论文部分内容阅读
对相联存储自学习控制系统（ＡＭＬＣＳ）提出了一种改进方案，可用于动态特性几乎完全未知且慢时变的工业过程。主要改进之点在于：１）过程预报模型及控制策略均取增量形式，以便克服阶跃扰动，并有助于减少二者所需用的相联存储系统（ＡＭＳ）的内存占用量；２）在按领先多步的子目标对当前控制增量进行优化过程中引入钝化因子，以使原属多变量优化的问题简化成单变量优化的问题；３）对ＡＭＳ提出了新的寻址机制，致使其内存占用量大为减少，同时又避免了原有杂凑寻址作法所引起的数据冲撞问题；４）在用于过程预报模型的ＡＭＳ中采用局部线性外推法进行主动学习，既使学习收敛过程得以通过预先扩大已训域而加快，又避免对控制品质产生不利影响。数字仿真结果表明了这种ＡＭＬＣＳ新方案的可行性和有效性。
An improved scheme for associative memory self-learning control system (AMLCS) is proposed for industrial processes with almost completely unknown dynamic characteristics and slow time-varying performance. The main improvements are: 1) Both the process forecasting model and the control strategy take incremental forms to overcome step disturbances and help to reduce the memory footprint of associated AMSs; 2 ) Introduce passivation factor in optimizing the current control increment in sub-goal of leading multi-step, in order to simplify the original Multivariable optimization problem to single variable optimization problem; 3) Propose new addressing to AMS Mechanism, resulting in a significant reduction in memory footprint while avoiding the data collision problems caused by the original hash-addressing approach; 4) Local linear extrapolation in AMS for process prediction models for proactive learning, both So that the process of learning convergence can be expedited by pre-expanding already-learned domain and avoid adversely affecting the quality of control. The numerical simulation results show the feasibility and effectiveness of this new AMLCS scheme.