摘要
针对一类连续搅拌反应釜(CSTR)存在强非线性和时变性,难以建立准确数学模型的问题,提出一种新的基于多策略改进麻雀算法优化深度极限学习机(OtSSA-DELM)的Hammerstein-Wiener模型的辨识建模的方法。针对麻雀算法像其他群智能算法一样后期寻优精度低、易陷入局部最优等缺点提出三点改进措施,首先利用正交阵列对麻雀种群初始化,再使用鱼鹰优化算法在第一阶段的全局勘探策略替换原始麻雀算法的探索者位置更新公式,最后采用t-分布变异策略替换原始麻雀算法的跟随者位置更新公式,并使用测试函数验证其改进的性能。使用改进的麻雀算法对DELM网络训练过程单层网络的输入权重和偏置因子进行寻优,解决DELM易陷入局部最优的缺点。最后利用该混合优化算法对Hammerstein-Wiener模型进行辨识实验,实验表明利用该混合优化算法相比于其他群智能算法优化DELM对Hammerstein-Wiener模型具有较高的辨识精度。
Aiming at the problem that it is difficult to establish an accurate mathematical model for a class of continuous stirred reactor(CSTR)due to its strong nonlinear and time variability,a new method for identification and modeling of Hammerstein-Wiener model based on multi-strategy improved Sparrow algorithm optimized Deep Extreme Learning Machine(OtSSA-DELM)is proposed.In view of the shortcomings of Sparrow algorithm like other swarm intelligence algorithms,such as low optimization accuracy in late optimization and easy to fall into local optimality,three improvement measures are proposed.Firstly,orthogonal array is used to initialize sparrow population,and then the Osprey optimization algorithm is used to replace the explorer position update formula of the original sparrow algorithm with the global exploration strategy in the first stage.Finally,the follower position update formula of the original Sparrow algorithm is replaced by T-mutation strategy,and its improved performance is verified by test function.The improved Sparrow algorithm is used to optimize the input weight and bias factor of the single-layer network in the training process of DELM network,which can solve the problem of DELM falling into local optimal.Finally,the identification experiment of Hammerstein-Wiener model is carried out by using this hybrid optimization algorithm.The experiment shows that the optimization of DELM by this hybrid optimization algorithm has higher identification accuracy than that of other swarm intelligent algorithms.
作者
盛斌
张军
Sheng Bin;Zhang Jun(Shanghai University of Electric Power,Shanghai 200090,China)
出处
《化工设备与管道》
CAS
北大核心
2024年第6期7-16,共10页
Process Equipment & Piping
基金
国家自然科学基金(61273190)。
