西太平洋副高形态指数的分解重构与集成预测

用小波分解和自适应神经模糊推理系统(ANFIS)相结合的方法,建立了西太平洋副热带高压形态指数月、季时间尺度的集成预报模型。由于小波分解可在信号的频域—时域内自由伸缩,准确地分解和重构带通、低通信号,因而能将复杂的副高指数时间序列分解为相对简单的周期分量信号,既简化了系统结构,又突出了信号特征。随后基于ANFIS模糊系统的非线性、容错性、自适应性和联想学习功能,建立各分量信号的独立预报模型,最后对分量预报结果进行集成。试验结果表明,该方法在保留预报对象主要特征的前提下,有效降低了预报难度,预报准确率和预报时效均较传统方法有明显的改进和提高。

WAVELET DECOMPOSITION AND COMPOSITIVE PREDICTION ON THE MODALITY INDEX OF THE WEST-PACIFIC SUBTROPICAL HIGH

Based on the method of associating wavelet decomposition with adaptive neuro-fuzzy inference system (ANFIS), a compositive prediction model on the modality index of the west-Pacific subtropical high(WPSH) on the monthly-seasonal scale was established. Signals can be freely extended/shrinked in frequency time domain and any pass-band and pass-low frequency branch can be accurately produced and reconstructed by means of wavelet decomposition. Therefore, the complex WPSH modality index time series signals can be separated into several relative simple band-pass signals, which both simplify the system structure and stand out the chief characters of signals. Subsequently, the independent prediction model of the decomposed signals were established based on the advantages of ANFIS model, such as non-linear, bearing-error, self-adapting and association-learning and the independent predicted results were integrated finally. The test results showed that under the premise of keeping the main characters of forecast body, the prediction difficulty on WPSH system had effectively been decreased, the precision and durative of the compositive prediction model were evidently improved and promoted compared with that of traditional prediction technique.