Prediction of Cloud Resources Demand Based on Fuzzy Deep Neural Network

In this information explosion age, processing and storing the vast data sometimes are intractable problems. To cope with the challenges, fog computing is proposed, which is good at handling the real-time tasks but the service price can be correspondingly more expensive. In order to harvest more profits and get more users, the companies that offer cloud services also carry out the reservation selling strategy, whose price is relatively cheaper. For the purpose of minimizing the cost of using cloud services, this paper proposes a fuzzy deep neural network (FDNN) based method to predict the demand of cloud resources. Besides the utilization of fuzzy logic, in the phase of network training, the back-propagation algorithm, adaptively varied learning rate and the dropout strategy are also employed. According to the network predictions, the customers are able to decide how many resources to reserve so as to minimize their expenses most. Simulation results based on the real data sets from Carnegie Mellon University show that the proposed method gives the economical predictions and outperforms the traditional deep neural network.