| With the continuous development and deployment of mobile communication networks,network planning has put forward higher requirements on the prediction of wave propagation characteristics,and there is an urgent need to establish more accurate and efficient wave propagation prediction models and algorithms.Traditional propagation models are usually calibrated based on statistical data and actual measurement data,but with the increasingly massive amount of data and the rich variety of influencing factors,traditional models have great limitations in handling these complex data.Based on this,the thesis combines traditional methods and machine learning approaches to predict the propagation characteristics of radio waves in urban cell environments,and constructs an accurate and efficient prediction model accordingly.The main research elements are as follows.(1)The empirical and deterministic models of radio wave propagation are simulated and compared,respectively.The results show that in more complex scenarios such as urban neighborhoods,deterministic wave propagation models such as ray tracing method have higher accuracy,but at the same time the higher the accuracy the longer the time will be.(2)Back Propagation Neural Network(BPNN)based urban area radio wave propagation model construction is investigated.The training data set is composed of path loss values obtained from ray tracing method simulation.The results show that BPNN is suitable for the construction of efficient and accurate radio wave propagation models in small areas or areas with simple environmental parameters,and is not suitable for large areas with drastic path loss variations.(3)The path loss data are quantified according to the wireless communication system performance parameters,and a communication quality class data set is established to enhance the applicability of the BPNN-based radio wave propagation model in urban areas.The results show that the training model construction time is reduced by 60% using the communication quality class data set,and the complexity of the network structure is reduced accordingly.(4)The performance of the BPNN,Support Vector Regression(SVR)and Random Forest(RF)based propagation models is further compared.The study shows that all three methods are able to construct a propagation model for urban areas,but in general,the RF-based model has the best performance and can accurately and efficiently predict the communication quality among communication nodes based on the transmitting antenna frequency,transmitting antenna height,horizontal coordinates of transmitting and receiving antennas,and propagation distance vector. |
