为保证旋转机械注水泵机组安全、稳定运行,应采用合适的预测模型对其状态评定参数进行预测。提出基于均值函数新息加权的遗传算法优化神经网络预测模型(MWGANN模型),用此模型能够优化神经网络结构参数,并可利用时间序列数据新旧程度的不同提高预测的精度和实时性。工业现场采集大型旋转注水泵机组振动烈度时间序列数据,应用MWGANN模型和基于人工经验设定神经网络结构参数的模型分别对其进行预测并比较,结果表明MWGANN模型在预测精度、预测实时性方面取得了较好的效果。 In order to ensure the safe and stable operation of rotary mechanical water injection pump units, the appropriate predictive models should be used to predict the state evaluation parameters. This paper proposes a new genetic algorithm (GA) neural network prediction model (MWGANN model) based on the weighted average of the mean value function, which can optimize the structural parameters of the neural network and improve the prediction accuracy and real-time performance by using the newness of the time series data. The industrial site collected time series data of vibration intensity of large rotary water injection pump units, predicted and compared MWGANN model and model based on artificial experience to set structural parameters of neural network respectively. The results show that the MWGANN model is obtained in prediction accuracy and real-time prediction A better effect.