耦合动态方程的神经网络模型在水质预测中的应用

Application of Neural Network Model Coupled with Dynamic Equation in Water Quality Prediction

水质变化趋势的有效预测对于水资源综合管理具有重要意义。针对现有数据驱动模型不能有效反映研究对象物理机理的问题,提出了一种耦合动态方程的神经网络模型,并给出了动态方程的耦合方法。分别从数值算例和实际案例2个方面对传统网络模型和机理先验前馈网络模型进行了对比计算分析,拟合程度指标和计算误差指标都表明机理性先验知识的加入可以提高网络模型的预测精度和非线性拟合能力。同时,该模型在水质预测中具有适用性和合理性。在样本数量一定的情况下,机理性先验知识的耦合是进一步提高网络计算精度的有效途径。

Precise prediction of water quality trend is of vital importance for water resources management. Commonly used data-driving models cannot reflect the physical characteristics of research objective. In view of this, a neural network coupled with dynamic equation is proposed in this paper, and the method to couple dynamic equation into model iteration is also given. A numerical case and a practical case are used to demonstrate the difference between network model with mechanism priori-knowledge and traditional network model. The results of fitting degree and cal- culation error indicate that the coupled priori-knowledge is able to improve calculation accuracy and enhance non- linear fitting. The proposed model is applicable and rational in water quality prediction. Sample size is the basis of neural network model application, and coupling mechanism priori knowledge under the circumstance of fixed sample size is an efficient approach to improving prediction accuracy.