Error-feedback stochastic modeling strategy for time series forecasting with convolutional neural networks

Despite the superiority of convolutional neural networks demonstrated in time series modeling and forecasting, it has not been fully explored on design network architecture tuning hyper-parameters. Inspired by incremental construction strategy for building a random multilayer perceptron, we propose novel Error-feedback Stochastic Modeling (ESM) to construct Convolutional Neural Network (ESM-CNN) forecasting task, which builds adaptively. The ESM suggests that filters neurons error-feedback connected layer are incrementally added steadily compensate prediction error during process, then filter selection is introduced enable ESM-CNN extract different size temporal features, providing helpful information at each iterative process prediction. performance justified its accuracy one-step-ahead multi-step-ahead tasks respectively. Comprehensive experiments both synthetic real-world datasets show proposed only outperforms state-of-art networks, but also exhibits stronger predictive power less computing overhead comparison trained deep models.