Enhanced Models And Methods For Power System Fault Diagnosis Utilizing Temporal Information Of Alarm Messages

When a fault occurs in a given power system,a flood of alarms from different monitoring systems will be uploaded to the dispatching center,and analyzed by the dispatchers.However,the amount of alarms is usually huge especially for multiple faults with protective device malfunctions.Under great pressures,it is difficult for the dispatchers to process information and judge the fault location accurately in a short period of time.In addition,the existence of missing/false alarms and wrong timestamps may confuse the stressed operators,giving rise to more difficulties in identifying fault section(s)and taking appropriate actions quickly.Therefore,it is very demanding to develop an effective and accurate method for power system fault diagnosis,in order to ensure the security and stability of the power system associated and to improve the reliability of power supply.The main issues of the existing models for fault diagnosis include:over reliance on the state information of protection relays(PRs)and circuit breakers(CBs)after the fault,limited fault information source,low information redundancy;low fault tolerance,may obtain false diagnosis results for complicated faults,i.e.,multiple faults involved with malfunction of PRs and CBs and false/missing alarms.Given this background,based on the existing artificial intelligence based models and methods,the uncertainties during the fault process are considered to improve the fault tolerance capability by developing a more reasonable and accurate model and taking account of temporal information in the alarm series.Some research results are obtained as follows:1)An overview is made on fault diagnosis in power systems.The aim and functions of the fault diagnosis system are first analyzed.The fault information utilized by the fault diagnosis system is then studied.The classification and sources of different kinds of information are identified,and the characteristics of the fault information source are introduced.A universal layered structure of power system fault diagnosis is constructed,in order to reasonably use the fault information according to the fault condition.2)The Shannon's information theory is employed to deal with uncertainties during the fault diagnosis.First,a communication channel model is presented for the fault diagnosis problem,and the action logics of PRs and CBs as well as the relationship among PRs,CBs and alarms are expressed analytically.A new optimization model for power system fault diagnosis is next developed with the minimization of information loss as the objective,and can well accommodate uncertainties from various sources.Then,an improved genetic algorithm is employed to solve the optimization problem.Finally,actual fault scenarios of Zhejiang power system are served for demonstrating tolerance capability and diagnosis speed of the developed method,and the diagnosis time is less than one second for complicated fault scenarios.The presented method could be used for on-line fault diagnosis of actual power systems.3)An effort is made to develop a novel analytic model for power system fault diagnosis so as to well utilize redundancy and temporal information of alarms.An alarm preprocessing procedure is proposed to preserve such properties without impairing diagnosis accuracy and efficiency.A new fault hypothesis is established to systematically take account of malfunctions of protective devices,missing and distorted alarm messages,and timestamp errors in the proposed model.An objective function is developed for fault diagnosis,and could accommodate various protection schemes and bus configurations.Based on the presented fault diagnosis method,a software package is developed.Several fault scenarios occurred in an actual power system in China are served for demonstrating the correctness and efficiency of the developed software package.4)A fuzzy Petri net model capable of handling temporal constraints is presented and applied to power system fault diagnosis.The places in the Petri net are assigned time attributes so as to take time-delay constraints of protective relays and circuit breakers into account,and a fuzzy weighted algorithm employed to improve the fault tolerance capability.The developed method could identify wrong and missing alarm messages,as well as timing inconsistencies,and evaluate the operating performance of protective relays and circuit breakers.The IEEE 39-bus power system is finally served for demonstrating the developed fault diagnosis method.Compared with some existing fault diagnosis methods,the developed method has stronger fault tolerant capability,and can be applied to on-line fault diagnosis of large-scale power systems.5)A reliable fault-tolerant event correlation analysis method is presented,so as to discover the correlation between fault hypotheses and received alarms,and hence make fault diagnosis.In the proposed method,a real-time fault diagnosis module(FDM)is developed,based on the temporal constraint network(TCN),to fully exploit the rich information in alarm timestamps.First,a hypothesis generation scheme gathers all the candidate hypotheses related to the received alarms;next,a feature selection scheme collects universal correlation features for each hypothesis-alarm pair;then,an ensemble model of extreme learning machines(ELMs)is employed to classify the hypothesis-alarm pairs using the selected features;finally,an expert system is developed for comprehensive evaluation of the ELM ensemble outputs.The TCN-based feature selection scheme detaches the FDM from specific power system topologies and protection placements,providing the method with generalization capability.Test results for fault cases in the IEEE 14-bus power system and an actual power network in China demonstrate that the developed method has high computational efficiency and diagnosis accuracy,and is suitable for online applications.Finally,several conclusions are obtained based on the research outcomes,and directions for future research indicated.