Modeling Electricity Consumption and Production in Smart Homes using LSTM Networks

Abstract

This paper presents a forecasting method of the electricity consumption and production in a household equipped with photovoltaic panels and a smart energy management system. The prediction is performed with a Long Short-Term Memory recurrent neural network. The datasets collected during five months in a household are used for the evaluations. The recurrent neural network is configured optimally to reduce the forecasting errors. The results show that the proposed method outperforms an earlier developed Multi-Layer Perceptron, as well as the Autoregressive Integrated Moving Average statistical forecasting algorithm.

Electricity production and consumption modeling through fuzzy logic

“…LSTM is a recurrent neural network and was exploited to predict electricity production and consumption by Bachici and Gellert. 20 LSTM has a memory component that transmits the information learned at the current timestep to the next timesteps. It is able to forget unnecessary information from the preceding state, forwarding only the relevant parts of the state to the output.…”

“…A prior classification and filtration of the data is recommended before being processed by the predictor. The data sets collected by Feilmeier 11 have been used to evaluate several predictors of different types, namely: a multilayer perceptron (MLP), 11 a Markovian predictor, 19 a predictor based on long short-term memory (LSTM), 20 a predictor based on a variation of the ARIMA algorithm 21,22 and a predictor based on the TBATS algorithm. 22 Gellert et al 19 compared a context-based predictor (Markovian predictor), an incremental predictor and a hybrid predictor (which combines the functionalities of both the Markovian and the incremental predictors).…”

“…LSTM is a recurrent neural network and was exploited to predict electricity production and consumption by Bachici and Gellert 20 . LSTM has a memory component that transmits the information learned at the current timestep to the next timesteps.…”

“…A prior classification and filtration of the data is recommended before being processed by the predictor. The data sets collected by Feilmeier 11 have been used to evaluate several predictors of different types, namely: a multilayer perceptron (MLP), 11 a Markovian predictor, 19 a predictor based on long short‐term memory (LSTM), 20 a predictor based on a variation of the ARIMA algorithm 21,22 and a predictor based on the TBATS algorithm 22 …”

“…It is able to forget unnecessary information from the preceding state, forwarding only the relevant parts of the state to the output. For nonlinearity, the LSTM is applying activation functions, the sigmoid function being used by Bachici and Gellert 20 with output between 0 and 1. The experiments have shown that the LSTM configured optimally for the electricity prediction task has four inputs, 10 first hidden layer neurons, 5 second hidden layer neurons, a learning rate of 0.01 and 50 epochs.…”