基于IQPSO-GA优化ANFIS模型的服务器故障预警方法

Server Fault Warning Method Based on IQPSO-GA Optimization of ANFIS Model

摘要:

针对服务器底层部分业务类硬件故障对系统稳定运行的影响,提出一种改进的量子行为粒子群优化(IQPSO)与遗传算法(GA)相结合的混合元启发式优化算法对自适应神经模糊推理系统(ANFIS)参数进行训练,以获得更准确的ANFIS规则进行硬件故障预警的方法。首先,通过分析服务器业务与硬件相关参数之间的映射关系,通过采集的数据集对ANFIS模型进行训练构造预测模型；其次,考虑ANFIS在梯度计算过程中存在容易陷入局部最优值的问题,设计了一种IQPSO算法结合GA中的交叉和变异算子操作混合元启发算法全局搜索ANFIS规则参数；最后,通过一组后处理样本数据集对所提方法有效性和稳定性进行了检验。实验结果表明,该方法可有效预警服务器硬件故障,基于所提混合元启发优化算法获得的ANFIS模型具备更快的收敛速度和更高的全局搜索精度,与传统ANFIS模型相比泛化精度提高了47%以上。

Abstract:

Aiming at the impact of the hardware failure of server bottom business on the stable operation of the system, a hybrid meta-heuristic optimization algorithm combining improved quantum behavior particle swarm optimization (QPSO) and genetic algorithm (GA) was proposed to train the parameters of adaptive neural fuzzy reasoning system (ANFIS). To obtain more accurate ANFIS rules for hardware fault warning. First, by analyzing the mapping relationship between server business and hardware related parameters, the ANFIS model is trained to construct the prediction model through the collected data set. Secondly, considering the problem that ANFIS is prone to fall into the local optimal value in the gradient calculation process, an IQPSO algorithm is designed to search ANFIS rule parameters globally by combining the crossover and mutation operators in GA mixed meta-heuristic algorithm. Finally, a set of post-processing sample data sets were used to test the effectiveness and stability of the proposed method. Experimental results show that the proposed method can effectively warn server hardware failures. The ANFIS model based on the proposed hybrid element heuristic optimization algorithm has faster convergence speed and higher global search accuracy, and the generalization accuracy is more than 47% higher than the traditional ANFIS model.