Communication Network Quality Model Of Geographic Grid And Data Forecasting

Mobile communications is a major application in daily life and the total number of smart phone users is getting bigger and bigger. Thus it is becoming increasingly important that mobile communication networks provide reliable performance to meet the needs of those increasingly number of users. To achieve this goal, network operators focus on developing efficient methods to monitor and maintain mobile communication networks. However, traditional ways of network optimization and maintenance lack both efficiency and accuracy since they solely rely on manual knowledge to locate problems in networks by investigating a high volume of data. The advent of data mining and data visualization technology make the intelligent network optimization Service possible. It greatly speed up the network maintenance staffs' response time and raise their working efficiency by evaluating network's quality of service(Qo S) and presenting the results in a visualization statistics way.This thesis is based on a project of China Mobile, named PM2.5 intelligent network management system. In the system, geographic grids are defined as the basic unit to indicate the situation of networks in different regions, which is a novel method to monitor and analyze the performance of cellular networks. In addition, through intelligent predication, network operators can have better monitoring of current network operation conditions. Generally speaking, the four main contributions of our work are summarized as follows:1. A network quality model of geographic grid is proposed. The quality of each geographic grid in network is denoted by a Quality Score(QS), which is based on communication cells' Qo S and the weight of the corresponding base station, and can provide guidance for network maintenance. By using Baidu map to demonstrate each grid's QS, operators can greatly improve their working efficiency.2. A combination model based on BP neural network and transmission model(Okumura-Hata and COST-231-Hata) is proposed. This new model can accurately calculate the field strength in each measuring point and be applicable to various scenes in city. Besides, it can also be used to estimate the coverage area of each base station.3. A novel prediction model based on multi-feature and EMD-SVR is proposed. To further improve accuracy in the process of prediction, it applies SVR to the decomposing results and EMD to predict possible data in the next point, and then combine them to obtain the final result. In addition, this model considers multi-dimensional environmental factors, such as weather, holidays, news number etc, as predicted features to improve the accuracy.4. An improved grid search algorithm is introduced to figure out the optimal parameters in the SVR model. The new grid search algorithm can adaptively adjust the search range and step size to efficiently calculate the optimal parameters in SVR.