摘要:
煤矿工作面设备通常由多个复杂系统模块组成，各模块间具有强耦合性，且设备故障机理复杂，在进行设备故障预测时需对设备的运行状态、环境数据、操作数据等进行实时监测，从而获取电气、机械、热工类多参数时序数据。提出一种基于时序数据对工作面设备进行故障预测的方法：首先，采用时序对齐算法将采集的设备监测数据对齐，即对监测数据的时间列重新排序，以时间列为关键值，各监测数据作为标签值填入，空缺值以前值填充；然后，根据故障表征现象和发生机理选取故障相关因素，通过Pearson相关系数分析法计算相关因素间的相关性，由此确定故障预测因素集；最后，采用长短期记忆（LSTM）网络建立工作面设备故障预测模型，以归一化的故障预测因素集作为LSTM预测模型的输入、故障作为输出，将迟滞时间段引入LSTM预测模型，实现了迟滞性故障的超前预测。以采煤机过热跳闸故障为例进行试验，通过分析得出故障预测因素集为{滚筒温度，滚筒电流，滚筒启停，牵引温度，变压器温度，摇臂温度}，当LSTM网络细胞层数为10、隐藏细胞数为10、学习率为0.001、迭代次数为1 500、1次读取样本个数为120时，采煤机过热跳闸故障的迟滞时间为30 min，采用测试集进行故障预测时可实现超前26 min预测，与迟滞时间相差4 min，表明采用LSTM网络可基于时序数据有效实现工作面设备故障的超前预测。
Abstract:
Coal mine working face equipment are usually consists of several complex system modules that have strong coupling among each other. Moreover, the equipment fault mechanism is complex. Therefore, when the equipment fault prediction is carried out, it is necessary to conduct real-time monitoring of equipment operation status, environmental data and operation data so as to obtain time series data of electrical, mechanical, thermal and other parameters. A method for fault prediction of working face equipment based on time series data is proposed. Firstly, the time series alignment algorithm is used to align the collected equipment monitoring data. The time columns of monitoring data are reordered, and the time columns are the key values. Each monitoring data is filled in as the label value, and the previous value is filled in the vacant value. Secondly, the fault-related factors are selected according to the fault characterization phenomenon and the occurrence mechanism. And the correlation between the relevant factors is calculated by Pearson correlation coefficient analysis method, thereby determining the fault prediction factor set. Finally, the long short-term memory(LSTM) network is used to establish a fault prediction model for working face equipment. The normalized set of fault prediction factor set is used as the input and the fault is used as the output of the LSTM prediction model. The delay time period is introduced into the LSTM prediction model to realize advanced prediction of delay faults. The test is carried out by taking the shearer overheating trip fault as an example. Through analysis, it is found that the fault prediction factor set is {drum temperature, drum current, drum start and stop, traction temperature, transformer temperature, rocker arm temperature}. When the number of LSTM network cell layers is 10, the number of hidden cells is 10, the learning rate is 0.001, the number of iterations is 1 500, and the number of samples read per time is 120, the delay time of shearer overheating trip fault is 30 min. When the test set is used for fault prediction, the advanced prediction time is 26 min , which is 4 min shorter than the delay time, indicating that the LSTM network can effectively achieve advanced fault prediction of working face equipment based on time series data.
百度学术
下载:
