基于SSA-BP与SSA的地下水污染源反演识别

应用基于SSA-BP神经网络替代模型的模拟-优化方法和SSA研究了地下水污染源位置及释放历史的反演识别问题。并在建立地下水水流模型时,应用Cholesky分解方法建立含水层渗透系数连续场,该方法相比于普通的参数分区方法更好地描述了水文地质参数的非均质性。结果表明:SSA-BP神经网络替代模型对模拟模型具有较高的逼近精度,其平均相对误差仅有3.21%。应用SSA求解优化模型,能够快速准确地识别出点污染源的位置及释放历史。SSA对污染源位置的反演识别相对误差在10%左右,对污染源源强的反演识别相对误差不超过4%。因此,本文所提出的方法是一种有效的地下水污染源识别方法,可为污染责任认定及污染修复方案的优化提供参考。

Inversion and identification of groundwater pollution sources based on SSA-BP and SSA

The simulation-optimization method based on SSA-BP neural network alternative model and SSA were applied to study the inverse identification of groundwater pollution source location and release history. And the Cholesky decomposition method was applied to establish the continuous field of aquifer permeability coefficients in the groundwater flow model, which better describes the non-homogeneity of hydrogeological parameters compared with the common parameter partitioning method. The results showed that the SSA-BP neural network alternative model has a high approximation accuracy for the simulation model, and its average relative error is only 3.21%. The relative error of SSA in the inverse identification of source location is about 10%, and the relative error of SSA in the inverse identification of source intensity does not exceed 4%. Therefore, the proposed method is an effective groundwater pollution source identification method, which can provide reference for pollution responsibility identification and pollution remediation plan optimization.