An Energy-Image Based Multi-Unit Power Load Forecasting System

Abstract

Electric energy is one of the most important energy sources for modern industry. The electrical power system is expected to achieve the dynamic balance between electricity generation and electricity consumption, to avoid thriftless excessive generation or power shortage. In this paper, we propose EMPLF, an energy-image based multi-unit power load forecasting system that applies Internet of Things (IoT) techniques on traditional electricity industry. To solve the inaccuracy caused by the diversity of unit power consumption, EM-PLF exploits multiple models to predict the power loads of different units in a factory. To gather the prediction supporting data, we design an embedded device platform to collect the fine-grained power consumption as energy-image snapshots. We also propose the power load prediction algorithm based on Long Short-Term Memory (LSTM) neural network, taking the time correlation of power loads into consideration. We implement and run our system in a real-world factory for more than one year and evaluate its performance with a 500-day real operation data set. The results demonstrate that EM-PLF significantly improves the prediction accuracy.