Intelligent Large-Scale AP Control with Remarkable Energy Saving in Campus WiFi System

A Unified Methodology to Predict Wi-Fi Network Usage in Smart Buildings

“…The key concept behind smart building management systems is the preemptive control of building infrastructure in order to save resources such as lighting, Heating, Ventilating and Air Conditioning (HVAC), elevators and even network infrastructure [3], [5], [6]. Some building management systems do not require precise occupancy information to be functional and capable of saving energy, especially HVAC systems, by using fixed building control schedules [7].…”

“…Some of them collect information on the building areas occupancy history to predict if they are occupied or not (occupancy detection) [3], [5], [8], [10]. Others use Wi-Fi information to predict the occupancy count of some building areas [1], [4], [6], [9]. In this scenario, several studies use Wi-Fi infrastructure combined with machine learning methods to predict occupancy of building areas, floors and rooms [1], [3]- [5], [8], [10], [11].…”

“…They do not necessarily use the association history information from the Wi-Fi network to build their dataset and create prediction systems, but rather other information such as channel utilization or bandwidth [1], [3]- [5], [8], [10]- [13]. Those studies used single-label or multi-label machine learning classification models and artificial neural networks (ANNs) to address the occupancy detection problem using Wi-Fi association history and developed mechanisms that decide whether an AP should be turned on or off [1], [6], [9], [12], [14]- [16]. There are others that use Wi-Fi association data to create single-label machine learning regression models to estimate the occupancy count that can also be used on Wi-Fi AP energy saving mechanisms [3], [17].…”

“…According to Cui et al [19], energy consumption in a Wi-Fi network is considerable. University wireless networks display a bimodal periodic behavior with daily and weekly cycles, and Wi-Fi Access Points (APs) may stay unused for extensive periods of time [6], [20], [21]. These long idle periods represent a considerable energy waste that presents an excellent optimization opportunity.…”

“…These long idle periods represent a considerable energy waste that presents an excellent optimization opportunity. That scenario allows the use of machine learning prediction models capable of delivering occupancy demand predictions for network APs throughout the day [6]. The Wi-Fi network controller can switch off the network interface of unused APs during idle time slots based on those predictions.…”