|dc.description.abstract||In the past, when predicting the electricity consumption of households, there are two categories: (1) using various types of sensors (such as motion sensors, temperature and humidity sensors, power sensors, etc.) to predict residents′ behaviors first, and then the electrical appliances used to estimate the final power consumption; (2) directly estimate the future electricity consumption from the data collected by the sensors, but both types of prediction accuracy are not satisfactory. In recent years, with the great leap of artificial intelligence and deep learning technology, for example, Google′s DeepMind organization began to develop the AlphaGo project in 2014, applying deep learning methods to let computers learn Go, and in March 2016 and played with the world champion. The game result was four wins (AlphaGo) and one loss. Therefore, this study applies the Long Short-Term Memory (LSTM) and uses the UK′s public electricity data as well as Taiwan′s limited power consumption data to analyze the power consumption records. It is assumed once households will encounter unsafe electricity usage conditions, residents can have more time to deal with the risks that may result in electrical fires.
In addition, this study analyzes the collection of smart meters used in countries such as the United Kingdom, France, Switzerland, the United States, Australia and India, and finds that the definitions and units of the fields used by countries′ electricity data are different. In the future, they can serve as the references for Taiwan′s smart meters deployment. By applying the proposed system, residents can stop their behaviors immediately if an alert is activated due to the unsafe electricity usage prediction.||en_US|