The mission of this group is to bring together utility professionals in the power industry who are in the thick of the digital utility transformation. 


How to stay on top of the Tsunami of IoT with an AI surfboard

image credit: © Jakub Jirsák |
Andy Skumanich's picture

Renewable Energy specialist, with AI/ML experience to address the most challenging energy issues.

  • Member since 2021
  • 8 items added with 3,857 views
  • Oct 27, 2021

This item is part of the Advances in Utility Digitalization - October 2021 SPECIAL ISSUE, click here for more

Written by: Dr. Andy Skumanich, & Dr. Manny Ghiassi

Massive amounts of data are flooding utilities and, really, any tech industry.  The big question is “how to make the best use of it”.  What that means is using all the IoT and data streaming, and pulse-taking, to extract the most useful Signal-to-Noise information which allows for actionable responses and decisions to optimize the enterprise.

We are past the point of no return to when a tech company could use their human resources to “crunch the numbers” and get their expert’s direction.  The complexity is just too great at this point.  The need for AI and Machine Learning is now an essential component to extract the maximum value.  The new problem is: which AI/ML? and how do we use it?  In our opinion there are too many AI/ML solutions out there which make for another overwhelming situation.  These solutions range from the open-source AI/ML code, to Commercial-off-the-shelf (COTS) software, to at the super charged end, the IBM Watson level.  Each of these solutions can provide an answer, but of course there are significant trade-offs in resources (head count & costs), reliability (accuracy of the info), and repeatability (can the solutions be reused or are they always in need of re-customizing). 

Your access to Member Features is limited.

These are not the easiest questions to answer and in some cases, it would seem to require that enterprises need to set up an AI department which is parallel to the IT department.  This is a daunting undertaking.  Any massive AI/ML set of algorithms with a large enough team can “solve” any given datarich problem, however, such solutions are often excessively costly and require continued, specialized, resources.  The question is how to balance the benefits with the costs and get a meaningful return on the investment in a relatively short time.  However, even with these challenges, it is already the case that forward-looking enterprises in other domains have moved ahead in incorporating the AI surfboard and have realized quantifiably better financial results.  

We examined various other adjacent technology domains which have subsets of both AI adopters and non-adopters, and where the difference can be quantified.  These domains include the telecommunications sector, the automobile sector (sub-set of electric and self-driving vehicles), and medical technologies.  It is interesting to note that the power utility sector is lagging with the implementation of AI/ML compared with these others.  In the case of AI adopters vs non-adopters, various research has determined that there are significant financial differences between the two subsets which can be greater than 20%.  Specifically, AI adopters in parallel business paths had increased gross margins by delta of +10 to +20%, compared with non-adopters, directly due to the AI/ML actions.  Also, the AI adopters reported direct-AI augmented EBIT by >20% as analyzed by McKinsey Global Institute analysis. [1]  

We recently published a paper at the Photo-Voltaic Specialists Conference which is the annual Solar Technology conference (PV SC 48) where we determined that for the DOE solar 2030 cost targets, the only way to achieve the solar cost reductions is to implement AI/ML enhancements. [2]

We’ve partnered with several utilities in joint projects and recognize some of the dynamics which can be holding the utility sector back.  The key is that taking risks in the energy supply domain is challenging at best, and substantially constrained by regulatory systems put in place to protect the energy distribution.  We’ve developed data which shows the value of quality AI/ML in forecasting demand for some of our engagements.  This indicates the potential, and some examples are shown in figures 1 & 2 for an electric utility and for a water utility.  The level of accuracy is certainly adequate to provide the type of AI/ML support.  The main challenge is that shifting an operational plan to include the AI/ML types of analysis needs to be done in the constrained environment where utilities operate – and that’s a very reasonable requirement.   

Fig. 1. Total electrical power vs Day for AI/ML: forecasted vs acutal [3] 



                                                                                    Fig. 2. Close agreement by Ghiassi AI/ML: forecasted vs acutal [4l  

We propose a simple solution as a concept: micro-grids for utilities. [ 5, 6]  Micro-grids are becoming increasingly popular and in fact necessary.  In the California market, the utilities have put in micro-grids in non-remote regions where it is more economical to do so.  There are various notable micro-grids and one of the recent ones is just outside of Yosemite National Park, where the complex terrain and the risk of wildfires, made a micro-grid the solution of choice.  The micro-grids can provide the ideal platform to test-out, and exercise the AI/ML surfboards, so that the learning can become more easily implemented.  

We’ve partnered with another regional utility, looking at their challenges to managing the bi-directional flow of power over the course of a year, and how can AI/ML improve that power management.  It should be noted that various larger utilities have key staff which are working directly to identify the modes of AI/ML implementation.  One of the (necessary) limitations is that utilities are reluctant to divulge and share their operational data.  This makes development projects more challenging, ones which could potentially use DOE R&D funding.  These cases again support the suggestion of micro-grid test beds which could be development and training grounds.  We would encourage the utility community to drive some of these solutions in partnership with the government agencies.  

So to answer the headline question – how to surf the IoT tsunami – we recommend to start with surfing at the local water-theme park.  Go for the bigger waves by then working with flexible local utilities where there is a supportive community and increase the size of the waves.  Then you can start “hanging10” and stay on top for what promises to be an exhilarating ride.

McKinsey Global Institute Analysis 2021, at Global survey: The state of AI in 2020 | McKinsey and related.

Skumanich, A., Ghiassi, M.  “Achieving significant cost reduction in solar through the targeted use of AI and Machine Learning”,  Proceedings of the PV SC 48th International Conference.

Ghiassi, M., et. al. (2006).  “Medium Term System Load Forecasting with a Dynamic Artificial Neural Network Model,” Electric Power Systems Research Vol. 76, 302-316.

Ghiassi, M., et. al. (2008).  “San Jose Urban Water Demand Forecasting with a Dynamic Artificial Neural Network Model,” Journal of Water Resources Planning and Management, Vol. 134, No. 2, 138146.

Skumanich, A., Ghiassi M.. (2020).The developing need for AI to optimize PV for increased

adoption”,  Proceedings of the PV SC 47th International conference  

Skumanich, A., Ghiassi M., (2020).  “AI and ML for Developing Non-Wire Alternatives”, Presentation at the 2020 InterSolar International Conference, Feb. San Diego  

Post script side-bar note:  

It can be done as a recent case study shows.  A distribution system operator (DSO) in Europe used an data analytics platform that provides risk-based asset management and investment planning. Using the AI, visualization, and simulations capabilities which incorporate data on voltage, load, grid topology, and more, the platform helped operators assess available capacity on the system and reliably plan for future needs. It has also helped them utilize assets more efficiently and prepare to manage the increase in distributed energy resources— but still keeping costs down and maintaining power quality. The DSO gained 50 percent efficiency and saved close to $10M over 10 years in capital expenditures.  These case studies are becoming increasingly common, but there is still a requirement for extensive support capabilities in developing the necessary data/AI/ML infrastructure.

Matt Chester's picture
Matt Chester on Oct 27, 2021

There are various notable micro-grids and one of the recent ones is just outside of Yosemite National Park, where the complex terrain and the risk of wildfires, made a micro-grid the solution of choice.  The micro-grids can provide the ideal platform to test-out, and exercise the AI/ML surfboards, so that the learning can become more easily implemented.

Can the lessons learned from these specific lessons be readily applied to more residential/urban microgrids, and vice versa? 

Andy Skumanich's picture
Andy Skumanich on Oct 27, 2021

Generally yes, all micro-grids are custom jobs, but they have elements in common, so leveraging that commonality will allow for learning transfer.

Andy Skumanich's picture
Thank Andy for the Post!
Energy Central contributors share their experience and insights for the benefit of other Members (like you). Please show them your appreciation by leaving a comment, 'liking' this post, or following this Member.
More posts from this member

Get Published - Build a Following

The Energy Central Power Industry Network is based on one core idea - power industry professionals helping each other and advancing the industry by sharing and learning from each other.

If you have an experience or insight to share or have learned something from a conference or seminar, your peers and colleagues on Energy Central want to hear about it. It's also easy to share a link to an article you've liked or an industry resource that you think would be helpful.

                 Learn more about posting on Energy Central »