This paper describes the identification and design of optimal Proportional plus Integral (PI) control system for a high voltage unit of an electrostatic precipitator (ESP) filter of a real cement kiln. The unbiased recursive least-squares algorithm is used to identify a linear discrete-time state-space model of the system based on state and measurements that were recorded during subjecting the system to a pseudo random binary sequence superimposed on its input. The state and input variables have been measured by using a dedicated electronic interface system especially designed for this purpose. Based on the identified model an optimal PI controller is designed by minimizing a quadratic performance index using dynamic programming algorithm. The controller is implemented on the real ESP filter and practical tests are presented to show the system performance.