A study on neural network based system identification with application to heating, ventilating and air conditioning (hvac)system

Abstract

Recent efforts to incorporate aspects of artificial intelligence into the design and operation of automatic control systems have focused attention on techniques such as fuzzy logic, artificial neural networks, and expert systems. Although LMS algorithm has been considered to be a popular method of system identification but it has been seen in many situations that accurate system identification is not achieved by employing this technique. On the other hand, artificial neural network (ANN) has been chosen as a suitable alternative approach to nonlinear system identification due to its good function approximation capabilities i.e. ANNs are capable of generating complex mapping between input and output spaces. Thus, ANNs can be employed for modeling of complex dynamical systems with reasonable degree of accuracy. The use of computers for direct digital control highlights the recent trend toward more effective and efficient heating, ventilating, and air-conditioning (HVAC) control methodologies. The HVAC field has stressed the importance of self learning in building control systems and has encouraged further studies in the integration of optimal control and other advanced techniques into the formulation of such systems. In this thesis we describe the functional link artificial neural network (FLANN), Multi-Layer Perceptron (MLP) with Back propagation (BP) and MLP with modified BP called the emotional BP and Neuro fuzzy approaches for the HVAC System Identification. The thesis describes different architectures together with learning algorithms to build neural network based nonlinear system identification schemes such as Multi-Layer Perceptron (MLP) neural network, Functional Link Artificial Neural Network (FLANN) and ANFIS structures. In the case of MLP used as an identifier, different structures with regard to hidden layer selection and nodes in each layer have been considered. It may be noted that difficulty lies in choosing the number of hidden layers for achieving a correct topology of MLP neural identifier. To overcome this, in the FLANN identifier hidden layers are not required whereas the input is expanded by using trigonometric polynomials i.e. with cos(nπu) and sin(nπu), for n=0,1,2,…. The above ANN structures MLP, FLANN and Neuro-fuzzy (ANFIS Model) have been extensively studied.