【作者】 彭清娥；

【导师】 曹叔尤；

【作者基本信息】 四川大学 ， 水力学及河流动力学， 2001， 博士

【摘要】 水土流失是当今世界人类生存环境面临的主要危机之一。中国是世界上水土流失最严重的国家之一，而西部又是中国水土流失最严重的地区。建设长江上游生态屏障，确保可持续发展和西部大开发战略的实施，需要对水土流失有更深刻的认识。本文结合国家自然科学基金委员会和水利部联合资助的重大项目“江河泥沙灾害形成机理及防治研究”(59890200)，运用人工神经网络及系统动力学理论对坡面产流产沙及流域产沙机理进行研究，取得了富有创新性和实际意义的结果。与国内外现有研究成果相比，本项研究在诸多方面都具有探索性和开创性。 在已有流域产沙模型研究成果的基础上，提出了运用神经网络理论和系统动力学理论研究流域产沙的必要性。 对人工神经网络BP网络算法中固定学习率η存在的弊端提出了改进方案，使BP算法中网络误差函数能够达到局部极小点，提高了算法的拟合精度。并运用改进的神经网络模型对流域产沙及坡面产流产沙进行了研究。得到了流域年均含沙量的BP网络模型、不同坡度顺耕及横耕次降雨产流产沙BP网络模型和荒坡六种不同利用方式的次降雨产流产沙BP网络模型。 首次运用系统动力学理论对流域产沙系统进行建模。采用系统动力学有关系统的、整体的观点，全面分析流域产沙系统的动态行为，建立了流域产沙系统的系统动力学模型。并进行一小流域产沙的模型仿真实验研究。 对于系统动力学模型参数的调试法存在的缺陷，提出了多种解决方案。一是 四J！D大学博士学位论文将神经网络模型与系统动力学相结合。二是将流域产沙的物理成因模型与系统动力学相结合。这不仅解决了系统动力学模型调试参数时存在盲目性的问题，而且提高了系统动力学模型的计算精度，加强了模型的科学基础。 以上研究成果为流域产沙模型研究注入了新的科学理论和方法，开辟了新的研究方向。这不仅有助于提高人类对流域产沙机理的认识，并为流域减沙治沙策略的制定提供实验平台和科学依据。丸更多还原

【Abstract】 At present water and soil lose is a crisis threatening mankind. China is one of the most senous countries in the world in soil erosion and the ~vest of China is the most part in whole country. To protect the zoology in the upper reaches of Yangtze River, develop the west of China continuously, profound knowledge about the soil erosion is needed. The mechanism of soil erosion of slopes and watershed are studied in this paper and new meaningful results are obtained. Comparing with the research at present, works presented in this paper have characteristics of exploring and initiating.Based on the current results the necessary of researching on soil erosion of slopes and watershed by neural network and system dynamics is suggested.The abuse of fixed learning rate q in BP algorithm of artificial neural network is discussed. A new method in choosing of q is found out. Local minimum point of error function can be got out by this method and it can improve the precision of the algorithm. Using the new algorithm some research on soil erosion of slopes and watershed has conducted. Three neural network models are developed, including the model of annual average sediment concentration in a watershed, the model of sediment concentration and runoff on different gradient ploughing horizontally or vertically, and the model of sediment concentration and runoff on waste slopes using in six different ways.It is the first time to propose a model on the soil erosion in watershed by system dynamics in this paper. Following the view of system, we analyzed the dynamic and physical behavior of soil erosion in watershed, and constructed the system dynamicsmodel. An emulation experiment research is carried out on a small watershed.To avoid the difficulty of deciding the parameter of the model, we found out two ways. One is using neural network model, and the other by physical mechanism soil erosion. It can not only overcome blindness of parameters choice, but also improve the precision of dynamics model.Above all, some new theories and methods in the field if the soil erosion in watershed are obtained. A new research direction is found. It gives us more knowledge of soil erosion and provides an experiment at approach to establish police for controlling of soil and water erosion.更多还原