KR102176188B1 - Electronic device for evaluating deterioration level of the road and the operating method thereof - Google Patents

Abstract

Disclosed are an electronic device for determining a road aging level based on artificial intelligence and a method of operating the same. The present invention generates a judgment model based on artificial intelligence for determining the aging grade of the road by using pre-configured training set data as having an effect on determining the aging grade of the road, and the generation of the judgment model is completed. Then, the present invention relates to an electronic device capable of determining an aging level of a road according to the determination model based on actual driving information acquired from a terminal mounted in a vehicle driving on the road, and an operation method thereof.

Description

Electronic device for determining road aging level based on artificial intelligence and its operation method {ELECTRONIC DEVICE FOR EVALUATING DETERIORATION LEVEL OF THE ROAD AND THE OPERATING METHOD THEREOF}

The present invention relates to an electronic device for determining a road aging grade based on artificial intelligence and a method of operating the same.

Recently, as the number of obsolete roads increases throughout the country, the risk of traffic accidents is increasing.

On aging roads, road damage such as potholes or pavement cracks, which are local holes that are caused by part of the road surface being broken or settled down, can easily occur, which can increase the risk of traffic accidents on aging roads. .

Therefore, there is a need to prepare for risks that may occur in the aged road by preliminarily grasping the degree of aging of the road. However, it is difficult to check the conditions of countless roads one by one and determine the aging grade of the road based on this.

If information that can be used to determine the aging grade of the road is received from a vehicle running on the actual road, and the aging grade of the road can be determined based on this, the actual road condition for countless roads. Even if you do not directly check the aging level, you will be able to more efficiently determine the aging grade for the road.

On the other hand, in recent years, machine learning-based artificial intelligence technology that can make a judgment model for determining a predetermined result based on some sample data has appeared. In this regard, machine learning-based artificial intelligence technology can be used to determine the aging grade for roads.

Therefore, there is a need for research on a technology capable of generating an artificial intelligence-based judgment model for determining the aging grade of a road, and determining the aging grade of the road based on this.

The present invention provides an electronic device capable of generating an artificial intelligence-based judgment model for determining an aging grade for a road by using pre-configured training set data as having an influence on determining the aging grade of a road, and an operating method thereof. I want to present it.

Through this, the electronic device and its operation method according to the present invention are intended to support the determination of the aging grade of the road according to the determination model based on actual driving information obtained from a terminal mounted in a vehicle running on the road. .

An electronic device for determining an artificial intelligence-based road aging grade according to an embodiment of the present invention stores vehicle type information in which a plurality of vehicle type information and predetermined different numerical values corresponding to each of the plurality of vehicle type information are stored. Negative, a one-hot vector storage unit in which a different N-dimensional one-hot vector is stored for each of the preset N (N is a natural number of 3 or more) aging grades, and for roads. Different M (M is a natural number of 2 or more) input information sets predetermined as input information for generating a judgment model for determining the aging grade-Each of the M input information sets are measured values for vibration and speed And vehicle type information refers to information consisting of a set-a training set storage unit in which M training sets matching each of the M input information sets and pre-designated aging grades are stored, the vehicle model After checking the numerical value corresponding to the vehicle type information included in each of the M input information sets, referring to the information storage unit, the measured value and vehicle type for vibration and humidity included in each of the M input information sets An input vector generator for generating a three-dimensional input vector corresponding to each of the M input information sets by constructing a three-dimensional vector having a numerical value corresponding to the information as a component, each of the M input information sets When a three-dimensional input vector corresponding to is generated, by referring to the one-hot vector storage unit, a one-hot vector for an aging class matching each of the M input information sets is checked, and the M inputs A one-hot vector identification unit for checking a one-hot vector corresponding to each of the information sets, a deep neural network consisting of two or more weight matrices for a three-dimensional input vector corresponding to each of the M input information sets-the deep Two or more weight matrices constituting the neural network are matrices configured such that the dimension of the vector calculated as the output of the deep neural network becomes N-dimensional- When an output vector generator for generating an N-dimensional output vector corresponding to each of the M input information sets, and an N-dimensional output vector corresponding to each of the M input information sets are generated by applying as input to, An N-dimensional output vector corresponding to each of the M input information sets is a predetermined activation function-The activation function is a component constituting an N-dimensional output vector corresponding to each of the M input information sets It is a function that converts values into values of 0 or more and 1 or less.- An operation vector generator and the M input information sets that generate an N-dimensional operation vector corresponding to each of the M input information sets by applying to as inputs When an N-dimensional operation vector corresponding to each of the M input information sets is generated, the N-dimensional operation vector corresponding to each of the M input information sets is maximally approximated to a one-hot vector corresponding to each of the M input information sets. And a model generation unit that generates the judgment model for determining an aging grade for a road by performing machine learning on the deep neural network.

In addition, in an embodiment of the present invention, in the method of operating an electronic device for determining an artificial intelligence-based road aging level, a plurality of vehicle type information and predetermined different numerical values corresponding to each of the plurality of vehicle type information are stored. Maintaining a vehicle type information storage unit, which maintains a one-hot vector storage unit in which a different N-dimensional one-hot vector is stored for each of the preset N (N is a natural number of 3 or more) aging grades. Step, M (M is a natural number of 2 or more) different input information sets predetermined as input information for generating a judgment model for determining the aging grade of the road-Each of the M input information sets is vibration, This refers to information consisting of a set of measured values for speed and vehicle type information-and a training set in which M training sets in which pre-designated aging grades are matched corresponding to each of the M input information sets Maintaining a storage unit, referring to the vehicle model information storage unit, checking a numerical value corresponding to vehicle model information included in each of the M input information sets, and then vibration included in each of the M input information sets And generating a three-dimensional input vector corresponding to each of the M input information sets by constructing a three-dimensional vector having a measured value for humidity and a numerical value corresponding to the vehicle type information as components, the M When a three-dimensional input vector corresponding to each of the input information sets is generated, by referring to the one-hot vector storage unit, a one-hot vector for an aging class matching each of the M input information sets is checked. , Checking a one-hot vector corresponding to each of the M input information sets, A deep neural network composed of two or more weight matrices by using a three-dimensional input vector corresponding to each of the M input information sets-the deep Two or more weight matrices constituting the neural network are used so that the dimension of the vector calculated as the output of the deep neural network is N-dimensional. Constructed matrices-generating an N-dimensional output vector corresponding to each of the M input information sets by applying as input to, when an N-dimensional output vector corresponding to each of the M input information sets is generated , An N-dimensional output vector corresponding to each of the M input information sets is a predetermined activation function.- The activation function includes 0 components constituting an N-dimensional output vector corresponding to each of the M input information sets. It is a function that converts to a value equal to or less than 1-generating an N-dimensional operation vector corresponding to each of the M input information sets by applying as input to and N corresponding to each of the M input information sets When a dimensional operation vector is generated, the N-dimensional operation vector corresponding to each of the M input information sets is maximally approximated to a one-hot vector corresponding to each of the M input information sets. And generating the judgment model for determining an aging grade for the road by performing machine learning.

The present invention provides an electronic device capable of generating an artificial intelligence-based judgment model for determining an aging grade for a road by using pre-configured training set data as having an influence on determining the aging grade of a road, and an operating method thereof. Can be presented.

Through this, the electronic device and the operation method thereof according to the present invention may determine a road aging grade according to the determination model based on actual driving information obtained from a terminal mounted in a vehicle running on a road.

1 is a diagram illustrating a structure of an electronic device for determining an artificial intelligence-based road aging grade according to an embodiment of the present invention.
2 is a flow chart illustrating a method of operating an electronic device for determining an artificial intelligence-based road aging grade according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. This description is not intended to limit the present invention to a specific embodiment, it is to be understood to include all changes, equivalents, and substitutes included in the spirit and scope of the present invention. While describing each drawing, similar reference numerals have been used for similar components, and unless otherwise defined, all terms used in the present specification including technical or scientific terms refer to common knowledge in the technical field to which the present invention belongs. It has the same meaning as commonly understood by someone who has.

In this document, when a part "includes" a certain component, it means that other components may be further included rather than excluding other components unless otherwise stated. In addition, in various embodiments of the present invention, each component, functional blocks or means may be composed of one or more sub-components, and the electrical, electronic, and mechanical functions performed by each component are electronic. A circuit, an integrated circuit, and an application specific integrated circuit (ASIC) may be implemented with various known devices or mechanical elements, and may be implemented separately or two or more may be integrated into one.

On the other hand, the blocks of the attached block diagram and the steps in the flowchart are computer program instructions that are mounted on a processor or memory of equipment capable of processing data such as a general-purpose computer, a special-purpose computer, a portable notebook computer, and a network computer to perform specified functions It can be interpreted as meaning. Since these computer program instructions can be stored in a memory provided in a computer device or in a memory readable by a computer, the functions described in the blocks in the block diagram or in the steps in the flowchart are produced as a product containing the instruction means to perform this. It could be. In addition, each block or each step may represent a module, segment, or part of code including one or more executable instructions for executing the specified logical function(s). In addition, it should be noted that in some alternative embodiments, functions mentioned in blocks or steps may be executed in a different order. For example, two blocks or steps shown in succession may be performed substantially simultaneously or may be performed in reverse order, and in some cases, some blocks or steps may be omitted.

1 is a diagram illustrating a structure of an electronic device for determining an artificial intelligence-based road aging grade according to an embodiment of the present invention.

Referring to FIG. 1, an electronic device 110 for determining an artificial intelligence-based road aging grade according to the present invention includes a vehicle type information storage unit 111, a one-hot vector storage unit 112, and a training set storage unit 113. ), an input vector generation unit 114, a one-hot vector identification unit 115, an output vector generation unit 116, an operation vector generation unit 117, and a model generation unit 118.

The vehicle model information storage unit 111 stores a plurality of vehicle model information and predetermined different numerical values corresponding to each of the plurality of vehicle model information.

For example, the vehicle model information storage unit 111 may store a plurality of vehicle model information and predetermined different numerical values corresponding to each of the plurality of vehicle model information as shown in Table 1 below.

Multiple vehicle model information Numerical value Avante One Sonata 2 Carnival 3 Granger 4 Genesis 5 ... ...

The one-hot vector storage unit 112 stores a one-hot vector of different N dimensions for each of the preset N (N is a natural number of 3 or more) aging grades.

For example, when N is 5, the one-hot vector storage unit 112 may store different 5-dimensional one-hot vectors for each of the five pre-set aging levels as shown in Table 2 below. I can.

5 aging grades 5-D One-Hot Vector Level 1 [1 0 0 0 0] Level 2 [0 1 0 0 0] Level 3 [0 0 1 0 0] Level 4 [0 0 0 1 0] Level 5 [0 0 0 0 1]

In the training set storage unit 113, different M input information sets (M is a natural number of 2 or more) and the M input information set in advance as input information for generating a judgment model for determining an aging grade for a road M training sets in which a pre-designated aging grade is matched as corresponding to each of the sets are stored. Here, each of the M input information sets means information in which measurement values for vibration and speed and vehicle model information are configured as one set.

For example, when M is 3, the training set storage unit 113 has three different input information sets and an aging grade previously designated as corresponding to each of the three input information sets as shown in Table 3 below. Matched three training sets may be stored.

3
Training sets 3 input set information Aging grade vibration speed Car type Training set 1 200dB(V) 40km/h Sonata Level 1 Training set 2 280dB(V) 50km/h Genesis Level 4 Training set 3 220dB(V) 45km/h Avante Level 2

Here, the reason why each input information set of the training set is composed of vibration, speed, and vehicle type is as follows. First, there may be a correlation between the degree of aging of the road and the vibration of the vehicle, such as a vibration generated in the vehicle increases because the road condition is worse as the road aging is severe. In this case, the vibration generated by the vehicle may vary slightly depending on the speed of the vehicle, and the difference may also occur depending on the vehicle type. For example, even when passing through the same porthole, a case where the vehicle speed is high may generate a greater vibration than that in a case where the vehicle speed is not. Eventually, the vibration generated by the vehicle may vary depending on the degree of aging of the road, and since this vibration is related to the driving speed of the vehicle and the vehicle type, the above-mentioned reference designated to create a judgment model for determining the aging grade for the road The input information set may consist of vibration, speed, and vehicle type.

The input vector generator 114 refers to the vehicle model information storage unit 111, checks a numerical value corresponding to vehicle model information included in each of the M input information sets, and then checks each of the M input information sets. By constructing a three-dimensional vector having as a component a numerical value corresponding to the measured values for vibration and humidity and vehicle model information included in, a three-dimensional input vector corresponding to each of the M input information sets is generated. .

When a three-dimensional input vector corresponding to each of the M input information sets is generated, the one-hot vector identification unit 115 refers to the one-hot vector storage unit 112, and the M input information sets A one-hot vector corresponding to each of the M input information sets is identified by confirming the one-hot vector for the aging grade matched to each.

The output vector generator 116 corresponds to each of the M input information sets by applying a 3D input vector corresponding to each of the M input information sets as input to a deep neural network composed of two or more weight matrices. N-dimensional output vector is generated.

Here, the two or more weight matrices constituting the deep neural network are matrices configured such that the dimension of a vector calculated as an output of the deep neural network is N-dimensional.

For example, as in the above example, assume that M is 3 and N is 5. At this time, the input vector generation unit 114 refers to the vehicle type information storage unit 111 shown in Table 1 above, and corresponds to vehicle type information included in each of the three input information sets,'Sonata, Genesis, Avante'. The numerical values can be checked as '2, 5, 1'respectively. In addition, the input vector generation unit 114 constructs a three-dimensional vector having as a component a numerical value corresponding to the measured value of vibration and humidity and vehicle model information included in each of the three input information sets. A three-dimensional input vector corresponding to each of the three input information sets may be generated. In this regard, the measured values for vibration and humidity included in the'input information set 1'are '200 (dB(V)) and 40 (km/h)', respectively, and the'Sonata' which is vehicle type information Since the numerical value is '2', the input vector generator 114 may generate'[200 40 2]' as a three-dimensional input vector corresponding to the'input information set 1'. In this way, the input vector generation unit 114 generates'[200 40 2], [280 50 5], [220 45 1]' as a three-dimensional input vector corresponding to each of the three input information sets. can do.

In this way, when'[200 40 2], [280 50 5], [220 45 1]' is generated as a three-dimensional input vector corresponding to each of the three input information sets by the input vector generator 114 , The one-hot vector identification unit 115 refers to the one-hot vector storage unit 112, and is assigned to the aging grades matched to each of the three input information sets. You can check the one-hot vector for this. At this time, in the one-hot vector storage unit 112 as shown in Table 2, the one-hot vector for '1st grade' is'[1 0 0 0 0]', and the one-hot vector for '4th grade' is Since'[0 0 0 1 0]' and the one-hot vector for '2nd grade' is stored as'[0 1 0 0 0]', the input vector generator 114 '[1 0 0 0 0], [0 0 0 1 0], and [0 1 0 0 0]' can be identified as one-hot vectors corresponding to each of the sets.

Then, the output vector generation unit 116 generates at least two input vectors'[200 40 2], [280 50 5], and [220 45 1], which are three-dimensional input vectors corresponding to each of the three input information sets. By applying as input to a deep neural network composed of weight matrices, a 5-dimensional output vector corresponding to each of the three input information sets is'[a ₁ a ₂ a ₃ a ₄ a ₅ ], [b ₁ b ₂ b ₃ b ₄ b ₅ ], [c ₁ c ₂ c ₃ c ₄ c ₅ ]'.

When the N-dimensional output vector corresponding to each of the M input information sets is generated, the operation vector generator 117 generates an N-dimensional output vector corresponding to each of the M input information sets with a predetermined activation function ( activation function) to generate an N-dimensional operation vector corresponding to each of the M input information sets.

Here, the activation function is a function for converting components constituting an N-dimensional output vector corresponding to each of the M input information sets into values of 0 or more and 1 or less.

In this case, according to an embodiment of the present invention, the activation function may be a softmax function such as Equation 1 below.

는 상기 M개의 입력 정보 세트들 각각에 대응되는 N차원의 연산 벡터를 구성하는 성분들 중 i번째 성분의 값으로,

는 상기 M개의 입력 정보 세트들 각각에 대응되는 N차원의 출력 벡터를 구성하는 성분들 중 i번째 성분의 값,

는 자연상수를 의미한다. here,

Is the value of the i-th component of the components constituting the N-dimensional operation vector corresponding to each of the M input information sets,

Is the value of the i-th component of the components constituting the N-dimensional output vector corresponding to each of the M input information sets,

Means natural constant.

When an N-dimensional operation vector corresponding to each of the M input information sets is generated, the model generation unit 118 generates an N-dimensional operation vector corresponding to each of the M input information sets. By performing machine learning on the deep neural network so as to maximally approximate the one-hot vector corresponding to each of them, the judgment model for determining the aging grade of the road is generated.

For example, as in the above example, M is 3, N is 5, and is a 5-dimensional output vector corresponding to each of the three input information sets'[a ₁ a ₂ a ₃ a ₄ a ₅ ], [b ₁ b ₂ b ₃ b ₄ b ₅ ], [c ₁ c ₂ c ₃ c ₄ c ₅ ]'is generated, the operation vector generator 117 is 5 corresponding to each of the three input information sets The dimensional output vector of'[a ₁ a ₂ a ₃ a ₄ a ₅ ], [b ₁ b ₂ b ₃ b ₄ b ₅ ], [c ₁ c ₂ c ₃ c ₄ c ₅ ]'is expressed in Equation 1 above. By applying as input to an activation function such as'[a ₍₁₎ a ₍₂₎ a ₍₃₎ a ₍₄₎ a ₍₅₎ ] _, a five-dimensional operation vector corresponding to each of the three input information sets is applied. , [b ₍₁₎ b ₍₂₎ b ₍₃₎ b ₍₄₎ b ₍₅₎ ], [c ₍₁₎ c ₍₂₎ c ₍₃₎ c ₍₄₎ c ₍₅₎ ]' can do.

In this way, by the operation vector generation unit 117, the five-dimensional operation vector corresponding to each of the three input information sets is'[a ₍₁₎ a ₍₂₎ a ₍₃₎ a ₍₄₎ a _(5)). ], [b ₍₁₎ b ₍₂₎ b ₍₃₎ b ₍₄₎ b ₍₅₎ ], [c ₍₁₎ c ₍₂₎ c ₍₃₎ c ₍₄₎ c ₍₅₎ ]' Then, the model generation unit 118 is a five-dimensional operation vector corresponding to each of the three input information sets'[a ₍₁₎ a ₍₂₎ a ₍₃₎ a ₍₄₎ a ₍₅₎ ], [b ₍₁₎ b ₍₂₎ b ₍₃₎ b ₍₄₎ b ₍₅₎ ], [c ₍₁₎ c ₍₂₎ c ₍₃₎ c ₍₄₎ c ₍₅₎ ]' Machine learning for the deep neural network so as to maximally approximate'[1 0 0 0 0], [0 0 0 1 0], [0 1 0 0 0]', which are one-hot vectors corresponding to each of the input information sets By performing, it is possible to generate the judgment model for determining the aging grade for the road.

In this case, according to an embodiment of the present invention, the model generation unit 118 corresponds to components constituting an N-dimensional operation vector corresponding to each of the M input information sets and each of the M input information sets. A loss value for each of the M input information sets is calculated by calculating a loss value based on a loss function according to Equation 2 below, based on the components constituting the one-hot vector. In addition, machine learning for the deep neural network may be performed so that the average value of the loss values for each of the M input information sets is minimized.

은 상기 손실 값으로,

는 상기 M개의 입력 정보 세트들 각각에 대응되는 N차원의 원-핫 벡터를 구성하는 성분들 중 i번째 성분의 값,

는 상기 M개의 입력 정보 세트들 각각에 대응되는 N차원의 연산 벡터를 구성하는 성분들 중 i번째 성분의 값을 의미한다.here,

Is the loss value,

Is the value of the i-th component of the components constituting the N-dimensional one-hot vector corresponding to each of the M input information sets,

Denotes a value of an i-th component among components constituting an N-dimensional operation vector corresponding to each of the M input information sets.

For example, as in the above example, M is 3, N is 5, and the five-dimensional operation vector corresponding to each of the three input information sets is'[a ₍₁₎ a ₍₂₎ a ₍₃₎ a ₍₄₎ a ₍₅₎ ], [b ₍₁₎ b ₍₂₎ b ₍₃₎ b ₍₄₎ b ₍₅₎ ], [c ₍₁₎ c ₍₂₎ c ₍₃₎ c ₍₄₎ c ₍₅₎ ]'is generated, and the one-hot vector corresponding to each of the three input information sets is'[1 0 0 0 0], [0 0 0 1 0], [0 1 0 0 0]' Suppose

At this time, the model generation unit 118 is a five-dimensional operation vector corresponding to each of the three input information sets'[a ₍₁₎ a ₍₂₎ a ₍₃₎ a ₍₄₎ a ₍₅₎ ], [b ₍₁₎ b ₍₂₎ b ₍₃₎ b ₍₄₎ b ₍₅₎ ], [c ₍₁₎ c ₍₂₎ c ₍₃₎ c ₍₄₎ c ₍₅₎ ]' And components constituting'[1 0 0 0 0], [0 0 0 1 0], [0 1 0 0 0]' which are one-hot vectors corresponding to each of the three input information sets , By calculating a loss value based on the loss function according to Equation 2, loss values for each of the three input information sets may be calculated as'L ₁ , L ₂ , L ₃ '. Then, the model generation unit 118 may perform machine learning on the deep neural network so that the average value of'L ₁ , L ₂ , L ₃ 'is minimized.

At this time, according to an embodiment of the present invention, the model generation unit 118 performs backpropagation processing so that the average of the loss values for each of the M input information sets is minimized, so that the deep neural network Machine learning can be performed.

According to an embodiment of the present invention, the electronic device 110 includes an information request unit 119, a determination input vector generation unit 120, a determination output vector generation unit 121, and a determination operation vector generation unit 122. ), a similarity calculation unit 123, a determination unit 124, and a display unit 125 may be further included.

After the generation of the determination model is completed, the information requesting unit 119, when a command to determine the aging level for the first road is received, for the first vehicle terminal mounted on the first vehicle running on the first road, the At the same time as requesting transmission of the vibration value and speed value measured from the first vehicle, the transmission of vehicle type information of the first vehicle is requested.

Here, in order to collect the information (vibration, speed, vehicle type) for determining the aging level of the first road, the first vehicle has the first vehicle terminal mounted inside the vehicle and dispatched on the first road. It may be a vehicle for collecting information, or a vehicle that has been previously negotiated to provide information for determining an aging level of a road among general vehicles, and may be a vehicle having the first vehicle terminal installed therein. In addition, the first vehicle terminal is a terminal mounted inside the first vehicle, and may have pre-stored vehicle type information of the first vehicle in a memory, and generated in the first vehicle using a predetermined vibration sensor. A vibration value of the first vehicle may be measured, and the speed value of the first vehicle may be collected by being connected to an On-Board Diagnostics (OBD) terminal of the first vehicle. Thus, when the first vehicle terminal receives an aging level determination command for the first road from the electronic device 110 in a situation where the first vehicle is driving on the first road, the vibration value of the first vehicle , After measuring the speed value, the vibration value of the first vehicle, the speed value, and vehicle type information of the first vehicle stored in the memory may be transmitted to the electronic device 110.

The determination input vector generation unit 120 receives a first vibration value and a first speed value measured by the first vehicle terminal by the first vehicle from the first vehicle terminal, and the vehicle type for the first vehicle. When the first vehicle model information is received as information, a first numerical value that is a numerical value corresponding to the first vehicle model information is checked with reference to the vehicle model information storage unit 111, and the first vibration value and the first speed value And constructing a three-dimensional vector having the first numerical value as a component to generate a three-dimensional input vector for determination.

When the three-dimensional determination input vector is generated, the determination output vector generation unit 121 applies the three-dimensional determination input vector as an input to the deep neural network on which machine learning of the determination model is completed, An N-dimensional determination output vector corresponding to the dimensional determination input vector is generated.

When the N-dimensional determination output vector is generated, the determination operation vector generation unit 122 corresponds to the N-dimensional determination output vector by applying the N-dimensional determination output vector as an input to the activation function. An N-dimensional judgment operation vector is generated.

When the N-dimensional determination operation vector is generated, the similarity calculation unit 123 calculates a similarity between the N-dimensional determination operation vector and the one-hot vector for each of the N aging grades.

At this time, according to an embodiment of the present invention, the similarity calculation unit 123 may have calculated a Hadamard product between the N-dimensional determination operation vector and the one-hot vector for each of the N aging grades. The Manhattan norm of the vector calculated at this time may be calculated as a similarity between the N-dimensional determination operation vector and the one-hot vector for each of the N aging grades.

Here, the Hadamard product means an operation that multiplies each component in a vector or matrix of the same size. For example, when there are vectors'[a b c]' and'[x y z]', a vector of'[ax by cz]' can be calculated by calculating the Hadamard product between the two vectors.

In addition, the Manhattan norm is an L1 norm representing the size of a vector or matrix, and can be calculated according to Equation 3 below.

은 맨해튼 노름을 의미하고, x_i는 벡터나 행렬에 포함되어 있는 i번째 성분을 의미한다.here

Denotes the Manhattan norm, and x _i denotes the i-th component of a vector or matrix.

The determination unit 124 checks a first one-hot vector whose similarity to the N-dimensional determination operation vector is calculated to be maximum among the one-hot vectors for each of the N aging grades, and the one-hot vector By checking the first aging grade, which is the aging grade corresponding to the first one-hot vector, from the storage unit 115, it is determined that the aging grade for the first road is the first aging grade.

When it is determined that the aging grade for the first road is the first aging grade, the display unit 125 displays information on the first aging grade on the screen.

Hereinafter, the operation of the determination input vector generation unit 120, the determination output vector generation unit 121, the determination operation vector generation unit 122, the similarity calculation unit 123, the determination unit 124, and the display unit 125 It will be described in detail with an example.

First, assume that the first vibration value received from the first vehicle terminal is '250dB(V)', the first speed value is '45km/h', and the first vehicle type information is'carnival'. At this time, the determination input vector generation unit 120 refers to the vehicle type information storage unit 111 as shown in Table 1 above, and sets '3', which is a numerical value corresponding to the first vehicle type information,'carnival' to a first numerical value. Can be confirmed by

Then, the determination input vector generation unit 120 is the first vibration value '250 (dB(V))', the first speed value '45 (km/h)', and the first numerical value '3'. By constructing a three-dimensional vector having as a component,'[250 45 3]' can be generated as a three-dimensional input vector for judgment.

In this way, when'[250 45 3]' is generated as the three-dimensional judgment input vector by the judgment input vector generation unit 120, the judgment output vector generation unit 121 is the three-dimensional judgment input By applying the vector'[250 45 3]' as an input to the deep neural network in which the machine learning of the judgment model has been completed, the 5-dimensional judgment output vector corresponding to the 3-dimensional judgment input vector is converted to'[d ₁ d ₂ d ₃ d ₄ d ₅ ]'.

Thereafter, the determination operation vector generation unit 122 applies the five-dimensional determination output vector'[d ₁ d ₂ d ₃ d ₄ d ₅ ]'as an input to the activation function. A five-dimensional determination operation vector corresponding to the determination output vector may be generated as'[d ₍₁₎ d ₍₂₎ d ₍₃₎ d ₍₄₎ d ₍₅₎ ]'.

In this way, when the five-dimensional determination operation vector called'[d ₍₁₎ d ₍₂₎ d ₍₃₎ d ₍₄₎ d ₍₅₎ ]'is generated, the similarity calculator 123 determines the five-dimensional '[D ₍₁₎ d ₍₂₎ d ₍₃₎ d ₍₄₎ d ₍₅₎ ]', which is an operation vector, and'one-hot vector for each of the five aging grades as shown in Table 2 By calculating the Hadamard product between [1 0 0 0 0], [0 1 0 0 0], [0 0 1 0 0], [0 0 0 1 0], [0 0 0 0 1]','[d ₍₁₎ 0 0 0 0], [0 d ₍₂₎ 0 0 0], [0 0 d ₍₃₎ 0 0], [0 0 0 d ₍₄₎ 0], [0 0 0 0 d _{(5 )} ]'can be calculated.

Then, the similarity calculation unit 123 is calculated according to Equation 3 above:'[d ₍₁₎ 0 0 0 0], [0 d ₍₂₎ 0 0 0], [0 0 d ₍₃₎ 0 0] , [0 0 0 d ₍₄₎ 0], [0 0 0 0 d ₍₅₎ ]', the Manhattan norms of'd ₍₁₎ , d ₍₂₎ , d ₍₃₎ , d ₍₄₎ , d _{( 5)} 'may be calculated as a similarity between the five-dimensional determination operation vector and the one-hot vector for each of the five aging grades.

At this time, the degree of similarity between the five-dimensional determination operation vector and the one-hot vector for each of the five aging grades'd ₍₁₎ , d ₍₂₎ , d ₍₃₎ , d ₍₄₎ , d _{( 5)} Assuming that'd ₍₄₎ ' of 'is calculated as the largest value, the determination unit 124 is the one-hot vector for each of the five aging grades,'[1 0 0 0 0], [ 0 1 0 0 0], [0 0 1 0 0], [0 0 0 1 0], [0 0 0 0 1]', which is the five-dimensional determination operation vector'[d ₍₁₎ d _{(2 )} d ₍₃₎ d ₍₄₎ d ₍₅₎ ]'can be identified as'[0 0 0 1 0]' as the first one-hot vector calculated to have the maximum degree of similarity. Then, the determination unit 124 may check the first aging grade, which is the aging grade, corresponding to the first one-hot vector “[0 0 0 1 0]” from the one-hot vector storage unit 115. . At this time, since the one-hot vector storage unit 115 as shown in Table 2 stores the aging level corresponding to'[0 0 0 1 0]' as '4th grade', the determination unit 124 1 It can be determined that the aging grade for the road is '4 grade'.

In this way, when it is determined by the determination unit 124 that the aging grade for the first road is the first aging grade, which is '4 grade', the display unit 125 indicates that the aging grade of the first road is '4 grade'. Information indicating that it can be displayed on the screen. Through this, the road manager can check the current aging grade of the first road, and based on this, can make a repair plan for the first road.

2 is a flow chart illustrating a method of operating an electronic device for determining an artificial intelligence-based road aging grade according to an embodiment of the present invention.

In step S210, a vehicle model information storage unit in which a plurality of vehicle model information and predetermined different numerical values corresponding to each of the plurality of vehicle model information are stored is maintained.

In step S220, a one-hot vector storage unit in which different N-dimensional one-hot vectors for each of the preset N (N is a natural number of 3 or more) aging levels are stored is maintained.

In step S230, different M (M is a natural number of 2 or more) input information sets (each of the M input information sets) that are predetermined as input information for generating a judgment model for determining the aging grade of the road. Means information consisting of one set of measurement values for vibration and speed, and vehicle model information) and M training sets in which pre-designated aging grades are matched as corresponding to each of the M input information sets. Maintain a storage of training sets.

In step S240, after checking a numerical value corresponding to the vehicle model information included in each of the M input information sets by referring to the vehicle model information storage unit, vibrations included in each of the M input information sets and A three-dimensional vector having a measured value for humidity and a numerical value corresponding to vehicle type information is constructed as a component, thereby generating a three-dimensional input vector corresponding to each of the M input information sets.

In step S250, when a three-dimensional input vector corresponding to each of the M input information sets is generated, referring to the one-hot vector storage unit, the aging grade matching each of the M input information sets is By checking the one-hot vector for, one-hot vector corresponding to each of the M input information sets is identified.

In step S260, a three-dimensional input vector corresponding to each of the M input information sets is a deep neural network consisting of two or more weight matrices (two or more weight matrices constituting the deep neural network are calculated as the output of the deep neural network. By applying as inputs to matrices configured such that the vector has an N-dimensional dimension), an N-dimensional output vector corresponding to each of the M input information sets is generated.

In step S270, when an N-dimensional output vector corresponding to each of the M input information sets is generated, an N-dimensional output vector corresponding to each of the M input information sets is a predetermined activation function (the activation function). Is a function that converts the components constituting the N-dimensional output vector corresponding to each of the M input information sets into values of 0 or more and 1 or less) as an input, so that each of the M input information sets Generate the corresponding N-dimensional operation vector.

In step S280, when an N-dimensional operation vector corresponding to each of the M input information sets is generated, an N-dimensional operation vector corresponding to each of the M input information sets is converted to each of the M input information sets. By performing machine learning on the deep neural network so as to maximally approximate the one-hot vector corresponding to, the judgment model for determining the aging grade of the road is generated.

In this case, according to an embodiment of the present invention, the activation function in step S270 may be a softmax function such as Equation 1 above.

At this time, according to an embodiment of the present invention, in step S280, components constituting an N-dimensional operation vector corresponding to each of the M input information sets and a circle corresponding to each of the M input information sets -A loss value for each of the M input information sets is calculated by calculating a loss value based on the loss function according to Equation 2 based on the components constituting the hot vector, and the M input information Machine learning may be performed on the deep neural network so that the average value of the loss values for each of the sets is minimized.

In addition, according to an embodiment of the present invention, the method of operating an electronic device for determining an aging level of a road based on artificial intelligence is, after the generation of the determination model is completed, when an aging level determination command for a first road is received, the Request for transmission of the vibration value and speed value measured from the first vehicle to the first vehicle terminal mounted on the first vehicle running on the first road and at the same time request transmission of vehicle type information of the first vehicle In the step of, the first vehicle terminal receives the first vibration value and the first speed value measured by the first vehicle from the first vehicle terminal, and at the same time, the first vehicle type information is converted into vehicle type information for the first vehicle. Upon receipt, referring to the vehicle model information storage unit, confirming a first numerical value that is a numerical value corresponding to the first vehicle model information, and having the first vibration value, the first speed value, and the first numerical value as components The step of generating a three-dimensional judgment input vector by constructing a three-dimensional vector, and when the three-dimensional judgment input vector is generated, the three-dimensional judgment input vector is converted into the machine learning of the judgment model. Generating an N-dimensional judgment output vector corresponding to the three-dimensional judgment input vector by applying it as an input to the deep neural network, when the N-dimensional judgment output vector is generated, the N-dimensional judgment output Generating an N-dimensional determination operation vector corresponding to the N-dimensional determination output vector by applying a vector as an input to the activation function. When the N-dimensional determination operation vector is generated, the N-dimensional determination operation vector Calculating a similarity between a determination operation vector and a one-hot vector for each of the N aging grades, a similarity with the N-dimensional determination operation vector among the one-hot vectors for each of the N aging grades By confirming the first one-hot vector calculated to be the maximum, and by checking the first aging grade corresponding to the first one-hot vector from the one-hot vector storage unit, The aging grade is the first furnace Determining that it is an aging grade, and when it is determined that the aging grade for the first road is the first aging grade, displaying information on the first aging grade on the screen.

At this time, according to an embodiment of the present invention, in the calculating step, the Manhattan of the vector calculated when the Hadamard product between the N-dimensional determination operation vector and the one-hot vector for each of the N aging grades is calculated. The norm may be calculated as a degree of similarity between the N-dimensional determination operation vector and the one-hot vector for each of the N aging grades.

In the above, a method of operating an electronic device for determining an artificial intelligence-based road aging grade according to an embodiment of the present invention has been described with reference to FIG. 2. Here, the method of operating the electronic device for determining the level of road aging based on artificial intelligence according to an embodiment of the present invention is based on the operation of the electronic device 110 for determining the level of road aging based on the artificial intelligence described with reference to FIG. 1. Since it may correspond to the configuration, a more detailed description thereof will be omitted.

The method of operating an electronic device for determining an artificial intelligence-based road aging level according to an embodiment of the present invention may be implemented as a computer program stored in a storage medium for execution through a combination with a computer.

In addition, the method of operating an electronic device for determining an artificial intelligence-based road aging level according to an embodiment of the present invention may be implemented in the form of program commands that can be executed through various computer means and recorded in a computer-readable medium. . The computer-readable medium may include program instructions, data files, data structures, and the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the present invention, or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -A hardware device specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of the program instructions include not only machine language codes such as those produced by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like.

As described above, in the present invention, specific matters such as specific components, etc., and limited embodiments and drawings have been described, but this is provided only to help a more general understanding of the present invention, and the present invention is not limited to the above embodiments. , If a person of ordinary skill in the field to which the present invention belongs, various modifications and variations are possible from these descriptions.

Therefore, the spirit of the present invention is limited to the described embodiments and should not be defined, and all things that are equivalent or equivalent to the claims as well as the claims to be described later fall within the scope of the spirit of the present invention. .

110: Electronic device for determining road aging grade based on artificial intelligence
111: vehicle model information storage unit 112: one-hot vector storage unit
113: training set storage unit 114: input vector generation unit
115: one-hot vector verification unit 116: output vector generation unit
117: calculation vector generation unit 118: model generation unit
119: information request unit 120: input vector generation unit for determination
121: judgment output vector generation unit 122: judgment operation vector generation unit
123: similarity calculation unit 124: determination unit
125: display

Claims (12)

A vehicle model information storage unit storing a plurality of vehicle model information and predetermined different numerical values corresponding to each of the plurality of vehicle model information;
A one-hot vector storage unit in which a one-hot vector of different N-dimensional predetermined for each of the preset N (N is a natural number of 3 or more) is stored;
Different M (M is a natural number of 2 or more) input information sets that are predetermined as input information to generate a judgment model for determining the aging grade of the road-Each of the M input information sets is based on vibration and speed. A training set storage unit that stores M training sets in which the measured values and vehicle type information for each of the M input information sets are matched with a pre-designated aging level corresponding to each of the M input information sets. ;
After checking a numerical value corresponding to vehicle model information included in each of the M input information sets with reference to the vehicle model information storage unit, a measurement value for vibration and speed included in each of the M input information sets And an input vector generator configured to generate a three-dimensional input vector corresponding to each of the M input information sets by constructing a three-dimensional vector having a numerical value corresponding to vehicle type information as a component.
When a three-dimensional input vector corresponding to each of the M input information sets is generated, a one-hot vector for an aging class matching each of the M input information sets is referred to the one-hot vector storage unit. A one-hot vector identification unit for identifying a one-hot vector corresponding to each of the M input information sets;
A deep neural network consisting of two or more weight matrices using a three-dimensional input vector corresponding to each of the M input information sets-Two or more weight matrices constituting the deep neural network have a dimension of a vector calculated as an output of the deep neural network Matrices configured to be N-dimensional-an output vector generator for generating an N-dimensional output vector corresponding to each of the M input information sets by applying as input to;
When an N-dimensional output vector corresponding to each of the M input information sets is generated, an N-dimensional output vector corresponding to each of the M input information sets is a predetermined activation function-the activation function is It is a function that converts the components constituting the N-dimensional output vector corresponding to each of the M input information sets into values of 0 or more and 1 or less.- Corresponds to each of the M input information sets by applying as input to An operation vector generator for generating an N-dimensional operation vector to be used; And
When an N-dimensional operation vector corresponding to each of the M input information sets is generated, an N-dimensional operation vector corresponding to each of the M input information sets is a circle corresponding to each of the M input information sets- A model generation unit that generates the judgment model to determine the aging grade of the road by performing machine learning on the deep neural network to maximize the hot vector
Electronic device for determining a road aging grade based on artificial intelligence comprising a. The method of claim 1,
The activation function is an artificial intelligence-based electronic device for determining the level of road aging, characterized in that it is a Softmax function as shown in Equation 1 below.
[Equation 1]

here,

Is the value of the i-th component of the components constituting the N-dimensional operation vector corresponding to each of the M input information sets,

Is the value of the i-th component of the components constituting the N-dimensional output vector corresponding to each of the M input information sets,

Means natural constant. The method of claim 2,
The model generation unit
Based on the components constituting the N-dimensional operation vector corresponding to each of the M input information sets and the components constituting the one-hot vector corresponding to each of the M input information sets, the following Equation 2 By calculating a loss value based on a loss function according to, a loss value for each of the M input information sets is calculated, and the average value of the loss values for each of the M input information sets is An electronic device for determining the level of road aging based on artificial intelligence that performs machine learning on the deep neural network to minimize it.
[Equation 2]

here,

Is the loss value,

Is the value of the i-th component of the components constituting the N-dimensional one-hot vector corresponding to each of the M input information sets,

Denotes a value of the i-th component among components constituting an N-dimensional operation vector corresponding to each of the M input information sets. The method of claim 1,
After the generation of the judgment model is completed, when an aging level judgment command for the first road is received, the vibration measured from the first vehicle for the first vehicle terminal mounted on the first vehicle driving on the first road An information request unit that requests transmission of a value and a speed value and simultaneously requests transmission of vehicle type information of the first vehicle;
When the first vibration value and the first speed value measured by the first vehicle terminal are received from the first vehicle terminal and the first vehicle type information is received as vehicle type information for the first vehicle, With reference to the vehicle model information storage unit, a first numerical value, which is a numerical value corresponding to the first vehicle model information, is checked, and a three-dimensional structure having the first vibration value, the first speed value, and the first numerical value as components A determination input vector generator for generating a three-dimensional determination input vector by constructing a vector;
When the three-dimensional determination input vector is generated, the three-dimensional determination input vector is applied as an input to the deep neural network in which the machine learning of the determination model is completed, so that N corresponding to the three-dimensional determination input vector A determination output vector generation unit that generates an output vector for determination of dimensions;
When the N-dimensional determination output vector is generated, by applying the N-dimensional determination output vector as an input to the activation function, an N-dimensional determination operation vector corresponding to the N-dimensional determination output vector is generated. A determination operation vector generator;
A similarity calculating unit for calculating a similarity between the N-dimensional determination operation vector and the one-hot vector for each of the N aging grades when the N-dimensional determination operation vector is generated;
Among the one-hot vectors for each of the N aging grades, a first one-hot vector with a maximum similarity to the N-dimensional determination operation vector is identified, and the first one-hot vector from the one-hot vector storage unit A determination unit determining that the aging grade for the first road is the first aging grade by checking the first aging grade, which is the aging grade corresponding to the one-hot vector; And
When it is determined that the aging grade for the first road is the first aging grade, a display unit that displays information on the first aging grade on the screen
Electronic device for determining an artificial intelligence-based road aging grade further comprising a. The method of claim 4,
The similarity calculation unit
The N-dimensional determination of the Manhattan norm of the vector calculated when the Hadamard product between the N-dimensional determination operation vector and the one-hot vector for each of the N aging grades is calculated An electronic device for determining a road aging grade based on artificial intelligence that calculates a degree of similarity between an operation vector and a one-hot vector for each of the N aging grades. Maintaining a vehicle model information storage unit in which a plurality of vehicle model information and predetermined different numerical values corresponding to each of the plurality of vehicle model information are stored;
Maintaining a one-hot vector storage unit in which a preset N-dimensional one-hot vector for each of N (N is a natural number of 3 or more) set in advance is stored;
Different M (M is a natural number of 2 or more) input information sets that are predetermined as input information to generate a judgment model for determining the aging grade of the road-Each of the M input information sets is based on vibration and speed. A training set storage unit in which M training sets matching a pre-designated aging level as corresponding to each of the M input information sets and the vehicle type information are stored in one set. Maintaining;
After checking a numerical value corresponding to vehicle model information included in each of the M input information sets with reference to the vehicle model information storage unit, a measurement value for vibration and speed included in each of the M input information sets And generating a three-dimensional input vector corresponding to each of the M input information sets by constructing a three-dimensional vector having a numerical value corresponding to vehicle type information as a component.
When a three-dimensional input vector corresponding to each of the M input information sets is generated, a one-hot vector for an aging class matching each of the M input information sets is referred to the one-hot vector storage unit. By confirming the one-hot vector corresponding to each of the M input information sets;
A deep neural network consisting of two or more weight matrices using a three-dimensional input vector corresponding to each of the M input information sets-Two or more weight matrices constituting the deep neural network have a dimension of a vector calculated as an output of the deep neural network Generating an N-dimensional output vector corresponding to each of the M input information sets by applying as input to the matrices configured to be N-dimensional;
When an N-dimensional output vector corresponding to each of the M input information sets is generated, an N-dimensional output vector corresponding to each of the M input information sets is a predetermined activation function-the activation function is It is a function that converts the components constituting the N-dimensional output vector corresponding to each of the M input information sets into values of 0 or more and 1 or less.- Corresponds to each of the M input information sets by applying as input to Generating an N-dimensional arithmetic vector; And
When an N-dimensional operation vector corresponding to each of the M input information sets is generated, an N-dimensional operation vector corresponding to each of the M input information sets is a circle corresponding to each of the M input information sets- Generating the judgment model for determining an aging grade for a road by performing machine learning on the deep neural network to maximize the proximity to a hot vector
A method of operating an electronic device for determining an artificial intelligence-based road aging grade comprising a. The method of claim 6,
The activation function is a method of operating an electronic device for determining a road aging grade based on artificial intelligence, characterized in that the activation function is a softmax function as shown in Equation 1 below.
[Equation 1]

here,

Is the value of the i-th component of the components constituting the N-dimensional operation vector corresponding to each of the M input information sets,

Is the value of the i-th component of the components constituting the N-dimensional output vector corresponding to each of the M input information sets,

Means natural constant. The method of claim 7,
The step of generating the judgment model
Based on the components constituting the N-dimensional operation vector corresponding to each of the M input information sets and the components constituting the one-hot vector corresponding to each of the M input information sets, the following Equation 2 By calculating a loss value based on a loss function according to, a loss value for each of the M input information sets is calculated, and the average value of the loss values for each of the M input information sets is A method of operating an electronic device for determining the level of road aging based on artificial intelligence that performs machine learning on the deep neural network to a minimum.
[Equation 2]

here,

Is the loss value,

Is the value of the i-th component of the components constituting the N-dimensional one-hot vector corresponding to each of the M input information sets,

Denotes a value of the i-th component among components constituting an N-dimensional operation vector corresponding to each of the M input information sets. The method of claim 6,
After the generation of the judgment model is completed, when an aging level judgment command for the first road is received, the vibration measured from the first vehicle for the first vehicle terminal mounted on the first vehicle driving on the first road Requesting transmission of a value and a speed value and simultaneously requesting transmission of vehicle type information of the first vehicle;
When the first vibration value and the first speed value measured by the first vehicle terminal are received from the first vehicle terminal and the first vehicle type information is received as vehicle type information for the first vehicle, With reference to the vehicle model information storage unit, a first numerical value, which is a numerical value corresponding to the first vehicle model information, is checked, and a three-dimensional structure having the first vibration value, the first speed value, and the first numerical value as components Generating a three-dimensional input vector for determination by constructing a vector;
When the three-dimensional determination input vector is generated, the three-dimensional determination input vector is applied as an input to the deep neural network in which the machine learning of the determination model is completed, so that N corresponding to the three-dimensional determination input vector Generating an output vector for dimensional determination;
When the N-dimensional determination output vector is generated, by applying the N-dimensional determination output vector as an input to the activation function, an N-dimensional determination operation vector corresponding to the N-dimensional determination output vector is generated. Step to do;
When the N-dimensional judgment operation vector is generated, calculating a similarity between the N-dimensional judgment operation vector and the one-hot vector for each of the N aging grades;
Among the one-hot vectors for each of the N aging grades, a first one-hot vector with a maximum similarity to the N-dimensional determination operation vector is identified, and the first one-hot vector from the one-hot vector storage unit Determining that the aging grade for the first road is the first aging grade by checking a first aging grade, which is an aging grade corresponding to a one-hot vector; And
If it is determined that the aging grade for the first road is the first aging grade, displaying information on the first aging grade on a screen
An electronic device operating method for determining an artificial intelligence-based road aging level further comprising a. The method of claim 9,
The calculating step
The N-dimensional determination of the Manhattan norm of the vector calculated when the Hadamard product between the N-dimensional determination operation vector and the one-hot vector for each of the N aging grades is calculated A method of operating an electronic device for determining a road aging level based on artificial intelligence that calculates a degree of similarity between an operation vector and a one-hot vector for each of the N aging classes. A computer-readable recording medium recording a computer program for executing the method of any one of claims 6 to 10 through combination with a computer. A computer program stored in a storage medium for executing the method of any one of claims 6 to 10 through combination with a computer.

Images