Field Strength Prediction Model With Modular Neural Networks

It is of great importance for the design and plan of a wireless communication system.Radio wave propagation environments becomes more complex along with the development of wireless communication technology,so some field strength prediction models based on statistical analysis in early time can hardly work no longer,whereas deterministic models based on site-specific environment information require large time consumption.So,field strength prediction model based on artificial neural network is proposed by some researchers and investigated in this thesis.Field strength prediction model based on neural network has the advantages of both statistical models and deterministic models,that is high prediction precision and fast computation speed.A field strength prediction model based on modular neural network is proposed in this thesis,which takes advantage of the characteristics of field strength distribution.A set of sample is collected as training data,and clustered with KMeans algorithm to dispose the input space,and then the subnet is trained.The modular neural network disposes the complex problem to several simple problems and solve them one by one,so it is more suitable to complex problems.Simulation is carried out with modular neural network in a specific scene,and the result is compared with that from single neural network.It is shown that modular neural network model is more accurate than single neural network and is more effective for field strength prediction.