The high computational cost associated with first principle based numerical reservoir models create a barrier for realizing their full potential. Currently, high fidelity proxy models, capable of mimicking the numerical reservoir models, are used to undertake some of the tasks. Nonetheless, large number of experimental simulations are required to generate modeling data. Moreover, experimental realizations are obtained by sampling static geological properties, hence, the resulting proxy model does not honor the dynamic nature of reservoirs, especially during immiscible water alternating gas injection (iWAG) process. In this study, a system identification based proxy modeling algorithm is developed for a reservoir under iWAG. A case study is used to evaluate the algorithm's performance in reproducing oil, gas and water production rates. By systematically designing the input pattern, only a single experimental run is used to generate modeling data. A blind validation and a cross plot analysis resulted in an average percentage fit of 89 and a tolerance level of up to 10, respectively. The algorithm can have potential application for optimization of injection scenario and hence for automatic well control during iWAG. © Published under licence by IOP Publishing Ltd.