Scholarlink

This study presents an integrated environmental noise monitoring and forecasting system for campus buildings, combining IoT-based sensing, Kalman filtering, and deep learning. A stacked Bidirectional Long Short-Term Memory (Bi-LSTM) model was trained on multivariate time-series inputs—including time-of-day encodings, foot traffic proxies, and class activity indicators—to predict hourly noise levels over a 7-day horizon. Noise data was collected using LM393 sensors and ESP8266 microcontrollers, transmitted to ThingSpeak via Wi-Fi, and analyzed in MATLAB. The system was deployed across three academic zones—hallway, classroom, and laboratory—achieving MAPE values of 9.30%, 10.20%, and 10.70%, respectively. To validate performance, the Bi-LSTM was benchmarked against unidirectional LSTM and SARIMA models using residual error and forecasting metrics (RMSE, MAE, MAPE). Results show that Bi-LSTM outperformed baseline models in capturing both structured and irregular noise fluctuations. These findings demonstrate the system’s applicability for proactive noise forecasting in Smart Campus settings and provide a scalable, data-driven foundation for informed policy planning.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep Learning-Based Forecasting of Environmental Noise in Campus Buildings Using Bi-LSTM and IoT Integration

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Reymark John A. Macapanas,
  • Apple Rose B. Alce,
  • Adrian P. Galido

摘要

This study presents an integrated environmental noise monitoring and forecasting system for campus buildings, combining IoT-based sensing, Kalman filtering, and deep learning. A stacked Bidirectional Long Short-Term Memory (Bi-LSTM) model was trained on multivariate time-series inputs—including time-of-day encodings, foot traffic proxies, and class activity indicators—to predict hourly noise levels over a 7-day horizon. Noise data was collected using LM393 sensors and ESP8266 microcontrollers, transmitted to ThingSpeak via Wi-Fi, and analyzed in MATLAB. The system was deployed across three academic zones—hallway, classroom, and laboratory—achieving MAPE values of 9.30%, 10.20%, and 10.70%, respectively. To validate performance, the Bi-LSTM was benchmarked against unidirectional LSTM and SARIMA models using residual error and forecasting metrics (RMSE, MAE, MAPE). Results show that Bi-LSTM outperformed baseline models in capturing both structured and irregular noise fluctuations. These findings demonstrate the system’s applicability for proactive noise forecasting in Smart Campus settings and provide a scalable, data-driven foundation for informed policy planning.