Senior decision-makers come together to connect around strategies and business trends affecting utilities.


Journey towards artificial intelligence in the Electric Sector.

image credit: Kristian Hammond
Cristian Rodriguez's picture
IT Business Leader Grupo SAESA

14 years of experience in Distribution, Transmission and Regulation, in southern Chile, Implementing computer projects. Expert in Automation with Artificial Intelligence.

  • Member since 2019
  • 2 items added with 5,008 views
  • Nov 29, 2019

This item is part of the AI & Machine Learning in Utilities - Winter 2019/20 SPECIAL ISSUE, click here for more



Artificial Intelligence or AI, already has a long trajectory, dating back to 1950. No doubt the mathematical algorithms have not been modified, the explosion of the current use of these technologies is associated with hardware growth, which allows even having cloud development environments. Thanks to these advances it can be said that technically, AI is everywhere.

In addition to this, big technology companies have bet on these cloud platforms, delivering tools at intermediate developer user level, simplifying it’s use and implementation.

Your access to Member Features is limited.


How to integrate this technology in an electrical company? It is certainly a question that can be addressed in different ways, but to give an initialization we will begin by indicating the large blocks of this business:

  • Power Generation.
  • Transmission.
  • Distribution.
  • Commercialization.
  • Regulation.

Each one of these blocks have different needs and we will deliver some notions of implementation of them. But first, it is important to know which are the most used components in AI:

  • Machine Learning.
  • Deep Learning.
  • Evidence-Based Reasoning.
  • Text Analysis.
  • Recommendation System.
  • Natural Language Generation.

Also, we must consider some additional components that allow its information generation to be a contribution to AI:

  • Robotic Process Automation (RPA)
  • Chat/Conversational Interfaces


Examples of AI

Power Generation.

Without a doubt, electricity generation with renewable methods (solar, wind, etc.) has been increasing its market share. A vital point of this type of generation is weather, because climatic conditions such as radiation, wind speed and rain, are conditions that affect the generation of energy. To visualize how these conditions may impact the generation, the most optimal way is to use Learning Machines, either by purchasing climate data packages or using platforms such as StormGeo or Nnergix, which have already solved this analysis on their platforms and have many years of climate data.

If we select an option, for example, wind generation, with these platforms or with their own predictive models, it certainly improves the quality or prediction of the weather forecast which can lead to an optimization of the available turn reserves.

Another option is when it has a generation mixture, for example, Diesel and Hydraulics, in which its usage is according to the arrangement of the reservoir level. For this, an improved rain forecast will allow it to have a better logistics on the use of Diesel and a more efficient stock management, given that diesel prices in some countries are very volatile.



One of the most important issues in utility companies in the electricity sector is Asset control, how to generate predictive maintenance plans and how they are associated with corrective plans. In some companies, network extension does not allow a manual inspection of all their facilities, it is for this reason that inspections with Drones or via Helicopters have arisen exponentially. Without a doubt this option allows inspecting a large amount of assets, with information that really contributes greatly to the generation of maintenance plans.

The problem begins when the company owns thousands of linear assets, which are transformed into thousands of kilometers of inspection, and therefore, thousands of images to analyze. The common mistake is that, at the beginning of these inspection services, it is decided to make videos, resulting in thousands of images that are transformed into hundreds of hours of video analysis.

How to make optimal and a real contribution to drone inspections? By using deep learning, which creates a neural network capable of analyze thousands of images in a few hours. To do this, you must train the machine to look for example, broken insulators, trees approaching the line and so on, having the ability to create dozens of classifiers.

In this solution, you can find several platforms that have solved the inspection process plus AI (Optelos), or look for a combination with OpenCV + Deep Learning Framework. This AI will detect the images with problems and classify them, this way the Maintenance Engineers will focus on generating a robust Maintenance Plan and dedicate themselves to processed images.


The transmission example can certainly be done in distribution, with a greater number of classifiers, but we will focus on a problem that takes most of the headaches, the “electrical failures in Dx”. For this we can dedicate hours in the predictive models that can be incorporated to help in this process. We will focus on the direct relationship with customers and the information that is vital and that helps to improve the final customer satisfaction, these are the estimated restoration times (ETR).

This solution has a low to medium level of implementation, but at a strategic level it is high. One of the most critical points that affects customers towards distribution, is the information that flows to them. I will explain how to generate this model in the company I work. First you need at least 2 years of electrical failures data, which must be geographically linked. These failures must also be classified by type and elements that generated the failure and some additional data, such as weather associated with the speed of the gusts of wind that were involved may be necessary. An important point is that the person who performs data science has business experience as it is vital for a positive outcome.

Based on the parameters of the faults and weather, a model of machine learning type Random Forest is created (you can explore to find a model that is more suitable for it). The benefit of this model is that, when a customer indicates the lack of supply, it can return an ETR from 5 to 30 minutes after the call without a crew arriving at the place.

In my case, this example was made with IBM Watson + RED Node and connected to the technical system and the DMS(Distribution management system) module, as it is important to have feedback if the client that is calling is not in an active fault. Then connect it to the IVR, Social Networks, or even create a Virtual Assistant.


Trying to show a single example in this block is complex, so we will focus on client/company communication. For this, the client can communicate via Social Networks (Facebook, Twitter, Instagram, etc.), phone, chats and e-mails. All this information, depending on the number of customers that the company has, can be transformed into Big Data, so the models of NLP (Natural Language Processing) comes to participate.

With these models, a customer can be classified by analyzing the text or tone and have automatic attention procedures in which customer needs are resolved without contacting an operator or executive.

For these solutions you can use the well renowned TensorFlow, which contains several modules and one of these is NLP. Now if your focus is not programming from cero, you can use paid tools as Azure, AWS, Google, IBM, bringing tools intermediate to basic user level to create these models in a very interactive way.


Depending on geographical location, regulation may be more complex than others. In my case, our company is regulated by the Country and these entities generate articles and trades that must be answered with dates committed and established by law.

It all depends on the magnitude of your company and the geographical area that is established in its operation, but given the various articles, trades and request for information, two important components of AI come in play: text analysis and natural language generation.

Text analysis using NLP allows classifying requests and guiding which platforms will be involved to obtain the information to respond. Natural language generation responds are based on facts, which are configured from legacy platform data.

No doubt the solution seems simple, but it needs an arduous training which all legal contexts are incorporated, but at the moment of working it will be like having a lawyer or supporting the work of the company's regulatory lawyers, saving fines for not complying with the information requested.

For this solution, you can use platforms such as LegalRobot or simply build your models with paid platforms such as Azure, AWS, Google, IBM.


As a final analysis, the inclusion of AI is more effective than ever, paid platforms or specific platforms in a solution, are very varied and can be included in all the Electricity business blocks. The most important thing is how you make this trip. For this, you do not need such a complex personnel structure, but if you need it, training Data Scientists and Automation experts who knows the business will undoubtedly pave this great path, the benefits are varied, most quantifiable, depending on the problem of each company you have before doing this journey.

What I can recommend is to follow it with something simple, then the need and learning will lead to more complex issues. My intention has always been to automate, not focused on replacing people, but focused, for example, in a Maintenance Engineer dedicated to his specialty and does not become an Excel Engineer.






No discussions yet. Start a discussion below.

Cristian Rodriguez's picture
Thank Cristian 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 »