Introduction Neural networks based HVAC predictive control At IFAC-WC 2014

1. Introduction Neural networks based HVAC predictive control At IFAC-WC 2014 Xiaobao Jia MESA (Mechatronics, Embedded Systems and Automation)Lab School of Engineering, University of California, Merced E: xjia3@ucmerced.edu Phone: 13925273161 Lab: CAS Eng 820 (T: 228-4398) Sep 8, 2014. Monday 4:00-6:00 PM Applied Fractional Calculus Workshop Series @ MESA Lab @ UCMerced

2. Slide-2/1024 Outline 1.Contribution 2. Experiment Setup 3. Predicted Mean Vote Index 4. Predictive Models 5. Predictive Control and Result 6. Conclusion AFC Workshop Series @ MESALAB @ UCMerced

3. 1. Contribution 1) Wireless sensor network 2) Model identified by means of a Multi-Objective Genetic Algorithm 3) a Model Based Predictive Control (MBPC) strategy is used to control the HVAC equipment 4) Experimental results show that this approach is feasible and robust, and able to obtain energy savings greater than 50% AFC Workshop Series @ MESALAB @ UCMerced

4. 2. Experiment Setup - building This building is University of Algarve, in the South of Portugal. Algarve has a temperate climate, with daily average temperatures of 23ºC and 13ºC, in August and December, respectively. AFC Workshop Series @ MESALAB @ UCMerced

5. 2. Experiment Setup – System layout AFC Workshop Series @ MESALAB @ UCMerced

6. 2. Experiment Setup – Sensor Layout Each areas has one Wireless Sensor Network (WSN) with sensors in all rooms to monitor the air temperature ( Tai ) and humidity ( Hai ), the globe temperature ( Tg ), the state of windows and doors (open/closed), and movement using a passive infra-red activity sensor. AFC Workshop Series @ MESALAB @ UCMerced

7. 3. Predict Mean Vote ASHRAE Definition: proposed the thermal sensation scale with the purpose of quantifying the thermal sensation of people. It uses an integer numerical coding to express the qualitative thermal sensation, by relating the integer range [-3,3] to the qualitative words cold, cool, slight cool, neutral, slight warm, warm, and hot A series of computation AFC Workshop Series @ MESALAB @ UCMerced

8. 3. Predict Mean Vote Assumption1: all occupants are assumed to be dressed similarly regarding the type of clothing they wear Assumption2: be performing similar activities like attending a lecture, sitting writing a research paper, or having breakfast at the cafeteria. AFC Workshop Series @ MESALAB @ UCMerced

9. 3. Predict Mean Vote By defining a context vector C and PMV, a set of input-output data pairs may be generated in order to train an RBF model to approximate the mapping. AFC Workshop Series @ MESALAB @ UCMerced

10. Slide-10/1024 4. Predictive Model - NARX Using the neural network static mapping , and the context vector, it is possible to predict the evolution of the PMV over a prediction horizon (ph) for each room AFC Workshop Series @ MESALAB @ UCMerced

11. Slide-11/1024 4. Predictive Model AFC Workshop Series @ MESALAB @ UCMerced

12. Slide-12/1024 4. Predictive Model – Model Design Cycle This data was divided into three sets: training set – X t, used to estimate the model parameters, generalization or test set - X g, used to implement an early-stopping and to assess each model in fresh data, and a validation set - X v, used to compare different designed models on fresh data. AFC Workshop Series @ MESALAB @ UCMerced

13. Slide-13/1024 4. Predictive Model-MOGA Result The two scatter plot show the performance obtained by the models in the RMSE (Xt) - RMSE (Xg ) space (left) and in the RMSE (Xg) - §(Xv ph) space (right). The black dots represent dominated solutions, the blue non-dominated solutions, and the red preferable solutions. AFC Workshop Series @ MESALAB @ UCMerced

14. Slide-14/1024 4. Predictive Model-MOGA Result The left plot shows the one-step-ahead model output (in blue), and the measured temperature (in red) over the whole data. The right plot illustrates the evolution of the RMSE of the chosen model over ph, which increases from 0.06ºC to 0.65ºC. As it can be seen, there is nearly a perfect matching. AFC Workshop Series @ MESALAB @ UCMerced

15. Slide-15/1024 5. Model Based Predictive Control and Result An approach to non-linear MBPC consists in discretising the control space into an appropriate finite set of control actions and performing a search for the optimal future control trajectory within the available set of control options. In that case Branch-and-Bound (BB) has been proposed and applied in practice to discrete non-linear MBPC problems . In order to maintain thermal comfort and simultaneously minimize the energy spent, the problem may be formulated as follows. AFC Workshop Series @ MESALAB @ UCMerced

16. Slide-16/1024 5. Model Based Predictive Control and Result the nonlinear MBPC algorithm under continuous operation, during 48 hours, in summer conditions. the context, for the PMV model used, was C = {0.65, 1.0, 0.1}.Regarding the MBPC system parameters, ch was set to 5 samples (25 min) and ph to 48 samples (4 h). AFC Workshop Series @ MESALAB @ UCMerced

17. Slide-17/1024 5. Model Based Predictive Control and Result In the upper plot the measured and one-step-ahead predicted relative humidity are shown. the middle plot for the inside air temperature, The bottom plot shows the computed and the one-step-ahead estimated PMV, where the upper limit was set to 0.5. AFC Workshop Series @ MESALAB @ UCMerced

18. Slide-18/1024 5. Model Based Predictive Control and Result illustrates the system operation, during 11 hours, in winter conditions. The context vector used is C = {1.0,1.0,0.08} AFC Workshop Series @ MESALAB @ UCMerced

19. Slide-19/1024 5. Model Based Predictive Control and Result The one-step-ahead predicted values are shown in dots. As it can be seen, again the predictions are very accurate, and the room was kept in thermal comfort, requiring for that only a 15% heating operation during the 11 hours period. AFC Workshop Series @ MESALAB @ UCMerced

20. 6. Conclusion • A MBPC control methodology using BB method was formulated and applied to control existing HVAC systems 2) Experimental results show that this approach is feasible and robust, and able to obtain energy savings greater than 50% under normal building occupation 3) The PMV methodology needs the estimation of Tr . The approach followed was the use of a globe thermometer, which can not be used in a commercial application. AFC Workshop Series @ MESALAB @ UCMerced

21. Slide-21/1024 Thank You! AFC Workshop Series @ MESALAB @ UCMerced