【作者】 唐海燕；

【导师】 齐维贵；

【作者基本信息】 哈尔滨工业大学 ， 电力电子与电力传动， 2010， 博士

【摘要】 随着社会的发展,高楼大厦不断兴建,电梯已经成为生产与生活中不可缺少的机电设备。如何让电梯更好的发挥其作用已成为备受关注的问题,电梯优化控制方法的研究也成为热点课题。目前,在电梯交通流的研究中,对其特性的分析较少。然而,作为电梯群控中的基础问题,对其特性的研究是很必要的。此外,理论基础扎实完善、简单有效和易于实现的群控算法是电梯群控调度的重要研究内容。良好的电梯速度控制方法是提高电梯控制水平的有效途径,而现有的速度控制方法主要停留在传统控制的基础上,智能控制方法的应用相对较少。电梯的交通分析是电梯控制的重要组成部分,本文首先对电梯的交通分析进行研究,并在此基础上对电梯群的优化控制方法进行研究;另一方面,从改善电梯的运行性能目标出发,对电梯的零速停靠控制方法进行研究。针对电梯交通的非线性,控制中的非线性和多目标性,在研究中采用智能控制理论,这将在很大程度上改善电梯的服务水平,提高电梯的运行质量。电梯交通流的预测是电梯群控的基本问题之一。针对电梯交通流时间序列的非线性和小样本的特征,利用支持向量机方法对电梯交通流时间序列进行预测。将混沌动力学系统的相空间延迟坐标重构理论引入预测中,解决时间序列预测模型输入维数确定困难的问题;应用小数据量法对电梯交通流时间序列的混沌特性进行判定,并应用庞卡莱截面法对其混沌特性进行验证;在此基础上构建电梯交通流混沌时间序列的支持向量机预测模型,并利用试验寻优的方法获得模型参数;利用实测的电梯交通流数据,分别针对进门厅客流时间序列和出门厅客流时间序列进行预测,并与RBF神经网络模型及传统的动平滑方法和指数平滑法进行比较。仿真结果表明,构建的模型能很好的跟踪电梯交通流的变化趋势,可以实现电梯交通流较好的拟合和预测。在电梯群控制研究中,针对不同的电梯交通模式采用相应的控制算法对电梯群进行优化调度,将大大改善电梯群控系统的性能。提出一种粒子群优化(Particle Swarm Optimization, PSO)模糊核聚类算法(Kernel Fuzzy C-Means, KFCM)对电梯交通模式进行识别,利用PSO算法代替KFCM算法的基于梯度下降的迭代过程,使算法具有较强的全局和局部搜索能力,并降低KFCM算法对初始值的敏感度。利用核方法将低维电梯交通流特征空间样本映射到高维特征空间,增加对交通流样本特征的优化,并使样本特征在高维特征空间线性可分,更加容易聚类。为了验证KFCM算法的有效性,采用某办公楼的电梯交通流数据作为测试样本,仿真结果表明该算法具有良好的性能指标,对电梯交通流的聚类效果更准确。电梯群控调度算法是电梯群控系统的核心,研究电梯群控策略对提高电梯的运行效率和综合性能有重要意义。在传统电梯群控制策略的基础上,基于电梯群控系统的非线性和多目标性,给出两种模糊控制策略,模糊区域多目标调度算法和模糊多规则多目标调度算法。结合目标函数对算法的实现过程进行描述,并就两种控制策略适合的交通模式进行讨论,通过仿真对各调度算法的性能进行验证及比较分析。电梯的零速停靠问题是电梯控制中的另一个重要问题。将预测控制应用于电梯的零速停靠中,利用预测的爬行距离增加到电梯速度曲线的匀速段,实现减小或消除爬行距离的目的,从而实现电梯的零速停靠。给出电梯速度曲线优化机理,分别建立电梯零速停靠的小波神经网络、RBF神经网络和BP神经网络预测模型;为提高小波神经网络的预测精度,采用遗传算法对小波网络的参数进行优化;通过仿真对三种预测模型的预测性能进行比较研究。对电梯采用零速停靠算法前后的运行性能进行仿真,验证本文方法的有效性。更多还原

【Abstract】 With the development of the society, there are more and more high buildings. Elevator has become one of the necessary mechanical and electrical equipments in the life and industry. How to make the elevators play the parts better has been concerned. The study on elevator control is one of the hot research subjects. In current, there are fewer studies on the analysis of the characteristic for elevator traffic flow. However, as the basic problem in elevator group control, the characteristic study for elevator traffic flow is necessary. In addition, the elevator group control algorithm, which is with integrated theory basis, simple and effective, and easily realized, is an important content in the elevator group control scheduling. A well elevator speed control method is an effective way to improve the elevator control level, while the present method is mainly on the basis of the traditional control, the application of the intelligent control method is relatively less.The elevator analysis is an important part of the elevator control, and which is studied firstly in the paper. On the basis of the elevator analysis, the optimal control of elevator group is studied; on the other hand, with the purpose to improve the running performance of the elevator, the zero speed parking control of elevator is studied. Due to the nonlinear of elevator traffic, the nonlinear and multi-objective of elevator control, intelligent control theory is used to improve the service level of the elevator as well as the running quality of the elevator.The prediction of elevator traffic flow is one of the basis problems in elevator group control. Combining with the nonlinear and small sample characteristic, support vector machine (SVM) is used to predict the elevator traffic flow time series. The takens’delay coordinate phase reconstruction of chaotic dynamical system is introduced, and the input dimension of the time series is ascertained. The small data set method is applied to analyze the chaotic property of elevator traffic flow time series, which is validated by Poincare section method. Then the prediction model of elevator traffic flow chaotic time series based on SVM is established, and the optimal model parameters are obtained by experiment. So the incoming and outgoing passenger flow time series are predicted respectively using the data collected in some building. Meanwhile, comparing research with RBF neural network model, traditional moving smoothing method and exponential smoothing method are given. Simulation results show that the trend of the factual traffic flow is better followed by traffic flow obtained by the proposed method. The fitting and prediction of elevator traffic flow with better effect can be realized. In the elevator group control system, when the elevator group is scheduled by suitable algorithm according to traffic mode, the performance of elevator group control system will be improved. The kernel fuzzy clustering (KFCM) algorithm based on particle swarm optimization (PSO) is proposed to realize the elevator traffic mode identification. The iterative process based on gradient descent in KFCM algorithm is replaced by PSO, which has stronger global and local search capability. Meanwhile the sensitivity to initial value of FCM is decreased. By using kernel method, the sample in the low-dimensional feature space is mapped into high-dimensional feature space. And the sample feature is optimized and can be linearly divided in high-dimensional feature space so that clustering could be performed efficiently. The elevator traffic flow data collected is regard as the test sample. The simulation results show that the algorithm proposed has better performance indices, and the clustering effect of traffic flow is more exact.Elevator group control scheduling algorithm is the core of elevator group control system, the research on which has important meaning to improve the running efficiency and holistic performance of elevator. On the basis of the traditional algorithm, based upon the uncertainty and the multi-target, two scheduling algorithms based on fuzzy control are proposed, including fuzzy area control scheduling algorithm with multi-objective and fuzzy multi-rule scheduling algorithm with multi-objective. The realization of algorithm is described combined with objective function, and the application range of the algorithm is discussed. The performances of the scheduling algorithm are validated by simulation.Zero speed parking problem of elevator is another basic problem in elevator control. The prediction control is applied in zero speed parking of elevator. The creeping-in distance predicted is added to the uniform motion stage to decrease or eliminate the distance, so the zero speed parking of elevator is realized. The optimal principle of elevator speed curve is given, and the prediction models based on wavelet neural network (WNN), RBF neural network and BP neural network are established respectively. The genetic algorithm is used to optimize the parameters to improve the prediction precision of WNN. Comparing research of the prediction performance among the three models is carried through by simulation. Meanwhile, the elevator running performance after using zero speed parking algorithm is obtained by simulation, and the effect of the proposed method is validated.更多还原