세미나 및 이벤트

A Combined Proportional Plus Integral (PI) and Neural Network Controller for

세미나 날짜 2000-10-19
작성자
손문숙
작성일
2000-10-19
조회
982
1. 제 목 : A Combined Proportional Plus Integral (PI) and Neural Network Controller for


a Heating Coil





2. 연 사 : Professor Douglas C. Hittle


Associate Head for Graduate Studies


Department of Mechanical Engineering, Colorado State University





3. 일 시 : 8월 14일(월) 오후 2시-3시





4. 장 소 : 301동 1512호





5. 내 용 : A brief tutorial will be presented on neural networks followed by a discussion of this


new application. A neural network is trained to produce the steady state output of a


PI controller that is modulating a control valve on a heating coil to maintain a


constant discharge air temperature. Inputs to the network are the set point, the inlet


air temperature, the inlet water temperature and the air flow rate. Training is


accomplished on a three-layer network of sigmoid units using back propagation.


Once the network is trained, it is placed in parallel with the PI controller. In this


configuration, the neural network can produce the desired steady state control signal


instantly, essentially replacing the integrator of the PI controller. Some proportional


gain is needed for good response to set point changes. The advantage of the


combined controller is that a PI controller can be tuned conservatively to avoid risk


of instability as plant gains change. While performance of such a controller would


initially be poor, once the network is trained and made part of the controller,


performance improves and provides better response than can be achieved with a PI


controller alone (even a carefully tuned one). The neural network acts as a non-linear


feed forward component and is a well matched augmentation to PI control for non-


linear heating, ventilating and air-conditioning (HVAC) systems. Results from


simulation studies and from experiments on a full scale HVAC system are presented.





6. 문 의 : 기계항공공학부 김민수 교수 (☏ : 880-8362)