Design And Implementation Of An Intelligent Appliances Control System Based On Activity Prediction Algorithm

As IOT technology becomes a central issue for technical staff,as an important application of IOT,smart home is developing rapidly and the development of smart home should be the top priority of the next phase.At present,the smart home system on the market can provide some environmental monitoring and remote control of home appliances,but has not yet reached a higher "smart" requirement,that is,the home appliance control system has no ability to accumulate experience and autonomous learning to behaviors of users and deduce the system’s optimal control solution.This thesis is designed to combine the development of embedded system software based on Raspberry Pi and intelligent control algorithm to make the smart home system really has its own "brain",which has practical significance.The project develops an intelligent home appliance control system based on activity prediction algorithm which realizes a remote control scheme of household appliances through the network.It can access most of the indoor environment of the Internet of Things and realize next-activity prediction according to users’ previous behaviors.The system consists of intelligent terminals and a control platform as slave computer and master computer respectively.The slave computer is composed of plenty of indoor environment sensors and a single chip microcomputer named Arduino,which is responsible for collecting various parameters such as temperature,humidity and wind speed in smart home and achieves data transmission in the way of Wi Fi cooperating with the wireless module ESP8266,in which process using the TCP protocol Socket to complete the communication.The master computer is a Java application,which would wait to receive requests from Arduino after setting up the running envirionment and store collected data in the database,realizing remote monitoring and intelligent management to the slave computer through activity-prediction algorithm.At the same time,this project introduce the word embedding method in nature language processing into the system.Using a similar idea,we first map the behavior of each user and behaviors close to it in a smart home into the corresponding vector space.Then a many-to-one LSTM network would be constructed.The prediction is obtained according to the output vector and its similar vectors by training the LSTM network.The result of experiments demonstrated that the accuracy of prediction was 82.3% after the word embedding algorithm was introduced and it is suitable for multi-user smart spaces,which is significantly improved compared to traditional prediction algorithm.This system make full use of various data in the smart home to make future decisions automatically for users based on their own habits and provide users with a convenient smart home environment experience.