Study On Medium-and-long-term Hydrologic Prediction Of Runoff

Because of the effect of many factors (such as climate variety, human movement, terrain change), the change of hydrologic phenomena is complex. The hydrologic system shows some complicated characteristics, such as stochastic, grey, chaos and so on. So the prediction of annual runoff, monthly runoff and daily runoff is difficult in hydrology study. Medium-and-long-term prediction of runoff is an important content on the field of water resource study. It plays an important role on the development of national economy. It is the basic reference of water resource planning and development, flood controlling and drought preventing, power plant operation and management. Based on runoff data of Bao Zhu-si power station and Da Du River, many models are used and analyzed in this paper. Firstly, during the prediction of annual runoff study, six models are introduced: linear and stationary auto-regression model, grey dynamic model, artificial neural network model, nearest neighbor bootstrapping regressive model, mean generating function model and wavelet network model. The feasibility of those models and their characteristics are studied with some examples. Secondly, during the prediction of monthly runoff study, six models are introduced: seasonal auto-regressive model, recession model, mixed linear regressive model, seasonal artificial neural network model, threshold auto-regressive model and wavelet network model. The feasibility and characteristics of those models are studied with some examples. Thirdly, during the expanded trend prediction of monthly runoff in a year study, three models are introduced: artificial neural network model, nearest neighbor bootstrapping regressive model, a new hybrid model with pursuit projection regressive and nearest neighbor bootstrapping regressive. The feasibility of those models is studied with some examples. Fourthly, during the prediction of daily runoff study, four models are introduced: seasonal auto-regressive model, stage stationary auto-regressive model, artificial neural network model and nearest neighbor bootstrapping regressive model. The practicability of those models is studied with some examples. Every chapter applies lots of utility models. The paper emphasizes particularly on the way of models'improve and parameter optimization. The expanded trend prediction of monthly runoff during a year is an innovation in this paper. Hybrid models are another innovation of this paper, such as the wavelet network model, and the new hybrid model with pursuit projection regressive and nearest neighbor bootstrapping regressive.