一种基于模块化神经网络的场强预测方法

Field Strength Prediction Based on Modular Neural Network

接收信号场强预测对无线通信网络的设计与规划非常重要.为此,提出了一种基于模块化神经网络的场强预测模型.对于给定的区域,选取一定数量的接收样本点,根据接收信号场强数据的分布特点,使用K均值(K-Means)聚类方法对全部样本点聚类,以实现对输入样本空间的分解,并建立对应的子神经网络模块.以某学校宿舍区域为例,选取了训练集和测试集样本点,通过对比单一神经网络模型和模块化神经网络模型的预测误差,发现模块化神经网络的预测结果优于单一神经网络,证明了所提出模型的有效性.

A wireless signal electronic field strength prediction model based on modular neural network is presented. For a given area,a number of receiving points is selected for the sample. All the sample points is clustered with the method of K-means according to the feature of distribution of received signal field strength,for the purpose of decomposing sample space of inputs and setting up corresponding sub-modules of neural network. Sample points of training set and test set is collected in an area of university building. Comparison is made between prediction error obtain from single BP neural network and from modular neural network. The result show that modular neural network model is more accurate than single BP neural network.