Research And Application Of Mobile Service Forecasting Model Based On Restricted Boltzmann Machine

With the rapid development of the information age and the gradual increasing demand for mobile, which makes the wireless resources of mobile networks become tense. In order to make reasonable use of limited mobile network resources, accurate traffic forecasting is of great significance.In this paper, the method of Deep Learning based on Restricted Boltzmann Machine (RBM) is studied, and the prediction of mobile network traffic is realized. RBM can be regarded as an undirected graph double-layer network model, which is an effective unsupervised feature extraction method, which is often used to extract image feature. On the basis of RBM, Conditional Restricted Boltzmann Machine (CRBM),which takes into account the inter-image correlation in the time domain,achieves more superior performance in image recognition.In this paper, the Deep Learning model based on RBM and CRBM is studied. This paper proposes a new RBM-based Deep Belief Network(DBN) time series forecasting model. The model consists of three parts:the bottom layer consists of two Gaussian Bernoulli RBM modules, the middle layer is the Feedforward Neural Network (FNN), the top of the error correction algorithm module. This model is applied to three time series datasets: [Energy], [Dollar] and [TAIEX], and compared with the traditional FNN model and the Gaussian-Bernoulli Deep Boltzmann Machine, GDBM), the results show that the proposed DBN model has a significant improvement in the prediction accuracy. Furthermore,considering the advantage of CRBM, the CRBM-based deep DBN model is proposed, which is suitable for inputting spatial pixels, such as images,and is not suitable for time series data input in time series prediction. A novel input representation is proposed to make the time series data suitable for the application in CRBM.Based on the above research, DBN model based on RBM and DBN model based on CRBM suitable for time series data are applied to mobile traffic forecasting. 10 MR datasets for users were selected and the time series were constructed in 20-minute time granularity. The results of the prediction made 80% of the cells reach 80% of the prediction accuracy.