基于卷积-长短期记忆神经网络的抽水蓄能机组健康性能趋势预测

A Health Performance Tendency Prediction Model of Pumped Storage Unit Based on Convolution Neural Network-long Short-term Memory Neural Network

为准确掌握抽水蓄能机组的健康性能水平,提出基于卷积-长短期记忆神经网络(CNN-LSTM)的机组健康性能趋势预测方法。首先,为有效地刻画机组的运行特性,构建基于高斯过程回归的机组健康状态模型;然后,设计可量化机组健康性能的指标因子;进一步融合CNN良好的局部特征提取能力和LSTM在时间序列预测方面的优势,提出基于CNN-LSTM的预测模型。对国内某抽水蓄能电站机组监测数据进行的试验结果表明,所提方法可较好地预测机组健康性能的发展趋势。

To accurately obtain the health performance level of a pumped storage unit(PSU),a health performance tendency prediction method based on convolution neural network-long short-term memory neural network(CNN-LSTM)is proposed.Firstly,a unit health state model based on Gaussian process regression was constructed to effectively characterize the operating characteristics of the PSU.Then,an index that can quantify the health performance of the PSU was proposed.Finally,by integrating the good local feature extraction ability of the CNN and the advantage of the LSTM in time series prediction,a prediction model based on CNN-LSTM was proposed.The experiments were conducted using monitoring data from a pumped storage station in China.The results show that the proposed method can betterly predict the future evolution of the PSU's health performance.