December 2, 2013 | Written by: IBM Research Editorial Staff
Share this post:
Heating Ventilation and Air Conditioning (HVAC) systems typically operate at 30 to 45 percent below their efficiency rating, wasting hundreds of combined Gigawatt hours of energy in the United States. Clothes dryers, refrigerators and dishwashers, to name a few other appliances are also known to waste significant amounts of energy due to inefficient usage and malfunctions.
Technical Staff Member,
Smarter Energy Group,
IBM Research – India
These statistics prompted scientists at IBM Research – India to seriously look at reducing energy wastage, and so developed SocketWatch, an autonomous appliance monitoring system. According to Tanuja Ganu, lead researcher for the project, “A significant amount of energy is wasted by electrical appliances when they operate inefficiently due to anomalies, incorrect usage or idling. It is important that this waste be minimized because in the coming years, appliance usage is projected to sharply increase. SocketWatchis attempting to address this problem.”
What is SocketWatch?
SocketWatch is a smart plug that is positioned between a wall socket and an attached appliance, like a refrigerator, a water heater or any industrial appliance. It learns the characteristics of the attached appliance by analyzing its active and reactive power consumption patterns. These patterns accurately represent the appliance’s normal operations. Any significant shift in these patterns would indicate inefficiencies in the appliance usage. Using machine learning algorithms embedded in the device, it monitors power consumption characteristics to spot energy leaked by malfunctioning or idling electrical appliances.
SocketWatch is inexpensive and easy to use; it neither requires enhancement to the appliances, nor to the power sockets or any communication infrastructure. Moreover, this decentralized approach avoids communication latency and costs, and preserves data privacy.
“For example, a refrigerator with a door gasket that’s dry and cracked will consume approximately 50 percent more electricity than its rated consumption. And an efficient air conditioner that is operating in a room with an open door (or leaky walls) would consume more than when it is operating in suitable conditions.
“Many of the available products in the market don’t identify external factors that impact appliance efficiency that warrant a communication network and a computing device like a smart phone for interfacing with users. This lack of knowledge results in substantial ownership costs and installation complexity,” Tanuja said.
|SocketWatch (sPlug) learns the normal behavioral model of appliance by using active & reactive power consumption data.
SocketWatch is designed to operate in a decentralized fashion, with all aspects of sensing, analytics, actuation and notification performed at the device. Ithas a learning phase and a monitoring phase. During the learning phase, SocketWatchsenses the electrical parameters (voltage, current, frequency and phase angle) to measure the power consumption patterns of the attached appliance. It then analyzes these measurements using resource-efficient machine learning algorithms to build a behavioral model of the appliance. This would include the different unique power levels of operation, durations and frequencies of those states, and transitions from one state to another.
During the monitoring phase, it compares the appliance consumption patterns against the learned model. The deviations are used to spot malfunctions and energy leakage. SocketWatch would then take appropriate corrective actions (such as turning off idling appliances) or alert the users.
“It is estimated that each of us can potentially save around 30 percent of the electricity we use while at work and at home,” Tanuja said.
While SocketWatch is ready to be licensed, IBM has not set a date on its availability. Watch this space for updates.
This work is conducted by IBM Research, India in collaboration with Universiti Brunei Darussalam, Brunei. Following researchers have contributed to this work – Tanuja Ganu, Deva P. Seetharam, Dwi Rahayu (intern at IBM Research, India), Rajesh Kunnath (Radio St
udio, India), Ashok Pon Kumar, Vijay Arya, Saiful A. Husain (Universiti Brunei Darussalam, Brunei) and Shivkumar Kalyanaraman.