Control methodology of domestic appliances in a smart grid

Type: Master's Assignment
Student: Haining Wu
Contacts: Albert Molderink, Maurice Bosman, Bert Molenkamp, Gerard Smit
Location: CAES, University of Twente


The control methodology of domestic devices in a smart grid, developed at the University of Twente, consists of three steps. The first step is predicting consumption and production patterns on a local scope, to determine the scheduling freedom within the smart grid. The second step is to aggregate the predicted patterns to a global scope and to make a plan for energy production/consumption on a local scope, based on a global objective, e.g. reduce usage of primary fuel of power plants. The third step is to real-time control devices on a local scope; to determine when and which appliances should be turned on/off or whether energy is stored, i.e. optimize production/consumption patterns with guarding comfort of users.


Problems arise when the real world situation deviates from the prediction, which may have severe impact on the runtime behavior of the system. Implementing Model Predictive Control (MPC) in the third step can somehow work around prediction errors. But when the local scheduler cannot deal with the errors, a re-planning on a global scope will be needed. The length of the horizon of MPC, when and on which level a re-planning should be performed are important topics for improving the current algorithm.

This assignment is to investigate current methodologies in planning and control steps and find ways to improve the quality and robustness of the real-time scheduling. An improved algorithm will be implemented and tested in the simulator, which has been developed at the University of Twente. An evaluation of performance and conclusions will be presented.

Research Questions

  1. Production/consumption patterns are predicted with uncertainty, how to quantify this uncertainty, and how to deal with it in planning and controlling?
  2. How to evaluate whether it is worth to postpone switching on a device?
  3. The goals of the three parties are not the same: the production companies are energy-centric, suppliers are profit-centric and households are cost-centric. How to satisfy these three parties simultaneously?
  4. MPC does not need to be done all the time. For example, it can be just needed in "rush hours" like the evening, when it is most likely the real world situation will deviate from the prediction. What time interval(s) is/are most efficient for MPC?
Free Joomla Templates designed by Web Hosting Top