The present invention provides a system and method for an online optimization for a non-linear cognitive wireless. The optimization is achieved by using qualitative model on the system which provides information in the form of monotonic influence diagram. The influence graph represents the state (control variables, goal variables and intermediate variables) and their dependency information using a graphical model incorporating monotonicities information. The present invention also incorporates learning features to build a compact history for reducing bad configurations online. The invention also provides a control engine which achieves the near optimum control of the cognitive wireless network by using the network state information along with the qualitative model.