Intelligent identification method for continuous longitudinal slope traffic risk elements based on deep learning

Abstract

The present application relates to the technical field of traffic data processing, in particular to a continuous longitudinal slope traffic risk factor intelligent identification method based on deep learning. The longitudinal slope characteristics of the continuous longitudinal slope are obtained, and the brake characteristics are identified according to the monitoring data of the vehicle brake in the continuous longitudinal slope. A basic traffic risk factor identifier is obtained, and the incremental learning data configuration is carried out according to the brake time and the longitudinal slope characteristics, and the incremental learning data is obtained. According to the brake frequency distribution, the incremental learning data is used for incremental learning of the basic traffic risk factor identifier, and a plurality of incremental identifiers are obtained. The vehicle characteristics are collected, combined with the longitudinal slope characteristics, and based on the plurality of incremental identifiers, the longitudinal slope traffic risk factor information is identified and output. The characteristics of different continuous longitudinal slope sections can be effectively adapted, and the longitudinal slope traffic risk factors can be comprehensively and accurately identified.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a continuous longitudinal slope traffic risk factor intelligent identification method based on deep learning. BACKGROUND

[0002] Traditional continuous longitudinal slope traffic risk factor identification technology mostly relies on artificial field investigation combined with historical accident data statistics, or uses a simple machine learning model and trains a small amount of fixed sample data. However, these technologies have obvious deficiencies. They ignore the dynamic behavior characteristics of vehicle driving, resulting in insufficient data utilization. The simple machine learning model is difficult to adapt to different characteristic road sections due to fixed training data, and has poor adaptability. Moreover, it is also affected by artificial subjective factors, data lag, and insufficient robustness of single models, resulting in low reliability of risk factor identification results and difficulty in meeting the demand for accurate identification. SUMMARY

[0003] The present application provides a continuous longitudinal slope traffic risk factor intelligent identification method based on deep learning to solve the technical problem that the prior art cannot meet the demand for accurate identification.

[0004] The technical solution of the present application to solve the above technical problems is as follows:

[0005] In a first aspect, the present application provides a continuous longitudinal slope traffic risk factor intelligent identification method based on deep learning, comprising: obtaining longitudinal slope characteristics of a continuous longitudinal slope, and identifying brake characteristics according to monitoring data of vehicle braking in the continuous longitudinal slope, wherein the brake characteristics include brake frequency distribution and brake time; obtaining a basic traffic risk factor identifier, incrementally learning data configuration according to the brake time and the longitudinal slope characteristics, and obtaining incrementally learning data; incrementally learning the basic traffic risk factor identifier using the incrementally learning data according to the brake frequency distribution, and obtaining a plurality of incrementally learning identifiers; collecting vehicle characteristics, combining the longitudinal slope characteristics, and identifying and outputting longitudinal slope traffic risk factor information based on the plurality of incrementally learning identifiers.

[0006] Optionally, obtaining longitudinal slope characteristics of a continuous longitudinal slope and identifying brake characteristics according to monitoring data of vehicle braking in the continuous longitudinal slope, comprises: obtaining longitudinal slope characteristics of a continuous longitudinal slope, wherein the longitudinal slope characteristics include longitudinal slope gradient and longitudinal slope length; obtaining vehicle braking monitoring data in a preset time range in the continuous longitudinal slope, identifying brake time sequence, and obtaining a historical brake time sequence set; calculating a plurality of average frequencies and average total time of braking according to the historical brake time sequence set, and obtaining brake frequency distribution and brake time as brake characteristics.

[0007] Optionally, the obtaining the basic traffic risk element identifier comprises: collecting sample longitudinal slope feature sets and sample vehicle feature sets according to safety monitoring data of a plurality of sample longitudinal slopes, and collecting proportions of longitudinal slope risk events occurring under different sample longitudinal slope features and sample vehicle features, and labeling as sample longitudinal slope traffic risk element information sets; constructing a basic traffic risk element identifier based on machine learning; and supervising training of the basic traffic risk element identifier using the sample longitudinal slope feature sets, the sample vehicle feature sets, and the sample longitudinal slope traffic risk element information sets, and completing the training after a test accuracy rate meets an accuracy rate threshold to obtain a trained and tested basic traffic risk element identifier.

[0008] Optionally, the incremental learning data configuration is performed according to the brake time and the longitudinal slope feature to obtain incremental learning data, comprising: obtaining a preset incremental learning data amount; obtaining average brake times of a plurality of sample longitudinal slopes in monitoring processing; correcting the preset incremental learning data amount according to a ratio of the brake time and the average brake time to obtain an incremental learning data amount; collecting an incremental longitudinal slope feature set, an incremental vehicle feature set, and an incremental longitudinal slope traffic risk element information set of a continuous longitudinal slope with the greatest similarity to the longitudinal slope feature according to the incremental learning data amount until the incremental learning data amount is met to obtain the incremental learning data.

[0009] Optionally, the incremental learning is performed on the basic traffic risk element identifier using the incremental learning data according to the brake frequency distribution to obtain a plurality of incremental identifiers, comprising: constructing a brake frequency interval according to a minimum brake frequency and a maximum brake frequency in the brake frequency distribution; processing average brake frequency intervals according to monitoring data of a plurality of sample longitudinal slopes; obtaining a preset incremental learning amount; correcting and calculating the preset incremental learning amount according to an interval length ratio of the brake frequency interval and the average brake frequency interval to obtain an incremental learning amount; and performing incremental learning on the basic traffic risk element identifier using the incremental learning data according to the incremental learning amount to obtain a plurality of incremental identifiers.

[0010] According to the incremental learning amount, the incremental learning is performed on the basic traffic risk element identifier using the incremental learning data to obtain a plurality of incremental identifiers, comprising: performing random data division a plurality of times in the incremental learning data according to the incremental learning amount as a total number of times to obtain a plurality of random incremental data; and performing incremental learning training on the basic traffic risk element identifier using the plurality of random incremental data respectively to obtain a plurality of incremental identifiers.

[0011] Optionally, the vehicle features are collected, combined with the longitudinal slope features, and based on a plurality of incremental identifiers, the longitudinal slope traffic risk element information is identified and output, including: acquiring the vehicle features of the current passing vehicle through the road monitoring system, wherein the vehicle features include the vehicle length; inputting the vehicle features and the longitudinal slope features into a plurality of incremental identifiers, identifying and outputting a plurality of identified longitudinal slope traffic risk element information, and fusing and processing to obtain the longitudinal slope traffic risk element information.

[0012] By implementing the present application, the longitudinal slope features of the continuous longitudinal slope are obtained, and the brake features are identified and obtained according to the monitoring data of the vehicle brake in the continuous longitudinal slope, wherein the brake features include brake frequency distribution and brake time. The longitudinal slope features provide basic road environment information for subsequent traffic risk identification. Different longitudinal slope gradients and lengths directly affect the vehicle driving state and risk level. The brake features can intuitively reflect the driving safety of the vehicle on the continuous longitudinal slope section. The brake frequency distribution reflects the intensity of the vehicle brake behavior, and the brake time reflects the duration of the brake behavior. The combination of the two provides key vehicle driving behavior data support for subsequent risk identification, making the risk identification more targeted.

[0013] The basic traffic risk element identifier is obtained, the incremental learning data is configured according to the brake time and the longitudinal slope features, and the incremental learning data is obtained. The basic traffic risk element identifier provides a stable initial model framework for subsequent incremental learning. It is trained based on a large number of samples and has certain general risk identification ability. By correcting the incremental learning data amount and collecting similar data through the brake time and the longitudinal slope features, the incremental learning data can better adapt to the actual situation of the current continuous longitudinal slope, avoiding data redundancy or deficiency, improving the efficiency and accuracy of subsequent incremental learning, and making the model adapt to the characteristics of the current section faster.

[0014] According to the brake frequency distribution, the incremental learning data is used to perform incremental learning on the basic traffic risk element identifier to obtain a plurality of incremental identifiers. The incremental learning amount is corrected based on the brake frequency distribution, which can match the number of incremental learning with the complexity of the vehicle brake behavior of the current section. When the brake frequency fluctuates greatly, the learning times can be appropriately increased to ensure the sufficiency of model learning. Multiple incremental identifiers are obtained by training the data through multiple random divisions, which avoids the contingency that may be caused by single training data division, enhances the robustness of the model, and improves the identification ability of the model for complex traffic risks.

[0015] The vehicle features are collected, combined with the longitudinal slope features, and based on multiple incremental identifiers, longitudinal slope traffic risk element information is obtained through identification output. The introduction of the vehicle features supplements the vehicle dimension information of traffic risk identification. Different lengths of vehicles face different risks when driving on continuous longitudinal slopes. Combined with the longitudinal slope features, the risk can be more comprehensively evaluated. The identification results of the multiple incremental identifiers are fused, the advantages of each identifier are integrated, the misjudgment or omission of a single identifier is reduced, and the reliability and accuracy of the longitudinal slope traffic risk element identification result are significantly improved, providing more accurate basis for traffic management and safety decision.

[0016] In summary, by implementing the present application, the characteristics of different continuous longitudinal slope sections can be effectively adapted, and the longitudinal slope traffic risk elements can be comprehensively and accurately identified, providing reliable data support and decision basis for road traffic safety management and control. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A flowchart of a continuous longitudinal slope traffic risk element intelligent identification method based on deep learning provided by the present application is shown.

[0018] Figure 2 In the continuous longitudinal slope traffic risk element intelligent identification method based on deep learning provided by the present application, a flowchart of incremental learning data for incremental learning of the basic traffic risk element identifier is shown. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0020] In the description of the present application, the terms "first" and "second" are used only for description purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features with "first" and "second" can explicitly or implicitly include one or more features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0021] In the description of the present application, the term "for example" is used to mean "serving as an example, instance, or illustration." Any embodiment described as "for example" in the present application is not necessarily to be construed as preferred or advantageous over other embodiments. The following description is presented to enable any person skilled in the art to make and use the application. In the following description, for the purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the present application. It will be apparent, however, to one skilled in the art that the present application can be practiced without using these specific details. In other instances, well-known structures and processes are not elaborated upon in order to avoid unnecessary detail, which can obscure the description of the present application. Thus, the present application is not intended to be limited by the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed herein.

[0022] As shown in Embodiment I, Figure 1 The embodiment of the present application provides a deep learning-based continuous longitudinal slope traffic risk factor intelligent identification method, which comprises the following steps:

[0023] S100: Obtain the longitudinal slope characteristics of the continuous longitudinal slope, and identify the brake characteristics according to the monitoring data of the vehicle brake in the continuous longitudinal slope, wherein the brake characteristics include brake frequency distribution and brake time;

[0024] S200: Obtain a basic traffic risk factor identifier, configure incremental learning data according to the brake time and the longitudinal slope characteristics, and obtain the incremental learning data;

[0025] S300: According to the brake frequency distribution, the incremental learning data is used to perform incremental learning on the basic traffic risk factor identifier, and a plurality of incremental identifiers are obtained;

[0026] S400: Collect vehicle characteristics, combine the longitudinal slope characteristics, and identify and output longitudinal slope traffic risk factor information based on the plurality of incremental identifiers.

[0027] In step S100 of the embodiment of the present application, the longitudinal slope characteristics of the continuous longitudinal slope are obtained, and the brake characteristics are identified according to the monitoring data of the vehicle brake in the continuous longitudinal slope, which comprises the following steps:

[0028] Obtain the longitudinal slope characteristics of the continuous longitudinal slope, wherein the longitudinal slope characteristics include the longitudinal slope gradient and the longitudinal slope length;

[0029] Obtain the vehicle brake monitoring data in the continuous longitudinal slope within a preset time range, identify the brake time sequence, and obtain a historical brake time sequence set;

[0030] According to the historical brake time sequence set, a plurality of average frequencies and average total time lengths of the brake are calculated, and the brake frequency distribution and the brake time are obtained as the brake characteristics.

[0031] In the embodiments of the present application, the purpose of step S100 is to provide comprehensive and key basic data support for subsequent continuous longitudinal slope traffic risk factor identification, specifically including two aspects: one is to clearly define the core road geometric properties of the continuous longitudinal slope, to define the basic environmental boundary for risk identification; the second is to extract the key driving behavior characteristics of the vehicle on the road section, which directly reflects the real-time safety state of the road section, and the combination of the two makes the subsequent risk identification more targeted and accurate.

[0032] Specifically, the longitudinal slope characteristics of the continuous longitudinal slope, i.e., the longitudinal slope gradient and the longitudinal slope length, need to be obtained. The longitudinal slope gradient reflects the inclination degree of the road section, and the longitudinal slope length reflects the extension range of the continuous inclined road section, and these two parameters are the basis for determining the risk-related factors such as vehicle driving resistance and braking load on the road section. The longitudinal slope characteristics of the continuous longitudinal slope can be obtained through road design drawings and on-site measurement. For example, assuming that there is a continuous longitudinal slope road section A on a mountainous highway, the longitudinal slope characteristics of the A road section can be: the road section includes 3 continuous slopes, the first slope is 5%, the second slope is 7%, and the third slope is 4%, the first slope length is 800 meters, the second slope length is 600 meters, and the third slope length is 1000 meters, and the total length of the continuous longitudinal slope is 2400 meters.

[0033] Further, vehicle brake monitoring data in a preset time range in the continuous longitudinal slope needs to be obtained, the brake time sequence is identified, and a historical brake time sequence set is obtained. That is, vehicle brake monitoring data in a preset time range, such as 1 month or 3 months, in the continuous longitudinal slope is selected, the specific time of each brake occurrence is identified from the data, and a historical brake time sequence set is formed to record the time distribution of vehicle braking behavior on the road section.

[0034] Specifically, in the above example, the high-definition camera, vehicle speed induction coil and other vehicle monitoring equipment arranged along the A road section can be used to collect brake monitoring data of all passing vehicles in the past 3 months, identify the specific time of each brake occurrence, and form a historical brake time sequence set. For example, on January 5, 8:12:30, a truck with license plate number GanA・XXXX1 brakes on the first slope, and the brake time is 3s; on January 5, 8:15:10, a car with license plate number GanA・XXXX2 brakes on the second slope, and the brake time is 4s, etc. The historical brake time sequence set is formed to record the time distribution of vehicle braking behavior on the road section.

[0035] Further, according to the historical brake time sequence set, a plurality of average frequencies and average total time of braking are calculated to obtain brake frequency distribution and brake time as brake characteristics.

[0036] Optionally, in the above example, the average frequency of braking can be calculated according to the slope segments. The first segment of the slope has an average of 28 brake events per day, the second segment of the slope has an average of 35 brake events per day, and the third segment of the slope has an average of 18 brake events per day, forming a "brake frequency distribution of slopes with different slopes". In another embodiment, the average frequency of braking can also be calculated according to time periods, i.e. in the A section, the average number of brake events per hour during the morning peak (7:00-9:00) is 32, the average number of brake events per hour during the flat peak (9:00-17:00) is 15, and the average number of brake events per hour during the evening peak (17:00-19:00) is 28, forming a "brake frequency distribution of different time periods".

[0037] Then, the duration of all braking behaviors is counted, and the average total duration is calculated as a brake time feature. For example, the average duration of each braking event of a truck on this section is 4.2 seconds, the average duration of each braking event of a car is 2.1 seconds, and the average total duration of all braking events of all vehicles is 3.0 seconds, which is the "brake time" feature of this section.

[0038] In step S200 of the embodiments of the present application, a basic traffic risk factor identifier is obtained, including:

[0039] According to the safety monitoring data of the plurality of sample longitudinal slopes, a sample longitudinal slope feature set and a sample vehicle feature set are collected, and the proportion of longitudinal slope risk events occurring under different sample longitudinal slope features and sample vehicle features is collected and labeled as a sample longitudinal slope traffic risk factor information set;

[0040] Based on machine learning, a basic traffic risk factor identifier is constructed;

[0041] The sample longitudinal slope feature set, the sample vehicle feature set, and the sample longitudinal slope traffic risk factor information set are used to supervise the training of the basic traffic risk factor identifier, and the training is completed after the test accuracy meets the accuracy threshold, thereby obtaining the trained and tested basic traffic risk factor identifier.

[0042] In the embodiments of the present application, the purpose of step S200 is to construct an initial model framework with general traffic risk identification capability, i.e. a basic traffic risk factor identifier. The basic traffic risk factor identifier can establish the correlation between "longitudinal slope features + vehicle features" and "longitudinal slope risk events" based on historical sample data, provide a stable and reliable basic model for subsequent incremental learning for specific continuous longitudinal slope sections, and avoid the problems of low efficiency and insufficient accuracy caused by training the model from scratch.

[0043] Firstly, sample data for training the basic traffic risk factor identifier needs to be collected and labeled, i.e., the safety monitoring data of multiple sample longitudinal slopes, the sample longitudinal slope feature set and the sample vehicle feature set are collected, and the proportion of longitudinal slope risk events occurring under different sample longitudinal slope features and sample vehicle features is collected and labeled as the sample longitudinal slope traffic risk factor information set.

[0044] Specifically, multiple continuous longitudinal slope road segments of different types are selected as sample longitudinal slopes, covering road segments of different slopes, different lengths, and different traffic flows to ensure the diversity and representativeness of the samples. Then, for each sample longitudinal slope, the longitudinal slope gradient, longitudinal slope length, and other parameters are obtained through road design data and field measurement to form a sample longitudinal slope feature set.

[0045] The features of the vehicles passing through each sample longitudinal slope are collected through road monitoring devices such as cameras and vehicle detectors. The features of the vehicles mainly include vehicle length, and in the case of monitoring device support, can further include vehicle type (classified as trucks, cars, buses, etc.), vehicle load, etc., to form a sample vehicle feature set.

[0046] The longitudinal slope risk events, such as vehicle rear-end collisions and brake failures, occurring within a certain time, such as 1 year, for each sample longitudinal slope are counted, and the proportion of risk events occurring under different combinations of "sample longitudinal slope features + sample vehicle features" is calculated. For example, under the combination of "slope 7% + length 2000 meters + truck (length 12 meters)", the risk event occurrence proportion is 8%, and these proportion data are labeled as the sample longitudinal slope traffic risk factor information set.

[0047] Further, a basic traffic risk factor identifier needs to be constructed based on machine learning. According to the task type of the basic traffic risk factor identifier, a neural network algorithm can be used for construction.

[0048] The basic traffic risk factor identifier mainly includes three main parts, namely the input layer, the hidden layer, and the output layer.

[0049] Among them, in the input layer, nodes are set according to the feature categories of the sample longitudinal slope feature set and the sample vehicle feature set, for example, the number of input layer nodes is set to 5, corresponding to 5 types of core feature parameters, which are: sample longitudinal slope features, including longitudinal slope gradient and longitudinal slope length; sample vehicle features, including vehicle length, vehicle load, and vehicle type.

[0050] The hidden layer is set to have two hidden layers, the first hidden layer contains 12 neurons, and the second hidden layer contains 8 neurons. Through two layers of nonlinear transformation, the deep correlation between input features and output risk information is fully excavated.

[0051] The number of nodes of the output layer is set to 1, which corresponds to the core content of the “sample longitudinal slope traffic risk element information set”, that is, the “proportion of longitudinal slope risk event occurrence” under different sample longitudinal slope characteristics and sample vehicle characteristics, and the output result is a value between 0 and 1, representing the proportion of risk event occurrence, such as 0.08, which is 8%.

[0052] In the parameter setting of the basic traffic risk element identifier, the ReLU activation function is used between the input layer and the first layer hidden layer to solve the gradient disappearance problem and enhance the model's feature extraction capability; the ReLU activation function is used between the first layer hidden layer and the second layer hidden layer; and the Sigmoid activation function is used between the second layer hidden layer and the output layer to map the output result to the 0-1 interval, matching the numerical range of the “risk event proportion”. The Adam optimizer is selected, the learning rate is set to 0.005, the mean square error loss function is used as the loss function, the L2 regularization coefficient is set to 0.001, and the sample batch size for each training is set to 32.

[0053] In the training of the basic traffic risk element identifier, the training data is the sample longitudinal slope characteristic set, the sample vehicle characteristic set and the sample longitudinal slope traffic risk element information set. The training data is divided into a training set and a test set according to a ratio of 2:8. The maximum number of training rounds is set to 1200 rounds.

[0054] In the last 20 rounds of training, the loss function value (mean square error) of the model is stable below 0.002, and there is no obvious downward or upward trend, indicating that the model has entered a stable state, and the training is stopped. Finally, the basic traffic risk element identifier is obtained.

[0055] In step S200 of the embodiment of the present application, the incremental learning data is configured according to the brake time and the longitudinal slope characteristics, and the incremental learning data is obtained, including:

[0056] A preset incremental learning data amount is obtained;

[0057] The average brake time of the sample longitudinal slope monitoring processing is obtained, the preset incremental learning data amount is corrected according to the ratio of the brake time to the average brake time, and the incremental learning data amount is obtained;

[0058] According to the incremental learning data amount, the incremental longitudinal slope characteristic set, the incremental vehicle characteristic set and the incremental longitudinal slope traffic risk element information set of the continuous longitudinal slope with the highest similarity to the longitudinal slope characteristics are collected until the incremental learning data amount is met, and the incremental learning data is obtained.

[0059] In the embodiment of the present application, obtaining the incremental learning data in step S200 provides strong adaptability and reasonable quantity of training data for subsequent incremental learning based on the basic traffic risk factor identifier. By combining the brake time of the current continuous longitudinal slope and the longitudinal slope characteristics, the amount of incremental learning data is dynamically adjusted and similar data is screened to avoid data redundancy or deficiency, ensuring that the incremental learning can accurately adapt to the characteristics of the current road section and improving the identification accuracy and efficiency of the subsequent incremental identifier.

[0060] First, a preset incremental learning data amount needs to be obtained, that is, an initial and general incremental learning data amount standard is set. The preset value is determined based on the experience data of past incremental learning of similar longitudinal slope road sections, and subsequent data amount correction provides a basic reference. For example, it is required to collect 100 groups of data.

[0061] Then, the average brake time of the monitoring processing in the multiple sample longitudinal slopes needs to be obtained. The preset incremental learning data amount is corrected according to the ratio of the brake time and the average brake time to obtain the incremental learning data amount. That is, from the existing safety monitoring data of the multiple sample longitudinal slopes, that is, the sample longitudinal slopes used when the basic traffic risk factor identifier is constructed, the brake time of each sample longitudinal slope is extracted, the average brake time of all sample longitudinal slopes is calculated, for example, the brake times of the multiple sample longitudinal slopes are 3.2 seconds, 2.8 seconds and 3.0 seconds, and the average brake time is 3.0 seconds.

[0062] Then, the average brake time of the monitoring processing in the multiple sample longitudinal slopes is obtained. The ratio of the brake time and the average brake time, that is, the ratio of the brake time of the current continuous longitudinal slope and the average brake time, is calculated, such as 4.5 ÷ 3.0 = 1.5.

[0063] Then, the preset incremental learning data amount is corrected by the ratio. The correction method can be incremental learning data amount = ratio of brake time and average brake time * preset incremental learning data amount. If the current brake time is greater than the average brake time (ratio > 1), it means that the vehicle brake behavior of the current road section is more frequent and the risk is more complex, so the data amount needs to be increased, such as adjusting the data amount to 100 x 1.5 = 150 groups. If the current brake time is less than the average brake time (ratio < 1), the data amount is appropriately reduced, and finally the incremental learning data amount adapted to the current road section is obtained.

[0064] Further, the incremental longitudinal slope characteristic set, the incremental vehicle characteristic set and the incremental longitudinal slope traffic risk factor information set of the continuous longitudinal slope with the largest similarity to the longitudinal slope characteristics need to be collected according to the incremental learning data amount until the incremental learning data amount is met.

[0065] That is, taking the longitudinal slope characteristics of the current continuous longitudinal slope as the benchmark, from the sample longitudinal slope database or the newly monitored longitudinal slope section, the continuous longitudinal slope with the smallest difference in longitudinal slope gradient, longitudinal slope length, and the current section is selected. For example, for the longitudinal slope in S100, the gradient is 7%, and the length is 2000 meters. Then, similar longitudinal slopes with a gradient of 6.8% to 7.2% and a length of 1900 to 2100 meters can be selected.

[0066] For the selected similar longitudinal slope, an incremental longitudinal slope characteristic set, an incremental vehicle characteristic set, and an incremental longitudinal slope traffic risk element information set are collected. The incremental longitudinal slope characteristic set is the gradient, length, and other parameters of the similar longitudinal slope; the incremental vehicle characteristic set is the length and other characteristics of the vehicles passing through the similar longitudinal slope; and the incremental longitudinal slope traffic risk element information set is the proportion of longitudinal slope risk events occurring under different combinations of “longitudinal slope characteristics + vehicle characteristics” in the similar longitudinal slope.

[0067] The above three types of data are continuously collected until the total amount of data reaches the incremental learning data amount determined in the foregoing step, such as 150 groups, to finally form the incremental learning data.

[0068] In step S300 of the embodiment of the present application, according to the brake frequency distribution, the incremental learning data is used to perform incremental learning on the basic traffic risk element identifier to obtain a plurality of incremental identifiers, including:

[0069] According to the minimum brake frequency and the maximum brake frequency in the brake frequency distribution, a brake frequency interval is constructed;

[0070] According to the monitoring data of a plurality of sample longitudinal slopes, an average brake frequency interval is processed and obtained;

[0071] A preset incremental learning number is obtained;

[0072] According to the interval length ratio of the brake frequency interval and the average brake frequency interval, the preset incremental learning number is corrected and calculated to obtain an incremental learning number;

[0073] According to the incremental learning number, the incremental learning data is used to perform incremental learning on the basic traffic risk element identifier to obtain a plurality of incremental identifiers.

[0074] In the embodiment 0 of the present application, the purpose of step S30 is to dynamically adjust the number of incremental learning by combining the brake frequency distribution of the current continuous longitudinal slope, to perform multiple optimization training on the basic traffic risk element identifier using incremental learning data with strong adaptability, and finally to obtain a plurality of incremental identifiers. This can make the model fully learn the complex characteristics of the brake behavior of the current section, improve the robustness and accuracy of the traffic risk element identification of the section, and avoid the identification deviation caused by the insufficient data or training times of a single model.

[0075] Specifically, first, the brake frequency interval needs to be constructed according to the minimum brake frequency and the maximum brake frequency in the brake frequency distribution. That is, based on the brake frequency distribution obtained in step S100, the minimum brake frequency and the maximum brake frequency are extracted, and the brake frequency interval of the current continuous longitudinal slope is constructed with the two values as the boundaries. For example, if the minimum frequency in the brake frequency distribution is 18 times / day and the maximum frequency is 35 times / day, then the brake frequency interval of the current section is [18 times / day, 35 times / day].

[0076] Further, the average brake frequency interval needs to be obtained according to the monitoring data of multiple sample longitudinal slopes. From the safety monitoring data of multiple sample longitudinal slopes used to construct the basic traffic risk factor identifier, the brake frequency distribution of each sample longitudinal slope is extracted, the brake frequency interval of each sample longitudinal slope is calculated respectively, and then the brake frequency intervals of all sample longitudinal slopes are statistically processed: the "minimum frequency average" and the "maximum frequency average" of all sample intervals are calculated, and the average brake frequency interval is formed with the two average values as the boundaries. For example, the minimum brake frequency average of 100 sample longitudinal slopes is 20 times / day, and the maximum brake frequency average is 30 times / day, then the average brake frequency interval is [20 times / day, 30 times / day].

[0077] Further, the preset incremental learning quantity needs to be obtained. An initial and general incremental learning frequency standard is set, which is determined based on the experience of past incremental learning of similar longitudinal slope sections, for example, the preset incremental learning quantity is 5 times, which provides a basis for subsequent frequency correction.

[0078] Further, the preset incremental learning quantity is corrected and calculated according to the interval length ratio of the brake frequency interval and the average brake frequency interval, and the incremental learning quantity is obtained.

[0079] First, the length of the brake frequency interval of the current section and the length of the average brake frequency interval of the sample longitudinal slope are calculated respectively. The brake frequency interval of the current section is the maximum brake frequency-minimum brake frequency, such as 35-18=17 times / day; the length of the average brake frequency interval of the sample longitudinal slope is the average maximum brake frequency-average minimum brake frequency, such as 30-20=10 times / day.

[0080] Then, the ratio of the lengths of the two intervals is calculated, that is, the length of the brake frequency interval of the current section ÷ the length of the average brake frequency interval of the sample longitudinal slope, such as 17÷10=1.7.

[0081] Next, the preset incremental learning number is corrected using the ratio of the interval length. Specifically, the incremental learning number = preset incremental learning number * the above interval length ratio. If the above interval length ratio is greater than 1, it means that the braking frequency of the current road segment fluctuates more and the vehicle driving status is more complex, so the number of incremental learning needs to be increased, such as 5 × 1.7 ≈ 9 times. If the above interval length ratio is less than 1, the number of learning is appropriately reduced, and finally the incremental learning number adapted to the current road segment is obtained.

[0082] Furthermore, according to the stated incremental learning quantity, the basic traffic risk element identifier needs to be incrementally learned using the incremental learning data to obtain multiple incremental identifiers.

[0083] In step S300 of this application embodiment, the basic traffic risk element identifier is incrementally learned using the incremental learning data according to the incremental learning quantity, to obtain multiple incremental identifiers, including:

[0084] Based on the total number of incremental learning iterations, the incremental learning data is randomly divided multiple times to obtain multiple random incremental data sets.

[0085] Multiple random incremental data sets are used to incrementally train the basic traffic risk element identifier, resulting in multiple incremental identifiers.

[0086] like Figure 2 As shown, in step S300 of this application embodiment, multiple differentiated incremental recognizers are generated by randomly dividing and training the incremental learning data multiple times. This is to avoid the random errors caused by a single data division method, improve the robustness and generalization ability of the model to identify traffic risk factors on the current continuous longitudinal slope, and ensure that the final output traffic risk identification result is more comprehensive and reliable.

[0087] First, the total number of data partitions needs to be determined using the "incremental learning quantity" already defined in S300. For example, if the incremental learning quantity is 9 times, then 9 data partitions are required.

[0088] For the "incremental learning data" configured in S200, the incremental learning data is randomly allocated each time it is partitioned into a training subset and a validation subset. The data partitioning ratio remains consistent each time, with the training subset accounting for 80% and the validation subset accounting for 20%, but the specific data allocation must be different.

[0089] After completing the same number of random partitions as the incremental learning, a corresponding number of "random incremental data" are obtained. For example, 9 partitions correspond to 9 random incremental data. Each random incremental data contains an independent training subset and a validation subset, and the sample distribution of each data is different, which can cover more diverse feature combination scenarios.

[0090] Furthermore, the "Basic Traffic Risk Element Identifier" is used as the initial model for each incremental learning iteration. For each random incremental data set, its training subset is input into the Basic Traffic Risk Element Identifier. By adjusting parameters such as the model's learning rate and neuron weights, the model learns the correlation patterns between features and risks in the current random incremental data. At the same time, a validation subset is used to monitor the model's training effect in real time to avoid overfitting.

[0091] Ultimately, multiple incremental recognizers are generated: After training on each random incremental data set, the parameters of the base recognizer are optimized to form an "incremental recognizer" adapted to the characteristics of that data set. This process is repeated until all random incremental data sets have been trained, resulting in "multiple incremental recognizers" equal to the number of incremental learners. Each recognizer can identify traffic risk factors on continuous longitudinal slopes based on the feature patterns it has learned.

[0092] In step S400 of this application embodiment, vehicle features are collected, and combined with the longitudinal slope features, based on multiple incremental recognizers, longitudinal slope traffic risk factor information is identified and output, including:

[0093] The road monitoring system obtains the vehicle characteristics of currently passing vehicles, including vehicle length.

[0094] The vehicle features and longitudinal slope features are input into multiple incremental recognizers, and the recognition outputs multiple longitudinal slope traffic risk element information are obtained. The longitudinal slope traffic risk element information is obtained by fusion processing.

[0095] In this embodiment of the application, step S400 above is to collect the key features of currently passing vehicles, combine them with the inherent longitudinal slope features of the road section, and utilize the collaborative recognition capabilities of multiple incremental recognizers to achieve accurate and comprehensive identification of traffic risk factors on continuous longitudinal slopes.

[0096] Specifically, this requires relying on road monitoring systems deployed along the roadside, including high-definition cameras, vehicle detection radar, and intelligent weighing equipment, to capture real-time data on vehicles traveling on continuous longitudinal slope sections and extract core vehicle characteristics. The focus is on acquiring the length parameters of each passing vehicle, along with data such as load and vehicle type. For example, by using camera images and image recognition technology to measure the distance between the front and rear ends of vehicles, a sedan's length can be determined to be 4.5 meters, and a truck's length to be 12 meters, thus forming the vehicle characteristic data for the current vehicle.

[0097] Then, the collected current vehicle features are integrated with the continuous longitudinal slope features obtained in S100 to form a complete feature input set. For example, a feature input set can be "vehicle length 4.5 meters + longitudinal slope gradient 7% + longitudinal slope length 2000 meters".

[0098] Next, the complete feature input set is synchronously transmitted into the multiple incremental recognizers obtained in S300. Each incremental recognizer independently analyzes the input vehicle and longitudinal slope features based on the feature rules learned by itself through different random incremental data, and outputs corresponding recognized longitudinal slope traffic risk element information, i.e., the risk event occurrence proportion predicted by the incremental recognizer, such as that the traffic risk proportion output by the alpha incremental recognizer is 8%, the traffic risk proportion output by the beta incremental recognizer is 7.5%, and so on.

[0099] Finally, the recognized longitudinal slope traffic risk element information output by the multiple incremental recognizers is fused by using a reasonable fusion strategy to eliminate the accidental errors of a single recognizer, and comprehensive and reliable longitudinal slope traffic risk element information is obtained, such as taking the average value, weighted average value, or screening the results consistent with the majority of recognizers, and so on. For example, the five incremental recognizers output 8%, 7.5%, 8.2%, 7.8%, and 8.1% respectively, and the final traffic risk proportion 7.92% is obtained by taking the average value, which is taken as the longitudinal slope traffic risk element information, representing the risk degree of the corresponding vehicle driving in the corresponding longitudinal slope.

[0100] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0101] Those skilled in the art should understand that the embodiments of the present application can provide a method, a system or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0102] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system) and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams can be realized by computer program instructions, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded computer or other programmable data processing devices to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices produce a machine that implements the functions described in the flowcharts and / or block diagrams.Figure 1 apparatuses that implement the functions specified in the flowchart Figure 1

[0103] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flowchart Figure 1 apparatuses that implement the functions specified in the flowchart Figure 1

[0104] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions that are executed on the computer or other programmable apparatus provide steps for implementing the flowchart Figure 1 apparatuses that implement the functions specified in the flowchart Figure 1

[0105] Although preferred embodiments of the application have been described, those skilled in the art will recognize that additional modifications and variations are possible without departing from the spirit and scope of the application.

[0106] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the application, the application can be practiced otherwise than as specifically set forth herein.​​​

Claims

1. A method for intelligent identification of continuous longitudinal slope traffic risk factors based on deep learning, characterized in that, The method comprises: obtaining longitudinal slope characteristics of a continuous longitudinal slope, and identifying brake characteristics from monitoring data of vehicle braking in the continuous longitudinal slope, wherein the brake characteristics comprise brake frequency distribution and brake time; obtaining a basic traffic risk factor identifier, and configuring incremental learning data according to the brake time and the longitudinal slope characteristics to obtain incremental learning data; incrementally learning the basic traffic risk factor identifier using the incremental learning data according to the brake frequency distribution to obtain a plurality of incremental identifiers; collecting vehicle characteristics, combining the longitudinal slope characteristics, and identifying and outputting longitudinal slope traffic risk factor information based on the plurality of incremental identifiers; wherein the configuring of the incremental learning data according to the brake time and the longitudinal slope characteristics to obtain the incremental learning data comprises: obtaining a preset incremental learning data amount; obtaining an average brake time of monitoring processing in a plurality of sample longitudinal slopes, and correcting the preset incremental learning data amount according to a ratio of the brake time to the average brake time to obtain an incremental learning data amount; collecting an incremental longitudinal slope characteristic set, an incremental vehicle characteristic set, and an incremental longitudinal slope traffic risk factor information set of a continuous longitudinal slope with the greatest similarity to the longitudinal slope characteristics according to the incremental learning data amount until the incremental learning data amount is met to obtain the incremental learning data.

2. The deep learning-based continuous longitudinal slope traffic risk factor intelligent identification method according to claim 1, characterized in that, obtaining longitudinal slope characteristics of a continuous longitudinal slope, and identifying brake characteristics from monitoring data of vehicle braking in the continuous longitudinal slope, comprising: obtaining longitudinal slope characteristics of a continuous longitudinal slope, wherein the longitudinal slope characteristics comprise longitudinal slope gradient and longitudinal slope length; obtaining vehicle brake monitoring data in a preset time range in the continuous longitudinal slope, identifying brake time sequences, and obtaining a historical brake time sequence set; calculating a plurality of average frequencies and average total time lengths of braking according to the historical brake time sequence set to obtain brake frequency distribution and brake time as brake characteristics.

3. The deep learning-based continuous longitudinal slope traffic risk factor intelligent identification method according to claim 1, characterized in that, obtaining a basic traffic risk factor identifier comprises: collecting a sample longitudinal slope characteristic set and a sample vehicle characteristic set according to safety monitoring data of a plurality of sample longitudinal slopes, and collecting proportions of longitudinal slope risk events occurring under different sample longitudinal slope characteristics and sample vehicle characteristics and labeling as a sample longitudinal slope traffic risk factor information set; constructing a basic traffic risk factor identifier based on machine learning; supervising training of the basic traffic risk factor identifier using the sample longitudinal slope characteristic set, the sample vehicle characteristic set, and the sample longitudinal slope traffic risk factor information set, completing training when a test accuracy rate meets an accuracy rate threshold to obtain a trained and tested basic traffic risk factor identifier.

4. The deep learning-based continuous longitudinal slope traffic risk factor intelligent identification method according to claim 1, characterized in that, incrementally learning the basic traffic risk factor identifier using the incremental learning data according to the brake frequency distribution to obtain a plurality of incremental identifiers, comprising: constructing a brake frequency interval according to a minimum brake frequency and a maximum brake frequency in the brake frequency distribution; processing an average brake frequency interval from monitoring data of a plurality of sample longitudinal slopes; obtaining a preset incremental learning quantity; correcting and calculating the preset incremental learning quantity according to an interval length ratio of the brake frequency interval to the average brake frequency interval to obtain an incremental learning quantity; According to the number of incremental learning, the base traffic risk factor identifier is incrementally learned by using the incremental learning data, and a plurality of incremental identifiers are obtained.

5. The deep learning-based continuous longitudinal slope traffic risk factor intelligent identification method according to claim 4, characterized in that, According to the number of incremental learning, the base traffic risk factor identifier is incrementally learned by using the incremental learning data, and a plurality of incremental identifiers are obtained, comprising: According to the total number of incremental learning, the incremental learning data is randomly divided multiple times to obtain multiple random incremental data; Respectively using multiple random incremental data, the base traffic risk factor identifier is incrementally learned and trained to obtain multiple incremental identifiers.

6. The deep learning-based continuous longitudinal slope traffic risk factor intelligent identification method according to claim 1, characterized in that, Collecting vehicle features, combining the longitudinal slope features, and based on the plurality of incremental identifiers, longitudinal slope traffic risk factor information is identified and output, comprising: Through the road monitoring system, the vehicle features of the current passing vehicle are obtained, wherein the vehicle features include the vehicle length; The vehicle features and the longitudinal slope features are input into the plurality of incremental identifiers, and a plurality of identified longitudinal slope traffic risk factor information is obtained by identification and output, and the longitudinal slope traffic risk factor information is obtained by fusion processing.

Images