Research And Analysis Of Perception And Control Model Of Smart Home Based On Deep Learning

In recent years, Smart home has been gained wide attention of academia and industries because of the continuous improvement of key technologies about Internet of Things. Smart home industry has become a filed with a broader prospect of development. Increasing number of smart home products bring convenience to users, but also increased the complexity of devices management. Building a control platform for smart home with the features of opening, centralization and virtualization is the important way to integrate smart devices, simplify operations and enhance home services. The core capability of smart home platform is to control the smart home devices intelligently, precisely, automatically and deal with challenges of the devices coordination, circumstance changing and diverse user needs.In order to solve the problem of smart home devices control, this research firstly collects the user data of household environment by questionnaire, online data collection website and home environment simulation platform. Then, form original data sets based on structured data model. Based on the original data sets, this paper proposes a model named DeepHome, which is a perception and control model of smart home based on deep learning algorithm. At first, DeepHome trains autoencoder network to get universal device models. Then, combines multiple device models to build a deep neural network and train this network by the method of supervised learning. At last, DeepHome gets the ability to predict the parameters of smart devices based on environmental data, and controls smart home devices automatically to provide services in home.Firstly, this paper introduces the background and significance of this research and analyzes the development prospects of the smart home industry and the problems to be solved urgently. Then, expounds the main contents and objectives of this research, and summarizes the progress and results of related research at home and abroad. After that, this paper introduces the concept and related research of the software-defined smart home platform. At the same time, it introduces the theory foundation and training method of the deep neural network used in this research. After the introduction of relevant technical background, the collection of home environment data that the research relies on is described in detail: the design and survey results of questionnaire, the design and implementation of home data online collection system and the design and implementation of smart home environment simulation platform. Next, this paper focuses on the depth of structure and training methods of DeepHome. Finally, this paper describes the performance of the control model under different environment and different parameters,such as prediction accuracy and convergence speed, and demonstrates the validity and feasibility of the model and its training method.