Data
Management Has Become the Most Complex Aspect of System Design and
Implementation
The amount of data being
collected or capable of being collected is increasing exponentially. This rapid
expansion of data retrieval results from the fact that more field devices are
being installed and that these field devices are becoming more
"intelligent" both in what power system characteristics they can
capture, and also in what calculations and algorithms they can execute which
result in even more data.
As distribution automation
extends communications to devices on feeders, as substation automation expands
the information available for retrieval by substation planners, protection
engineers and maintenance personnel, and as more power system asset information
is stored electronically in Geographical Information Systems and AM/FM systems,
even more varieties and volumes of data will be need to be maintained and
managed.
Data management is a
complex issue, encompassing many aspects of data accuracy, acquisition and
entry, storage and access, consistency across systems, maintenance, backup and
logging, and security. These are discussed in the following sections.
Data management must address a complex set of issues which include the following services:
1. Validation of source data
and data exchanges
2. Ensuring data is up-to-date
3. Management of time-sensitive
data flows and timely access to data by multiple different users
4. Management of data
consistency and synchronization across systems
5. Management of data formats
in data exchanges
6. Management of transaction
integrity (backup and rollback capability)
7. Management of the naming of
data items (namespace allocation and naming rules)
10. Data Entry
11. Data
Storage and Access Management
12. Data
Consistency across Multiple Systems
13. Database
Maintenance Management
No single cross-industry technology addresses all of these issues, but multiple solutions are available for different aspects. These include the best practices discussed in the following sections.
Validation of data should be performed at each step in a process, and “flags” should be included with the data to indicate the “quality” of the data. This quality flag will normally have to be carried forward with the data to each additional step, with additional validation included whenever feasible and pertinent. For instance, power flow data from field equipment may only be validated for reasonability within a substation, but compared with other data for bias or inconsistency in a state estimation process at a control center.
Validation of data is not a technology per se, but is a theme that should run through all system designs.
Many functions are sensitive to the time when data last updated. For these functions, a number of mechanisms are available for handling this sensitivity. These mechanisms include:
1. Time
stamping data as to when it was last updated
2. Flagging
data as “out-of-date” if it exceeds a specified time limit.
3. Extrapolating
or interpolating data values from other more timely data
Management of time-sensitive data flows entails ensuring that systems can handle the streams of data in a timely manner. Although the primary solution to this issue is to ensure that the systems and communications are appropriately sized, this sizing can entail addressing the following questions:
1. Is
the flow of data relatively constant or are there bursts of data? (Queuing
theory can be used to establish sizing for this issue).
2. Is
maximum time of data flows critical? If so, then the system should be designed
for deterministic delivery times (as opposed to statistical delivery times) and
sized to handle the maximum throughput. An example is the requirement for
maximum response time of 10 ms for protective relaying. In this case,
non-absolute protocols like Ethernet are not feasible unless they are used in a
switched network (so that essentially there never is a conflict).
3. Is
it acceptable to base the timeliness of data flows on statistical
characteristics? For example, access by Market Participants to market
information often has statistical requirements, e.g. 95% of Market Participants
must be able to access information within one second 99% of the time. The
primary solution is the provision of adequate bandwidth, processor power, and
well-deigned applications to ensure these contractual requirements are met.
Additional solutions include alternative or backup sources of the time-sensitive
information, as well as measures to prevent denial-of-service security attacks.
In addition, performance logging should be implemented so that proof of meeting
the statistical characteristics are available.
4. Must
multiple users be accommodated with contractual commitments for access within
specific time windows (e.g. access within 10 seconds after each hour, or
response to a request within 30 seconds)? The primary solution is the provision
of adequate bandwidth, processor power, and well-deigned applications to ensure
these contractual requirements are met.
In either case, very precise specifications, statistical analysis of data flows, and rigorous factory and field testing are the most effective ways of ensuring this requirement is met.
Management of data consistency is becoming more crucial as more functions are automated, and as more decisions are made in “real-time” in which there is no leisure to check the consistency of data.
No single solution exists for ensuring data consistency; normally it is a combination of technologies and best practices applied consistently across a system. These technologies and best practices include:
· Clear identification of primary sources of data
· Publish/subscribe, so that systems, applications, and
databases can “subscribe” to data, which will be “published” to them whenever
it changes
· Validation of data consistency through a variety of
data checks
· Alarming on inconsistencies so that they can be
corrected
· Automated procedures for establishing or
re-establishing consistent data after systems are restarted or re-initialized
Mismatches in data formats or structures are frequently a cause of incompatible data exchanges. For instance, one system will be updated so that a new attribute is added to a data item, or a new data element is deleted from a list of data. The other system usually cannot manage these mismatches in expectations, and either shuts down with an error, or blithely continues processing what it thinks it can do (sometimes correctly and sometimes incorrectly), but does not inform anyone of the inconsistency.
The use of object models to establish both the structure of each object and the contents of each data exchange can alleviate many of these problems. In addition, if metadata models of the objects are available in electronic format (e.g. as XML), then systems can automatically detect and correct mismatches.
Many systems require one-step transactions, where if the transaction fails for some reason, the results must be either ignored or rolled back. Two-step transactions are even more complex, since sometimes the results of the second transaction must also roll back the results of the first transaction.
Utility operations systems have not often needed this capability except peripherally. For example in updating the SCADA database, roll back to a previous version is needed if some error was introduced into the updated version of the database. However, this type of capability is certainly required in the market operations, and increasingly in other control center functions.
Two-step transactions and roll-back have been implemented in a number of products.
Data, whether it is
monitored from the field or retrieved from some other system, needs to have
appropriate precision and accuracy. In particular, distribution automation
functions require more accurate data than system operators who merely need to
get an estimate of conditions. This data accuracy requirement implies that (to
the appropriate degree for the functions using the data):
1. The
source of the data must be precise, with accurate CTs and PTs, appropriately
scaled analog-to-digital conversions, appropriate resolution of the digital
values (8-bits, 12 bits, or more bits to represent the analog value), and
appropriately precise calculations of derived values (e.g. var, VA, kW).
2. The
data sources must be accurately calibrated.
3. The
data must be converted correctly to engineering units or other measurement
units.
4. Data
should be validated for reasonableness, if not for more precise checks. For
instance, within IEDs, certain checks for consistency could be used to detect
inconsistent data, while at the control center State Estimation at both the
transmission and the distribution levels could be used.
Data acquisition has many pitfalls
as well for data management. Some of these pitfalls can be ameliorated by new
information technologies, while others require old-fashioned carefulness. The
main problem is the sheer volume of data now required. Some IED controllers
contain hundreds (even thousands) of points. Although many are not needed by
SCADA operations, they are useful for engineers and maintenance personnel.
Given this magnitude of data items, data management must be viewed as a
significant aspect in the design of new and upgraded systems. In particular:
1. Object-oriented
protocols (e.g. UCA (IEC61850) for field equipment, CIM for power system data,
and XML for general data) should be required. These protocols require that the
IED which control the field devices are organized with well-known point names
which are self-describing and can be “browsed” for information much like Web
pages. This self-description allows applications to link automatically to the
correct data with minimal human intervention (thus avoiding one of the main
causes of error).
2. Data
objects required by SNMP MIBs should be retrieved from field equipment,
regardless whether object-oriented or point-oriented protocols are used.
One of the most difficult
areas for managing data is data entry, in which humans type information into a
system or database. The primary issue is trying to maintain high levels of
accuracy. Again, information technology has begun to address this issue:
1. Data entry by humans should
be avoided as much as possible. For instance, no data should be entered twice:
a single source of each type of data should be established and used to populate
other databases.
2. Object-oriented data should
be used, so that applications and database tools can be used as much as possible
to automate the data entry process.
3. Data should be validated as
much as possible during the data entry process. This validation should include
reasonability and consistency checking as well as format checking. Roll-back and
two-step entry procedures should be used where critical functions must have
accurate data.
4. Applications using this
data should validate it as well, to avoid “crashes” and security violations.
5. Logs should be kept of all
data entries.
6. Data backup of all
important data should be provided.
Data storage and access
management is another critical area in the design of systems. Often systems
have been developed for one purpose, then added-to for another purpose. Later,
other applications need data, so an additional jury-rig is added. Some key
recommendations for avoiding this problem (or migrating away from it) include:
1. The
real-time database in the SCADA system, which is focused on providing timely but
limited amounts of data to operators, should not be used as a source of data
for other systems. Rather the front-end data acquisition and control (DAC)
system should be structured to supply the required real-time data to SCADA
system, but also provide other kinds of data to other applications without
impacting the SCADA system itself.
2. The
DAC should support access to field equipment by planners, protection engineers,
and technicians
3. Object-oriented
protocols should be used for all data exchanges between systems. With data
having well-defined names, managing the access to the data is easier and more
likely to be correct. These object-oriented protocols include the UCA
(IEC61850), the CIM, and XML information exchange models (see next section).
4. Currently
some of these object-oriented protocols are not completely interoperable or
consistent in their structure. In particular, the data names and structures of
IEC61850 and the CIM need to be harmonized. This activity is taking place in the
IEC.
5. Standard
formats and methodologies for Application Program Interfaces (APIs) for data
access also need to be formalized. Currently the CIM specifies the Generic
Interface Definition (GID). The GID identifies explicitly which features of existing
APIs (such as DAF
and DAIS)
will be implemented to exchange data implemented in CIM-based databases, to
extend these capabilities to include features needed in utility operations, and
to specify the exact formats to use when implemented over different types of
middleware (e.g. Corba or Microsoft COM).
6. Electronic
registers should be developed which contain the metadata models of the
object-oriented data. This is discussed in more detail in the section on
Information Exchange Management.
7. As
major assets are purchased, their characteristics should be entered into an
electronic asset database (e.g. AM/FM system), possibly using bar codes (to
avoid the fallible human data entry process). They should then be tracked
throughout their life as they move from the warehouse to one or more field
locations over time. This method could provide the accuracy so often missing
but badly needed in the asset databases.
Data consistency across
multiple systems is vital for reliable automation of functions. If data is
inconsistent, then the applications will have inconsistent and probably
incorrect results. Many factors impact data consistency, but some methodologies
can be used to help insure consistency. These include:
1. Use
of publish/subscribe application services, in which every application that
requires certain data “subscribes” to it. Then, whenever the source data is
updated, it is “published” to all subscribers simultaneously.
2. Data
should carry “quality” indications as it is passed from its source to other
systems. These quality indications could include “invalid”, “out-of-date”,
“manually-entered”, and “calculated”. More complex indications could include
multiple timestamps indicating when the data was first created, as well as when
it arrived at each database, or indications of what parameters where used in
calculating its value, etc.
3. Applications
should validate the input data, possibly by analysis (e.g. State Estimation),
possibly by accessing multiple sources of similar data that can be
cross-checked.
4. Error
handling mechanisms should be in place to notify the Network Manager of loss of
data, inaccessibility of data, invalid data, and other data quality
indications.
Database maintenance is one
of the most difficult jobs to perform accurately all the time. Again, the best
method for managing this effort is to provide as many tools as possible for
capturing the data automatically, and then verifying it before the database is
released for use by the control center systems. Many of these tools and
methodologies have been mentioned in previous sections:
1. Object-oriented
data using self-description techniques for correctness; eventually for
automatic update capabilities
2. Cross-database
consistency checking
3. Data
quality indications and time-stamping
4. Updating
of metadata registers and notification to applications to access this register
to determine if data exchange formats have been modified
Data management will never
achieve perfection. Therefore, critical data should always be backed up, all
changes should be logged, and some means should be available to “roll back” to
a previous version if necessary.
The logging requirement can
also be critical for auditing as utility operations are being scrutinized in
more and more detail due to the market environment.
The management of
applications is also necessary in order to ensure that functions operate
correctly and accurately. Most of this management must involve the individual
applications themselves, but general management methodologies can be
recommended. These include:
1. The
Unified Modeling Language (UML) methodology described in the previous section
should be used/required for all new application development or upgrades of
existing applications. Use of this methodology will help ensure that the
application is properly integrated with other applications.
2. All
new and updated applications should be very thoroughly tested to ensure they
execute correctly under both normal and error conditions. Applications which
have not been properly tested will not be trusted by users, often leading to
the applications being “turned off” or ignored (this has been the fate of many
power system network analysis applications).
3. The
status of applications should be monitored by the SNMP Manager, with
appropriate levels of notifications sent to users (e.g. alarm if a critical
function is impacted, warnings if applications are “taking too long” to process
data, “not responding” notifications, etc.).