基于LSTM的风机高速轴温度预测及预警方法研究

Research on LSTM based on Temperature Prediction and Warning Method for High-speed Shaft of Wind Turbines

在风力发电机高速轴温度监视中,最普遍的是设置定值阈值报警,但其触发报警条件单一,当设备温度超过设定温度阈值时触发报警,这种方式只考虑了设备最极限的温度值,无法对设备的缓慢劣化发出趋势预警。本文对风机SCADA系统(数据采集与监视控制系统)进行大数据挖掘,提出了一种基于长短期记忆神经网络(LSTM)的风机高速轴温度预测模型及预警方法,在LSTM输出层后依序加入Dropout层、全连接层和线性回归层进行温度预测,计算预测值与实际值的残差及残差均值,以残差均值为中心线,设置预警逻辑,采用滑动窗口的计数方式,实现高速轴温度趋势预警,解决现有风机高速轴温度监测因风资源不稳定、波动大和非线性的特点,而无法实现高精度提前预测和趋势预警问题。

In the temperature monitoring of wind turbine high-speed shaft,the most common is to set a fixed threshold alarm,but its triggering the alarm condition is single,when the equipment temperature exceeds the set temperature threshold,the alarm is triggered.This method only considers the maximum temperature value of the equipment,can not send a trend warning for the slow deterioration of the equipment.Based on the big data mining of wind turbine SCADA system(data acquisition and monitoring control system),this paper proposed a temperature prediction model and early warning method of wind turbine high-speed shaft based on Long Short-Term Memory neural network(LSTM).Dropout layer,full connection layer and linear regression layer are added in order after the LSTM output layer to predict the temperature.Calculation the actual and estimated values of residual error and the residual mean,with the center line of the residual averages,set up the early warning and logic,the sliding window way of counting,the tendency of high speed shaft temperature early warning,to solve the existing fan shaft temperature monitoring because of the wind resource is not stable,fluctuation and the characteristics of nonlinear,cannot achieve high accuracy prediction and trend in advance warning.