关键词: 人工神经网络, 反向传播模型, 径流趋势预测

Abstract: Artificial neural network (ANN), as a computing model possessing high-nonlinear mapping ability, has been widely applied in lots of field, for example mode recognizing, automatal control and so on. In the numerical values forecast, this computing method can be used for predicting by means of studying sample data and it need not predefining sample's model. The forecasting for runoff variation trend is one of the important matters in water resource research field, and is at the holding position in long-term programming about water resource utilization. At present, the statistical models combining with experience indexes are mainly used for establishing predicting equation in long-term forecasting for runoff. These models basically belong to linear models, whose forecasting effect is not very satisfactory for forecasting largely fluctuating runoff change process. Back-Propagation model in ANN, B-P model for short, composed by nonlinear transform cell possess better nonlinear mapping ability, its structure is simple and its capability is favorable. So B-P neural network net is used for predicting runoff variation trends. The Longyangxia Key Water Control System is located on the upper Yellow River in the northeastern Qinghai-Tibet Plateau, 1 688 km down from the source of the Yellow River. As the first of the stairstep power station along the Longyangxia Gorge to Qingtongxia Gorge river section, its reservoir can hold 24.7 billion m³ of water has been playing an very important role in providing power, storing flood and resisting ice running and irrigating, etc. in the northwestern China. The upper Yellow River basin above Longyangxia Gorge is located in the northeastern Qinghai-Tibet Plateau, between 95°50'~102°52'E, 32°20'~36°30'N, with a water collection area of 13.14×10⁴ km². The Tangnag Hydrometric Station, upstream about 110 km, is the monitoring station of runoff into the Longyangxia Reservoir. Runoff has been observing since 1956 and there has more than forty years data now. The inflow to the reservoir mainly generates in the upper Yellow River basin above Tangnag, so studying the variation characteristics and forecasting the future trend on runoff at Tangnag can provide an important base for controlling the inflow to the reservoir. Therefore, the inflow into the Longyangxia Reservoir is chosen as the research object and used for forecasting the change trend on inflow in this paper, and the result is compared with Time-Sequence Analysis Method in common use to analyze and examine the application prospect and advantage of ANN method in hydrological forecast field.

Key words: artificial neural network, "Back-Propagation" model, runoff trend forecasting