Big data-based traffic flow and accident correlation analysis method and system

Abstract

The application provides a traffic flow and accident correlation analysis method and system based on big data, and belongs to the technical field of traffic safety active warning.The three-layer time sequence detection framework of the wave speed disorder window, the road surface dry-wet state change chain and the oscillation monotonous increasing period of the vehicle head time interval is constructed, the oscillation increasing period is verified for effectiveness in the state change transition period, and then the wave speed disorder window is set in the union and superposition, so that the high-precision identification of the accident precursor interval of the road section downstream of the signal intersection is realized.Compared with the existing method, the causal coupling relationship among the traffic wave propagation anomaly, the road surface adhesion coefficient mutation and the deterioration of the microscopic following behavior is first introduced into the unified analysis framework, and the false alarm rate of the single feature warning is significantly reduced.

Description

Technical Field

[0001] This invention belongs to the field of proactive traffic safety early warning technology, specifically involving a method and system for analyzing the correlation between traffic flow and accidents based on big data. Background Technology

[0002] Traffic accidents are rarely isolated events; they are closely causally related to the evolution of traffic flow and the deterioration of road surface physical conditions. In the field of intelligent transportation systems research, mining accident precursor features from multi-source traffic data to achieve proactive early warning of potential risks has become one of the key technical paths to improve road traffic safety. In recent years, the correlation mechanism between traffic flow and accidents has been explored from different dimensions. At the macro level, a large number of studies have established statistical regression models or machine learning models based on historical accident data to analyze the correlation between accident incidence and macro parameters such as traffic flow, speed, and occupancy. These methods, based on a large number of historical accident samples, output the probability distribution of accident risk and can be used for road network-level safety situation assessment. At the micro level, the relationship between vehicle following behavior (headway, acceleration and deceleration, etc.) and accident risk has attracted attention. Studies have found that excessively small following distances and increased speed oscillations both increase the probability of rear-end collisions. However, existing studies generally analyze traffic wave speed characteristics, road surface condition characteristics, and vehicle following behavior characteristics as independent factors, lacking a systematic joint exploration of the temporal coupling relationship between the three before an accident, making it difficult to form a unified identification framework, thus resulting in deficiencies in the timeliness and accuracy of early warning.

[0003] For example, CN119672963B discloses an intelligent safety operation and maintenance method and system for urban tunnels that combines multiple data sources. It focuses on the periodic wave impact phenomenon of traffic flow at tunnel entrances and exits caused by traffic light phase switching. It systematically characterizes the peak characteristics of the entrance by constructing a waveform feature mapping matrix and combines multi-source detector data to assess traffic congestion risk and traffic safety levels. By combining the periodic characteristics of traffic waves with tunnel safety operation and maintenance, and achieving a forward-looking assessment of congestion risk through similarity analysis of the wave feature matrix, it represents a breakthrough in the field of macro-level tunnel safety management. However, this patent mainly targets the traffic light connection area at tunnel entrances, limiting its application to urban tunnels and not extending to the identification of accident precursors in downstream sections of general signalized intersections. More importantly, this technical solution focuses on the accumulation pattern of traffic waves at the entrance section, without delving into the fluctuation pattern of the headway in the micro-following behavior of vehicles, nor does it introduce the moderating effect of road surface physical conditions (such as changes in dry and wet conditions) on accident risk. The safety assessment results are essentially still based on the engineering characteristics of traffic flow, lacking attention to driver behavior adaptability and sudden changes in road surface adhesion coefficient. At the level of accident cause identification, it has not yet moved from traffic flow engineering indicators to the coupling analysis of driving behavior response and changes in road surface environment.

[0004] Furthermore, CN102419905A discloses a method for determining the traffic impact range of highway accidents based on traffic wave theory, applying traffic wave theory to assess the spatiotemporal impact range of highway traffic accidents. This method analyzes the actual accumulation and dissipation process of traffic waves after an accident, calculating the traffic wave velocity at each stage and combining it with a flow-density relationship diagram to determine the temporal boundaries and spatial range of the accident's impact at each stage, providing decision support for accident emergency management and traffic guidance. It mathematically models the evolution mechanism of accident morphology at the theoretical level, possessing clear physical interpretability. However, it is essentially a post-accident assessment—the analysis starts at the moment the accident has already occurred, and the scope of the accident's impact is inferred by measuring the congestion wave generated at the scene after the accident. It is an impact assessment of an event that has already occurred and lacks predictive capabilities for the future. It cannot identify precursory signals such as abnormal wave velocity, sudden changes in road conditions, or deteriorating driving behavior before an accident occurs, making it difficult to meet the technical requirements of proactive safety warnings. Summary of the Invention

[0005] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0006] In view of the aforementioned existing problems, the present invention is proposed.

[0007] To address the aforementioned technical problems, the present invention provides the following technical solution: A big data-based traffic flow and accident correlation analysis method includes: recording the traffic light phase switching time and the first surge in downstream cross-section traffic flow for each signal cycle; calculating the wave velocity time difference for each signal cycle; marking the corresponding time period as a wave velocity disorder window when the wave velocity time difference exhibits a non-monotonic change within multiple consecutive signal cycles; discretizing the road surface condition into dry and wet categories within the wave velocity disorder window to generate a state change chain; extracting the headway sequence within a preset duration after the start of the state change chain; counting the number of occurrences of local extreme points within the sliding time window from the headway sequence as the oscillation count; filtering time periods where the oscillation count monotonically increases with time; spatiotemporally overlaying the wave velocity disorder window with the time periods where the oscillation count monotonically increases; validating the time periods where the oscillation count monotonically increases using the transition period corresponding to the state change chain; and outputting the overlaid continuous time period as an accident correlation precursor interval.

[0008] In a preferred embodiment of the present invention, the wave velocity time difference is obtained only when the first surge in flow is later than the signal light phase switching time. The wave velocity time difference of the signal cycle is retained only when the first surge in flow is later than the signal light phase switching time. Three signal cycles with consecutive original signal cycle numbers and all retaining the wave velocity time difference are selected, and it is determined whether there is an extreme value relationship between the wave velocity time difference of the three signal cycles, where the intermediate signal cycle is greater than the two adjacent signal cycles or the intermediate signal cycle is less than the two adjacent signal cycles. When an extreme value relationship exists, the time period between the signal light phase switching time of the first signal cycle and the signal light phase switching time of the third signal cycle is marked as the wave velocity disturbance window.

[0009] As a preferred embodiment of the present invention, the discretization of road surface conditions into dry and wet categories includes: collecting dielectric constant at fixed time intervals within a wave velocity disturbance window; determining dryness based on dielectric constant below a first threshold, dampness based on dielectric constant between the first and second thresholds, and water accumulation based on dielectric constant above the second threshold; dampness and water accumulation are collectively referred to as wet road surfaces; and defining dry to damp, dry to water accumulation, damp to dry, and water accumulation to dry as state change chains.

[0010] As a preferred embodiment of the present invention, the method for obtaining the first traffic surge moment is as follows: the time interval of each signal cycle is from the traffic light phase switching moment of the current signal cycle to the traffic light phase switching moment of the next signal cycle, and all headway distances falling within the corresponding time interval are extracted; the arithmetic mean of all headway distances within the signal cycle is calculated as the average headway distance; for a signal cycle in which the average headway distance of the previous signal cycle is valid, the first headway distance in the headway distances of the current signal cycle that is less than a preset proportion of the average headway distance of the previous signal cycle is found in chronological order, and the end moment of the headway distance is taken as the first traffic surge moment.

[0011] As a preferred embodiment of the present invention, the time extraction method after the start of the state change chain is as follows: search for the first occurrence of the state change chain within the wave velocity disorder window; if no state change chain exists, terminate the process and output an empty set; otherwise, take the time of the first sampling point after the state change occurs as the start time of the state change chain; extract a time interval of a preset duration from the start time of the state change chain, extract all the headway distances whose start times are within the time interval from the downstream section detection data to form a headway distance sequence, and record the start time of each headway distance.

[0012] As a preferred embodiment of the present invention, the occurrence count of local extreme points within the statistical sliding time window includes: setting a sliding time window with a window length of the first window length and a step size of the first step length, taking the start time of the first train headway in the train headway time distance sequence as the left boundary and the end time of the last train headway time distance as the right boundary, and sliding sequentially from the left; for each sliding time window, selecting all train headway times that overlap with the time interval covered by the sliding time window to form a subsequence; when the number of elements in the subsequence is greater than or equal to a preset minimum number, detecting local extreme points within the subsequence, i.e., satisfying that the previous train headway time distance is less than the current train headway time distance and the current train headway time distance is greater than the next train headway time distance, or the previous train headway time distance is greater than the current train headway time distance and the current train headway time distance is less than the next train headway time distance, counting once for each such current train headway time distance to obtain the oscillation count of the sliding time window.

[0013] In a preferred embodiment of the present invention, when screening the time period in which the number of oscillations increases monotonically with time, the number of oscillations of each sliding time window is arranged in order of the start time of the sliding time window to obtain an oscillation number sequence. The longest continuous subsequence is selected such that the number of oscillations of the next oscillation is greater than or equal to the number of oscillations of the previous oscillation, and at least one position satisfies the greater than relationship. If there are multiple longest continuous subsequences of the same length, the one with the earliest start time is selected. The time period between the start time of the first sliding time window corresponding to the subsequence and the end time of the last sliding time window is recorded as the time period in which the number of oscillations increases monotonically.

[0014] In a preferred embodiment of the present invention, the transition period corresponding to the state change chain is a time period extending from the start time of the state change chain for a preset transition duration; the spatiotemporal superposition is as follows: the overlap duration of the time period with monotonically increasing oscillation frequency and the transition period is calculated; if the overlap duration is less than the product of the preset transition duration and the preset overlap ratio threshold, the time period with monotonically increasing oscillation frequency is discarded; otherwise, the retained time period with monotonically increasing oscillation frequency and the wave velocity disorder window are joined on the global time axis, and the longest continuous interval is selected from the union result as the accident-related precursor interval; if the union result is empty or the length of the longest continuous interval is less than the preset duration threshold, an empty set is output.

[0015] As a preferred embodiment of the present invention, when there is no state change chain within the wave velocity disorder window, or no headway time interval within a preset time after the start of the state change chain, or the number of headway time intervals in the headway time interval sequence is less than a preset minimum number, the subsequent steps are terminated and an empty set is output.

[0016] On the other hand, this invention also provides a traffic flow and accident correlation analysis system based on big data, including: a wave speed disturbance detection module, which records the signal light phase switching time and the first jump time of downstream cross-section traffic flow in each signal cycle, calculates the wave speed time difference of each signal cycle, and marks the corresponding time period as a wave speed disturbance window when the wave speed time difference shows a non-monotonic change in multiple consecutive signal cycles; a state time distance extraction module, which discretizes the road surface state into dry and wet categories within the wave speed disturbance window, generates a state change chain, and extracts the headway time distance sequence within a preset time period after the start of the state change chain; an oscillation monotonic filtering module, which counts the number of occurrences of local extreme points within the sliding time window from the headway time distance sequence as the oscillation number, and filters the time period in which the oscillation number monotonically increases with time; and an overlay precursor interval module, which spatiotemporally overlays the wave speed disturbance window with the time period in which the oscillation number monotonically increases, and uses the transition period corresponding to the state change chain to verify the validity of the time period in which the oscillation number monotonically increases, and outputs the overlaid continuous time period as the accident correlation precursor interval.

[0017] The beneficial effects of this invention are as follows: Compared with the prior art, the technical effects of this invention are as follows: This invention constructs a three-layer time-series detection architecture consisting of a wave velocity disorder window, a road surface wet / dry state change chain, and a monotonically increasing time interval of vehicle headway oscillation. It utilizes the state change transition period to verify the effectiveness of the oscillation increasing time interval before combining it with the wave velocity disorder window, thus achieving high-precision identification of accident precursor intervals in downstream road sections of signalized intersections. Compared with existing methods, this invention incorporates for the first time the causal coupling relationship between traffic wave propagation anomalies, road surface adhesion coefficient mutations, and micro-level car-following behavior into a unified analysis framework, significantly reducing the false alarm rate of single-feature warnings. Simultaneously, it adopts a union rather than intersection superposition strategy, effectively tolerating small offsets of each feature on the time axis, avoiding the loss of precursor signals due to single sensor positioning errors, and significantly improving the timeliness and robustness of warnings while maintaining a high recall rate. Attached Figure Description

[0018] Figure 1 This is a flowchart of the traffic flow and accident correlation analysis method based on big data described in this invention.

[0019] Figure 2 This is a structural diagram of a traffic flow and accident correlation analysis system based on big data, as described in this invention.

[0020] Figure 3 A schematic diagram of the structure of an electronic device for implementing the traffic flow and accident correlation analysis method based on big data in this embodiment of the invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0022] like Figure 1 As shown, the traffic flow and accident correlation analysis method based on big data described in this invention includes: S1: Record the signal light phase switching time and the first jump time of downstream cross-section flow in each signal cycle, calculate the wave velocity time difference in each signal cycle, and mark the corresponding time period as the wave velocity disorder window when the wave velocity time difference changes non-monotonicly in multiple consecutive signal cycles.

[0023] S1.1: In this embodiment, a roadside data acquisition unit is installed at the target signalized intersection. This unit is connected to the output interface of the signal controller via wired or wireless means. Using a global synchronization clock (such as a GPS timing module) as a reference, it reads the red light end time of each signal cycle in real time and records it as the signal light phase switching time, denoted as... subscript Number the signal period. =1,2,3,…

[0024] Simultaneously, a geomagnetic coil vehicle detector is installed at a cross-section L meters downstream of the stop line to continuously collect vehicle presence signals passing through this cross-section and automatically analyze the headway sequence between adjacent vehicles. Here, L is equal to the product of the reference travel time and average vehicle speed (typically 2 seconds) and the historical average vehicle speed of this road segment, and L is limited to between 50 meters and 200 meters; if the average vehicle speed is unknown or historical data is unavailable, L is defaulted to 100 meters. Once set, L remains constant during system operation. The wave velocity time difference, i.e., the time difference between the traffic light phase switching moment and the moment of the first surge in traffic flow at the downstream cross-section, essentially reflects the travel time required for the traffic wave to propagate from the stop line to the downstream cross-section. Given a fixed distance L, a smaller wave velocity time difference indicates faster traffic wave propagation; a larger wave velocity time difference indicates slower traffic wave propagation. Therefore, the relative change in the wave velocity time difference is directly equivalent to the relative change in the traffic wave velocity.

[0025] For each signal cycle Define the time interval as The headway of all trains falling within this interval is extracted. The time corresponding to the first traffic surge is recorded when the following conditions are met. In the headway time sequence of the current signal period i, find the first headway time in chronological order. , making ,in, As a preset ratio, this embodiment uses 0.5. Based on the engineering experience that the headway of the first wave of vehicles usually decreases to about half of the headway of the free flow vehicles after the initial wave dissipates, it can be adjusted within the range of 0.4 to 0.6 according to the actual traffic flow characteristics. This is the arithmetic mean of the headway distances of all vehicles in the previous signal cycle, requiring at least three vehicles to have passed in the previous cycle. Meanwhile, the vehicles found in the current cycle... The corresponding vehicle must be one of the first five vehicles in the current signal cycle to avoid false triggering due to fluctuations in distant vehicles. If there is no headway meeting the above conditions in the current cycle, no signal will be generated in this cycle. For the first signal cycle (i=1), since there is no average headway from the previous cycle, this cycle does not generate the first surge in traffic flow, nor does it participate in the wave velocity time difference calculation. If a valid average headway from the previous cycle cannot be obtained for several consecutive signal cycles, for example, due to the absence of vehicles at night, the system will continue to wait until the first valid cycle appears before starting the detection of the subsequent three consecutive valid cycles.

[0026] It should be noted that this invention uses the average headway of the previous cycle as a dynamic benchmark and limits the headway attenuation ratio, which can accurately identify the traffic flow change point in the discrete release of the first wave of vehicles caused by the activation of the traffic lights, and avoid false triggering caused by random noise or abnormal behavior of a single vehicle.

[0027] S1.2: For each record Given the signal period i (called the effective period), calculate the wave velocity time difference of that period. This difference represents the time it takes for the wavefront to propagate from the parking line to the downstream section. According to common knowledge, when the propagation distance L is constant, this time difference is inversely proportional to the wave velocity. If... If the time difference is abnormal, such as when the downstream detector is disturbed and the jump time is earlier than the green light activation time, it is discarded and not included in subsequent extreme value judgments; otherwise, the difference is retained as input for subsequent analysis. By eliminating non-positive wave velocity time differences, invalid data caused by detector false alarms or vehicle queue overflows are eliminated.

[0028] S1.3: Take three consecutive periods of the original signal, all of which are valid periods, and denote them as m-1, m, and m+1. The corresponding wave velocity time differences are respectively... Determine whether any of the following extreme value relationships are satisfied: Peak relationship: and Valley value relationship: and ;in, Tolerance thresholds are recommended to be 0.2 seconds or twice the standard deviation of wave velocity time differences over 10 consecutive stable cycles at the intersection. If the implementer has not explicitly set this threshold... The default is... =0.2 seconds. This invention does not limit the specific value and can be adjusted according to the noise level at the site. If any of the above relationships are satisfied, it is considered that the wave velocity time difference has undergone a non-monotonic change within three consecutive cycles.

[0029] If an invalid period appears in a continuous signal cycle, i.e., the moment of the first surge in flow cannot be obtained or the wave velocity time difference is ≤0, then the search for three consecutive valid periods starts again from the next valid period after the invalid period, and the previous part is not included in the extreme value judgment. For example, assuming that period numbers 1, 2, 3 are valid, 4 is invalid, and 5, 6, 7 are valid, then extreme value judgments are performed on (1,2,3) and (5,6,7) respectively, without crossing invalid periods for combination.

[0030] The above method uses local extrema over three consecutive cycles, rather than a simple rate-of-change threshold, to effectively distinguish between the gradual trend of normal traffic wave speeds and transient disturbances caused by sudden events. It should be noted that in reality, wave speed anomalies do not necessarily require only three cycles; this is used here as a simplified detection rule.

[0031] S1.4: When the above non-monotonic change is detected, the wave velocity disorder window will be adjusted. The time marked as the signal light phase switching time from the (m-1)th cycle At the time of signal light phase switching in the (m+1)th cycle The continuous time period between [time periods]. Also record the start time of this window. and end time If multiple separate wave velocity disturbance windows exist on the same time axis, the start and end times of each window are recorded for subsequent correlation analysis between road conditions and vehicle headway.

[0032] It should be noted that the length of the wave velocity disturbance window is typically 60–180 seconds, based on the start and end times of three consecutive effective cycles actually detected. When extracting the headway time-distance sequence in subsequent steps, if a longer analysis time window (e.g., 300 seconds) is required, it should be limited by the actual boundaries of the wave velocity disturbance window; a full 300-second window is not mandatory. 300 seconds is a typical time window for the pre-accident period, sufficient to capture the evolution of the headway time-distance oscillation.

[0033] S2: Within the wave velocity disorder window, the road surface state is discretized into two categories: dry and wet, generating a state change chain, and extracting the headway sequence within a preset time period after the start of the state change chain.

[0034] S2.1: Within the wave velocity disturbance window marked in step S1.4, the dielectric constant ε of the road surface material is continuously collected at fixed time intervals using a road surface dielectric constant sensor pre-embedded in the same downstream section. This sensor measures in real time based on frequency domain reflectometry or time domain reflectometry, and the measured value directly reflects the road surface moisture content. In this embodiment, a fixed time interval of 1 second is used, as the road surface humidity changes relatively slowly, and a 1-second sampling interval is sufficient to accurately capture the moment of dry-wet transition.

[0035] To convert a continuous analog quantity into a discrete state, two thresholds are preset: the first threshold... Second threshold The above values ​​are for illustrative purposes only. In actual deployment, they need to be calibrated through on-site experiments, for example, by using the mean value measured on a dry road surface plus two standard deviations. The mean value of waterlogged road surface measurements minus 2 standard deviations is used as... If calibration is not possible, a default value may be used, but it should be corrected in conjunction with local road surface materials. These adjustments are all within the scope of this invention. Specifically, the determination rules are as follows: If... If the road surface condition at that sampling time is then assigned the value "dry"; if If the value is less than 20, the value is assigned as "damp"; if the value is greater than 20, the value is assigned as "waterlogged".

[0036] The above three categories cover three typical road surface conditions: dry, wet (slippery but without water film), and water accumulation (with obvious water film or thin water layer). Among them, wet and water accumulation together constitute the generalized wet road surface, which is used for subsequent change chain analysis.

[0037] S2.2: Arrange the dry / wet / waterlogged labels in chronological order to form a discrete state sequence, and perform a sliding window comparison to find the first occurrence of the state change chain. The current time interval under consideration is the wave velocity disturbance window marked in step S1. If the wave velocity disorder window does not exist, for example, because no non-monotonic change was detected within three consecutive valid periods, the process will be terminated directly and an empty set will be output, and no further steps will be executed.

[0038] The state change chain is defined as a transition pattern between four adjacent states, such as dry → damp, dry → waterlogged, damp → dry, and waterlogged → dry. In this embodiment, if the states of two consecutive sampling points (with a 1-second interval) conform to any of the above patterns, it is determined to be a change chain. The absolute time of the first sampling point after the change occurs, i.e., the time when the new state first appears, is taken as the starting time of the state change chain, denoted as . .

[0039] If no of the above four state change chains are detected within the wave velocity disturbance window, such as the entire window being dry, or only switching repeatedly between wet and waterlogged without involving dryness, then it is determined that there is no road surface state change event in the current window, the process is terminated and an empty set is output, and no further steps are executed.

[0040] It should be noted that the above operation is intended to detect potential accident precursors caused by sudden changes in road surface humidity (dry-wet transition). If the road surface condition remains unchanged throughout the entire window, such as being consistently dry or consistently wet / waterlogged, no precursor interval will be output due to the lack of sudden changes in the road surface adhesion coefficient, which falls within the scope of the design.

[0041] The above operations are intended to detect potential accident precursors caused by sudden changes in road surface humidity. If the wave speed disorder window is not present, it indicates that no traffic wave speed abnormality has been detected, and no further analysis is performed in this embodiment of the invention.

[0042] S2.3: From the start time of the state change chain Begin by selecting a time interval of a preset duration, which does not exceed the end time of the wave velocity disturbance window. The difference; if the remaining duration of the wave velocity disturbance window is less than 300 seconds, the end time of the wave velocity disturbance window is used as the truncation endpoint. In this embodiment, the preset duration is normally 300 seconds, but if the remaining duration of the wave velocity disturbance window is less than 300 seconds, the remaining duration is taken. Within this time interval, the original vehicle headway data collected by the same geomagnetic coil vehicle detector is retrieved. Each vehicle headway is defined. start time This refers to the absolute moment when the front end of the following vehicle arrives at the detection section, corresponding to the time difference between the departure of the rear edge of the preceding vehicle and the arrival of the front edge of the following vehicle. This represents the total number of train headway times in the headway sequence. Specifically, only those that satisfy the following conditions are retained. The headway distance constitutes the headway sequence. And simultaneously record each corresponding If no vehicles pass through the interval (i.e., M=0), it is determined that subsequent oscillation counts cannot be performed, the process is terminated, and an empty set is output.

[0043] After extracting the headway time sequence, calculate the total time span of the sequence. .like If the time interval is less than the window length of the sliding time window, it is determined that no valid sliding time window can be constructed, the process is immediately terminated, and an empty set is output. Furthermore, if the number of extracted headway times M is less than 3, the process is also terminated.

[0044] It should be noted that anchoring the time window for extracting the headway to the starting point of the road condition change chain can ensure the causal alignment of traffic flow characteristics and sudden changes in road condition over time.

[0045] S3: Count the number of local extreme points within the sliding time window from the time distance sequence of the vehicle head as the number of oscillations, and filter the time period in which the number of oscillations increases monotonically with time.

[0046] S3.1: If satisfied ,and Window length greater than or equal to the sliding time window Construct an absolute timeline with the left boundary as... The right boundary is That is, the end time of the last locomotive headway.

[0047] Furthermore, the window length of the sliding time window is In this embodiment, 10 seconds is taken, with a step size of [missing value]. In this embodiment, a 2-second window is used. It should be noted that a 10-second window length can cover about 3 to 5 vehicles, which is sufficient to statistically analyze local extreme values; the 2-second step size ensures the continuity between sliding windows.

[0048] From the left boundary of the time axis Begin by generating the first... The coverage area of ​​each sliding time window is ,in, And satisfy That is, the window does not extend beyond the right boundary. For each window, filter out all headway values ​​that meet one of the following conditions. Start time Located within the window interval, end time Located within the window interval, meaning it overlaps with the window in terms of time and start time. The window start time must be less than or equal to the window end time, meaning the headway completely covers the entire window. These... Arranged in chronological order, forming the first... Subsequence of windows .

[0049] By anchoring the train headway sequence to the absolute time axis and sliding the time window with a fixed step size, the time alignment and translation invariance of the oscillation count statistics can be ensured.

[0050] S3.2: For the first Subsequences corresponding to each sliding time window Let the number of times the front of the car is included be . .like If a valid predecessor-current-successor triple cannot be formed within the window, the oscillation count of this window is directly reduced. Assign a value of 0. If Then, local extremum detection is performed within the subsequence: for subsequences with index , , Headway ,in The global index of the locomotive's time distance in the original sequence H is used to compare it with the time distance of the immediately preceding locomotive. Time difference with the next car The size relationship. If one of the following two conditions is met: Peak conditions: Valley condition: Then determine This is a local extremum point, and the number of oscillations within this window is... Increment by 1, where the two boundary points of the subsequence (the first and last elements) are not included in the extreme value judgment. After traversing all intermediate elements in the subsequence, the number of oscillations of the window is obtained. (A non-negative integer).

[0051] The oscillation frequency here refers to the total frequency of peaks and troughs in the headway time sequence within the sliding time window, used to quantify the severity of headway fluctuations. Although called oscillation, it does not require the headway time to exhibit periodic back-and-forth changes; as long as there are continuous rises and falls, it is counted. This definition is consistent throughout the text and should not be understood using the general physical meaning of oscillation.

[0052] It can be seen that the present invention uses local extreme points rather than amplitude thresholds to quantify the oscillation behavior of the headway, and can adapt to the fluctuation characteristics under different traffic flow levels without parameters.

[0053] S3.3: Arrange all sliding time windows in chronological order of their start times to obtain the oscillation sequence. Each Corresponding to the A window. Find all continuous subsequences that satisfy the non-decreasing condition, that is, for any two consecutive elements in the subsequence... and ,Require Furthermore, in this subsequence, there exists at least one pair of adjacent elements that satisfy... This is to exclude completely flat sequences.

[0054] Extract all such continuous subsequences from the sequence and compare the lengths of each subsequence (i.e., the number of windows it contains). Select the longest continuous subsequence; if multiple longest continuous subsequences of the same length exist, select the one with the smallest starting window index (i.e., the earliest start time). Let the start time of the first window corresponding to the selected subsequence be denoted as . The end time of the last window is Then the absolute time period covered by the subsequence The period marked as a monotonically increasing oscillation number is denoted as... If the entire sequence If there are no strictly monotonically increasing continuous subsequences of length 2 or longer (i.e., the longest subsequence is 1), then it is determined that there is no time interval with a monotonically increasing number of oscillations. If marked as empty, this time period will be ignored in subsequent step S4. If there is only one sliding time window and its oscillation count is greater than 0, whether to use it as a monotonically increasing time period can be determined by the implementer based on the application scenario; this invention does not adopt this approach by default.

[0055] It should be noted that, by screening the longest continuous subsequence with strictly monotonically increasing oscillation count, the embodiments of the present invention can identify the key transformation process of traffic flow from disordered fluctuations to frequent oscillations. Moreover, this feature is highly consistent with the physical process of drivers gradually losing stable control of vehicle speed on slippery roads. Compared with simple threshold or mean changes, it can provide a better early warning of potential accident risks.

[0056] S4: Spatiotemporally superimpose the wave velocity disorder window with the monotonically increasing oscillation number time period, and use the transition period corresponding to the state change chain to verify the validity of the monotonically increasing oscillation number time period, and output the superimposed continuous time period as the accident-related precursor interval.

[0057] S4.1: Read the start and end times of the marked wave speed disturbance window from step S1. This window corresponds to the time period defined by the phase switching times numbered m-1, m, and m+1 of three consecutive valid signal cycles, reflecting the unstable period of traffic flow caused by the non-monotonic change in traffic wave speed.

[0058] S4.2: Starting from the initial time of the state change chain obtained in step S2, continue for a preset transition time. Define the transition period for: ; This transition period does not directly participate in the union of time intervals, but is used to verify the validity of periods with a monotonically increasing number of oscillations. In this embodiment, 60 seconds is used because the impact on following stability is most significant within the first 60 seconds after the road surface changes from dry to wet. This value can be adjusted according to actual needs; for example, it is recommended to increase the value appropriately at higher speeds and use a smaller value at lower speeds. This invention only provides an example.

[0059] S4.3: Obtain the monotonically increasing oscillation period from step S3. If it is empty, it is determined that there is no valid oscillating increasing feature, the process is terminated and an empty set is output; otherwise, continue to execute S4.4.

[0060] S4.4: If If not empty, then calculate the transition period. Overlap duration: ; in, This represents the starting point of a period in which the number of oscillations is monotonically increasing. This represents the end of a period in which the number of oscillations is monotonically increasing. This marks the beginning of the transition period. This marks the end of the transition period.

[0061] like If the period of monotonically increasing oscillation frequency is deemed insufficiently correlated with the transition period of road surface condition changes, it is discarded (considered invalid) and not included in subsequent superposition. To preset the overlap ratio threshold, this embodiment sets it to 0.5, which is set according to the statistical calculation of the overlap ratio between the effective precursor interval and the transition period in historical data.

[0062] After the above screening, if there are multiple retained periods of monotonically increasing oscillation frequency (i.e., multiple non-overlapping periods)... (interval), then each Perform a union operation with the wave velocity disorder window; if only one is retained... Then, a union operation is performed with the wave velocity disorder window. The specific union rule is as follows: The union of the two time intervals involved in the operation (each interval being a continuous interval) is calculated on the global time axis. If there is overlap or adjacency (interval ≤ 2 seconds is considered adjacency), they are merged into a single continuous interval. All union results (which may come from multiple...) are then processed. ), select the longest continuous interval as If all unions are empty, then Empty.

[0063] It's important to note that the technical intent behind first calculating the overlap between the oscillation-increasing period and the transition period, and then performing a union operation with the wave velocity disorder window, is that the transition period—a time following a change in road conditions—is a high-probability period during which driving behavior may be affected. If the oscillation-increasing period does not sufficiently overlap with this transition period, it indicates that the oscillating behavior is likely not caused by the current change in road conditions, and therefore it is discarded. The subsequent union operation is necessary because there may be slight misalignments between the wave velocity disorder window and the retained oscillation-increasing period on the actual timeline. For example, the wave velocity disorder window might end slightly earlier than the oscillation-increasing period, or the two might alternate. Using union can cover a more complete precursor interval, avoiding interval breaks due to time positioning errors. These two methods are not contradictory; overlap verification filters causal relationships, while the union operation tolerates slight timeline shifts, jointly improving the robustness of the warning.

[0064] S4.5: If the union result is empty, or the length of the longest continuous interval is less than the preset duration threshold, it is determined that there are no statistically significant accident-related precursors in the current detection period, and an empty set is output, indicating no precursors; otherwise, the longest continuous interval is output as the accident-related precursor interval. This output can be used to trigger the early warning system, such as sending warning information to roadside variable message signs or broadcasting safety reminders to connected vehicles.

[0065] In this embodiment, the preset duration threshold is set to 30 seconds, which is the lower quartile of the historical incident precursor interval length. This threshold is used to filter out excessively short, occasional overlapping intervals to avoid false alarms. In actual deployment, it can be calibrated based on historical incident data, with a recommended range of 10 to 120 seconds.

[0066] The threshold values ​​mentioned in the above embodiments are exemplary values ​​and need to be calibrated according to actual needs in practical applications. If no calibration conditions are available, the default values ​​provided by this invention can be used, but it should be recognized that the default values ​​may lead to performance degradation in different scenarios. Those skilled in the art can achieve adaptive parameter adjustment based on the logical framework disclosed in this invention, combined with conventional parameter optimization methods (such as grid search and cross-validation), without any creative effort.

[0067] Furthermore, in practical deployment, the method described in this invention is typically deployed as a continuously running online analysis module. After each signal cycle of data acquisition and processing is completed, regardless of whether the final output is an accident-related precursor interval or an empty set, the system automatically slides to the next signal cycle and repeats steps S1 to S4. When the output is an empty set, it indicates that within the current analysis time window, no coupling feature simultaneously satisfying the three conditions of wave velocity turbulence, road surface condition change, and monotonically increasing headway turning frequency was detected. However, this does not indicate a system malfunction or absolute traffic safety; the system will remain in monitoring mode and continue processing subsequent cycles.

[0068] If data loss occurs, such as due to prolonged absence of a vehicle in the downstream detector or a malfunction in the dielectric constant sensor, resulting in the inability to acquire a valid wave velocity time difference for N consecutive signal cycles (i.e., the inability to generate a jump moment in multiple consecutive cycles in step S1), the system can enter an idle waiting mode. Once the first valid cycle is detected, the system will automatically resume the complete three-layer detection process. The value of N can be set according to actual needs; a default value of N=10 is recommended.

[0069] The output accident-related precursor intervals can be used to trigger downstream early warning systems (such as displaying warning information on roadside variable message signs, broadcasting to connected vehicles, etc.). After the warning is triggered, the system does not require additional reset and continues to operate according to the signal cycle.

[0070] like Figure 2 As shown, the present invention also provides a traffic flow and accident correlation analysis system based on big data, comprising: The wave velocity disturbance detection module records the signal light phase switching time and the first jump time of the downstream section flow in each signal cycle, calculates the wave velocity time difference in each signal cycle, and marks the corresponding time period as the wave velocity disturbance window when the wave velocity time difference changes non-monotonicly in multiple consecutive signal cycles. The state-time distance extraction module discretizes the road surface state into two categories, dry and wet, within the wave velocity disorder window, generates a state change chain, and extracts the vehicle headway sequence within a preset time period after the start of the state change chain. The oscillation monotonic filtering module counts the number of local extreme points within a sliding time window from the vehicle headway time distance sequence as the oscillation count, and filters the time period in which the oscillation count monotonically increases with time. The superimposed precursor interval module spatiotemporally superimposes the wave velocity disorder window with the monotonically increasing oscillation number time period, and uses the transition period corresponding to the state change chain to verify the validity of the monotonically increasing oscillation number time period, and outputs the superimposed continuous time period as the accident-related precursor interval.

[0071] The traffic flow and accident correlation analysis device based on big data provided in this embodiment of the invention can execute the traffic flow and accident correlation analysis method based on big data provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0072] Figure 3This is a schematic diagram of an electronic device for implementing the traffic flow and accident correlation analysis method based on big data, as described in this embodiment of the invention. The electronic device is intended to represent various forms of digital computers, such as laptops, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0073] like Figure 3 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0074] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0075] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as traffic flow and accident correlation analysis methods based on big data.

[0076] In some embodiments, the big data-based traffic flow and accident correlation analysis method can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the big data-based traffic flow and accident correlation analysis method described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the big data-based traffic flow and accident correlation analysis method by any other suitable means (e.g., by means of firmware).

[0077] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0078] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0079] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0080] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0081] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0082] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0083] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0084] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for analyzing the correlation between traffic flow and accidents based on big data, characterized in that, include: Record the signal light phase switching time and the first jump in downstream cross-sectional flow rate for each signal cycle, calculate the wave velocity time difference for each signal cycle, and mark the corresponding time period as the wave velocity disorder window when the wave velocity time difference changes non-monotonically within multiple consecutive signal cycles. Within the wave velocity disorder window, the road surface state is discretized into two categories: dry and wet, generating a state change chain, and extracting the headway sequence within a preset time period after the start of the state change chain. The time extraction method after the start of the state change chain is as follows: search for the first occurrence of the state change chain within the wave velocity disorder window. If no state change chain exists, terminate the process and output an empty set; otherwise, take the time of the first sampling point after the state change occurs as the start time of the state change chain; extract a time interval of a preset duration from the start time of the state change chain, extract all the headway distances with start times within the time interval from the downstream section detection data to form a headway distance sequence, and record the start time of each headway distance; The number of occurrences of local extreme points within the sliding time window in the time distance sequence of the vehicle head is counted as the number of oscillations, and the time period in which the number of oscillations increases monotonically with time is selected. The wave velocity disorder window is spatiotemporally superimposed with the time period of monotonically increasing oscillation number, and the validity of the time period of monotonically increasing oscillation number is verified by using the transition period corresponding to the state change chain. The superimposed continuous time period is output as the accident-related precursor interval. The transition period corresponding to the state change chain is a time period extending from the start time of the state change chain for a preset transition duration; the spatiotemporal superposition is as follows: calculate the overlap duration between the time period of monotonically increasing oscillation frequency and the transition period. If the overlap duration is less than the product of the preset transition duration and the preset overlap ratio threshold, then discard the time period of monotonically increasing oscillation frequency; otherwise, find the union of the retained time period of monotonically increasing oscillation frequency and the wave velocity disorder window on the global time axis, and select the longest continuous interval from the union result as the accident-related precursor interval; If, after validity verification, there is no time period with monotonically increasing oscillation count that is retained, or if the length of the longest continuous interval is less than the preset duration threshold, then an empty set is output.

2. The method for analyzing the correlation between traffic flow and accidents based on big data according to claim 1, characterized in that, The wave velocity time difference is only retained when the first surge in flow rate is later than the signal light phase switching time. Take three signal periods with consecutive original signal period numbers and retain the wave velocity time difference in each, and determine whether there is an extreme value relationship between the wave velocity time difference of the three signal periods, where the middle signal period is greater than the two adjacent signal periods or the middle signal period is less than the two adjacent signal periods. When an extreme value relationship exists, the time period between the signal light phase switching time of the first signal cycle and the signal light phase switching time of the third signal cycle is marked as the wave speed disturbance window.

3. The method for analyzing the correlation between traffic flow and accidents based on big data according to claim 1, characterized in that, The discretization of road surface conditions into two categories, dry and wet, includes: Dielectric constants are collected at fixed time intervals within the wave velocity disturbance window. The dielectric constant is judged as dry if it is below the first threshold, as wet if it is between the first and second thresholds, and as water accumulation if it is above the second threshold. Dampness and standing water are collectively referred to as wet road surfaces; The process of changing from dry to wet, from dry to waterlogged, from wet to dry, and from waterlogged to dry is defined as a state change chain.

4. The traffic flow and accident correlation analysis method based on big data according to claim 2, characterized in that, The method for obtaining the moment of the first jump in traffic is as follows: The time interval for each signal cycle is from the moment the signal light phase changes in the current signal cycle to the moment the signal light phase changes in the next signal cycle. Extract the headway of all vehicles that fall within the corresponding time interval. The arithmetic mean of all headway distances within the signal period is calculated as the average headway distance. For a signal cycle in which the average headway of the previous signal cycle is valid, the first headway in the current signal cycle that is less than a preset proportion of the average headway of the previous signal cycle is found in chronological order, and the end time of the headway is taken as the time of the first surge in traffic.

5. The traffic flow and accident correlation analysis method based on big data according to claim 4, characterized in that, The number of occurrences of local extreme points within the statistical sliding time window includes: Using the start time of the first locomotive time distance in the locomotive time distance sequence as the left boundary and the end time of the last locomotive time distance as the right boundary, a sliding time window with a window length of the first window length and a step size of the first step length is set, and the window slides sequentially from the left. For each sliding time window, all train headway times that overlap with the time interval covered by the sliding time window are selected to form a subsequence. When the number of elements in the subsequence is greater than or equal to the preset minimum number, local extrema are detected within the subsequence, i.e., the previous train headway time is less than the current train headway time and the current train headway time is greater than the next train headway time, or the previous train headway time is greater than the current train headway time and the current train headway time is less than the next train headway time. Each such current train headway time is counted once to obtain the number of oscillations of the sliding time window.

6. The traffic flow and accident correlation analysis method based on big data according to claim 5, characterized in that, When screening for time periods where the number of oscillations increases monotonically with time, the oscillation counts of each sliding time window are arranged in order of their start times to obtain an oscillation count sequence. The longest continuous subsequence is selected where the number of oscillations in each subsequent subsequence is greater than or equal to the number of oscillations in the previous subsequence, and at least one position satisfies the greater than relationship. If there are multiple longest continuous subsequences of the same length, the one with the earliest start time is selected. The time period between the start time of the first sliding time window corresponding to the subsequence and the end time of the last sliding time window is recorded as the time period where the number of oscillations increases monotonically.

7. The method for traffic flow and accident correlation analysis based on big data according to claim 1, characterized in that, When there is no state change chain within the wave velocity disorder window, or no train headway within a preset time after the start of the state change chain, or the number of train headway times in the train headway sequence is less than a preset minimum number, the subsequent steps are terminated and an empty set is output.

8. A traffic flow and accident correlation analysis system based on big data, based on the traffic flow and accident correlation analysis method based on big data as described in any one of claims 1 to 7, characterized in that: Also includes: The wave velocity disturbance detection module records the signal light phase switching time and the first jump time of the downstream section flow in each signal cycle, calculates the wave velocity time difference in each signal cycle, and marks the corresponding time period as the wave velocity disturbance window when the wave velocity time difference changes non-monotonicly in multiple consecutive signal cycles. The state-time distance extraction module discretizes the road surface state into two categories, dry and wet, within the wave velocity disorder window, generates a state change chain, and extracts the vehicle headway sequence within a preset time period after the start of the state change chain. The oscillation monotonic filtering module counts the number of local extreme points within a sliding time window from the vehicle headway time distance sequence as the oscillation count, and filters the time period in which the oscillation count monotonically increases with time. The superimposed precursor interval module spatiotemporally superimposes the wave velocity disorder window with the monotonically increasing oscillation number time period, and uses the transition period corresponding to the state change chain to verify the validity of the monotonically increasing oscillation number time period, and outputs the superimposed continuous time period as the accident-related precursor interval.

Images