Learning Automata-Based Data Aggregation Tree Construction Framework for Cyber-Physical Systems

A high degree of energy efficiency and real-time for data transmission is required in cyber-physical systems (CPSs). Data aggregation is an efficient technique to conserve energy by reducing the amount of transmission data. To optimize real-time communication under constraints of power consumption and data aggregation performance of each node in CPS, this paper presents a learning automata (LA)-based degree-bounded bottle-neck data aggregation tree (DBBDAT) construction framework to minimize the maximum delay on data aggregation trees with bounded degree, which is an NP-hard problem. We model the network of CPS as a connected weighted and directed graph to form a network of LA. Degree-bounded data aggregation trees are constructed first by the action selection of each automaton. Then, the action vector of each automaton is updated by linear reward-inaction learning algorithm, and at last DBBDAT is constructed based on a threshold. Simulation results show that our approach significantly outperforms integer linear programming (ILP)-based method in terms of time complexity. Compared with ILP-based method, it can obtain an optimal solution or a suboptimal solution with guaranteed approximation ratios, and can control the trade-off between accuracy and cost by choosing appropriate learning rate and threshold. Its distributed implementation is simple and it can efficiently solve the problem for the sparse graph in practice.