基于 LSTM 循环神经网络的核电设备状态预测-《计算机技术与发展》

文章信息/Info

Title:: Nuclear Power Equipment Status Prediction Based on LSTM Recurrent Neural Network

Author(s):: GONG An; MA Guang-ming; GUO Wen-ting; CHEN Chen; School of Computer &Communication Engineering,China University of Petroleum,Qingdao 266580,China

Keywords:: nuclear power equipment; time series data; recurrent neural networks; state forecast; deep learning

摘要:: 核电站的规模随着经济的发展日益扩大,核电设备运行状态的研究已成为数据挖掘的重要研究领域。核电设备是高可靠性和高安全性的复杂系统,多年的设备运行产生了大量的时间序列数据。为了解决核电设备运行状态难以准确预测等问题,提出了一种基于长短期记忆(long short-term memory,LSTM)循环神经网络的核电设备状态预测方法。首先除去原数据中噪点明显的数据,然后使用 z-score 标准化方法对数据进行预处理,然后实现 LSTM 的网络结构设计、网络训练和预测,最后对预测结果进行比较分析。考虑到核电设备各个部件运行产生的数据种类繁多,选择与核电设备运行状态相关的重要数据主泵电机绕组温度作为研究对象。通过与 GRU、RNN 等模型进行对比实验,表明了该算法对核电设备的运行状态有更高的预测精度。

Abstract:: Nuclear power plants grow in size as the economy grows. The research on the running state of nuclear power equipment has become an important field of data mining. Nuclear power equipment is a complex system with high reliability and security. Years of equipment operation have produced a large amount of time series data. In order to solve the problem that it is difficult to predict the operational status of nuclear power equipment accurately,a nuclear power equipment state prediction method based on long-term memory (LSTM) recurrent neural network is proposed. Firstly,the data with obvious noise points in the original data are removed,and then preprocessed by z-score standardization method. After that,the network structure design,network training and prediction of LSTM are implemented. Finally,the predicted results are compared and analyzed. Considering the variety of data reflecting the operating status of nuclear power equipment,choose the main pump motor winding temperature related to the operating status of nuclear power equipment as a experiment object. Through comparison experiments with GRU and RNN,the proposed algorithm has higher prediction accuracy for the operating status of nuclear power equipment.