自适应惩罚策略及其在交通信号优化中的应用

针对约束优化问题的求解,设计了一种处理约束条件的自适应惩罚策略,用于将具有不等式约束和等式约束的优化问题转变为仅包含决策变量上、下限约束的优化问题。该策略通过引入约束可行测度、可行度的概念来描述决策变量服从于不等式约束和等式约束的程度,并以此构造处理约束条件的自适应惩罚函数,惩罚值随着约束可行度的变化而动态自适应地改变。为了检验该惩罚策略的有效性,针对单路口交通信号优化问题进行了应用研究,并用三种不同算法进行了大量的仿真计算,结果表明所设计的自适应策略在具有高度约束条件的城市交通信号优化问题中具有良好的效果。     

预测控制约束边界效应与解决方法研究

约束预测控制（Constrained model predictive control,CMPC）中,因约束的存在,优化过程中最优控制作用可能会在可行域的边界取值,也就是说会 有一个或多个变量饱和,即约束边界效应. 而过程控制中操纵变量饱和是我们不希望出现的. 对此,首先基于稳态模型,对期望值位于可行域内时最优解必在期望值处达到给出证明;同时证明了期望值在可行域外时最优解可转化为期望值到可行域的投影. 其次,针对变量在动态及稳态过程中饱和的情况提出了改善控制性能的措施——调整目标函数;终端约束的加入,为预测控制系统稳定性提供了保障. 通过对包含约束的连续搅拌釜式反应器（Continuous stirred tank reactor,CSTR）系统进行仿真实验,验证了所提方法的正确性,并说明了对目标函数进行适当调整,可有效改善系统的控制性能.     

基于惩罚和修复混合策略的约束处理方法

约束处理是约束优化的关键问题，特别是非线性约束的处理一直缺少特别有效的解决方法，将惩罚函数法与修复策略结合使用，可以有效地避免迭代过程中大量非可行解的产生，使得约束优化问题在惩罚函数和修复算子的协同作用下收敛于全局最优，较好地解决了在遗传算法约束优化问题中单独使用惩罚和修复方法时一些难以解决的问题。基于随机方向法构造的修复算子作用效果显著，采用多个测试函数对算法进行检验，均能较好地收敛于可行域中的最优解，验证了算法的可靠性。     

一种非参数惩罚函数的优化演化算法

对约束优化问题的处理通常使用惩罚函数法,使用普通惩罚函数法的困难存在于参数的选取。该文提出一种基于演化算法的非参数罚函数算法,对违反约束条件动态地进行惩罚,由适应值的设定来平衡群体中可行解和不可行解的比例,使群体较好地向最优解逼近,使用实数编码的多父体单形杂交演化策略来实现新算法,通过对测试函数的检验,该算法具有稳健、高效、简洁易于实现等特点。     

一种基于新约束处理方法的遗传算法

针对目前的约束处理方法中存在的问题,提出一种新的约束处理方法。该方法通过可行解和不可行解混合交叉的方法对问题的解空间进行搜索,对可行种群和不可行种群分别进行选择操作。避免了惩罚策略中选取惩罚因子的困难,使得约束处理问题简单化。实例测试结果表明,该约束处理方法的有效性。     

神经网络多步预测控制及其在精馏塔中的应用研究

本文提出一种基于多神经网络并行预测模型的多变量协调预测控制,利用预测误差实时反馈校正各个神经网络预测模型的参数,对于对象的时变、未建模干扰及模型失配引起的误差均有很好的适应性。针对存在耦合的被控对象,本文在优化性能指标中采用多变量协调优化策略,对被控变量集及操作变量集优化,使被控变量达到优化值和使部分操作变量达到优化值。在单步预测控制的基础上,提出基于多RBF神经网络并行预测模型的多变量协调预测控制,提高了预测控制的鲁棒性及抗干扰能力。将此方法应用于精馏塔控制中,在保证主要产品质量合格的前提下,对操作变量进行约束,使部分操作变量达到优化值,从而减少能耗,提高经济效益。仿真结果表明,基于神经网络预测模型的多变量协调预测控制具有很好的动态特性、鲁棒性及显著的节能降耗效果。     

解约束优化问题的一种新的罚函数模型

罚函数法是进化算法中解决约束优化问题最常用的方法之一,它通过对不可行解进行惩罚使得搜索逐步进入可行域.罚函数常定义为目标函数与惩罚项之和,其缺陷一方面在于此模型的罚因子难以控制,另一方面当目标函数值与惩罚项的函数值的差值很大时,此模型不能有效地区分可行解与不可行解,从而不能有效处理约束.为了克服这些缺点,首先引入了目标满意度函数与约束满意度函数,前者是根据目标函数对解的满意度给出的一个度量,而后者是根据约束违反度对解的满意度给出的一个度量.然后将两者有机结合,定义了一种新的罚函数,给出了一种新的罚函数模型.并且设置了自适应动态罚因子,其随着当前种群质量和进化代数的改变而改变.因此它很易于控制.进一步设计了新的杂交和变异算子,在此基础上提出了解决约束优化问题的一种新的进化算法.通过对6个常用标准测试函数所作的数据仿真实验表明,提出的算法是十分有效的.     

约束进化算法及其应用研究综述

约束优化问题广泛存在于科学研究和工程实践中,其对应的约束优化进化算法也成为了进化领域的重要研究方向。约束优化进化算法的本质问题是如何有效地利用不可行解和可行解的信息,平衡目标函数和约束条件,使得算法更加高效。首先对约束优化问题进行定义;然后详细分析了目前主流的约束进化算法,同时,基于不同的约束处理机制,将这些机制分为约束和目标分离法、惩罚函数法、多目标优化法、混合法和其他算法,并对这些方法进行了详细的分析和总结;接着指出约束进化算法亟待解决的问题,并明确指出未来需要进一步研究的方向;最后对约束进化算法在工程优化、电子和通信工程、机械设计、环境资源配置、科研领域和管理分配等方面的应用进行了介绍。     

一种改进的基于约束支配的多目标进化算法

针对传统优化方法在处理带约束的多目标优化问题上的不足进行了分析，将多目标进化算法以及约束支配的概念结合起来，重新定义了种群个体间的支配关系，避免了罚函数法因惩罚系数不合适而出现优化结果为非可行解的情况。并且结合惩罚值改进了选择算子和适应值分配机制，避免出现早熟收敛。同时，采用精共策略，让精英个体参与遗传操作，加快算法收敛速度。通过算例分析可知，将多目标进化算法以及约束支配的概念应用到浮筒配置优化方案是可行的、有效的。     

动态评价粒子群优化及风电场微观选址

提出了动态评价方法处理一类约束优化问题.将目标函数值和约束违反量进行动态归一化处理,再进行加权求和,动态评价解的优化性能.不仅解决了惩罚因子确定困难的问题,而且增加了优化算法的多样性,提高了优化算法搜索全局最优解的能力.将动态评价方法引入粒子群算法,求解风电场微观选址优化问题.仿真结果表明,动态评价方法提高了风电场发电量和风能利用效率.此外,该方法可广泛应用于其他优化算法以求解约束优化问题.     

Input-output linearizing control of constrained nonlinear processes

An input-output linearization strategy for constrained nonlinear processes is proposed. The system may have constraints on both the manipulated input and the controlled output. The nonlinear control system is comprised of: (i) an input-output linearizing controller that compensates for processes nonlinearities; (ii) a constraint mapping algorithm that transforms the original input constraints into constraints on the manipulated input of the feedback linearized system; (iii) a linear model predictive controller that regulates the resulting constrained linear system; and (iv) a disturbance model that ensures offset-free setpoint tracking. As a result of these features, the approach combines the computational simplicity of input output linearization and the constraint handling capability of model predictive control. Simulation results for a continuous stirred tank reactor demonstrate the superior performance of the proposed strategy as compared to conventional input-output linearizing control and model predictive control techniques.     

Control of ball mill grinding circuit using model predictive control scheme

Ball mill grinding circuits are essentially multivariable systems with high interaction among process variables. Traditionally grinding circuits are controlled by detuned multi-loop PI controllers that minimize the effect of interaction among the control loops. Detuned controllers generally become sluggish and a close control of the circuit is not possible. Model Predictive Controllers (MPC) can handle such highly interacting multivariable systems efficiently due to its coordinated approach. Moreover, MPC schemes can handle input and output constraints more explicitly and operation of the circuits close to their optimum operating conditions is possible. Control studies on a laboratory ball mill grinding circuit are carried out by simulation with detuned multi-loop PI controllers, unconstrained and constrained model predictive controllers and their performances are compared.     

双层预测控制中基于操作变量增量的多优先级稳态目标计算

本文给出一种双层结构预测控制(MPC)中多优先级稳态目标计算(SSTC)的描述方法.在可行性阶段,被控变量(CV)和外部目标(ET)的软约束以及操作变量(MV)的硬约束被统一表述为关于MV增量的约束,将软约束(包括ET的期望上下界、CV的操作上下界、以及ET的跟踪)进行放松,保证放松以后MV增量约束集的相容性.在经济优化阶段,在MV增量约束集中寻找经济最优的MV增量值.该算法在已有文献的基础上,对ET/CV的等式/不等式约束统一处理.仿真算例证实了该算法的有效性.     

Aggregation-based fuzzy dual-mode control for nonlinear systems with mixed constraints

A new receding horizon dual-mode control method is proposed for a class of discrete-time nonlinear systems represented by Takagi–Sugeno (T–S) fuzzy models subject to mixed constraints including hard input constraint and soft state constraint. On the one hand, our receding horizon scheme is based upon an online optimisation that utilises optimised sequence plus local linear feedback. On the other hand, due to the consideration of computation burden, an amplitude decaying aggregation strategy is introduced to reduce the number of optimisation variables. The proposed controller is obtained using semi-definite programming, which can be easily solved by means of linear matrix inequalities. A numerical example is given to verify the feasibility and efficiency of the proposed method.     

Comparison of PI and MPC for control of a gas recovery unit

This study compares PI and MPC controls via a computer simulation for a gas recovery unit (GRU), which consists of three distillation columns operated in series: a de-ethanizer, a depropanizer and a debutanizer. In addition, the de-ethanizer feed is preheated by the bottoms product from the de-ethanizer, which causes additional process coupling. Rigorous models are developed for the columns including column pressure dynamics and heat transfer dynamics. The process is a highly coupled system and has interactive constraints that determine the feasible operating regions. A decentralized PI control system with override controls for the constraints was designed and implemented on the GRU simulator and was compared with an industrial MPC controller. The MPC controller was observed to outperform the decentralized control system due to its multivariable constraint control capability. Since the simulator is available to other university researchers, it can serve as a challenge problem for multivariable control and identification. Three MPC controllers with different strategies for controlling the bottom level of the first column were implemented on the GRU process. The first MPC controller does not directly control the level, the second one moves the setpoint to the PI level controller, and the third one controls the level directly by manipulating the flow. The results show that including level into the MPC controller improves composition control for cases in which the manipulated variable for the level control has a significant impact on compositions.     

An algorithmic method for the selection of multivariable process control structures

In process systems, the selection of suitable sets of manipulated and controlled variables and the design of their interconnection, known as the control structure selection problem, is an important structural optimisation problem. The operating performance of a plant depends on the control structure selected as well as the characteristics of the disturbances acting on the plant. The economic penalty associated with the variability of main process variables close to active constraints is used in this work in order to develop a quantitative measure for the ranking of alternative control structures. Based on this measure, a general methodology is presented for the generation of promising control structures where general centralised, linear time invariant, output feedback controllers are used to form the closed loop system. The special case of optimal static output feedback controllers is further investigated in this paper. Furthermore, the problem of selecting proper weights in forming quadratic integral performance indices in designing optimal multivariable controllers is addressed. The validity and usefulness of the method is demonstrated through a number of case studies.     

Model predictive pressure control of steam networks

The control scheme of industrial power plants leads typically to a complex multivariable control structure with active constraints to be taken care of. Then Model Predictive Control method (MPC) handles multivariate control problems naturally and optimal control result is calculated considering actuator limitations and constraints of process variables. MPC is applied to control the pressure stability in a multilevel steam network. The system is demonstrated in a simulator environment. MPC can also be used as a convenient tool for analyzing and designing the structure of the steam network. A power plant simulator controlled by MPC helps to decide the location and the capacity of steam levelling components needed to stabilize the operation of the process.     

Virtual control with hard constraints

Practical control problems are always subject to plant state and/or input constraints, which make designing an effective controller a challenging task. This paper introduces a novel virtual control approach to handling the presence of hard constraints in control systems by utilizing virtual mechanisms in the form of nonlinear springs and dampers. The augmented virtual mechanisms are to assist in better shaping the closed‐loop responses, especially when operating near the constrained boundary. A linear quadratic regulator based model predictive control method is utilized to develop stabilizing controllers that not only achieve desired system performance, but also meet the imposed hard constraints. The basic idea is to dramatically increase control penalty by way of tuning the spring and damper effect when the constrained state/input response is close to its hard constraint. The proposed method is applied to a balancing ball problem to demonstrate its applicability and effectiveness, and the simulation results validate the proposed concept.     

Probabilistic tubes in linear stochastic model predictive control

This paper considers constrained control of linear systems with additive and multiplicative stochastic uncertainty and linear input/state constraints. Both hard and soft constraints are considered, and bounds are imposed on the probability of soft constraint violation. Assuming the plant parameters to be finitely supported, a method of constraint handling is proposed in which a sequence of tubes, corresponding to a sequence of confidence levels on the predicted future plant state, is constructed online around nominal state trajectories. A set of linear constraints is derived by imposing bounds on the probability of constraint violation at each point on an infinite prediction horizon through constraints on one-step-ahead predictions. A guarantee of the recursive feasibility of the online optimization ensures that the closed loop system trajectories satisfy both the hard and probabilistic soft constraints. The approach is illustrated by a numerical example.