EC Window Energy Optimization Design Team Muhannad Al Hinai, James Goddard, Benjamin Hellier, Thomas Wurm Design Advisor Prof. Sagar Kamarthi Abstract Electrochromic smart windows are designed to generate energy savings by automatically altering their tint level to regulate the amount of sun light entering a room. The existing smart window controllers optimize energy consumption, but they push window tint levels beyond the users comfort zone. In present work constraints are added to the smart window controllers to ensure that the user s desired comfort levels are accounted for in various lighting and weather conditions. These methods are tested using a room fitted with an electrochromic window and a simulator that mimic the outside weather conditions. The main objective of this project is to study the feasibility of optimizing energy consumption of a room fitted with electrochromic windows while keeping the window control settings within the user defined acceptable limits. Level of Energy Consumption Typical Energy consumption curve for summer Typical human comfort curve Typical energy consumption curve for winter Level of Human Comfort Low Tint Smart Window Tint Level High Tint For more information, please contact sagar@coe.neu.edu
The Need for Project Improving green technology usability will increase its acceptance and implementation. Around the world societies are becoming increasingly concerned with energy efficiency, sustainability, and environmental impact. These concerns manifest themselves in the form of aggressive government funding for green initiatives. In response to this, there have been major technological advances in recent years which seek to reduce overall energy consumption. One area which has lagged behind this trend in terms of widespread adoption is smart window technology. Electrochromic windows have been shown to provide significant energy savings, however, one of the biggest resistances for this technology comes from the room occupants, who desire to, but cannot change the controller settings. To gain widespread user acceptance of electrochromic technology, the window control systems must accommodate occupant comfort preferences while maintaining energy efficiency. The Design Project Objectives and Requirements Design a control algorithm to determine the appropriate window tint level. Design Objectives The objective of this project is to develop an adaptive response-surface based algorithm which determines the appropriate levels for control variables (room temperature and light transmissivity of electrochromic window) in order to minimize energy consumption for given a set of environmental conditions and occupant comfort parameters. Design Requirements The response surface is a function of 4 input variables: outside temperature, room temperature, outside brightness, and window tint level. The inputs are tested at 3 levels each to create varying outputs, which represent energy consumption of the room. The 4 inputs at 3 levels each create 81 data points that are used to train the machine learning method. Training allows the machine learning method to learn which tint level generates the best energy efficiency for any set of input values within the user comfort zone. Certain input values must also be constrained to ranges that are acceptable to users. Outside temperature and brightness are uncontrollable by users because they are determined by real-world weather conditions. Room temperature and window tint level are
controllable and will be constrained to ensure that user comfort preferences are met. Design Concepts Considered The team considered including The team initially considered using five input variables in the other variables in the model system. Room brightness was considered in addition to the four and using different modeling previously mentioned variables. This variable was eliminated because techniques. the team determined that lighting is fairly standard within a room and would not affect energy consumption the way an HVAC system (room temperature variable) does. The elimination of room brightness reduces the number of data points from 243 to 81. The 81 data points drive the machine learning method to set the appropriate window tint level. The machine learning method was chosen for this process because it most accurately determines the relationship between the inputs and output. Originally the team considered using a linear regression model to express this relationship. It would be much easier to create and integrate a linear regression model into the system. However, a linear regression requires that covariates be linearly independent and they are related to a response variable linearly. Since the input variables of this system cannot be expected to be linearly independent and linearly related to the response variables, a more complex model is necessary. Recommended Design Concept Response surface methods will be used to describe the relationship between the environmental variables and energy consumption. The machine learning method will use the response surface data to determine the appropriate window tint level for any set of inputs. 1 Design Description In order to achieve the project objective, the team has collected energy data using a weather and office environment simulator, in which the simulated office chamber employs an electrochromic window. A mechanical engineering team designed and built the test chambers. The mechanical engineers also contributed to the data collection; a collaborative schedule was prepared for a 3 week testing period in which the team members of both groups conducted tests on the test bed. The collected data is used to create a response surface (Rep. 5.2), which is a mathematical relationship between the input and output variables. The average energy usage is determined for 4 input variables (outside temperature, room temperature, outside brightness, and
window tint level) at 3 levels each, for a total of 81 tests. For any combination of inputs there is energy consumption. The machine learning method (Rep. 5.3) is trained to calculate this energy consumption level. (2) Analytical Investigations The data analysis is completed using response surface methods and the machine learning method. The response surface represents the output of the system, energy consumption, which needs to be minimized for every combination of input variables. An Excelbased algorithm is used to find the optimal energy value for each patch of response surface by choosing a point on the surface and traveling in the direction of the greatest downward slope until the minimum energy consumption point is found. When the minimum value for each response surface has been found, it is possible to determine the appropriate window tint level. Every minimum output value has a corresponding value relating to the window tint level. (3) Experimental Investigations In order to collect the necessary data the team worked in collaboration with a mechanical engineering team that built an energy test bed to measure energy consumption data. The test bed consists of two chambers, one which simulates external weather conditions, and another which simulates an office room. In between the two chambers is an electrochromic window. The two chambers are set to predetermined values as defined by the 3 levels of the 4 input variables. This creates 81 different tests with unique energy consumption output values. When the chambers and electrochromic window are set to the appropriate settings the environment is allowed to run until an accurate measure of the average energy consumption can be recorded. (4) Key Advantages of Recommended Concept The energy test bed allows the team to collect data without conducting field studies. Testing the electrochromic window in a simulated environment allows the project team to have more control over the variables, whereas testing the window in an actual office building has many complications. Testing for various weather conditions would require much more time; in the lab different seasonal conditions are changed in minutes instead of months.
Once the data is collected the response surface and machine learning method models will provide the most accurate interpretation of the tests results. The response surface is a direct mathematical expression of the variable relationships. Using the response surface one can easily find an optimal output value. The machine learning method uses all of the provided values to train itself to determine the appropriate window tint level. The machine learning method is a good fit to this application because it can handle complex variable relationships and develop predictive techniques for previously unseen input values. Financial Issues The only cost in the project is the software used to create the machine learning method, $1,100. The machine learning method software cost $1,100, which is the only cost in designing the control algorithm for the window. However, the test bed created by the mechanical engineering group cost a total of $4,050. Recommended Improvements Extensive data collection can provide greater accuracy and detail to the machine learning method, thus improving the control algorithm. The scope of this project is simply to provide human comfort in environments using electrochromic windows, but there is much more work to be undertaken to improve electrochromic window systems. The greater goal is to optimize the energy efficieny of electrochromic windows on a large scale with autonomous, automated control systems. This project aids the greater goal by showing that it is feasible to make electrochromic window systems adaptive to varying weather conditions and changing user preferences.