Driving style classification and identification method under urban typical steering working condition

Abstract

The invention provides a driving style characteristic classification and identification method under an urban typical steering condition. The method comprises the following steps: constructing a plurality of urban typical driving experiment scenes and collecting driving steering data and car-following data; respectively processing the driving data under each working condition by using Kalman filtering, a sliding time window, an analytic hierarchy process and a principal component analysis method, and obtaining multi-dimensional characteristic parameters; providing a multi-working-condition characteristic parameter coupling method considering subjective driving style evaluation of the driver to couple the multi-dimensional characteristic parameters; using a k-means + + algorithm optimized by a Cuckoo Search algorithm to perform unsupervised clustering on the data samples, and dividing driving styles into a cautious type, a general type and a radical type; training a BP neural network optimized by a genetic algorithm by using a data set containing a driving style, and constructing a driving style characteristic identification model under an urban typical steering working condition; and the identification accuracy is verified in a multi-working-condition coupled city scene. According to the invention, the identification accuracy of the driving style can be improved, and personalized driving requirements are met.

Description

Technical Field

[0001] The present invention relates to the field of intelligent vehicles, and in particular to a method for classifying and identifying driving styles under typical urban steering conditions. Background Art

[0002] The driver's style during driving is an important indicator for evaluating a driver's driving characteristics, and different drivers have different driving styles. In urban roads, the driver's steering characteristics can often effectively reflect the driver's driving style. With the continuous advancement of intelligent connected vehicle technology, the demand for human-computer interaction is also increasing. Drivers hope that the vehicle can make adaptive adjustments based on their own driving style to meet personalized driving needs and experience. The core of the above process is to quickly obtain driving characteristic parameters and accurately identify the driving style type. Therefore, accurate identification of the driver's driving style has important guiding significance for the design of intelligent vehicles.

[0003] At present, most of the classification and identification methods of driver driving style characteristics identify the driver's driving style into several limited categories by analyzing the driver's driving behavior data, facial fatigue level, psychological characteristic data, etc. However, most of the research on such driving style characteristics uses a single working condition scenario for experiment, only considers the driver's own driving behavior, selects the same characteristic parameters and data processing methods, and does not combine driving behavior with road scenes for research, which makes such methods often have low driving style identification accuracy and poor robustness in practical applications. In addition, the methods used by some current research on driver characteristics are often too complicated, and do not take into account the characteristics that vehicles are mainly driven on standardized urban roads, making it difficult to meet actual personalized driving needs. Summary of the invention

[0004] In order to solve the problems that the current driver characteristic identification methods have a single or complex applicable scenario and poor identification robustness, the present invention provides a driving style characteristic classification and identification method under typical urban steering conditions, which can realize the identification of driving style characteristics for different types of drivers under typical urban steering conditions and provide fast and accurate steering driving style information for intelligent vehicle driving assistance systems.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A method for classifying and identifying driving style characteristics under typical urban steering conditions of the present invention comprises the following steps:

[0007] Step 1: Construct experimental driving scenarios, including three conditions: right-angle turns at urban intersections, lane change following a vehicle on general urban roads, and emergency obstacle avoidance steering.

[0008] Step 2, collecting steering data and following data of the driver under corresponding working conditions during the driving process through sensors;

[0009] Step 3: Statistically process the steering data and car-following data collected under each working condition to obtain the steering characteristic parameters and car-following characteristic parameters under each working condition. Further process the parameters through filtering, data enhancement, principal component analysis, and hierarchical analysis to obtain multi-dimensional characteristic data;

[0010] Step 4: firstly assign corresponding basic weights to the multidimensional characteristic parameters under each working condition according to the frequency of driving behaviors under each working condition in actual driving, then perform subjective weight correction on the multidimensional characteristic parameters according to the driver's evaluation of his personal driving behavior style under each working condition, and finally couple them into characteristic parameters describing the driver's steering behavior under typical urban working conditions;

[0011] Step 5, based on the steering behavior characteristic parameters, unsupervised clustering is performed using the k-means++ algorithm optimized by Cuckoo Search to generate a data set containing the characteristic parameters and the driving style types to which the samples belong;

[0012] Step 6: introduce the data set input into the BP neural network optimized by the genetic algorithm for training, build a driver characteristic identification model under the corresponding scenario, and perform driver characteristic identification.

[0013] Preferably, in step 2, the collection method is: the steering wheel and the pedals are connected to the simulation main computer, and the steering wheel angle, steering wheel angular velocity, throttle opening, and brake pedal pressure value during the driver's driving process are obtained through the communication interface of the steering wheel motion sensor and the pedal pressure sensor; an AIR sensor is set in the working condition established by the simulation main computer to collect parameters during the driver's driving process including: longitudinal vehicle speed, longitudinal acceleration, lateral acceleration, impact degree, lateral error, reaction time error, collision time TTC, headway TH, and emergency obstacle avoidance duration.

[0014] Preferably, in step 2, the steering data and following data collected under different working conditions are: under the right-angle bend condition at an urban intersection, three indicators, namely, steering style index, longitudinal style index and tracking index, are collected. The steering style index includes parameters: mean steering wheel angle, standard deviation of steering wheel angle, peak steering wheel angular velocity; the longitudinal style index includes parameters: mean longitudinal vehicle speed, mean longitudinal acceleration, mean throttle opening; the tracking index includes parameters: lateral error, mean yaw rate, and reaction time.

[0015] Under the following and lane changing conditions on general urban roads, the characteristic parameters obtained through statistical processing include: mean steering wheel angular velocity, steering wheel angular velocity variation coefficient, steering wheel angle peak, mean throttle opening, mean brake pedal pressure, mean longitudinal vehicle speed, mean longitudinal acceleration, and mean impact degree.

[0016] Under the emergency obstacle avoidance steering condition, the characteristic parameters obtained through statistical processing include: TTC at the start of obstacle avoidance, the proportion of dangerous vehicle headway, the mean of the product of the steering wheel angle and the vehicle speed, the standard deviation of the product of the steering wheel angle and the vehicle speed, and the duration of emergency obstacle avoidance lane change.

[0017] Preferably, the emergency obstacle avoidance lane change duration T c :

[0018]

[0019] t1 and t2 are the start and end times of the emergency obstacle avoidance behavior, respectively;

[0020] The product of the steering wheel angle and the vehicle speed p:

[0021] p=δ sw V x

[0022] δ sw is the steering wheel angle, V x is the vehicle speed;

[0023] The headway time TH is expressed as:

[0024]

[0025] d represents the relative distance between the front vehicle and the ego vehicle;

[0026] Jerk:

[0027]

[0028] a is acceleration;

[0029] Collision time TTC:

[0030]

[0031] V r is the relative speed between the vehicle and the front vehicle;

[0032] Dangerous headway ratio δ TH ;

[0033]

[0034] n THT is the number of data items whose TH is less than a certain second. c Duration of emergency lane change to avoid obstacles.

[0035] Preferably, in step 3, the steering data and following data collected from each working condition are filtered to eliminate external noise and erroneous data caused by driver errors. Specifically, for the right-angle bend condition at an urban intersection, the hierarchical analysis method is used to determine the weight of each parameter under the indicator, and each indicator is normalized within the experimental section, and all products are summed after multiplying with the corresponding weight to obtain three indicator values ​​as three-dimensional indicators that describe the driver's steering characteristics under the condition; for the following and lane changing condition on general urban roads, the data is subjected to a KMO test to verify its correlation, and then the principal component analysis method is used to select three principal components from high to low according to the principal component contribution rate as the three-dimensional indicators that describe the driver's steering characteristics under the condition; for the emergency obstacle avoidance steering condition, the system measurement equation is established as follows: the estimated value of the system state is obtained by iterative calculation according to the discrete Kalman filter algorithm, the data is enhanced using a sliding time window, and the three-dimensional feature parameter set is obtained by dimensionality reduction through principal component analysis.

[0036] Preferably, in step 4, firstly, the multi-dimensional characteristic parameters under each working condition are assigned corresponding basic weights based on the frequency of occurrence of driving behaviors under each working condition:

[0037]

[0038] Among them, ω i represents the basic parameters of the ith working condition, c i Indicates the number of times the i-th operating condition occurs in the experiment;

[0039] Then the evaluation of the impact of a certain working condition on personal driving style is divided into three levels: important, generally important, and not very important. The basic weight is corrected by subjective weight through the driving test personnel's evaluation of personal driving style under three typical working conditions. Finally, the corrected weight is assigned to the characteristic parameters of each working condition and added together to obtain the coupling characteristic parameters under typical urban steering conditions, which is expressed as:

[0040] M=[ω1k1ω2k2…ω m k m ]×[P1 P2 P3]

[0041] Where M represents the column vector composed of multidimensional parameters after coupling, ω i k i represents the corrected weight of the i-th working condition, k i Represents the subjective correction coefficient, P i Represents a column vector consisting of m-dimensional feature parameters describing the steering style characteristics under the i-th working condition.

[0042] Preferably, in step 5, the several steering behavior characteristic parameter samples obtained in step 4 are used as the original driver characteristic data set, and the original data set is unsupervised clustered using the k-means++ algorithm optimized by Cuckoo Search; different types of cluster centers are recorded and a data set containing characteristic parameters and their corresponding driving style types is generated, and the driving styles are divided into three categories: cautious, general, and aggressive.

[0043] As a preferred choice, the k-means++ algorithm optimized by Cuckoo Search is different from the traditional k-means clustering. It introduces the heuristic algorithm Cuckoo Search to optimize the alternative mean clustering according to the self-iteration process of distance, uses the Levy Flight method to perform a global search, selects the sample point with the highest fitness as the clustering center, and abandons the existing center with a certain probability in each round of iteration to avoid the clustering result falling into the local optimal solution.

[0044] Preferably, performing unsupervised clustering using the k-means++ algorithm optimized by Cuckoo Search comprises the following steps:

[0045] 1. Initially randomly select m cluster centers,

[0046] 2. Calculate the fitness value of each cluster center according to the working conditions,

[0047] 3. Select the cluster center with the largest fitness value as the current optimal solution to vector X best middle,

[0048] 4. Use the Levy flight method to perform a new round of iterations to obtain new m cluster centers.

[0049] 5. Record the optimal solution of this round to X best middle,

[0050] 6. Randomly discard the cluster center with a pre-set probability pa and reselect,

[0051] 7. Record the current optimal solution to X again best middle,

[0052] 8. Repeat 1-7 above until the convergence condition is met, and select the cluster center with the optimal fitness value as the final cluster reference;

[0053] 9. Based on the multidimensional parameter values ​​of the cluster centers, the steering styles of drivers under typical urban steering conditions are defined as cautious, general, and aggressive.

[0054] Preferably, the data set is used as the input of the neural network, the number of hidden layers and the number of neurons are set according to the scene type and the amount of data, the driving style type is encoded and used as the only output of the neural network, and a genetic algorithm is used to optimize the weights and thresholds of the iterative network during the training process to obtain a driver steering characteristic identification model under the scene conditions corresponding to the data set.

[0055] Beneficial Effects

[0056] 1. The present invention takes into account a variety of actual working conditions under urban steering behavior and couples the characteristics of each working condition. Compared with the existing method, it can make a more detailed description of the driver's steering behavior characteristics and is more practical.

[0057] 2. Under the turning condition of an urban intersection, taking into account the relatively smooth transition of parameter changes in this condition, in order to better describe the style differences between different drivers, the present invention divides the characteristic parameters into a turning style index, a longitudinal style index and a tracking index, fully considering the correlation and difference between multiple parameters, and being able to make a hierarchical description of the driving style.

[0058] 3. Under the condition of urban intersections, the hierarchical analysis method is used to evaluate driving style. In addition to the vehicle parameter output, the driver's subjective evaluation of the scene conditions is additionally considered. The three types of indicators are given weights that are suitable for the driver himself, which can better reflect the person's subjective driving style and operating intention.

[0059] 4. In urban emergency obstacle avoidance conditions, taking into account the urgency of the scene, the TTC at the start of obstacle avoidance, the proportion of dangerous vehicle headway, the mean of the product of the steering wheel angle and the vehicle speed, the standard deviation of the product of the steering wheel angle and the vehicle speed, and the duration of emergency obstacle avoidance lane change are selected as five-dimensional feature parameters. Compared with existing methods, it is more concise and efficient, and highlights the main behavioral characteristics in the obstacle avoidance process.

[0060] 5. In urban emergency obstacle avoidance conditions, considering the urgency and volatility of the conditions, Kalman filtering is used to eliminate abnormal noise, effectively retaining the data style characteristics, and improving clustering stability in urgent situations; a sliding time window is used to enhance data samples, highlight data features, and improve the efficiency of feature data acquisition and the degree of data feature differentiation; the principal component analysis method is used to reduce the dimension of feature data, which simplifies the preprocessing process of clustering and identification while retaining most of the data features, improves the recognition efficiency, and helps to achieve real-time style recognition.

[0061] 6. A multi-condition characteristic parameter coupling method considering the driver's subjective driving style evaluation is proposed. Compared with the existing method, it additionally considers the driver's subjective evaluation of his own driving style. The obtained coupling characteristic parameters can more accurately describe the driver's own steering behavior characteristics.

[0062] 7. Cuckoo Search is introduced to optimize the k-means++ clustering process, which effectively avoids the shortcomings of k-means clustering that are affected by the selection of initial center points, easily fall into local optimal solutions, and are greatly affected by noise. The clustering results have higher discrimination, converge faster when processing large amounts of data, and are more robust.

[0063] 8. Genetic algorithm is introduced to optimize the training process of BP neural network. In the process, the network weights and thresholds are dynamically iterated, and the network fitness is higher. While retaining the advantages of BP neural network, the model accuracy and iteration stability are enhanced. Compared with the disadvantage of ordinary BP neural network that the recognition accuracy will drop rapidly when the input difference is not large, GABP network can realize accurate recognition of driving style when the input difference is not obvious, which is more in line with actual steering conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0065] Figure 1 A flow chart of a driving style classification and identification method according to an embodiment of the present invention;

[0066] Figure 2 A schematic diagram of a turning test scenario at an urban intersection according to an embodiment of the present invention;

[0067] Figure 3 A schematic diagram of a general urban lane change and steering test scenario according to an embodiment of the present invention;

[0068] Figure 4 A schematic diagram of an emergency obstacle avoidance steering test scenario according to an embodiment of the present invention;

[0069] Figure 5 A test system structure diagram of an embodiment of the present invention;

[0070] Figure 6 A schematic diagram of a simulation scene of an embodiment of the present invention;

[0071] Figure 7 A schematic diagram of a vehicle-driver model according to an embodiment of the present invention;

[0072] Figure 8 This is a schematic diagram of a clustering effect of an embodiment of the present invention;

[0073] Fig. 9 A structural diagram of a neural network identification model according to an embodiment of the present invention;

[0074] Fig.10 The figure is a schematic diagram of a recognition effect according to an embodiment of the present invention. DETAILED DESCRIPTION

[0075] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0076] The invention application provides a method for classifying and identifying driving styles under typical urban turning conditions, which solves the problem that the current driver characteristic identification methods have a single or complicated applicable scenario and poor identification robustness, and strives to achieve accurate identification of driving style characteristics under typical urban turning conditions, and provide fast and accurate steering driving style information for intelligent vehicle driving assistance systems.

[0077] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0078] like Figures 1 to 10 As shown, a method for classifying and identifying driving style characteristics under typical urban turning conditions includes:

[0079] Step 1: Construct a driving scenario model. In this embodiment, three typical urban driving scenarios are selected, including a right-angle turn scenario at an urban intersection, a lane-changing scenario following a vehicle on a general urban road section, and an emergency obstacle avoidance steering scenario.

[0080] A right-angle turn at a city intersection Figure 2 As shown in the figure, the road in this scenario consists of six 90-degree right-angle turns, forming a closed loop road similar to the shape of an "8". The test vehicle is initially located in the center of the scenario, and the driver completes left and right turns in three circular arcs with different curvatures in turn. This scenario simulates the driver's steering operation in different right-angle turns.

[0081] The following lane-changing scenes on general urban roads are as follows: Figure 3 As shown, it contains two lanes, each of which has an obstacle, obstacle 1 and obstacle 2. The driver needs to change lanes when driving the main vehicle close to obstacle 1 and obstacle 2. This scenario simulates the driver's lane changing behavior on urban roads.

[0082] Emergency obstacle avoidance steering scenario Figure 4As shown in the figure, there is a long straight double lane. The front car is a vehicle set by the system, and the rear car is a vehicle controlled by the driver through the simulator. The initial speed of the two cars is 0, and the distance between the two cars is 50m. The rear car follows the front car and accelerates at the starting point of the road until the front car reaches 60km / h. When the rear car follows the front car at 60km / h for 20s, the front car performs emergency braking. After the front car issues an emergency braking warning, the rear car starts emergency obstacle avoidance lane change, and records the current time t1, which is considered to be the beginning of obstacle avoidance behavior. The front car decelerates from 60km / h to 0. When the steering wheel of the front car returns to the center, it is considered that the lane change behavior is over at this time, and the current time t2 is recorded.

[0083] Step 2-3, collecting the steering data and following data of the driver under different working conditions during the driving process, and performing statistical processing including filtering, data enhancement, principal component analysis, and hierarchical analysis on the collected data to obtain multidimensional feature numbers under each working condition.

[0084] In this embodiment, Figure 2 , Figure 3 , Figure 4 The established scenario is connected to the simulation host computer through the steering wheel and pedals, and the steering wheel angle, steering wheel angular velocity, throttle opening average, and brake pedal pressure value during the driver's driving process are obtained through the communication interface of the steering wheel motion sensor and the pedal pressure sensor. The AIR sensor is set in the working condition established by the simulation host computer to collect parameters during the driver's driving process, including longitudinal vehicle speed, longitudinal acceleration, lateral acceleration, impact degree, lateral error, reaction time error, collision time (TTC), headway (TH), and emergency obstacle avoidance duration.

[0085] The reason for selecting the above parameters is that

[0086] Steering wheel angle, steering wheel angular velocity, longitudinal vehicle speed, longitudinal acceleration and yaw rate: These characteristic parameters are closely related to driving style and can reflect the driver's operating habits and style tendencies in different driving situations. Generally speaking, a larger steering wheel angle, a larger steering wheel angular velocity, a higher longitudinal vehicle speed, longitudinal acceleration and a faster yaw rate tend to indicate an aggressive driving style, which means that the driver is more decisive when turning and is more aggressive in driving. On the contrary, smaller values ​​of these physical quantities indicate a more conservative driving style, showing that the driver's operation is more stable and cautious.

[0087] Lateral error, yaw rate and reaction time: This characteristic parameter is closely related to the driver's driving ability during driving. Specifically, when the driver can keep the vehicle along the predetermined path well and the lateral error is small, it means that the driver's control of the vehicle is more precise and can effectively reduce deviations from the predetermined route, reflecting a strong driving control ability. The yaw rate refers to the degree to which the car deviates from the expected trajectory during driving. Generally, the smaller the yaw rate means the smaller the deviation from the predetermined track, reflecting a stronger driving control ability. Similarly, a shorter reaction time means that the driver can quickly perceive changes in the external environment and make timely adjustments, so that when faced with emergencies, they can better avoid dangerous situations.

[0088] Throttle opening: This parameter is data that directly describes the driver's operating behavior. A cautious driver will only lightly step on the accelerator when accelerating and accelerate slowly, while an aggressive driver will step on the accelerator hard to get a greater acceleration.

[0089] Emergency lane change duration: refers to the time from the start to the end of the emergency lane change behavior, which is used to measure the total time the driver changes lanes. A shorter emergency lane change duration is usually associated with a more aggressive driving style. The mathematical expression is:

[0090] T c =t2-t1

[0091] Product of steering wheel angle and vehicle speed: There is a coupling relationship between steering wheel angle and longitudinal speed. Generally, the greater the longitudinal speed, the smaller the steering wheel angle. However, aggressive drivers tend to drive at higher speeds with larger steering wheel angles. Therefore, the product of steering wheel angle and vehicle speed can be used to characterize the driver's driving style. The mathematical expression is:

[0092] p=δ sw V x

[0093] Headway: Refers to the time distance between the front and rear vehicles, usually used to measure the driver's following behavior. Shorter headway is usually associated with a more aggressive driving style. The mathematical expression of headway is:

[0094]

[0095] Where d represents the relative distance between the front vehicle and the ego vehicle.

[0096] Jerk: refers to the rate of change of acceleration, and is also one of the indicators to measure the aggressiveness of driving style. Its mathematical expression is:

[0097]

[0098] Where a is the acceleration.

[0099] Time to Collision (TTC): refers to the time required for the vehicle to collide with the vehicle in front without deceleration at the current vehicle speed. The mathematical expression of the time to collision is:

[0100]

[0101] Among them, V r is the relative speed between the vehicle and the front vehicle.

[0102] The proportion of dangerous headway (δ TH ): When TH is less than 2(s), the following behavior is not safe enough. TH A larger value indicates that the sample is at a smaller following distance or a larger following speed most of the time, and its driving style is more aggressive. TH The mathematical expression is as follows:

[0103]

[0104] where n TH T is the number of data with a TH less than 1.5 seconds. c Duration of emergency lane change to avoid obstacles.

[0105] Furthermore, statistical processing including filtering, data enhancement, principal component analysis and hierarchical analysis method are performed according to the collected original parameters to filter out external noise and erroneous data caused by driver's operating errors, and obtain multi-dimensional feature data under various working conditions.

[0106] Statistical treatment:

[0107] In statistics, mean, standard deviation, coefficient of variation and peak value can effectively reflect the characteristics of data. Therefore, in the scenario of driver characteristic identification, the characteristics of driving behavior can be described by calculating the statistical indicators of the above parameters.

[0108] The mean can reflect the central tendency of the moment. It determines the average level of a set of data. Its mathematical expression is:

[0109]

[0110] Among them, μ represents the mean, X i Represents the i-th sample data, N is the number of sample data. It is represented by the subscript mean in the feature parameter.

[0111] The standard deviation shows the degree of fluctuation of the data. It measures the dispersion of the data points relative to the mean. The mathematical expression of the standard deviation is:

[0112]

[0113] Among them, σ represents the standard deviation. It is represented by the subscript std in the characteristic parameter.

[0114] The coefficient of variation (CV) is the ratio of the standard deviation to the mean, which is used to measure the relative dispersion of data. Its mathematical expression is:

[0115]

[0116] Where CV stands for coefficient of variation, which is represented by the subscript cv in characteristic parameters.

[0117] The peak value refers to the maximum or minimum value in a set of data, which can reflect the extreme situation of the data. The mathematical expression of the peak value is:

[0118] Peak=max(x1,x2,…,x N )

[0119] or

[0120] Peak=min(x1,x2,…,x N )

[0121] Wherein, Peak represents the peak value, which is represented by the subscript peak in the characteristic parameter.

[0122] Specifically,

[0123] For three different typical urban driving scenarios, corresponding feature parameters are selected and corresponding data processing solutions are adopted, as shown below:

[0124] For right-angle bend conditions at urban intersections:

[0125] The selected indicators include the mean steering wheel angle (θ mean ), steering wheel angle standard deviation (θ std ), steering wheel angular velocity peak (ω peak ), mean longitudinal speed (v mean ), mean longitudinal acceleration (a mean ), throttle opening average (k mean ), mean lateral error (d mean ), mean yaw rate (m mean ), reaction time (R t ).

[0126] When the driver performs driving behavior under this condition, after collecting and processing the raw data and obtaining the above parameters, normalization is performed to map them to the interval [0, 1] to unify the dimensions. The mathematical expression is:

[0127]

[0128] Among them, x represents a statistical parameter before treatment, x' represents its normalized value, and x max , x min They respectively represent the maximum and minimum values ​​of the parameters within the data interval.

[0129] Subsequently, the normalized parameters of the steering style index, the vertical style index, and the tracking index are evaluated using the hierarchical analysis method. The specific process is as follows:

[0130] According to the experimental situation and the driver's driving experience survey, the weight of each parameter under this indicator is determined by the hierarchical analysis method, and the weight is assigned according to the following table:

[0131]

[0132] Arrange the corresponding parameters in rows / columns, fill in the corresponding scales, and obtain the judgment matrix

[0133]

[0134] It has the following characteristics:

[0135] (1)a ij The meaning of this is that compared with index j, the importance of i

[0136] (2) The scale on the main diagonal is 1

[0137] (3)a ij >0 and satisfies a ij ×a ji =1

[0138] (4)a ik =a ij ×a jk

[0139] According to the above matrix, the weight vector can be obtained

[0140]

[0141] The score of the indicator can be obtained by multiplying the parameters of the above indicator classification with their weight vectors. The mathematical expression is:

[0142]

[0143] Among them, S is the score of this type of indicator, x i The i-th parameter value contained in this type of indicator, w i is the weight corresponding to the parameter value.

[0144] Finally, [S1, S2, S3] are obtained as the three-dimensional characteristic parameters to characterize the driver's steering style characteristics under right-angle bend conditions at urban intersections.

[0145] For the following lane changing conditions on general urban roads:

[0146] The selected indicators include the mean steering wheel angular velocity (ω mean ), steering wheel angular velocity variation coefficient (ω cv ), steering wheel angle peak (θ peak ), throttle opening average (k mean ), brake pedal pressure value average (b mean ), mean longitudinal speed (v mean ), mean longitudinal acceleration (m mean ), mean impact strength (J mean ).

[0147] When the driver performs driving behavior under this condition, after collecting and processing the original data and obtaining the above parameters, the KMO test is performed on the data to verify its correlation. The mathematical expression is:

[0148]

[0149] Among them, r ij Represents the simple correlation coefficient of the data, α ij Represents the partial correlation coefficient of the data. Generally speaking, when KM0>0.6, the data is considered to be highly correlated and principal component analysis can be used.

[0150] After passing the KM0 correlation test, the parameters are subjected to principal component analysis. The steps are as follows:

[0151] 1. Construct a sample matrix x of size n×p, where n is the number of samples and p is the number of feature parameters.

[0152]

[0153] 2. Standardize the original data by the mean and standard deviation and calculate the covariance matrix, where X is the standardized matrix. ij is the element in the i-th row and j-th column of matrix x. R is the calculated covariance matrix, r ij is the element in the i-th row and j-th column of the matrix R.

[0154]

[0155] 3. Calculate the eigenvalue and eigenvector of R. Where λ is the eigenvalue and a is the eigenvector.

[0156] λ1≥λ2≥…≥λ p ≥0

[0157]

[0158] 4. Calculate the principal component contribution rate and the cumulative contribution rate.

[0159]

[0160] The first, second, ..., mth (m≤p)th principal component corresponding to the eigenvalue. The i-th principal component: F = a1*X1+a2*X2+...+a i *X i , (i=1, 2, …, m). The principal component coefficients are determined by the eigenvectors a1, a2, a3, ….

[0161] According to the contribution rate sorting, the first principal component, the second principal component, and the third principal component are selected from high to low as three-dimensional indicators to describe the driver's steering characteristics under this working condition.

[0162] For emergency obstacle avoidance steering conditions:

[0163] The selected indicators include the time to start avoiding obstacles (TTC), the ratio of dangerous headway (δ TH ), the mean of the product of the steering wheel angle and the vehicle speed (p mean ), the standard deviation of the product of steering wheel angle and vehicle speed (p std ), emergency obstacle avoidance lane change duration (T c )

[0164] When the driver performs driving behavior under this condition, the raw data is collected and processed to obtain the above parameters, and then Kalman filtering is performed. Under the emergency obstacle avoidance steering condition, the sampling time is short, the data fluctuates greatly, and there is a lot of noise. The Kalman filtering algorithm can effectively smooth the data and eliminate noise.

[0165] In this embodiment, a Kalman filter algorithm is performed on some parameters including vehicle speed and acceleration. Taking vehicle speed and acceleration as an example, the steps are as follows:

[0166] First, establish the system state equation of the vehicle longitudinal motion:

[0167] X k =A·X k-1 +G·ξ k-1 , k = 2, 3, ..., N

[0168] The system status is X K =[v K a k ]T,v k is the speed of the following vehicle a k is the acceleration of the following vehicle, ξ kis the random acceleration of the rear vehicle, and the matrices A and G can be obtained from Newton’s laws of motion:

[0169]

[0170] Where T is the sampling period. The measurement equation of the system is:

[0171] Z k =H·X k +δ k ,k=2,3,…,N

[0172] where ξ k and δ k is a white noise that conforms to the normal distribution, and its covariance is S and Y respectively. According to the discrete Kalman filter algorithm, the estimated value of the system state can be obtained by iterative calculation Iteration process:

[0173]

[0174] Where K k is the residual gain, P k - and P k are the prior and a posteriori estimation error covariance matrices respectively, and their initial values ​​can be set to zero matrices.

[0175] For the original data after Kalman filtering, due to the short emergency obstacle avoidance time and the small amount of driving behavior data, it is difficult to collect a large amount of data for classification in reality. Therefore, the sliding time window is used to increase the total amount of data, and the data of a certain style of driving behavior is enhanced to increase the distinction between different types of style data. The sliding time window needs to be used for data enhancement, and the steps are as follows:

[0176] 1. The n-dimensional driving behavior data time series x obtained after collection and filtering t Segment according to the time window τ. Get N driving behavior data sets. The i-th driving behavior data set can be expressed as y i ={x τ}, where y i is the i-th driving behavior dataset, x τ is the element in the i-th driving behavior dataset, and t is the sampling time.

[0177] 2. For y1, y2, ..., y N , (i=1, 2, ..., N) N driving behavior data sets are processed. Since the data is discrete in time series, a series of discrete data must be statistically processed to obtain a specific parameter to better characterize the type and reflect the driver's characteristics. Statistical processing includes calculating the mean and standard deviation. Then, the principal component analysis method is used for dimensionality reduction.

[0178] According to the contribution rate, the first principal component, the second principal component, and the third principal component are selected from high to low as the three-dimensional indicators to describe the driver's steering characteristics under this working condition.

[0179] Step 4: Couple the characteristic parameters of the above working conditions.

[0180] The present embodiment discloses a method for coupling characteristic parameters of multiple working conditions taking into account the subjective evaluation of the driver's driving style to couple the characteristic parameters of each working condition: a real urban road section is modeled, and multiple driving test subjects drive freely in the road section according to their personal preferences for a period of time, and the test subjects record the number of occurrences under three typical turning conditions. When the data sample is large enough, the average number of occurrences of each working condition is obtained, and the basic weight of each working condition is calculated. The evaluation of the impact of a certain working condition on the individual's driving style is divided into three levels: important, generally important, and not very important. The basic weight is corrected subjectively through the driving test personnel's evaluation of the individual's driving style under three typical working conditions. Finally, the corrected weights are assigned to the characteristic parameters of each working condition and added together to obtain the coupling characteristic parameters under typical urban turning conditions.

[0181] Based on the multidimensional feature data obtained above, the corresponding basic weights are assigned to the multidimensional feature parameters under each working condition according to the frequency of driving behaviors under each working condition in the actual driving experiment. The mathematical expression for weight calculation is:

[0182]

[0183] Among them, ω i represents the basic parameters of the ith working condition, c i Indicates the number of times the i-th operating condition occurs in the experiment.

[0184] Then, the subjective weight correction of the multi-dimensional characteristic parameters is performed according to the driver's evaluation of his personal driving behavior style under various working conditions. The correction coefficient follows the following Table 1:

[0185] Table 1 Correction coefficients

[0186] evaluate important General Important Not very important <![CDATA[Correction factor k i > 1.0 0.5 0.25

[0187] The calculation expression of coupled multidimensional characteristic parameters is:

[0188] M=[ω1k1 ω2k2 … ω m k m ]×[P1 P2 P3]

[0189] Among them, M represents the column vector composed of multidimensional parameters after coupling, ω i k i represents the corrected weight of the i-th working condition, P iRepresents a column vector consisting of m-dimensional feature parameters describing the steering style characteristics under the i-th working condition.

[0190] Step 5, obtain the processed coupling parameters, use the k-means++ algorithm optimized by Cuckoo Sea rch to perform unsupervised clustering on them, and preliminarily divide the driving styles corresponding to the driving data into three categories: cautious, general, and aggressive, and generate a data set containing the characteristic parameters and the driving style types to which their samples belong.

[0191] Cuckoo Sea rch is a heuristic algorithm that can effectively solve optimization problems. Its optimization process follows the Levy flight process. In the real world, this flight feature can find the target very effectively. Driving tests show that using Cuckoo Sea rch for clustering can effectively optimize the clustering process and obtain clusters with higher discrimination. The above method divides the data into three categories without other special considerations, and divides them into three categories according to the actual natural driving style: cautious, general and aggressive. The clustering steps are as follows:

[0192] 1. Initially randomly select m cluster centers (bird nests);

[0193] 2. Calculate the fitness value of each cluster center (selected according to the working conditions, usually the root mean square error RMSE);

[0194] 3. Select the cluster center with the largest fitness value as the current optimal solution to vector X best middle;

[0195] 4. Use the Levy flight method to perform a new round of iteration to obtain m new cluster centers;

[0196] 5. Record the optimal solution of this round to X best middle;

[0197] 6. Randomly discard the cluster center with a pre-set probability pa and reselect;

[0198] 7. Record the current optimal solution to X again best middle;

[0199] 8. Repeat the above process until the convergence condition is met, and select the cluster center with the optimal fitness value as the final cluster reference;

[0200] 9. Based on the multidimensional parameter values ​​of the cluster centers, the steering styles under typical urban steering conditions are defined as cautious, general, and aggressive;

[0201] 10. After clustering is completed, the program records the sample data set containing driving styles.

[0202] Step 6: Use the above sample data set to train the BP neural network and build a driver characteristic recognition model for the corresponding scenario. In order to accurately identify the driving style when the characteristic parameter difference is not obvious, the training process uses genetic algorithm optimization to obtain weights and thresholds that are more in line with the scenario conditions. The specific parameter settings are as follows:

[0203] Genetic algorithm: The number of individuals is initially set to 50, the maximum genetic generation is 5, the number of binary bits for transcoding is 10, the generation gap, inheritance probability, and mutation probability are set to 0.95, 0.7, and 0.01 respectively. The objective function is set to the sum of the fitting output values ​​of the BP neural network.

[0204] BP neural network: Set the number of training times to 1000, the training target value to 0.01, and the learning rate to 0.1. The training set and validation set are divided into 70% and 30% of the selected data set. The number of input layer nodes is adapted to the sample dimension, and the output layer output is a 1*3 orthogonal matrix. The number of hidden layer neurons is initially set to 30, with a single hidden layer and the hidden layer activation function is the Sigmoid function. The initial weights and thresholds are adapted to the initial individual parameter values ​​of the genetic algorithm. The error back propagation uses the minimum mean square error mse, and the gradient descent process uses the Levenberg-Marquardt algorithm. The neural network structure is as follows: Figure 6 shown.

[0205] After multiple iterations of the genetic algorithm, the weights and thresholds under the optimal fitness are obtained. The above values ​​are saved and the driver steering style characteristic identification model corresponding to the data set is constructed.

[0206] A driving steering style characteristic identification model is constructed for the typical urban steering scene conditions mentioned above and connected to the driving simulation system, such as Figure 5 The corresponding identification model is used to identify the steering style of driving test subjects with known driving style characteristics under typical urban steering conditions with multi-scenario coupling to verify its accuracy.

[0207] Through the above steps, a set of driver style characteristic identification models under typical urban turning scenarios is finally obtained, and a driving style characteristic classification and identification method under typical urban turning conditions is proposed.

[0208] The present invention provides a method for classifying and identifying driving style characteristics under typical urban steering conditions. There are many methods and ways to implement the technical solution. The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be considered as the scope of protection of the present invention. All components not specified in this embodiment can be implemented using existing technologies.

Claims

1. A method for classifying and identifying driving style characteristics under typical urban steering conditions, characterized in that: The steps include: Step 1: Build a driving scenario model, including right-angle turns at urban intersections, lane-changing with other vehicles on general urban roads, and emergency obstacle avoidance and steering scenarios; Step 2, collecting the steering data and following data of the driver under different driving conditions during the driving process; Step 3, filtering and statistically processing the steering data and the car-following data respectively to obtain steering characteristic parameters and car-following characteristic parameters under various working conditions, and further processing them by hierarchical analysis method, principal component analysis or data enhancement to obtain multidimensional characteristic data; Step 4: firstly assign corresponding basic weights to the multidimensional characteristic parameters under each working condition according to the frequency of driving behaviors under each working condition in actual driving, then perform subjective weight correction on the multidimensional characteristic parameters according to the driver's evaluation of his personal driving behavior style under each working condition, and finally couple them into characteristic parameters describing the driver's steering behavior under typical urban working conditions; Step 5, based on the steering behavior characteristic parameters, unsupervised clustering is performed using the k-means++ algorithm optimized by Cuckoo Search to generate a data set containing the characteristic parameters and the driving style types to which the samples belong; Step 6: introduce the data set input into the BP neural network optimized by the genetic algorithm for training, build a driver characteristic identification model under the corresponding scenario, and perform driver characteristic identification.

2. The method according to claim 1, characterized in that In step 2, the acquisition method is: the steering wheel and the pedal are connected to the simulation main computer, and the steering wheel angle, steering wheel angular velocity, throttle opening, and brake pedal pressure value during the driver's driving process are obtained through the communication interface of the steering wheel motion sensor and the pedal pressure sensor; An AIR sensor is set up in the working conditions established by the simulation main computer to collect parameters of the driver during the driving process, including: longitudinal vehicle speed, longitudinal acceleration, lateral acceleration, impact degree, lateral error, reaction time error, collision time TTC, headway TH and emergency obstacle avoidance duration.

3. The method according to claim 2, characterized in that In step 2, the steering data and following data collected under different working conditions are as follows: under the right-angle bend working condition at an urban intersection, three indicators are collected: steering style index, longitudinal style index, and tracking index. The steering style index includes parameters: steering wheel angle mean, steering wheel angle standard deviation, and steering wheel angular velocity peak value; the longitudinal style index includes parameters: longitudinal vehicle speed mean, longitudinal acceleration mean, and throttle opening mean; the tracking index includes parameters: lateral error, yaw rate mean, and reaction time; Under the condition of following a vehicle and changing lanes on a general urban road, the characteristic parameters obtained through statistical processing include: mean steering wheel angular velocity, steering wheel angular velocity variation coefficient, steering wheel angle peak, mean throttle opening, mean brake pedal pressure, mean longitudinal vehicle speed, mean longitudinal acceleration, and mean impact degree; Under the emergency obstacle avoidance steering condition, the characteristic parameters obtained through statistical processing include: TTC at the start of obstacle avoidance, the proportion of dangerous vehicle headway, the mean of the product of the steering wheel angle and the vehicle speed, the standard deviation of the product of the steering wheel angle and the vehicle speed, and the duration of emergency obstacle avoidance lane change.

4. The method according to claim 3, characterized in that The duration of the emergency obstacle avoidance lane change is T c : T c =t2-t1 t1 and t2 are the start and end times of the emergency obstacle avoidance behavior, respectively; The product of the steering wheel angle and the vehicle speed p: p=δ sw V x δ sw is the steering wheel angle, V x is the vehicle speed; The headway time TH is expressed as: d represents the relative distance between the front vehicle and the ego vehicle; Jerk: a is acceleration; Collision time TTC: V r is the relative speed between the vehicle and the front vehicle; Dangerous headway ratio δ TH ; n TH T is the number of data items whose TH is less than a certain second. c Duration of emergency lane change to avoid obstacles.

5. The method according to any one of claims 1 to 4, characterized in that: In step 3, the steering data and following data collected under each working condition are filtered to eliminate external noise and erroneous data caused by driver's operating errors. Specifically, for the right-angle bend condition at the urban intersection, the hierarchical analysis method is used to determine the weight of each parameter under the index, and each index is normalized in the experimental section, and all products are summed after multiplying with the corresponding weight to obtain three index values ​​as three-dimensional indicators describing the driver's steering characteristics under the condition; for the following and lane changing condition on general urban roads, the data is subjected to the KMO test to verify its correlation, and then the principal component analysis method is used to select three principal components from high to low according to the principal component contribution rate as the three-dimensional indicators describing the driver's steering characteristics under the condition; for the emergency obstacle avoidance steering condition, the measurement equation of the system is established as follows: the estimated value of the system state is obtained by iterative calculation according to the discrete Kalman filter algorithm, the data is enhanced by using the sliding time window, and the three-dimensional feature parameter set is obtained by dimensionality reduction through principal component analysis.

6. The method according to claim 5, characterized in that In step 4, firstly, the corresponding basic weights are assigned to the multi-dimensional feature parameters under each working condition based on the frequency of driving behavior under each working condition: Among them, ω i represents the basic parameters of the ith working condition, c i represents the number of times the i-th working condition occurs in the experiment; then the evaluation of the impact of a certain working condition on personal driving style is divided into three levels: important, generally important, and not very important. The basic weight is corrected by subjective weight through the driving test personnel's evaluation of personal driving style under three typical working conditions. Finally, the corrected weights are assigned to the characteristic parameters of each working condition and added together to obtain the coupling characteristic parameters under typical urban steering conditions, which are expressed as: M=[ω1k1 ω2k2 … ω m k m ]×[P1 P2 P3] Where M represents the column vector composed of multidimensional parameters after coupling, ω i k i represents the corrected weight of the i-th working condition, k i Represents the subjective correction coefficient, P i Represents a column vector consisting of m-dimensional feature parameters describing the steering style characteristics under the i-th working condition.

7. The method according to claim 1, characterized in that In step 5, the steering behavior characteristic parameter samples obtained in step 4 are used as the original driver characteristic data set, and the k-means++ algorithm optimized by Cuckoo Search is used to perform unsupervised clustering on the original data set; different types of cluster centers are recorded and a data set containing characteristic parameters and their corresponding driving style types is generated, and the driving styles are divided into three categories: cautious, general, and aggressive.

8. The method according to claim 7, characterized in that The k-means++ algorithm optimized by Cuckoo Search is different from the traditional k-means clustering. It introduces the heuristic algorithm Cuckoo Search to optimize the alternative mean clustering according to the self-iteration process of distance, uses the Levy Flight method to perform a global search, selects the sample point with the highest fitness as the clustering center, and abandons the existing center with a certain probability in each round of iteration to avoid the clustering result falling into the local optimal solution.

9. The method according to claim 8, characterized in that Unsupervised clustering using the k-means++ algorithm optimized by Cuckoo Search includes the following steps: 1) Initially randomly select m cluster centers, 2) Calculate the fitness value of each cluster center according to the working conditions, 3) Select the cluster center with the largest fitness value as the current optimal solution to vector X best middle, 4) Use the Levy flight method to perform a new round of iterations to obtain m new cluster centers. 5) Record the optimal solution of this round to X best middle, 6) Randomly discard the cluster center with a pre-set probability pa and reselect, 7) Record the current optimal solution to X again best middle, 8) Repeat 1)-7) until the convergence condition is met, and select the cluster center with the optimal fitness value as the final cluster reference; 9) According to the multidimensional parameter values ​​of the cluster centers, the steering styles of drivers under typical urban steering conditions are defined as cautious, general, and aggressive.

10. The method according to claim 1, characterized in that The data set is used as the input of the neural network, the number of hidden layers and the number of neurons are set according to the scene type and the amount of data, the driving style type is encoded and used as the only output of the neural network, and a genetic algorithm is used to optimize the weights and thresholds of the iterative network during the training process to obtain a driver steering characteristic identification model under the scene conditions corresponding to the data set.

Images