雨衰时间序列的混沌识别与预测

Chaos Identification and Prediction for Rain Attenuation Time Series

针对传统模型在高频段雨衰预测时存在参数计算复杂、实时性差的问题,设计了一种新的非线性动态预测模型.通过混沌识别证明了雨衰时序具备混沌的动力学特性及采用混沌预测方法的可行性.该方法以雨衰前导数据为训练样本建立非线性自适应滤波器,可忽略不同地理区域降雨分布差异性的影响.仿真结果表明,嵌入维数是影响预测精度的最主要因素,在满足嵌入维数为8、重构时延为3 s、采样间隔为1 s条件时预测相对误差可达0.05以下.同时预测的雨衰时间概率分布与ITU-R模型相比有较好的一致性,验证了所提方法具备参数配置简单、可用度高的优点.

To cope of the issue of intricate parameters and low real-time characteristic of traditional models in predicting rain attenuation,a new nonlinear dynamic model is proposed.Through the strict proof for chaotic characteristic of rain attenuation time series,chaos prediction method is deemed to be feasible.By this method,the leader rain attenuation data is obtained as training samples to establish a nonlinear adaptive filter,so as to break through the influence of rainfall distribution from different climatic regions.Simulations show that embedding dimension is the most important factor affecting the prediction accuracy and relative error increases with the increase of embedding dimension.Moreover,the prediction relative error of proposed model is less than 0.05 when embedding dimension is eight,delay is three seconds,and sampling interval is one second.The comparison with experimental results also represents that the proposed method has advantages of real-time forecast and high availability.Significantly,the prediction using the proposed model shows good agreement with ITU-R model in terms of exceedance probability.