One of the major architectural components shared between cognitive radios and networks is a cognitive engine supporting a cognitive process. Many of the decision making will need an optimization core for determining the optimum assignment for the control variables. This needs to be on-line to take care of the dynamics of the network and user contexts with strict time budgets. Moreover the overall network system is highly non-linear and often the state relations are analytically not tractable. An efficient search algorithm using a system model incorporating aspects of system ontology is be of value. Here a signed graph based methodology is proposed to incorporate the knowledge of influence relationship between state variables to guide a heuristic optimization process. The proposed methodology is illustrated using a typical wireless link optimization problem.