One of the main aspects that affect the performance of hybrid ad hoc networks is the discovery and selection of Internet gateways. Similar to conventional ad hoc route discovery, this procedure could be accomplished employing three different strategies: proactive, reactive and hybrid. In the proactive gateway discovery, the gateway periodically generates Modified Router Advertisement (MRA) messages. The election of the time between two consecutive MRAs (T) may greatly impact on the network performance (a low T could lead to unnecessary control packets meanwhile a high T could induce nodes keep invalid routes to external hosts). The optimum value of T depends on the network conditions as the load, the node density or the node mobility. In order to dynamically adjust the T interval, in this work we present a new adaptive scheme which is supported by a control system based on a genetic algorithm. The simulation results show that the proposed algorithm is able to decrease the network congestion and achieve lower end-to-end delay.