Building a forecasting model for time-series data is a tough but very valuable research topic in recent years. High variation of time-series features must be considered appropriately for an accurate prediction. For weather forecasting, which is continuous, dynamic and chaotic, it's difficult to extract the most important information present in the knowledge base and determine the importance of each feature. In this paper, taking tropical cyclone (TC) as an example, we present an integrated similarity retrieval model to forecast the intensity of a tropical cyclone using neural network, which is adopted to generate a set of appropriate weights for various associated features of a tropical cyclone. A time adjustment function is used for time-series consideration. The experimental results show that this integrated approach can achieve a better performance.
展开▼
