Neural net controller for industrial boiler systems

Abstract

Boiler is the heart of the chemical process industries like fertilizer manufacturing plant. Automatic boiler control system is desirable for safe, economic and reliable operation of the industry. Conventional three elements Proportional plus Integral plus Derivative (PID) controller are generally used in the different control loops of the boiler. Main drawback of the use of this type of the controller is the adjustment of controller's parameters such as gain, reset time and dead time due to change of operating point of the process. However, the application of Self Tuning Controller (STC) and Model Reference Adaptive Controllers (MRAC) can overcome tllis drawback in the case of linear process plant. The boiler plant is highly nonlinear plant. Thus a neural network based integrated control system is proposed to control an industrial boiler. A 120 ton per hour capacity boiler of the Zia Fertilizer Company Limited (ZFCL), Ashuganj, Bangladesh is taken as reference boiler for the case study. The process lIlverse dynamic modelling technique is applied to design the proposed controller. A multilayer feedforward and diagonal recurrent neural network is trained to identify the unknown inverse dynamic model of the boiler plant by general backpropagation and dynamic back propagation training algorithm respectively. The training data was collected from the computer data bank of the reference boiler. After investigating the peIformance of both network, the feedforward network arcllitecture is selected for the proposed controller. Using the weights of the network a new, software controller is then developed for integrated control system of the ZFCL boiler. The developed controller is tested by using the boiler input - output data that are not used during the training. The output response of the developed controller is compared with that of the existing PID controller. Both responses are very close to each other. The developed controller output is then converted into signal using pulse width control teclmique. 'nlese signals can then be used for on-line regulation of the control valves through parallel port of the computer. The average value of tile pulse indicates the percentage of the valve opening.