CNN-LSTM-model-for-energy-usage-forecasting

A CNN-LSTM encoder-decoder with univariate input is used to make multi-step predictions for time-series energy usage data

Introduction

Long Short-Term Memory or LSTM is a special type of Recurrent Neural Network (RNN) that can be used for time-series forecasting. LSTM networks are capable of learning features from input sequences of data and can be used to predict multi-step sequences. In this example, a CNN-LSTM architecture is used for multistep time-series energy usage forecasting. A one-dimensional convolutional neural network (1D CNN) is used to read and encode the input sequence. An LSTM network is then used as a decoder to make ode-step prediction for each value in the output sequence. For an excellent introduction to LSTMs, look up Deep Learning for Time Series Forecasting.

Getting Started

The CNN_LSTM_univariate_multistep_output_github file is the main file for running the LSTM model. The publicly available time-series energy usage data of IIT, Delhi is sourced from here. Infromation on the data can be found in this article.

Note: The notebook files have been tested on Google Colab.

Model

The model is a CNN-LSTM architecture encoder-decoder architecture:

Farecasting

Multi-step energy usage forecasting (12 weeks) with CNN-LSTM encoder-decoder model with a Root Mean Squared Error (RMSE) of 6.29.