A Smart Industrial Electrical Energy Analytics and Forecasting System

The rise of digital smart energy meters with advanced industrial communication protocols has availed opportunities to easily harvest utility energy data in the modern industry 4.0 era. Industrial utility load profile data aggregated over time can be used as a dataset for training of machine learning models for the prediction of future consumption and accrued energy bill costs in an industrial setup. This paper presents a smart industrial electrical energy analytics and forecasting system that utilizes ultra-modern machine learning techniques to predict energy consumption and estimated energy bill based on historical data. An electronic data acquisition unit that comprises a Raspberry Pi 4B, an industrial energy protocol converter, and a 3-phase smart energy meter was developed and deployed for data collection. Readings were stored locally on the Raspberry Pi every 5 min and synched to the cloud for redundancy purposes. Machine learning models were developed using the logged data to predict future energy consumption patterns. Two time-series machine learning forecasting algorithms, i.e., Facebook Prophet and Auto-Regressive Integrated Moving Average (ARIMA) were employed in training the model using the train dataset and exhibited Mean Absolute Percentage Error (MAPE) of 17.72 and 18.86, respectively, when tested with unseen data. A web dashboard was developed to visualize readings from the data acquisition unit as well as forecasted energy trends from which different energy analysis and insights can be generated.