KR102489586B1 - Recurrent Neural Network-Based Optimal Sensing Duty Cycle Control Method and Apparatus for Wireless Sensor Networks - Google Patents

Abstract

A recurrent neural network-based optimal sensing duty cycle control method and device in a wireless sensor network are provided. The recurrent neural network-based optimal sensing duty cycle control device in a wireless sensor network provided in the present invention comprises: a data generation unit that generates data for recurrent neural network learning by creating a virtual digital twin using the location of a sensor node for object detection and information on an object movement model detected through the sensor node; a duty cycle calculation unit that calculates a duty cycle mode depending on the moving direction and speed of the object detected through the sensor node; and a recurrent neural network learning unit that predicts an optimal duty cycle mode using the calculated duty cycle mode, whether the object is detected through the sensor node, and the location of the sensor node. Accordingly, object tracking accuracy can be improved.

Description

Recurrent Neural Network-Based Optimal Sensing Duty Cycle Control Method and Apparatus for Wireless Sensor Networks}

The present invention relates to a method and apparatus for adjusting an optimal sensing duty cycle based on a recurrent neural network in a wireless sensor network.

A wireless sensor network is a technology that configures an ad-hoc network between sensor nodes wirelessly without a base station and collects and exchanges data.

An application of the wireless sensor network is an object tracking wireless sensor network that detects and tracks objects on the network. Object Tracking The wireless sensor network can perform monitoring in both indoor and outdoor environments and can track various objects according to the purpose.

Wireless sensor networks require techniques to increase the energy efficiency of sensor nodes in order to extend the lifespan of the network. In order to increase energy efficiency, prior art has proposed a duty cycle method in which a sensor node periodically switches between a sleep mode and a wakeup mode.

An object tracking wireless sensor network requires a technique to increase the accuracy of object tracking. In the case of sleep mode in the duty cycle method, data is not collected. Since an object cannot be detected if the object is within the sensing radius during the corresponding period, consideration of the movement tendency of the object is required when the duty cycle method is applied.

Therefore, in a wireless sensor network for object tracking, a technique for selecting a duty cycle mode capable of maximizing the energy efficiency of a sensor node while satisfying the required object detection requirements is required.

Korean Patent Registration No. 10-2018117 (2019.08.29)

An object of the present invention is to provide a method and apparatus for determining a duty cycle mode for maximizing energy efficiency of a sensor node in a wireless sensor network environment. In addition, the optimal duty cycle mode is determined by using the position of the sensor node for object detection, the information of the object movement model detected through the sensor node, and the duty cycle mode according to the moving direction and speed of the object detected through the sensor node. It is intended to provide a predictive recurrent neural network learning method and apparatus.

In one aspect, the optimal sensing duty cycle control device based on a recurrent neural network in a wireless sensor network proposed in the present invention uses information about the position of a sensor node for object detection and an object movement model sensed through the sensor node to simulate virtual A data generation unit that generates a digital twin of and generates data for recurrent neural network learning, a duty cycle calculator that calculates a duty cycle mode according to the moving direction and speed of an object sensed through the sensor node, and the calculated duty cycle and a recurrent neural network learning unit that predicts an optimal duty cycle mode using the mode and whether an object is detected through the sensor node and the location of the sensor node.

The data generation unit creates a virtual digital twin using real environment information including the arrangement of sensor nodes, an object movement model, and a sensing model, and each sensor node in the digital twin determines whether an object is detected, the location of the sensor node, It stores the object position, the time when the object enters the sensing area of the sensor node, and the time when the object leaves the sensing area of the sensor node.

The duty cycle calculation unit calculates a duty cycle that satisfies the object detection requirements by using information on the movement direction and speed of the object and the time required for the object to pass through the sensing area of the sensor node when the object is inside the sensing area of the sensor node. Calculate.

The recurrent neural network learning unit reconstructs information extracted from the digital twin generated by the data generation unit and the duty cycle mode calculated by the duty cycle calculation unit Extraction data reconstruction unit, LSTM (Long short-term memory) based recurrent neural network A learning unit that learns weights to minimize a cross entropy loss function in order to reduce the difference between the correct label and the predicted label for the data reconstructed by the extraction data reconstructing unit using a learning unit, and whether an object is detected and a sensor according to the learning result of the learning unit and a duty cycle mode predictor for predicting an optimal duty cycle for the location of the node.

In another aspect, the optimal sensing duty cycle control method based on a recurrent neural network in a wireless sensor network proposed in the present invention is a location of a sensor node for object detection through a data generator and an object movement model sensed through the sensor node. Generating a virtual digital twin using information of to generate data for recurrent neural network learning, calculating a duty cycle mode according to the moving direction and speed of an object sensed through the sensor node by a duty cycle calculator, and The recursive neural network training unit predicts an optimal duty cycle mode using the calculated duty cycle mode, whether an object is detected through a sensor node, and a location of a sensor node.

According to the embodiments of the present invention, a duty cycle mode with increased energy efficiency and object tracking accuracy can be calculated through a duty cycle mode derivation method that considers the moving speed of an object and satisfies the object detection requirements, and through this, the lifespan of the network and increase object tracking accuracy. In addition, according to embodiments of the present invention, by extracting a data set for learning a recurrent neural network through a digital twin in which a real environment is implemented as a virtual environment, time and cost for collecting a data set can be reduced. In addition, according to the embodiments of the present invention, by predicting the duty cycle optimized for each sensor node by taking information on the location of sensor nodes in the network and whether each sensor node detects an object as input through a recurrent neural network , it can reduce the complexity of calculation and increase the possibility of utilization in the real environment for real-time moving objects.

1 is a diagram showing the configuration of an optimal sensing duty cycle adjusting device based on a recurrent neural network in a wireless sensor network according to an embodiment of the present invention.
2 is a flowchart illustrating a method for adjusting an optimal sensing duty cycle based on a recurrent neural network in a wireless sensor network according to an embodiment of the present invention.
3 is a diagram for explaining an operating method of a predefined discontinuous duty cycle and a detailed area of a sensor node in an object tracking wireless sensor network environment according to an embodiment of the present invention.
4 is a diagram for explaining a duty cycle calculation process for satisfying object detection requirements in consideration of object motion in an object tracking wireless sensor network environment according to an embodiment of the present invention.
5 is a diagram for explaining a process of reconstructing information extracted from a digital twin in a data generation step for recurrent neural network learning according to an embodiment of the present invention.
6 is a diagram for explaining a process of learning a recurrent neural network and predicting a duty cycle mode through the recurrent neural network according to an embodiment of the present invention.

An embodiment of the present invention relates to a method for selecting an optimized duty cycle mode of sensor nodes using a Recurrent Neural Network (RNN) in a wireless sensor network (WSN) environment, and more particularly, Adjust the optimal duty cycle so that the sensor of the wireless sensor network can satisfy the minimum detection requirements for moving objects, generate training data of the recurrent neural network using supervised learning in the digital twin reflecting the real environment, and create the digital twin. It relates to a recurrent neural network model structure that predicts an optimal duty cycle according to the motion of an object using data extracted through Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

First, terms mainly used in the present invention will be described as follows.

A duty cycle according to an embodiment of the present invention is an operation method of periodically switching between a wakeup mode and a sleep mode, and is expressed as a ratio of a wakeup mode to a sensing period. keep for a long time

A wake up mode according to an embodiment of the present invention means a state in which a sensor node using a duty cycle method is activated, and data can be collected and transmitted/received.

A sleep mode according to an embodiment of the present invention means a state in which a sensor node using a duty cycle method is inactivated, and data collection and transmission/reception cannot be performed.

A recurrent neural network (RNN) according to an embodiment of the present invention is a type of deep learning and refers to an artificial neural network specialized in sequential data processing.

A sensing area (sensing coverage) according to an embodiment of the present invention refers to an area in which a sensor node can sense information in a sensor network.

LSTM (Long short-term memory) according to an embodiment of the present invention is a recurrent neural network technique designed to alleviate the problem that learning is not smooth when the sequence length is increased in the basic recurrent neural network.

A loss function according to an embodiment of the present invention means a loss caused by an error between an estimator and a parameter.

Weight according to an embodiment of the present invention means a parameter used for model learning in deep learning.

Cross entropy according to an embodiment of the present invention is a type of loss function and represents an expected value of a class of an output data set with respect to input data as a probability.

1 is a diagram showing the configuration of an optimal sensing duty cycle adjusting device based on a recurrent neural network in a wireless sensor network according to an embodiment of the present invention.

In the proposed wireless sensor network, the optimal sensing duty cycle control device based on the recurrent neural network includes a data generator 110, a duty cycle calculator 120, and a recurrent neural network learning unit 130.

The data generation unit 110 according to an embodiment of the present invention uses the information of the object movement model sensed through the positions of a plurality of sensor nodes (1, 2, ??, n _s ) for object detection and the sensor nodes, A virtual digital twin 111 is created to generate data for learning the recurrent neural network.

), 센서 노드의 위치(

), 객체 위치(

), 객체가 센서 노드의 센싱 영역에 들어오는 시간(

), 객체가 센서 노드의 센싱 영역에서 나가는 시간(

)의 정보를 저장한다. The data generator 110 according to an embodiment of the present invention creates a virtual digital twin 111 by using real environment information including an arrangement of sensor nodes, an object movement model, and a sensing model. In the created digital twin, each sensor node determines whether an object is detected (

), the position of the sensor node (

), object location (

), the time the object enters the sensing area of the sensor node (

), the time the object leaves the sensing area of the sensor node (

) to store the information.

)를 계산하고, 이때 감지되는 객체의 이동 방향 및 속력에 따른 듀티 사이클 계산에는 객체 감지 요구 조건(

)을 고려한 값을 제공한다. The duty cycle calculation unit 120 according to an embodiment of the present invention is a duty cycle mode according to the moving direction and speed of an object detected through a sensor node (

) is calculated, and the object detection requirement (

) is provided.

The duty cycle calculation unit 120 according to an embodiment of the present invention uses information on the moving direction and speed of the object when the object is inside the sensing area of the sensor node and the time required for the object to pass through the sensing area of the sensor node. Calculate the duty cycle that meets the object detection requirements. A duty cycle calculation process in consideration of object detection requirements according to an embodiment of the present invention will be described in more detail with reference to FIG. 4 .

The recurrent neural network learning unit 130 according to an embodiment of the present invention predicts an optimal duty cycle mode using the calculated duty cycle mode, whether an object is detected through a sensor node, and the location of the sensor node.

The recurrent neural network learning unit 130 according to an embodiment of the present invention includes an extraction data reconstruction unit 131, a learning unit 132, and a duty cycle mode prediction unit 133.

The recurrent neural network learning unit 130 according to an embodiment of the present invention uses the data stored in the data generator for learning the recurrent neural network to predict the optimal duty cycle mode when an object is detected and the location of a sensor node is input. Reorganize for purpose. A recurrent neural network can be learned using the reconstructed data, and an optimal duty cycle mode can be predicted through the recurrent neural network.

The extraction data reconstruction unit 131 according to an embodiment of the present invention reconstructs information extracted from the digital twin generated by the data generator 110 and the duty cycle mode calculated by the duty cycle calculator 120.

The learning unit 132 according to an embodiment of the present invention uses a long short-term memory (LSTM)-based recurrent neural network to reduce the difference between the correct answer label and the prediction label for the data reconstructed in the extraction data reconstruction unit. We learn the weights to minimize the entropy loss function.

The duty cycle mode predictor 133 according to an embodiment of the present invention predicts whether an object is detected and an optimal duty cycle for the location of a sensor node according to the learning result of the learning unit 132 . An extraction data reconstruction process according to an embodiment of the present invention will be described in more detail with reference to FIG. 5 .

2 is a flowchart illustrating a method for adjusting an optimal sensing duty cycle based on a recurrent neural network in a wireless sensor network according to an embodiment of the present invention.

In the proposed wireless sensor network, the recurrent neural network-based optimal sensing duty cycle control method creates a virtual digital twin using the location of the sensor node for object detection through the data generator and the information of the object movement model detected through the sensor node. generating data for recurrent neural network learning (210), a duty cycle calculation unit calculating a duty cycle mode according to the moving direction and speed of an object detected through the sensor node (220), and a recurrent neural network learning unit A recurrent neural network learning step 230 of predicting an optimal duty cycle mode using the calculated duty cycle mode, whether an object is detected through a sensor node, and the location of the sensor node is included.

In step 210, a virtual digital twin is created using the information of the location of the sensor node for object detection and the object movement model sensed through the sensor node through the data generator to generate data for learning the recurrent neural network.

The data generation unit according to an embodiment of the present invention creates a virtual digital twin using real environment information including an arrangement of sensor nodes, an object movement model, and a sensing model. In the created digital twin, each sensor node stores information on whether an object is detected, the location of the sensor node, the location of the object, the time when the object enters the sensing area of the sensor node, and the time when the object leaves the sensing area of the sensor node.

The duty cycle calculation unit according to an embodiment of the present invention calculates a duty cycle mode according to the moving direction and speed of the object detected through the sensor node, and the duty cycle calculation according to the moving direction and speed of the object detected requires object detection. Provides a value considering conditions.

In step 220, the duty cycle calculation unit calculates a duty cycle mode according to the moving direction and speed of the object sensed through the sensor node.

The duty cycle calculation unit according to an embodiment of the present invention, when the object is inside the sensing area of the sensor node, uses information on the movement direction and speed of the object and the time it takes for the object to pass through the sensing area of the sensor node to meet the object detection requirements. Calculate the duty cycle that satisfies

In step 230, the recurrent neural network learning unit predicts an optimal duty cycle mode using the calculated duty cycle mode, whether an object is detected through a sensor node, and the location of the sensor node.

The recurrent neural network learning unit according to an embodiment of the present invention includes an extraction data reconstruction unit, a learning unit, and a duty cycle mode prediction unit.

The recurrent neural network learning unit according to an embodiment of the present invention reorganizes the data stored in the data generator to suit the purpose for learning the recurrent neural network to predict the optimal duty cycle mode when an object is detected and the location of the sensor node is input. do. An extraction data reconstruction process according to an embodiment of the present invention will be described in more detail with reference to FIG. 5 . A recurrent neural network can be learned using the reconstructed data, and an optimal duty cycle mode can be predicted through the recurrent neural network.

Step 230 according to an embodiment of the present invention reconstructs information extracted from the digital twin generated by the data generator and the duty cycle mode calculated by the duty cycle calculator through the extraction data reconstruction unit of the recurrent neural network learning unit. Cross entropy to reduce the difference between the correct answer label and the predicted label for the data reconstructed in the extraction data reconstruction unit using a long short-term memory (LSTM) based recurrent neural network through the learning unit of the recurrent neural network learning unit. Learning weights to minimize a loss function and predicting an optimal duty cycle for the location of a sensor node and whether an object is detected according to a learning result of the learning unit through a duty cycle mode prediction unit of the recurrent neural network learning unit. do.

3 is a diagram for explaining an operation method of a predefined discontinuous duty cycle and a detailed area of a sensor node in an object tracking wireless sensor network environment according to an embodiment of the present invention.

)의

개의 불연속적인 모드로 구분한다. 모드가

인 센서 노드의 센싱 주기(

)(311)는 웨이크업 모드 기간(

)(312), 슬립 모드 기간(

)(313)으로 구성된다. 고정된 웨이크업 모드 기간(

)(312)을 사용하고 선택된 모드에 따라 슬립 모드 기간(

)(313)을 조절한다. 따라서 모드가

일 때 듀티 사이클은 수학식 1로 나타낼 수 있다: Referring to Figure 3 (a), the duty cycle mode is predefined 0 ~ (

)of

divided into two discrete modes. mode is

The sensing cycle of the sensor node (

) 311 is the wake-up mode period (

) 312, sleep mode period (

) (313). Fixed wakeup mode duration (

) 312 and depending on the selected mode, the sleep mode period (

) (313). So the mode is

When , the duty cycle can be expressed as Equation 1:

Equation 1

의 선택된 모드

가

에 가까울수록 슬립 모드 기간(

)(313)의 길이가 길어져 낮은 듀티 사이클이 되고, 0에 가까울수록 슬립 모드 기간(

)(313)의 길이가 짧아져 높은 듀티 사이클이 되며 0인 경우에는 센서 노드를 항상 웨이크업 모드로 유지한다. of Equation 1

selected mode of

go

The closer to , the sleep mode period (

) 313 becomes longer, resulting in a lower duty cycle, and the closer to 0, the sleep mode period (

) 313 is shortened, resulting in a high duty cycle, and when it is 0, the sensor node is always maintained in a wakeup mode.

Referring to FIG. 3(b), the detailed areas of each sensor node include a sensing area 321 where the sensor node can detect an object, and an uncertain area 322 that cannot be detected because the object is outside the sensing area but is within a certain distance. ), and an area 323 outside the uncertain area.

)와 듀티 사이클 모드의 관계를 설명한다. 객체의 위치와 객체가 센서 노드의 영역을 지나가는 속도에 따라 각각의 센서 노드의 듀티 사이클 모드를 결정한다. 객체가 센서 노드의 바깥 영역(323)에 있는 경우에는 가장 낮은 듀티 사이클의 주기(

)(333)를 선택한다. 객체가 센서 노드의 불확실한 영역(322)에 있는 경우에는 약간 높은 듀티 사이클이지만 여전히 낮은 듀티 사이클의 주기(

)(332)를 선택한다. 객체가 센서 노드의 센싱 영역 안(321)에 있는 경우에는 객체가 센싱 영역을 통과하는 속도에 따라서

에서

사이의 듀티 사이클 모드(331)를 선택한다. Referring to FIG. 3(c), the sensing period of the sensor node according to the position of the object (

) and the duty cycle mode are explained. The duty cycle mode of each sensor node is determined according to the position of the object and the speed at which the object passes through the area of the sensor node. If the object is in the outer region 323 of the sensor node, the period of the lowest duty cycle (

) (333). If the object is in the uncertain region 322 of the sensor node, a slightly higher duty cycle but still a period of a lower duty cycle (

) (332). If the object is in the sensing area 321 of the sensor node, according to the speed at which the object passes through the sensing area

at

selects the duty cycle mode 331 between

4 is a diagram for explaining a duty cycle calculation process for satisfying object detection requirements in consideration of object motion in an object tracking wireless sensor network environment according to an embodiment of the present invention.

Referring to FIG. 4 , an object moving direction 411 and an object moving speed 412 when the object is inside the sensing area 413 are shown. The time for an object to pass through the sensing region can be expressed as Equation 2:

Equation 2

은 객체가 센싱 영역을 나오는 시간,

는 객체가 센싱 영역을 들어가는 시간으로 그 차인

는 객체가 센서 노드

의 센싱 영역을 통과하는데 소요되는 시간이다. of Equation 2

is the time the object leaves the sensing area,

is the time the object enters the sensing region, the difference

If the object is a sensor node

is the time required to pass through the sensing region of

가

보다 같거나 큰 경우(41)에는 수학식 3에 의해 듀티 사이클 모드 주기의 길이가 결정된다: According to an embodiment of the present invention, the time the object passes the sensing area

go

In the case of greater than or equal to (41), the length of the duty cycle mode period is determined by Equation 3:

Equation 3

에 한 번은 객체가 감지될 수 있도록 듀티 사이클 모드(431)를 선택한다. 따라서 객체가 센싱 영역에 들어왔을 때 가장 낮은 듀티 사이클 주기인

은 수학식 3에 의해서 결정된다.While an object is within the sensing area, the sensor

Select the duty cycle mode 431 so that an object can be detected once in . Therefore, when an object enters the sensing area, the lowest duty cycle period

Is determined by Equation 3.

가

보다 같거나 긴 시간 동안 소요되는 경우(422)에는 요구되는 감지 요구 조건

을 만족해야 하며 이를 충족하는 최소 듀티 사이클 모드(432)를 갖게 된다. According to an embodiment of the present invention, the time the object passes the sensing area

go

If it takes the same or longer time (422), the required detection requirement

must be satisfied and will have the minimum duty cycle mode 432 that satisfies this.

가

보다 작은 경우(421)에는 최소 감지 요구 조건

을 만족해야 하고 이는 매우 높은 듀티 사이클 모드(431)를 갖게 된다. According to an embodiment of the present invention, the time the object passes the sensing area of the sensor node

go

If less than (421), the minimum detection requirement

must satisfy and this will have a very high duty cycle mode (431).

의 센싱 영역을 통과하는 시간

에 따른 최소 객체 감지 요구 조건은 수학식 4에 의해 결정된다: Therefore, if the object is a sensor node

time to pass through the sensing area of

The minimum object detection requirement according to is determined by Equation 4:

Equation 4

은 센서 노드

의 최소 객체 감지 요구 조건이다. 객체가 센서 노드

의 센싱 영역을 통과할 때 감지 요구 조건을 만족하는 최대 센싱 주기

는 수학식 5에 의해 결정된다: in Equation 4

silver sensor node

is the minimum object detection requirement of object is a sensor node

The maximum sensing period that satisfies the sensing requirements when passing through the sensing area of

is determined by Equation 5:

Equation 5

)는 수학식 6에 의해 결정된다: Therefore, the optimal duty cycle mode (

) is determined by Equation 6:

Equation 6

가

보다 작은 경우에는 모드 0을 선택하여 항상 웨이크업 모드 상태를 유지한다. 최대 센싱 주기

가

과

사이인 경우에는 짧은 센싱 주기인 모드

을 선택한다. 수학식 6을 통해 계산된 최적의 듀티 사이클 모드(

)가 객체 감지 요구 조건(

)을 만족하는 것을 수학식 7을 통해 확인할 수 있다: max sensing cycle

go

If it is less than , mode 0 is selected to always maintain the wakeup mode state. max sensing cycle

go

class

, the short sensing cycle mode

Choose The optimal duty cycle mode calculated through Equation 6 (

) is the object detection requirement (

) can be confirmed through Equation 7:

Equation 7

5 is a diagram for explaining a process of reconstructing information extracted from a digital twin in a data generation step for recurrent neural network learning according to an embodiment of the present invention.

), 센서 노드의 객체 감지 여부(

)를 순환 신경망의 입력으로 사용한다. 객체 감지 여부(

)는 객체가 감지된 경우에는 1, 감지되지 않은 경우에는 0의 이진 정보로 구성된다. 각각의 타임 스텝

에서 입력 타임 스텝 벡터는

로 구성된다.

은 학습 영역에 있는 센서들의 수를 의미한다. The location of the sensor node stored in the data generation step according to the embodiment of the present invention (

), whether the sensor node detects an object (

) is used as the input of the recurrent neural network. Object detection (

) is composed of binary information of 1 when the object is detected and 0 when the object is not detected. each time step

where the input time step vector is

consists of

denotes the number of sensors in the learning area.

를

로 수정하여 사용한다: Referring to FIG. 5 and Equation 8, the maximum sensing period used for learning the recurrent neural network

cast

Modify to use:

Equation 8

)와 센서 노드의 객체 감지 여부(

)는 실용적이지만 객체가 센싱 영역을 통과하는 시간을 명확히 추정하기에는 어려울 수 있다는 점을 고려한다. 따라서 수학식 8에서

는 순환 신경망 학습 데이터 생성 시 사용되는 최대 센싱 주기이고

의 범위를 갖는 센싱 주기 마진

을 이용하여 최대 센싱 주기를 보수적으로 설정한다. 수학식 8을 참조하여 순환신경망 학습 단계에 사용되는 최적의 듀티 사이클 모드(

)는 [수학식 9]의

로 수정되어 사용된다: Equation 8 is the position of the sensor node used as an input of the recurrent neural network (

) and whether the sensor node detects an object (

) is practical, but consider that it may be difficult to unequivocally estimate the time an object passes through the sensing region. Therefore, in Equation 8

is the maximum sensing period used when generating recurrent neural network training data, and

Sensing cycle margin with a range of

Set the maximum sensing period conservatively using . Referring to Equation 8, the optimal duty cycle mode used in the recurrent neural network learning step (

) of [Equation 9]

Modified to use:

Equation 9

)(510)은 입력 타임 스텝 벡터

개를 하나의 데이터(511)로 간주한다. 이에 상응하는 출력 데이터 셋(

)(520)는 수학식 9를 참조하여 타임 스텝

에서 각각의 센서 노드들에 최적의 듀티 사이클 모드(521)로 구성된다. 순환 신경망 데이터 셋은

개(522)의 데이터로 구성된다. According to an embodiment of the present invention, the input data set of the recurrent neural network (

) 510 is the input time step vector

A dog is regarded as one piece of data 511 . The corresponding output data set (

) 520 is the time step with reference to Equation 9

It is composed of an optimal duty cycle mode 521 for each sensor node in . The recurrent neural network data set is

It consists of data of dog 522.

6 is a diagram for explaining a process of learning a recurrent neural network and predicting a duty cycle mode through the recurrent neural network according to an embodiment of the present invention.

)를 최소화하는 방향으로 가중치(

)를 학습한다. According to an embodiment of the present invention, a long short-term memory (LSTM) based recurrent neural network 610 uses a cross entropy loss function (

) in the direction of minimizing the weight (

) to learn.

The present invention relates to a method and apparatus for selecting an optimized duty cycle mode of sensor nodes using a recurrent neural network in a wireless sensor network environment, and more particularly, to a sensor of a wireless sensor network that satisfies a minimum detection requirement for a moving object. It relates to a method for adjusting an optimal duty cycle to be satisfied, generating learning data of a recurrent neural network in a digital twin, and learning a recurrent neural network to predict an optimal duty cycle.

According to an embodiment of the present invention, it is possible to maximize energy efficiency and increase object tracking accuracy by providing a duty cycle derivation method capable of satisfying an object detection requirement by considering the position of an object, the movement speed of the object, and the like. In addition, the time and cost required for learning can be greatly reduced by using a virtual digital twin that reflects the real environment, and the optimal duty cycle can be effectively predicted for multi-object and various sensor network environments through a recurrent neural network.

The devices described above may be implemented as hardware components, software components, and/or a combination of hardware components and software components. For example, devices and components described in the embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA), It may be implemented using one or more general purpose or special purpose computers, such as a programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. A processing device may run an operating system (OS) and one or more software applications running on the operating system. A processing device may also access, store, manipulate, process, and generate data in response to execution of software. For convenience of understanding, there are cases in which one processing device is used, but those skilled in the art will understand that the processing device includes a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that it can include. For example, a processing device may include a plurality of processors or a processor and a controller. Other processing configurations are also possible, such as parallel processors.

Software may include a computer program, code, instructions, or a combination of one or more of the foregoing, which configures a processing device to operate as desired or processes independently or collectively. The device can be commanded. Software and/or data may be any tangible machine, component, physical device, virtual equipment, computer storage medium or device, intended to be interpreted by or provide instructions or data to a processing device. can be embodied in Software may be distributed on networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer readable media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. Program commands recorded on the medium may be specially designed and configured for the embodiment 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. - includes hardware devices specially configured to store and execute program instructions, such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language codes that can be executed by a computer using an interpreter, as well as machine language codes such as those produced by a compiler.

As described above, although the embodiments have been described with limited examples and drawings, those skilled in the art can make various modifications and variations from the above description. For example, the described techniques may be performed in an order different from the method described, and/or components of the described system, structure, device, circuit, etc. may be combined or combined in a different form than the method described, or other components may be used. Or even if it is replaced or substituted by equivalents, appropriate results can be achieved.

Therefore, other implementations, other embodiments, and equivalents of the claims are within the scope of the following claims.

Claims (8)

a data generator for generating data for learning a recurrent neural network by creating a virtual digital twin using information on a location of a sensor node for object detection and an object movement model sensed through the sensor node;
a duty cycle calculation unit calculating a duty cycle mode according to the moving direction and speed of the object sensed through the sensor node; and
Recurrent neural network learning unit predicting an optimal duty cycle mode using the calculated duty cycle mode, whether an object is detected through a sensor node, and the location of a sensor node
including,
The duty cycle calculator,
When the object is inside the sensing area of the sensor node, a duty cycle that satisfies the object detection requirement is calculated using information on the moving direction and speed of the object and the time required for the object to pass through the sensing area of the sensor node,
The object detection requirements are,
Determine the length of the duty cycle mode period according to the case where the object is outside the sensing area of the sensor node, the uncertain area of the sensor node, and the case inside the sensing area,
In case the object is inside the sensing area of the sensor node, a duty cycle mode in which the object is sensed at least once while passing through the sensing area by using the time required for the object to pass through the sensing area according to the moving direction and speed of the object. Determining the length of the period to determine the maximum sensing period for the object, and if the maximum sensing period for the object is smaller than the sensing period of the sensor node to always keep the sensor in a wake-up mode state
A sensing duty cycle control device based on a recurrent neural network. According to claim 1,
The data generator,
Create a virtual digital twin using real environment information including sensor node arrangement, object movement model, and sensing model,
In the digital twin, each sensor node stores information on whether an object is detected, the location of the sensor node, the location of the object, the time when the object enters the sensing area of the sensor node, and the time when the object leaves the sensing area of the sensor node.
A sensing duty cycle control device based on a recurrent neural network. delete According to claim 1,
The recurrent neural network learning unit,
an extraction data reconstruction unit for reconstructing information extracted from the digital twin generated by the data generator and the duty cycle mode calculated by the duty cycle calculator;
a learning unit learning weights to minimize a cross entropy loss function in order to reduce a difference between an answer label and a prediction label for the data reconstructed in the extraction data reconstruction unit using a long short-term memory (LSTM)-based recurrent neural network; and
A duty cycle mode predictor for predicting an optimal duty cycle for object detection and sensor node location according to the learning result of the learning unit.
Recurrent neural network-based sensing duty cycle control device comprising a. Generating data for recurrent neural network learning by creating a virtual digital twin using information of a location of a sensor node for object detection and an object movement model sensed through a sensor node through a data generator;
calculating, by a duty cycle calculation unit, a duty cycle mode according to a moving direction and speed of an object sensed through the sensor node; and
A recursive neural network learning step in which the recursive neural network learning unit predicts an optimal duty cycle mode using the calculated duty cycle mode, whether an object is detected through a sensor node, and the location of the sensor node.
including,
The step of calculating the duty cycle mode according to the moving direction and speed of the object sensed through the sensor node by the duty cycle calculator,
When the object is inside the sensing area of the sensor node, a duty cycle that satisfies the object detection requirement is calculated using information on the moving direction and speed of the object and the time required for the object to pass through the sensing area of the sensor node,
The object detection requirements are,
Determine the length of the duty cycle mode period according to the case where the object is outside the sensing area of the sensor node, the uncertain area of the sensor node, and the case inside the sensing area,
In case the object is inside the sensing area of the sensor node, a duty cycle mode in which the object is sensed at least once while passing through the sensing area by using the time required for the object to pass through the sensing area according to the moving direction and speed of the object. Determining the length of the period to determine the maximum sensing period for the object, and if the maximum sensing period for the object is smaller than the sensing period of the sensor node to always keep the sensor in a wake-up mode state
A sensing duty cycle control method based on a recurrent neural network. According to claim 5,
The step of generating data for learning the recurrent neural network by creating a virtual digital twin using the information of the location of the sensor node for object detection and the object movement model sensed through the sensor node through the data generator,
Create a virtual digital twin using real environment information including sensor node arrangement, object movement model, and sensing model,
In the digital twin, each sensor node stores information on whether an object is detected, the location of the sensor node, the location of the object, the time when the object enters the sensing area of the sensor node, and the time when the object leaves the sensing area of the sensor node.
A sensing duty cycle control method based on a recurrent neural network. delete According to claim 5,
The recursive neural network learning step in which the recursive neural network learning unit predicts an optimal duty cycle mode using the calculated duty cycle mode and whether an object is detected through a sensor node and a location of a sensor node,
reconstructing information extracted from the digital twin generated by the data generator and the duty cycle mode calculated by the duty cycle calculator through an extraction data reconstruction unit of the recurrent neural network learning unit;
Through the learning unit of the recurrent neural network learning unit, a cross entropy loss function is used to reduce the difference between the correct answer label and the prediction label for the data reconstructed in the extraction data reconstruction unit using a long short-term memory (LSTM) based recurrent neural network. learning weights to minimize; and
Predicting, through a duty cycle mode prediction unit of the recurrent neural network learning unit, an optimal duty cycle for whether an object is detected and a location of a sensor node according to a learning result of the learning unit
Recurrent neural network-based sensing duty cycle control method comprising a.

Images