Bidgely’s been leading the Utility AI and disaggregation space for eight years now, and it’s exciting to see how other vendors and energy utilities are embracing the technology. We want to extend a huge welcome to all companies attempting energy disaggregation science. There is no doubt that load disaggregation has crossed the chasm of early adoption. Many of the RFPs and RFIs floated over the last 24 months included load disaggregation, both for customer interactions and for utility internal decisions like Grid planning, DSM targeting, and Load profiling & forecasting, and so it is natural to see other vendors attempting to stamp their presence in the disaggregation space.
The adoption of AI in the energy space not only increases energy savings, drives targeted demand response programs and achieves more efficient grid optimization; but also enables the personalized experience that consumers have come to expect, thanks to the application of AI to everyday interactions by the likes of Netflix, Google and Amazon.
While a number of smaller companies have been working on energy load disaggregation for a few years, we have recently seen the likes of the larger organizations like Tendril and Oracle (OPower) either acquire disaggregation capabilities or share details about their experiments with disaggregation. Bidgely’s load disaggregation and appliance-level insights are offered by leading utilities through alert programs and websites for over 10M homes. The benefit of load disaggregation has specifically penetrated large deployments for Home Energy Assessments or Audits, as well as Field Audits. AI-powered surveys use customers’ actual usage to augment home energy assessments–allowing utilities to offer personalized promotions and recommendations, while eliminating the need to burden customers with 10-15 upfront questions. As the pioneers of smart meter disaggregation, and commanding an IP portfolio of 14 patents, we are proud to see the market adoption come this far. We’re also committed to extending our principles of scalability and enhanced utility services through our universal disaggregation.
What are you buying?
You’re not just buying a technology; you’re buying the value that it provides for your organization. Your disaggregation vendor should be able to address multiple use cases. Make sure you understand how the value of disaggregation will be realized in the solution being offered. Many utilities think showing itemized energy breakdown does all of the work. It’s a lot like buying a drone to take a selfie—there has to be a much bigger reason you are buying a drone (unless you are really buying it for playing around with – equivalent to the difference between innovation projects testing out disaggregation because they don’t believe it works vs. buying it to solve real business challenges).
To understand the value, explore how AI serves your key objectives. Then, dig deeper to see if your solution can extend from theory to reality. For example, you may ask yourself – How does load disaggregation or AI help me give my customers more personalized energy spend (similar to itemized credit cards, bank statements and phone bills)? How will load disaggregation increase my customer satisfaction? How will this help my DSM program targeting? Can my vendor tell me exactly which homes have an EV; or which homes have electric heating vs. gas heating; or which homes are using their cooling on weekday afternoons, so we can confidently offer them a free Smart Thermostat without lowering the TRC of the program?
Load disaggregation is not a one trick pony – it’s a swiss army knife that can be used in many ways — make sure your vendor isn’t just using it to dress up their existing solution. In most scenario where your vendor has an early immature solutions, they won’t qualify on following questions:
- Can you disaggregate non Smart Meter data (many utilities are just starting or are in the middle of the deployment that will run for 2-5 years)?
- Can you disaggregate gas loads?
- Can you disaggregate solar production?
- Can you disaggregate loads with less than 12 or even 6 months of data?
What is accurate enough?
It’s easy to ask “what is your accuracy?” And vendors love to throw out numbers like 93%. But the first question to ask is – what application am I trying to serve and what accuracy and recency is needed to serve that application? If you are trying to detect a garage door that has been left open in real time – AMI smart meter data is neither high enough resolution or real time enough to serve that application. A whopping $300/home starter fee is needed for real time hardware-based load disaggregation solutions. But, personalizing interactions, targeting customers for the right programs, increasing DSM efficiency, and factoring EV and PV load into grid planning – can all be done with existing Smart Meter data.
Still, the question of accuracy does not go away – Customers will first question how you did this and then they might question the accuracy of it? At Bidgely, we define accuracy at two levels: (a) Detection and (b) Estimation. Detecting an appliance right in a home is crucially important. Why? Because consumers need to have confidence in the solution. Also, they know what they use – so if you incorrectly detect EV or Pool pump in a home that does not have one, you lose customer confidence. Ultimately, false positives hurt lot more than false negatives. Achieving 90% or higher detection accuracy is critical. Estimation accuracy, on the other hand, refers to the accuracy of the predicted kWh consumed per load. While important, this accuracy level is less critical than detection accuracy — your estimation accuracy needs may actually vary by appliance type. Hope our readers are now more intelligent to ask the right questions.
Another important consideration for accuracy: – how many homes have been measured? Testing something on a few 100 homes is easy to both test and optimize, but hard when you test it on a larger data set that is a true representation of overall population. Evaluations need to cover thousands of homes, consisting of a variety of appliance brands, age, weather, user behavior, etc in order to give enough confidence that the technology exposure will not bomb back.
Where rubber meets the road
Nothing is a bigger proof than large customer adoption. The biggest question to ask is: Is your solution being used to touch millions of customers proactively every month OR is it buried three layers deep on the website and only provides energy breakdown by asking the customers upfront to answer 5-10 questions. If it’s the latter, you know it’s a statistical model claiming to be load disaggregation. However, if you can get a vendor to expose the disaggregated load to millions of consumers in an opt-out fashion, you may have just found your winner. At Bidgely, your reputation is our reputation. We’ve developed that with 8 years, $30M in R&D, 30+ Data scientists, and passion for AI that runs deep into our culture. It’s built in, not bolted on.