Application Study Of Complexity Theory In River Runoff Time Series Analysis

One of the most important tasks of water resources and hydrology is description and forecasting of river runoff course that is basis of water resources planning, disposition, regulation and control. Supported by the National Natural Science Foundation of China and the Major State Basic Research Development Program of China, make runoff series of the Yellow River main stream as an example, applied complexity theory and its methods to study complexity, long relativity, forecasting ability and forecasting methods of river runoff variety. Main characteristic and innovation results as follows:1. Based on the thought that runoff is a swarm many factors coupling, analysis and forecasting modeling method of runoff time series-'decoupling, purification, coupling' are put forward, complexity theory frame of runoff series analysis is built up, which enrich water resources and hydrology system theory and provide a new technology for runoff evolvement law.2. Aimed at problems of relationship dimension, Lyapunov index, K entropy arithmetic and so on relying on time series length, complexity measure analysis methodadapting limited data is introduced to analyze river runoff complexity. Adopted empty ratio rule to improve complexity measure arithmetic, preferably excessive coarseness is avoided. Results indicate much or little runoff variety reflect runoff order or out-of-order evolvement course, low value abnormity variety of approximate entropy measure after high value reflects runoff evolvement dynamics character before runoff peak and paddy value, which provide a new analysis method to recognize runoff series dynamics character and provide a new qualitative measure to forecast runoff peak and paddy value.3. Aimed at tendency and instability characters of runoff time series, detrended fluctuation analysis method is introduced to analyze runoff time series long relativity of the Yellow River main stream. Firstly recursion figure method is adopted to diagnose forecasting ability of measured and natural runoff time series of the Yellow River. Results indicate certain long relativity exist in measured and natural runoff time series of the Yellow River main stream, and they can be forecasted in short time, which provide a new method to study long relativity and forecasting ability of runoff time series.4. Aimed at runoff many factors coupling characters, least factor gather participating runoff evolvement is provided; phase space reconstruction technology and independent component analysis method are introduced to realize decoupling and purification of runoff time series and to develop phase space reconstruction technology range. Results indicate there is three factors affecting runoff change at least of the Yellow River upstream, midstream and downstream. Among natural runoff affecting factors, one factor variety has being obvious law and the other two variety being complex. It provides a new thought to study affecting factors of runoff time series.5. Considered runoff series length and runoff peak break characters, adopted least squares support vector machine arithmetic in small-sample machine learning theory to forecast and model. Aimed at problems of parameter optimization, training and testingspeed of support vector machine arithmetic, a least squares support vector machine runoff forecasting model based on chaos optimization peak identification is built up; a least squares support vector machine arithmetic based on chaos optimization peak identification is put forward to separate sample adopting phase space reconstruction technology and independent component analysis, which provide technology guarantee for arithmetic learning and forecasting ability. Simulation results indicate this arithmetic not only has strong learning and forecasting ability but also has high training and testing speed, which provide a new tool for runoff time series forecasting.