Stable Forecasting of Environmental Time Series via Long Short Term Memory Recurrent Neural Network

In a recent decade, deep neural networks have been applied for many research areas after achieving dramatic improvements of accuracy in solving complex problems in vision and computational linguistics area. However, some problems, such as environmental modeling, are still limited to benefit from the deep networks because of its difficulty in collecting sufficient data of learning process. In this paper, aside from the accuracy issue, we raise another property - stability - of the deep networks useful for even such data-limited problems, especially in time-series modeling. Recurrent neural networks with memory cell structures, a deep network, can be deemed as a more robust network structure for long-term forecasting under coarse data observation and associated uncertainties, including missing values and sampling/measurement errors. The stability in forecasting is induced from balancing impact of inputs over all time steps in the networks. To analyze this property in various problem conditions, we adapt the recurrent networks with memory structure to environmental time-series problems, such as forecasting water pollution, air pollution, and ozone alarm. In the results, the recurrent networks with memory showed better performance of forecasting in non-stationary environment and long-term time lags.