Anomaly analysis on indoor office spaces for facility management using deep learning methods 리뷰

• This study analyzes the feasibility of anomaly analysis using deep learning, based on data collected by IoT devices on the conditions of indoor spaces, for facility management.
• Three IoT devices가 사용됨
  1. temperature
  2. humidity
  3. brightness
• LSTM model 사용
  • LSTM is representative recurrent neural network method.

1. Introduction

• open office의 시장이 커지고 있음
• open office 관련 연구가 많이 진행되고 있음
• 사람과 관련되어 있다보니 privacy랑 밀접하게 연관되어있음
• In this study, we used a deep learning method to perform anomaly analysis on the data collected using IoT devices for achieving facility management of open offices.

2. Data acquisition using IoT devices

2.1 IoT device fabrication

• temperature, humidity, brightness를 측정할 수 있는 모듈을 3개 제작함
• 특정 위치에 모듈을 설치함

2.2 IoT device installation

• 1, 3 모듈은 같은 공간에 설치함
• 2번은 떨어진 곳에 설치함
• Research should be conducted to determine how to maximize the collection of spatial information in restrictive locations.

2.3 Data acquisition

• ThingSpeak을 활용하여 데이터 수집
• 수집한 데이터를 통해서 낮시간에 온도와 습도가 대체로 낮은 것을 확인할 수 있었음
  • 에어컨 틀었기 때문
• In deep learning, the data undergo a scaling process; therefore, we could not assign units such as lux to the data. Nonetheless, each data record defines both the brightness change in the spaces and the difference between direct and indirect lightings. => 무슨 말인지 확실하게 이해 안됨
  • scaling process를 겪어야 해서 lux 단위를 쓰지 못한다...?
  • 각 저장 기록이 밝기의 변화와 직간접적 빛의 차이를 의미한다는 것도... ㅜ

3. Methods

3.1 Long short-term memory

• The LSTM is considerably suitable for time-series sequences.

• RNNs exhibit limitations in handling long-term dependencies, because the gradient either vanishes or explodes for the training process.

• An LSTM network comprises a forget gate, input gate, and output gate, and can appropriately handle long-term dependencies.

• Given the previous hidden state, previous cell state, and current input feature, an LSTM network determines each gate value as follows:

  • $f_r = \sigma(W_f \cdot [C_{t-1}, h_{t-1}, x_t] + b_f) \\ i_t = \sigma(W_i \cdot [C_{t-1}, h_{t-1}, x_t] + b_i) \\ o_t = \sigma(W_o \cdot [C_{t}, h_{t-1}, x_t] + b_o)$

  • f, i, o: forget, input, output gate values respectively

  • sigma: the activation function

    • In LSTM, the logistic sigmoid function is utilized as the activation function.
    • Because these gates appropriately control long-term dependencies, LSTM network can properly resolve the vanishing or exploding gradient problem.

3.2 Deep learning model and input feature

• Two LSTM layers were constructed, and a dense layer and output layer were configured.
  • LSTM 2개
  • dense layer 1개
  • output layer 1개
• The ReLU activation function and the mean squared error loss function were used.
• The optimizer used Adam algorithm.
• If the learning condition is regression, ReLU, MSE, Adam are favorable option for training.
• Normalization was performed to correct the differences in data scales before the training process.

4. Result and discussion

4.1 Case 1

4.2 Case 2

• 온도와 습도 데이터를 동시에 활용해서 온도를 예측하는 경우 MAPE가 9.28%임
  • 이는 온도와 습도 데이터의 변화가 비슷하기 때문이다.

4.3 Case 3

• brightness와 brightness 데이터를 동시에 활용하여 brightness를 예측하는 경우 MAPE는 8.28%임
  • 너무 당연한 얘기다. (유리 한장 있고 없고의 차이이므로)

5. Summary and conclusions

• Changes in office environments and the advent of new technologies necessitate changes in the facility management methods.
  • 무슨 얘기를 하고자 하는건지 모르겠다.
• brightness와 온도를 활용하여 습도의 패턴을 찾을 수 있다. (예측은 안되고, 패턴은 어느 정도 찾을 수 있음)
• brightness와 습도를 활용하여 온도를 예측할 수 있다.