عنوان مقاله
شبکه های عصبی تناسبی-انتگرالی-مشتقی برای سیستم های تاخیر زمانی
فهرست مطالب
چکیده
مقدمه
ساختار شبکه عصبی تناسبی-انتگرالی-مشتقی
الگوریتم شبکه عصبی تناسبی-انتگرالی-مشتقی
مثال ها
نتیجه گیری
بخشی از مقاله
ساختار شبکه عصبی تناسبی-انتگرالی-مشتقی
شبکه های شبکه عصبی تناسبی-انتگرالی-مشتقی شامل یک ساختار 2-3-1 می باشند. این شبکه سه لایه دارد که لایه داخلی، لایه پنهان و لایه خروجی می باشد. لایه داخلی دو تا نورون دارد، لایه پنهان سه تا نورون و لایه خروجی یک نورون دارد. نورون در درون شبکه به ترتیب تناسبی، انتگرالی و مشتقی می باشند.
کلمات کلیدی:
PID neural networks for time-delay systems Huailin Shu a,,, Youguo Pi b a Department of Electrical and Mechanical Engineering, Guangzhou University, Guangzhou 510091, PR China b Automation Engineering R & M Center, GuangDong Academy of Sciences, Guangzhou, GuangDong 510070, PR China Abstract PID neural network (PIDNN) is a new kind of networks. It consists of three layers and its hidden layer's units are proportional (P), integral (I) and derivative (D) neurons. PIDNN's weights are adjusted by the back-propagation algorithms and it perform a perfect function in process control. In this paper, we introduce PIDNN structure and algorithm and give examples in which PIDNN is used to control time-delay systems. © 2000 Elsevier Science Ltd. All rights reserved. Keywords: PID neural network; Neurocontrol; Time-delay system 1. Introduction There are a lot of time-delay systems in industry processes but it is difficult to design the controllers for them because the time-delay property. These systems generally have larger overhead, longer adjusting time and are not stable. In classical control theory the Smith method can be used to construct controllers if the transfer function of the system has been known. But, the transfer function of a practical system is not easy to measure or to complete. As is well known conventional PID controllers have many advantages so that they are most widely used in various fields of the industry, especially in the processes of chemical industry. Although PID controllers have strong abilities they are not suitable for the control of long time-delay systems, in which the P, I, and D parameters are difficult to chose. Artificial neural networks can perform adaptive control through learning processes. But there are some problems, which should be solved in practice. The main problems are the slow learning speed, the long weight convergence time and uncertain property. PID neural network (PIDNN) is a new kind of networks. It utilizes the advantages of both PID control * Corresponding author. E-mail addresses: hlshu@guangzu.edu.cn youguopi@a.gis.sti.gd.cn (Y. Pi). (H. Shu), and neural structure. It consists of proportional (P), integral (I) and derivative (D) neurons and its weights are adjusted by the back-propagation algorithms. It can control different systems through quick learning process and has perfect performances (Shu, 1997, 1998a,b,c; Shu & Li, 1998; Shu, 1999a,b. The rest of the paper is organized as follows. Section 2 presents the structure of PIDNN. Section 3 specifies the algorithm of PIDNN. System simulation examples are introduced in Section 4, including the performance behavior comparing between PIDNN and conventional PID controllers. Finally, the conclusion is given in Section 5. 2. Structure of PIDNN PIDNN consists of a 2-3-1 structure. It has three layers, which are input-layer, hidden-layer and outputlayer. The input-layer has two neurons, the hiddenlayer has three and the output-layer has only one. The neurons in the net are proportional (P) neuron, integral (I) neuron and derivative (D) neuron, respectively. The input-layer has two P neurons, one receives system setting input and another connects system output. The hidden-layer has three different neurons, the first is P neuron, the second is I neuron and the third is D neuron. The output- layer only has one neuron which completes the control output duty. The network structure and the control system are shown in Fig. 1.
