CN113947259A - Method for predicting vehicle speeds of drivers in different styles based on GRU neural network - Google Patents

Method for predicting vehicle speeds of drivers in different styles based on GRU neural network Download PDF

Info

Publication number
CN113947259A
CN113947259A CN202111321247.XA CN202111321247A CN113947259A CN 113947259 A CN113947259 A CN 113947259A CN 202111321247 A CN202111321247 A CN 202111321247A CN 113947259 A CN113947259 A CN 113947259A
Authority
CN
China
Prior art keywords
driving
gru
vehicle speed
data
vehicle
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111321247.XA
Other languages
Chinese (zh)
Inventor
王家林
宋大凤
董世营
刘奇芳
刘浩然
刘嘉琪
褚洪庆
高炳钊
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jilin University
Original Assignee
Jilin University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Jilin University filed Critical Jilin University
Priority to CN202111321247.XA priority Critical patent/CN113947259A/en
Publication of CN113947259A publication Critical patent/CN113947259A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W40/00—Estimation or calculation of non-directly measurable driving parameters for road vehicle drive control systems not related to the control of a particular sub unit, e.g. by using mathematical models
    • B60W40/10—Estimation or calculation of non-directly measurable driving parameters for road vehicle drive control systems not related to the control of a particular sub unit, e.g. by using mathematical models related to vehicle motion
    • B60W40/105—Speed
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Computation (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Software Systems (AREA)
  • Business, Economics & Management (AREA)
  • Health & Medical Sciences (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Automation & Control Theory (AREA)
  • Transportation (AREA)
  • Mechanical Engineering (AREA)
  • Evolutionary Biology (AREA)
  • Development Economics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Game Theory and Decision Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Traffic Control Systems (AREA)

Abstract

The invention discloses a GRU neural network-based method for predicting vehicle speeds of drivers in different styles, which comprises the following steps: dividing the driving database to obtain a plurality of driving sub-databases corresponding to different driving styles; respectively constructing GRU vehicle speed prediction models corresponding to different driving styles; respectively extracting training data from each driving sub-database, preprocessing the training data, and sliding time step by time step according to a time sequence to generate a training sample sequence database; respectively training GRU vehicle speed prediction models corresponding to the driving styles by using a training sample sequence database; judging the current driving style at certain intervals, and introducing a GRU vehicle speed prediction model corresponding to the current driving style to predict the vehicle speed when the current driving style is continuously unchanged within a preset time period. According to the invention, different GRU vehicle speed prediction models are constructed for prediction tasks in different driving style states, and prediction results of the driving style prediction models are compared without distinguishing, so that the prediction stability and the prediction precision can be effectively improved.

Description

Method for predicting vehicle speeds of drivers in different styles based on GRU neural network
Technical Field
The invention relates to the technical field of automobile speed prediction, in particular to a method for predicting speeds of drivers in different styles based on a GRU neural network.
Background
At present, Model Predictive Control (MPC) in the field of automobile control is studied deeply and extensively, and MPC not only considers current state information in the using process, but also further uses the prediction information obtained by a prediction model to make a plan in advance. All MPC algorithms have a common definition requirement that a model be required to predict future development of variables within a finite range. The current phase of research on automotive control systems has focused on predicting the state variables of the vehicle itself, which in most cases can be summarized as making a speed prediction.
Prediction of velocity generally refers to estimating a velocity value over a time domain by reasonable analysis of some historical data. To date, velocity prediction models have evolved into two broad categories: parametric methods and nonparametric methods. Since the speed of the main vehicle is influenced by the driving mode of the driver and the surrounding environment, the evolution of the vehicle speed has strong random and nonlinear properties, and the speed prediction by a nonparametric method has become the mainstream trend at present. In recent years, with the optimization of algorithms and the rapid increase of computing power and speed, the deep learning method has gained wide attention and great achievement in various engineering fields again. The vehicle speed prediction is a typical prediction task, the evolution of the vehicle speed along with time is dynamic, and an original Recurrent Neural Network (RNN) algorithm and a variant thereof are introduced into the construction of a network neural network prediction model, so that the internal rule of the vehicle speed change can be effectively analyzed and memorized.
The most widely applied recurrent neural network in the field of vehicle speed prediction at the present stage is a long-short term memory network (LSTM), and has the defects of excessive parameter quantity and incapability of parallel training while having good performance, so that a large amount of training data cannot be well processed in a short time, and therefore a more efficient training model needs to be introduced to complete a vehicle speed prediction task. On the other hand, in the conventional vehicle speed prediction process, only the states of the vehicle and the surrounding vehicle are considered, and the influence of the driving characteristics of the driver on the vehicle speed is ignored. Therefore, it is highly desirable to provide a vehicle speed prediction method with higher efficiency while ensuring prediction accuracy.
Disclosure of Invention
In view of the above, the invention provides a method for predicting vehicle speeds of drivers in different styles based on a GRU neural network, which is used for constructing different gating cycle neural network (GRU) vehicle speed prediction models for prediction tasks in different driving style states, comparing prediction results of driving style prediction models without distinguishing, and effectively improving the stability and the accuracy of prediction.
In order to achieve the purpose, the invention adopts the following technical scheme:
a method for predicting vehicle speeds of drivers in different styles based on a GRU neural network comprises the following steps:
s1, dividing the driving database by adopting a pre-constructed driving style classification model to obtain a plurality of driving sub-databases corresponding to different driving styles;
s2, respectively constructing GRU vehicle speed prediction models corresponding to different driving styles;
s3, respectively extracting training data from each driver sub-database, preprocessing the training data, and sliding the training data step by step according to a time sequence to generate a training sample sequence database;
s4, respectively training GRU vehicle speed prediction models corresponding to driving styles by using the training sample sequence database;
and S5, judging the current driving style by adopting the driving style identification model at certain intervals, and introducing a GRU vehicle speed prediction model corresponding to the current driving style to predict the vehicle speed when the driving style identification model is continuously unchanged in a preset time period.
Compared with the prior art, the invention discloses and provides a method for predicting the vehicle speeds of drivers in different styles based on a GRU neural network, and the method has the following beneficial effects:
1. by introducing a more optimized network structure GRU, the training time of the model is reduced by more than 10% on the premise of not changing the original prediction precision, the model prediction time is reduced, and the model prediction efficiency is improved.
2. On the premise of considering the states of the vehicle and the surrounding vehicle, the influence of the driving characteristics of the driver on the change of the vehicle speed is introduced, different vehicle speed prediction models are built for prediction tasks in different driving style states, different GRU vehicle speed prediction models are introduced along with the change of the driving style of the driver, and the prediction stability and the prediction precision are effectively improved by comparing the prediction results of the different driving style prediction models.
Further, the method also comprises the following steps:
and S6, updating the actual vehicle speed information with larger error in the prediction process and the historical information thereof into the training sample sequence database, and retraining and adjusting parameters of each GRU vehicle speed prediction model by using the updated training sample sequence database after a preset time interval.
Further, the method also comprises the following steps: s1 includes the steps of:
s12, carrying out preliminary clustering on the feature data in the driving database by using a self-organizing mapping neural network;
s13, clustering the primary clustering result again by using a K-means clustering method to obtain a plurality of driving sub-databases corresponding to different driving styles; the driving sub-databases include an aggressive driving database, a mediocre driving database, and a conservative driving database.
Further, the method also comprises the following steps: s12 includes the steps of:
s1201, normalizing the characteristic data in the driving database to obtain dimensionless pure quantity;
s1202, introducing an SOM model, and setting the geometric shape in the SOM model to be a rectangle;
s1203, randomly initializing weights of the SOM model, and initializing the node weights into smaller random numbers;
s1204, setting an initial neighborhood radius sigma, and determining nodes to be included in a winning neighborhood according to the neighborhood radius sigma;
s1205, setting a proximity function type, and calculating the respective updating amplitude of each node through the proximity function;
s1206, setting an initial learning rate alpha, and determining the magnitude of each iteration weight updating;
s1207, randomly taking an input sample XiPerforming a first iteration; traversing each node in the competition layer and calculating XiSelecting the node with the minimum distance as a winner node according to the similarity between the node and the node;
s1208, updating the Weight of the nodes in the preferential neighborhood with the preferential node as the center and sigma as the radius range;
s1209, after each iteration is finished, carrying out attenuation processing on sigma and alpha;
s1210: and after one round of iteration is finished, returning to S1207 until the set iteration times are met, and obtaining a primary clustering result.
Further, the method also comprises the following steps: s13 includes:
s1301, randomly selecting 3 samples { mu ] from the primary clustering result1，μ2，μ3As an initial centroid;
s1302, calculating each sample YiWith the center of mass mujDistance between (dis | | | Y)j-μj||2Sample YjClosest to the centroid mujFall into one category;
s1303, updating the centroid
Figure BDA0003345662580000041
Wherein ci is a clustering set in the iteration process, and i belongs to {1,2,3 };
and S1304, returning to S1302 until the maximum iteration number is reached or the centroid updating amplitude is smaller than a threshold value, and finally obtaining the driver sub-databases of the three driving styles.
Further, the method also comprises the following steps: the training data extracted in S3 includes input data and output data, where the input data includes: the data information continuously changing along with time and the state category information discretely distributed along with time, and the output data are vehicle speed signals; the data information continuously changing with time includes: historical vehicle speed, historical acceleration, engine speed, engine torque and relative vehicle distance of a front vehicle; the state category information discretely distributed over time includes gear information.
Further, the method also comprises the following steps: the preprocessing in S3 includes: and carrying out difference and normalization processing on the data information continuously changing along with time.
Further, the method also comprises the following steps: in S3, after preprocessing the data, the sequence X of 10S history data is set to { X ] with 15S as one cycle1,X2,X3,X4,X5,X6,X7,X8,X9,X10The speed signal value Y of 5s later is { Y ═ Y }as the input of the GRU speed prediction model training process1,y2,y3,y4,y5The GRU is used as the output of a GRU vehicle speed prediction model; wherein, Xj＝{x1,x2,x3,x4,x5,x6}，x1，x2，x3，x4，x5，x6Respectively representing six characteristics of historical speed, historical acceleration, engine speed, engine torque, relative vehicle distance of a front vehicle and gear information; y is1,y2,y3,y4,y5The vehicle speed value for the next 5 seconds.
Further, the method also comprises the following steps: the GRU vehicle speed prediction model in the S2 comprises a GRU layer and a full connection layer;
the GRU layer comprises an update gate, a reset gate, a candidate state and a hidden state; the GRU layer internal expression is as follows:
Zt＝σ(Wz[ht-1,xt]+bz)
rt＝σ(Wr[ht-1,xt]+br)
Figure BDA0003345662580000042
Figure BDA0003345662580000043
wherein x istAs an input vector at the current time, ht-1Is the hidden state vector at the previous moment, WzAnd bzRespectively, the weight matrix and the offset vector of the update gate, sigma is sigmoid function, ZtTo update the activation vector of the gate, rtFor the activation vector of the reset gate, Wr and brWeight matrix and offset vector, z, respectively, of reset gatetAs candidate state vector, WhAnd bhWeight matrix and offset vector, h, respectively, controlling the candidate state informationtIs a hidden state vector at the current moment;
and the full connection layer converts the calculation result output by the GRU layer into a final calculation result of the required dimension.
Further, the method also comprises the following steps: in S5, the current driving style determination process includes: continuously collecting driving speed information, acceleration information and vehicle following distance information in the driving process, and extracting vehicle speed distribution characteristics, acceleration derivative values along with time and average vehicle following relative distance values in the process; and judging the driving style once every 5s, if the driving style does not change after the driving lasts for 1min, determining that the driving style is stable, and introducing a GRU vehicle speed prediction model corresponding to the driving style to predict the vehicle speed.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
FIG. 1 is a flow chart of a method for predicting vehicle speeds of drivers of different styles based on a GRU neural network provided by the invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in FIG. 1, the embodiment of the invention discloses a method for predicting vehicle speeds of drivers in different styles based on a GRU neural network, which comprises the following steps:
s1, dividing the driving database based on a pre-constructed driving style classification model to obtain a plurality of driving sub-databases corresponding to different driving styles;
s2, respectively constructing GRU vehicle speed prediction models corresponding to different driving styles;
s3, respectively extracting training data from each driver sub-database, preprocessing the training data, and generating a training sample sequence database by time step sliding according to a time sequence;
s4, respectively training GRU vehicle speed prediction models corresponding to driving styles by using a training sample sequence database;
and S5, judging the current driving style by adopting the driving style classification model at certain intervals, and introducing a GRU vehicle speed prediction model corresponding to the current driving style to predict the vehicle speed when the current driving style is continuously unchanged within a preset time period.
Next, a process of constructing the driving style classification model according to the embodiment of the present invention will be described in detail.
The driving styles of different drivers can be summarized into aggressive drivers, mediocre drivers and conservative drivers according to the operating characteristics and behavior tendencies of the different drivers. Different types of drivers may have distinct and different driving characteristics during driving, including the severity of acceleration and deceleration, the distribution of speed intervals, and the control of following distance. General definition: the process of acceleration and deceleration is more violent, the vehicle tends to run at high speed under the same road condition, and the vehicle-following distance is closer to define as an aggressive driver; the acceleration and deceleration processes are relatively smooth, the vehicle tends to run at a medium and low speed under the same road condition, and the conservative driver can be defined as the driver with a long vehicle following distance; after the two driving styles are defined, the transition type driving style between the two driver states is defined as a mediocre driver. In the case of the above definition, the identification of the driver style can also include other reasonable reference-value driver behavior characteristic variables.
The driving behavior characteristics which can be used for identifying the style of the driver can be finally obtained by collecting a large amount of historical data of the driving process of the driver or directly calling data information in the existing database and combining a related data analysis method to relatively accurately summarize and extract the driving behavior data of different drivers. The process completes the data preparation process of driver data identification, and then a proper identification model is selected for training, so that good classification effect on historical data can be achieved. Finally, the deployment of the classification model is finished, and the purpose of good identification of newly acquired data can be achieved.
In conclusion, the whole process of identifying the driving style of the driver is divided into four processes of analyzing and processing historical big data, summarizing and extracting style identification characteristic quantities, selecting and training a driving style classification model and re-identifying the driving style after new data is input.
1. Analysis processing of historical big data and induction and extraction of style identification characteristic quantity
The driving style of the driver has a great influence on the driving state during driving, and these influences are directly reflected on the behavior data of the vehicle. A large number of researches show that drivers with aggressive driving styles have large occupation ratio in high-speed driving intervals in the whole driving process, and can finish the realization of the required vehicle speed in a short time, which is reflected in that the acceleration and deceleration processes are violent in the speed change, and the acceleration can be defined as large change rate of the acceleration along with the time, and the aggressive drivers are more inclined to the vehicle distance driving in the following scene. Based on the above intuitive idea, the present embodiment selects the statistics shown in table 1 as the feature quantity for identifying the driving style of the driver.
TABLE 1
Figure BDA0003345662580000071
2. Model selection and training
(1) Selection of identification models
The process of identifying the driving style of the driver is three processes of extracting features of driving data, carrying out cluster analysis on selected features and matching the driving process of the driver with a cluster center, and comprises the following specific steps:
s11, constructing a driving style classification model based on a self-organizing map neural network (SOM) and a K-means clustering method;
s12, performing preliminary clustering of the driver characteristic data through a self-organizing map neural network (SOM);
s13, performing secondary clustering on the primary clustering result by selecting a K-means clustering method, realizing final identification on the style of the driver, and obtaining a plurality of driving sub-databases corresponding to different driving styles.
The SOM neural network has a two-layer structure of an output layer and an input layer, wherein the input layer is determined by input data, and the network structure is mainly determined by the structure of the output layer. The output layer of the SOM neural network has a relatively free geometric structure, and may be a one-dimensional structure or a high-dimensional structure. The total node number of the output layer refers to an empirical formula:
Figure BDA0003345662580000081
wherein N is the total number of nodes and N is the number of sample points. The output layer of the embodiment is a two-dimensional square structure with the side length of
Figure BDA0003345662580000082
And if the condition that the integer division cannot be carried out occurs, rounding up.
After the SOM neural network finishes the initial clustering, the K-means method continues to finish the clustering work. In the present embodiment, the styles of drivers are classified into three types, an aggressive type, a mediocre type, and a conservative type, and thus the cluster center of K-means is set to 3 types.
(2) Training of models
The training process of the SOM neural network comprises the following steps:
and S1201, normalizing the characteristic data in the driving database to obtain dimensionless pure quantity.
S1202, introducing an SOM model, calling a MiniSom library in Python, introducing MiniSom classes, setting the geometric shape in the SOM model to be a rectangle, and setting the side lengths of the two sides to be respectively
Figure BDA0003345662580000083
Figure BDA0003345662580000084
S1203, randomly initializing weights of the SOM model, and initializing the node weights into smaller random numbers.
S1204, setting an initial neighborhood radius sigma, and determining nodes to be included in the winning neighborhood according to the neighborhood radius sigma. Nodes centered at the winning node and within a radius of sigma are updated. Here σ must be greater than 0, otherwise no neuron will be updated; and σ cannot be larger than the side length of the 2-dimensional output plane.
And S1205, setting a type of a proximity function, and calculating the respective updated amplitude of each node through the proximity function. The adjacent function is used for determining the influence of the winner node on the adjacent nodes, and the basic idea is that the closer to the winner node, the larger the updating amplitude is; the more distant the winner node is, the smaller the update amplitude is. More common is a Gaussian function method and a Bubble function method, the former has an updating amplitude which is continuously reduced along with the increase of the distance from the winner node, and the latter has the same updating amplitude of the node in the whole winner neighborhood.
And S1206, setting the initial learning rate alpha, and determining the magnitude of each iteration weight updating. Alpha can be selected from [0,1], and the influence of each iteration on weight updating is determined; the smaller alpha is, the smaller the amplitude of the weight update of each iteration is, and the larger alpha is, the larger the amplitude of the weight update of each iteration is.
S1207, randomly taking an input sample XiPerforming a first iteration; traversing each node in the competition layer and calculating XiAnd similarity between the nodes is represented by Euclidean distance, and the node with the minimum distance is selected as a winner node.
S1208, updating the Weight of the nodes in the preferential neighborhood with the preferential node as the center and sigma as the radius.
Wv(t+1)＝Wv(t)+θ(v,t)·α(t)·(D(i)-Wv(t))。
Wherein s denotes a state corresponding to the current iteration number; v denotes the current node serial number; wv(s) is the current weight of the vth node; wv(s +1) indicates a new weight value after the nth node is updated; θ (v, s) is a constraint on the update based on the distance to the optimal node, i.e., the return value of the proximity function; α(s) indicates a learning rate corresponding to the current state; d (t) refers to the vector corresponding to the current input.
And S1209, after each iteration is finished, carrying out attenuation processing on sigma and alpha. The most common formula for the decay function is:
Figure BDA0003345662580000091
wherein T represents the current iteration number, and T is the total iteration number/2, and finally:
Figure BDA0003345662580000092
Figure BDA0003345662580000093
s1210: and after one round of iteration is finished (the iteration times are plus 1), returning to S1207 until the set iteration times are met, and obtaining a primary clustering result.
The process of clustering the primary clustering result again by using the K-means clustering method comprises the following steps:
s1301, randomly selecting 3 samples [ mu ] from the primary clustering result1，μ2，μ3As an initial centroid;
s1302, calculating each sample YiWith the center of mass mujDistance between (dis | | | Y)j-μj||2Sample YjClosest to the centroid mujFall into one category;
s1303, updating the centroid
Figure BDA0003345662580000101
Wherein ci is a clustering set in the iteration process, and i belongs to {1,2,3 };
and S1304, returning to S1302 until the maximum iteration number is reached or the centroid updating amplitude is smaller than a threshold value, and finally obtaining the driver sub-databases of the three driving styles.
(3) Checksum improvement of output results
And judging whether the three clustering centers can well divide the driving style or not according to the characteristic values of the three clustering centers obtained by final clustering, and if a clustering result with confused driving style types exists, improving the identification result of the driving style by considering the methods of increasing the driving data of the driver again for training and increasing new identification characteristic quantities to obtain a final driving style classification model.
3. Re-identification of driving style after new data entry
After the perfect driving style classification model is obtained, the driving style of the newly input driving data can be identified in real time in the driving process of the driver.
Next, the above-mentioned S1-S6 will be further described.
And S1, dividing the driving database into 3 sub-databases by using the constructed driving style classification model, wherein the 3 sub-databases correspond to a radical driver driving database, a mediocre driver driving database and a conservative driver driving database respectively.
And S2, respectively constructing GRU vehicle speed prediction models corresponding to different driving styles.
The speed prediction aimed by the invention means that the speed distribution after a period of time domain is predicted through reasonable analysis of historical relevant variable data. The speed under the normal driving state has the characteristics of strong randomness and nonlinearity, so that a nonparametric model is adopted to carry out excavation of the speed change rule and prediction of the future speed. The embodiment selects a gated neural network (GRU) to predict the future vehicle speed.
The GRU neural network screens and inherits historical data through a door mechanism to obtain a speed change rule contained in historical information, and accurate prediction of vehicle speed is achieved.
The gated recurrent neural network-based model comprises a GRU layer and a full connection layer.
The GRU layer structure comprises an updating gate, a resetting gate, a candidate state and a hidden state; the internal expression of the GRU hidden layer is as follows:
Zt＝σ(Wz[ht-1,xt]+bz)
rt＝σ(Wr[ht-1,xt]+br)
Figure BDA0003345662580000111
Figure BDA0003345662580000112
wherein x istAs an input vector at the current time, ht-1Is the hidden state vector at the previous moment, WzAnd bzRespectively, the weight matrix and the offset vector of the update gate, sigma is sigmoid function, ZtTo update the activation vector of the gate, rtFor the activation vector of the reset gate, Wr and brWeight matrix and offset vector, z, respectively, of reset gatetAs candidate state vector, WhAnd bhWeight matrix and offset vector, h, respectively, controlling the candidate state informationtIs the hidden state vector at the current moment.
The full connection layer converts the calculation result output by the GRU layer into a final calculation result of a required dimension, and the output dimension of the full connection layer in the embodiment of the invention is 5.
Setting an error evaluation function of the GRU vehicle speed prediction model as a root mean square error, wherein the expression is as follows:
Figure BDA0003345662580000113
wherein m represents the total number of predictions,
Figure BDA0003345662580000114
representing the predicted value of velocity, yiRepresenting the true speed value.
And S3, respectively extracting training data from each driver sub-database, preprocessing the training data, and sliding the training data step by step according to the time sequence to generate a training sample sequence database.
The training data is mainly divided into two parts, one part is input data and is used for generating a predicted value through calculation in a feedforward process, the other part is a label part and is used for calculating a predicted error and further propagating and updating network weights in a reverse mode, in the embodiment, the output part is a vehicle speed signal value, and the input part is a variable signal related to the vehicle speed signal value.
Variables that affect vehicle speed variation can be divided into two categories: the first type is data information continuously changing along with time, most representative is historical vehicle speed information, the information does not have a fixed value range, has large change amplitude and strong volatility, and needs to mine an evolution rule in the data and an internal relation with speed change. The second information is state type information which is distributed discretely along with time, such as gear information during driving, the data has a very clear value range, the same value type has continuity in time, the change of different states is step, and the type information can be used as a type constraint condition of speed change. In the embodiment, the continuous input data comprises historical speed data, historical acceleration data, relative vehicle distance data of a front vehicle, engine torque data and engine rotating speed data; the discrete input data includes gear information.
The following processing may be performed for the discrete data and the continuous data described above: continuous input information continuously changing along with time, including historical speed data, historical acceleration data, vehicle engine rotating speed data, engine torque data and relative vehicle distance data of a front vehicle, needs to be subjected to reasonable difference and normalization processing, so that fitting and convergence of a next model are facilitated; the discrete information is used as constraint limiting information, does not need special processing, and can participate in the training of the model in the form of original label data.
The input data variable value ranges of different characteristics have great difference, and the distance of sample points in a data space is dominated by individual characteristic values, so that all data need to be mapped into the same scale for input information of multiple characteristics. All data in the input data are positive values, and the data are normalized to a [0,1] interval in the embodiment. For each feature vector, the normalized formula is as follows:
Figure BDA0003345662580000121
wherein x isscaleNormalized value, xminRepresents the minimum value, x, in the feature datamaxRepresents the maximum value in the feature quantity, and after processing, all data in the feature quantity are mapped to [0,1]]Within the range.
Specifically, the generation process of the training sample sequence database is as follows:
the input data at each moment comprises the historical speed, the historical acceleration, the engine speed, the engine torque, the relative distance between the front vehicles and the gear information]Six characteristics, taking 15 seconds as a period, and taking a historical data sequence X of 10 seconds of the history as { X ═ X }1,X2,X3,X4,X5,X6,X7,X8,X9,X10Is input, where Xj＝{x1,x2,x3,x4,x5,x6And the information is the preprocessed vehicle state information. The predicted vehicle speed value Y in the next 5 seconds is { Y ═ Y1,y2,y3,y4,y5Is the predicted output, where y1,y2,y3,y4,y5The vehicle speed value for the next 5 seconds. And each sequence sample is generated in a sliding way according to time step by time step, and a training sample sequence database is established.
And S4, respectively training GRU vehicle speed prediction models corresponding to the driving styles by using the training sample sequence database.
And training the corresponding GRU vehicle speed prediction model by using the training sample sequence database established in the step S3 for different driving styles. Flattening 15-second input sample sequence features into one-dimensional vectors of size 15Nf，NfThe number of features representing the input data, N in this embodiment f6. And inputting data to obtain predicted data, optimizing a mean square root error loss function by using a Stochastic Gradient Descent (SGD) optimization algorithm, and reversely propagating a result to update network parameters until the GRU prediction model converges, wherein the model training is considered to be finished.
And S5, judging the current driving style by adopting the driving style identification model at certain intervals, and introducing a GRU vehicle speed prediction model corresponding to the current driving style to predict the vehicle speed when the current driving style is continuously unchanged within a preset time period.
And S51, judging the driving style of the driver.
In the driving behavior process of a driver, continuously collecting driving speed information, acceleration information and vehicle following distance information of the driver, extracting vehicle speed distribution characteristics, acceleration derivative values along with time and average vehicle following relative distance values in the process, judging the driving style of the driver once by adopting a driving style classification model at intervals of 5s, considering that the driving style of the driver is stable when the driving style of the driver does not change for 1min, and introducing a GRU vehicle speed prediction model corresponding to the driving style.
S52, predicting future vehicle speed
And extracting past 10s speed data in the collected historical speed data, inputting 10 historical speed values into a GRU (general-purpose vehicle) speed prediction model by taking each second as a time interval, and outputting 5 future speed values.
S53, driving style supervision and correction
The style characteristics of the driver can change along with time and the surrounding environment, so that the style identification of the driver is continuously carried out in the process of continuously collecting the driving data. When the driving style identification result of the driver changes and lasts for more than 1min, the GRU vehicle speed prediction model is switched, and the step S52 is repeated.
In a more advantageous embodiment, it further comprises:
in a more advantageous embodiment, it further comprises:
s6, online adjustment of model parameters:
and updating the actual vehicle speed information with larger error and historical information thereof in the prediction process into a training sample sequence database, and retraining and adjusting parameters of each GRU vehicle speed prediction model by using the updated training sample sequence database after a preset time interval. And finally, optimizing and upgrading the model by online adjusting the GRU vehicle speed prediction model parameters so as to enhance the generalization and prediction precision of the model.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (10)

1. A method for predicting vehicle speeds of drivers in different styles based on a GRU neural network is characterized by comprising the following steps:
s1, dividing the driving database based on a pre-constructed driving style classification model to obtain a plurality of driving sub-databases corresponding to different driving styles;
s2, respectively constructing GRU vehicle speed prediction models corresponding to different driving styles;
s3, respectively extracting training data from each driver sub-database, preprocessing the training data, and sliding the training data step by step according to a time sequence to generate a training sample sequence database;
s4, respectively training GRU vehicle speed prediction models corresponding to driving styles by using the training sample sequence database;
and S5, judging the current driving style by adopting the driving style identification model at certain intervals, and introducing a GRU vehicle speed prediction model corresponding to the current driving style to predict the vehicle speed when the driving style identification model is continuously unchanged in a preset time period.
2. The method for predicting vehicle speeds of drivers of different styles based on a GRU neural network as claimed in claim 1, further comprising:
and S6, updating the actual vehicle speed information with larger error in the prediction process and the historical information thereof into the training sample sequence database, and retraining and adjusting parameters of each GRU vehicle speed prediction model by using the updated training sample sequence database after a preset time interval.
3. The method for predicting vehicle speeds of drivers of different styles based on a GRU neural network as claimed in claim 1, wherein S1 comprises the following steps:
s11, constructing a driving style classification model based on a self-organizing mapping neural network and a K-means clustering method;
s12, carrying out preliminary clustering on the feature data in the driving database by using a self-organizing mapping neural network;
s13, clustering the primary clustering result again by using a K-means clustering method to obtain a plurality of driving sub-databases corresponding to different driving styles; the driving sub-databases include an aggressive driving database, a mediocre driving database, and a conservative driving database.
4. The method for predicting vehicle speeds of drivers of different styles based on a GRU neural network as claimed in claim 3, wherein S12 comprises the following steps:
s1201, normalizing the characteristic data in the driving database to obtain dimensionless pure quantity;
s1202, introducing an SOM model, and setting the geometric shape in the SOM model to be a rectangle;
s1203, randomly initializing weights of the SOM model, and initializing the node weights into smaller random numbers;
s1204, setting an initial neighborhood radius sigma, and determining nodes to be included in a winning neighborhood according to the neighborhood radius sigma;
s1205, setting a proximity function type, and calculating the respective updating amplitude of each node through the proximity function;
s1206, setting an initial learning rate alpha, and determining the magnitude of each iteration weight updating;
s1207, randomly taking an input sample XiPerforming a first iteration; traversing each node in the competition layer and calculating XiSelecting the node with the minimum distance as a winner node according to the similarity between the node and the node;
s1208, updating the Weight of the nodes in the preferential neighborhood with the preferential node as the center and sigma as the radius range;
s1209, after each iteration is finished, carrying out attenuation processing on sigma and alpha;
s1210: and after one round of iteration is finished, returning to S1207 until the set iteration times are met, and obtaining a primary clustering result.
5. The method for predicting vehicle speeds of drivers of different styles based on a GRU neural network as claimed in claim 4, wherein S13 comprises:
s1301, randomly selecting 3 samples { mu ] from the primary clustering result1，μ2，μ3As an initial centroid;
s1302, calculating each sample YiWith the center of mass mujDistance between (dis | | | Y)j-μj||2Sample YjClosest to the centroid mujFall into one category;
s1303, updating the centroid
Figure FDA0003345662570000021
Wherein ci is a clustering set in the iteration process, and i belongs to {1,2,3 };
and S1304, returning to S1302 until the maximum iteration number is reached or the centroid updating amplitude is smaller than a threshold value, and finally obtaining the driver sub-databases of the three driving styles.
6. The method of claim 1, wherein the training data extracted at S3 comprises input data and output data, wherein the input data comprises: the data information continuously changing along with time and the state category information discretely distributed along with time, and the output data are vehicle speed signals; the data information continuously changing with time includes: historical vehicle speed, historical acceleration, engine speed, engine torque and relative vehicle distance of a front vehicle; the state category information discretely distributed over time includes gear information.
7. The method for predicting vehicle speeds of drivers of different styles based on the GRU neural network as claimed in claim 6, wherein the preprocessing in S3 comprises: and carrying out difference and normalization processing on the data information continuously changing along with time.
8. The method for predicting vehicle speeds of drivers with different styles based on GRU neural network as claimed in claim 6, wherein in S3, after preprocessing the data, 15S is taken as a period, and 10S historical data sequence X ═ X is taken1,X2,X3,X4,X5,X6,X7,X8,X9,X10The speed signal value Y of 5s later is { Y ═ Y }as the input of the GRU speed prediction model training process1,y2,y3,y4,y5The GRU is used as the output of a GRU vehicle speed prediction model; wherein, Xj＝{x1,x2,x3,x4,x5,x6}，x1，x2，x3，x4，x5，x6Respectively representing six characteristics of historical speed, historical acceleration, engine speed, engine torque, relative vehicle distance of a front vehicle and gear information; y is1,y2,y3,y4,y5The vehicle speed value for the next 5 seconds.
9. The method for predicting vehicle speeds of drivers of different styles based on the GRU neural network as claimed in claim 1, wherein the GRU vehicle speed prediction model in S2 comprises a GRU layer and a full connection layer;
the GRU layer comprises an update gate, a reset gate, a candidate state and a hidden state; the GRU layer internal expression is as follows:
Zt＝σ(Wz[ht-1,xt]+bz)
rt＝σ(Wr[ht-1,xt]+br)
Figure FDA0003345662570000031
Figure FDA0003345662570000032
wherein x istAs an input vector at the current time, ht-1Is the hidden state vector at the previous moment, WzAnd bzRespectively, the weight matrix and the offset vector of the update gate, sigma is sigmoid function, ZtTo update the activation vector of the gate, rtFor the activation vector of the reset gate, Wr and brWeight matrix and offset vector, z, respectively, of reset gatetAs candidate state vector, WhAnd bhWeight matrix and offset vector, h, respectively, controlling the candidate state informationtIs a hidden state vector at the current moment;
and the full connection layer converts the calculation result output by the GRU layer into a final calculation result of the required dimension.
10. The method for predicting vehicle speeds of drivers with different styles based on the GRU neural network as claimed in claim 1, wherein in S5, the current driving style is determined by the following steps: continuously collecting driving speed information, acceleration information and vehicle following distance information in the driving process, and extracting vehicle speed distribution characteristics, acceleration derivative values along with time and average vehicle following relative distance values in the process; and judging the driving style once every 5s, if the driving style does not change after the driving lasts for 1min, determining that the driving style is stable, and introducing a GRU vehicle speed prediction model corresponding to the driving style to predict the vehicle speed.
CN202111321247.XA 2021-11-09 2021-11-09 Method for predicting vehicle speeds of drivers in different styles based on GRU neural network Pending CN113947259A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111321247.XA CN113947259A (en) 2021-11-09 2021-11-09 Method for predicting vehicle speeds of drivers in different styles based on GRU neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111321247.XA CN113947259A (en) 2021-11-09 2021-11-09 Method for predicting vehicle speeds of drivers in different styles based on GRU neural network

Publications (1)

Publication Number Publication Date
CN113947259A true CN113947259A (en) 2022-01-18

Family

ID=79337112

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111321247.XA Pending CN113947259A (en) 2021-11-09 2021-11-09 Method for predicting vehicle speeds of drivers in different styles based on GRU neural network

Country Status (1)

Country Link
CN (1) CN113947259A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115879646A (en) * 2023-02-15 2023-03-31 山东捷讯通信技术有限公司 Reservoir water level prediction method, device, medium and equipment
CN117131955A (en) * 2023-10-27 2023-11-28 北京航空航天大学 Short-time vehicle speed prediction method considering multiple constraint conditions
CN117325875A (en) * 2023-12-01 2024-01-02 北京航空航天大学 Vehicle long-term speed prediction method based on individual driving characteristics
CN117349386A (en) * 2023-10-12 2024-01-05 吉玖(天津)技术有限责任公司 Digital humane application method based on data strength association model

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115879646A (en) * 2023-02-15 2023-03-31 山东捷讯通信技术有限公司 Reservoir water level prediction method, device, medium and equipment
CN115879646B (en) * 2023-02-15 2023-11-07 山东捷讯通信技术有限公司 Reservoir water level prediction method, device, medium and equipment
CN117349386A (en) * 2023-10-12 2024-01-05 吉玖(天津)技术有限责任公司 Digital humane application method based on data strength association model
CN117349386B (en) * 2023-10-12 2024-04-12 吉玖(天津)技术有限责任公司 Digital humane application method based on data strength association model
CN117131955A (en) * 2023-10-27 2023-11-28 北京航空航天大学 Short-time vehicle speed prediction method considering multiple constraint conditions
CN117131955B (en) * 2023-10-27 2024-01-16 北京航空航天大学 Short-time vehicle speed prediction method considering multiple constraint conditions
CN117325875A (en) * 2023-12-01 2024-01-02 北京航空航天大学 Vehicle long-term speed prediction method based on individual driving characteristics
CN117325875B (en) * 2023-12-01 2024-02-02 北京航空航天大学 Vehicle long-term speed prediction method based on individual driving characteristics

Similar Documents

Publication Publication Date Title
CN113947259A (en) Method for predicting vehicle speeds of drivers in different styles based on GRU neural network
CN107862864B (en) Driving condition intelligent prediction estimation method based on driving habits and traffic road conditions
CN106228185B (en) A kind of general image classifying and identifying system neural network based and method
Wu et al. A data mining approach for spatial modeling in small area load forecast
CN111291678B (en) Face image clustering method and device based on multi-feature fusion
CN107909179B (en) Method for constructing prediction model of running condition of plug-in hybrid vehicle and vehicle energy management method
JP4681426B2 (en) Apparatus and method for analyzing relation between operation and quality in manufacturing process, computer program, and computer-readable recording medium
CN101556650B (en) Distributed self-adapting pulmonary nodule computer detection method and system thereof
CN109472088B (en) Shale gas-conditioned production well production pressure dynamic prediction method
CN111814897A (en) Time series data classification method based on multi-level shape
CN116187835A (en) Data-driven-based method and system for estimating theoretical line loss interval of transformer area
JP4653547B2 (en) Apparatus and method for analyzing relation between operation and quality in manufacturing process, computer program, and computer-readable recording medium
CN113762370A (en) Depth network set generation method combined with Gaussian random field
US7797180B2 (en) Method and system for comparing populations of entities to make predictions about business locations
CN111539444A (en) Gaussian mixture model method for modified mode recognition and statistical modeling
CN114912195B (en) Aerodynamic sequence optimization method for commercial vehicle
CN111639688A (en) Local interpretation method of Internet of things intelligent model based on linear kernel SVM
CN110837853A (en) Rapid classification model construction method
Jozová et al. Bayesian Mixture Estimation without Tears.
CN112949524B (en) Engine fault detection method based on empirical mode decomposition and multi-core learning
Purnomo et al. Synthesis ensemble oversampling and ensemble tree-based machine learning for class imbalance problem in breast cancer diagnosis
CN113656707A (en) Financing product recommendation method, system, storage medium and equipment
Cho et al. Data clustering method using efficient fuzzifier values derivation
De Moura et al. Extraction of vehicle behaviors at intersections
Shah et al. Fuzzy based selection of PWARX model for the nonlinear hybrid dynamical systems

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination