基于静态权值组合集成模型的WSN时钟偏差估计

针对无线传感器网络(WSN)时钟同步精度低、复杂度高等问题，提出了一种基于静态权值组合集成模型的时钟偏差预测方法。对传感器节点的时间戳观测值进行有放回抽样，将面向回归问题的AdaBoost.RT集成学习算法的误差函数和阈值调整方法进行改进，并以改进的AdaBoost.RT算法作为集成框架，采用离散神经元网络(DPNN)作为弱学习机构建集成局域模型对时间偏差进行有效预测。实验表明，对于长期预测，AdaBoost.RT模型和改进型AdaBoost.RT模型的预测效果相对于DPNN全局模型提升了20%。此外，在长期观测和短期观测两种情况下，AdaBoost.RT改进型模型的预测效果要优于AdaBoost.RT模型，能够更有效的减小时间估计偏差。

Clock bias estimation for WSN network based on static weights integration model

Aiming at the low accuracy and high complexity in the clock synchronization procedure for WSN wireless sensor network, this paper proposes a clock error prediction method based on static weight composite integration model. Sample back into the observed time stamp value of sensor nodes, and then improve the error function and threshold adjustment method of the algorithm for the regression problems based AdaBoost.RT integrated learning algorithm. Lastly, the improved AdaBoost. RT algorithm will be used as an integration framework, and DPNN will be used as weak machine learning to build integrated local model to effectively predict the time deviation. Experiments show that for long-term prediction, prediction effect of the AdaBoost.RT model and the improved AdaBoost.RT model increased by 20% compared with DPNN global model. In addition, in both the long-term and short-term observations, the predicted effect of the improved AdaBoost.RT model is superior to AdaBoost.RT model, and it can more effectively reduce clock estimation bias of the WSNs.