The simulation derives behavior from the state of neural network, in order to produce desired behavior the state should be known first, in general case you need to search it if you don't have a reliable method to reach it from any other state.