Gated Recurrent Unit Network-based Cellular Trafile Prediction

With the development of 5G and big data, network traffic is growing exponentially year by year. Effective computing resource allocation and network traffic control become an increasingly important issue. If mobile network operators still use traditional traffic control strategies and methods, it will not be able to process heavy and flexible traffic and will easily lead to high packet loss and poor Quality of Service(QoS). With the improvement of computing power of hardware devices, neural networks have been widely used in many fields. A long short-term memory (LSTM) neural network is one of the famous methods for solving time series data. However, LSTM usually needs to spend a lot of training time and computing resources because of its internal structure. In mobile networks, traffic forecasting is a real-time problem. Therefore, we proposed a Gated Recurrent Unit (GRU) based model to predict the traffic of the base station. GRU replaces the forget gate and input gate in the LSTM with an update gate and combines the cell state and the hidden state to reduce the complexity of the architecture. The simulation results show that the proposed GRU-based model has better traffic prediction performance than LSTM and can greatly reduce training time.