Complex Event Processing in Power Distribution Systems: A Case Study

Debnath Mukherjee Deepti Shakya Prateep Misra Tata Consultancy Services Limited Plot A2, M2 & N2, Sector V, Block GP Salt Lake Electronic Complex, Kolkata, India 700091 {debnath.mukherjee, deepti.shakya, prateep.misra}@tcs.com
The authors are not aware of any reported fault categorization and detection work based on CEP technology. CEP platforms for phasor data concentrators and stream processing have been reported in [2]. However it does not mention specific use of CEP for fault detection and classification. Analysis of cause of faults and fault location is being considered in future work.

Abstract
Complex Event Processing (CEP) is an emerging discipline. This paper focuses on the application of CEP for fault detection and classification in 11kV radial distribution system using data collected from a Phasor Measurement Unit (PMU). The analysis has been done by monitoring the electrical quantities in the 11kV radial distribution system simulated using Matlab Simulink. The PMU is placed at the substation and transmits data to the Command and Control Center. The data is analyzed to identify the signatures of different types of faults and based on that rules have been designed to categorize them. An architecture stack has been designed based on a commercial CEP product (Tibco Business Events) to implement the fault detection. In this paper we present the architecture, the 11kV distribution system simulation and share our experience about fault detection. The paper outlines the categorization and analysis of various types of faults like Single Line to Ground (SLG) fault, Double Line to Ground (DLG) fault, and Three phase (3) fault using CEP software. Thus it shows a real life application of CEP software for fault classification with low computational time.

Phasor Measurement Units

PMUs are widely used in transmission networks for applications on Wide area monitoring and protection, Voltage instability analysis and prediction. Suitably engineered PMUs can also be integrated in to distribution level IEDs (Intelligent Electronic Devices). The paper examines such distribution level PMUs for fault detection and classification.

Problem Statement

Keywords
Complex Event Processing, Phasor Measurement Units, Architecture, Simulation, fault detection, fault classification

Introduction

Complex Event Processing (CEP) is an emerging technology which can be used to detect interesting patterns amongst events being received in real time. There are two related fields: Event Stream Processing is concerned with time-ordered sequence of events. CEP is of wider scope which includes partially ordered set of events known as event clouds [1]. In this paper we describe how CEP can be used for detection and classification of faults in 11kV distribution system using PMU. Fault detection and clearing is important for any network from safety and reliability perspective. In particular, we describe techniques that can be applied to distribution systems that monitor phasors using Phasor Measurement Units (PMU). The unique contribution of this work is efficient fault detection and categorization using CEP technology.

The problem under consideration is detection of occurrence and type of faults in the 11kV radial distribution system. This distribution system is simulated using Matlab Simulink. Three phase, 13 bus balanced radial distribution system is simulated in which a PMU is placed at bus 1. The PMU is simulated using the three phase measurement block of Simulink library. Appendix A shows the system details. Table 4 mentioned in appendix A shows the line and load data for the test system. The system is simulated at frequency of 60Hz. In this simulation, different types of faults are created at different locations. At bus 1, three phase voltage and current waveforms are sampled with sampling frequency of 1kHz, in form of magnitude and phase angles. The PMU sends this data to Command and Control station where it is processed for fault detection and classification using CEP software. The format of the transmitted synchronized data follows the standard mentioned in IEEE Standard C37.118 [3]. PMU aggregates the signals i.e three phase voltage and current in terms of magnitude and phase angles. These data are then sent to a Command and Control station where the data is processed for fault detection and classification using CEP software. The format of the transmitted synchronized data is similar as mentioned in [3]. The captured data is then used to detect the type of fault occurred in the distribution system.

Using PMUs to detect faults

Related Work

Let A, B, C and G represent phase a, b, c and ground. Different types of fault that can occur in the distribution system are: Single Line to Ground Fault (SLG) Line to Line fault Three phase Fault (L-L) (3L)

Current fault detection technology is based on current transformer based relaying. The relay co-ordination ensures that the impacted zone is minimum. The shortcoming of this is relay setting and relaying accuracy may differ, sometimes not providing optimum fault clearance.

Double Line to Ground Fault (DLG) i.e. ABG, BCG and CAG. i.e. ABC.

Whenever any one of these faults occur in the system, the voltage and current of a phase will show change in terms of decrease and increase in magnitude from its normal values respectively. If the fault occurs the phase angle for that particular voltage also show variation from its normal value. Similarly, phase angle for that phases current can also show changes from its normal value. These changes can be used as the key parameters for fault classification. As per [4], thresholds have been decided for these parameters for fault classification. For voltage, if 1.1pu >|V|> 0.8pu else if |V| < 0.8pu else if |V| > 1.1pu in system. then the system is in normal state then there is a Fault in System. then capacitor switching or transients occurs

The first case shows the variation in voltage and current parameter for phase A with respect to change in fault resistance from low to high value at one particular location i.e. at bus-25. While the second case shows changes in voltage and current parameter for phase A measured at bus 1 for change in fault location i.e at bus 21 and bus 31. Case I: Study of the effect of variation in fault resistance at end of bus-25 of the system shown in Appendix A. This case considers that AG fault occurs at the end of bus 25. Faults with different fault resistances are created at this location and their effects are analyzed at bus-1 using PMU. Table.1. shows the variation in three phase voltage and current data monitored by PMU at bus 1, with the change in fault resistances. It is observed that for low fault resistance the voltage magnitude is lower than 0.8p.u.but as the fault resistance is increased it has reached above this value. Thus for high resistance faults, the voltage magnitude remains within the normal range. Table 1 also show changes observed in phase angle (VA) for phase A for low as well as high resistance faults. The change observed is dependent on the instant of occurrence of fault as well as on the fault resistance. Thus, Table-1 shows that phase angle for phase A voltage changes even for high fault resistances but stays within normal limits as this parameter is more dependent on the instant of occurrence of fault rather than the severity of fault. Fig.1. show the variation in three phase voltage magnitude when high resistance faults occur at end of bus 25.
1 0.8 0.6 0.4

where |V| represents the phase voltage magnitude and pu stands for per unit. So, when the system is running under normal condition, the voltage will be less than 1.1pu and will be greater than 0.8pu. When a fault occurs, the voltage magnitude will be less than 0.8pu. Voltage magnitude show changes for low fault resistances, but if there is a high resistance fault in the system then the voltage magnitude may remain in the normal band i.e. greater than 0.8pu; but the current for that particular phase will show noticeable change from its normal value. Similarly, the phase angle change for voltage is considered to be normal for a change of 1 degree while for current, phase angle change of 10 degrees is acceptable. Two case studies for SLG fault on phase A, are discussed in this paper. When SLG fault on phase A occurs in the system, the magnitude and phase angle for phase A voltage and phase A current will show changes from their normal values. This change depends on the severity of the occurrence of the fault. The other two phases i.e. phase B and C will also suffer changes but the values will remain within a normal range. Two cases for SLG fault i.e. AG are considered. These cases discuss the change in voltage as well as current waveform monitored by PMU placed at bus-1 with respect to change in fault resistance as well change in fault location.
Table.1. Magnitude and phase angle variation in phase A Voltage and Current with respect to fault resistance when SLG occurs at bus 25
C rre t (p )) u n .u

V lta e (p ) o g u

0.2 0 -0.2 -0.4 -0.6 -0.8 -1 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

0.1

Time (sec.)

Fig.1. Three phase voltage waveform when AG fault occurs at instant 0.0376sec and is cleared at 0.0708sec with fault resistance of 50 at end of bus25.
5000 4000 3000

Fault Resistance (ohm) 1 10 50 100 150

Voltage Phase voltage (VA) (VA) (pu) 0.73 0.86 0.95 0.95 0.96 (degrees) -8.73 -9.96 -5.54 -4.32 -3.87

Current (IA) (pu) 5709.4 4100.5 2090 1654.9 1500.6

Phase current (IA) (degrees) -59.98 -37.92 -25.14 -25.48 -26.28

2000 1000 0 -1000 -2000 -3000 -4000 -5000 0 0.01 0.02 0.03 0.04 0.05 0.06 Time (sec.) 0.07 0.08 0.09 0.1

Fig.2. Three phase current waveform when AG fault occurs at instant 0.0376sec and is cleared at 0.0708sec with fault resistance of 50 at end of bus25.

Similarly, three phase current is also monitored at bus 1 using PMU. Table-1, show changes in magnitude as well as change in

its phase angle for phase A current observed at bus 1, with respect to change in fault resistance. It is observed that magnitude of phase A current shows significant change even for high resistance faults. Similarly, its phase angle is also showing changes from its normal value for low as well as high resistance faults. In Fig.2. Phase A current is showing abrupt change from its normal value for AG fault, while the currents for other two phases remains within normal range. Similar studies can be done for Double line to ground fault, Line to line fault and three phase fault. Case II: Study of variation in fault location as well as fault resistance at end of bus-21 and bus-31. AG fault is simulated at two different locations. First a fault was simulated at distance of 2.5km from monitoring end i.e. at the end of bus 21 then cleared. Then a second fault was simulated at 11km i.e. at the end of bus 31. Table.2 and Table 3. show the change in voltage and current parameters for phase A monitored at bus 1.
Table2. Magnitude and phase angle variation in phase A Voltage with respect to change in fault resistance as well as fault location when SLG occurs at bus 21&31 Fault Resistance (ohm) 1 10 50 100 150 Voltage (VA1) (pu) 0.51 0.86 0.95 0.95 0.96 Phase voltage (VA1) (degrees) -20.99 -14.87 -5.54 -4.32 -3.87 Voltage (VA2) (pu) 0.79 0.87 0.94 0.95 0.96 Phase voltage (VA2) (degrees) -6.83 -8.23 -5.38 -4.28 -3.85

Table3. Magnitude and phase angle variation in phase A current with respect to change in fault resistance as well as fault location when SLG occurs at bus 21&31

Fault Current Resistance (IA1) (ohm) 1 10 50 100 150 (pu) 10646.6 5334 2175.7 1682.1 1515.5

Phase current (IA1 ) (degrees) -62.71 -29.54 -22.18 -24.1 -26.53

Current (IA2) (pu) 4510.6 3613.8 2057.6 1648.7 1498.5

Phase current (IA2 ) (degrees) -58.34 -40.72 -25.97 -26.42 -26.53

Table.3. shows the current magnitude and its phase angle variation for phase A only when AG fault occurs at two different locations. Let IA1 and IA2 represent the current magnitude monitored by PMU at bus 1 due to faults located at bus end 21 & 31 respectively. Similarly, IA1 and IA2 represents the change in phase angle for phase A current respectively. It shows that IA1 & IA2 show severe changes for low resistance faults at both locations. However, it is observed that for same resistance faults, IA1 will be of higher value than IA2. Thus, change in current magnitude is more prominent when fault occurs near to monitoring location and its magnitude gets reduced as the distance of occurrence of fault increases from the monitoring end. It was observed that at two locations i.e. at bus 21 & bus 31, the PMU placed at bus 1 recorded comparatively less change in magnitude for IA1 & IA2 when the high resistance fault occurred compared to low resistance fault. Similarly, change in phase angle (IA1 & IA2 for current show variation with respect to change in fault resistance as well as fault location. The severity decreases with the increase in distance of fault location with respect to bus 1 as well with the variation in fault resistances. Similarly, cases can be studied for double line to ground fault, Line to line fault and three phase fault. For high resistance faults, three phase voltage magnitude may remain in normal limit. In that case three phase current magnitude can be observed for fault classification. Thus, if the current magnitude for a phase is higher than defined normal limit then it indicates a SLG fault for that particular phase. Similarly, if two phase or three phase current magnitude are higher than normal limit its DLG fault or three phase fault respectively. The phase angle change for three phase voltage and current can be examined for additional information. These parameters can be observed to know the instant of occurrence of fault. If the fault occurs at zero crossing, the phase change will be near to zero on the other hand this change will be maximum if the fault occurs at the peak value. These values can also be used to distinguish among the DLG fault and L-L faults.

Table.2. shows the comparison between the change in voltage magnitude and its phase angle at bus 21 and 31 with respect to increase in fault resistance for SLG fault at phase A. Let VA1 and VA2 represent the voltage magnitude monitored when the fault occurs at bus end 21 & 31 respectively. Similarly, VA1 and VA2 represents the phase angle for phase A voltage respectively. As it is AG fault, the other two phases may suffer change from its normal values in terms of magnitude and phase angle but within normal range. It is observed that for low fault resistances the voltage dip is more severe (as measured by PMU) when the fault occurs at bus end 21 in comparison to fault occurring at bus end 31. However if a high resistance fault occurs at these two locations, the voltage magnitude for phase A remains higher than 0.8p.u as measured by the PMU. Considering that each fault condition occurs at the same instant at both locations, change in phase angle for voltage A decreases with increase in fault resistance. It is also observed that for same fault resistance, change in phase angle (VA) decreases as the fault location increases from monitoring bus 1.

Solution

This section is organized as follows: Section 6.1 gives the solution overview and Section 6.2 onwards gives the details about the different modules of the solution.

fired. This is known as Run to Completion in Tibco Business Events. One class of CEP software, such as Tibco Business Events [5,6], includes rule-based inference engines which are enhanced to support event streams. Tibco Business Events receive event streams through the Tibco Rendezvous bus software which enables event producers to send events. Some CEP software such as Esper, Streambase and others use continuous queries to detect complex events. Continuous queries are SQL-like queries which are registered with the CEP server before the real-time data comes through, and act upon the real time data producing output with low latency. Some software such as Tibco Business Events offer both queries and rules. The Tibco Business Events CEP Engine was used for this problem. Each type of fault was mapped to a Business Event rule: There were separate rules for line A to ground fault, line B to ground fault, line C to ground fault., Double line to ground faults and three phase fault. The architecture diagram of the solution is shown in Figure 3.

6.1

Solution Overview

In this section we discuss the solution architecture for implementing fault detection using PMUs. The architecture diagram of the solution is shown in Figure 3. The PMUs are placed at the sub-stations from where its messages are communicated to the command and control station (CCS). The latter receives messages from multiple PMUs and processes them to determine if there is any fault. The message sent from the PMU has the following fields (only those required for fault detection are shown): 1. 2. 3. Voltage magnitude and phase angle for each of the phases, Current magnitude and phase angle for each of the phases. Timestamp of measurement

From these message attributes, the pattern that a fault has occurred has to be detected. The pattern for a Single Line to Ground Fault (phase A to ground) is shown below (IF Line to Neutral Voltage for Phase A < 0.8 pu OR IF Phase A current exceeds the current threshold) AND IF Other phase (B,C) voltage and current magnitudes are within normal limits THEN IT IS A SINGLE LINE TO GROUND FAULT IN PHASE A The logic consists of two parts: Part 1 implements the logic that either the line A voltage is below normal threshold OR the line A current is above normal threshold. Part 2 implements the logic that other phase (B and C) current and voltage related parameters are within normal limits. The two parts are joined by an AND operator. Similar rules hold for the double line to ground fault and three phase faults: for double line to ground faults, two of the phases will show abnormal characteristics and for three phase faults, all the three phases show abnormal characteristics. The logic above can be mapped to rule-based systems. A rule consists of two parts: an IF part (known also LHS) and a THEN part (known as action). The IF part contains a condition, which when satisfied fires an action which is contained in the THEN part. Rules are evaluated in cycles: a number of input conditions may be satisfied in the first cycle which then fires actions which may change the value of some variables that are considered in the IF part of some rules. In the next cycle, some more rules may get triggered because some variables have been changed as mentioned before (in some cases same rules already fired before may be fired again if the variables in the condition part of the rule have different values than before as a result of executing some actions ). In this way evaluation is done in cycles until no new rules are
Substn

Display

CEP Server Command and Control Station

Bus Data Collector

Substations
Substn

Fig.3: Architecture of fault detection

The architecture of the system consists of the following building blocks: At the substation level, a PMU sends a message to the command and control center At the command and control station, the message from the PMU are sent to a data collector which is an independent server program The data collector packages the PMU message as an event and sends to the CEP server (Tibco Business Events in this case) via the message bus (Tibco Rendezvous). The CEP Server executes the rules and sends to the display and also to pager/mobile alerts.

It also generates events that are sent to external systems that can take automated action based on the event.

In the following sub-sections we explain the various components of the architecture.

6.2

Data Collector

6.5

Prototype of Solution

The Data Collector is a Java program that adapts the input signal from the PMU to a format suitable for transmitting to the CEP Server via Rendezvous. The Data Collector takes the input signal from the PMU and converts it to the event format. The event format is the same as that of the PMU message: consisting of the voltage, voltage phase angle, current and current phase angle for each of the phases and the timestamp. The formatted data is sent to the Tibco Rendezvous bus in the Tibco-specific format. The Rendezvous bus is a daemon that runs on each machine participating in distributed computation. While in the prototype both Data Collector and CEP Server were hosted on the same computer, if the Data Collector and CEP Server are hosted on separate machines, the Rendezvous daemon need to be running on both the machines.

6.3

CEP Server

The CEP server hosts the rules for detecting faults. There are three rules for line to ground faults (for the three phases), three rules for double line to ground faults (for AB to Ground, BC to Ground and CA to Ground faults) and one rule for the Three Phase Fault case. Please see Section 6.1 for the structure of the rules. One of the requirements of the CEP Server was that the display should not show the same type of fault if it occurs in consecutive cycles; and the display should indicate when a fault is cleared. We found that this required a Concept object to be used. We now explain Business Events Concepts and how it was used in our prototype. A Concept in Business Events is similar to a class in objectorientation theory except that methods cannot be defined on it. A Concept can have attributes and it can inherit from other Concepts. Concepts can be instantiated just as Classes can be instantiated similar to objects in Object Oriented Design theory. A Concept called PMU with two fields was defined: a Boolean field FAULT (which is true if a fault occurred in the last cycle) and a String field FAULTTYPE (indicates the type of fault observed in the last evaluation). An instance of this Concept was initialized for each PMU during application startup. Using this concept and its two fields (which hold the state of the PMU) the requirement that the display should not show the same type of fault if it occurs in consecutive cycles, was satisfied. Additionally if no fault occurred, a no fault rule was designed to reset the fields of the Concept instance and also indicate when a previous fault is cleared.

A prototype of the solution has been built. A MATLAB simulation was used to generate data required for fault classification. An 11kV balanced distribution system with one radial feeder is simulated using Matlab Simulink. In the simulation, a PMU is placed at bus 1 to monitor the three phase voltage and current waveforms. The PMU is simulated using Simulink library. Three types of fault were simulated (Line A to Ground, AB to Ground and Three Phase) with the change in fault resistance as well as fault location. These fault conditions cause change in phase voltage and current in terms of magnitude and phase angle. This change is captured by PMU at bus 1. The data is sampled at frequency of 1000Hz. Then average per cycle is computed for voltage as well as current waveform, such that one data sample per 60 Hz cycle is output. The files of data generated by MATLAB were read by the data collector and sent to Rendezvous from where Tibco Business Events picked up the events and generated the faults and fault outputs. The prototype has been designed keeping in mind that network communication will be used to transfer data between the PMU and the data collector.

Results and Experience

The experience of using MATLAB to simulate faults is explained in Section 5. The key results and experience of using CEP software are explained below: How using rules simplified the development and enabled us to develop a prototype in a short time The CEP software demonstrated two features that simplified development: 1) 2) Definition of events and Concepts using a userfriendly IDE made it easy to develop the solution. Presence of rule definition capabilities made developing the solution easy. Rules can be defined by non-technical business users also.

Some amount of state maintenance is required unlike traditional rule approaches Simply using rule engines is not sufficient for this problem. Some amount of state has to be maintained about each PMU preferably in memory as explained earlier (see Concepts in Section 6.3). Simplicity of messaging framework during connection setup phase The CEP server was integrated with the Rendezvous messaging infrastructure. We found the Rendezvous messaging API simpler than standardized API like JMS (Java Message Service) during the setup phase, because as compared to obtaining a Connection, ConnectionFactory and Session, there was only one call to Tibrv.open().

6.4

Display

The design can support both command line output and SCADA (Supervisory Control and Data Acquisition) human machine interface.

The performance of Fault Detection using CEP Server We carried out an analysis of the performance of the fault detection system. Fault data was fed in at a high rate to the CEP Server by the data collector. Each fault was detected in less than a millisecond. 3000 faults were detected in 1422 milliseconds, giving an average fault detection time of 0.474 millisecond per fault. The actual fault detection rule execution time is expected to be less than 0.474 ms because the latter includes the time for the client program to post to the Business Events engine via Rendezvous. The hardware used was Intel Core 2 Duo 2.33 GHz CPU with 2 GB RAM.

BUS 1with PMU BUS 20 BUS 27 BUS 28 L 27 BUS 29 BUS 30 BUS 21 BUS 22 BUS 23 BUS 25 BUS 26

BUS 31 BUS 24

Future Work
Fig.4. 13 bus 11kV balanced distribution system. Table.4. Line and load data for the distribution system End Buses of Lines Bus x Bus y Line Length (kms) Real Power (kW) 1 20 21 22 23 22 25 20 27 28 27 30 20 21 22 23 24 25 26 27 28 29 30 31 1 1.5 3 5 2.5 3 4 1 1.5 2.5 4 5 108.0 94.0 81.0 108.0 108.0 102.0 41.0 108.0 162.0 68.0 68.0 95.0 Reactive Power (kvar) 52.0 46.0 39.0 52.0 52.0 50.0 20.0 52.0 79.0 33.0 33.0 46.0 Load at Bus y

In this work, only one PMU is placed in the distribution system. This work can be extended by placing more PMUs optimally in the distribution system for increasing the reliability of the solution. This fault classification algorithm can further be extended to fault location algorithm. Lines to line faults are not covered in this paper and will be considered in future work. The current work is based on a set Voltage and Current level for fault determination. However, features can be incorporated to detect continuous Under voltage conditions and Overloaded conditions in the network and distinguished from high resistive faults. Radial network is assumed. Algorithm could be modified to adapt to in-feeds from local generation, alternate switching routes.

Acknowledgments

The authors would like to acknowledge the contributions of Sumit Kumar Ray, Narayanan Rajagopal, and Ranjeet Vaishnav of TCS for their valuable contributions in this work.

10 REFERENCES
[1] David Luckham, The Power Of Events: An Introduction to Complex Event Processing In Distributed Enterprise Systems, Addison Wesley [2] Downloaded from http://openpdc.codeplex.com/wikipage?version=63 on 24th February, 2010 [3] IEEE Standard for Synchrophasors for Power Systems, IEEE Power Engineering Society, pp 1-57, March,2006 [4] IEEE Recommended Practice for Monitoring Electric Power Quality, IEEE Standards Coordinating Committee Report, pp. 1-70, June,1995. [5] TIBCO Software Inc., Tibco Business Events User Guide [6] TIBCO Software Inc., Tibco Business Events Language Reference

11 APPENDIX-A
The data corresponding to the practical 13 Bus Distribution System is tabulated in Table.4. Base kVA: 1000; Base kV: 11 Conductor Type: ACSR Line Resistance: 0.0086 p.u/km Line Reactance: 0.0037 p.u/km