An adaptive PI controller for room temperature control with level-crossing sampling Burkhard Hensel, Volodymyr Vasyutynskyy, Joern Ploennigs, Klaus Kabitzsch Dresden University of Technology Chair for Technical Information Systems, 01062 Dresden, Germany Email: {burkhard.hensel, volodymyr.vasyutynskyy, joern.ploennigs, klaus.kabitzsch}@tu-dresden.de Abstract—Event-based sampling allows saving energy in the sensor transmitter by avoiding unnecessary messages. One im- portant application is room temperature control with wireless sensors. Optimizing the controller parameters of a PI controller for this application is a difficult task, because usually no process model is available and challenging issues like actuator saturation have to be taken into account. Adaptive controllers offer the possibility to tune themselves automatically. In this paper, an adaptive PI controller based on pattern recognition is proposed, designed for room temperature control, sensor energy efficiency, and level-crossing sampling. The implementation is much easier than that of most other adaptive controllers and robustness to disturbances and noise is high. The focus of this paper lies rather on the basic idea, simulations and practical issues than on theoretical investigations. I. I NTRODUCTION Room temperature control loops using wireless sensor net- works allow high quality control with lower costs than using wired sensors, especially if the building automation system is not installed in the construction phase of the building. Reduction of the energy consumption of the nodes is one of the most investigated research issues in that field. Much energy can be saved by reducing the message count because sending messages requires much more energy than computing [1]. It has been shown that level-crossing sampling (also called send- on-delta, deadband, or Lebesgue sampling) allows a reduction of messages compared to periodic sampling while assuring the same control quality [2]. Therefore, level-crossing sampling can be used in common commercial building automation system technologies, e. g. LonWorks and EnOcean. The main idea of level-crossing sampling is that a new message from the sensor to the controller is only sent if the controlled (measured) signal has changed from the last sent value at least by a threshold Δ lc : |y m (t n ) - y m (t n-1 )|≥ Δ lc (1) where y m (t) is the measured signal, and t n and t n-1 are two subsequent time instances at which a message is sent. Usually the sensor wakes up periodically, measures then the current value of the controlled variable (here: room temperature) and decides according to (1) whether a message has to be sent [2]. Adaptive control allows near-optimal control without man- ual process identification because the controller sets its pa- rameters itself based on available information from past con- trol actions. Additionally, adaptive controllers change their parameters automatically if the process changes. This allows to install an untuned controller in each room of a building, and after start-up each controller optimizes itself according to the room it has to control. This allows cost reductions compared to manual tuning and higher control performance than using always-working, safe, but conservative settings. Possible reasons for differences between the rooms are the varying room size, wall material, window area, leaking doors or windows, heating and cooling equipment, duct architecture, sensor/actuator location and sensor inertia. Reasons for process changes are variations of the flow temperature from the central heat generator, larger changes of furniture (energy storages), refurbishments (e. g. new windows, new fac ¸ade insulation), sensor/actuator replacement, variations of the air flow, and changes in the HVAC (heating, ventilation and air condition- ing) system. The main contribution of this paper is an adaptive control algorithm for usage with level-crossing sampling and special emphasis on typical problems of room temperature control, i. e. actuator saturation, typical disturbances, and processes of unknown order. The goal of the adaptation is good control performance in combination with energy efficiency of the sensor. The focus of this paper is the basic idea, practical problems and simulation results; a more theoretical investiga- tion is currently done by the authors. This article is structured as follows. Section II gives an overview on possibilities for adaptive control based on nonuniformly sampled signals. Section III defines precisely the objective of the adaptive controller. Improvements over an older tuning rule are presented in section IV. The new adaptive controller is explained in section V. The approach is verified using simulations in section VI. While the algorithm is based on reference step responses, in section VII disturbance compensation is briefly discussed. Finally, section VIII draws the conclusions. II. OVERVIEW:POSSIBILITIES FOR ADAPTIVE CONTROL WITH LEVEL- CROSSING SAMPLING To the authors’ knowledge, there are only few works to- wards adaptive control with nonuniform sampling. Pawlowski et al. used a gain-scheduling controller based on outside temperature and outside wind speed together with several event-based sampling schemes for controlling the temperature in a greenhouse [3]. Dormido et al. published an autotuner 197 UKACC International Conference on Control 2012 Cardiff, UK, 3-5 September 2012 978-1-4673-1560-9/12/$31.00 ©2012 IEEE