Application of Model Predictive Control To A Geothermally Heated Bridge Deck

Abstract

The purpose of this study was to design and implement a Model Predictive Control based control system for the Oklahoma State University Geothermal Smart Bridge Project. The control algorithm presented in this thesis utilizes weather forecasts generated by the National Weather Service Rapid Update Cycle forecasting model. Weather forecasts are used to determine when snow or ice accumulation is likely to form on the bridge surface. Weather forecasts are also used in conjunction with a first principle bridge deck model to predict the future bridge surface temperature trajectory. Weather measurement data for the bridge site was provided by the Oklahoma Mesonet. A cyclic with line search optimization algorithm implemented as part of this work calculates control moves by minimizing a quadratic objective function with linear constraints. Object Oriented Programming was extensively used in control system software development. The work reported in this thesis represents a major advance in heated bridge technology. The control systems used in the previous heated bridge projects utilized On/Off feedback techniques with local weather information. The control system developed in this thesis utilizes advanced weather forecast and model predictive control techniques. The advantage of using this control system is that the heating system is engaged before the icing conditions reach the bridge and the bridge temperature is increased with optimized heat inputs. As a result, the bridge can be pre-heated to a desired temperature before the icing events, which guarantees the ice-free bridge condition. Real time results presented for four icing events during the winter of 2003-2004 demonstrated successful application of the control system. System performance proved most sensitive to the availability of weather forecast and measurement data. Extended delays of either compromised the ability of the control system to adequately preheat the bridge prior to an icing event. Improving the reliability of forecast and measurement data is the highest priority for future work. Improved optimization weighting factors were developed using data collected during one of the four icing events. Proper selection of the optimization weights over the prediction and control horizons proved critical to following the desired bridge temperature trajectory.