Efficient Abnormal Building Consumption Detection by Deep Learning LSTM IOT Data Classification

This research deals with the method of efficient detection of abnormal building energy consumption and energy losses for further intelligent building energy management and respect of energy transition obligations. Automatic energy loss detection has been proposed using Deep Learning (DL) Long Short Term Memory (LSTM) networks specially dedicated for building data provided by Internet of Things (IoT) objects. The article presents the proposed neural network, a description of the data classification, as well as features of processing the input and output data of the classification. A practical evaluation of the classification method was made on a real IoT building sensors dataset provided by the multifunctional administrative building in the Paris area, France. To train the neural network for accurate classification, the process of initial manual input data processing was presented. A two-month dataset was used for training, and a two-week dataset was used to test the efficiency of the obtained neural network. The subsequent in-depth data analysis showed the high efficiency and significant accuracy of the obtained method in the classification and research of abnormal building energy consumption.