Neural Network Algorithm for the Knapsack Allocation Problem in Unmanned Systems

The knapsack allocation problem in unmanned systems has been extensively studied in operations research and management science literature, and finding efficient solutions has consistently posed a challenge in academia. This paper introduces a deterministic annealing neural network (DANN) algorithm to tackle this issue. The algorithm is composed of two main components: a globally convergent iterative process and a convergence path formed by the minima of barrier functions. This paper demonstrates that as the barrier factor decreases from a sufficiently large real number to zero, an excellent near-optimal resolution to the challenge can be achieved along a series of convergence paths formed by the minima of barrier functions. Simulation results highlight the superiority of the proposed algorithm over existing methods.