论文部分内容阅读
水质模型被广泛应用于水环境管理和决策,但却面临着计算时间和模型应用效率等多方面的问题;利用函数映射和逼近等方法来建立水质模型的输入-输出响应关系,可有效减少计算成本并显著改善模型效率.水质模型的输入-输出响应函数关系有多种形式,本文以其中的2种为例,并分别基于2个水质模型(零维总磷模型、WASP/EUTRO5)的案例,分析和验证了神经网络模型在响应关系逼近中的适用性.案例的结果表明:神经网络函数可以有效地用于水质模型输入-输出响应关系的逼近;当网络规模超出阈值大小时,神经网络函数逼近的准确度和泛化度对网络规模不敏感.在案例研究的基础上,推导和讨论了在神经网络模型函数映射过程中所可能出现的非敏感参数的欺骗效应,以及可能由此导致的过度预测或过低预测问题;并建议在神经网络函数逼近中,应只包含水质模型的敏感参数,以防止降低神经网络模型的准确度.
Water quality model is widely used in water environment management and decision-making, but it faces many problems such as calculation time and model application efficiency. Using the function mapping and approximation to establish the input-output response relationship of water quality model can effectively reduce the calculation Cost and significantly improve the model efficiency.The relationship between input and output response function of water quality model has many forms, two kinds of them are taken as examples in this paper and are respectively based on two water quality models (WASP / EUTRO5) , The applicability of the neural network model in the response relation approximation is analyzed and verified.The results of the case show that the neural network function can be effectively used to approximate the input-output response of the water quality model, and when the network size exceeds the threshold, the neural network The accuracy and generalization of function approximation are insensitive to the network size.Based on the case studies, the deception effect of non-sensitive parameters that may occur in neural network model function mapping process is deduced and discussed, Over-forecasting or over-forecasting; and suggested that in the neural network function approximation, only the sensitive parameters of the water quality model should be included , To prevent reducing the accuracy of the neural network model.