基于协同滤波和BP神经网络的电力作业人员运动工况辨识

Motion conditions identification of power operating person based on collaborative filtering and BP neural network

针对预警装置靠近相同带电体时不同工况下报警阈值选择的误差造成误报、漏报问题,提出基于协同滤波和反向传播(BP)神经网络对电力作业人员爬塔、爬坡、水平行走三种工况辨识的模型。利用加速度与气压传感器采集作业人员头部的加速度与气压值,对数据进行协同滤波,提取有效相对高度值,然后对有效高度值进行一阶拟合得到拟合参数。最后根据拟合参数建立BP神经网络模型识别三种工况。选取室外杆塔、斜坡及水平路面为实验平台,每种工况采集400组数据,随机抽取350组数据进行训练, 50组数据进行验证。验证结果表明:训练样本准确率达到94.95 %,测试样本准确率达到94.67 %,满足室外工况辨识的要求。

Aiming at the problem of false alarm and miss alarm caused by errors of alarm threshold choice under different operating conditions when warning devices approaching the same charged body,a model based on collaborative filtering and BP neural network is proposed to identify three working conditions such as towers climbing,slope climbing,and horizontal walking.Collecting the acceleration and air pressure values of the operator’s head by acceleration and air pressure sensors.Then the effective relative height value is extracted by collaborative filtering and fitting parameters are obtained by first-order fitting.Finally,according to the fitting parameters,BP neural network model is established to identify three working conditions.Outdoor towers,slopes and horizontal road surfaces are selected as experimental platforms and 400 groups of data are collected for each working condition.Then randomly select 350 groups of data for training,50 groups of data for verification.The results show that the accuracy of training samples and test samples is 94.95 % and 94.67 %,which meets the requirements of outdoor working conditions identification.