The generation of reservoir simulation models that match field production data has been a long-time industry challenge. Two related workflows for assisted history match are presented. One workflow minimizes the misfit between simulated versus history data with a global optimizer, by adjusting reservoir and well parameters in the simulation model. An alternate workflow reduces the number of numerical simulations required by the optimization by using a comprehensive nonlinear proxy model to seed the optimizer solution set. The nonlinear proxy neural network is trained with a small set of numerical simulations from experimental design. The neural network quickly characterizes production sensitivities to reservoir parameters and generates valid solution sets. The neural network is an excellent proxy for the numerical simulator. An example from a water injection project illustrates the approach with matches for individual well fluid rates.