基于改进鲸鱼算法优化LSTM的化工过程故障诊断方法

Method of chemical process fault diagnosis based on MAWOA⁃LSTM

鲸鱼优化算法是一种群体智能优化算法,文中针对基本鲸鱼优化算法收敛速度慢、收敛精度低、易陷入局部最优等问题,从几个方面对其进行了改进:通过Tent混沌对种群进行初始化来增加种群多样性;融入非洲秃鹫算法的群体最优和次优策略与探索阶段位置更新策略,以避免算法早熟以及陷入局部最优;采用一种新的非线性收敛因子替代鲸鱼优化算法原本的线性收敛因子,平衡算法的全局探索和局部开发;引入了非线性自适应增量惯性权重,更好地平衡了全局搜索能力与局部搜索能力;最终得到一种混合非洲秃鹫算法的改进鲸鱼优化算法(MAWOA)。在对4种基准测试函数进行的对比试验中显示,MAWOA具有较快的收敛速度和较高的收敛精度。将MAWOA算法应用于长短期记忆(LSTM)网络的超参数寻优中,构建MAWOA-LSTM故障诊断模型。结合田纳西伊斯曼(TE)化工数据集进行故障诊断,通过与LSTM、WOA-LSTM等模型进行准确率对比,验证了所提算法的优越性。

Whale optimization algorithm(WOA)is a swarm intelligence optimization algorithm.In allusion to the problems of slow convergence speed,low convergence accuracy,and susceptibility to local optima in the basic whale optimization algorithm,improvements have been made in several aspects:initializing the population by means of Tent chaos to increase population diversity;integrating population optimization and suboptimal strategies with the African vulture optimization algorithm(AVOA),as well as location update strategies during the exploration phase,to avoid algorithm premature convergence and falling into local optima;adopting a new nonlinear convergence factor to replace the original linear convergence factor of the WOA,balancing the global exploration and local development of the algorithm;introducing a nonlinear adaptive incremental inertia weight to better balance the global search ability and local search ability,so as to obtain an improved WOA with a mixed AVOA(MAWOA).The experimental comparison on four benchmark test functions shows that MAWOA has a faster convergence speed and higher convergence accuracy.Then,the MAWOA algorithm is applied to the hyperparameter optimization of long short-term memory(LSTM)network,and a MAWOA-LSTM fault diagnosis model is constructed.The model is applied to fault diagnosis on the Tennessee Eastman(TE)chemical data set,and the accuracy is compared with LSTM,WOA-LSTM,etc.to verify the superiority of the proposed algorithm.