Due to the lack of widely stable and reliable water quality parameters on-line instrumentation, it is difficult to implement closed-loop control of water quality and optimize the operation for wastewater treatment plant. In this paper, a nonlinear dynamic soft-sensing multi-model based on PLS is proposed to solve the problem of multi-variable, non-linear and time-varying uncertainty in wastewater treatment process, through selection of such auxiliary variables easily received as water flow and quality, the dissolved oxygen and oxygen aeration. The methodology integrate dynamic ARX with Fuzzy C-means identifies operating conditions of time-varying and uncertainty in the wastewater treatment process. NNPLS is used to establish a number of non-linear model in different operating conditions and the whole non-linear system. The proposed method is applied in soft-sensing of effluent quality component concentration in wastewater treatment plant. Simulation results indicate that the method which establishes a multi-variable model of water quality indicators is more precise than traditional linear PLS model