Deep Learning-based DeepAR for Time Series Probabilistic Forecasting

Real-world applications bring forth innovations for time series forecasting. Here let’s consider two use cases. An electricity company needs to forecast the energy consumptions of millions of individual households in a geographical area. The predictions can enable the electricity company to plan and allocate resources at household level. The second use case comes from e-commerce businesses (like Amazon.com). They need to forecast the inventories of individual products. They even need the forecasts for new products that have no historical data. In both cases, prediction intervals are needed to quantify uncertainty.

These use cases demand modeling techniques that can forecast multiple periods. The techniques should model multiple related time series all together, i.e., the global models. The techniques should provide prediction intervals. In short, the techniques should provide multi-period, multi-series forecasts with uncertainty.

This chapter introduces the DeepAR framework proposed by Salinas et. al. (2020) [2]. DeepAR is a Deep learning-based model that uses an autoregressive (AR) recurrent network architecture and incorporate probabilistic forecasting. The authors demonstrated that the forecast accuracy of DeepAR has outperformed over many forecasting methods on a wide variety of data sets, and is…