压缩机实际能头特性的深度学习网络预测方法

1. 中国石油大学(北京)机械与储运工程学院;2. 中国石化新疆煤制天然气外输管道有限责任公司湖广分公司

压缩机;深度学习;BP 算法;特性预测;校核能头

Prediction method for compressor real energy head characteristics based on deep learning network
WANG Fuxi1,2, LI Xiaoping1, CHEN Xinguo1, WANG Wei1, GONG Jing1

1. College of Mechanical and Transportation Engineering, China University of Petroleum (Beijing);2. Huguang Branch of SINOPEC Xinjiang Gas Pipeline Transportation Co. Ltd.

compressor, deep learning, BP algorithm, characteristic prediction, check energy head

DOI: 10.6047/j.issn.1000-8241.2020.04.015

备注

压缩机运行特性与原厂测试特性存在差异,为了指导压缩机的安全稳定运行,结合压缩机特性计算方法与部分实际特性,建立了基于深度学习网络的压缩机实际能头特性预测模型。将大量不同工况下的压缩机实际能头数据作为深度学习网络的训练样本,在训练完成后利用未训练样本对模型精度进行了检验,得到最大相对误差为2.60%、最小相对误差为0.32%、平均相对误差为0.78%。由深度学习网络所绘制的能头曲线与实际的能头曲线有着良好的一致性。深度学习网络模型改进了传统神经网络的缺陷,具有良好的预测精度与泛化计算能力,为压缩机性能的评估与预测提供了新方法。(图5表2,参[28]

The operating characteristics of the compressor may differ from the factory test characteristics. In order to guide the safe and stable operation of the compressor, a prediction model of compressor real energy head characteristics based on deep learning network was established considering the calculation method of compressor characteristics and some real characteristics. A large number of real energy head data of compressors under different working conditions was used as training samples of the deep learning network, and after training, the model precision was checked by untrained samples, concluding the maximum relative error of 2.60%, the minimum relative error of 0.32% and the average relative error of 0.78%. The real energy head curve drawn by the deep learning network is well consistent with the actual energy head curve. The established deep learning network model improves the defects of the traditional neural network, with good prediction accuracy and generalization calculation ability, providing a new method for compressor performance evaluation and prediction. (5 Figures, 2 Tables, 28 References)

·