Detecting Presence from a WiFi Router’s Electric Power Consumption by Machine Learning

 

ABSTRACT

Presence and occupancy detection in residential and office environments is used to predict movement of people, detect intruders and manage electric power consumption. Specifically, we are developing methods to improve demand side electrical power management by reducing electrical power waste in unoccupied spaces. In this work we conduct an extensive analysis on the applicability of using a WiFi router’s electrical power consumption in different types of environments to determinate the number or people present in a space. We show the importance of a moving average filter for electrical load time series data, confirm the correlation between control packets and increased minimal router power consumption and present our results on the accuracy of our approach. We conclude that a WiFi router’s power consumption can improve presence detection in home environments and occupancy estimation in office environments, and where possible, should be analysed separately from the aggregated power consumption.

 

 

EXISTING SYSTEM :

This means that multiple technologies will be needed to improve electric production, management and consumption if we are to transition to a sustainable and smart electric grid. With this in mind, our work focuses on reducing electrical waste on the demand side by creating a cheap and easily implementable system, which would be able to correctly   estimate a room’s occupancy and in turn adjust or turn off unnecessary plug-in devices. There are several reasons why we consider this problem important. We focus on electric waste in residential and office environments, since the residential and commercial sectors make up approximately 2/3 of national power consumption which translate to the fact that buildings take up between 40 to 60% of electric power consumption in developed countries . Second, we target electrical energy waste because it is the least objectionable aspect of demand response since it equally benefits consumers and power companies, while having no detrimental effect on the consumer like restricted power usage. Also, multiple papers report high potential power savings from reducing electrical power waste. Lastly, we focus on presence detection to achieve our goal, since it has been shown that electrical power consumption in both regular and green buildings is not highly correlated with either temperature, or humidity, or occupancy . This means that often an unoccupied building will consume the same amount of electrical energy as an occupied building, despite the energy difference of lighting, heating and cooling. This is due to the often rigid nature of Heat, Air Ventilation, and Cooling (HAVC) policies which are usually manually set based on time or the administrator’s intuition.  Because of this, multiple studies into human presence monitoring for electric power consumption and control have been suggested in recent years. These fall into several groups. First is the use of camera monitoring systems such as the one in  where the authors achieve 80% occupancy estimation accuracy and report 14% energy savings, which are not significantly affected by the 20% error rate. While this approach can achieve high accuracy rates, it is very dependent on positioning. In our own experiments, we used cameras to determine the ground truth value of the number of people occupying a room. The issues with this approach includes: problems in correct identification due to blending with the background, blind spots in the cameras’ field of view and low picture resolution and sample frequency due to storage limitations. While some of these problems might be addressed with better and more expensive hardware (fisheye lens cameras, on sight processing) the price is significantly increased. Also, we must address the problem of privacy; since in each one of our experiments the participants expressed vocal disagreement to having a camera system monitoring them. The second group of approaches focuses on different mixes of small embedded sensors like passive infrared sensor (PIR) and door sensors to monitor presence in real time . The authors in  use this approach to infer occupancy with 88% accuracy, pointing out that for only 25 dollars’ worth of sensors it is possible to reduce the electrical energy consumption by 28% in HVAC. This methods lacks the cost and privacy shortcomings of using a camera, but it does not address installation complexity and PIR sensitivity issues. Similar to cameras, we have also tried using PIR sensors, positioned at strategic entrances and doors, to monitor the number of people entering or leaving a house or office. We repeatedly found that sensors would mispredict  due to people loitering next to them, leaving doors open and the fact that animals have no problems triggering the PIR sensors. While we are fully confident in the accuracy of previously presented research, we consider that the potential to implement these approaches is highly dependent on the layout of the rooms, line of sight and the positioning of the sensors.  Lastly, there are the electric energy monitoring approaches which monitor the aggregated electric power consumption of a building the determine occupancy.  the electric power consumption from smart meters to estimate the points in time at which someone is at home. Electric load monitoring usually falls into two categories; non-intrusive load monitoring (NILM) in the form of smart meters which measure the aggregate power consumption of an apartment or building and intrusive load monitoring (ILM), usually in the form of smart plugs, measuring the power consumption of single device.

 

 

 

 

PROPOSED SYSTEM :

In this paper we look at ways to expand the benefits of an ILM approach (load controllability, low price, easy instability and privacy) with a high accuracy presence detection functionality. In our previous work . we showed initial results that it was possible to estimate with high accuracy the number of people in an office environment by looking at the electric power consumption of several routers. This is different from other approaches which look at the WiFi signal’s interactions with the surrounding area. In  the channel state information between 2 WiFi routers to determine occupancy, while the authors in off the shelf components and achieve similar results by using the Received Signal Strength (RSS) of the WiFi router. One can also probe the traffic or connect directly to the WiFi router itself to determine the number of number of active devices if privacy is ignored. We now present our full results and define our contributions as following: Applicability in different environments

We have gathered and analysed 4 different data sets gathered from 2 different office environments as well as 2 different types of home environments. We show that our system can achieve high presence detection accuracy in all environments, and is suited for occupancy detection in cases with higher number of people (4 or more). We demonstrate that cross-validation between different environments is possible and discuss the limitations of our approach. Finally, we recommend that it should be used as a separate feature in ILM presence detection.

Pre-filtering, feature extraction and classification

We expand on our previous feature extraction and show more clearly which electrical properties are important in classification. We show that by looking at more electrical signal characteristics it is possible to improve classification accuracy and discuss the importance of a moving average filter when analyzing time series data.

CONCLUSION

We have shown that by using a WiFi router’s power consumption it is possible to achieve up to 92.27% classification accuracy when predicting if anyone is present in a home environment and 93.31% when predicting the exact number of people in an office environment. The method is well suited in a work environment and complimentary to other  smart plug based occupancy determining methods, especially when a rolling average filter is used which significantly improves classification accuracy. In future work we plan on implementing our presence detection algorithm onto a smart plug and look into ways to make a distributed smart plug system which would not need a central processing unit.