A Multi-step Time Series Prediction Strategy in Deep Learning: Combination of Recursive Strategy and Multi-output by Dense Layer Strategy

In practical applications, when solving the problem of time series prediction, it is often necessary to predict the data of multiple time points in the future according to the observed data. It’s a problem called multi-step time series prediction. Now there are some solutions to handle the problem, but each solution has its own advantages and is insufficient. As a result, the goal of this paper is to combine the recursive strategy and the multi-output strategy to propose a new method for solving the multi-step time series prediction problem to gain better performance than either of them alone. In the research, the author will use House Hold Energy data from Kaggle and conduct a contrast experiment to reflect the advantages of the combined strategy. The experiment results show that the combined strategy outperforms the recursive and multi-output strategies.