following defects: Fisher optimal segmentation and its improved algorithm belong to enumeration algorithms, with low calculation efficiency, and the setting of different class diameters and intra-class deviations has a great influence on the calculation results; the traffic signal control time period division method based on rule and manual threshold setting is not universal; directly using Kmeans, K-Shape and other algorithms, because it does not consider the changing characteristics of time series data, will cause the signal control time period division results to be inconsistent with the actual traffic flow law. The process steps of some other methods are too cumbersome and difficult to be put into actual engineering use. Summary of the invention

[0004] In view of the above-mentioned deficiencies of the prior art, the purpose of the present invention is to provide a method for automatically dividing multiple time periods of intersection time signal control, so as to solve the problem of low computational efficiency caused by enumerating multiple combinations in the prior art method, and the problem of unreasonable segmentation results caused by ignoring the periodic characteristics and noise of time series data. The method of the present invention inspects and repairs the 5-minute traffic flow at the lane level of the intersection throughout the day, calculates the total traffic flow of the key lanes of the intersection based on the channelization information and optional constraints (whether the right-turn traffic flow is considered), and divides the control period based on the total traffic flow of the key lanes, so that the signal timing plan is more reasonable.

[0005] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] The invention provides a method for automatically dividing multiple time periods of intersection timing signal control, comprising the following steps:

[0007] 1) Processing intersection traffic flow data;

[0008] 2) Calculate the traffic flow of key lanes in the lane group;

[0009] 3) Discretize the total flow sequence of key lanes into a character code sequence;

[0010] 4) Sampling and clustering the character coding sequences;

[0011] 5) Signal control time interval segmentation and adjustment based on clustering label sequence.

[0012] Furthermore, the step 1) specifically includes:

[0013] 11) Obtain intersection traffic flow data from the traffic flow history database as input data. The data is the lane-level traffic flow statistics with a 5-minute granularity for the whole day for 2 consecutive weeks or more. Perform data inspection and repair by lane on a daily basis. Specifically, for a certain lane, generate a negative value sequence with a 5-minute timestamp, traverse the negative value sequence, and determine one by one whether the timestamp of the negative value sequence contains the timestamp set of the input data. If it does, extract the traffic statistics of the corresponding timestamp from the input data to replace the negative number of the negative value sequence; if it does not contain, it means that the 5-minute traffic flow data of the corresponding timestamp of the input data is missing; check the result of the negative value sequence after processing. If all the negative values ​​of the negative value sequence are replaced by non-negative values, there is no data missing, and no repair is required to proceed to step 13); otherwise, proceed to step 12);

[0014] 12) When a single time segment is missing, the average traffic flow of the previous and next 5 minutes is used for replacement and repair; when multiple consecutive 5-minute data are missing, the adjacent non-missing data is used for replacement. When the first and last elements are encountered, the values ​​of the adjacent time segments at the other end are directly selected for replacement one by one, and then step 13) is entered; if the number of consecutive missing data exceeds the threshold, the group of data is discarded and re-selected, and the data is extracted again by returning to step 11);

[0015] 13) Calculate the average of the complete traffic flow data or the patched traffic flow data of the same lane on different dates within the same 5 minutes, and proceed to step 14);

[0016] 14) The traffic flow mean sequence obtained in step 13) is converted into a standard passenger car equivalent sequence, and the proportion and number of specific vehicle models are obtained from the traffic flow history database.

[0017] Furthermore, the step 2) specifically includes:

[0018] 21) According to the input channelization data, lane turns are classified into left turn, straight and right turn, among which the U-turn lane belongs to left turn;

[0019] 22) Calculate the maximum traffic flow in each lane group, sum the maximum traffic flow of each lane group in 5 minutes to obtain the total traffic flow of the key lanes of the intersection, and obtain the time series of the total traffic flow of the key lanes of the intersection in units of standard passenger car equivalent.

[0020] Furthermore, in step 3), the total flow sequence data of the key lanes is converted into symbol sequence data using a symbol aggregation approximation method, which specifically includes:

[0021] 31) Standardize the total flow sequence data of the key lanes to make them conform to the standard normal distribution to simplify the symbolization process and ensure the comparability between the total flow sequences of different key lanes; set the total flow time series of the key lanes of the intersection to be V = {v1, v2, …, v i}, where v i represents the total flow of the key lane in the i-th 5-minute period, i = 1, 2, ..., 288; after standardized calculation, the new sequence C = {c1, c2, ..., c i}, where c i represents the normalized value of the total flow of the key lane in the i-th 5-minute period, i = 1, 2, ..., 288; calculate the overall sequence mean μ and (overall) standard deviation δ, the expressions are as follows:

[0022]

[0023] Among them, n = 288;

[0024] The elements in sequence V are standardized one by one to obtain sequence C. The calculation method of each element in sequence C is:

[0025]

[0026] 32) Convert the total traffic sequence data of the key lane into a PAA segmented sequence representation, divide the total traffic sequence of the key lane into several subsequences of equal length, the length of the subsequence is 288 / w, denoted as s, where w represents the number of subsequences, and then calculate the mean of each subsequence to obtain a low-dimensional vector representation; specifically: a time series C of length n is converted into a reduced-dimensional sequence of length w represented by PPA in

[0027] 33) The continuous numerical value represented by PAA is converted into discrete symbols by the symbol aggregation approximation method, and each PAA coefficient is mapped to a symbol through the normal distribution table lookup method to form a symbol sequence; all symbols are connected in series to form a string to form a symbolic representation of the total flow sequence of the key lanes, that is, the character coding sequence.

[0028] Furthermore, in step 4), the character encoding sequence obtained in step 3) is clustered using the Kmeans algorithm, and the input parameters include the desired number of clusters k, where k is less than the dimension reduction sequence The length w of the function is to set a maximum number of iterations m. The specific steps are as follows:

[0029] 41) k and m are used as input parameters for Kmeans clustering. K samples are randomly sampled as the initial cluster centers, and then the sequence is calculated one by one. The distance between the elements in and the k cluster centers is taken, and the data point with the smallest distance is put into the corresponding cluster, and the mean of the k clusters is recalculated;

[0030] 42) Repeat step 41) until the mean of all clusters no longer changes or the maximum number of iterations m is reached, and the dimension reduction sequence The symbol of the corresponding element position is changed to the number of the cluster to which it belongs, and the cluster label sequence K = {k1, k2, ..., k w}, where k i Indicates the cluster number corresponding to the i-th subsequence, i = 1, 2, …, w.

[0031] Furthermore, the step 5) specifically includes:

[0032] 51) By controlling the minimum time period T min The minimum number of elements u in the calculation period is calculated by subsequence length s, and the expression is as follows:

[0033]

[0034] For the clustering label sequence K = {k1, k1, ..., k w} to perform initial cutting and determine whether the number of elements with the same cluster number and continuous index is greater than / equal to u. If so, the time interval corresponding to the start index and the end index is used as a control period, otherwise it needs to be merged into the adjacent control period; the control period split point sequence B = {b1, b2, ..., b r}, the splitting points do not include 0:00 and 23:55, and the corresponding number of control periods is r+2, where r is the number of elements in the splitting point sequence;

[0035] 52) Determine whether r+2 is equal to the desired number of control time periods t. If not, proceed to step 53); if r+2=t, the calculation ends;

[0036] 53) Adjust the number of input clusters k of the Kmeans algorithm, iterate from k+1 to the length value w of the dimension reduction sequence, repeat step 52), and take the time period with the smallest difference between the calculated result and the expected control time period t as the final result.

[0037] Beneficial effects of the present invention:

[0038] 1. The present invention first standardizes the data to avoid algorithm instability caused by artificially setting subjective thresholds, thereby affecting the accuracy of the time period division results; uses the symbolic aggregation approximation method (SAX) to reduce the dimension of the traffic time series, which is more efficient than the enumeration algorithm and can quickly extract features from the time series and complete the time period division task, which is more practical than the enumeration algorithm that is longer and more time-consuming; uses the Kmeans algorithm to randomly sample and cluster the reduced-dimensional sequence, and then performs time interval adjustment, merging and other operations on the obtained results, which is more in line with the actual engineering application of traffic signal control.

[0039] 2. The present invention starts from the principle of the timing signal control cycle calculation model and uses the total flow of the key lane group of the intersection compared with the commonly used total number of vehicles at the intersection, making the time period division more reasonable and more suitable for the timing signal control mode.

[0040] 3. The present invention uses traffic flow data with a 5-minute granularity. Compared with data with a 15-minute granularity, the impact on traffic flow when switching control schemes in adjacent time periods is lower, which can reduce the possibility of causing short-term congestion. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0042] In order to facilitate the understanding of those skilled in the art, the present invention is further described below in conjunction with embodiments and drawings. The contents mentioned in the implementation modes are not intended to limit the present invention.

[0043] Reference Figure 1 As shown, a method for automatically dividing multiple time periods of intersection timing signal control of the present invention comprises the following steps:

[0044] 1) Process the intersection traffic flow data; specifically including:

[0045] 11) Obtain intersection traffic flow data from the traffic flow history database as input data. The data is the lane-level traffic flow statistics with a 5-minute granularity for the whole day for 2 consecutive weeks or more. Perform data inspection and repair by lane on a daily basis. Specifically, for a certain lane, generate a negative value sequence with a 5-minute timestamp, traverse the negative value sequence, and determine one by one whether the timestamp of the negative value sequence contains the timestamp set of the input data. If it does, extract the traffic statistics of the corresponding timestamp from the input data to replace the negative number of the negative value sequence; if it does not contain, it means that the 5-minute traffic flow data of the corresponding timestamp of the input data is missing; check the result of the negative value sequence after processing. If all the negative values ​​of the negative value sequence are replaced by non-negative values, there is no data missing, and no repair is required to proceed to step 13); otherwise, proceed to step 12);

[0046] 12) When a single time segment is missing, the average traffic flow of the previous and next 5 minutes is used for replacement and repair; when multiple consecutive (not exceeding the threshold) 5-minute data are missing, the adjacent non-missing data is used for replacement. When the first and last elements are encountered, the values ​​of the adjacent time segments at the other end are directly selected for replacement one by one, and then go to step 13); if the number of consecutive missing data exceeds the threshold, the group of data is discarded and reselected, and the data is extracted again by returning to step 11);

[0047] 13) Calculate the average of the complete traffic flow data or the patched traffic flow data of the same lane on different dates within the same 5 minutes, and proceed to step 14);

[0048] 14) The traffic flow mean sequence obtained in step 13) is converted into a standard passenger car equivalent (Passenger Car Unit, pcu) sequence, and the proportion and number of specific vehicle models are obtained from the traffic flow history database.

[0049] 2) Calculate the traffic flow of the key lanes in the lane group; specifically include:

[0050] 21) According to the input channelization data, lane turns are classified into left turn, straight and right turn, among which the U-turn lane belongs to left turn;

[0051] 22) Calculate the maximum traffic flow in each lane group (i.e., the critical lane flow), sum up the maximum traffic flow of each lane group for 5 minutes to obtain the total flow of the critical lanes of the intersection, and obtain the time series of the total flow of the critical lanes of the intersection in units of standard passenger car equivalent (pcu).

[0052] 3) Discretize the total traffic flow sequence of key lanes into a character code sequence

[0053] Wherein, in step 3), the total flow sequence data of the key lanes is converted into symbolic sequence data using the symbolic aggregate approximation method (SAX), which specifically includes:

[0054] 31) Standardize the total flow sequence data of the key lanes to make them conform to the standard normal distribution to simplify the symbolization process and ensure the comparability between the total flow sequences of different key lanes; set the total flow time series of the key lanes of the intersection to be V = {v1, v2, …, v i}, where v i represents the total flow of the key lane in the i-th 5-minute period, i = 1, 2, ..., 288; after standardized calculation, the new sequence C = {c1, c2, ..., c i}, where c irepresents the normalized value of the total flow of the key lane in the i-th 5-minute period, i = 1, 2, ..., 288; calculate the overall sequence mean μ and (overall) standard deviation δ, the expressions are as follows:

[0055]

[0056] Among them, n = 288;

[0057] The elements in sequence V are standardized one by one to obtain sequence C. The calculation method of each element in sequence C is:

[0058]

[0059] 32) Convert the total traffic sequence data of the key lane into a PAA (Piecewise Aggregate Approximation) segmented sequence representation, divide the total traffic sequence of the key lane into several subsequences of equal length, the length of the subsequence is 288 / w, denoted as s, where w represents the number of subsequences, and then calculate the mean of each subsequence to obtain a low-dimensional vector representation; specifically: a time series C of length n is converted into a reduced-dimensional sequence of length w represented by PPA in

[0060] 33) The continuous numerical value represented by PAA is converted into discrete symbols by the symbol aggregation approximation method, and each PAA coefficient is mapped to a symbol through the normal distribution table lookup method to form a symbol sequence; all symbols are connected in series to form a string to form a symbolic representation of the total flow sequence of the key lanes, that is, the character coding sequence.

[0061] 4) Sampling and clustering the character coding sequences;

[0062] In step 4), the Kmeans algorithm is used to cluster the character encoding sequence obtained in step 3), and the input parameters include the desired number of clusters k, k is less than the dimensionality reduction sequence The length w of the function is to set a maximum number of iterations m. The specific steps are as follows:

[0063] 41) k and m are used as input parameters for Kmeans clustering. K samples are randomly sampled as the initial cluster centers, and then the sequence is calculated one by one. The distance between the elements in and the k cluster centers is taken, and the data point with the smallest distance is put into the corresponding cluster, and the mean of the k clusters is recalculated;

[0064] 42) Repeat step 41) until the mean of all clusters no longer changes or the maximum number of iterations m is reached, and the dimension reduction sequence The symbol of the corresponding element position is changed to the number of the cluster to which it belongs, and the cluster label sequence K = {k1, k2, ..., k w}, where k i Indicates the cluster number corresponding to the i-th subsequence, i = 1, 2, …, w.

[0065] 5) Segmentation and adjustment of signal control time interval based on clustering label sequence; specifically including:

[0066] 51) By controlling the minimum time period T min The minimum number of elements u in the calculation period is calculated by subsequence length s, and the expression is as follows:

[0067]

[0068] For the clustering label sequence K = {k1, k1, ..., k w} to perform initial cutting and determine whether the number of elements with the same cluster number and continuous index is greater than / equal to u. If so, the time interval corresponding to the start index and the end index is used as a control period, otherwise it needs to be merged into the adjacent control period; the control period split point sequence B = {b1, b2, ..., b r}, the splitting points do not include 0:00 and 23:55, and the corresponding number of control periods is r+2, where r is the number of elements in the splitting point sequence;

[0069] 52) Determine whether r+2 is equal to the desired number of control time periods t. If not, proceed to step 53); if r+2=t, the calculation ends;

[0070] 53) Adjust the number of input clusters k of the Kmeans algorithm, iterate from k+1 to the length value w of the dimension reduction sequence, repeat step 52), and take the time period with the smallest difference between the calculated result and the expected control time period t as the final result.

[0071] The present invention has many specific application paths. The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements can be made without departing from the principle of the present invention. These improvements should also be regarded as the protection scope of the present invention.

Claims

1. A method for automatically dividing multiple time periods of intersection time signal control, characterized in that: Here are the steps: 1) Processing intersection traffic flow data; 2) Calculate the traffic flow of key lanes in the lane group; 3) Discretize the total flow sequence of key lanes into a character code sequence; 4) Sampling and clustering the character coding sequences; 5) Signal control time interval segmentation and adjustment based on clustering label sequence.

2. The method for automatically dividing multiple time periods of intersection timing signal control according to claim 1 is characterized in that: The step 1) specifically includes: 11) Obtain intersection traffic flow data from the traffic flow history database as input data. The data is the lane-level traffic flow statistics with a 5-minute granularity for the whole day for 2 consecutive weeks or more. Perform data inspection and repair by lane on a daily basis. Specifically, for a certain lane, generate a negative value sequence with a 5-minute timestamp, traverse the negative value sequence, and determine one by one whether the timestamp of the negative value sequence contains the timestamp set of the input data. If it does, extract the traffic statistics of the corresponding timestamp from the input data to replace the negative number of the negative value sequence; if it does not contain, it means that the 5-minute traffic flow data of the corresponding timestamp of the input data is missing; check the result of the negative value sequence after processing. If all the negative values ​​of the negative value sequence are replaced by non-negative values, there is no data missing, and no repair is required to proceed to step 13); otherwise, proceed to step 12); 12) When a single time segment is missing, the average traffic flow of the previous and next 5 minutes is used for replacement and repair; when multiple consecutive 5-minute data are missing, the adjacent non-missing data is used for replacement. When the first and last elements are encountered, the values ​​of the adjacent time segments at the other end are directly selected for replacement one by one, and then step 13) is entered; if the number of consecutive missing data exceeds the threshold, the group of data is discarded and re-selected, and the data is extracted again by returning to step 11); 13) Calculate the average of the complete traffic flow data or the patched traffic flow data of the same lane on different dates within the same 5 minutes, and proceed to step 14); 14) The traffic flow mean sequence obtained in step 13) is converted into a standard passenger car equivalent sequence, and the proportion and number of specific vehicle models are obtained from the traffic flow history database.

3. The method for automatically dividing multiple time periods of intersection timing signal control according to claim 2 is characterized in that: The step 2) specifically includes: 21) According to the input channelization data, lane turns are classified into left turn, straight and right turn, among which the U-turn lane belongs to left turn; 22) Calculate the maximum traffic flow in each lane group, sum the maximum traffic flow of each lane group in 5 minutes to obtain the total traffic flow of the key lanes of the intersection, and obtain the time series of the total traffic flow of the key lanes of the intersection in units of standard passenger car equivalent.

4. The method for automatically dividing multiple time periods of intersection timing signal control according to claim 3 is characterized in that: In the step 3), the total flow sequence data of the key lanes is converted into symbol sequence data by using a symbol aggregation approximation method, which specifically includes: 31) Standardize the total flow sequence data of the key lanes to make them conform to the standard normal distribution to simplify the symbolization process and ensure the comparability between the total flow sequences of different key lanes; set the total flow time series of the key lanes of the intersection to be V = {v1, v2, …, v i }, where v i represents the total flow of the key lane in the i-th 5-minute period, i = 1, 2, ..., 288; after standardized calculation, the new sequence C = {c1, c2, ..., c i }, where c i represents the normalized value of the total flow of the key lane in the i-th 5-minute period, i = 1, 2, ..., 288; calculate the overall sequence mean μ and standard deviation δ, the expression is as follows: Among them, n = 288; The elements in sequence V are standardized one by one to obtain sequence C. The calculation method of each element in sequence C is: 32) Convert the total traffic sequence data of the key lane into a PAA segmented sequence representation, divide the total traffic sequence of the key lane into several subsequences of equal length, the length of the subsequence is 288 / w, denoted as s, where w represents the number of subsequences, and then calculate the mean of each subsequence to obtain a low-dimensional vector representation; specifically: a time series C of length n is converted into a reduced-dimensional sequence of length w represented by PPA in j = s × (i-1) + 1; 33) The continuous numerical value represented by PAA is converted into discrete symbols by the symbol aggregation approximation method, and each PAA coefficient is mapped to a symbol through the normal distribution table lookup method to form a symbol sequence; all symbols are connected in series to form a string to form a symbolic representation of the total flow sequence of the key lanes, that is, the character coding sequence.

5. The method for automatically dividing multiple time periods of intersection time signal control according to claim 4 is characterized in that: In step 4), the Kmeans algorithm is used to cluster the character encoding sequence obtained in step 3), and the input parameters include the expected number of clusters k, k is less than the dimension reduction sequence The length w of the function is to set a maximum number of iterations m. The specific steps are as follows: 41) k and m are used as input parameters for Kmeans clustering. K samples are randomly sampled as the initial cluster centers, and then the sequence is calculated one by one. The distance between the elements in and the k cluster centers is taken, and the data point with the smallest distance is put into the corresponding cluster, and the mean of the k clusters is recalculated; 42) Repeat step 41) until the mean of all clusters no longer changes or the maximum number of iterations m is reached, and the dimension reduction sequence The symbol of the corresponding element position is changed to the number of the cluster to which it belongs, and the cluster label sequence K = {k1, k2, ..., k w }, where k i Indicates the cluster number corresponding to the i-th subsequence, i = 1, 2, …, w.

6. The method for automatically dividing multiple time periods of intersection time signal control according to claim 5 is characterized in that: The step 5) specifically includes: 51) By controlling the minimum time period T min The minimum number of elements u in the calculation period is calculated by subsequence length s, and the expression is as follows: For the clustering label sequence K = {k1, k1, ..., k w } to perform initial cutting and determine whether the number of elements with the same cluster number and continuous index is greater than / equal to u. If so, the time interval corresponding to the start index and the end index is used as a control period, otherwise it needs to be merged into the adjacent control period; the control period split point sequence B = {b1, b2, ..., b r }, the splitting points do not include 0:00 and 23:55, and the corresponding number of control periods is r+2, where r is the number of elements in the splitting point sequence; 52) Determine whether r+2 is equal to the desired number of control time periods t. If not, proceed to step 53); if r+2=t, the calculation ends; 53) Adjust the number of input clusters k of the Kmeans algorithm, iterate from k+1 to the length value w of the dimension reduction sequence, repeat step 52), and take the time period with the smallest difference between the calculated result and the expected control time period t as the final result.

Images