Research on Link Quality Estimation Method Based on Incremental Learning

链路质量估计是无线传感器网络建立多跳传输路径的基础，受干扰、噪声和多径效应等因素影响，采用离线训练的方法很难建立适用于各种条件的链路质量估计模型。文章提出一种基于增量学习的链路质量估计方法LEIL，以在线获取的链路质量指标LQI和信噪比SNR作为输入特征，PRR作为预测目标，在线建立链路估计模型。以EFDT为基础，采用均方误差MSE作为纯度指标，实现了决策回归树的增量生成。实验结果显示，LEIL具有较高的预测精度，不同链路质量条件下平均的MAE为0.0528。与现有方法相比，LEIL可以自适应环境，无需重新建立训练数据集，减小了建立链路估计模型的成本，适用于各种工作条件的无线传感器网络。Link quality estimation is the foundation for establishing multi-hop transmission paths in wireless sensor networks. Due to factors such as interference, noise, and multipath effects, it is difficult to establish a link quality estimation model suitable for various conditions using offline training methods. This paper proposes an incremental learning-based link quality estimation method LEIL, which takes the online-obtained link quality index LQI and signal-to-noise ratio SNR as input features, PRR as the prediction target and establishes a link estimation model online. Based on EFDT and using mean square error (MSE) as the purity indicator, an incremental generation of decision regression trees has been achieved. The experimental results show that LEIL has high prediction accuracy, with an average MAE of 0.0528 under different link quality conditions. Compared with existing methods, LEIL can adapt to the environment without rebuilding the training dataset, reducing the cost of establishing a link estimation model, and is suitable for wireless sensor networks under various working conditions.