Constructing an Intelligent Environmental Monitoring and Forecasting System: Fusion of Deep Neural Networks and Gaussian Smoothing

Abstract

To enhance monitoring of environmental indicators like temperature, humidity, and carbon dioxide (CO 2) concentration in data centers, this study evaluates various deep neural network (DNN) models and improves their forecast accuracy using Gaussian smoothing. Initially, multiple DNN architectures were assessed. Following these evaluations, the optimal algorithm was selected for each indicator: CNN fortemperature, LSTM for humidity, and a hybrid LSTM-GRU model for CO 2 concentration. These models underwent further refinement through Gaussian smoothing and re-training to enhance their forecasting capabilities. The results demonstrate that Gaussian smoothing significantly enhanced forecast accuracy across all indicators. For instance, R 2 values notably increased: the temperature forecast improved from 0.59925 to 0.98012, humidity from 0.63305 to 0.99628, and CO 2 concentration from 0.71204 to 0.99855. Thus, this study highlights the potential of DNN models in environmental monitoring after Gaussian smoothing, providing precise forecasting tools and real-time monitoring support for informed decision-making in the future.