KR101827124B1 - System and Method for recognizing driving pattern of driver - Google Patents

Abstract

The present invention can define the risk of the driver by analyzing the driving tendency of the learned driver model by learning and classifying the driver's pattern by applying the driver's pattern to the multi-tree genetic programming algorithm method using the vehicle data provided by the vehicle, The present invention provides a driving pattern recognition system and method for a driver capable of real-time processing of a risk of non-existent data.
A driving pattern recognition system for a driver according to an embodiment of the present invention includes a data collection unit for collecting driving data of a vehicle, data for sorting major sensors having a large influence on the driving pattern of the driver by analyzing the driving data of the vehicle A data learning unit for generating a model for each driver's operation pattern in the driving data of the received vehicle through a reduction unit, a machine learning algorithm, and a service for executing a service using the model for the generated driving pattern of the driver And an execution unit.

Description

Technical Field [0001] The present invention relates to a driving pattern recognition system,

BACKGROUND OF THE INVENTION 1. Field of the Invention [0002] The present invention relates to a driving pattern recognition system and method for a driver, and more particularly, to a technique for recognizing a driving pattern of a driver using vehicle data provided by a vehicle.

Customer satisfaction with regard to the driving performance of a vehicle depends on how much the vehicle travels in accordance with the customer's tendency. However, since the performance characteristic of the vehicle is determined to be a performance characteristic for the same vehicle type, there may be a difference between the driving behavior of the customer and the response of the vehicle. Accordingly, the customer often complains about the driving performance of the vehicle.

That is, it is possible to maximize the satisfaction of the customer regarding the driving performance by grasping the driving behavior of the customer and controlling the shift so that the vehicle responds to the driving propensity of the customer.

Accordingly, a method of learning the driving behavior of the customer for a long period of time and controlling the shift according to the learned driving propensity has been developed. The method of controlling the shift according to the learned driving propensity is performed under the assumption that the driving propensity of the customer is constant.

However, the tendency of the driver is not always constant, but varies depending on the driver's emotions, changes in the willingness to drive instantaneously, road conditions, etc. Therefore, the learned driving propensity is different from the actual driver's tendency at any moment.

Accordingly, when the shift control is performed in accordance with the learned driving propensity, there is a problem that the actual will of the driver can not be reflected in the shift, but rather causes dissatisfaction of the driver.

For example, when the driver is very bright with the accelerator pedal, the conventional shift control device does not distinguish whether the driver has stepped on the accelerator pedal or the accelerator pedal has been stepped on by the driver in order to accelerate the vehicle I did not. As a result, the shift control is performed in an erroneous manner, causing a dissatisfaction of the driver.

[Patent Literature] Korean Laid-Open Patent Publication No. 2015-0034899.

The present invention can define the risk of the driver by analyzing the driving tendency of the learned driver model by learning and classifying the driver's pattern by applying the driver's pattern to the multi-tree genetic programming algorithm method using the vehicle data provided by the vehicle, The present invention provides a driving pattern recognition system and method for a driver capable of real-time processing of a risk of non-existent data.

Other objects and advantages of the present invention will become apparent from the following description, and it will be understood by those skilled in the art that the present invention is not limited thereto. It is to be easily understood that the objects and advantages of the present invention can be realized by means of the means shown in the claims and combinations thereof.

A driving pattern recognition system for a driver according to an embodiment of the present invention includes a data collection unit for collecting driving data of a vehicle, a data reduction unit for selecting sensors that affect the driving pattern of the driver by analyzing the driving data of the vehicle, A data learning unit for generating a model of an operation pattern of each driver from the driving data of the received vehicle through a machine learning algorithm, and a service execution unit for executing a service using the model of the generated driving pattern of the driver .

A method for recognizing a driving pattern of a driver according to an embodiment of the present invention includes collecting driving data of a vehicle, selecting sensors that affect an operation pattern of a driver by analyzing driving data of the vehicle, Generating a model of an operation pattern of each driver based on the received operation data of the vehicle, and executing a service using the generated model of the driver's operation pattern.

This technique can define the risk of the driver by learning and classifying the driver's pattern by applying the driver's pattern to the multi-tree genetic programming algorithm method using the vehicle data provided by the vehicle, learning and analyzing the driving tendency of the learned driver model , And real-time processing of the risk of untrained data.

In addition, the technology can develop a driving assist system (ADAS) to improve the safety and fuel efficiency of the vehicle, and can classify the non-learning operation data immediately, thereby increasing the utilization of the driving assistance system (ADAS) It is a technology that can be.

1 is a block diagram illustrating a driving pattern recognition system for a driver according to an embodiment of the present invention.
2 is a view for explaining a method of operating a driving pattern recognizing system according to an embodiment of the present invention.
3 is a view for explaining a structure of an internal tree constituting an entity in a multi-tree genetic programming algorithm method of a driver's traveling pattern recognition system according to an embodiment of the present invention.
4 is a view for explaining a method of designing a multi-tree genetic programming algorithm of a driver's driving pattern recognition system according to an embodiment of the present invention.
FIG. 5 is a diagram illustrating a method for designing a multi-tree genetic programming algorithm of a driver's driving pattern recognition system according to an embodiment of the present invention.
6 is a diagram illustrating a selection operator in a multi-tree genetic programming algorithm according to an embodiment of the present invention.
7 and 8 are diagrams illustrating math operators in a multiple tree genetic programming algorithm according to an embodiment of the present invention.
9 is a diagram illustrating a mutation operator in a multi-tree genetic programming algorithm according to an embodiment of the present invention.
10 is a diagram illustrating a decision process for classifying a driver pattern in a multi-tree genetic programming algorithm according to an embodiment of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS The advantages and features of the present invention, and how to accomplish it, will be described with reference to the embodiments described in detail below with reference to the accompanying drawings. However, the present invention is not limited to the embodiments described herein but may be embodied in other forms. The embodiments are provided so that those skilled in the art can easily carry out the technical idea of the present invention to those skilled in the art.

In the drawings, embodiments of the present invention are not limited to the specific forms shown and are exaggerated for clarity. Although specific terms are used herein, It is to be understood that the same is by way of illustration and example only and is not to be taken by way of limitation of the scope of the appended claims.

The expression " and / or " is used herein to mean including at least one of the elements listed before and after. Also, the expression " coupled / connected " is used to mean either directly connected to another component or indirectly connected through another component. The singular forms herein include plural forms unless the context clearly dictates otherwise. Also, as used herein, "comprising" or "comprising" means to refer to the presence or addition of one or more other components, steps, operations and elements.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

1 is a block diagram illustrating a driving pattern recognition system for a driver according to an embodiment of the present invention.

Referring to FIG. 1, a driving pattern recognition system of a driver includes a data collecting unit 100, a data reducing unit 200, a data learning unit 300, and a service executing unit 400.

The data collecting unit 100 collects vehicle driving data. The sensor data of the in-vehicle electronic control units (ECUs) is collectively collected and transmitted to the data reduction unit 200, the data learning unit 300, and the service execution unit 400.

The data reduction unit 200 analyzes the vehicle data received from the data collecting unit 100 to select major sensors having a large influence on the driver's operation pattern. The selected main sensors are generated as a list, and the generated lists are transmitted to the data learning unit 300. Here, the main sensors, which greatly affect the driver's driving pattern, may include sensors such as an accelerator pedal sensor, a steering sensor, and the like, which can be sensed by sensors provided in the vehicle according to the driver's integrated driving habits.

The data learning unit 300 generates a model for each driver's operation pattern from the vehicle data received from the data collection unit 100 through a machine learning algorithm. At this time, the data learning unit 300 shortens the generation time of the model for the driving pattern of the driver by using the main sensors selected by the data reduction unit 200, and improves the quality of the model for the driving pattern of the driver.

The service execution unit 400 executes the service using the model of the driver's operation pattern generated by the data learning unit 300. [ When the new vehicle data is received from the data collection unit 100, the service execution unit 400 applies the generated new vehicle data to the generated model of the driver's operation pattern to grasp the current driving behavior and state of the driver, And generates a service execution instruction according to the state.

2 is a view for explaining a method of a multi-tree genetic programming algorithm in a machine learning algorithm of a driving pattern recognition system according to an embodiment of the present invention.

Referring to FIG. 2, the driver's travel pattern recognition system generates an entity (candidate model) set. Here, a method for generating an entity set is to generate an initial entity set by randomly generating a tree constituting an entity (S11).

Next, the driver's travel pattern recognition system assigns arbitrarily selected features (sensors) to each evolution group (S13). Here, each evolution group can use only the assigned sensor data for the entity (candidate model).

Next, the driver's travel pattern recognition system evaluates the fitness of the individual (S15). That is, we evaluate the numerical suitability of how each entity represents the problem (driver pattern). The method of evaluating the fitness of these objects is based on learning vehicle data and the degree of modeling of the driver 's driving pattern can be used as an evaluation index.

Next, the driving pattern recognition system of the driver determines the termination condition (S17).

For example, the driver's travel pattern recognition system may be a conditional repetition for the termination of a multi-tree genetic programming algorithm, or a repetition of 1000 times, or a case where all the entities are the same shape. End the algorithm and make the final decision.

Next, the driver's travel pattern recognition system selects multiple trees (S19). In other words, individuals with high fitness for selecting individuals with high fitness to survive in the next generation can be duplicated (replicated), and low individuals can be eliminated and disappear.

Next, the driver's traveling pattern recognition system crosses multiple trees (S21). That is, crossing of multiple trees is a process of exchanging the traits of the selected entities, thereby creating a reassorted child entity of the dominant individuals.

Next, the driver's travel pattern recognition system generates a mutation of the multiple tree (S23).

That is, it is a process in which an arbitrary transition phenomenon occurs in the traits of each individual, and it is possible to search a range in which a problem can not be searched by recombination of existing entities.

Next, the driver's travel pattern recognition system redistributes the weight of the data (S25).

That is, the redistribution of the data weights is performed by statistically evaluating the difficulty of each data based on the fitness of the individuals and giving a weight based on the degree of difficulty (S27).

For example, at the start of a multi-tree genetic programming algorithm, weights of the same value are all 1, but data that is difficult to model as the iterations progress (the problem of fewer modeled objects) becomes heavier, The weight is lowered. After the weighting of the data is redistributed, the fitness of the individuals is evaluated by reflecting these weights at the stage of the individual fitness evaluation.

Next, the driving pattern recognition system of the driver makes a final decision by voting the multiple tree (S29).

In addition, when the repetition of the multi-tree genetic programming algorithm is completed in step 17, the driver's traveling pattern recognition system selects the object having the highest fitness in each evolution group, collects all the decisions of the selected objects, The final intention of

3 is a view for explaining a structure of an internal tree constituting an entity in a multi-tree genetic programming algorithm method of a driver's traveling pattern recognition system according to an embodiment of the present invention.

Referring to FIG. 3, the ensemble multi-tree genetic programming algorithm of the present invention is similar to a conventional multi-tree genetic programming algorithm in that one tree is not an entity but a plurality of trees It consists of one entity.

The number of trees constituting one entity means the number of evolution groups (1 to n).

In addition, one tree consists of an operator node and a leaf node, an operator node stores a mathematical symbol, a leaf node stores one of the sensor data of the vehicle as a variable, and a numerical value to be measured is inputted or an arbitrary constant Can be input.

This one tree T ₁ can be expressed by the following equation ( ₁ ).

[Equation 1]

The formulas generated in each tree are represented by a single function value and can be classified according to whether the function value is negative or positive.

4 is a view for explaining a method of designing a multi-tree genetic programming algorithm of a driver's driving pattern recognition system according to an embodiment of the present invention.

Referring to FIG. 4, the driver's travel pattern recognition system assigns a set of features (sensors) arbitrarily selected from the entities to each evolution group.

Specifically, the driver's travel pattern recognition system randomly samples m sets in the selected (full state) feature set S31 and assigns them to the respective evolution groups (evolution groups 1 to m) (S33, S35) , And the optimal models selected in each evolution group are configured as one final model (S37).

FIG. 5 is a diagram illustrating a method for designing a multi-tree genetic programming algorithm of a driver's driving pattern recognition system according to an embodiment of the present invention.

Referring to FIG. 5, the driver's travel pattern recognition system randomly samples in a selected (full state) feature set S51 and allocates the same to each evolution group (S53), and statistically data The learning difficulty level is defined (S55). In the case of data in which many individuals fail to model, the learning difficulty level is determined to be high, and in the opposite case, the low learning difficulty level is determined. This learning difficulty is assigned to each data in the form of a weight. When evaluating the fitness of an individual, weights are reflected in the classification accuracy calculation of each individual model (S57 to S59).

In other words, the object acquires a higher fitness than that when the data of low weight is classified when the data of high weight is successfully classified.

6 is a diagram illustrating a selection operator in a multi-tree genetic programming algorithm according to an embodiment of the present invention.

Referring to FIG. 6, the multiple tree genetic programming algorithm includes a selection operator for evolution of a candidate set.

The multi-tree genetic programming algorithm designs a method for selecting candidate entities with good fitness for each evolutionary group to the next generation.

Elitism ensures that individuals with the best fit are always passed on to the next generation to preserve the dominant individual. Thus, the entity to be preserved records the highest fitness and includes all the non-identical entities. The criterion of non-identical objects is the combination of classification results of learning data. One data classification result judges other objects as different objects.

Tournament is a conventional universal selection operator that selects an excellent entity as the next generation candidate based on probability and fitness. Specifically, the fitness of two arbitrarily selected entities is compared with each other, and an entity having a high fitness is copied to a candidate set of the next generation. This process is repeated until 90% of the total number of individuals including x% of individuals selected by the Elitism is filled.

Reconstruct involves placing entities that are completely new and randomly constructed into the next-generation candidate entity set. 90% created by Elitism and Tournament and 10% created by Reconstruct become the next generation of individuals, and even if the generation progresses, the sets retain the same number of individuals.

7 and 8 are diagrams illustrating math operators in a multiple tree genetic programming algorithm according to an embodiment of the present invention.

Referring to Figures 7 and 8, the multiple tree genetic programming algorithm includes a mating operator for evolution of candidate groups.

We design the mating technique, which is a method by which the dominant individuals generate developed individuals through transgenic transformation, to be applicable to the multi-tree genetic programming algorithm method. Mixing (A-> A ') or swapping (Swapping; B-> B') the entire tree is performed on two arbitrarily selected trees. This process is repeated several times based on probability. In this case, internal mating (C) in which two trees are selected and executed in one evolution group is defined as external mating (D) when selected in another evolution group, and the probability of each mating is defined differently. For effective search of problems and conservation of beneficial objects, internal breeding can be defined as an occurrence probability of 80%, and external breeding can be defined as a probability of occurrence of 5%.

9 is a diagram illustrating a mutation operator in a multi-tree genetic programming algorithm according to an embodiment of the present invention.

Referring to Figures 9 (a), 9 (b) and 9 (c), the multiple tree genetic programming algorithm includes a mutation operator for the evolution of candidate groups.

The multi-tree genetic programming algorithm can be performed by selecting three ways of growing a subtree, pruning a subtree, or modifying an operator as an arbitrary node in an arbitrary tree in an object .

10 is a diagram illustrating a decision process for classifying a driver pattern in a multi-tree genetic programming algorithm according to an embodiment of the present invention.

Referring to FIG. 10, in the multi-tree genetic programming algorithm, the decision for classifying the driving pattern of the driver satisfies the termination condition of the multi-tree genetic programming algorithm, the optimum candidate solution E of each group among the derived candidate solution models, (F, classifier) combined by voting and derive a final classification result (F, classifier) that is combined by voting. In addition, the classifier classifier By way of example, it is possible to obtain higher classification accuracy.

As described above, according to the present invention, the driver's pattern is applied to the multi-tree genetic programming algorithm method using the vehicle data provided by the vehicle and is learned and classified. By learning and analyzing the driving tendency of the learned driver model, And it is a technology capable of real-time processing of the risk of unlearned data.

In addition, the technology can develop a driving assist system (ADAS) to improve the safety and fuel efficiency of the vehicle, and can classify the non-learning operation data immediately, thereby increasing the utilization of the driving assistance system (ADAS) It is a technology that can be.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, Various modifications and variations may be made without departing from the scope of the appended claims.

Claims (11)

delete delete Collecting vehicle driving data;
Analyzing the vehicle operation data to select sensors related to the driver's operation pattern, generating a list of the selected sensors, and reducing the vehicle operation data based on the generated list;
Generating a model for each driver's operation pattern from the reduced driving data of the vehicle and the driving data of the vehicle received through the machine learning algorithm; And
And executing a service using the generated model of the driver's operation pattern,
The machine learning algorithm using a multi-tree genetic programming algorithm and generating a set of entities;
Assigning to each evolution group an arbitrarily selected feature set from the set of generated entities;
Evaluating the fitness of the entity;
Determining an end condition of the multi-tree genetic programming algorithm;
Selecting multiple trees in the multi-tree genetic programming algorithm;
Crossing the multiple tree;
Generating a mutation of the multiple tree; And
Redistributing the weights of the data based on the fitness of the entity
Wherein the driver recognizes the travel pattern of the driver. delete delete The method of claim 3,
Wherein generating the set of entities comprises:
And generating a set of entities by randomly generating the trees constituting the entity. The method of claim 3,
The step of assigning to each evolution group
And using only the assigned sensor data for the object. The method of claim 3,
Evaluating the fitness of the entity comprises:
Wherein the degree of modeling of the driving pattern of the driver is used as an evaluation index based on the learning vehicle data. The method of claim 3,
Wherein crossing the multiple tree comprises:
Wherein the selected entities exchange traits of each other, thereby generating a child entity of which the traits of the dominant entities are recombined. The method of claim 3,
The step of redistributing the weights of the data comprises:
Wherein the degree of difficulty of each data is statistically evaluated based on the fitness of the individuals, and a weight based on the degree of difficulty is given. The method of claim 3,
Selecting one entity having the highest fitness from each of the evolution groups, collecting all decisions of the selected entity, and determining one final decision through a majority vote.

Images