Smartphone power management based on ConvLSTM model

Abstract

The battery capacity of smartphones is limited, and power saving is crucial. In this paper, we propose a power management approach depends on the users’ usage patterns. To save power, we need to know in which time interval the most power is consumed. The convolutional long short-term memory network (ConvLSTM) model provides learning long sequence dependencies based on time series data. The battery usage data is processed with the ConvLSTM model to predict every 30 min of the next 24-h remaining battery capacity of the smartphone. Then, the rate of power consumption is calculated for each hour interval according to predicted values of the remaining battery capacity. Depending upon the rates of power consumption, the value of numeric features (screen brightness, media volume, etc.) is reduced and non-numeric features (GPS, Wi-Fi, etc.) are turned off. Our approach saved much power at some time interval, besides saved up to 21% power saving overall. The performances of CNN, RNN, LSTM, CNN-LSTM, ConvLSTM, and ARIMA were also evaluated. We developed our approach as a smartphone application to forecast the remaining battery capacity and reduce battery consumption effectively for any smartphone user without an initial condition or any other restriction.