Research And Application Of Deep Learning In Mobile Sevice Forecasting

With the development of mobile communication technology and the popularization of smart phones,the mobile traffic presents explosive growth.In order to meet the demand of large-scale mobile users and business traffic with limited mobile resources and accurate prediction of mobile traffic,an effective solution can be provided for the rational allocation of mobile network resources so as to ensure the good service quality of users.Therefore,finding a reasonable and effective method for forecasting mobile traffic is of great significance.As a kind of time series problem,the mobile service forecasting problem has the characteristics of diversity and time-varying,and the traditional statistical methods cannot meet the requirements.The vigorous development of deep learning in all fields puts forward a new idea for the prediction of mobile services.With its depth and multi-level feature extraction,the implied relevance between data can be tapped.Based on this,this paper studies how to use the idea of deep learning to solve the problem of time series prediction,and proposes two models and applies them to the scenario of mobile service prediction.In this paper,we propose two time-series models based on two typical neural networks of deep learning:Recurrent Neural Network(RNN)and Deep Belief Nets(DBN).Firstly,based on the LSTM(Long-Short-Term Memory),a common variant of RNN,the one-step time series prediction model and the multi-step time series prediction model are respectively proposed.The input stability can be improved by differencing the data,finally,based on the standard dataset,the proposed LSTM-based temporal prediction model can improve the prediction accuracy compared with the traditional persistence model.Secondly,a time series prediction model based on DCRBM with differentiated parameters is proposed based on Conditional Restricted Boltzmann Machine(CRBM).Based on the proposed objective function of DCRBM model,the update formula of each parameter of the model is derived and implemented.Finally,testing on standard dataset,the proposed DCRBM-based prediction model has obvious prediction accuracy improvement compared with the CRBM model.Finally,two kinds of time series forecasting models based on deep learning proposed in this paper are applied to the prediction of mobile traffic.The measurement reports reported by 10 cell's users in the current network are selected as data sources,the virtual grids are divided,time series are constructed according to a time granularity,and two models are respectively used for cell level prediction and virtual grid level prediction.The experimental results show that both models can achieve more than 80%forecasting accuracy,and the DCRBM-based time series prediction model has higher forecasting accuracy and stronger stability than the LSTM-based time series prediction model.