针对水文序列周期识别的困难,提出首先对原序列处理,再识别周期的新思路。同时提出两种新方法:一种是模拟延长序列法,即通过建模延长原序列,再应用最大熵谱分析法(MESA)对延长序列识别周期;另一种方法是构建主频序列法,应用小波重构法重构原序列主频部分,然后应用MESA对重构序列进行周期识别。结合实例,运用多种方法对同一序列进行周期识别。分析结果表明:由于受序列长度偏短、偏态性、复杂随机成分等因素的影响,传统单一处理方法(周期图法、FFT、MESA、小波分析)周期识别效果并不理想,而使用两种新方法可以有效地减小或消除上述因素的影响,周期识别效果有明显改善。
In this thesis, a new idea for estimating the periodicities of hydrological time series is put forward. Two new methods, series simulating and prolonging method (SSAP) and main frequency series reconstructing method (MFSR), are developed. The former is to build a suitable model to prolong the length of the original series, and the latter is to reconstruct the frequency series of the original series. Maximum Entropy Spectral Analysis (MESA) is used to analyze the periods of new series. Various period-estimating methods (including traditional ones and newly developed ones) are used to identify which is better for same time series. Results show that the traditional methods (Fast Fourier Transform, MESA and Wavelet Analysis) are not as good as expected because of the influences of short length of series, complex random series and skew distribution, etc. However, using new methods (SSAP, MFSR) are not so strongly affected. It can be concluded that the new methods would greatly improve the estimation of results and, to some extent, eliminate disturbance.