Tunnel traffic flow big data mining method and system combined with time series

Abstract

The application provides a tunnel traffic flow big data mining method and system combined with time series, relates to the technical field of traffic flow data analysis, and first acquires a time series data set corresponding to tunnel traffic flow big data, which contains a plurality of traffic flow time series segments collected under continuous time dimensions; then constructs a time series correlation chain based on time series adjacent relations, fills in features of adjacent segment time gaps, and obtains a time series correlation chain set; then performs time series evolution feature extraction processing on each time series correlation chain to obtain a time series evolution feature set; calls a pre-trained time series pattern mining model to perform abnormal evolution pattern matching processing on the time series evolution feature set, generates an abnormal pattern mining result, and finally generates a traffic flow regulation suggestion set based on the abnormal pattern mining result, adjusts the regulation content in combination with influence range analysis, makes the regulation suggestion more targeted and operable, and can effectively improve the efficiency and accuracy of tunnel traffic flow regulation.

Description

Technical Field

[0001] This invention relates to the field of traffic flow data analysis technology, and more specifically, to a method and system for mining big data on tunnel traffic flow that combines time series data. Background Technology

[0002] In tunnel traffic management, the accurate analysis and mining of traffic flow data is directly related to the safety and efficiency of tunnel passage. Currently, most methods for processing tunnel traffic flow data focus on the analysis of static data or only consider simple time period divisions, while ignoring the continuous change of traffic flow over time.

[0003] Current technologies typically divide the day into several fixed time periods and collect traffic flow data for each period separately. However, this approach fails to capture the dynamic changes in traffic flow over continuous time. For example, during morning and evening rush hours, traffic flow fluctuates rapidly, and complex correlations may exist between different time periods, which traditional methods struggle to reveal. Furthermore, when traffic anomalies occur, such as accidents or congestion, their evolution over time cannot be effectively captured and analyzed. This results in existing data analysis methods failing to provide comprehensive and accurate information when facing complex tunnel traffic flow changes, hindering effective decision-making for traffic flow control and ultimately impacting tunnel traffic management. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for mining tunnel traffic flow big data by combining time series data, the method comprising:

[0005] Obtain a time-series data set corresponding to the big data of tunnel traffic flow. The time-series data set contains multiple traffic flow time-series segments collected in a continuous time dimension. Each traffic flow time-series segment records the traffic flow change information of a specific monitoring area in the tunnel within a corresponding time interval.

[0006] A temporal association chain is constructed based on the temporal adjacency relationship between the traffic flow time series segments. During the construction process, feature filling processing is performed on the time gaps of adjacent segments to obtain a set of temporal association chains. The temporal association chain is used to characterize the association relationship of traffic flow change information in different time intervals, and each temporal association chain contains a node sequence and a correlation strength parameter between nodes.

[0007] For each of the time-series association chains, a time-series evolution feature extraction process is performed. Features are extracted from the time dimension and the association dimension respectively and then fused together to obtain a time-series evolution feature set corresponding to each of the time-series association chains. The time-series evolution feature set includes the change trend features and association strength change features of the time-series association chains in the time dimension.

[0008] A pre-trained temporal pattern mining model is invoked to perform abnormal evolution pattern matching processing on each of the temporal evolution feature sets, generating abnormal pattern mining results. The abnormal pattern mining results include the matched abnormal evolution pattern type and the identification information of the corresponding temporal association chain.

[0009] Based on the results of the abnormal pattern mining, a set of traffic flow control suggestions is generated. The control content is adjusted in combination with the impact range analysis corresponding to the abnormal pattern. Each suggestion in the set of traffic flow control suggestions corresponds to one of the abnormal evolution pattern types, and each suggestion contains control direction information and control implementation step information for the corresponding abnormal evolution pattern.

[0010] In another aspect, embodiments of the present invention also provide a tunnel traffic flow big data mining system that combines time series, including a processor and a machine-readable storage medium. The machine-readable storage medium is connected to the processor. The machine-readable storage medium is used to store programs, instructions or code. The processor is used to execute the programs, instructions or code in the machine-readable storage medium to implement the above-described method.

[0011] Based on the above, by acquiring a time-series data set containing multiple traffic flow time-series segments across a continuous time dimension, the traffic flow changes in a specific monitoring area within the tunnel were recorded across different time intervals. A time-series correlation chain was constructed based on temporal adjacency relationships, and feature imputation was performed on the time gaps between adjacent segments, effectively solving the gap problem that may exist in time-series data. The constructed time-series correlation chain set can accurately represent the correlation relationships of traffic flow changes across different time intervals. For each time-series correlation chain, time-series evolution feature extraction was performed, extracting features from both the time and correlation dimensions and then interactively fusing them. The resulting time-series evolution feature set comprehensively reflects the changing trend and correlation strength characteristics of the time-series correlation chains across the time dimension. A pre-trained time-series pattern mining model was used for anomaly evolution pattern matching, which can quickly and accurately generate anomaly pattern mining results, providing an effective means for timely detection of traffic flow anomalies. Finally, a set of traffic flow control suggestions was generated based on the anomaly pattern mining results, and the control content was adjusted in conjunction with the impact range analysis, making the control suggestions more targeted and operable, effectively improving the efficiency and accuracy of tunnel traffic flow control. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the execution flow of the tunnel traffic flow big data mining method combining time series data provided in an embodiment of the present invention.

[0013] Figure 2 This is a schematic diagram of exemplary hardware and software components of the tunnel traffic flow big data mining system that combines time series data, provided in an embodiment of the present invention. Detailed Implementation

[0014] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a method for mining tunnel traffic flow big data by combining time series data, provided in one embodiment of the present invention. The following is a detailed description of this method for mining tunnel traffic flow big data by combining time series data.

[0015] Step S110: Obtain the time series data set corresponding to the big data of tunnel traffic flow. The time series data set contains multiple traffic flow time series segments collected in a continuous time dimension. Each traffic flow time series segment records the traffic flow change information of a specific monitoring area in the tunnel within the corresponding time interval.

[0016] In this embodiment, taking traffic flow monitoring of a city's tunnel complex as an example, the tunnel management system can collect traffic flow data from various monitoring areas at preset time intervals (e.g., every 5 minutes). The monitoring areas can be different sections within the tunnel, such as the tunnel entrance section, the middle driving section, and the exit section. Each traffic flow time series segment corresponds to the traffic flow change of a monitoring area within a time interval. For example, for monitoring area A at the tunnel entrance section, from 8:00 AM to 8:05 AM, the traffic flow gradually increases from an initial 0 to a certain value; the traffic flow change information during this process constitutes a traffic flow time series segment. Therefore, all eligible traffic flow data can be retrieved from the database and organized into a time series data set. Each element in this time series data set is a traffic flow time series segment, containing information such as the monitoring area identifier, time interval, and traffic flow change sequence.

[0017] Step S120: Construct a temporal association chain based on the temporal adjacency relationship between the traffic flow time series segments. During the construction process, feature filling processing is performed on the time gaps of adjacent segments to obtain a set of temporal association chains. The temporal association chains are used to characterize the association relationship of traffic flow change information in different time intervals, and each temporal association chain contains a node sequence and a correlation strength parameter between nodes.

[0018] Step S121: Extract time interval information of all traffic flow time series segments from the time series data set, wherein the time interval information includes the start time and end time of each traffic flow time series segment.

[0019] In this embodiment, for the time-series data set of the aforementioned tunnel group, each traffic flow time-series segment is traversed, and its start and end times are extracted. For example, one time-series segment in monitoring area A starts at 8:00 and ends at 8:05; another time-series segment starts at 8:05 and ends at 8:10; and another time-series segment in monitoring area B starts at 8:03 and ends at 8:08, and so on. This time interval information is organized into a list for convenient subsequent analysis.

[0020] Step S122: Determine the temporal adjacency relationship between any two traffic flow time segments based on the time interval information. The temporal adjacency relationship is used to indicate whether there is a continuous connection between the time intervals of the two traffic flow time segments. If there is an interval between the end time of the previous traffic flow time segment and the start time of the next traffic flow time segment, it is determined that there is a temporal gap between the two traffic flow time segments.

[0021] In this embodiment, the extracted time interval information is compared pairwise. For example, in monitoring area A, the 8:00-8:05 segment and the 8:05-8:10 segment have a seamless transition between the end time 8:05 and the start time 8:05 of the former segment, with no time gap. However, in monitoring area A, the 8:05-8:10 segment and the 8:03-8:08 segment in monitoring area B have a seamless transition between the start time 8:05 and the end time 8:08 of the former segment, and the end time 8:10 of the former segment is also not a seamless transition between the start time 8:03 and the start time 8:03 of the latter segment. There is a direct connection. Between monitoring area A (8:05-8:10) and monitoring area B (8:03-8:08), if we look at the next possible segment of monitoring area A, assuming there is another segment of monitoring area C with a start time of 8:11, then between monitoring area A (8:05-8:10) and monitoring area C (8:11-8:16), there is a 1-minute interval between the previous end time 8:10 and the next start time 8:11. Therefore, it is determined that there is a time gap between these two segments.

[0022] Step S123: For two traffic flow time segments with a time gap, feature interpolation processing is performed using traffic flow change information of adjacent segments to generate a transition feature sequence to fill the time gap. The time length of the transition feature sequence is consistent with the time length of the time gap.

[0023] In this embodiment, for the 8:05-8:10 segment of monitoring area A and the 8:11-8:16 segment of monitoring area C with time gaps, the time gap is 1 minute (i.e., 60 seconds; assuming the time unit is seconds, the time length is 60). First, traffic flow change information for the 8:05-8:10 segment of monitoring area A is extracted, for example, the flow rate is a1 at 8:05, a2 at 8:06, ..., a6 at 8:10; for the 8:11-8:16 segment of monitoring area C, the flow rate is c1 at 8:11, c2 at 8:12, ..., c6 at 8:16. Using an interpolation method, a transition feature sequence is generated based on the flow rate values ​​of a6 and c1 and the length of the time gap. For example, assuming the time interval is 1 minute, i.e., one time step (here, we assume the time step is 1 minute), then the time length of the transition feature sequence is 1, and its flow value can be obtained by linear interpolation of a6 and c1. For example, the flow value of the transition feature sequence in the time interval of 8:10-8:11 is (a6+c1) / 2 (of course, other interpolation methods can also be used, such as polynomial interpolation, etc., depending on the actual needs). Thus, the transition feature sequence is obtained, and its time length is consistent with the time length of the time interval.

[0024] Step S124: For two traffic flow time series segments that have the temporal adjacency relationship, calculate the similarity of traffic flow change information in the two traffic flow time series segments, and use the similarity as the temporal correlation strength between the two traffic flow time series segments.

[0025] Step S1241: Extract traffic flow change information from two traffic flow time segments that have the temporal adjacency relationship. If a transition feature sequence exists, merge the transition feature sequence with the traffic flow change information of the next traffic flow time segment to obtain the first traffic flow change sequence and the subsequent traffic flow change sequence, respectively.

[0026] In this embodiment, taking the 8:00-8:05 and 8:05-8:10 segments of monitoring area A as examples, these two segments are temporally adjacent and do not have a transitional feature sequence. Traffic flow change information for the 8:00-8:05 segment of monitoring area A is extracted. Assuming the time step of this segment is 1 minute, the traffic flow change sequence is [q1, q2, q3, q4, q5] (q1 is the traffic flow at 8:00, q2 is the traffic flow at 8:01, and so on); the traffic flow change sequence for the 8:05-8:10 segment is [q6, q7, q8, q9, q10]. The former is taken as the first traffic flow change sequence, and the latter as the subsequent traffic flow change sequences. For example, the 8:05-8:10 segment of monitoring area A and the 8:11-8:16 segment of monitoring area C generate a transition feature sequence [q11] after step S123 (assuming a time step of 1 minute). Then, the traffic flow change sequence [q6, q7, q8, q9, q10] of the 8:05-8:10 segment of monitoring area A is taken as the first traffic flow change sequence. The transition feature sequence [q11] is merged with the traffic flow change sequence [c1, c2, c3, c4, c5] of the 8:11-8:16 segment of monitoring area C to obtain the subsequent traffic flow change sequence [q11, c1, c2, c3, c4, c5].

[0027] Step S1242: Perform time dimension length unification processing on the first traffic flow change sequence and the subsequent traffic flow change sequence. For the first traffic flow change sequence and the subsequent traffic flow change sequence after unification, calculate the traffic flow value difference at the corresponding position step by step to obtain a difference sequence composed of multiple time step difference values.

[0028] In this embodiment, for the 8:00-8:05 segment (sequence length 5) and the 8:05-8:10 segment (sequence length 5) in the aforementioned monitoring area A, since both have the same length, no length unification processing is required. The difference is calculated step-by-step. Assuming padding is used, the length of the first sequence is extended to 6, for example, by adding a value identical to the last element at the end, or by interpolation. Then, the difference is calculated step-by-step. For example, if the first sequence is extended to [q1, q2, q3, q4, q5, q5], and the subsequent sequences are [q11, c1, c2, c3, c4, c5], then each element of the difference sequence is q11-q1, c1-q2, c2-q3, c3-q4, c4-q5, c5-q5. This yields the difference sequence.

[0029] Step S1243: Calculate the overall fluctuation degree of the difference sequence based on the differences in traffic flow values ​​at all time steps, and characterize the overall fluctuation degree by statistically analyzing the distribution concentration of all difference values ​​in the difference sequence.

[0030] In this embodiment, the distribution of all difference values ​​in the obtained difference sequence is statistically analyzed. The variance or standard deviation of the difference values ​​can be calculated. A smaller variance indicates a more concentrated set of differences and lower overall volatility; a larger variance indicates higher volatility. For example, in a difference sequence [d1, d2, d3, d4, d5, d6], the sum of the squares of the differences between each difference value and the mean is calculated, and then divided by the number of difference values ​​(here, 6) to obtain the variance. The magnitude of the variance characterizes the overall volatility.

[0031] Step S1244: Determine the basic value of similarity based on the overall fluctuation level. The lower the overall fluctuation level, the higher the basic value of similarity.

[0032] In this embodiment, a mapping relationship between the degree of fluctuation and the basic value of similarity is preset. For example, when the variance is 0, the basic value of similarity is 1; when the variance is v1, the basic value of similarity is 0.8; when the variance is v2, the basic value of similarity is 0.5, etc. (here, v1 and v2 are thresholds set according to actual conditions). Based on the variance of the calculated difference sequence, the corresponding basic value of similarity is found. For example, if the calculated variance is v0, if v0 is less than v1, the basic value of similarity is 1; if v1 ≤ v0 < v2, the basic value of similarity is 0.8, and so on.

[0033] Step S1245: Based on the location correlation of the monitoring areas corresponding to the two traffic flow time series segments, the base value of the similarity is corrected. If the monitoring areas corresponding to the two traffic flow time series segments are adjacent, the base value of the similarity is increased. If the monitoring areas corresponding to the two traffic flow time series segments are not adjacent, the base value of the similarity remains unchanged.

[0034] In this embodiment, the positional correlation of each monitoring area within the tunnel is predefined. For example, the positional correlation of adjacent monitoring areas (such as the monitoring areas at the tunnel entrance and the middle driving section) is 1, the positional correlation of areas separated by one monitoring area is 0.5, and the positional correlation of areas separated by two or more monitoring areas is 0. The positional correlation of the monitoring areas corresponding to the two traffic flow time-series segments is then examined. For example, if monitoring area A (entrance section) and monitoring area B (middle driving section) are adjacent and have a positional correlation of 1, then the similarity baseline value is increased by a certain percentage, such as 0.1. If monitoring area A and monitoring area C (exit section) are not adjacent and have a positional correlation of 0, then the similarity baseline value remains unchanged.

[0035] Step S1246: Use the corrected similarity as the temporal correlation strength between the two traffic flow time series segments.

[0036] In this embodiment, the basic similarity value obtained in step S1244 is corrected in step S1245, and the resulting value is the temporal correlation strength between two traffic flow time series segments. For example, if the basic similarity value is 0.8 and the location correlation is 1, and the value is increased by 0.1 after correction, the temporal correlation strength is 0.9.

[0037] Step S125: Based on the distribution of correlation strength of similar segments in historical traffic flow data, dynamically calibrate the currently calculated temporal correlation strength to eliminate the influence of outliers on the correlation strength.

[0038] In this embodiment, correlation strength data for segments similar to the current two traffic flow time-series segments (similar means the monitoring area type and time interval type are the same, such as both being entry segment monitoring area segments during the morning rush hour) are retrieved from the historical traffic flow database. The distribution of these correlation strengths is statistically analyzed, and their mean and standard deviation are calculated. Then, the currently calculated time-series correlation strength is compared with the mean and standard deviation. If the difference between the current correlation strength and the mean exceeds three times the standard deviation, it is considered an outlier and is calibrated. The calibration method can be to adjust it to the mean plus or minus three times the standard deviation (depending on the direction of deviation), or to use other statistical methods for correction to eliminate the influence of outliers.

[0039] Step S126: Based on the preset association strength threshold, select traffic flow time series segments whose time series association strength meets the threshold requirement to obtain a set of effective association pairs. Each association pair in the set of effective association pairs contains two traffic flow time series segments and a calibrated time series association strength.

[0040] In this embodiment, a preset association strength threshold, such as 0.6, is used. All calculated and calibrated temporal association strengths are iterated through, and traffic flow time-series segments with a temporal association strength greater than or equal to 0.6 are selected to form a valid association pair set. For example, the temporal association strength of the 8:00-8:05 and 8:05-8:10 segments in monitoring area A is 0.9, which is greater than 0.6, and is therefore selected; the temporal association strength of two segments in monitoring area B is 0.5, which is less than 0.6, and is not selected. Each element in the valid association pair set contains the identifiers of the two traffic flow time-series segments and the calibrated temporal association strength.

[0041] Step S127: Using one traffic flow time sequence segment in each of the effective association pairs as the starting node and the other traffic flow time sequence segment as the subsequent node, connect multiple effective association pairs sequentially according to the order of time intervals. If nodes are repeated during the connection process, retain the connection path with higher association strength to construct an initial time sequence association chain.

[0042] In this embodiment, for each association pair in the set of valid association pairs, their temporal order is determined, with the earlier segment as the starting node and the later segment as the subsequent node. These association pairs are then connected sequentially to form a chain structure. For example, valid association pair 1: segment A (8:00-8:05) and segment B (8:05-8:10), association strength 0.9; valid association pair 2: segment B (8:05-8:10) and segment C (8:10-8:15), association strength 0.8; valid association pair 3: segment A (8:00-8:05) and segment D (8:03-8:08), association strength 0.7. During the connection process, segment A connects to segment B, and segment B connects to segment C, forming a chain. However, if segment A connects to segment D, since segment A has already served as the starting node in the previous chain, and the time interval of segment D (8:03-8:08) overlaps with that of segment A (8:00-8:05), there is a node duplication (segment A duplication). At this point, comparing the association strength of the two paths, the sum of the association strength of segment A-segment B-segment C (0.9+0.8=1.7) is greater than the association strength of segment A-segment D (0.7). Therefore, the path of segment A-segment B-segment C is retained, and the initial temporal association chain is constructed.

[0043] Step S128: Perform redundant node removal processing on the initial temporal association chain, delete nodes in the initial temporal association chain whose traffic flow change information repetition with adjacent nodes exceeds a preset repetition threshold, and update the temporal association strength between nodes before and after node removal.

[0044] For example, step S1281: Starting from the starting node of the initial temporal association chain, each node is selected as the current node in sequence, and the subsequent node adjacent to the current node is determined. The subsequent node is the node that immediately follows the current node in time in the temporal association chain.

[0045] In this embodiment, for the initial temporal association chain constructed above, for example, the chain is segment A (8:00-8:05) - segment B (8:05-8:10) - segment C (8:10-8:15) - segment D (8:15-8:20), starting from the starting node segment A, the current node is segment A, and the next node is segment B; then the current node is segment B, and the next node is segment C; the current node is segment C, and the next node is segment D; the current node is segment D, and there is no next node (assuming the chain ends here).

[0046] Step S1282: Extract the traffic flow change information of the current node and the traffic flow change information of the subsequent nodes. The traffic flow change information includes the traffic flow value and traffic flow change trend at each time step.

[0047] In this embodiment, taking the current node as segment B (8:05-8:10) and the subsequent node as segment C (8:10-8:15) as an example, the traffic flow change information of segment B is extracted, including the traffic flow value at 8:05, the traffic flow value at 8:06, ..., the traffic flow value at 8:10, and the traffic flow change trend at each time step (e.g., whether it is increasing, decreasing, or remaining stable); similarly, the traffic flow change information of segment C is extracted, including the traffic flow value at 8:10, the traffic flow value at 8:11, ..., the traffic flow value at 8:15, and the corresponding change trend.

[0048] Step S1283: Compare the two types of traffic flow change information step by step, and count the proportion of time steps with the same traffic flow value to the total number of time steps. At the same time, count the proportion of time steps with the same traffic flow change trend to the total number of time steps.

[0049] In this embodiment, it is assumed that segment B has 5 time steps (8:05-8:10, one time step per minute), and segment C also has 5 time steps (8:10-8:15, one time step per minute). A time-step comparison is performed, for example, the traffic flow values ​​at 8:05 (segment B) and 8:10 (segment C), the traffic flow values ​​at 8:06 (segment B) and 8:11 (segment C), and so on. The number of time steps with the same traffic flow value is counted; assuming there are 3, then the proportion of time steps with the same traffic flow value is 3 / 5. Similarly, the number of time steps with the same traffic flow trend is counted; assuming there are 4, then the proportion of time steps with the same traffic flow trend is 4 / 5.

[0050] Step S1284: Calculate the comprehensive value of repetition. Add the proportion of the same traffic flow value and the proportion of the same traffic flow change trend according to the preset weight to obtain the repetition of traffic flow change information between the current node and the subsequent node.

[0051] In this embodiment, the weight of the proportion of the same traffic flow value is preset to 0.6, and the weight of the proportion of the same traffic flow change trend is 0.4. Then the comprehensive value of repetition = 0.6×(3 / 5)+0.4×(4 / 5)=0.6×0.6+0.4×0.8=0.36+0.32=0.68.

[0052] Step S1285: Compare the calculated repeatability with a preset repeatability threshold. If the repeatability exceeds the preset repeatability threshold, mark the subsequent node as a redundant node.

[0053] In this embodiment, the preset repetition threshold is 0.7. Since the calculated repetition is 0.68, which is less than 0.7, the subsequent node segment C is not marked as a redundant node. Alternatively, if the calculated repetition is 0.75, which exceeds 0.7, then the subsequent node is marked as a redundant node.

[0054] Step S1286: If the repetition does not exceed the preset repetition threshold, then retain the subsequent node and continue to judge the next node.

[0055] In this embodiment, since the repetition degree of fragment B and fragment C is 0.68, which does not exceed the threshold of 0.7, fragment C is retained, and the next node fragment C is judged to determine its subsequent node fragment D. The above steps S1281-S1285 are repeated.

[0056] Step S1287: Traverse all nodes in the initial temporal association chain, mark all redundant nodes, and then delete all marked redundant nodes.

[0057] In this embodiment, all nodes in the initial temporal association chain are traversed, and the repetition rate of each node and its successor nodes is calculated and marked. If, in another chain, a successor node is marked as a redundant node, then after the traversal is complete, all marked redundant nodes are removed from the initial temporal association chain. For example, if the initial temporal association chain is segment E—segment F—segment G—segment H, where the repetition rate of segment F and segment G exceeds a threshold, and segment G is marked as a redundant node, then segment G is deleted, and the chain becomes segment E—segment F—segment H.

[0058] Step S1288: Determine the adjacency relationship of the remaining nodes after deleting the redundant node. If the deleted redundant node is located between the previous and subsequent retained nodes, then the previous and subsequent retained nodes become new adjacent nodes.

[0059] In this embodiment, for the chain segment E after deleting segment G—segment F—segment H, since segment G was deleted, the original successor node of segment F was segment G, and the successor node of segment G was segment H. Now that segment G has been deleted, the preceding retained node segment F and the following retained node segment H become new adjacent nodes.

[0060] Step S1289: Calculate the temporal association strength between the preceding and subsequent retained nodes. Combine the association strength between the preceding retained node and the original redundant node, the association strength between the original redundant node and the subsequent retained node, and the similarity of traffic flow change information between the preceding and subsequent retained nodes to obtain a new temporal association strength.

[0061] In this embodiment, it is assumed that the association strength between the preceding retained node segment F and the original redundant node segment G is s1, the association strength between the original redundant node segment G and the subsequent retained node segment H is s2, and the similarity of traffic flow change information between the preceding retained node segment F and the subsequent retained node segment H is s3. The new temporal association strength can be calculated using a certain formula, such as s_new=(s1+s2)×s3 / 2 (this formula can be adjusted according to the actual situation, such as considering weights). For example, if s1=0.8, s2=0.7, and s3=0.6, then s_new=(0.8+0.7)×0.6 / 2=1.5×0.6 / 2=0.45.

[0062] Step S12810: Reconnect the remaining nodes after deleting redundant nodes according to the original time interval sequence, and add new temporal association strength to form an optimized temporal association chain.

[0063] In this embodiment, the remaining nodes after deleting redundant nodes are rearranged according to the order of time intervals, such as segment E - segment F - segment H. Then, new temporal association strengths are added between adjacent nodes. For example, the association strength between segment E and segment F remains unchanged (assuming they were not deleted), and the association strength between segment F and segment H is the calculated s_new = 0.45, thus forming the optimized temporal association chain.

[0064] Step S129: Integrate all the optimized temporal association chains, record the number of nodes, total time span and average association strength of each temporal association chain, and form the temporal association chain set.

[0065] In this embodiment, all optimized temporal correlation chains are collected, and statistics are performed on each chain. For example, an optimized temporal correlation chain contains 3 nodes, with time intervals from 8:00-8:05, 8:05-8:10, and 8:10-8:15, for a total time span of 15 minutes (from 8:00 to 8:15). The correlation strengths between the nodes are 0.9 and 0.8, respectively, and the average correlation strength is (0.9+0.8) / 2=0.85. This information is recorded and, together with other optimized temporal correlation chains, forms a temporal correlation chain set.

[0066] Step S130: Perform temporal evolution feature extraction processing on each of the temporal association chains, extract features from the time dimension and the association dimension respectively, and perform interactive fusion to obtain a temporal evolution feature set corresponding to each of the temporal association chains. The temporal evolution feature set includes the change trend features and association strength change features of the temporal association chains in the time dimension.

[0067] Step S131: Divide each of the time-series association chains into multiple consecutive evolutionary stages according to the chronological order of the time intervals. Each evolutionary stage contains at least two adjacent traffic flow time-series segment nodes, and the time span of each evolutionary stage remains consistent.

[0068] In this embodiment, each temporal association chain in the aforementioned set of temporal association chains is divided according to the chronological order of its time intervals. For example, a temporal association chain might have time intervals from 8:00-8:05, 8:05-8:10, 8:10-8:15, and 8:15-8:20, with a total time span of 20 minutes. This is divided into two evolutionary stages, each with a time span of 10 minutes. The first evolutionary stage contains the first two nodes (8:00-8:05 and 8:05-8:10), and the second evolutionary stage contains the last two nodes (8:10-8:15 and 8:15-8:20). This ensures that the time span of each evolutionary stage is consistent, and that each stage contains at least two adjacent nodes.

[0069] Step S132: Extract traffic flow change information of all traffic flow time sequence nodes in each evolution stage, calculate the average rate of change of traffic flow change information in each evolution stage, the average rate of change is obtained by dividing the total change of traffic flow value in the stage by the stage time span, and use the average rate of change as the basic feature in the stage.

[0070] In this embodiment, taking the first evolutionary stage (8:00-8:05, 8:05-8:10) as an example, traffic flow change information for these two nodes is extracted. Assume the traffic flow change sequence from 8:00-8:05 is [q1, q2, q3, q4, q5], and the traffic flow change sequence from 8:05-8:10 is [q6, q7, q8, q9, q10]. The total change in traffic flow values ​​within each stage is (q5-q1) + (q10-q6), and the stage time span is 10 minutes (from 8:00 to 8:10). The average rate of change = total change / 10, and this average rate of change is used as the basic characteristic within this evolutionary stage.

[0071] Step S133: Analyze the transition of traffic flow change information between two adjacent evolution stages, calculate the difference between the average rate of change of the previous evolution stage and the average rate of change of the next evolution stage, and count the number of fluctuations in traffic flow values ​​during the transition process. The difference and the number of fluctuations are respectively used as independent components of the transition characteristics between stages.

[0072] In this embodiment, taking the aforementioned two evolutionary stages as examples, the average rate of change in the first evolutionary stage is v1, and the average rate of change in the second evolutionary stage is v2, with a difference of v2-v1. Then, the number of traffic flow fluctuations during the transition process (i.e., the traffic flow change from the last time step of the first evolutionary stage to the first time step of the second evolutionary stage) is analyzed. For example, the traffic flow value at the last time step of the first evolutionary stage (8:10) is q10, the traffic flow value at the first time step of the second evolutionary stage (8:11) is q11, then at 8:12 it is q12, and so on. The number of times the traffic flow value increases and then decreases, or decreases and then increases, during the transition process (assuming the transition time range is 8:10-8:15, i.e., the time interval of the second evolutionary stage) is counted; this is the number of fluctuations. The difference v2-v1 and the number of fluctuations are considered as two independent components of the inter-stage transition characteristics.

[0073] Step S134: Based on the basic features within each evolutionary stage and the transitional features between each evolutionary stage in each of the time-series association chains, construct the evolutionary trend curve of the time-series association chain, and plot the evolutionary trend curve with the midpoint of each evolutionary stage as the horizontal axis and the basic features within each evolutionary stage as the vertical axis.

[0074] Step S1341: Extract the start time and end time from the time interval information of each evolution stage, calculate the midpoint between the start time and the end time, and use the midpoint as the midpoint of the evolution stage.

[0075] In this embodiment, taking the time interval of the first evolutionary stage from 8:00 to 8:10 as an example, the start time is 8:00, the end time is 8:10, and the midpoint is 8:05. 8:05 is taken as the midpoint of this evolutionary stage. The time interval of the second evolutionary stage is from 8:10 to 8:20, the start time is 8:10, the end time is 8:20, and the midpoint is 8:15. 8:15 is taken as the midpoint of this evolutionary stage.

[0076] Step S1342: Using the midpoint of each evolutionary stage as the horizontal axis coordinate value and the basic features within each evolutionary stage as the vertical axis coordinate value, mark the coordinate points corresponding to each evolutionary stage in a two-dimensional coordinate system. Each coordinate point contains both horizontal and vertical axis coordinate values.

[0077] In this embodiment, the midpoint of the first evolutionary stage is 8:05, and the basic characteristic (average rate of change) within the stage is v1, so the coordinate point is (8:05, v1); the midpoint of the second evolutionary stage is 8:15, and the basic characteristic within the stage is v2, so the coordinate point is (8:15, v2). These two coordinate points are marked in a two-dimensional coordinate system.

[0078] Step S1343: Check if there is any abnormal deviation of the marked coordinate points. If the difference between the ordinate value of any coordinate point and the ordinate values ​​of the two adjacent coordinate points exceeds the preset deviation threshold, then the coordinate point is determined to be an abnormal coordinate point.

[0079] In this embodiment, it is assumed that there is a third evolutionary stage, with the intermediate time point being 8:25. The basic feature within this stage is v3, and the coordinate point is (8:25, v3). Now, each coordinate point is examined. For example, for the coordinate point (8:15, v2), the difference between its ordinate and the values ​​of its two adjacent coordinate points (8:05, v1) and (8:25, v3) is calculated, namely |v2-v1| and |v2-v3|. If both of these differences exceed a preset deviation threshold (for example, the preset deviation threshold is Δv), then (8:15, v2) is determined to be an abnormal coordinate point.

[0080] Step S1344: Correct the abnormal coordinate points by replacing the vertical coordinate value of the abnormal coordinate point with the average value of the vertical coordinate values ​​of two adjacent normal coordinate points, while keeping the horizontal coordinate value of the abnormal coordinate point unchanged, thus obtaining the corrected set of coordinate points.

[0081] In this embodiment, assuming the coordinate point (8:15, v2) is determined to be an abnormal coordinate point, and its adjacent normal coordinate points are (8:05, v1) and (8:25, v3), then the corrected vertical axis coordinate value is (v1+v3) / 2, and the horizontal axis coordinate value remains 8:15, resulting in the corrected coordinate point (8:15, (v1+v3) / 2). After correcting all coordinate points, the corrected set of coordinate points is obtained.

[0082] Step S1345: Connect all coordinate points in the corrected coordinate point set sequentially using a smooth curve connection method to avoid sharp broken lines during the connection process, thus forming a preliminary evolution trend curve.

[0083] In this embodiment, a smooth curve is used to connect the corrected coordinate points, such as a Bézier curve or a spline curve, to ensure a smooth transition without sharp breaks in the curve. For example, connecting coordinate points (8:05, v1), (8:15, (v1+v3) / 2), and (8:25, v3) forms a preliminary evolutionary trend curve. This evolutionary trend curve can reflect the changing trend of the basic characteristics of the temporal association chain in different evolutionary stages.

[0084] Step S1346: Perform local optimization processing on the preliminary evolution trend curve. For the segment whose fluctuation amplitude exceeds the preset fluctuation threshold, adjust the segment using the curve trend of the adjacent segment. After completing the local optimization, the final evolution trend curve is obtained. The evolution trend curve is used to reflect the changes in the basic characteristics of the temporal association chain in different evolution stages.

[0085] In this embodiment, the preliminary evolutionary trend curve is analyzed segment by segment, and the fluctuation amplitude of each segment is calculated (e.g., the difference between the maximum and minimum vertical axis values ​​of the curve within that segment). If the fluctuation amplitude of a segment exceeds a preset fluctuation threshold (e.g., Δv_threshold), the segment is adjusted using the curve trend of its adjacent segments. For example, if a segment has a large fluctuation amplitude, and the curve trend of its preceding segment is a steady increase while the curve trend of its following segment is a steady decrease, then the curve of that segment is adjusted to make the transition smoother and reduce the fluctuation amplitude below the threshold. After completing all local optimizations, the final evolutionary trend curve is obtained.

[0086] Step S135: Extract the slope change features and inflection point distribution features of the evolution trend curve. The slope change features are obtained by calculating the slope change amplitude of the corresponding curve segments in adjacent stages. The inflection point distribution features are obtained by identifying the locations where the curve slope changes from positive to negative or from negative to positive and the distribution of these locations. The slope change features and inflection point distribution features are combined as the evolution trend features.

[0087] In this embodiment, the evolution trend curve is divided into segments corresponding to each evolution stage, such as one segment for each evolution stage. The slope change amplitude of adjacent segments is calculated. For example, if the slope of the first segment is k1 and the slope of the second segment is k2, the slope change amplitude is |k2-k1|. The slope change amplitudes of all adjacent segments are combined to obtain the slope change characteristics. Then, the inflection points of the curve are identified, i.e., the points where the slope changes from positive to negative or from negative to positive. The number of these points and their distribution time intervals are statistically analyzed to obtain the inflection point distribution characteristics. The slope change characteristics and the inflection point distribution characteristics are combined to form the evolution trend characteristics.

[0088] Step S136: Calculate the change magnitude of the temporal association strength between all adjacent traffic flow temporal segment nodes in each temporal association chain to obtain the association strength change sequence. The change magnitude is obtained by the absolute value of the difference between the subsequent association strength and the previous association strength.

[0089] In this embodiment, for adjacent node pairs in a temporal association chain, such as node 1 and node 2 having an association strength of s1, node 2 and node 3 having an association strength of s2, and node 3 and node 4 having an association strength of s3, the change ranges are |s2-s1| and |s3-s2|, respectively. Arranging the above change ranges in order, we obtain the association strength change sequence, for example, [|s2-s1|, |s3-s2|, ...].

[0090] Step S137: Analyze the fluctuation period characteristics and peak frequency characteristics of the correlation strength change sequence. The fluctuation period characteristics are obtained by identifying the time interval of repeated fluctuation patterns in the sequence. The peak frequency characteristics are obtained by statistically analyzing the ratio of the number of times values ​​exceeding a preset peak threshold appear in the sequence to the total time step. The fluctuation period characteristics and peak frequency characteristics are combined as the correlation strength change characteristics.

[0091] In this embodiment, the correlation strength change sequence is analyzed to identify recurring fluctuation patterns, such as the pattern [0.2, 0.3, 0.2] in the sequence. The time interval between these recurrences (i.e., every few time steps) is counted to obtain the fluctuation period feature. Then, the number of times values ​​exceeding a preset peak threshold (e.g., 0.4) occur in the sequence is counted, and divided by the total time step length (i.e., the length of the sequence) to obtain the peak frequency feature. These two features are combined to form the correlation strength change feature.

[0092] Step S138: Perform interactive fusion processing on the basic features within the stage, the independent components of the transition features between stages, the evolution trend features, and the correlation strength change features, calculate the correlation between different features, adjust the fusion weight of each feature according to the correlation, and generate the fused comprehensive features.

[0093] In this embodiment, the correlations among the independent components of the basic features within a stage and the transitional features between stages (such as differences and fluctuation counts), evolutionary trend features (slope change features and inflection point distribution features), and correlation strength change features (fluctuation cycle features and peak occurrence frequency features) are first calculated. For example, the correlation coefficient between the basic features within a stage and the slope change features, and the correlation coefficient between the differences in the transitional features between stages and the peak occurrence frequency features are calculated. The fusion weights of each feature are adjusted according to the magnitude of the correlation; the weights of features with high correlation can be appropriately increased, and the weights of features with low correlation can be appropriately decreased. Then, the above features are fused using a weighted concatenation method to generate a fused comprehensive feature. This comprehensive feature contains information from each original feature and has undergone weight adjustment.

[0094] Step S139: Integrate the comprehensive features with each original feature to form a set of temporal evolution features corresponding to each temporal association chain.

[0095] In this embodiment, the fused comprehensive features are integrated with the original features, such as the basic features within a stage, the independent components of the transition features between stages, evolution trend features, and correlation strength change features, to form a time-series evolution feature set. For example, the time-series evolution feature set includes comprehensive features, differences and fluctuation times of basic features within a stage and transition features between stages, slope change features and inflection point distribution features of evolution trend features, and fluctuation cycle features and peak occurrence frequency features of correlation strength change features. Each feature is a multi-dimensional numerical set, which are used together for subsequent abnormal pattern matching.

[0096] Step S140: Call the pre-trained temporal pattern mining model to perform abnormal evolution pattern matching processing on each of the temporal evolution feature sets, and generate abnormal pattern mining results. The abnormal pattern mining results include the matched abnormal evolution pattern type and the identification information of the corresponding temporal association chain.

[0097] Step S141: Input each of the time-series evolution feature sets into the feature input layer of the time-series pattern mining model, and perform dimension normalization processing on each feature in the time-series evolution feature set to convert it into a standard feature vector that conforms to the model processing format.

[0098] In this embodiment, the feature input layer of the temporal pattern mining model receives each temporal evolution feature set and performs dimensionality normalization on each feature in the set. For example, the numerical range of the basic feature (average rate of change) within a stage may be between 0 and 10, while the numerical range of the correlation strength change feature may be between 0 and 1. Through normalization (such as min-max normalization, which transforms the feature values ​​to between 0 and 1), the dimension of each feature conforms to the processing format of the model, generating a standard feature vector. Each element of the standard feature vector corresponds to a normalized feature value, and its dimension is consistent with the input dimension required by the model.

[0099] Step S142: Input the standard feature vector into the feature attention layer of the time-series pattern mining model. The feature attention layer calculates the importance weight of each feature in abnormal pattern matching. The importance weight is obtained through correlation analysis between the feature and historical abnormal pattern features. The higher the correlation, the greater the importance weight.

[0100] In this embodiment, the feature attention layer of the temporal pattern mining model receives standard feature vectors and calculates the correlation between each feature and historical anomaly pattern features. Historical anomaly pattern features are extracted from a set of historical anomaly temporal evolution features and stored in the model's database. For example, if a feature (such as the slope change feature of the evolution trend feature) has a high correlation with historical anomaly pattern features, it indicates that the feature is important in anomaly pattern matching, and its importance weight is larger; conversely, features with lower correlation have smaller weights. The importance weight is calculated for each feature in this way.

[0101] Step S143: Based on the calculated importance weights, weight each feature in the standard feature vector to obtain a weighted feature vector.

[0102] In this embodiment, based on the importance weights calculated in step S142, each element (i.e., the normalized value of each feature) in the standard feature vector is weighted. For example, if the standard feature vector is [v1, v2, v3, ...] and the corresponding importance weights are [w1, w2, w3, ...], then the weighted feature vector is [v1×w1, v2×w2, v3×w3, ...]. This weighted feature vector highlights the influence of important features.

[0103] Step S144: Input the weighted feature vector into the feature mapping layer of the time-series pattern mining model. The feature mapping layer maps the weighted feature vector to a preset pattern feature space through a preset mapping rule. The pattern feature space is a high-dimensional feature space constructed based on historical abnormal pattern features, thus obtaining the pattern space feature vector.

[0104] In this embodiment, the feature mapping layer of the time-series pattern mining model receives the weighted feature vectors and maps them into the pattern feature space according to a preset mapping rule (such as linear mapping, nonlinear mapping, etc., the specific mapping rule is determined during model training). The pattern feature space is a high-dimensional space where each point corresponds to a feature vector. This pattern feature space is constructed based on a large number of historical abnormal pattern features, making similar abnormal patterns relatively close in space. Through mapping, a pattern space feature vector is obtained, which has a specific coordinate position in the pattern feature space.

[0105] Step S145: Call the pattern library of the time series pattern mining model, calculate the similarity between the pattern space feature vector and each standard pattern feature vector in the pattern library. The similarity is obtained by comparing the spatial distance between the pattern space feature vector and each standard pattern feature vector in the pattern library in the pattern feature space. The closer the spatial distance, the higher the similarity. The pattern library stores standard pattern feature vectors corresponding to a variety of preset abnormal evolution patterns, and each abnormal evolution pattern corresponds to a unique pattern identifier.

[0106] Step S1451: Read the standard pattern feature vectors corresponding to all preset abnormal evolution patterns and the pattern identifiers corresponding to each abnormal evolution pattern from the pattern library of the time series pattern mining model, forming a corresponding set of standard pattern feature vectors and pattern identifiers.

[0107] In this embodiment, the pattern library of the time-series pattern mining model stores various abnormal evolution patterns, such as "sudden increase in traffic", "sudden decrease in traffic", and "fluctuation in traffic". Each pattern corresponds to a standard pattern feature vector and a pattern identifier (such as ID1, ID2, ID3, etc.). These standard pattern feature vectors and corresponding pattern identifiers are read from the pattern library to form a corresponding set, which facilitates the subsequent calculation of similarity.

[0108] Step S1452: For each standard pattern feature vector in the corresponding set, determine the coordinate position of the standard pattern feature vector in the pattern feature space, and at the same time determine the coordinate position of the pattern space feature vector in the pattern feature space.

[0109] In this embodiment, the pattern feature space is a high-dimensional space, and each standard pattern feature vector and pattern space feature vector has a corresponding coordinate position. For example, the coordinates of the standard pattern feature vector V1 in the pattern feature space are (x1, y1, z1, ...), and the coordinates of the pattern space feature vector V are (x, y, z, ...).

[0110] Step S1453: Calculate the spatial distance between two coordinate positions by traversing all dimensions of the two vectors, counting the differences in coordinate values ​​in each dimension, and then obtaining the spatial distance based on the differences.

[0111] In this embodiment, the spatial distance between V and V1 is calculated by traversing each dimension and calculating the differences between x-x1, y-y1, z-z1, etc. Then, the square root of the sum of the squares of the above differences (Euclidean distance), or by using other distance measurement methods (such as Manhattan distance, cosine distance, etc.), is used to obtain the spatial distance.

[0112] Step S1454: Determine the initial value of similarity based on the calculated spatial distance. The smaller the spatial distance, the higher the initial value of similarity.

[0113] In this embodiment, a mapping relationship between a spatial distance and an initial similarity degree is preset. For example, when the spatial distance is 0, the initial similarity degree is 1; when the spatial distance is d1, the initial similarity degree is 0.8; when the spatial distance is d2, the initial similarity degree is 0.5, etc. (d1 and d2 are preset distance thresholds). According to the calculated spatial distance, the corresponding initial similarity degree is found. For example, if the spatial distance is d0, and d0 < d1, then the initial similarity degree is 1; if d1 ≤ d0 < d2, then the initial similarity degree is 0.8, and so on.

[0114] Step S1455: Combine the time span of the timing correlation chain corresponding to the current timing evolution feature set to adjust the initial similarity degree. If the difference between the time span of the timing correlation chain and the time span corresponding to the standard mode feature vector is less than the preset difference threshold, then increase the initial similarity degree; if the difference between the time span of the timing correlation chain and the time span corresponding to the standard mode feature vector is greater than or equal to the preset difference threshold, then decrease the initial similarity degree.

[0115] In this embodiment, the time span corresponding to the standard mode feature vector is determined during model training. For example, the standard time span of the "sudden increase in traffic" mode is 15 minutes. The time span of the current timing correlation chain is 20 minutes, and the difference is 5 minutes. The preset difference threshold is 10 minutes. Since 5 minutes is less than 10 minutes, the initial similarity degree is increased, for example, increased by 0.1. If the difference is greater than or equal to 10 minutes, the initial similarity degree is decreased, for example, decreased by 0.1.

[0116] Step S1456: Use the adjusted similarity degree as the final similarity degree between the pattern space feature vector and the corresponding standard mode feature vector.

[0117] In this embodiment, the initial similarity degree obtained in step S1454, after being adjusted in step S1455, the obtained value is the final similarity degree between the pattern space feature vector and the corresponding standard mode feature vector. For example, the initial similarity degree is 0.8, the time span difference is less than the threshold, and after adjustment, it is increased by 0.1, and the final similarity degree is 0.9.

[0118] Step S1457: Record the final similarity degrees corresponding to all standard mode feature vectors, and at the same time associate the corresponding pattern identifiers to form an association list of similarity degrees and pattern identifiers.

[0119] In this embodiment, the final similarity score of each standard pattern feature vector is associated with its corresponding pattern identifier to form a list. For example, the final similarity score corresponding to pattern identifier ID1 is 0.9, the final similarity score corresponding to pattern identifier ID2 is 0.7, the final similarity score corresponding to pattern identifier ID3 is 0.8, and so on. This list is the association list between similarity score and pattern identifier.

[0120] Step S146: Sort all the calculated similarities and select the abnormal evolution pattern type and pattern identifier corresponding to the standard pattern feature vector with the highest similarity.

[0121] In this embodiment, the similarity in the association list of similarity and pattern identifier is sorted from high to low. The standard pattern feature vector with the highest similarity is found, and its corresponding abnormal evolution pattern type and pattern identifier are the most matching abnormal patterns in the current time-series evolution feature set. For example, the highest similarity after sorting is 0.9, the corresponding pattern identifier is ID1, and the abnormal evolution pattern type is "sudden increase in traffic".

[0122] Step S147: Check whether the similarity of the abnormal evolution pattern type with the highest similarity exceeds the preset matching threshold. If it exceeds the preset matching threshold, then determine that the abnormal evolution pattern type is the abnormal evolution pattern type matched by the current time-series evolution feature set.

[0123] In this embodiment, the preset matching threshold is 0.7. Since the highest similarity is 0.9, which exceeds 0.7, the abnormal evolution pattern type ("sudden increase in traffic") is determined to be the abnormal evolution pattern type matched by the current time-series evolution feature set.

[0124] Step S148: If the preset matching threshold is not exceeded, the abnormal evolution pattern type with the second highest similarity is taken as the candidate pattern type, and the difference features between the current time-series evolution feature set and the candidate pattern type are recorded.

[0125] In this embodiment, assuming the highest similarity is 0.65, which does not exceed the preset matching threshold of 0.7, the abnormal evolution pattern type corresponding to the second highest similarity (e.g., 0.6) is selected as the candidate pattern type. Simultaneously, the differences between the current temporal evolution feature set and the standard pattern feature vectors of the candidate pattern types are analyzed, and the difference features are extracted, such as different values ​​for a certain feature or different feature distributions, and these difference features are recorded.

[0126] Step S149: Record the matched abnormal evolution pattern type or candidate pattern type and the corresponding temporal association chain identification information and difference features, and organize them according to a preset format to generate the abnormal pattern mining results.

[0127] In this embodiment, the matched abnormal evolution pattern type (or candidate pattern type), the corresponding temporal association chain identifier information (such as the chain number), and the differential features (if any) are organized according to a preset format (such as JSON or XML format) to generate abnormal pattern mining results. For example, the abnormal pattern mining result is: {"Abnormal Evolution Pattern Type": "Traffic Surge Type", "Temporal Association Chain Identifier": "Chain 1", "Differential Features": null} (if there are no differential features).

[0128] Step S150: Generate a set of traffic flow control suggestions based on the abnormal pattern mining results, and adjust the control content in combination with the impact range analysis corresponding to the abnormal pattern. Each suggestion in the set of traffic flow control suggestions corresponds to one of the abnormal evolution pattern types, and each suggestion contains control direction information and control implementation step information for the corresponding abnormal evolution pattern.

[0129] Step S151: Analyze the abnormal pattern mining results and extract all abnormal evolution pattern types, corresponding time-series association chain identification information, and differential features contained therein.

[0130] In this embodiment, the abnormal pattern mining results are parsed to extract the abnormal evolution pattern type (e.g., "traffic surge type"), the corresponding time-series correlation chain identifier (e.g., "chain 1"), and differential features (if any). For example, this information is obtained by parsing fields from the abnormal pattern mining results in JSON format.

[0131] Step S152: For each of the abnormal evolution mode types, retrieve the basic control strategy associated with the abnormal evolution mode type from the preset control strategy library. The control strategy library stores the mapping relationship between each abnormal evolution mode type and the corresponding basic control strategy. The basic control strategy includes control objectives and basic control measures.

[0132] In this embodiment, the control strategy library stores basic control strategies corresponding to each type of abnormal evolution pattern. For example, the basic control strategy for the "sudden increase in traffic flow" type aims to reduce traffic flow in the area, and basic control measures include issuing traffic warnings, guiding vehicles to detour, and adjusting the duration of traffic lights at tunnel entrances. For the extracted "sudden increase in traffic flow" abnormal evolution pattern type, the corresponding basic control strategy is retrieved from the control strategy library.

[0133] Step S153: Based on the identification information of the time-series association chain corresponding to each of the abnormal evolution mode types, query the tunnel monitoring area information involved in the traffic flow time-series segment corresponding to the time-series association chain. The tunnel monitoring area information includes the location of the monitoring area, the number of lanes in the area, and the traffic flow carrying capacity limit of the area.

[0134] In this embodiment, based on the identifier information of the time-series association chain (such as "chain 1"), the tunnel monitoring area involved in the traffic flow time-series segment corresponding to the time-series association chain is queried. For example, the traffic flow time-series segment corresponding to chain 1 involves monitoring area A (tunnel entrance section) and monitoring area B (middle driving section). The information of these two monitoring areas is queried: monitoring area A is located 500 meters before the tunnel entrance, has 3 lanes, and a traffic flow capacity limit of 2000 vehicles per hour; monitoring area B is located in the middle section of the tunnel, has 4 lanes, and a traffic flow capacity limit of 2500 vehicles per hour.

[0135] Step S154: Based on the tunnel monitoring area information, analyze the surrounding monitoring areas that may be affected by the abnormal evolution pattern type, and determine the range of the affected surrounding monitoring areas by calculating the traffic flow interaction frequency between the target monitoring area and the surrounding monitoring areas.

[0136] Step S1541: Retrieve traffic flow interaction data between the target monitoring area and all surrounding monitoring areas within a historical time period from the tunnel traffic management system. The traffic flow interaction data includes the traffic flow value from the target monitoring area to the surrounding monitoring areas and the traffic flow value from the surrounding monitoring areas to the target monitoring area per unit time.

[0137] In this embodiment, the target monitoring area is monitoring area A (tunnel entrance section), and the surrounding monitoring areas include monitoring area B (middle driving section), monitoring area C (exit section), and monitoring area D (adjacent tunnel entrance section), etc. Traffic flow interaction data between monitoring area A and these surrounding monitoring areas is retrieved from the tunnel traffic management system database for historical time periods (e.g., morning rush hour of the past month). For example, within a unit of time (1 hour), the traffic flow from monitoring area A to monitoring area B is 1500 vehicles, and the traffic flow from monitoring area B to monitoring area A is 200 vehicles; the traffic flow from monitoring area A to monitoring area D is 500 vehicles, and the traffic flow from monitoring area D to monitoring area A is 100 vehicles, etc.

[0138] Step S1542: For each surrounding monitoring area, calculate the traffic flow interaction frequency between the target monitoring area and the surrounding monitoring area. The traffic flow interaction frequency is obtained by dividing the sum of the two-way traffic flow values ​​per unit time by the unit time length.

[0139] In this embodiment, for monitoring area B, the total two-way traffic flow value per unit time is 1500 + 200 = 1700 vehicles, and the unit time length is 1 hour, so the traffic flow interaction frequency is 1700 / 1 = 1700 vehicles / hour. For monitoring area D, the total two-way traffic flow value is 500 + 100 = 600 vehicles, and the traffic flow interaction frequency is 600 / 1 = 600 vehicles / hour.

[0140] Step S1543: Compare the traffic flow interaction frequency of each surrounding monitoring area with the preset traffic flow interaction frequency threshold.

[0141] In this embodiment, the preset traffic flow interaction frequency threshold is 1000 vehicles / hour. The traffic flow interaction frequency in monitoring area B is 1700 vehicles / hour, exceeding the threshold; the traffic flow interaction frequency in monitoring area D is 600 vehicles / hour, which does not exceed the threshold.

[0142] Step S1544: If the traffic flow interaction frequency of any surrounding monitoring area exceeds the traffic flow interaction frequency threshold, then the surrounding monitoring area is determined to be a highly correlated surrounding area that may be affected by the abnormal evolution pattern.

[0143] In this embodiment, the traffic flow interaction frequency of monitoring area B, 1700 vehicles / hour, exceeds the threshold of 1000 vehicles / hour, so monitoring area B is determined to be a highly correlated surrounding area.

[0144] Step S1545: If the traffic flow interaction frequency of any surrounding monitoring area does not exceed the traffic flow interaction frequency threshold, but the surrounding monitoring area is directly adjacent to the target monitoring area and there are no other monitoring areas between them, then the surrounding monitoring area is determined to be a low-association surrounding area that may be affected by the abnormal evolution pattern.

[0145] In this embodiment, the traffic flow interaction frequency of monitoring area D is 600 vehicles / hour, which does not exceed the threshold. However, monitoring area D is not directly adjacent to the target monitoring area A (there is monitoring area B and the tunnel body in between), so it is not determined to be a low-association surrounding area. Assuming there is another monitoring area E that is directly adjacent to monitoring area A, with a traffic flow interaction frequency of 800 vehicles / hour, which does not exceed the threshold, then monitoring area E is determined to be a low-association surrounding area.

[0146] Step S1546: If the traffic flow interaction frequency of any surrounding monitoring area does not exceed the traffic flow interaction frequency threshold and is not directly adjacent to the target monitoring area, then the surrounding monitoring area is determined to be an unaffected surrounding area.

[0147] In this embodiment, the monitoring area D is not directly adjacent to the target monitoring area A, and the traffic flow interaction frequency does not exceed the threshold. Therefore, the monitoring area D is determined to be an unaffected surrounding area.

[0148] Step S1547: Integrate the highly correlated and low-correlation surrounding areas to form the affected surrounding monitoring area range, and mark the correlation type of each affected surrounding monitoring area.

[0149] In this embodiment, the highly correlated surrounding area is designated as monitoring area B, and the low-correlation surrounding area is designated as monitoring area E. These areas are integrated to form the affected surrounding monitoring area range. The correlation type of monitoring area B is marked as "high correlation", and the correlation type of monitoring area E is marked as "low correlation".

[0150] Step S155: Integrate the tunnel monitoring area information and the affected surrounding monitoring area range into the corresponding basic control strategy, adjust the control object and control range in the basic control strategy, expand the control object to the affected surrounding monitoring area, and adjust the control range according to the number of lanes in the area to obtain a preliminary personalized control strategy for a specific monitoring area and surrounding area.

[0151] In this embodiment, the basic control strategy originally targeted monitoring area A, but now expands to include monitoring areas B (high correlation) and E (low correlation). Regarding the control range, monitoring area A has 3 lanes, monitoring area B has 4 lanes, and monitoring area E has 2 lanes. The traffic light duration adjustment in the basic control measures, originally targeting the 3 lanes in monitoring area A, now targets the 4 lanes in monitoring area B, with the adjustment magnitude adjusted according to the proportion of lanes (e.g., the more lanes, the longer the traffic light duration); for the 2 lanes in monitoring area E, the adjustment magnitude is correspondingly reduced. This results in a preliminary personalized control strategy, targeting monitoring areas A, B, and E, with the control range adjusted according to the number of lanes in each area.

[0152] Step S156: Analyze the potential impact of each of the preliminary personalized control strategies on traffic flow in the target monitoring area and the affected surrounding monitoring areas. By simulating the changes in traffic flow after the implementation of the control measures, determine whether the control measures will lead to new traffic flow anomalies.

[0153] In this embodiment, traffic flow simulation software is used to input the content of a preliminary personalized control strategy and simulate traffic flow changes after the control measures are implemented. For example, the system simulates the changes in traffic flow in areas A, B, and E after issuing traffic warnings, guiding vehicles to detour, and adjusting traffic light durations. It checks whether new traffic flow anomalies will occur, such as whether traffic flow in monitoring area E, which was originally low, will suddenly increase after the control measures are implemented, or whether traffic flow in monitoring area B will suddenly decrease, leading to new congestion.

[0154] Step S157: If the simulation results show that it will lead to new traffic flow anomalies, adjust the intensity of the control measures in the initial personalized control strategy, reduce the intensity of the control measures that may cause new anomalies, and repeat the impact simulation until the simulation results show that it will not lead to new traffic flow anomalies.

[0155] In this embodiment, if the simulation results show that the traffic flow in monitoring area E increases sharply after regulation, exceeding its capacity limit and indicating a new anomaly, the intensity of the regulation measures in the initial personalized regulation strategy will be adjusted. For example, the number of vehicles guided to detour to monitoring area E may be reduced, or the adjustment range of the traffic light duration in monitoring area E may be decreased. Then, the simulation will be repeated until the simulation results show that the traffic flow in each monitoring area is within the normal range and no new anomalies occur.

[0156] Step S158: If the simulation results show that it will not lead to new traffic flow anomalies, then retain the current preliminary personalized control strategy and use it as the optimized personalized control strategy.

[0157] In this embodiment, after adjustment and re-simulation, the simulation results show that the traffic flow in monitoring areas A, B, and E are all within the normal range and no new anomalies have appeared. Therefore, the current preliminary personalized control strategy is retained and used as the optimized personalized control strategy.

[0158] Step S159: Integrate all optimized personalized control strategies, add corresponding abnormal evolution mode type identifier, monitoring area identifier and implementation priority to each strategy, and form the traffic flow control suggestion set. Each suggestion in the traffic flow control suggestion set includes abnormal evolution mode type, corresponding monitoring area information and optimized control strategy content.

[0159] In this embodiment, all optimized personalized control strategies are collected, and each strategy is assigned an abnormal evolution mode type identifier (e.g., "sudden traffic surge"), a monitoring area identifier (e.g., identifiers for monitoring areas A, B, and E), and an implementation priority (determined based on the severity and scope of the anomaly, e.g., control priority for highly correlated areas is higher than for low-correlation areas). These strategies are then integrated to form a set of traffic flow control recommendations. For example, one recommendation might be: the abnormal evolution mode type is "sudden traffic surge," the corresponding monitoring areas are A, B, and E, and the optimized control strategy is to issue traffic warnings, guide some vehicles to detour to unaffected areas, increase the traffic light duration by 5 seconds per cycle in monitoring area A, by 8 seconds per cycle in monitoring area B, and by 3 seconds per cycle in monitoring area E, with a high implementation priority. All these recommendations are then integrated to form a set of traffic flow control recommendations.

[0160] Figure 2The illustration shows exemplary hardware and software components of a tunnel traffic flow big data mining system 100 that incorporates time series data, which can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 can be used in the tunnel traffic flow big data mining system 100 incorporating time series data and to perform the functions described in this application.

[0161] For example, the tunnel traffic flow big data mining system 100 incorporating time series data may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the tunnel traffic flow big data mining system 100 incorporating time series data may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The tunnel traffic flow big data mining system 100 incorporating time series data also includes an I / O interface 150 between the computer and other input / output devices.

[0162] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for mining tunnel traffic flow big data by combining time series data, characterized in that, The method includes: Obtain a time-series data set corresponding to the big data of tunnel traffic flow. The time-series data set contains multiple traffic flow time-series segments collected in a continuous time dimension. Each traffic flow time-series segment records the traffic flow change information of a specific monitoring area in the tunnel within a corresponding time interval. A temporal association chain is constructed based on the temporal adjacency relationship between the traffic flow time series segments. During the construction process, feature filling processing is performed on the time gaps of adjacent segments to obtain a set of temporal association chains. The temporal association chain is used to characterize the association relationship of traffic flow change information in different time intervals, and each temporal association chain contains a node sequence and a correlation strength parameter between nodes. For each of the time-series association chains, a time-series evolution feature extraction process is performed. Features are extracted from the time dimension and the association dimension respectively and then fused together to obtain a time-series evolution feature set corresponding to each of the time-series association chains. The time-series evolution feature set includes the change trend features and association strength change features of the time-series association chains in the time dimension. A pre-trained temporal pattern mining model is invoked to perform abnormal evolution pattern matching processing on each of the temporal evolution feature sets, generating abnormal pattern mining results. The abnormal pattern mining results include the matched abnormal evolution pattern type and the identification information of the corresponding temporal association chain. Based on the results of the abnormal pattern mining, a set of traffic flow control suggestions is generated. The control content is adjusted in combination with the impact range analysis corresponding to the abnormal pattern. Each suggestion in the set of traffic flow control suggestions corresponds to one of the abnormal evolution pattern types, and each suggestion contains control direction information and control implementation step information for the corresponding abnormal evolution pattern.

2. The method for mining tunnel traffic flow big data by combining time series data according to claim 1, characterized in that, The process involves constructing a temporal association chain based on the temporal adjacency relationship between the traffic flow time-series segments. During the construction process, feature imputation is performed on the time gaps between adjacent segments to obtain a set of temporal association chains, including: Extract time interval information of all traffic flow time series segments from the time series data set, wherein the time interval information includes the start time and end time of each traffic flow time series segment; The temporal adjacency relationship between any two traffic flow time segments is determined based on the time interval information. The temporal adjacency relationship is used to indicate whether there is a continuous connection between the time intervals of the two traffic flow time segments. If there is an interval between the end time of the previous traffic flow time segment and the start time of the next traffic flow time segment, it is determined that there is a temporal gap between the two traffic flow time segments. For two traffic flow time segments with a time gap, feature interpolation is performed using traffic flow change information of adjacent segments to generate a transition feature sequence that fills the time gap. The time length of the transition feature sequence is consistent with the time length of the time gap. For two traffic flow time series segments that have the aforementioned temporal adjacency relationship, the similarity of traffic flow change information in the two traffic flow time series segments is calculated, and the similarity is used as the temporal correlation strength between the two traffic flow time series segments. Based on the distribution of correlation strength of similar segments in historical traffic flow data, the currently calculated temporal correlation strength is dynamically calibrated to eliminate the influence of outliers on the correlation strength. Based on the preset association strength threshold, traffic flow time series segments that meet the threshold requirements are selected to obtain a set of effective association pairs. Each association pair in the set of effective association pairs contains two traffic flow time series segments and a calibrated time series association strength. Taking one traffic flow time series segment in each of the effective association pairs as the starting node and the other traffic flow time series segment as the subsequent node, multiple effective association pairs are connected sequentially according to the order of time intervals. If nodes are repeated during the connection process, the connection path with higher association strength is retained to construct an initial time series association chain. Redundant nodes are removed from the initial temporal association chain. Nodes in the initial temporal association chain whose traffic flow change information repetition with adjacent nodes exceeds a preset repetition threshold are deleted. At the same time, the temporal association strength between nodes before and after node removal is updated. All optimized temporal association chains are integrated, and the number of nodes, total time span, and average association strength of each temporal association chain are recorded to form the temporal association chain set.

3. The method for mining tunnel traffic flow big data by combining time series data according to claim 2, characterized in that, For two traffic flow time-series segments that have the aforementioned temporal adjacency relationship, the similarity of traffic flow change information in the two traffic flow time-series segments is calculated, and the similarity is used as the temporal correlation strength between the two traffic flow time-series segments, including: Traffic flow change information is extracted from two traffic flow time segments that have the temporal adjacency relationship. If a transition feature sequence exists, the transition feature sequence is merged with the traffic flow change information of the next traffic flow time segment to obtain the first traffic flow change sequence and the subsequent traffic flow change sequence, respectively. The first traffic flow change sequence and the subsequent traffic flow change sequences are processed to unify the length of the time dimension. For the first traffic flow change sequence and the subsequent traffic flow change sequences after unification, the traffic flow value difference at the corresponding position is calculated step by step to obtain a difference sequence composed of multiple time step difference values. Based on the differences in traffic flow values ​​at all time steps, the overall volatility of the difference sequence is calculated, and the overall volatility is characterized by the distribution concentration of all difference values ​​in the difference sequence. The basic value for the degree of similarity is determined based on the overall degree of fluctuation. The lower the overall degree of fluctuation, the higher the basic value for the degree of similarity. The base value of similarity is adjusted by combining the location correlation of the monitoring areas corresponding to the two traffic flow time series segments. If the monitoring areas corresponding to the two traffic flow time series segments are adjacent, the base value of similarity is increased; if the monitoring areas corresponding to the two traffic flow time series segments are not adjacent, the base value of similarity remains unchanged. The corrected similarity is used as the temporal correlation strength between the two traffic flow time series segments.

4. The method for mining tunnel traffic flow big data by combining time series data according to claim 1, characterized in that, The step of performing temporal evolution feature extraction processing on each of the temporal association chains involves extracting features from both the time dimension and the association dimension, and then interactively fusing them to obtain a temporal evolution feature set corresponding to each of the temporal association chains, including: Each of the time-series association chains is divided into multiple consecutive evolutionary stages according to the chronological order of the time intervals. Each evolutionary stage contains at least two adjacent traffic flow time-series segment nodes, and the time span of each evolutionary stage remains consistent. Extract traffic flow change information of all traffic flow time sequence nodes within each evolution stage, calculate the average rate of change of traffic flow change information within each evolution stage, the average rate of change is obtained by dividing the total change of traffic flow value within the stage by the stage time span, and use the average rate of change as the basic feature within the stage. The transition of traffic flow change information between two adjacent evolutionary stages is analyzed. The difference between the average rate of change of the previous evolutionary stage and the average rate of change of the next evolutionary stage is calculated. At the same time, the number of fluctuations in traffic flow values ​​during the transition is counted. The difference and the number of fluctuations are respectively used as independent components of the transition characteristics between stages. Based on the basic features within each evolutionary stage and the transitional features between each evolutionary stage in each of the time-series association chains, an evolutionary trend curve of the time-series association chain is constructed. The midpoint of each evolutionary stage is used as the horizontal axis coordinate, and the basic features within each evolutionary stage are used as the vertical axis coordinate to plot the evolutionary trend curve. The slope change characteristics and inflection point distribution characteristics of the evolution trend curve are extracted. The slope change characteristics are obtained by calculating the slope change amplitude of the corresponding curve segments in adjacent stages. The inflection point distribution characteristics are obtained by identifying the locations where the curve slope changes from positive to negative or from negative to positive and the distribution of these locations. The slope change characteristics and inflection point distribution characteristics are combined as the evolution trend characteristics. Calculate the change magnitude of the temporal association strength between all adjacent traffic flow temporal segment nodes in each temporal association chain to obtain the association strength change sequence. The change magnitude is obtained by the absolute value of the difference between the subsequent association strength and the previous association strength. The fluctuation period characteristics and peak frequency characteristics of the correlation strength change sequence are analyzed. The fluctuation period characteristics are obtained by identifying the time interval of repeated fluctuation patterns in the sequence. The peak frequency characteristics are obtained by statistically analyzing the ratio of the number of times values ​​exceeding a preset peak threshold appear in the sequence to the total time step. The fluctuation period characteristics and peak frequency characteristics are combined as the correlation strength change characteristics. The independent components of the basic features within the stage, the transition features between the stages, the evolution trend features, and the correlation strength change features are interactively fused to calculate the correlation between different features. The fusion weights of each feature are adjusted according to the correlation to generate the fused comprehensive features. The comprehensive features are integrated with the original features to form a set of temporal evolution features corresponding to each temporal association chain.

5. The method for mining tunnel traffic flow big data by combining time series data according to claim 4, characterized in that, Based on the basic characteristics within each evolutionary stage and the transitional characteristics between each evolutionary stage in each temporal association chain, an evolutionary trend curve of the temporal association chain is constructed. The curve is plotted with the midpoint of each evolutionary stage as the horizontal axis and the basic characteristics within each stage as the vertical axis, including: Extract the start time and end time from the time interval information of each evolutionary stage, calculate the midpoint between the start time and the end time, and use the midpoint as the midpoint of the evolutionary stage. Using the midpoint of each evolutionary stage as the horizontal axis coordinate value and the basic characteristics within each evolutionary stage as the vertical axis coordinate value, the coordinate points corresponding to each evolutionary stage are marked in a two-dimensional coordinate system, and each coordinate point contains both horizontal axis coordinate values ​​and vertical axis coordinate values. Check if there are any abnormal deviations in the marked coordinate points. If the difference between the ordinate value of any coordinate point and the ordinate values ​​of the two adjacent coordinate points exceeds the preset deviation threshold, then the coordinate point is determined to be an abnormal coordinate point. The abnormal coordinate points are corrected by replacing the ordinate value of the abnormal coordinate point with the average of the ordinate values ​​of two adjacent normal coordinate points, while keeping the abscissa value of the abnormal coordinate point unchanged, thus obtaining the corrected set of coordinate points. All coordinate points in the corrected coordinate point set are connected sequentially using a smooth curve connection method to avoid sharp broken lines during the connection process, thus forming a preliminary evolution trend curve; The preliminary evolution trend curve is locally optimized. For segments whose fluctuation amplitude exceeds a preset fluctuation threshold, the curve trend of adjacent segments is used to adjust the segment. After the local optimization is completed, the final evolution trend curve is obtained. The evolution trend curve is used to reflect the changes in the basic characteristics of the temporal association chain in different evolution stages.

6. The method for mining tunnel traffic flow big data by combining time series data according to claim 1, characterized in that, The pre-trained temporal pattern mining model is invoked to perform abnormal evolution pattern matching processing on each of the temporal evolution feature sets, generating abnormal pattern mining results, including: Each of the time-series evolution feature sets is input into the feature input layer of the time-series pattern mining model, and each feature in the time-series evolution feature set is subjected to dimension normalization processing to convert it into a standard feature vector that conforms to the model processing format. The standard feature vector is input into the feature attention layer of the time-series pattern mining model. The feature attention layer calculates the importance weight of each feature in the abnormal pattern matching. The importance weight is obtained through the correlation analysis between the feature and the historical abnormal pattern features. The higher the correlation, the greater the importance weight. Based on the calculated importance weights, each feature in the standard feature vector is weighted to obtain a weighted feature vector. The weighted feature vector is input into the feature mapping layer of the time-series pattern mining model. The feature mapping layer maps the weighted feature vector to a preset pattern feature space through a preset mapping rule. The pattern feature space is a high-dimensional feature space constructed based on historical abnormal pattern features, thus obtaining the pattern space feature vector. The pattern library of the time-series pattern mining model is called, and the similarity between the pattern space feature vector and each standard pattern feature vector in the pattern library is calculated. The similarity is obtained by comparing the spatial distance between the pattern space feature vector and each standard pattern feature vector in the pattern library in the pattern feature space. The closer the spatial distance, the higher the similarity. The pattern library stores standard pattern feature vectors corresponding to a variety of preset abnormal evolution patterns, and each abnormal evolution pattern corresponds to a unique pattern identifier. All calculated similarities are sorted, and the abnormal evolution pattern type and pattern identifier corresponding to the standard pattern feature vector with the highest similarity are selected. Check whether the similarity of the abnormal evolution pattern type with the highest similarity exceeds a preset matching threshold. If it exceeds the preset matching threshold, then determine that the abnormal evolution pattern type is the abnormal evolution pattern type matched by the current time-series evolution feature set. If the preset matching threshold is not exceeded, the abnormal evolution pattern type with the second highest similarity is selected as the candidate pattern type, and the difference features between the current time-series evolution feature set and the candidate pattern type are recorded. Record the matched abnormal evolution pattern type or candidate pattern type and the corresponding temporal association chain identification information and difference features, and organize them according to a preset format to generate the abnormal pattern mining results.

7. The method for mining tunnel traffic flow big data by combining time series data according to claim 6, characterized in that, The step of calling the pattern library of the time-series pattern mining model and calculating the similarity between the pattern space feature vector and each standard pattern feature vector in the pattern library includes: Read the standard pattern feature vectors corresponding to all preset abnormal evolution patterns and the pattern identifiers corresponding to each abnormal evolution pattern from the pattern library of the time series pattern mining model to form a corresponding set of standard pattern feature vectors and pattern identifiers. For each standard pattern feature vector in the corresponding set, determine the coordinate position of the standard pattern feature vector in the pattern feature space, and at the same time determine the coordinate position of the pattern space feature vector in the pattern feature space; To calculate the spatial distance between two coordinate positions, it is necessary to traverse all dimensions of the two vectors, count the difference in coordinate values ​​in each dimension, and then obtain the spatial distance based on the difference. The initial value of similarity is determined based on the calculated spatial distance; the smaller the spatial distance, the higher the initial value of similarity. Based on the time span of the temporal association chain corresponding to the current temporal evolution feature set, the initial value of the similarity is adjusted. If the difference between the time span of the temporal association chain and the time span corresponding to the standard pattern feature vector is less than a preset difference threshold, the initial value of the similarity is increased. If the difference between the time span of the temporal association chain and the time span corresponding to the standard pattern feature vector is greater than or equal to the preset difference threshold, the initial value of the similarity is decreased. The adjusted similarity is taken as the final similarity between the pattern space feature vector and the corresponding standard pattern feature vector; Record the final similarity scores corresponding to all standard pattern feature vectors, and associate them with the corresponding pattern identifiers to form a list of associations between similarity scores and pattern identifiers.

8. The method for mining tunnel traffic flow big data by combining time series data according to claim 1, characterized in that, The process of generating a set of traffic flow control suggestions based on the abnormal pattern mining results, and adjusting the control content in conjunction with the impact range analysis corresponding to the abnormal patterns, includes: The abnormal pattern mining results are analyzed to extract all abnormal evolution pattern types, corresponding identification information and differential features of the time-series association chains contained therein; For each of the aforementioned abnormal evolution mode types, a basic control strategy associated with that abnormal evolution mode type is retrieved from a preset control strategy library. The control strategy library stores the mapping relationship between each abnormal evolution mode type and the corresponding basic control strategy. The basic control strategy includes control objectives and basic control measures. Based on the identification information of the temporal correlation chain corresponding to each of the aforementioned abnormal evolution mode types, query the tunnel monitoring area information involved in the traffic flow time series segment corresponding to the temporal correlation chain. The tunnel monitoring area information includes the location of the monitoring area, the number of lanes in the area, and the traffic flow carrying capacity limit of the area. Based on the information of the tunnel monitoring area, the surrounding monitoring areas that may be affected by this abnormal evolution pattern type are analyzed. By calculating the traffic flow interaction frequency between the target monitoring area and the surrounding monitoring areas, the range of the affected surrounding monitoring areas is determined. The tunnel monitoring area information and the affected surrounding monitoring area range are integrated into the corresponding basic control strategy. The control object and control range in the basic control strategy are adjusted. The control object is expanded to the affected surrounding monitoring area, and the control range is adjusted according to the number of lanes in the area, so as to obtain a preliminary personalized control strategy for a specific monitoring area and surrounding area. The impact of each of the aforementioned preliminary personalized control strategies on traffic flow in the target monitoring area and the affected surrounding monitoring areas is analyzed. By simulating the changes in traffic flow after the implementation of the control measures, it is determined whether the control measures will lead to new traffic flow anomalies. If the simulation results show that it will lead to new traffic flow anomalies, then adjust the intensity of the control measures in the initial personalized control strategy, reduce the intensity of the control measures that may trigger new anomalies, and repeat the impact simulation until the simulation results show that it will not lead to new traffic flow anomalies. If the simulation results show that it will not lead to new traffic flow anomalies, the current preliminary personalized control strategy will be retained and used as the optimized personalized control strategy. All optimized personalized control strategies are integrated, and each strategy is given a corresponding abnormal evolution mode type identifier, monitoring area identifier, and implementation priority to form the traffic flow control suggestion set. Each suggestion in the traffic flow control suggestion set includes the abnormal evolution mode type, corresponding monitoring area information, and optimized control strategy content.

9. The method for mining tunnel traffic flow big data by combining time series data according to claim 8, characterized in that, The analysis based on the tunnel monitoring area information determines the potential impact of this abnormal evolution pattern on surrounding monitoring areas. This is achieved by calculating the traffic flow interaction frequency between the target monitoring area and surrounding monitoring areas to identify the affected surrounding monitoring areas, including: Traffic flow interaction data between the target monitoring area and all surrounding monitoring areas within a historical time period is retrieved from the tunnel traffic management system. The traffic flow interaction data includes the traffic flow value from the target monitoring area to the surrounding monitoring areas and the traffic flow value from the surrounding monitoring areas to the target monitoring area per unit time. For each surrounding monitoring area, the traffic flow interaction frequency between the target monitoring area and the surrounding monitoring area is calculated. The traffic flow interaction frequency is obtained by dividing the sum of the two-way traffic flow values ​​per unit time by the unit time length. The traffic flow interaction frequency of each surrounding monitoring area is compared with the preset traffic flow interaction frequency threshold. If the traffic flow interaction frequency of any surrounding monitoring area exceeds the traffic flow interaction frequency threshold, then the surrounding monitoring area is determined to be a highly correlated surrounding area that may be affected by abnormal evolution patterns. If the traffic flow interaction frequency of any surrounding monitoring area does not exceed the traffic flow interaction frequency threshold, but the surrounding monitoring area is directly adjacent to the target monitoring area and there are no other monitoring areas between them, then the surrounding monitoring area is determined to be a low-association surrounding area that may be affected by the abnormal evolution pattern. If the traffic flow interaction frequency of any surrounding monitoring area does not exceed the traffic flow interaction frequency threshold and is not directly adjacent to the target monitoring area, then the surrounding monitoring area is determined to be an unaffected surrounding area. The highly correlated and low-correlated surrounding areas are integrated to form the affected surrounding monitoring area range, and the correlation type of each affected surrounding monitoring area is marked.

10. A tunnel traffic flow big data mining system combining time series data, characterized in that, The tunnel traffic flow big data mining system combining time series includes a processor and a memory, the memory and the processor are connected, the memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the tunnel traffic flow big data mining method combining time series as described in any one of claims 1-9.

Images