关键词: 旋转控制头, 轴承温度, 遗传算法, 神经网络, 置信区间

Abstract:

Rotary contol head（RCH）bearing assembly withstands great dynamic load，and severe heat and abrasion resulting
from the friction force. Shorter equipment life may arise because of bearing failure caused by excessive bearing temperature.
Aiming to overcome the difficulty in precise calculating and measuring，due to various influence factors on RCH bearing
temperature，a method based on GA-BP（the optimized algorithm of BP neural network based on genetic algorithm，GA-BP）is
proposed to predict RCH bearing temperature. The bench test data of an outboard cooling and lubrication pump station RCH
was used for training and testing，and traditional neural network model（BP）was used for comparison. Results show that，the
GA-BP prediction model can realize adaptive control for RCH bearing temperature prediction process. The linear correlation
between prediction value and the expectative output comes up to 0.991 48. 95% confidence interval and mean，max，min
absolute percentage error were contrasted between GA-BP and BP，and the result shows that the approximation capability，
convergence and generalization ability of GA-BP are better than BP. With high prediction accuracy and good stability，GA-BP
model can help monitor the bearing running state，and optimization of the cooling and lubrication stuctures. The GA-BP model
has an important guiding significance in improving the overall performance of RCH.

Key words: rotary control head, bearing temperature, genetic algorithms, neural networks, confidence interval