Research On Optimization Algorithms For 5G Refined Coverage Prediction And Weak Coverage Area Identification

In order to achieve seamless high-quality coverage of 5G networks,the number of 5G base stations and energy consumption have increased exponentially compared to traditional mobile communication networks.Therefore,operators have put forward higher requirements for wireless network planning and optimization technologies to reduce network construction and operating costs.Currently,the main goal is to achieve continuous wide-area coverage of 5G networks,wireless network planning puts forward refined requirements for coverage prediction based on empirical models.Moreover,in order to meet the 5G network optimization requirements for Massive Multiple-Input Multiple-Output(mMIMO)weight optimization and coverage simulation technology.It is necessary to improve the traditional method of optimizing weak coverage problems.To solve the above problems,the thesis optimizes the traditional empirical model correction algorithm and the weak coverage area identification algorithm.The main research work is as follows:1)An improved algorithm for empirical model correction based on feature engineering and Multiple Linear Regression-Deep Neural Network(MLR-DNN)is proposed.Firstly,a feature engineering scheme based on the wireless signal propagation characteristics is proposed.It solves the problem of weak feature expression of drive test data,and improves the upper limit of the performance of the correction model.Secondly,the MLR-DNN model is proposed.It includes primary correction based on MLR model and secondary correction based on DNN model.In the primary correction part,an MLR model is constructed for the 5G wireless signal path loss prediction problem based on the 5G Urban Macrocell(UMa)model.And the least squares method is used to fit the linear relationship between the feature and the path loss.In addition,the prediction result of the MLR model is used as a high-order feature to describe the linear relationship,and a new data set is created through feature combination.In the secondary correction part,a reasonable DNN model and training strategy are designed to fit the nonlinear relationship between the characteristics and the path loss.Through comparative experiments,the standard deviations of the empirical model prediction errors corrected by the traditional MLR model and the DNN model are 8.63dB and 5.51dB,respectively,while the standard deviation of the empirical model prediction errors corrected by the improved algorithm is reduced to 4.65dB.The result meets the requirement of refined coverage forecasting that the standard deviation of prediction error of the empirical model is less than 5dB.2)An optimization algorithm for identifying weak coverage areas based on Density-Based Spatial Clustering of Applications with Noise(DBSCAN)and computational geometry is proposed.Firstly,an optimization method for identifying poor quality clusters based on DBSCAN clustering is proposed.The optimal parameter combination of the clustering algorithm is automatically selected according to the silhouette coefficient.The algorithm takes an average of 35ms,which solves the high cost and low efficiency of traditional network optimization manpower analysis problem.Secondly,an optimization algorithm for cell coverage area extraction based on drive test data is proposed.Also,it builds a grid index to optimize the storage structure.The calculation speed of the relationship between poor quality points and cell coverage is increased by 33 times.Finally,an optimization algorithm for extracting weak coverage areas based on computational geometry is proposed.It uses the ray method to accurately locate the target cell cluster for mMIMO weight optimization,according to the actual coverage area of the cell.In addition,it uses the convex hull method to accurately extract the simulation area for the weak coverage optimization.Compared with the circumscribed rectangle method,the proportion of redundant grids in the area extracted by the convex hull method is reduced by 29.1%.It saves coverage simulation computing resources and satisfies the refinement of the 5G network optimization method based on mMIMO weight optimization and coverage simulation.