Application Of Neural Network Data Classification Algorithm In IOT

The Internet of Things is a novel paradigm which is rapidly gaining ground and soon to become the current hot topic in the scenario of modern wireless telecommunications. The basic concept of Internet of Things is the variety of things and objects around us are widely exchange, communication and collaboration to each other to achieve common goals by the world’s only mode of address. The Internet of Things is a great concept, it can often be seen as an extension of the existing Internet, and so it will inherit many of the existing network resources and research results. However, due to the huge the amount of data and resources, environmental constraints, it also brings a lot of new problems. Data classification techniques studied in this article is to ease the pressure of huge data in the Internet of Things dimension to the data acquisition, data transmission and data processing.As an important topic in data mining, data classification intended to generate a classification function or model which can map the date to a class in the specified category. This article selected BP neural network algorithm for data classification, after a detailed derivation of how it works, we found that BP algorithm have network convergence slow problem and easy to fall into local minimum defects. To solve these problems, this paper uses method which combines a variable learning rate and momentum factor to improve the traditional BP algorithm. In network training experiments, we found that the improved algorithm improve the convergence speed of the network in a certain extent. Data classification results show that the BP neural network classification for the cube in the Internet of Things has a higher classification success rate.Although the algorithm has been improved, but we find that while the network accuracy down to a certain extent, the convergence rate is at a very low level. And the time used in training is too long to accept in many special applications from Internet of Things. So in this paper, we finish the training of BP network in the cloud computing platform. Cloud computing technology usually is considered as the cornerstone of the system of the internet of Things, its application directly accelerate the expansion of the Internet of Things. In order to achieve the training of BP network in Hadoop platform, this paper gives a MapReduce method by the characteristics of data classification application. Learning from the network training results in Hadoop cluster computing platform which has deployed, we can found that the BP network learning algorithm is effective to achieve the parallel decomposition by Mapreduce method. Compared with the time consumed by the whole process, the experimental results show that the BP neural network under the cloud computing platform can greatly shorten the network training time, so base stations in IOT which with limiting energy and processing power can use cloud computing technology to achieve neural network’s training. The basic principles of neural networks mimic the human brain network, trying to have intelligent learning ability and logical analysis just like human being, but the human brain has86billion neurons, which is the current computer can not simulate. However, we can learn from the human brain structure and learning principles to simulate the network as the brain thinking as much as possible. With the computer performance, some problem which considered as a bottleneck problem in the current time may be solved in future. Combination of theoretical and experimental results, we find that the BP neural network algorithm is expected to break the computational bottleneck in future Internet of Things system, and won its rapid development.