基于GRNN的GSM-R场强覆盖预测算法

Prediction Algorithm of GSM-R Field Intensity Coverage Based on GRNN

在比较Hata模型修正方法和GRNN算法场强覆盖预测效果的基础上,仿真分析了训练集构成和平滑因子选择对GRNN算法预测效果的影响,给出了训练集构成和平滑因子选择的方法;提出了电波传播环境相似系数来表征GRNN模型在不同环境下的适用性。仿真结果表明,两种传播环境的相似系数越大,由一种环境下测试数据确定的GRNN在另一环境下的预测精度越高。

In this paper, the prediction accuracies of field intensity coverage were compared between the Hata modified model and the generalized regression neural network （GRNN） algorithm, and then some simulations were made to analyze the effect of the composition of training set and the smoothing factor on prediction accuracy of GRNN algorithm. Further, some guidelines for choosing the training set and smoothing factor were given. Finally, the paper suggested that： the applicability of GRNN model with different environment can be represented by similarity coefficient of radio propagation environment. The conclusion drawn from the simulation experiment result is that： the greater the similarity coefficient of two propagation environments is, the more accurate in another environment the prediction of GRNN model determined by testing data in one environment will become.