Multi-model ensemble forecasting in high dimensional chaotic system

This paper describes a new forecasting technique based on multi-model ensemble in high-dimensional chaotic system. A chaotic model is built from the time series reconstruction in the time-delayed high-dimensional phase space. The chaotic model forecasts are made by the adaptive multi-local models constructed based on the dynamical neighbors found in this space. We utilize several different predictive local models (including MLP ANN) and ensemble their model forecasts to create a more accurate hybrid model. This proposed method was implemented and tested for building a storm surge model for the North Sea. The model results showed that the multi-model ensemble model has a significant increase on the forecasting accuracy compared to standard chaotic model or global neural network model.