A Novel Machine Learning Approach to Optimize Energy Forecasting in Industry

The rising energy costs and increasing environmental concerns have compelled industries to seek innovative solutions for optimizing their energy consumption. This paper presents a data-driven approach utilizing Internet of Things (IoT) sensors to collect energy data and a comparative study on various time series forecasting techniques to improve energy forecasting in industries. This experiment is based on public energy usage dataset from steel company which produces various types of steel plates, iron plates and coils. In this paper, we have discussed about the data preprocessing techniques including effective feature engineering and use various Machine Learning (ML) techniques to identify patterns and trends. Also, we have compared the results with previous research outcomes on various time series forecasting models such as deep learning, and other ensemble techniques tested with this dataset for improved energy consumption forecasts. This paper shows how including additional features improves the prediction accuracy of the models allowing industries to achieve optimal energy utilization. The results shows that XGBoost model consistently outperformed other models in energy forecasting with lowest RMSE value of 0.037 and MAE of 0.015 comparing with other deep learning and tree based techniques. The proposed methodology has the potential to significantly reduce energy costs by finding the area to save energy, improve sustainability, and enhance operational efficiency in various industrial sectors.