基于FCM-ELM-BBPS的预测控制参数整定

Predictive Control Parameter Tuning Algorithm Based on FCM-ELM-BBPS

模型预测控制设计参数选择显著影响被控系统性能,目前基于专家经验的主流参数整定方法会出现控制器鲁棒性差、计算成本高等缺点.为了解决上述问题,提出一种基于模糊C均值-极限学习机-裸骨粒子群(Fuzzy C-means-Extreme Learning Machine-Bare Bones Particle Swarm,FCM-ELM-BBPS)的参数整定算法.通过模糊C均值算法(Fuzzy C-means,FCM)聚类进行数据预处理,将被控系统复杂数据根据自身特征进行聚类,以降低神经网络的训练误差,提高预测精度;针对每一类特征数据,利用极限学习机(Extreme Learning Machine,ELM)建立预测控制参数与性能指标的映射关系模型,并进一步获得参数整定规则;采用裸骨粒子群(Bare Bones Particle Swarm,BBPS)优化算法进行预测控制参数整定,通过采用高斯分布来更新粒子位置,加快目标函数的收敛速度,从而有效地减少参数寻优时间;分别进行仿真和水箱系统实验验证,证明了提出算法的有效性.实验结果表明,本文提出的算法与现有方法相比,更具有优越性,其中整定时间减少了34.84%,同时在调整时间等时域性能指标上提升了43.98%.

The design parameter selection of model predictive control significantly affects the performance of the controlled system.The current mainstream parameter tuning methods based on expert experience have the disadvan-tages of poor controller robustness and high calculation cost.To solve the above problems,this paper proposes a pa-rameter tuning algorithm based on Fuzzy C-means-Extreme Learning Machine-Bare Bones Particle Swarm(FCM-ELM-BBPS).Firstly,Fuzzy C-means(FCM)clustering is used to preprocess the data,and the complex data of the controlled system is clustered according to its own characteristics,so as to reduce the training error of the neural net-work and improve the prediction accuracy.Secondly,for each kind of characteristic data,the Extreme Learning Ma-chine(ELM)was used to establish the mapping relationship model between predictive control parameters and perfor-mance indices,and the parameter tuning rules were further obtained.Then the Bare Bones Particle Swarm(BBPS)optimization algorithm is used to tune the predictive control parameters.The Gaussian distribution is adopted to up-date the particle position,which accelerats the convergence of the objective function and effectively reducs the pa-rameter optimization time.Finally,simulation and experiment of the water tank system are carried out respectively to prove the effectiveness of the proposed algorithm.Experimental results show that,compared with existing methods,the proposed algorithm has more advantages,in which the tuning time is reduced by 34.84%,and the time domain performance indices such as the adjustment time are improved by 43.98%.