© 2019 SourceMedia. All rights reserved.

Big Data, Analytics Energize Electric Utilities

For utilities, a perfect storm is just that: a rare confluence of various weather elements that coalesces to create an event of unusual magnitude. A perfect storm can often mean huge costs and losses to the company from damage to its infrastructure. But utilities are among the industries best-positioned to reap the benefits of another type of perfect storm, the rise of big data and streaming analytics.

The sheer size of a utility -- with its customer base and its generating and transmission grid – is well-suited to the voluminous big data that is now capturable. This data can drive insights that, when multiplied across hundreds of thousands of customers and billions of dollars in revenue and investment, can generate significant efficiencies in an industry with annual revenues of more than $300 billion.

Rise of the Smart Meter

A key element of the industry’s transformation is the smart meter, which can deliver energy-consumption data points that are orders of magnitudes greater than older analog devices. U.S. utilities have installed more than 50 million such devices, covering more than 43% of the country’s homes, according to the Edison Foundation’s Institute for Electric Innovation.

Analyzing this data can help improve:

  1. Energy efficiency
  2. Revenue protection – including theft detection
  3. Load forecasting to optimize utility companies’ buying decisions
  4. Grid utilization, as well as outage prediction and detection
  5. Customer experience

A more recent trend upping the ante is the ability to not only warehouse and analyze this data on a static, historical basis – but subject the data to analytics for real-time processing and insights.
One way of analyzing static data is with batch analysis. An example of this approach is clustering customers by daily energy-consumption behavior. With hundreds of thousands, if not millions, of customers, these patterns are at the surface individualized. But at a global level, they share common patterns of energy use that allow the utility to distribute them among a relative few different consumption clusters. This data can help with load balancing and energy-efficiency programming.

Four Examples of Customer Daily Consumption Summaries

Utilities generate these descriptive models of customers that share similar consumption patterns, leading to analytical insights. The utilities can then use these models to create “scores” based on a target variable, such as who is more apt to benefit from energy-reduction programs. This is the basis of predictive analytics, where you can forecast with a high degree of probability future behavior.

Which customers do you target with energy-reduction programs? Combining the scores with other types of data can help identify those individual customers likely to respond to these offers. Utilities can base that determination partly on customers that are anomalies to the model, for example, those that use more energy than the model predicts is typical of their cluster of peers.

You can also look at the data over time to identify fluctuations, known as Time Series Decomposition. A smart meter allows you to slice and dice the consumption data to determine:

Utilities can translate these consumption patterns into formulas that again will identify anomalies the utility can target for improved efficiency and lower usage. The aberrations could include a heating or cooling cycle that is an outlier in the magnitude of energy usage or the frequency of these cycles throughout the day, compared with similar customers.

To fully reap the fruits of big data and predictive analytics, electric utilities can build a streaming analytics infrastructure that uses real-time data to help them make the right decisions at the right time. FICO has used this technology since the early 1990s for the financial services industry, allowing predictive model scores for payment card fraud with response times that are measured in milliseconds. These models enable the real-time determination of whether card transactions should be allowed or declined based on probability of fraud.

Addressing Fraud

So where can this help utilities? How about fraud? Utilities lose as much as $6 billion annually from stolen power, amounting to as much as 3% of their revenue in some regions. One significant source of theft is illegal marijuana farms, which have huge power requirements for indoor lighting and temperature control. Other cases are opportunistic, as when contractors or homeowners install bypass lines. In these situations, measurements of consumption can lead to predictions of where fraud is occurring on the network.

Streaming analytics will make predictions by understanding consumption throughout the day as well as differences among customers over time. As an example, streaming analytics can clearly distinguish between two customers with the same monthly consumption but vastly different usage patterns.

Streaming analytics can consolidate huge amounts of transaction data into a transaction profile that contains recursive variables and features for the problem it is addressing (whether it’s fraud, revenue assurance or network assurance). So a utility can utilize a meter-transaction profile and not have to send huge volumes of data points for each hour of each day for every customer back to a central location for eventual analysis. Streaming data updates the profile based on each received transaction point where the recursive variables properly weight the new observable data with the older estimates.

This temporal data compression allows the profile to include measurement of variables – daily and weekly usage, peak and off-peak consumption, usage variation, short- and long-term usage ratios, etc. – that allows the production of real-time scores to provide the utility with actionable decisions to make about the individual customer. This customer insight is of particular use in deregulated markets where pricing packages and discounts grounded in analytics can boost business and reduce customer churn.

With these transaction profiles, profile variables and scores can be developed to determine risk of fraud across the entire customer base by identifying the anomalies, or outliers, in this population. To ensure that anomalies are relevant, these analytics need to be self-calibrating such that abnormality is based on the current real-time behavior versus dated historical data. This means that the analytics dynamically update themselves in real-time-- learning where anomalies are in patterns and variances “on the fly” without corrective intervention from an outside source. This saves time and resources with a high degree of accuracy.

This allows the utility to determine with a high probability that lower power usage is not a problem with the infrastructure but is within the bounds of lower usage due to a change in weather pattern – for example, a temperature of 70 degrees on a January day in the Northeast, as in the first chart below. In the lower graphic, the utility can identify with a good deal of certainty that theft of service is occurring because the anomalies are outside the bounds fueled by real-time data.

Theft of electricity is a major problem. Commonwealth Edison Company is in the midst of a program to install smart meters at all of its customers in its northern Illinois territory by 2018, more than 750,000 and counting. The company has both an efficiency program and a revenue-protection unit that uses its “smart” grid to detect fraud and theft.

Predicting Failures

Utilities can apply the same principles and processes to detecting outages and predicting failures. Streaming analytics allows a utility to react to network issues before receiving calls from unhappy customers and to avoid or more quickly begin to repair an outage. Utilities are finding it advantageous to use predictive analytics and streaming data to greatly improve their prediction of such events and prevent potential failures. When failures happen, they are typically preceded by fluctuations in electric usage, voltage and other equipment parameters. Utilities can profile short-term behavior and typical long-term behavior to detect the probability of possible failure in the near future – becoming more proactive on inventory management and repair of critical network infrastructure.

The bottom line is predictive analytics can translate into better customer service and higher profitability by giving utilities the ability to respond in real-time and with the proper insight to tackle the inevitable issues of outages, theft and utilization spikes.

For reprint and licensing requests for this article, click here.