核技术
覈技術
핵기술
NUCLEAR TECHNIQUES
2015年
7期
070606-1-070606-5
,共1页
偏离泡核沸腾比%神经网络%误差反向传播算法%事故分析
偏離泡覈沸騰比%神經網絡%誤差反嚮傳播算法%事故分析
편리포핵비등비%신경망락%오차반향전파산법%사고분석
DNBR%Neural network%BP algorithm%Accident analysis
在压水堆事故分析中,通常采用系统分析程序、热流密度计算程序和子通道分析程序分步计算堆芯偏离泡核沸腾比(Departure from Nucleate Boiling Ratio, DNBR)。利用该方法计算的堆芯DNBR结果准确性较好,但是计算过程繁琐、费时。对于系统分析程序自带的堆芯DNBR简化计算模型,由于其根据堆芯限制线偏微分近似得到,适用范围较窄,准确性也难以保证。利用神经网络中的误差反向传播(Back Propagation, BP)算法,基于堆芯核功率、入口温度、流量和压力4个变量对应的一系列DNBR值,选取部分数据进行训练并建立模型,以达到快速和准确地预测堆芯DNBR的目的。根据误差分析,建立的计算模型具有较好的准确性,而且在部分失流事故和汽机停机事故下可较好地预测堆芯DNBR。
在壓水堆事故分析中,通常採用繫統分析程序、熱流密度計算程序和子通道分析程序分步計算堆芯偏離泡覈沸騰比(Departure from Nucleate Boiling Ratio, DNBR)。利用該方法計算的堆芯DNBR結果準確性較好,但是計算過程繁瑣、費時。對于繫統分析程序自帶的堆芯DNBR簡化計算模型,由于其根據堆芯限製線偏微分近似得到,適用範圍較窄,準確性也難以保證。利用神經網絡中的誤差反嚮傳播(Back Propagation, BP)算法,基于堆芯覈功率、入口溫度、流量和壓力4箇變量對應的一繫列DNBR值,選取部分數據進行訓練併建立模型,以達到快速和準確地預測堆芯DNBR的目的。根據誤差分析,建立的計算模型具有較好的準確性,而且在部分失流事故和汽機停機事故下可較好地預測堆芯DNBR。
재압수퇴사고분석중,통상채용계통분석정서、열류밀도계산정서화자통도분석정서분보계산퇴심편리포핵비등비(Departure from Nucleate Boiling Ratio, DNBR)。이용해방법계산적퇴심DNBR결과준학성교호,단시계산과정번쇄、비시。대우계통분석정서자대적퇴심DNBR간화계산모형,유우기근거퇴심한제선편미분근사득도,괄용범위교착,준학성야난이보증。이용신경망락중적오차반향전파(Back Propagation, BP)산법,기우퇴심핵공솔、입구온도、류량화압력4개변량대응적일계렬DNBR치,선취부분수거진행훈련병건립모형,이체도쾌속화준학지예측퇴심DNBR적목적。근거오차분석,건립적계산모형구유교호적준학성,이차재부분실류사고화기궤정궤사고하가교호지예측퇴심DNBR。
Background: In safety analysis of pressurized water reactor (PWR), departure from nucleate boiling ratio (DNBR) is usually calculated by three codes: a system transient analysis code, a heat flux calculation code and a subchannel analysis code, or by simplified model through a partial derivative approximation of the core DNB limit lines, but either procedure has problems of cumbersome or low accuracy.Purpose: The aim of this study is to gain a simple DNBR calculation method with high accuracy.Methods: A 3-layers back propagation (BP) neural network was proposed with a training data set to quickly predict DNBR using four variables of reactor coolant system (nuclear power, core inlet temperature, mass flow rate and pressure).Results: The error of the developed BP network is very small, and has similar results compared with the subchannel code calculations in two typical events.Conclusion: The trained BP network is accurate enough to be used in predicting DNBR, even in transient conditions.