A Review on Forecasting Models and Anomaly Detection for Household Energy Consumption

Smart meters collect data from household appliances to achieve optimal energy consumption from the consumer and utility perspective. The data collected should be analyzed and processed to design an effective control structure. Detecting anomalous behaviour in the appliances becomes vital to maintain energy demands and optimize energy consumption. This article discusses energy consumption forecasting and anomaly detection, in which various statistical and intelligent methods are compared. The performance of the various forecasting models is evaluated with multiple time series evaluation metrics. Statistical forecasting models like autoregressive integrated moving average (ARIMA) and seasonal autoregressive integrated moving averages (SARIMA) are compared with intelligent methods such as univariate long short term memory (uni-LSTM) models and multivariate long short term memory (multi-LSTM) models. From the results and analysis performed, it is concluded that the multi-LSTM models give better results and can detect anomalous behaviour more efficiently compared to their counterparts.