基于灰色小波核偏最小二乘的传感器网络数据预测融合研究

Data prediction and fusion in a sensor network based on grey wavelet kernel partial least squares

针对如何降低传感器网络中采集的非平稳、非线性信号的数据传输量,提出了一种基于灰色Morlet小波核偏最小二乘(GMWKPLS)的预测融合模型。该模型把灰色模型预测的思想融入到核偏最小二乘(KPLS)中,采用构造的Morlet小波核函数进行数据变换,将输入映射到高维非线性的特征空间,在特征空间中,利用线性偏最小二乘方法构造预测融合模型。通过对齿轮箱断齿工况升速过程中的振动信号进行分析,结果表明,该模型使用滑动窗方法不断更新建模数据进行动态预测,预测精度高,可大大降低数据传输量,获得显著的节能收益。通过与灰色RBF核偏最小二乘(GRBFKPLS)和RBF核偏最小二乘(RBFKPLS)预测模型对比,GMWKPLS性能最佳,预测误差范围在±0.15%以内。

In order to reduce the amount of data of non-stationary and nonlinear signals collected in a sensor network,a grey Morlet wavelet kernel partial least squares（GMWKPLS） model was proposed.In this model,grey prediction theory was firstly introduced into kernel partial least squares（KPLS）.Then,the input-output data were mapped to a nonlinear higher dimensional feature space with Morelt kernel transformation.Finally,a prediction and fusion model was constructed with linear partial least squares.Moreover,the moving window method was utilized to update samples continuously in this dynamical prediction model.The model was validated using vibration signals of gear tooth breakage with rising speed.The results showed that the model can execute dynamic multi-step prediction,and has higher precision prediction;thus,it can obviously reduce the data amount in a sensor network and save energy;compared with grey RBF kernel partial least squares（GRBFKPLS） and RBF kernel partial least squares（RBFKPLS）,GMWKPLS is best in prediction performance,and the prediction errors are with in ±0.15%.