核动力系统运行工况预判的深度学习方法研究

Study for Distinguishing and Predicting Operation Conditions of Nuclear Power System Based on the Deep Learning Method

为建立核动力系统运行工况的高精度实时判别与运行监测参数的长时间在线预测方法,本研究基于神经网络模型,针对核动力系统运行状态预判开展了两方面工作。首先,基于核动力系统过去15个时间步(步长1 s)的监测参数变化,对运行工况进行实时诊断判别,同时,采用搜索算法对判别模型的参数进行优化,提高模型对运行工况的识别精度;之后,对核动力系统的关键运行参数进行超前预测。结果表明:优化后模型的诊断判别准确率稳定在0.99以上;在100个时间步的长时间序列下能够实现对于参数变化趋势的有效预测;对比支持向量机、K-近邻、多层前馈等多种经典的算法可知,改进的循环网络——注意力机制网络联合模型在核动力系统的运行预判方面表现优异。本研究所建立的运行工况预判方法可为保障核动力系统安全运行的辅助判断决策与超实时监测感知提供工程应用参考。

In order to establish a high-precision identification method for nuclear power system operating conditions and a long-term prediction method for operation monitoring parameters,two aspects of work for nuclear power system operating state identification and prediction are carried out based on the neural network model in this study.Firstly,based on the changes of monitoring parameters in past 15 time steps(step length:1 second),the real-time judgment model of operating conditions was carried out.Meanwhile,the grid search algorithm was used to optimize parameters of the model to improve identification accuracy for operating conditions.Then,the key operating parameters of nuclear power system are predicted in advance.The results show that identification accuracy of the optimized model is more than 0.99.In the long time series of 100 time steps,the effective prediction of parameters variation trend can be realized.Compared with support the vector machine,K-nearest neighbor,multilayer feedforward and other classical algorithms,it can be seen that the improved the cyclic net-attention mechanism network joint model has excellent performance in the operating condition identification and prediction of nuclear power system.The method established in this study can provide important engineering application reference for auxiliary judgment decision and super-real-time monitoring perception to ensure the safe operation of nuclear power system.