Solving Complex Real Time Engineering Problems by Artificial Immune System: Case Study of Dynamic Stochastic Knapsack Problem

This paper describes an artificial immune system approach (AIS) to modeling phenomenon characterized by uncertainty, and real-time decision-making. The proposed AIS represents combination of optimization techniques and neural networks. The AIS develops antibodies (the best control strategies) for different antigens (different "scenarios"). This task is performed using some of the optimization or heuristics techniques. Then a set of antibodies is combined to create Artificial Immune System (Artificial Neural Network is constructed). We tested proposed approach in the case of Real-Time optimization of the Dynamic Stochastic Knapsack Problem. A Petri net model of the knapsack is used to generate arrival patterns for different classes of items arriving to the knapsack and calculates the rewards. Dynamic programming optimization is employed with an objective towards maximizing the reward to perform offline optimization of the knapsack. The optimal values of the input variables (number of items of each class) thus obtained are used to build and train a MultiLayer perceptron network. The Neural Network weights and configuration obtained can be used to perform optimal accept/reject decisions in real-time. The preliminary results are very promising.