1

Design Aspects for large PQ Monitoring Systems in future Smart Grids

Jos Mara Romero Gordon, Jan Meyer and Peter Schegner, Senior Member, IEEE
monitoring campaigns. Due to the growing number of measurement sites, state-of-theart technologies in PQ monitoring reach limits and certain issues become more and more evident. Most important are: (a) Power Quality instruments (PQI) are quite expensive. In addition to the one-time costs for the instruments, the nonneglible, recurring costs for handling and maintenance during the lifetime of the instrument have to be taken into account. (b) The manufacturers tend to develop easy and straightforward tools, but with insufficient scalability. PQIs are sold out-of-box with proprietary software realizing a complete functionality range from low-level communication up to detailed statistical analysis. In case of a few dozen of PQIs analysis is still easy and timeefficient. If data volume in central database storage increases and statistical analysis needs more and more computation power the system becomes rapidly inefficient. Query times increase very fast and simultaneous analysis for multiple measurement sites lasts longer and longer. (c) The proprietary software uses in most cases a native manufacturer-dependent file format and storage methodology. Easy interoperability between different instruments of different brands or even different types from one manufacturer is virtually not available. Usually a device-independent analysis needs time and spaceconsuming export/import activities. From the DNOs point of view an easy handling of PQIs (e. g. setup) as well as an efficient data collection and analysis are mandatory factors in the decision process for a future mass roll-out. If adequately addressed, much more positive feedback is expected for future PQ monitoring in smart grids. The paper tries to give a few hints on new development strategies and is not based on imagination or prophecies, but on the experience of the authors in pros and cons of PQ monitoring systems during the last decade. III. SOME IMPORTANT CHALLENGES (A CLOSER LOOK) Today compliance of supply voltage with actual standards is rarely controlled and measured at customer level in the LV networks. However in Europe requirements exist regarding the product quality of electricity [2]. It is specified by voltage quality parameters. At the moment it is not feasible for economic reasons to monitor voltage quality at each supply point in LV or MV networks. Basically DNO cannot assure the delivery of a product that meets the specification. Usually measurements are carried out only in response to a customer complaint. The DNO acts as a service provider between the
Abstract-- Power quality (PQ) has become a non-negligible issue for DNOs. More and more sites are being monitored permanently throughout their networks covering almost all voltage levels. Due to the continuously increasing number of measurement sites, certain issues become more and more evident. PQ analyzers usually run only in combination with proprietary software and data transfer methods. The interoperability between different brands is very limited. The efficiency of data handling and data analysis decreases significantly with growing number of measurement sites. On the other hand, new smart meters including more or less PQ functionality are being installed in distribution networks. The use of their data in an efficient way helps to improve the value of these meters for PQ management in future smart grids significantly. This however will depend on the long-term prospective of their design. The paper deals with these new challenges and is focused on key issues of future PQ monitoring, its requirements and implementation possibilities. The paper addresses things like interoperability, distributed storage as well as new and smart monitoring and analysis approaches. Index Terms Power quality, smart meter, smart grid

I. ABBREVIATIONS DNO EHV LV RDBS PMD PQI Distribution network operator Extra high voltage Low voltage Relational Database System Power monitoring device (device whose main function is metering and monitoring of electrical parameters, like RMS-values or power factor) Power quality instrument (instrument whose main function is to measure, monitor and/or ascertain PQ parameters in power supply systems, and whose measuring methods (class A or class S) are defined in IEC 61000-4-30 Power quality II. CURRENT SITUATION (AN INTRODUCTION)

PQ -

oday power quality is more and more in the focus of DNOs as well as regulatory authorities. The number of permanently installed PQ-measurement equipment will increases continuously in the networks from LV distribution networks up to EHV transmission networks [e.g. 1]. Intentions and level of data analysis differ significantly between the
Jos Mara Romero Gordon is an industrial engineer within the Power Quality Department of ENDESA, Spain, e-mail: josemaria.romero@endesa.es Jan Meyer and Peter Schegner are with the Institute of Electrical Power Systems and High Voltage Engineering, TU Dresden, 01062 Dresden, Germany, e-mail: jan.meyer@tu-dresden.de

978-1-4577-1002-5/11/$26.00 2011 IEEE

energy consumers and producers and is responsible for the quality of electricity supply. The technical options of the network operators to improve the PQ are limited. Therefore they need more and specific information on the PQ situation in their whole network. The verification of product quality compliance for a large number of customers becomes more and more important in the future. Several global changes can be observed in the distribution networks. The number of distributed generators increases very fast. Usually they consist of a self-commutating inverter that emits harmonics in the higher frequency range. New operating strategies, like self-balanced, islanded LV networks are discussed. This may lead to an increased variability of short circuit power, which means a higher variability of network disturbances caused by one specific installation. Furthermore new appliances, like energy saving lamps or variable speed drive techniques, are introduced. Due to the increasing energy efficiency this new equipment tends to have a lower damping effect. Contrary it is a source of disturbances due to the power electronics that is used for input circuits. As a result of the above mentioned trends e.g. voltage control will become more and more complicated or harmonic levels will increase in a wide spectrum. The monitoring and management of power quality becomes more and more important in future networks. Most new smart meters are able to detect and record events when voltage is outside of predefined limits. Only if this violation of limits (e.g. voltage too high or too low) can raise an alarm, the control center is able to react and to take appropriate countermeasures. For millions of such smart meters central data handling and processing becomes more and more difficult. But only if the future systems are able to process and to analyze synchronous recordings from millions of meters, an improved operation of the networks becomes possible. A close link between PQ measurements and network control is an essential requirement for an improved operation of future grids considering an adequate power quality. The functionality of modern PQIs could be divided into some basic functions, which are widely standardized, and a high number of vendor individual functions and concepts, e.g. with or without human interface. In this way the vendors try to implement user specific demands as best as possible and to increase their market share. From the DNOs point of view this leads to large differences of PQI handling, e.g. how a PQI has to be setup, how the measured data have to be handled and how data has to be analyzed. Moreover look and content of measurement reports vary significantly and a comparison becomes more and more difficult. This situation is not satisfying for the DNO and leads to high acquisition and handling costs for the whole PQ measurement and analysis process. Future smart meters with integrated PQ functionality need a standardized, easy-to-use frontend for PQ analysis to improve the easiness of handling PQ monitoring significantly. The 4 challenges mentioned above are very important from the author's point of view. However, much more challenges for future PQ monitoring and management do exist. The valued reader is asked to complete the list of challenges. Any feedback is highly appreciated.

IV. NEW REQUIREMENTS (AN OVERVIEW) A. Modular and flexible functionality From an economic point of view it is not wise to integrate the full PQ functionality in each smart meter. The authors propose a multi-level approach for PQ monitoring functionality integration into smart meters. The following PQ monitoring functionality levels (PQMFL) are introduced: PQMFL-I encloses only the measurement und registration of the RMS values of the voltage and voltage interruptions. Methods have to comply with the relevant standards. PQMFLI guarantees the observation of a basic quality level. PQMFL-II includes the complete functionality of PQMFL-I. In addition it covers all main voltage quality parameters according to EN 50160. This includes flicker, voltage unbalance and voltage harmonics with a reduced frequency range. PQMFL-II functionality meets most requirements of normal sensitive consumers and should be used e.g. for the standard household customers. Currents are not considered in this level. PQMFL-III includes the complete functionality of PQMFLII. In addition different aggregation intervals (shorter and longer than the standard interval of 10 minutes) should be possible. PQMFL-III can be used for more individual measurements, e.g. for more sensitive customers with power quality contracts. Currents are still not considered in this level. PQMFL-IV includes the complete functionality of PQMFLIII. Furthermore it includes the measurement of current quality parameters. It should cover a higher frequency range to allow measurements behind the actual standards. It should be used for monitoring very sensitive consumers with high power quality requirements and producers. PQMFL-IV allows detailed analyses of causes for power quality disturbances and to identify possible sources. In general it has to be guaranteed that the functionality levels are downward compatible. It means that in level PQMFL-IV the functionality of all other PQMFL levels is included in an absolute identical way. It should also be possible to change the PQMFL functionality of smart meters without a new installation. For transparent information of customers about PQ an optional human interface module for local visualization of power quality data should be available. It would be useful that the described PQMFLs are also available in standalone PMDs and PQIs. To guarantee a vendor independent definition of the different PQMFLs a suitable way could be the addition of a new instrument class in IEC 61000-4-30. E.g. in addition to the already defined classes A and S, a new class M could be added, which covers the above described four levels of power quality monitoring functionality. B. Cost-effective integration of PQ functionality The price of PQIs is too high for an area-wide permanent power quality monitoring. Furthermore the installation of multiple devices at the supply points needs more space, more wiring, etc. and has higher power consumption. For a mass roll-out it is only cost-effective to integrate PQ monitoring functionality into the smart meters. In this way already existing hardware of smart meters could be utilized, such as

the analog signal conditioning, power supply unit and the communication interface. Especially the co-use of the communication interface and herby the use of the whole communication infrastructure for the PQ data handling creates a real comparative advantage. Altogether this enables a significant reduction of costs related to equipment, installation and operation. The price per unit with integrated PQ functionality of levels PQMFL-I and PQMFL-II has to be only slightly higher than the price of the meter without PQfunctionality. C. Open and standardized interfaces One of the biggest challenges in the field of PQ monitoring is a missing standard for the data interface and communication. Two different interfaces have to be considered (cf. 1) and 2) in the text below). For both interfaces security issues, like protection against illegal modifications or hacking of communication links are of particular importance. However they are not direct in the focus of the paper and are therefore not considered here. 1) Plug-in interface for the PQ-module The PQ functionality should be integrated into an as small as possible module that is simply plugged into the smart meter. This way the number of smart meters with PQ functionality can be adapted to the DNO's needs in a very flexible way. The smart meter has to provide a simple hardware interface for analog signals (voltages and currents) and for data communication between PQ-module and smart meter. From the view point of the DNO these interfaces should be standardized to allow easy interchangeability of PQ-modules of different PQMFLs and PQ-modules as well as smart meters of different vendors. That will reduce the maintenance cost and speeds up maintenance time. A fixed integration of PQMFL-I and PQMFL-II functionality into the smart meter may lead to a further cost reduction. However, to change the PQ functionality level needs an exchange of the whole smart meter. 2) Communication interface for data exchange Power quality monitoring produces a tremendous amount of data. Today each vendor uses proprietary interfaces to his PQIs and his own setup software and data handling software. For data exchange some vendors implement in their analysis software an interface for manual data export on file basis. This solution becomes ineffective with an increasing number of monitoring devices and is therefore only limited suitable for future mass-rollouts. Furthermore it is not unusual that even one vendor provides different interfaces and setup software for different PQIs within his own product portfolio. As a result for the DNO data handling becomes time consuming and errorprone. This is not acceptable for efficient PQ monitoring in future grids. Another drawback of this is the missing comparability of measured and aggregated PQ parameters. To guaranty a reliable and cost-efficient treatment of PQ data, international standards are necessary. This could be either standards for file formats or, which would be better, standards direct for the communication interface. Some ideas for the communication interface will be given in the next chapter.

D. Zero-configuration approaches for setup Today the setup of PQIs is a complex and time consuming process. As before mentioned each vendor uses his one setup software with his own philosophy. The DNO's staff has to be familiar with these different software and setup philosophies. This is really confusing for them and the reason for mistakes during the setup process. The situation becomes more critical with an increasing number of PQIs from different vendors. Different setups of different PQIs may finally lead to incomparable measurement data. Another problem is the correct setting of thresholds. Normally a test measurement has to be carried out to calculate these settings. PQ-modules for smart meters should not need an individual setup. They have to provide a useful and unique standard setup out of the box. Thresholds and other, more location specific, setup parameters should be self-adapting by intelligent algorithms. Depending on the local measurement conditions and based on the standard setup the PQ-modules have to find their optimal settings. Independent on individual settings the measurement data have to be comparable.. If this is not possible (e.g. due to different aggregation values), this has to be indicated by the smart meter. In general only for a few selected smart meters an individual, manual setup should be necessary. To identify the measurement location, the PQ-module should use the unique identification of the smart meter. E. Intelligent data management and distributed data storage PQ monitoring produces a huge amount of data. Up to now usually the data is stored in separated files at a central location within the intranet of the DNO. He is responsible to define a clear directory structure and unique simple-to-understand filenames. Another solution is a vendor specific central database, which offers fast data retrieval and vendor specific analysis. The integration of foreign PQIs in such databases is problematic. As mentioned before, both solutions become more and more inefficient with an increasing number of PQIs. Moreover communication infrastructure may not be able to handle this huge data amount. The transfer of the data from the PQIs to the storage system or database had to be solved by the DNO itself. Either he employs enough staff who travels around and reads out the data from the PQIs or he upgrades his digital communication network, what is not always possible and/or generates disproportional high costs. Therefore both solutions are also not adequate. In the future, the handling of PQ data has to become more flexible. A scalable, multi-level approach for data storage is proposed by the authors (Fig. 1). The blue highlighted part is the classical approach that is already realized in many monitoring systems today. Measurement data are stored nearly exclusive in one central storage site. The number of hierarchy levels varies. Depending on the number of devices and performance of communication links sometimes data of devices is directly transferred into a central database (only first and last level do exists). Nowadays the scalability of the top level is usually limited.

Scalability (1..A, 1..B, 1..C) Central storage location h-1

h-2

...

h-A

...
Data concentrator II-1 Individual PQDevice I-1 II-2 II-B

I-2

I-C

Fig. 1. Principle schema of scalable approach (PQ-Device can be a PQI, PMD or smart meter with PQ-functionality)

In future monitoring systems a more advanced approach of distributed storage at all hierarchy levels (green highlighted in Fig. 1) may be beneficial. Most of the data should be stored in the PQ-module of the smart meter itself. Only high aggregated information should be continuously transferred to a central site. Raw PQ data should only be transferred if necessary and only from selected smart meters (e.g. sensitive customers with contracts). The data transfer should be initialized by the smart meter (push-principal) following an approach that randomizes the send time of individual smart meter to avoid a simultaneous transfer request of a huge amount of meters. To optimize the traffic, different storage levels in combination with data concentrators could be introduced. In general the storages have to be organized as intelligent circular buffer. Data should be step-wise aggregated according to a specific time period schema. All information is available for a certain time (first period), e.g. to respond to or to analyze customer complaints. After the first time interval is lapsed, the storage and buffer system starts to aggregate the information depending on the elapsed storage time. Finally after the last period the data is deleted. Under normal operating conditions of the measurement system no action from the user regarding data management should be necessary. For an individual measurement campaign the DNO and/or the involved consultants should specify the requirements for data storage carefully already in the planning stage. Only by this way inadequate, non-sustainable storage solutions can be avoided. F. New approaches for data analysis Today only the preprocessing of the data (according to IEC 61000-4-30) is realized inside the PQI (e.g. different aggregation intervals, calculation of flicker or harmonics). The statistical evaluation, the data flagging and the assessment of compliance e.g. according to EN 50160 are usually carried out offline by vendor specific software (post-processing). Implementation of post-processing may differ from vendor to vendor, which increases the complexity of data analysis for the DNO. Furthermore most of the reports are related to one site only. Aggregated overall site reports or indices are not available. In general the implementation of analysis could follow a similar approach as for data storage (cf. Fig. 1). By distributing processing functionality to the different hierarchy levels as well as using scalability within each level

performance and speed of processing of large data amounts could be increased significantly. In future PQ-modules for smart meters could be more intelligent. Based on the standardized and comparable preprocessing of the measured data, e.g. according to IEC 61000-4-30, higher aggregated and easy to interpret indices have to be calculated from the measured data directly by the PQ-modules. Qualitative overall information about the PQ status may be indicated by different states corresponding to traffic lights (red, yellow, green). New aggregation methods and indices are necessary to provide such a top-down analysis methodology. Finally transparency to the customer could be achieved by presenting this highly aggregated information via an optional human interface. The smart meters with PQ functionality should no longer be simple "data collecting devices". Smart meters with PQ functionality should reach a new functional level compared to today's PQIs. A certain level of autonomy together with some control functionality could significantly improve the value of PQ monitoring to the DNOs. However a lot of research is still necessary to identify the control functionality that should have to be integrated into smart meters. Such devices could be the foundation of the power quality management in future grids. V. SOME ASPECTS OF PQ INTEGRATION INTO SMART METERS IN DETAIL A. Communication issues Currently PQIs and PMDs use a wide range of different communication protocols. Some of them are closed while others are publicly available. The paper deals with the most popular, pointing out some of their benefits and drawbacks. It should be pointed out that power quality analyzers mostly transfer files, while many already established protocols are just designed for control purposes. The protocols are used to send single items of information. In general, our experiences show that frame length is in many cases proportional to transfer speed. Next some important protocols are discussed in detail: 1) MODBUS The frame length is limited to 255 bytes, so files have to be split into very short pieces. For each frame a request is raised by a master and the slave gives a response. This process involves a delay given by the physical layer. Thus this protocol tends to be very slow, even on fast DSL or cable lines. Speed is only adequate on LANs. 2) IEC 60870-5-10x The frame lengths are also limited to 255 bytes, so results similar to MODBUS are expected. 3) DNP The frame lengths are limited to 292 bytes, so the same as for above applies. When not using TCP/IP, acknowledge is not always required, so transfer speed can be improved.

4) ZMODEM This protocol accommodates network and timesharing system delays by continuously transmitting data unless the receiver interrupts the sender to request retransmission of garbled data. Therefore it is quite robust but at the same time fast. It's really designed for transferring files, not commands. 5) FTP It is a file transfer protocol built upon TCP/IP. If frame lengths are kept long in TCP/IP, it delivers efficient and fast file transfers. 6) HTTP By using common GET, PUT and POST methods its particularly easy to transfer files to a server and get detailed success and error information. XML-RPC, SOAP and REST are just well-established examples of this approach. Nevertheless, it has to be pointed out that plain formats such as XML are not optimized for size, so there often is a length overhead that is not negligible on low speed and low reliable links. The authors suggest implementing several protocols at once. For instance, short-frame protocols could still be used for online monitoring and integration into a SCADA system. At the same time suitable protocols for file transfers shall also be provided by the PQ device. FTP protocol is starting to be widely implemented by PQ manufacturers. Usually files are generated on a daily or monthly basis. It would be advisable to provide flexible intervals for generating these files, ranging from minutes up to months. This can avoid reading one file multiple times. Other protocols such as ZMODEM and HTTP are able to limit the time window in the request, so they are more flexible than just FTP. B. Integration Issues It's neither practical nor economical feasible to change all PQIs at the same time when its technology becomes obsolete. Within DNOs it's quite common to have cohabitation between ancient and very new devices. Therefore, it's a mandatory for the DNO to be able to integrate old and new equipment in a sustainable manner. Nevertheless, our proposal is to develop a future standard covering among other things an open communication standard. After establishing such a standard, integration of smart meters with PQ-modules from different vendors with already existing PQIs isn't an issue anymore. Case study from ENDESA As it was stated before, certain communication protocols are not well-suited for file transfers. At the same time, data formats vary among devices. It is reasonable to convert this information into a common format and to treat all devices as general and uniform PQIs. Even though this intermediate transformation may involve an overhead, it allows independency of data analysis and data collection from physical devices. In order to improve transfer speed it would be advisable to use slow protocols on a local base. This approach may be

implemented in protocol converters or, in a more flexible way, by using embedded computers. These devices are cheap but powerful machines. They can be easily programmed and upgraded according to current necessities. This has been ENDESA's strategy for a set of hundreds of PQIs. By using lightweight embedded computers without movable parts, PQIs from different brands have been integrated and uniformly addressed by TCP/IP. For instance, whereas just MODBUS protocol was available, simple and robust scripts have been developed for transfer of data into the embedded computer at LAN speed. In a next step the downloaded flat files can be uploaded to a central server for data processing. It doesnt matter whether physical links are UTP cable, RS-232, RS-485 or even Bluetooth SPP, in all cases devices are easily accessible by a TCP port. The embedded computers could be part of the distributed storage and analysis hierarchy. With increasing number of devices in future the functionality of the embedded computers can be upgraded including storage and post-processing functions as described in chapter IV (cf. Fig. 2).

Fig. 2

Integration of PQ Devices at ENDESA a) Current Scheme b) Proposed Scheme

C. Data Storage Issues One problem that becomes evident with increasing number of devices is how data of PQIs is stored and handled. In contrast to simple meters or PMDs, PQIs may generate hundreds of variables at a time. Moreover, depending on their configuration, the data amount of a single PQI may vary in wide ranges.

Communication Appliance

Central System

Communication Appliance

Central System

Traditional ways of storing information in relational databases systems (RDBSs) do not seem adequate, at least without creative solutions. Many commercial software packages just rely on RDBSs with high reputation and wrongly assume the underlying design doesn't matter. It's quite common to store data in very simple tables consisting of timestamps, variable id and value as exemplarily shown in Table I.
TABLE I TRADITIONAL WAYS OF STORING INFORMATION IN RELATIONAL DATABASES (EXAMPLE)

parameter including all relevant information (e.g. 201 could represent voltage RMS, L1-N, mean). Pros and Cons of RDBSs A relational database is a complex system with many features. Moreover, not all of them are able to store vectors in an efficient and simple way. Indeed, they were not designed with this target in mind. Another important drawback of RDBSs is scalability. Even though some kinds of clustering can be developed, they usually require constant maintenance work and do not scale well at all. It's quite common to implement just vertical scaling based on more powerful computers. However, it doesn't mean that certain RDBSs are not able to handle the structures mentioned above. It only means that most probably the size of data tables will be limited to a few thousand devices at most for reasonable handling. In the authors opinion the requirements of new smart meters for PQ data management will exceed the capabilities of RDBSs significantly in case of central data storage. Distributed Key-Value Stores A new technology has emerged together with the new requirements from the WWW. It just consists of storing any information for a certain key. Vector structures fit very well in this format. These stores have been designed with horizontal scalability in mind. This means that clusters can be easily deployed by adding new nodes. The design is consistent based on redundancy and reliability concepts from the beginning. Another important advantage of this technology is data analysis. Instead of relying on expensive servers, queries may be distributed among cluster's members and collected when presented to a client. One of these implementations is called "Map/Reduce", which is a de-facto standard among large WWW enterprises. Moreover, by sharing desktop computers between common office applications and zero-cost virtual computing machines, DNOs would have access to a tremendous processing power. Virtual machines could run in the background on common desktop computers, leveraging tons of distributed storage, memory and CPU power. In fact, desktop computers are underutilized and much more could be extracted from them. Unimaginable complex operation could be undertaken in the background, even reducing centralized computing costs. D. Analysis issues As mentioned before a central data storage of all measured parameters may not be straightforward for a mass roll-out of smart meters with PQ functionality. The PQ-modules should include pre-processing as well as post-processing functions. Under normal operating conditions only highly aggregated information should be transferred to a central site. Additional a pre-defined set of events (e.g. limit violations, interruptions ) should be defined, for which the corresponding data is pushed to the central site too. The fast availability of this information is e.g. important for managing customer complaints. The DNO should always be informed of the event before a customer calls the service center. The above described data handling strategy ensures that communication infrastructure and storage systems as well as the stuff analyzing the data will not be overloaded.

Start 01/05/10 10:50 01/05/10 11:00

End 01/05/10 11:00 01/05/10 11:10

PQ-par. ID 301 301

Value 123678 123781

This design is not effective. Assuming a PQI with 400 variables means that a single measurement generates 400 rows in the table. Indexes become very large in a short time and an awful performance can be expected. Case study from ENDESA The approach developed by one of the authors is based on vector storage. It has been successfully implemented in ENDESA's network and is still fast and efficient after more than 6 years of continuous recordings. Instead of storing single variables, a complete set of them is stored sequentially as a vector. This simple design reduces the necessary space at least by factor 100 compared to the above discussed database approach. The extraction of the relevant information is a bit more difficult, but far more efficient and faster. Furthermore the approach can store information without knowing anything about its meaning.
TABLE II
VECTOR-TABLE FORMAT (EXAMPLE)

Variables vector-ID 1 2 Start 01/05/10 10:50 01/05/10 11:00

Variables vector-table PQ-parameter IDs [ 201, 202, 203, 500, 501, 503, , 790 ] [ 202, 203, 500, 501, 503, , 817 ] Time-series vector-table End Variables vector-id 01/05/10 1 11:00 01/05/10 11:10 1 Values [1234, 7891, , 1271 ] [1225, 7767, , 1267 ] ...

In fact, it's a simple and efficient way of storing time-series data. Table II shows an example how the vector-table format looks like. Each parameter-id represents a specific PQ

If the DNO intends to monitor the PQ status of a whole network with many sites online, new indices with following properties are necessary: Comparability between different PQ-parameters, sites and networks Direct relation between individual PQ-index and the margin between actual levels and specified limits (DNO planning levels or limits given by standards), which quantifies the quality reserve Clear and easy-to-understand indices, tables and graphs, especially in case of many sites While for events such aggregated possibilities of presentation already exist (e.g. DISDIP tables), for variations a new and flexible index system has to be introduced. Our proposed PQ indices are based on normalization and aggregation and are derived from the methodology in [3]. The calculation is always based on a standard compliant pre-processing (e.g. 95%-percentiles for 1 week according to EN 50160). Tab. III shows the individual PQ-indices for different 95-%-percentile of THD for 1 week (m).
TABLE III EXAMPLE OF PQ-INDEX FOR THD

Each small cube in Fig. 3 represents one individual PQ-index. All individual PQ-indices can always be presented by a large cube with the 3 dimensions: measurement site, time, PQparameter. Possible aggregation strategies can be directly derived from the dimensions of the cube. E.g. a site index aggregates all individual PQ-indices (green dimension) per week and site. Multi-dimensional aggregation is also possible. Aggregating all 3 dimensions will always result in one global index representing the minimum margin between actual levels and defined limits for all PQ-parameters (green), weeks (blue) and sites (red) considered. The global index is independent from the number of individual indices, which means independent from the number of sites and measurement time. The choice of an aggregation method mainly depends on the purpose of the analysis and can be flexible adjusted by the DNO. The aggregation algorithm works as follows: If all individual indices are positive, the smallest one is kept, if one or more of the individual indices are negative, the group index represents the sum of the negative individual indices. Case study from a German DNO The following example is taken from a project that was realized together with a local DNO in 2009/2010. It should illustrate some aspects of the proposed index system for PQanalysis of large permanent monitoring campaigns, especially the Top-Down approach. Starting with one single index the user can stepwise go deeper into the data down to the most detailed level (raw data). The monitoring system uses the classical approach with central data storage and analysis. During the years 2007 to 2012 about 110 revenue meters in MV network has to be replaced. The DNO decided to buy a revenue meter with basic PQ functionality. The PQ option costs about 30 per meter. It should be noted that the functionality with actual firmware is small and complies only partly with class B according to IEC 61000-4-30. The measurement data is transferred to a central storage by using the existing metering infrastructure. For a device-independent and automatic PQ analysis the aforementioned index system was applied. The example analysis is based on the data of 22 meters for the time interval from July to December 2008 based on 10minute-mean values for selected harmonics, voltage THD, supply voltage magnitude and flicker. The single indices are calculated using the 95-%-percentiles based on weekly intervals according to EN 50160.
TABLE IV REDUCTION OF DATA SIZE (BOTTOM UP)

m - Actual value g - Limit acc. to EN 50160 rE - Individual index The colour code is a more qualitative representation of the existing margin. In this example green is used for a reserve larger than 50 %, orange for a reserve between 0 % and 50 % and red if the limits are exceeded and no reserve exists at all. The individual index (rE) is direct proportional to the existing margin (e.g. rE = 0.53 means a reserve of 53 %) and can be directly compared with the individual indices for other PQparameters. For an assessment according to EN 50160 82 individual PQindices have to be calculated for each site and each week. Compared to the size of the measurement data this is already a significant decrease, but considering many sites and weeks a further aggregation is useful.

(e ti .g m .w e ee ks )

Raw data points: 10-minute-mean values Normalization: Single PQ indices Aggregation 1: Site indices Aggregation 2: Net index / Global index

app. 14 million values app. 13700 values app. 570 values 1 value

Fig. 3

Schematic model of PQ index system

The net index (global index) rN = 0.17 represents the minimum of all individual PQ-indices. It means that for each considered PQ parameter at each of the selected sites for each selected week at least a reserve of 17% exists compared to the limits of

EN 50160. On the other hand the best-performing site has a reserve of 69% (maximum individual index) and more than 50% (median value of individual indices) have a PQ reserve better than 56%. These values are easy to understand and usually enough for routinely reporting e.g. in control centers within the DNO. The presentation of weekly site indices including trend information in a GIS may also be a suitable way for integration of more detailed, easy-to-interpret PQ information into the control system. Another graphical way for data presentation is shown in Fig. 4.
42
50 37

0.65 0.6 0.55 0.5 0.45 0.4 0.35 0.3 0.25 0.2 Messort Measurement site 5 10 15 20

52 Kalenderwoche Calendar week 47 45 42 37

40 52

network as a whole. PQ monitoring should become actively integrated into the day-to-day network operation business and should be considered in new concepts for smart grids. A possible way could be the introduction of flexible multilevel strategies for data storage and data analysis. This means not only a hierarchical structure from device level to site level, but also a horizontal scalability for each level. Storage solutions may use many different storage locations within the IT infrastructure of a DNO with high redundancy. The utilization of many computers (parallel computing approach) can enable a powerful analysis of the huge datasets in reasonable time. Furthermore the introduction of appropriate standards (e.g. for communication issues) can further increase cost-effectiveness of PQ monitoring in the future. Last but not least it is pointed out, that the ideas presented in the paper are intended to be impulses for a constructive discussion of the long-term trend in PQ monitoring and the future needs. Comments and opinions on the presented ideas are highly appreciated. VII. REFERENCES
[1] [2] [3] Council of European Energy Regulators (CEER), 4th Benchmarking Report on Quality of Electricity Supply 2008, December 2008. EN 50160 Voltage characteristics of electricity supplied by public electricity networks, 3rd edition, 2010. V.J. Gosbell, B.S.P. Perrera, H.M.S.C. Herath, Unified power quality index (UPQI) for continuous disturbances 10th International Conference on Harmonics and Quality of Power (ICHQP), vol 1, p.312-316, 6-9 Oct. 2002.

47 42 35 37

Fig. 4

Plot of site indices for 22 sites and 19 weeks (Colors are not related to color code in Tab. 3 !)

VIII. BIOGRAPHIES
Jos Mara Romero Gordon is an industrial engineer within the Power Quality Department of ENDESA, the largest electric utility in Spain and part of the ENEL conglomerate. His expertise covers measurement, monitoring, and analysis of the whole range of power-quality disturbances. He received his Master of Science at the University of Seville, School of Engineering. Previously to his activities in ENDESA he had worked in the railway and mining sectors. e-mail: josemaria.romero@endesa.es Jan Meyer studied Electrical Power Engineering at Technische Universitt Dresden (Germany), where he received the Dipl.-Ing. He received the Ph.D. with a thesis on the statistical assessment of power quality in distribution networks. Since 1995 he is with the Institute of Electrical Power Systems and High Voltage Engineering at the Technische Universitt Dresden. He is member of several working groups. His fields of interest are all aspects of design and management of large power quality monitoring campaigns and theory of network disturbances, especially harmonics. e-mail: jan.meyer@tu-dresden.de Peter Schegner studied Electrical Power Engineering at the Darmstadt University of Technology (Germany), where he received the Dipl.-Ing. After that he worked as system engineer in the field of power system control and became a member of the scientific staff at the Saarland University (Germany), receiving the Ph.D. with a thesis on the earth-fault distance protection. Then he worked as head of the development department of protection systems at AEG, Frankfurt A.M., Germany. Later he became a full Professor of Electrical Power Systems at the Technische Universitt Dresden (Germany). e-mail: peter.schegner@tu-dresden.de

Each colored square represents a single site index per site and week. Aggregation applies for PQ-parameters only (cf. Fig. 3, green axes). From this plot sites with worse PQ reserve (blue ones, sites 12, 13) or best PQ reserve (dark red ones, sites 5, 17) can be easily identified. Moreover trends or abrupt changes in power quality can be clearly recognized by the colored pattern. Much more possibilities of numeric and graphic presentation with different levels of detail are possible, but due to limited space they are not shown in this paper. The application of the described methodology allows a higher efficiency of PQ monitoring within the DNO. While control center may be responsible for routine monitoring, PQ department is only noticed in case of problems (events, exceeding limits or decreasing trends). VI. CONCLUSIONS Monitoring PQ will become an important part of the operation and management in future grids in all voltage levels. The biggest challenge is its efficient implementation. Costs for PQ devices and for their operation (including maintenance, communication links, IT infrastructure, data analysis ) have to adequate. Future PQ monitoring systems have to provide benefits for the DNO. A simple accumulation of a huge amount of data, which becomes more and more difficult to handle, is definitely not enough. On the other hand the individual analysis of single sites has to be superseded by global approaches covering the