표제지
요약문
목차
제1장 서론 13
제1절 연구배경과 목적 13
제2절 연구범위와 방법 16
제2장 이론적 배경 22
제1절 기계학습 기반 소음 예측 22
제2절 인공신경망 구조 및 알고리즘 최적화 25
제3절 도로교통소음 예측 모델 34
제3장 연구자료와 방법 41
제1절 소음 자료 수집 41
제2절 시뮬레이션 기반 데이터 생성 47
제3절 학습 자료 구성 49
제4장 실외 소음도 예측 모델 54
제1절 신경망 모형 작성 및 모델 성능 비교 54
제2절 데이터 세트에 따른 성능 비교 67
제3절 소결 76
제5장 실내 소음도 예측 모델 79
제1절 소음 예측 모델 작성 및 성능 평가 81
제2절 변수별 예측 성능 영향도 평가 85
제3절 소결 90
제6장 결론 91
참고문헌 96
ABSTRACT 104
Table 2.1. Comparison between accuracy and time complexity of different Neural Networks having different number of hidden layers 33
Table 2.2. Comparative road traffic noise prediction models across countries 37
Table 2.3. Road surface correction factor, DStr₀ of RLS-90[이미지참조] 40
Table 3.1. Predictor variables for outdoor noise prediction 52
Table 3.2. Predictor variables for indoor noise prediction 53
Table 4.1. Data set composition for each scenario 60
Table 4.2. Results of baseline model performance 63
Table 4.3. Architecture summary of the prediction model 1 64
Table 4.4. Architecture summary of the prediction model 2 65
Table 4.5. Hyper-parameter specifications for model 1, 2 66
Table 4.6. Cross-validation MAE and MSE metrics by scenario 68
Table 4.7. Scenario-specific test performance summary 70
Table 4.8. Predicted values from simulation model 74
Table 5.1. Architecture summary of the indoor noise prediction model 81
Table 5.2. Hyperparameter specifications for indoor noise prediction models 82
Table 5.3. 5-fold validation outcomes measured by MAE 83
Table 5.4. Ranked importance of variables for the outdoor noise prediction model 86
Table 5.5. Ranked importance of variables for the indoor noise prediction model 88
Figure 1.1. Flow diagram for outdoor noise prediction study 19
Figure 1.2. Overview of the research process for ANN model development 21
Figure 2.1. The graph of a single-layer perceptron 25
Figure 2.2. The graph of a multi-layer perceptron 26
Figure 2.3. Information processing of the neural network 28
Figure 2.4. Small learning rate : Stuck in Local Minimum 29
Figure 2.5. Large learning rate : Divergence 29
Figure 2.6. Comparison of activation functions (a) Sigmoid, (b) Hyperbolic Tangent and (c) ReLU 30
Figure 2.7. Workflow of GIS and noise mapping 36
Figure 3.1. Layout types of building and road arrangement (a) Horizontal, (b) Vertical, (c) Mixed 42
Figure 3.2. Measurement points : lower and upper floors 43
Figure 3.3. Sound attenuation for building stories 44
Figure 3.4. Measurement points : indoor and outdoor 45
Figure 3.5. (a) Outdoor and (b) indoor noise measurements 46
Figure 3.6. Flowchart of the data generation based on simulation 47
Figure 3.7. (a) Site boundary and (b) Traffic segmentation 49
Figure 4.1. Visualization of forward propagation 56
Figure 4.2. Five-fold cross validation method 61
Figure 4.3. Cross-Validation MSE by scenario and fold 69
Figure 4.4. Actual vs Predicted values scatter plot for (a) Scenario 1 and (b) Scenario 2 72
Figure 4.5. Actual vs Predicted values scatter plot for (a) Scenario 3 and (b) Scenario 4 73
Figure 5.1. Deep Neural Network Model Architecture for indoor noise prediction 80
Figure 5.2. Comparison of test results between (a) Scenario 1 and (b) Scenario 2 84
Figure 5.3. Box plot of variable importance for the outdoor noise prediction model 87
Figure 5.4. Box plot of variable importance for the indoor noise prediction model 89