Development of the effectiveness measure for an advanced alarmsystem using signal detection theory

Since many alarms which are activated during major process deviations or accidents in nuclear power plants can result in negative effects for operators, various types of advanced alarm systems that can select important alarms for the identification of process deviation have been developed to reduce the operator's workload. However, the irrelevant selection of important alarms could distract the operator from correct identification of process deviation. Therefore, to evaluate the effectiveness of the advanced alarm system, a tradeoff between the alarm reduction rate (how many alarms are reduced?) and informativeness (how many important alarms that are conducive to identifying process deviation are provided?) of an advanced alarm system should be considered. In this paper, a new measure is proposed to evaluate the effectiveness of an advanced alarm system with regard to the identification of process deviation. Here, the effectiveness measure is the combination of informativeness measure and reduction rate, and the informativeness measure means the information processing capability performed by the advanced alarm system including wrong rejection and wrong acceptance, and it can be calculated using the signal detection theory (SDT). The effectiveness of the prototype alarm system was evaluated using the loss of coolant accident (LOCA) scenario, and the validity of the effectiveness measure was investigated from two types of the operator response, such as the identification accuracy and the operator's preference for the identification of LOCA. From the investigation, it can be concluded that the effectiveness measure is appropriately used to evaluate an advanced alarm system because the operator can perform the identification of process deviation more correctly and easily when the effectiveness score is sufficiently high     

基于事件触发机制的核电站智能诊断专家系统置信规则库的研究

为提高核电站故障诊断的准确性和及时性,提出采用以核电站主要设备运行参数的报警状态为事件的触发机制对核电站设备故障进行诊断,利用核电站相关运行参数的信息集合建立事件触发下的核电站智能诊断专家系统置信规则库,而构建规则库采用故障机理模型与核电模型等相结合的方式,即在故障引起的报警下进行描述故障的征兆集合提取、规则的表示和规则变量的设定。通过在核电模型中人为引入故障,利用基于事件触发机制的核电站智能诊断专家系统进行故障诊断。诊断结果表明,本系统诊断出的故障类型与在核电模型中引入的故障类型一致,验证了本系统诊断结果的准确性,证明了此规则库的有效性和可行性。     

Development of advanced I&C in nuclear power plants: ADIOS and ASICS

In this paper Automatic Startup Intelligent Control System (ASICS) that automatically controls the PWR plant from cold shutdown to 5% of reactor power and Alarm and Diagnosis-Integrated Operator Support System (ADIOS) that is integrated with alarms, process values, and diagnostic information to an expert system focused on alarm processing are described. Nuclear Power Plant is manually controlled from cold shutdown to 5% according to the general operation procedures for startup operation of nuclear power plant. Alarm information is the primary sources to detect abnormalities in nuclear power plants or other process plants. The conventional hardwired alarm systems, characterized by one sensor-one indicator may lead the control room operators to be confused with avalanching alarms during plant transients. ASICS and ADIOS are designed to reduce the operator burden. The advances in computer software and hardware technology and also in information processing provide a good opportunity to improve the control systems and the annunciator systems of nuclear power plants or other similar process plants. It is very important to test and evaluate the performance and the function of the computer- or software-based systems like ASICS and ADIOS. The performance and the function of ASICS and ADIOS are evaluated with the real-time functional test facility and the results have shown that the developed systems are efficient and useful for operation and operator support.     

方家山核电厂火灾频繁误报警原因分析及策略

火灾报警系统是核电厂消防系统的重要组成部分,运行人员通过火警主机CRT及DCS二层报警系统判断现场是否出现火警情况,火灾报警系统的可靠性决定了机组运行是否存在潜在火灾风险。本文阐述了方家山机组运行以来,对"方家山火灾频繁误报警"的报警进行分类统计和原因分析,针对典型的、占比较高的缺陷类型制定了应对策略,并对相关的优化改造效果进行了经验总结,希望能对其他核电厂有一定的借鉴和指导意见。     

Development strategies of an expert system for multiple alarmprocessing and diagnosis in nuclear power plants

The development strategies of a prototype expert system, called ESAPD, for multiple alarm processing and diagnosis in nuclear power plants are described. The main objectives of the system are to assist operators in identifying a primary causal alarm among multiple fired alarms and to diagnose the plant malfunction quickly. The overall plant-wide diagnosis is performed at the alarm processing stage which can identify a primary causal alarm and can diagnose possible failure modes and failed systems and automatic interlock actions. The knowledge base for the alarm processing is represented as object-oriented concepts. The specific root cause diagnosis for the primary causal alarm can be performed at the alarm diagnosis stage. The system can provide operators with the possible causes of the primary causal alarm, emergency actions, and follow-up treatments. The diagnostic method adopted in this system is a hypothesize-and-test paradigm     

基于自适应AR模型的核电站松动件报警方法

为能快速准确地检测到核电站一回路零部件的松动或脱落,提出1种基于自适应AR（auto-regressive）模型的松动件报警方法。该方法利用自适应AR模型跟踪一回路中背景噪声的变化,先对信号进行白化处理,再计算白化后信号的短时均方根（RMS）,设置RMS动态阈值实现报警。采用秦山核电站一号机组背景噪声和松动件碰撞信号叠加进行了仿真试验,结果表明,该方法能够在低信噪比和噪声复杂变化的条件下快速检测出松动件碰撞信号。     

秦山核电厂松动部件监测系统典型误报警事件分析和处理

松动部件监测系统（LPMS）是核电厂监测一回路中是否有松脱件的基本安全工具，误报警是困扰松动部件监测的一个最重要的问题。为了甄别误报警事件，分析了秦山核电厂运行过程中的典型报警事件数据和工况信息，确认主泵启停过程中的报警、雷雨天气触发的报警、堆顶风机切换触发的报警、主系统在升温升压过程的报警为误报警，并针对这几类典型的误报警给出了解决方案。      

核电厂松脱事件报警及应急响应研究

通过对松脱部件事件报警逻辑及国内外松脱部件监测法规和标准中关于松脱事件报警及应急响应的研究，结合国内某核电厂热试期间发生的松脱事件报警及报警事后应急响应，详细分析了该核电厂松脱事件过程。松脱事件报警分为绝对阈值和相对阈值报警，本文提出在系统报警逻辑中增加了计数因子和通道复核来提高报警准确率；GB/T 11807、IEC 60988和ASME均要求事件报警后采取措施确定报警是否是松脱部件所致，以及松脱部件对反应堆一回路系统产生的潜在危害进行评估；RG 1.133则要求确认松脱事件报警后应上报核管理委员会，GB/T 11807和IEC 60988则根据具体运行电厂确定，对于是否上报未做明确规定。基于国内某核电厂出现松脱事件报警及其应急响应，建议建立适合于国内核电厂松脱事件报警及应急响应机制，对于运行中出现的松脱事件及时确认，并对设备部件可能造成的损失情况进行及时评估，并对是否上报及采取何种措施做出进一步的规定。      

Application of neural networks to multiple alarm processing anddiagnosis in nuclear power plants

A feasibility study of multiple alarm processing and diagnosis using neural networks is presented. The backpropagation network (BPN) algorithm is applied to the training of multiple alarm patterns for the identification of faults in a reactor coolant pump (RCP) system. The general mapping capability of the neural network makes it possible to identify a fault easily. A number of case studies are performed, with emphasis on the applicability of the neural network to the pattern recognition of multiple alarms. Based on the case studies, the neural network can identify the cause of multiple alarms well, although untrained, incomplete/sensor-failed or time-varying alarm symptoms are given. Also, multiple faults are easily identified with a given alarm pattern     

A New Diagnosis Method Using Alarm Annunciation for Nuclear Power Plants

We discuss the methodology diversity for diagnosis reasoning in an autonomous operation system, and propose a new diagnosis method using an alarm annunciation system. The combination of annunciated alarms is expected to be peculiar to the anomalous phenomenon or accident. Moreover, as the state of affairs is developing, each appearance of the pattern is changing with time peculiarly to each anomaly or accident. The matter is utilized for the new diagnosis method. The patterns of annunciated alarms with progress of the events are prepared in advance under the condition of the anomalies or accidents by use of a plant simulator. The diagnostic reasoning can be done by comparing the obtained combination of annunciated alarms with the reference templates by using pattern matching method. On the other hand, we have another method, called COBWEB used for conceptual classification in cognitive science, to reason for diagnosis. We have carried out the experiments using the loop type LMFBR plant simulator to obtain the various combinations of annunciated alarms with progress of the events under the conditions of anomalies and accidents. The examined cases were related to the anomalies and accidents in the water/steam system of the LMFBR power plant. The simulation examination showed that each change of the pattern of annunciated alarms is specific to each anomaly or accident, and we have applied the pattern matching technique and COBWEB methods into the diagnostic reasoning using annunciated alarms. We could show the capability of these two methods to reason and focus among various candidates of causes of anomalies with gradually improved conviction degree as time passes from the occurrence of anomalies. It was also confirmed that these methods are effective in diagnosis reasoning as a way the operators are doing the diagnosis reasoning in existing plants.     

核电厂松脱部件报警案例分析

针对某核电机组在大修热试阶段多次出现松脱部件报警的现象，在关键位置临时安装探头实施了现场试验，通过分析撞击信号阵发波到达传感器的时差对采集试验数据进行定位分析。结果表明，松脱部件报警事件的起源与核电机组热试相关状态改变有关；报警触发源位于主管道冷段上（距离压力容器入口1 m处），事件是由于该位置的甩击限制器自发或间接将信号传导至主管道所导致的。此报警事件的成功诊断，不仅解决了工程实际问题，而且对现有松脱部件监测系统探测区间的完善具有重要意义。      

基于MFM和SDG的二回路系统报警分析研究

将多层流模型（MFM）和符号有向图(SDG)相结合,对核动力装置二回路系统的报警信号进行分析。利用MFM对核动力装置进行抽象、分层,运用SDG的理论对MFM中的设备符号进行逻辑分析。对二回路系统的两种故障状态进行报警仿真分析,结果表明：两种方法的结合能准确地判断源故障,清晰显示报警信号传递路径,有助于操纵员进行判断。     

基于模糊熵的核电站瞬态识别方法

为保障核电站安全经济运行,需及时准确地识别核电站出现的异常。本文通过处理关键变量的时间序列数据,对瞬态过程进行识别：利用模糊熵度量时间序列复杂度的能力,判断系统是否处于正常状态;进而利用互模糊熵度量两时间序列相似度的能力,对出现的瞬态进行类型识别。利用模块式高温气冷堆核电站仿真机的数据验证了本方法的可行性和有效性,结果表明本文方法可有效进行瞬态识别,且不需复杂的训练过程。     

Generation Method of Educational Materials Using Qualitative Reasoning

Central Research Institute of Electric Power Industry has developed a nuclear power plant educational system in which educational materials for several events are included. The system effectively teaches operators by tailoring the event explanations to their knowledge levels of understanding. The preparation of the educational materials, however, is laborious and this becomes one of the problems in the practical use of the system. Discussed in the present paper is a basic explanation generation method using qualitative reasoning. This has been developed to solve the problem.

Qualitative equations describing a recirculation pumps trip were transformed into production rules. These were stored in the knowledge base of an event explanation generation system together with explanation sentences. When an operator selects a certain variable's time-interval in which he wants to know the reasons for a variable change, the inference engine searches for the rule which satisfies both the qualitative value and qualitative differential value concerned with this time-interval. Then the event explanation generation section provides explanations by combining the explanation sentences attached to the rules. This paper demonstrates that it is possible to apply qualitative reasoning to such complex reactor systems, and also that explanations can be generated using the simulation results from a transient analysis code.     

核电站中跌落零件冲击信号的时频综合报警方法

在对核电站中跌落零件冲击信号和复杂背景噪声进行时域分析和周围图法功率谱估计的荐,提出了跌落零件时频综合报警方法。该方法首先将背景噪声带通滤波,然后对信号时域预报警及频域报警确认。     

核临界事故报警系统布置分析方法研究

本文阐述了布置核临界事故报警系统的意义和原则,分析了核临界事故可能发生的机理,初步建立了一套核临界事故情景假设分析方法。研究了最小临界事故源项计算方法以及三维剂量场分布计数的方法,采用各设备最小临界事故剂量场分布最小值等高线图的方法来从众多剂量场分布图中优化选取合适的核临界事故报警系统布置点位,以确保其可以覆盖到每个具有核临界事故风险的设备,并对核临界事故报警系统探头类型选择的原则和方法进行了分析。     

辐射监测仪器中假报警的抑制

假报警是辐射监测仪器设计中应考虑的一个重要问题。本文从假报警的产生入手,根据统计涨落的连续信号作用于设定的报警阈时的数学分析,提出了抑制假报警的设计方案。特别是偶然事件对导致决定报警与否的判定依据的不等权贡献方案,有可能实现对假报警的事实上的近于完全抑制。     

RETRAN Code Analysis of Tsuruga-2 Plant Chemical Volume Control System (CVCS) Reactor Coolant Leakage Incident

In the Chemical Volume Control System (CVCS) reactor primary coolant leakage incident, which occurred in Tsuruga-2 (4-loop PWR, 3,423 MWt, 1,160 MWe) on July 12, 1999, it took about 14 hours before the leakage isolation. The delayed leakage isolation and a large amount of leakage have become a social concern. Effective procedure modification was studied. Three betterments were proposed based on a qualitative analysis to reduce the pressure and temperature of the primary loop as fast as possible by the current plant facilities while maintaining enough subcooling of the primary loop. I analyzed the incident with RETRAN code in order to quantitatively evaluate the leakage reduction when these betterments are adopted. This paper is very new because it created a typical analysis method for PWR plant behavior during plant shutdown procedure which conventional RETRAN transient analyses rarely dealt with. Also the event time is very long. To carry out this analysis successfully, I devised new models such as an Residual Heat Removal System (RHR) model etc. and simplified parts of the conventional model.

Based on the analysis results, I confirmed that leakage can be reduced by about 30% by adopting these betterments. Then the Japan Atomic Power Company (JAPC) modified the operational procedure for reactor primary coolant leakage events adopting these betterments.     

核电厂设备状态多参数同步监测技术研究

核电厂设备稳定的运行状态以及长期的运行数据积累为实现数据驱动的设备状态智能监测建立了良好的数据基础。本文提出一种基于多参数关联关系的设备状态智能监测方法,该方法包括建模、训练和推断3个步骤,建立数据驱动的设备状态智能监测和预警模型。首先识别并分析系统设备监测参数、参数监测内容和关联关系,设计建立监测参数的关联关系模型;其次,采集并筛选设备正常运行历史数据作为训练数据,基于反向传播（BP）前馈神经网络训练关联关系模型;最终,实时采集设备监测参数实测值,并基于模型推断各参数的预测值,监控实测值与预测值的偏差,当偏差超过预定的阀值时发出预警信息。本文以某电厂热交换器和主给水泵为例进行建模并验证,结果表明本文提出的监测模型可以有效同步监测设备参数微小异常变化,预警早期异常,同时保持极低的误报警率。      

Event sequence assessment of deep snow in sodium-cooled fast reactor based on continuous Markov chain Monte Carlo method with plant dynamics analysis

Margin assessment of a nuclear power plant against external hazards is one of the most important issues after Fukushima Dai-ichi Nuclear Power Plant Accident. In this paper, a new approach has been developed to assess the plant status during external hazards and countermeasures against them in operation quantitatively and stochastically. A continuous Markov chain Monte Carlo (CMMC) method is applied and coupled with a plant dynamics analysis. In the CMMC method, a subsequence plant status is determined by the latest state (Markov chain) and the status is evaluated from the plant dynamics analysis. A failure or success of safety function of plant component is also evaluated stochastically based on a latest state of plant or hazard. A numerical investigation of plant dynamics analysis against a snow hazard is also carried out in a loop type sodium-cooled fast reactor so as to assess the margin against the hazard.