本项目采取实例研究与理论分析相结合的方法，对地下水DNAPLs污染源反演识别研究前沿中尚待解决的问题开展研究。探索分别应用差分进化极限学习机(DE-ELM)法及集对分析(SPA)法建立多相流数值模拟模型的替代模型，并通过DNAPLs污染源反演识别的实际应用考察其适用性和优缺点。研究建立地下水污染源反演识别的0-1混合整数非线性规划优化模型，能对离散整数变量（污染源位置）和连续型变量（释放强度和初始释放时间）同步进行优化求解。将同伦算法与粒子群算法结合起来对非线性规划优化模型进行求解，探索能够搜索到全局最优且不依赖于初值选择的有效解法。研究对模拟模型中的污染源和模型参数分别单独进行优化识别的途径，探索设计合理的反馈修正迭代过程，逐步对污染源和模型参数进行修正改善，最终求得有关污染源和模型参数的近似值作为反演问题的解。丰富和拓展地下水污染源反演识别的理论基础和技术内涵。

Some forefront problems which have yet not been solved on inversion identification of DNAPLs pollution sources in groundwater will be studied by the combination of case study and theoretical analysis in this project. Differential evolution extreme learning machine (DE-ELM) method and set pair analysis (SPA) method will be adopted to formulate the surrogate models of the multi-phase flow simulation model separately, and the merits and demerits of the surrogate models will be evaluated through the practical application of inversion identification of DNAPLs pollution sources in groundwater. A 0-1 mixed integer nonlinear programming optimization model will be formulated to search optimal solutions of discrete integer variables (source locations) and continuous variables (release intensity and initial release time) synchronously. The homotopy algorithm and particle swarm algorithm will be combined to solve the nonlinear programming optimization model for exploring effective solutions, which are global optimal and not depend on initial value selection. The research will be done for identifying pollution sources and model parameters separately, and the pollution sources and the model parameters will be corrected and improved step by step by means of the feedback correction iteration process. In the end, the approximate values of pollution sources and model parameters will be taken as the solutions of inversion identification. These research results will enrich and develop the theoretical basis and technical connotation for the inversion identification of pollution sources in groundwater.