基于深度卷积神经网络的堆芯换料方案性能评价研究

Evaluation of Core Refueling Loading Patternwith Deep Convolutional Neural Network

为精确高效地评价堆芯换料方案的性能,本文基于深度卷积神经网络算法提出了一种换料方案——堆芯关键参数预测方法。该预测方法引入Inception-ResNet的卷积网络结构以提高网络深度和学习效率,通过学习基于换料经验生成的大量堆芯换料方案,拟合得到换料方案与堆芯关键参数之间的映射关系。针对某二代改进型机组的实验结果表明,该预测方法对测试集中堆芯换料方案的临界硼浓度的平均预测误差为0.86 ppm,功率峰因子与核焓升因子平均相对误差分别为0.54%与0.38%,平均每个换料方案关键参数预测用时0.0005 s左右。上述结果表明本文提出的预测方法具有较好的泛化能力和较高的可靠性,为换料方案优化提供了一种快速评价的方法。

In order to evaluate the key parameters of core refueling loading pattern accurately and efficiently,a method based on the deep convolutional neural network algorithm was proposed in this paper.Applying the Inception-ResNet blocks,the proposed method can effectively accelerate the learning process and improve the neural network depth.Through learning the sufficient samples of the core refueling loading pattern,corresponding relationship between the key parameters and refueling loading patterns can be established.Taking the improved Gen-ⅡPWR as the numerical results,it can be observed that the average error of the critical boron concentration is 0.86 ppm,and the average relative errors of the pin-power peak and nuclear enthalpy-rise factor are 0.54%and 0.38%respectively.Moreover,the time costed for the key parameter prediction of one-single refueling loading pattern is only 0.0005 second.Therefore,the method proposed in the present work has very high performance and high reliability in predicting the key parameters of the refueling loading pattern,which can provide a fast-evaluation method for the refueling loading pattern optimization.