CN115547047A - A car-following model for intelligent networked vehicles based on attention model - Google Patents

Abstract

The present invention relates to the technical field of car-following vehicles, in particular to an attention model-based intelligent networked car-following model, and a data acquisition module for acquiring historical vehicle-following data, the historical vehicle-following data including vehicle information, Vehicle distance information; the model building module is used to construct a BP neural network model based on the obtained historical vehicle following data and using a neural network algorithm; it is also used to construct a Gipps following vehicle model based on the obtained historical vehicle following data; The linear combination module is used to perform linear combination based on the constructed BP neural network model and the Gipps car-following model to generate a corresponding linear combination prediction model; Car-following data, predicting the current car-following speed of the car-following vehicle. This solution can realize the true prediction of the following speed of the following vehicle under the premise of ensuring the safety of the following vehicle.

Description

A car-following model for intelligent connected vehicles based on attention model

technical field

The invention relates to the technical field of car-following vehicles, in particular to an attention model-based intelligent networked car-following model.

Background technique

Car-following is an important part of micro-traffic flow theory, which studies the car-following behavior of vehicles on a single lane. The study found that car-following behavior has an important impact on the characteristics of traffic flow and traffic simulation results, and car-following is also an indispensable key component in intelligent networked vehicles and adaptive cruise control systems.

Most of the existing car-following models based on deep learning predict the speed of the next moment based on the data at the current moment. The prediction of the car-following speed at the next moment through this deep learning method can indeed be more realistic Restore the speed of the car-following vehicle at the next moment, and truly reflect the situation of the car-following scene. However, for the behavior of car-following, not only the authenticity, but also safety must be considered. After all, in the process of car-following In the process, there is no guarantee that there will be no accidents during the car-following process.

Based on this, an attention-based car-following model for intelligent connected vehicles is needed, which can realize the true prediction of the car-following speed of the car-following vehicle while taking into account the safety of the car-following vehicle.

Contents of the invention

The present invention intends to provide an intelligent networked car-following model based on an attention model, which can realize the real prediction of the car-following speed of the car-following vehicle under the premise of ensuring the safety of the car-following vehicle.

In order to achieve the above object, the present invention adopts the following technical solution: a car-following model of intelligent networked vehicles based on attention model, including:

The data acquisition module is used to obtain historical vehicle following data, and the historical vehicle following data includes vehicle information and vehicle distance information; the vehicle information includes the speed of the following vehicle, the acceleration of the following vehicle, the speed of the preceding vehicle and the acceleration of the vehicle in front;

The model construction module is used to construct a BP neural network model based on the obtained historical vehicle following data and using neural network algorithms; it is also used to construct the Gipps following vehicle model according to the obtained historical vehicle following data;

The linear combination module is used to perform a linear combination according to the constructed BP neural network model and the Gipps car-following model to generate a corresponding linear combination prediction model;

The speed prediction module is used to use the linear combination model to predict the current speed of the following vehicle according to the vehicle following data at the previous moment.

The principle and advantages of this scheme are: in this scheme, the historical vehicle following data is first collected, and then the corresponding BP neural network model and Gipps following model are respectively constructed using the collected historical vehicle following data, and then the linear combination is used to The prediction method linearly combines the two models to form a new linear combination prediction model, and finally uses the vehicle following data at the previous moment to bring it into the new linear combination prediction model, and outputs the corresponding vehicle following at the current moment speed.

In this scheme, the linear combination prediction model is used in the prediction of vehicle following speed, that is, the coupling of BP neural network model and Gipps car following model, so that the final vehicle following speed can ensure that the vehicle is At the same time, it can restore the real car-following speed of the car-following vehicle as much as possible. Compared with a single model, for example, the car-following speed predicted by the BP neural network model is closer to reality, but the corresponding security It cannot be guaranteed, and the car-following speed predicted by the Gipps car-following model can indeed ensure the safety of vehicles, but the corresponding car-following speed is not realistic enough, which makes the corresponding data in the test relatively unreal and accurate. The corresponding test results are not representative. This scheme couples the two models so that the final car-following data is as close as possible to the real car-following speed under the premise of ensuring safety. more convincing and reliable.

Preferably, as an improvement, it also includes a BP neural network model update module, which is used to initialize the weight parameters and bias parameters in the BP neural network model, and the historical vehicle following data is input to the BP neural network model, after continuous training Learn and update weight parameters and bias parameters to form a new BP neural network model;

The model calibration module is used to calibrate the parameters in the Gipps car-following model by using historical car-following data to form a new Gipps car-following model.

Beneficial effects: The data update of the BP neural network model and the Gipps car-following model is realized through the settings of the BP neural network model update module and the model calibration module, so that the parameters corresponding to the two models meet the requirements, which makes the use of these two The corresponding output results of the model will be more accurate and reliable.

Preferably, as an improvement, it also includes an optimal weighting coefficient matching module, which is used to match the corresponding optimal weighting coefficient according to the formed linear combination prediction model and in combination with the prediction requirements; the prediction requirements include safety and real sex.

Beneficial effects: In this solution, when operators use the linear combination forecasting model, different scenarios correspond to different requirements. According to the forecasting requirements, some need to consider safety, and some need to consider reality. Based on these two points, both Carry out the optimal weighting coefficients corresponding to the prediction requirements. By matching the optimal weighting coefficients of these two requirements, it is possible to predict the car-following speed that meets the requirements in the first time when performing speed prediction, which greatly improves the following speed. The predicted speed of the gallop.

Preferably, as an improvement, it also includes a data screening module, which is used to select qualified vehicle following data from the following database to obtain corresponding historical vehicle following data.

Beneficial effect: by screening the car-following data in the car-following database, the validity and reliability of the data are greatly improved, and at the same time, the data is universal.

Preferably, as an improvement, the data screening module includes:

The first selection module is used to select vehicle following data under representative general road traffic environmental conditions in the following database;

The second selection module is used to select vehicle following data in the form of following on the same lane from the vehicle following data selected by the first selection module to form corresponding historical vehicle following data.

Beneficial effect: when screening the car-following data, the first thing is to select representative general road traffic environment conditions, so that the selected data will be more representative and universal, so that the application of the whole system in the specific prediction and use Stronger.

Description of drawings

FIG. 1 is a logical block diagram of an attention model-based car-following model for intelligent connected vehicles in Embodiment 1 of the present invention.

detailed description

The following is further described in detail through specific implementation methods:

The embodiment is basically as shown in accompanying drawing 1: a car-following model of an intelligent networked car based on an attention model, comprising:

The data acquisition module is used to obtain historical vehicle following data, and the historical vehicle following data includes vehicle information and vehicle distance information; the vehicle information includes the speed of the following vehicle, the acceleration of the following vehicle, the speed of the preceding vehicle and The acceleration of the vehicle in front; when acquiring data, it is necessary to select the data.

The data screening module is used to select the car-following data that meets the requirements in the car-following database, and obtain the corresponding historical car-following data;

The data screening module includes:

The first selection module is used to select vehicle following data under representative general road traffic environmental conditions in the following database;

The second selection module is used to select vehicle following data in the form of following on the same lane from the vehicle following data selected by the first selection module to form corresponding historical vehicle following data.

Specifically include: 1) Selecting car-following data under representative general road traffic conditions (roads, drivers, driving tracks, weather, etc.);

2) When selecting data, take the car-following vehicle and the car-followed vehicle as a whole, and regard them as a car-following unit;

3) The driving characteristics of the two vehicles in each car-following unit are: following in the same lane (the car-following vehicle neither changes lanes nor overtakes);

4) When the distance between the car-following vehicle and the car-followed vehicle in a car-following unit is too large, it is considered that the car-following feature of this piece of data is not obvious, and the driving trajectory data of these two vehicles are screened out;

5) Only select the car-following unit whose car-following behavior lasts for 26 seconds, and extract the driving trajectory data of the two vehicles in this car-following unit. When using specific data, 70% of the historical car-following data are used for model training, and other data are used for testing the model.

其中其中l为隐层的节点数，n为输入层的节点数，m为输出层的节点数，a为1至10之间的一个数，本实施例中取为6，因此隐层共有8个节点。BP神经网络通常采用Sigmoid可微函数和线性函数作为网络的激励函数。本文选择S型正切函数tansig作为隐层神经元的激励函数。预测模型选取S型对数函数tansig作为输出层神经元的激励函数。The model construction module is used to construct a BP neural network model based on the obtained historical vehicle following data using a neural network algorithm; in the present embodiment, the BP neural network model uses BP neural network technology to predict the following speed, Concretely at first construct a three-layer BP neural network model, including an input layer, a hidden layer and an output layer. 1. Vehicle distance information is used as the input of the input layer, so the input layer has 5 nodes, and the output is the vehicle following speed corresponding to the current moment, so there is one node in total. For the hidden layer, this embodiment uses the following formula to determine the hidden layer Number of nodes:

Among them, l is the number of nodes in the hidden layer, n is the number of nodes in the input layer, m is the number of nodes in the output layer, and a is a number between 1 and 10, which is 6 in this embodiment, so there are 8 hidden layers nodes. BP neural network usually adopts Sigmoid differentiable function and linear function as the activation function of the network. This paper chooses the sigmoid tangent function tansig as the activation function of the hidden layer neurons. The prediction model selects the S-type logarithmic function tansig as the excitation function of the output layer neurons.

It is also used to construct the Gipps car-following model based on the obtained historical car-following data;

In the present embodiment, the basic expression of the Gipps car-following model is as follows:

Among them: α _n+1 , V _n+1 , b _n+1 , b _n are the parameters to be calibrated. α _n+1 represents the maximum acceleration that vehicle n+1 can adopt; V _n+1 represents the expected speed of vehicle n+1 in this traffic environment; b _n+1 and b _n represent the speed of vehicle n+1 respectively. Maximum deceleration and maximum deceleration of vehicle n.

It also includes the BP neural network model update module, which is used to initialize the weight parameters and bias parameters in the BP neural network model, and input the historical vehicle following data into the BP neural network model, and update the weight parameters and bias after continuous training and learning parameters to form a new BP neural network model;

The model calibration module is used to calibrate the parameters in the Gipps car-following model by using historical car-following data to form a new Gipps car-following model. In this embodiment, the parameters V _n and α _n in Gipps are calibrated using the toolbox of MATLAB and the collected car-following data. Through the speed, acceleration, and distance of the car-following vehicle, the speed of the car and acceleration determine the values of the V _n and α _n parameters in the Gipps model.

The linear combination module is used to perform a linear combination according to the constructed BP neural network model and the Gipps car-following model to generate a corresponding linear combination prediction model;

The speed prediction module is used to use the linear combination model to predict the current speed of the following vehicle according to the vehicle following data at the previous moment.

It also includes an optimal weighting coefficient matching module, which is used to match the corresponding optimal weighting coefficient according to the formed linear combination prediction model and combining with prediction requirements; the prediction requirements include safety and authenticity.

Embodiment two

Compared with Embodiment 1, this embodiment is different in that it also includes a working condition identification module, which is used to identify the working condition of the following vehicle's following path after obtaining the vehicle following data at the previous moment information;

The similarity matching module is used to match and judge the similarity between the historical working condition information in the database and the working condition information according to the identified working condition information, and judge whether the corresponding similarity is greater than the preset similarity threshold. Exist, then call out the historical vehicle information corresponding to the historical operating condition information and the historical real current following speed, if not exist, then predict the current following speed of the following vehicle by the speed prediction module;

A processing module, configured to: if the identified historical working condition information is the same as the working condition information of the car-following vehicle, the current following speed of the car-following vehicle is the historical real current following speed corresponding to the historical working condition information ;

If the identified historical working condition information is similar to but not the same as the working condition information of the following vehicle, then according to the working condition information of the following vehicle, the corresponding working condition type is identified, and the corresponding working condition type is called adjust the strategy;

The adjustment module is used to adjust the historical real current car-following speed corresponding to the historical working condition information according to the corresponding adjustment strategy, and then the corresponding current car-following speed of the following car is the adjusted historical real current car-following speed speed;

The storage module is used to store historical working condition information, corresponding historical vehicle following data, and historical real current following speed; it is also used to adjust the historical real current eradication speed according to the type of working condition.

In this scheme, after obtaining the vehicle following data at the previous moment, the working condition information corresponding to the path of the following vehicle will be identified, and then the historical working condition information similar to the working condition information will be identified from the database. Matching with the condition information, that is, calculating and judging the similarity of the historical working condition information. Only when the corresponding similarity is greater than the preset similarity threshold, will the corresponding historical vehicle information and corresponding historical working condition information be called out. The history is true and the current car-following speed, otherwise, the speed prediction module will be used to predict the current car-following speed of the car-following vehicle. In this embodiment, when matching the similarity of the working condition information, not only the similarity of the vehicle driving environment, but also the similarity of the vehicle information is considered.

Afterwards, the similarity will be judged. If the corresponding historical working condition information is the same as the current working condition information, in this embodiment, the same working condition information means that the corresponding vehicle driving environment and vehicle information are the same. Then it will be determined that the car-following data this time should be the same as the historical working condition information, that is, the corresponding current following car-following speed of the car-following vehicle should be the historical real current following car-following speed corresponding to the historical working condition information.

Of course, if the corresponding historical working condition information is similar to but not the same as the working condition information of the car-following vehicle, that is, the corresponding vehicle driving environment is the same, but the corresponding vehicle information is not the same, then it will be based on the working condition information of the car-following vehicle. , identify the corresponding working condition type, and call the adjustment strategy corresponding to the working condition type, and then adjust the historical real current car-following speed corresponding to the historical working condition information according to the corresponding adjustment strategy, and then the corresponding The current car-following speed of the car-following vehicle is the adjusted historical real current car-following speed. Among them, the adjustment strategy means that different types of working conditions have different adjustment strategies. For example, in the same road section, the corresponding adjustment strategies are different under different weather conditions, such as sunny days and snowy days. When the real current car-following speed is adjusted, the historical real current car-following speed will be reduced as much as possible, fully considering that the snowy road surface is relatively slippery, and the braking distance of the vehicle will be longer during the car-following process. By reducing the historical real current car-following speed The current car-following speed of the car-following vehicle will be smaller than the historical real current car-following speed, and on sunny days, it is possible to maintain or appropriately increase the corresponding historical real current car-following speed, so that the car-following vehicle's current car-following speed Compared with the historical real current car-following speed, the current car-following speed will be greater or unchanged. When making specific adjustments, it will match the braking distance corresponding to the historical real current car-following speed under the working condition type according to the corresponding working condition type, and calculate the corresponding braking distance under the current working condition type and the working condition The difference of the braking distance of each type, and the corresponding adjustment ratio is formed according to the corresponding difference. The adjustment amount of the historical real current car-following speed is the value corresponding to the product of the corresponding historical real current car-following speed and the adjustment ratio. The adjustment amount of the corresponding historical real current car-following speed is associated with the corresponding braking distance, which greatly improves the accuracy of the adjustment amount and greatly improves the safety and accuracy of the corresponding adjusted car-following speed.

Of course, the corresponding adjustment strategies are also different under other types of working conditions. For example, snowy days and rainy days are also slippery road environments, but the corresponding slippery degrees are different. The corresponding adjustment strength during adjustment will also be different.

This solution extracts the data corresponding to similar historical working condition information, and matches different adjustment strategies by judging the working condition type, so as to obtain the working condition information that is more in line with the car-following vehicle The current car-following speed is lower, as far as possible to ensure the accurate judgment of the current car-following speed, and to ensure the safety during the car-following process, which greatly improves the safety of car-following.

What is described above is only an embodiment of the present invention, and common knowledge such as specific technical solutions and/or characteristics known in the solutions are not described here too much. It should be pointed out that for those skilled in the art, without departing from the technical solutions of the present invention, some modifications and improvements can also be made, which should also be regarded as the protection scope of the present invention, and these will not affect the implementation of the present invention effect and utility of patents. The scope of protection required by this application shall be based on the content of the claims, and the specific implementation methods and other records in the specification may be used to interpret the content of the claims.

Claims (5)

1. The utility model provides an intelligent networking car following model based on attention model which characterized in that: the method comprises the following steps:

the data acquisition module is used for acquiring historical vehicle-following data, and the historical vehicle-following data comprises vehicle information and vehicle distance information; the vehicle information comprises the speed of a following vehicle, the acceleration of the following vehicle, the speed of a preceding vehicle and the acceleration of the preceding vehicle;

the model building module is used for building a BP neural network model by utilizing a neural network algorithm according to the obtained historical vehicle following data; the method is also used for constructing a Gipps following model according to the acquired historical vehicle following data;

the linear combination module is used for carrying out linear combination according to the constructed BP neural network model and the Gipps car-following model to generate a corresponding linear combination prediction model;

and the speed prediction module is used for predicting the current following speed of the following vehicle according to the vehicle following data at the last moment by using the linear combination model.

2. The intelligent networked car following model based on the attention model as claimed in claim 1, wherein: the system also comprises a BP neural network model updating module, a weight parameter updating module and a bias parameter updating module, wherein the BP neural network model updating module is used for initializing the weight parameter and the bias parameter in the BP neural network model, inputting historical car following data into the BP neural network model, and updating the weight parameter and the bias parameter through continuous training and learning to form a new BP neural network model;

and the model calibration module is used for calibrating parameters in the Gipps following model by using the historical car following data to form a new Gipps following model.

3. The intelligent networked car following model based on the attention model as claimed in claim 2, wherein: the optimal weighting coefficient matching module is used for matching a corresponding optimal weighting coefficient according to the formed linear combination prediction model and in combination with the prediction requirement; the prediction requirements include security and authenticity.

4. The intelligent internet vehicle following model based on the attention model as claimed in claim 3, wherein: the system further comprises a data screening module, wherein the data screening module is used for selecting the car following data meeting the requirements from the following database to obtain the corresponding historical car following data.

5. The intelligent networked car following model based on the attention model as claimed in claim 4, wherein: the data screening module comprises:

the first selection module is used for selecting vehicle following data under the typical general road traffic environment condition from a following database;

and the second selection module is used for selecting the following data in the following form on the same lane from the vehicle following data selected by the first selection module to form corresponding historical vehicle following data.

Images