海域水质模型参数动态反演方法研究初探

海域水质模型长周期数值模拟中,模型参数全时段统一赋值的方法忽略了参数随时间动态变化的物理特性,降低了模型的可靠性,增加了海域水质模型验证工作的难度.本文建立了将数据驱动模型和水质模型有机结合的参数动态反演的新方法:以水质模型多参数设计工况的数值模拟,构建海域内部观测点污染物浓度响应解集,并将解集划分为若干时段;应用基于人工神经网络的数据驱动模型归纳建立观测点每一时段内污染物浓度同多个模型参数之间的非线性关系;将实测资料带入关系中,进行模型参数随时间变化的动态反演.以渤海海域水质模型为例,采用“孪生”实验验证参数动态反演新方法的可行性,结果表明该方法是有效的,能够保证模拟周期内较高的数值精度,提高了模型的准确性.

Dynamic inversion of parameters for sea water quality model

The uniform evaluation method for parameters in the long period’s simulation of marine water quality model ignored the parameters’ physical meaning of temporal dynamic characteristic,resulted in the reduction of model reliability,and magnified the model calibration’s difficulty.In this paper,a data-driven model based on artificial neural network was developed to inverse the model parameters’ values dynamically.The approach firstly constructed the solution space of internal observation site by the numerical simulation results of multi-parameter design cases,and the solution space was divided into some temporal segments.The nonlinear relationship between the pollutant concentration of every temporal segment and model parameters was generalized by the data driven model based on the artificial neuron network.The filed data was imported into the relationship for the dynamic inversion of model parameters.A case study in the Bohai Sea and twin experiment was presented for the validating dynamic inversion method.The verification results showed that the method was efficient for the parameter dynamic inversion,to guarantee the numerical precision in the simulated period and to improve the accuracy of model.