CN-122020293-A - Driving behavior prediction method integrating track fractal characteristics and emotion evolution

CN122020293ACN 122020293 ACN122020293 ACN 122020293ACN-122020293-A

Abstract

The invention discloses a driving behavior prediction method integrating fractal characteristics of tracks and emotion evolution, which relates to the technical field of driving behavior analysis, and comprises the steps of S1, obtaining multi-modal driving data of vehicle motion parameters, traffic situation indexes, driving track coordinate sequences, driver physiological signal sequences and emotion state label sequences marked synchronously, preprocessing the multi-modal driving data to generate tested sample data, S2, extracting fractal dimension characteristics of driving tracks through a fractal geometric analysis method based on the driving track coordinate sequences, and S3, constructing and training a hidden Markov model based on the vehicle motion parameters, the traffic situation indexes and the emotion state label states, wherein the emotion state is taken as a hidden state of the hidden Markov model. The fractal dimension of the driving track is provided as a quantization index, key discrimination information except the traditional motion parameters is provided for the model, and the prediction precision is remarkably improved.

Inventors

GUO YONGQING
Meng Tianyang
WEI FULU
GUO YANYONG
ZHANG XIN
WANG XIAOYUAN
XU CHUANHONG
LIU XIAOLONG
ZHAO WENYUE
Sun Lizu

Assignees

山东理工大学

Dates

Publication Date: 20260512
Application Date: 20260123

Claims (10)

1. The driving behavior prediction method integrating the track fractal characteristics and the emotion evolution is characterized by comprising the following steps of: S1, acquiring multi-mode driving data of vehicle motion parameters, traffic situation indexes, driving track coordinate sequences, driver physiological signal sequences and synchronously marked emotion state tag sequences, preprocessing the multi-mode driving data, and generating tested sample data; s2, extracting fractal dimension characteristics of the driving track by a fractal geometric analysis method based on the driving track coordinate sequence; s3, constructing and training a hidden Markov model based on the vehicle motion parameters, traffic situation indexes and emotion state label states, taking the emotion states as hidden states of the hidden Markov model, taking driving behavior states as observable states, and outputting emotion state vectors at the current moment; S4, constructing and training a GRU model integrating fractal features and emotion evolution, wherein an input layer receives standardized vehicle motion parameters, traffic situation indexes, a driver physiological signal sequence, fractal dimension features and emotion state vectors, and the GRU model is processed by a plurality of layers of GRU units and takes predicted driving behavior categories as output; s5, inputting the sample data of the test after pretreatment into a GRU model integrating fractal characteristics and emotion evolution after training, and outputting prediction results of six driving behaviors of accelerating and following, constant-speed following, decelerating and following, accelerating and lane changing, constant-speed lane changing and decelerating and lane changing.
2. The driving behavior prediction method for combining fractal features of trajectories and emotion evolution according to claim 1, wherein the step S1 specifically includes: The multi-mode driving data are acquired through a high-simulation driving simulation experiment platform, the driving simulation experiment platform constructs a simulation scene containing urban arterial roads, suburban expressways and commercial neighborhood roads based on UC-win/Road software, is provided with a force feedback steering wheel, a real vehicle pedal and a multi-channel data synchronous acquisition system, and synchronously acquires the vehicle motion parameters, a driving track coordinate sequence and a driver physiological signal sequence at a sampling frequency of not lower than 50 Hz.
3. The driving behavior prediction method for combining trajectory fractal features with emotion evolution according to claim 1, wherein said step S1 further comprises a data preprocessing step of: Uniformly converting the driving track coordinate sequence into a vehicle body coordinate system taking a vehicle as a center through coordinate transformation; dividing continuous time sequence data according to a preset fixed-time window to generate sample fragments input by a model; normalizing the vehicle motion parameters and normalizing the physiological signals of the driver; encoding the emotional state tag sequences into corresponding category values.
4. The driving behavior prediction method for combining trace fractal features with emotion evolution according to claim 1, wherein the fractal geometry analysis method is a box counting method, and specifically comprises the following steps: setting a series of square grids with different side lengths epsilon for covering a two-dimensional track curve formed by mapping a driving track coordinate sequence; Calculating the minimum number of square boxes N (epsilon) required to completely cover the two-dimensional track curve for each set side length epsilon; and calculating the fractal dimension D of the driving track according to the number N (epsilon) of the boxes and the corresponding side length epsilon.
5. The driving behavior prediction method for combining trajectory fractal features and emotion evolution according to claim 1, wherein in the step S3, a hidden markov model is constructed specifically by the steps of: Defining a set of hidden states Corresponding to six basic emotional states of anger, happiness, sadness, fear, surprise and aversion respectively, defining an observation state set Six driving behaviors of accelerating following, uniform following, decelerating following, accelerating lane changing, uniform lane changing and decelerating lane changing are respectively corresponding; Based on the vehicle motion parameters and the traffic situation indexes, according to a preset driving behavior discrimination rule, discriminating the driving operation at each moment into a specific driving behavior state, thereby generating a driving behavior state observation sequence synchronous with the emotion state label sequence Wherein I is a defined driving behavior state set; Determining parameter sets for hidden Markov models The parameter set lambda consists of an initial state probability distribution pi, a state transition probability matrix A and an observation probability matrix B.
6. The driving behavior prediction method combining trajectory fractal features and emotion evolution according to claim 5, wherein in step S3, the training hidden markov model is specifically performed by the steps of: Taking the driving behavior state observation sequence O as input, and adopting Baum-Welch algorithm to carry out parameter set of the hidden Markov model Performing iterative optimization until the likelihood function of the model to the observation sequence converges; For a given observation sequence O, trained model parameters are utilized Calculating the probability of the driver in various emotional states at each moment by a forward-backward algorithm; And taking the probability distribution of all the predefined emotional states calculated at each moment as the emotional state vector representing the emotional state of the driver at the moment.
7. The driving behavior prediction method for combining fractal features of trajectories and emotional evolution according to claim 1, wherein in the step S4, constructing a GRU model for combining fractal features and emotional evolution specifically includes: Splicing and aligning the preprocessed vehicle motion parameters, traffic situation indexes, the driver physiological signal sequence, the fractal dimension characteristics extracted in the step S2 and the emotion state vectors output in the step S3 to form unified time sequence feature vectors; Constructing a neural network taking a gating circulation unit as a core time sequence processing layer, wherein the input dimension of the network is matched with the dimension of a time sequence feature vector, and the output dimension is matched with the number of driving behavior categories to be predicted; and configuring a final output layer of the network as a Softmax layer, and mapping the final hidden state of the network into the prediction probabilities of various driving behaviors.
8. The driving behavior prediction method for merging fractal features of trajectories and emotional evolution according to claim 7, wherein in step S4, training a GRU model for merging fractal features and emotional evolution specifically includes: Inputting the time sequence feature vector into a network, sequentially carrying out time sequence feature extraction and state update through the gating circulating unit layer, and processing through the Softmax output layer to obtain a prediction probability distribution corresponding to the driving behavior category at each moment; Calculating error loss between the prediction probability distribution and the corresponding real driving behavior class labels through a cross entropy loss function; And adopting an Adam optimizer to iteratively update the parameters of the network according to the gradient.
9. The driving behavior prediction method for merging fractal features of trajectories and emotional evolution according to claim 8, wherein in step S4, training the GRU model for merging fractal features and emotional evolution specifically further includes: introducing an observation probability matrix and a behavior transition probability matrix of the hidden Markov model obtained by training in the step S3 as logic priori knowledge; in the loss calculation step, a regularization term is added for constraining the output probability distribution of the prediction model to be consistent with the logic priori knowledge, so that collaborative modeling of emotion evolution rules is realized.
10. The driving behavior prediction method for combining trajectory fractal features and emotion evolution according to claim 9, wherein in step S5, specifically: generating the sample data of the test after the pretreatment into an input sample containing the time sequence feature vector in the same way as the training phase; inputting the input sample into the GRU model integrating fractal characteristics and emotion evolution after training, and obtaining the prediction probability of each driving behavior class corresponding to each moment through model forward calculation; And outputting a final driving behavior prediction result according to the prediction probability, wherein the result is one of six behaviors of accelerating and following, uniformly following, decelerating and following, accelerating and lane changing, uniformly lane changing and decelerating and lane changing.

Description

Driving behavior prediction method integrating track fractal characteristics and emotion evolution Technical Field The invention relates to the technical field of driving behavior analysis, in particular to a driving behavior prediction method for fusing trace fractal characteristics and emotion evolution. Background The driving behavior prediction is a key technology in the fields of intelligent transportation and automatic driving, and is characterized in that accurate inference of future operation intention of a driver is realized through multi-source data such as vehicle track, environment information and the like. At present, related researches mainly follow a data driving paradigm, and a machine learning and deep learning model is utilized to fit and predict a historical track sequence; The invention patent application with the application number of CN117786526A discloses a vehicle track prediction method considering the emotion state characteristics of a driver, and aims to solve the problems that the existing vehicle track prediction technology is mostly based on the physical motion characteristics of the vehicle, nonlinear influence of emotion fluctuation of the driver on a decision process is not fully considered, prediction accuracy is limited in a complex traffic scene, how the emotion state structurally affects the geometric form of the driving track in space is not deeply disclosed, and quantitative modeling of an emotion dynamic evolution process and the relationship between the emotion dynamic evolution process and a behavior mode is lacking. In addition, although individual researches introduce fractal theory into the traffic field, such as the patent publication of application number CN101290713A, the method aims at excavating periodicity or trend, has no direct correlation with the geometric complexity of a microscopic bicycle driving track in a two-dimensional space, and does not relate to psychological state factors of a driver, and another category of technologies such as application number CN110379193B is completely focused on an automatic driving behavior planning strategy under a specific scene, belongs to a traditional control method of regular or optimized driving, does not analyze the spatial structural characteristics of the track, and does not consider the emotion of the driver. The technical scheme has the following limitations that in the aspect of emotion modeling, single emotion state recognition or simple feature fusion is mostly adopted, deep modeling of multi-category emotion dynamic transfer and a coupling mechanism of the multi-category emotion dynamic transfer and behavior decision is lacked, in the aspect of track feature characterization, traditional kinematic parameters or coordinate point sequences are generally relied on, space structure complexity and trans-scale self-similar characteristics generated by emotion disturbance of a track cannot be quantified from a geometric angle, and a model is difficult to capture a deep mechanism of a behavior mode under emotion driving, so that a driving behavior prediction method for fusing the fractal features of the track and emotion evolution is provided. Disclosure of Invention In order to solve the technical problems, the driving behavior prediction method integrating the track fractal characteristics and the emotion evolution is provided, and the technical scheme solves the problem of lack of deep modeling on multi-category emotion dynamic transfer and a coupling mechanism between the multi-category emotion dynamic transfer and behavior decision. In order to achieve the above purpose, the invention adopts the following technical scheme: A driving behavior prediction method integrating track fractal characteristics and emotion evolution comprises the following steps: S1, acquiring multi-mode driving data of vehicle motion parameters, traffic situation indexes, driving track coordinate sequences, driver physiological signal sequences and synchronously marked emotion state tag sequences, preprocessing the multi-mode driving data, and generating tested sample data; s2, extracting fractal dimension characteristics of the driving track by a fractal geometric analysis method based on the driving track coordinate sequence; s3, constructing and training a hidden Markov model based on the vehicle motion parameters, traffic situation indexes and emotion state label states, taking the emotion states as hidden states of the hidden Markov model, taking driving behavior states as observable states, and outputting emotion state vectors at the current moment; S4, constructing and training a GRU model integrating fractal features and emotion evolution, wherein an input layer receives standardized vehicle motion parameters, traffic situation indexes, a driver physiological signal sequence, fractal dimension features and emotion state vectors, and the GRU model is processed by a plurality of layers of GRU units and takes predicted driving