Estimation of break location and size for loss of coolant accidents using neural networks

In this work, a probabilistic neural network (PNN) that has been applied well to the classification problems is used in order to identify the break locations of loss of coolant accidents (LOCA) such as hot-leg, cold-leg and steam generator tubes. Also, a fuzzy neural network (FNN) is designed to estimate the break size. The inputs to PNN and FNN are time-integrated values obtained by integrating measurement signals during a short time interval after reactor scram. An automatic structure constructor for the fuzzy neural network automatically selects the input variables from the time-integrated values of many measured signals, and optimizes the number of rules and its related parameters. It is verified that the proposed algorithm identifies very well the break locations of LOCAs and also, estimate their break size accurately.     

Applying particle swarm optimization algorithm for tuning a neuro-fuzzy inference system for sensor monitoring

A neuro-fuzzy inference system (ANFIS) tuned by particle swarm optimization (PSO) algorithm has been developed for monitoring the relevant sensor in a nuclear power plant (NPP) using the information of other sensors. The antecedent parameters of the ANFIS that estimates the relevant sensor signal are optimized by a PSO algorithm and consequent parameters use a least-squares algorithm. The proposed methodology to monitor sensor output signals was demonstrated through the estimation of the nuclear power value in a pressurized water reactor using as input to the ANFIS six other correlated signals. The obtained results are compared to two similar ANFIS using one gradient descendent (GD) and other genetic algorithm (GA), as antecedent parameters' training algorithm.     

Application of neural networks to signal prediction in nuclearpower plant

The feasibility of using an artificial neural network for signal prediction is studied. The purpose of signal prediction is to estimate the value of the undetected next-time-step signal. In the prediction method, which is based on the idea of autoregression, a few previous signals are input to the artificial neural network, and the signal value of next time step is estimated from the outputs of the network. The artificial neural network can be applied to a nonlinear system and has fast response. The training algorithm is a modified backpropagation model, which can effectively reduce the training time. The target signal of the simulation is the steam generator water level in a nuclear power plant. The simulation result shows that the predicted value follows the real trend well     

Prediction of major transient scenarios for severe accidents of nuclear power plants

It is very difficult for nuclear power plant operators to predict and identify the major severe accident scenarios following an initiating event by staring at temporal trends of important parameters. In this regard, a probabilistic neural network (PNN) that has been applied well to the classification problems is used in order to classify accidents into groups of initiating events such as loss of coolant accidents (LOCA), total loss of feedwater (TLOFW), station blackout (SBO), and steam generator tube rupture (SGTR). Also, a fuzzy neural network (FNN) is designed to identify their major severe accident scenarios after the initiating events. The inputs to PNN and FNN are initial time-integrated values obtained by integrating measurement signals during a short time interval after reactor scram. An automatic structure constructor for the fuzzy neural network automatically selects the input variables from the time-integrated values of many measured signals, and optimizes the number of rules and its related parameters. In cases that an initiating event develops into a severe accident, this may happen when plant operators do not follow the appropriate accident management guidance or plant safety systems do not work, the proposed algorithm showed accurate classification of initiating events. Also, it well predicted timings for important occurrences during severe accident progression scenarios, which is very helpful to perform severe accident management.     

基于BP神经网络的核探测器故障诊断方法研究

核探测器是核设施放射性监测的重要设备,为了保障该设备的持续稳定运行,本研究针对闪烁体探测器提出了一种基于BP神经网络的在线智能故障诊断方法。采用小波包变换将探测器输出信号从时域变换至频域后提取特征向量,将得到的特征向量作为BP神经网络故障诊断模型的输入,再通过误差梯度下降法对该模型的参数进行优化,最终利用最优的诊断模型完成故障类型的智能识别与分类,并将该方法与统计诊断方法和基于支持向量机的故障诊断方法进行横向的对比研究。研究结果表明,新方法的平均诊断准确率均优于上述两种方法。因此,该方法的应用能有效地提高核探测器的故障诊断准确率。     

The autoassociative neural network in signal analysis: I. The data dimensionality reduction and its geometric interpretation

In complex and risky plants, such as the nuclear reactors, the analysis of the signals released by the many sensors which monitor the plant represents a difficult task due to the high-dimensionality of the data. This paper is the first of two in which we tackle the problem of the dimensionality reduction by the nonlinear principal components analysis as performed by an autoassociative neural network (AANN). This network filters the many input data and releases at the bottleneck output a relatively small number of signals which capture the significant properties of the original data, thus realizing the data reduction.In the present paper, we show that the network ability in correctly reproducing as output the given input after a passage through the bottleneck layer (which by definition should have fewer nodes than either input or output layers) could be conceived as a topological mapping between abstract spaces. Apart from the less critical choice of the number of nodes in the mapping and demapping layers, the topological mapping will be successful – and the AANN will be able to perform the required data reconstruction – provided that the number of nodes of the bottleneck layer is related to the dimensionality d of the abstract projection space. We show how to obtain a numerical estimate d* for the real dimension d. This numerical estimate will firmly base the choice of the number of nodes f of the bottleneck layer, thus avoiding the usual troubling trial-and-error procedure. The power of the proposed approach is demonstrated firstly on a few geometrical cases and then on the analysis of nuclear transients simulated by the classic Chernick’s model.     

模糊神经网络原理及其在核爆监测中的应用

本文实现了一种利用基于ＢＰ算法的模糊神经网络（ＦＮＮ）建立模糊模型的方法，这种方法可自动生成非线性系统的模糊模型，ＦＮＮ被用来产生模糊规则和隶属函数，用数值计算和异或（ＸＯＲ）问题对该方法的可行性和置信度进行了检验，结果表明，模糊神经网络能显著地改进模糊系统设计中的准确度，可靠性，缩短设计周期，降低系统成本，ＦＮＮ用于核爆对人员伤害估计并可将网络简化成一种用于信号极点提取的规则神经网络（ＲＮＮ）初     

Integrating neural network technology and noise analysis

The integrated use of neural network and noise analysis technologies offers advantages not available by the use of either technology alone. The application of neural network technology to noise analysis offers an opportunity to expand the scope of problems where noise analysis is useful and unique ways in which the integration of these technologies can be used productively. The two-sensor technique, in which the responses of two sensors to an unknown driving source are related, is used to demonstration such integration. The relationship between power spectral densities (PSDs) of accelerometer signals is derived theoretically using noise analysis to demonstrate its uniqueness. This relationship is modeled from experimental data using a neural network when the system is working properly, and the actual PSD of one sensor is compared with the PSD of that sensor predicted by the neural network using the PSD of the other sensor as an input. A significant deviation between the actual and predicted PSDs indicate that system is changing (i.e., failing). Experiments carried out on check valves and bearings illustrate the usefulness of the methodology developed.     

A novel channel selection method for CANDU refuelling based on the BPANN and GA techniques

A novel channel selection method for CANDU refuelling based on the back-propagation artificial neural network (BPANN) and genetic algorithm (GA) techniques is developed. In this method, GA is used as an “optimization tool” and BPANN as a refuelling “simulator” used to predict the core parameters. Based on this method an automatic refuelling channel selection program for CANDU reactors has been developed and tested by the refuelling simulation of the Qinshan Phase III CANDU reactor for 400 effective full power days. The numerical results show that the average properties of the time-dependent core are very close to the reference one and the refuelling channel selection method possesses superior computational efficiency.     

The autoassociative neural network in signal analysis: III. Enhancing the reliability of a NN with application to a BWR

This is the third and last of a series of papers trying to unveil the opaqueness of neural networks structure through a geometrical approach [Marseguerra M., Zoia, A., 2005a. The autoassociative neural network in signal analysis: I. The data dimensionality reduction and its geometric interpretation. Ann. Nucl. Energy 32, 1191–1206, Marseguerra, M., Zoia, A., 2005b. The autoassociative neural network in signal analysis: II. Application to on-line monitoring of a simulated BWR component. Ann. Nucl. Energy 32, 1207–1223]. Artificial neural networks (NN) provide a powerful tool in the operation of complex systems, such as nuclear power plants, in that they are suitable to determine the relationship between measured variables and control parameters on the basis of input-output examples. However, their major drawback is the fact that they always provide an output to the user, regardless of the appropriateness of the input. In this paper, we propose to adopt an autoassociative neural network (AANN) to work in cooperation with the NN to first assess the well-posedness of the desired neural model and to successively establish the appropriateness of the input data. The neural algorithm has been applied to a nuclear problem: the estimation of the reactivity forcing function parameters from the values of the measured neutron flux in a BWR reactor (provided by a reduced-order literature model). In this example, the AANN was able to suggest through geometrical considerations how to decompose the dataset in order to obtain a successful training for the NN and thereafter to validate the input data, thus enhancing the reliability of the NN model output.     

Signal reconstruction by a GA-optimized ensemble of PCA models

On-line sensor monitoring allows detecting anomalies in sensor operation and reconstructing the correct signals of failed sensors by exploiting the information coming from other measured signals. In field applications, the number of signals to be monitored is often too large to be handled effectively by a single reconstruction model. A more viable approach is that of decomposing the problem by constructing a number of reconstruction models, each one handling an individual group of signals. To apply this approach, two problems must be solved: (1) the optimal definition of the groups of signals and (2) the appropriate combination of the outcomes of the individual models. With respect to the first problem, in this work, Multi-Objective Genetic Algorithms (MOGAs) are devised for finding the optimal groups of signals used for building reconstruction models based on Principal Component Analysis (PCA). With respect to the second problem, a weighted scheme is adopted to combine appropriately the signal predictions of the individual models. The proposed approach is applied to a real case study concerning the reconstruction of 84 signals collected from a Swedish nuclear boiling water reactor.     

集成神经网络方法在蒸汽发生器故障诊断中的应用

针对蒸汽发生器传统故障检测与诊断方法的不足，提出了基于集成神经网络的蒸汽发生器故障检测与诊断的新方法。该方法采用两个神经网络。一个神经网络作为蒸汽发生器的动力学模型，用于蒸汽发生器的重要运行参数的预测，其原理是通过检测蒸汽发生器运行参数监测信号值与相应的蒸汽发生器神经网络模型预测值之间的偏差来确定是否发生了异常，如果某一参数偏差超过了预先给定的极限，就认为发生了异常。另一个神经网络作为故障分类模型，用以对蒸汽发生器故障进行分类，给出故障的类型。由两个神经网络监测和诊断结果的融合给出蒸汽发生器故障较为清晰的信息。仿真结果表明，该方法能够提高蒸汽发生器监测与诊断的能力。     

基于能量分布特征的地震事件自动识别

研究了地震信号在小波包变换下的特性,依据地震事件识别中“历史事例对比法”的思想,根据不同震源地震信号频率时变特性的不同．提出了基于“能量分布特征”的特征值,同时采用该特征值用神经网络方法对地震事件进行识别分类。该方法不依赖于系统的数学模型,而是直接利用各频率成分能量的变化提取特征值作为神经网络的输入特征向量来进行事件的识别,避免了对地震信号、传播途径准确建模的困难,简便、直观地完成了事件的识别。实验证明,该方法的事件识别率可达到99％以上．是一种有效的地震事件识别方法。     

遗传算法优化的BP神经网络在EDXRF中对钛铁元素含量的预测

在能量色散X荧光分析(EDXRF)技术中,受均匀效应、颗粒效应和基体效应等的干扰,定量分析精度受到影响。本文针对这一问题提出了遗传算法(GA)优化BP神经网络(GA-BP)的混合算法,该算法无需考虑元素浓度和射线强度之间的复杂关系。遗传算法优化BP神经网络的目的是为了获得更好的网络初始权值和阈值,其基本思想是:将初始化的BP神经网络均方根误差的倒数编码为遗传算法中个体的适应度;初始的权值和阈值用遗传算法中的个体代替,然后通过选择、交叉和变异操作挑选出最优个体,最后通过解码用最优的权值和阈值创建一个新的BP网络模型。攀枝花矿区5类矿样中钛和铁含量的整体预测和分类预测实验表明,分类预测效果远好于整体预测。预测值与化学分析值比较结果表明,其中76.7%的样品相对误差小于2%,表明了该方法在元素间基体效应校正上的有效性。     

Disturbance Identification Method of Pressurized Water Reactor Plant Operated with Automatic Frequency Control by Applying Projection Operator Technique

The identification method applying the projection operator for less small disturbances which do not instantly lead the alarm initiation has been extended to diagnose PWR plants operated with automatic frequency control. The authors previously proposed the non-linear transformation of the observed signals to satisfy with the linear relationship between the disturbances and the observed signals by utilizing several additional observed signals which are closely related with the non-linear characteristics of PWR plant. In addition, a method is proposed to compensate the fluctuation of observed signals due to automatic frequency control by utilizing both the input signal of the main steam valve control system and the input signal of the control rod control system. The effectiveness of the extended method has been examined by several computer experiments using a simple PWR plant simulation code.     

Neural Network Prediction of Disruptions Caused by Locked Modes on J-TEXT Tokamak

Prediction of disruptions caused by locked modes using the Back-Propagation （BP） neural network is completed on J-TEXT tokamak. The network, which is based on the BP neural network, uses Mirnov coils and locked mode coils signals as input data, and outputs a signal including information of prediction of locked mode. The rate of successful prediction of locked modes is more than 90%. For intrinsic locked mode disruptions, the network can give a prewarning signal about 1 ms ahead of the locking-time. For the disruption caused by resonant magnetic perturbation （RMPs） locked modes, the network can give a prewarning signal about 10 ms ahead of the locking-time.     

Critical heat flux function approximation using genetic algorithms

Function approximation is the problem of finding a system that best explains the relationship between input variables and an output variable. We propose two hybrid genetic algorithms (GAs) of parametric and nonparametric models for function approximation. The former GA is a genetic nonlinear Levenberg-Marquardt algorithm of parametric model. We designed the chromosomes in a way that geographically exploits the relationships between parameters. The latter one is another GA of nonparametric model that is combined with a feedforward neural network. The neuro-genetic hybrid here differs from others in that it evolves diverse input features instead of connection weights. We tested the two GAs with the problem of finding a better critical heat flux (CHF) function of nuclear fuel bundle which is directly related to the nuclear-reactor thermal margin and operation. The experimental result improved the existing CHF function originated from the KRB-1 CHF correlation at the Korea Atomic Energy Research Institute (KAERI) and achieved the correlation uncertainty reduction of 15.4% that would notably contribute to increasing the thermal margin of the nuclear power plants.     

Comparisons between the various types of neural networks with the data of wide range operational conditions of the Borssele NPP

This paper addresses a trend monitoring in operating nuclear power plant by use of two types of Recurrent Neural Networks (RNN). The interesting feature of the RNN is intrinsic dynamic memory that reflects the current output as well as the previous inputs and outputs are gradually quenched. The first one Elman type of RNN which has a feed-back from hidden layer to the input layer neurons while in the Jordan type, from the outputs of the neural net to the inputs of the neural net. In this paper the theoretical assessment of the both RNNs is given. Both topological structures including Back Propagation (BP) neural network were implemented to the Borssele NPP. Learning achieved from 30% to 100% nominal power at the starting period of the new core 30 September 2001. After learning period the reactor operation is followed by the neural network. Paper will present the reactor system, the real time data collection and the merits of the three types of the neural network applied while in the learning and continuous processing of the changing of the operational conditions.