Abstract
Multivariate time series forecasting is extensively studied throughout theyears with ubiquitous applications in areas such as finance, traffic,environment, etc. Still, concerns have been raised on traditional methods forincapable of modeling complex patterns or dependencies lying in real word data.To address such concerns, various deep learning models, mainly Recurrent NeuralNetwork (RNN) based methods, are proposed. Nevertheless, capturing extremelylong-term patterns while effectively incorporating information from othervariables remains a challenge for time-series forecasting. Furthermore,lack-of-explainability remains one serious drawback for deep neural networkmodels. Inspired by Memory Network proposed for solving the question-answeringtask, we propose a deep learning based model named Memory Time-series network(MTNet) for time series forecasting. MTNet consists of a large memorycomponent, three separate encoders, and an autoregressive component to trainjointly. Additionally, the attention mechanism designed enable MTNet to behighly interpretable. We can easily tell which part of the historic data isreferenced the most.