CN-118824036-B - Road self-adaption method, device and medium suitable for ICV driving model

CN118824036BCN 118824036 BCN118824036 BCN 118824036BCN-118824036-B

Abstract

The invention discloses a road self-adaptation method, a device and a medium suitable for an ICV driving model, wherein the method comprises the steps of obtaining a pre-training sample data set, preprocessing the pre-training sample data set, training an intelligent driving model by adopting the pre-training sample data set, judging whether a vehicle is in an unfamiliar road section, obtaining vehicle track data of an unfamiliar road as a self-adaptation sample set if the vehicle is in the unfamiliar road section, preprocessing the self-adaptation sample set, adopting a self-adaptation fine tuning strategy to finely tune the intelligent driving model according to the preprocessed self-adaptation sample set, obtaining a new intelligent driving model, and adding the obtained self-adaptation sample set into the pre-training sample data set to update the pre-training sample data set. The invention adopts the self-adaptive fine tuning strategy to fine tune the model parameters, so that the model parameters are suitable for the road environment of the unfamiliar road section, the safety and the adaptability of the ICV on the unfamiliar road section are improved, and the model can be widely applied to the fields of intelligent traffic, intelligent driving models and the like.

Inventors

HUANG LING
XU ZONGQING
HUANG XINGYU
Zhong Haochuan

Assignees

华南理工大学

Dates

Publication Date: 20260505
Application Date: 20240603

Claims (8)

1. The road self-adaptation method suitable for the ICV driving model is characterized by comprising the following steps of: acquiring a pre-training sample data set, and preprocessing the acquired pre-training sample data set; training an intelligent driving model by adopting the preprocessed pre-training sample data set; If the vehicle is in the unfamiliar road section, acquiring vehicle track data of the unfamiliar road as an adaptive sample set, and preprocessing the acquired adaptive sample set; according to the preprocessed self-adaptive sample set, adopting a self-adaptive fine tuning strategy to fine tune the intelligent driving model to obtain a new intelligent driving model; adding the obtained adaptive sample set into the pre-training sample data set to update the pre-training sample data set; the adoption self-adaptation fine setting strategy carries out fine setting to intelligent driving model, obtains new intelligent driving model, includes: the parameters of the intelligent driving model are finely adjusted by adopting an adaptive fine adjustment strategy, wherein the weight updating method adopts a root mean square propagation method, and a self-adaptive performance is introduced A variable representing the passage of a reference intelligent driving model The ratio of performance improvement after wheel fine tuning is expressed as: In the formula, The migration performance of the model is represented, Representing the passing of Model performance after round iteration; with mean square error MSE and adaptive performance Measuring the performance improvement of the intelligent driving model after parameter fine adjustment, judging that the model achieves the optimal performance when the difference between the adaptive performance of two adjacent iterations is smaller than a preset threshold value, ending the parameter adjustment of the model, and outputting the intelligent driving model with the optimal performance; For each training sample, the weight is updated by adopting a root mean square propagation method, and the expression is as follows: In the formula, As the amount of change in the parameter, To fine tune the learning rate; is a number preventing denominator from being 0; Is the first The gradient of the secondary training is that, Is in front of The cumulative square gradient of the secondary training, Is an attenuation factor.
2. A road adaptation method for an ICV driving model according to claim 1, wherein the acquiring a pre-training sample dataset and preprocessing the acquired pre-training sample dataset comprises: acquiring a preset number of vehicle track data as a pre-training sample data set; processing abnormal data in the pre-training sample data set by adopting a threshold method; and performing expansion processing on the data set after the exception processing by adopting a preset data expansion mode.
3. The road adaptation method applicable to an ICV driving model according to claim 1, wherein the intelligent driving model is a data-driven-based intelligent driving model; The training of the intelligent driving model by the pre-training sample data set after pretreatment comprises the following steps: training the intelligent driving model by adopting the preprocessed pre-training sample data set, and measuring the training effect by a preset evaluation standard to select the model with the best performance as a reference intelligent driving model.
4. The road adaptation method for an ICV driving model according to claim 1, wherein the determining whether the vehicle is in a strange road section comprises: And acquiring positioning data of the vehicle according to a preset period, and judging whether the road section where the vehicle is currently positioned is a strange road section according to the positioning data.
5. The road adaptive method suitable for the ICV driving model of claim 1, wherein the adaptive sample set data is obtained by two modes, namely a vehicle-to-vehicle communication mode and a vehicle-to-vehicle communication mode.
6. A road adaptation device adapted for use in ICV driving models for implementing the method of any one of claims 1-5, comprising: the data set construction module is used for acquiring a pre-training sample data set and preprocessing the acquired pre-training sample data set; The model pre-training module is used for training an intelligent driving model by adopting the pre-training sample data set after the pretreatment; The self-adaptive data acquisition module is used for judging whether the vehicle is in an unfamiliar road section, if so, acquiring vehicle track data of the unfamiliar road as a self-adaptive sample set, and preprocessing the acquired self-adaptive sample set; The model fine adjustment module is used for fine adjusting the intelligent driving model by adopting a self-adaptive fine adjustment strategy according to the preprocessed self-adaptive sample set to obtain a new intelligent driving model; And the data set updating module is used for adding the obtained adaptive sample set into the pre-training sample data set to update the pre-training sample data set.
7. A road adaptation device suitable for ICV driving models, comprising: At least one processor; At least one memory for storing at least one program; the at least one program, when executed by the at least one processor, causes the at least one processor to implement the method of any one of claims 1-5.
8. A computer readable storage medium, in which a processor executable program is stored, characterized in that the processor executable program is for performing the method according to any of claims 1-5 when being executed by a processor.

Description

Road self-adaption method, device and medium suitable for ICV driving model Technical Field The invention relates to the fields of intelligent traffic, intelligent driving models and the like, in particular to a road self-adaptation method, device and medium suitable for an ICV driving model. Background With the rapid development of intelligent traffic systems, intelligent internet-connected vehicles (INTELLIGENT AND Connected Vehicle, ICV) are gradually showing great potential in terms of improving traffic efficiency, enhancing driving safety and the like as an important ring, so that intelligent driving model research on ICV attracts attention of relevant researchers and becomes a current hot research problem. The ICV is widely applied and industrialized, and the safety of the ICV in different road environments needs to be ensured, namely, the intelligent driving model can adapt to different road environments. In order to solve the problem, the existing intelligent driving model based on data driving mainly comprises the steps of optimizing a network structure and expanding a training data set. The method for optimizing the network structure often depends on the complexity and a large amount of calculation force of the original pre-training intelligent driving model network structure, so that the training efficiency is increased, the adaptation performance is improved relatively poorly, and the adaptation requirement of ICV can not be well met. The data set expansion method mainly comprises joint updating and incremental updating, wherein the joint updating is a method for updating a model by combining historical data and new data, can thoroughly solve the model adaptability problem of different scenes, but has very high calculation and storage cost and poor practical application capability along with the increase of the data set, and the incremental updating is a method for updating the model by the new data in a small scale, so that the training efficiency is higher. Disclosure of Invention In order to solve at least one of the technical problems existing in the prior art to a certain extent, the invention aims to provide a road self-adaptation method, a device and a medium suitable for an ICV driving model. The first technical scheme adopted by the invention is as follows: A road adaptation method suitable for ICV driving models, comprising the steps of: acquiring a large number of pre-training sample data sets, and preprocessing the acquired pre-training sample data sets; training an intelligent driving model by adopting the preprocessed pre-training sample data set; If the vehicle is in the unfamiliar road section, acquiring vehicle track data of a small number of unfamiliar roads as an adaptive sample set, and preprocessing the acquired adaptive sample set; according to the preprocessed self-adaptive sample set, adopting a self-adaptive fine tuning strategy to fine tune the intelligent driving model to obtain a new intelligent driving model; The obtained adaptive sample set is added to the pre-training sample data set to update the pre-training sample data set. Further, the acquiring the pre-training sample data set and preprocessing the obtained pre-training sample data set includes: acquiring a preset number of vehicle track data as a pre-training sample data set; processing abnormal data in the pre-training sample data set by adopting a threshold method; and performing expansion processing on the data set after the exception processing by adopting a preset data expansion mode. Further, the intelligent driving model is an intelligent driving model based on data driving; The training of the intelligent driving model by the pre-training sample data set after pretreatment comprises the following steps: training the intelligent driving model by adopting the preprocessed pre-training sample data set, and measuring the training effect by a preset evaluation standard to select the model with the best performance as a reference intelligent driving model. Further, the determining whether the vehicle is in a strange road section includes: And acquiring positioning data of the vehicle according to a preset period, and judging whether the road section where the vehicle is currently positioned is a strange road section according to the positioning data. Further, the self-adaptive sample set data acquisition modes include a vehicle-to-vehicle communication mode and a vehicle-to-vehicle communication mode. Further, the fine tuning of the intelligent driving model by adopting the adaptive fine tuning strategy to obtain a new intelligent driving model includes: The method adopts a root mean square propagation method (RMSprop), introduces a variable named as self-adaptive performance alpha (n), and the variable represents the performance improvement proportion of the basic intelligent driving model after n (n > 0) rounds of fine tuning, and is expressed as follows: Where MSE 0 represents the migration performance o