Dark lager beer tastes good. With a little bit of imagination we can also use beer to illustrate something that few Big Data system vendors want to admit when promoting meter data analytics systems to electricity distribution companies. Big Data is missing a key thing to be successful in the meter data analytics paradigm: data.
The origin of this bold claim can begin with some parallels that can be drawn between beer and electricity. Bring enough propeller heads together and the discussion may turn to the role of In-Grid Data Analytics and how beer is a lot like electricity:
- Electricity is easy to consume but takes resources, skill, time, and effort to generate
- A complex network of transformers and wires are required to distribute electricity, and beer needs trucks, kegs, and bars
- Both don't last long, don't store well, and should be consumed quickly
- At least one movie has been made about stealing electricity
- When it comes to Big Data, analyzing the consumption patterns of end users of beer will make it difficult to find problems with upstream distribution and theft, just like electricity.
This last point begs the question: will the billions invested by distribution companies in smart metering, meter data analytics (MDA) and Big Data help them find losses and theft in the distribution grid?
To combat theft, the beer distribution company might install surveillance cameras in the shipping bay, in delivery trucks, and near storage and refrigeration units. It is far more difficult to combat electricity theft because the electricity constantly flows through hundreds of miles of power lines strung throughout the countryside. There simply is no cost effective way to install enough monitors or sensors to determine exactly where and when pilfering takes place.
Smart meters and MDA are often cited as providing a means or identifying all types of energy theft. Indeed a smart meter may send an alert if the end user tampers with the meter. MDA systems use various algorithms to try to spot anomalies; for example, low voltage outliers among multiple consumers on the same line may indicate a meter bypass. Smart meters and MDA certainly can help.
But questions arise:
- What can be done if no smart meters exist due to budgetary limits?
- What can be done if the smart meter is not registered in the billing system?
- What can be done if the meter has been installed wrong (i.e. wiring incorrect either accidentally or intentionally)?
- If the billing system is incorrectly set up (e.g. wrong billing multipliers).
Even if the smart meter is installed correctly and registered correctly, most people are still smarter than smart meters and the serious electricity thief is likely to start looking for weaknesses in the distribution system upstream from the point of sale.
The analogy with beer may be a 'smart pub:' install video cameras or an alarm system if an unauthorized patron tampers with a keg or doesn't pay for his beer. If someone were to siphon beer out of a keg, the pressure (voltage) might be lower than expected at the tap, and a system could be designed to find this -even though there are probably fewer pubs willing to pay for this than there are utilities willing to pay for smart meters. At some point, though, the serious beer thief is more likely to look for weaknesses in the distribution system upstream from the point of sale.
MDA systems do fairly well when it comes to finding anomalies in consumption patterns that might point to an underlying problem. When the consumption patterns of a large percentage of electricity users are cross-referenced with Customer Data Analytics (CDA), it may be possible to find additional unauthorized consumption. In recent years, major Big Data system vendors that are targeting smart meter infrastructure (SM!) have chosen to invest heavily in CDA and MDA systems. (Also this and this)
A concern raised by some distribution companies, however, is that the Big Data CDA/MDA approach simply generates another set of data that cannot be acted upon, or generates too many false positives. Data quality, communications errors, and the large number of alerts and flags can make it difficult to pinpoint the problem. The even bigger challenge, however, is that theft and losses that occur upstream are very difficult to measure using data collected by a smart meter. There is no easy way to audit the distribution grid using this approach.
Return to the beer analogy for a moment: checking the consumption patterns of a large percentage of beer drinkers will never give you an indication of where and when beer may have been diverted upstream. There is no easy way to know when or where a delivery truck was hijacked and kegs of beer stolen from the back. There's simply no data to help. It is the same problem with using Big Data analytics of energy consumption: these Big Data solutions lack data.
The data that is still needed for success in the fight against theft and losses in the electricity distribution grid is called In-Grid Data. Every data source that provides load data from within the grid may be classified as In-Grid Data.
A number of vendors have introduced sensors that can be used to generate In-Grid Data by installing a sensor on the medium voltage lines. Some vendors have introduced sensors to be installed on the low voltage side of transformers. In general, this data can be used to compare delivered energy against the aggregate consumption measured by the downstream billing meters.
Problem solved, right? Just install sensors on MV and LV lines to permanently audit energy consumption and metering at each point. The cause of any loss (including theft) can always be found if there is enough In-Grid Data between the generator and the point of sale. In our beer analogy, we can install a camera in every truck, warehouse and refrigerator -this would give us a constant monitoring system from the brewery to the point of sale.
Sadly, the problem is not solved because unlike beer in a fleet of delivery trucks, it is generally cost prohibitive to install permanent In-Grid Data sensors throughout the distribution grid. Most companies already struggle with the business case for smart metering-adding more in-grid sensors needs to be carefully planned to focus scarce budget dollars on the highest risk areas.
In-grid data -the better way
A distribution transformer in North America may have up to 10 consumers on one transformer, so a distribution company with 1,000,000 consumers may require over 100,000 LV transformer meters. Fully loaded cost for each installation: maybe $1000. If the distribution company wants to buy MV sensors to measure the line load for every 10 transformers (100 consumers), they would need 10,000 line sensors. Fully loaded cost for each installation: maybe $2000 per phase. The total cost can be $120M or more.
Spot checks can help
The best approach to obtaining useful In-Grid Data is to perform spot checks on the distribution system, and focus precious resources on the highest risk areas.
Rather than a camera on each delivery truck and so on, we can instead take statistically relevant samples of how much beer was loaded onto the trucks, how much came off, how much went into the bar, and how much was sold. Those In-Grid Data points can then be used to determine if there are losses in the distribution system. In high-risk areas, more sampling will serve to pinpoint the cause of the loss.
Similarly, spot checks may be carried out in high-risk segments of the electricity distribution grid. Measuring the line loads for a week or two in strategically selected locations will give a good data set for consumption patterns in that area. Combined with additional information about the consumers in the same area, this In-Grid Data can then be used to determine the likelihood of losses.
This is not an unusual thing to do: risk assessments and risk-based decision-making is widely used in the financial and insurance industries where Big Data problems abound. When fraud and abuse occurs in these industries, the company will collect more data selectively -e.g. by carrying out a field investigation in a particular neighborhood to see if an insurance claimant has a valid claim.
The highest risk areas may be selected using the results of MDA and CDA systems, of outage management systems, asset management systems, and a wide range of other metrics. Of course, suitable case management and investigation management tools are required to keep track of the In-Grid Data and results.
Meter Data plus In-Grid Data
The lessons we can learn from protecting our fictional beer distribution system can help us reduce losses in our real life electricity distribution grid. We just need to keep in mind that Big Data based solely on smart meter data might not be enough -only with In-Grid Data analytics can we be confident that losses of all types can be reduced.
Problem solved. Cheers!
Written by Rudi Carolsfeld and Mike Rowling
Mike Rowling is Chief Technical Officer at Awesense Inc. He has more than 20 years of experience solving Data Analytics problems. Mike has been a volunteer at the Great Canadian Beer Festival since inception.
Rudi Carolsfeld is Executive Vice President of Global Sales & Alliances at Awesense Inc. He has more than 20 years of experience solving Smart Grid problems. Rudi lives in Victoria BC and enjoys visiting the Great Canadian Beer Festival.