Future dynamics prediction from short-term time series by anticipated learning machine

Predicting short-term time series has significant practical applications over different disciplines. A dynamics-based data-driven method, Anticipated Learning Machine (ALM) is proposed to achieve precise future-state predictions based on short-term but high-dimensional data. From nonlinear-dynamical systems theory, ALM can transform recent spatial information of high-dimensional variables into future temporal information of any target variable, thereby overcoming the small-sample problem and achieving multi-step-ahead predictions. Experiments on real-world datasets demonstrate superior performances of ALM.