5 Mins

Optimising the future of clean energy investment

At Gridcog we sometimes get asked if we use 'AI'... now we can answer yes because we just produced this AI-generated 80s electro-pop banger!

More seriously, we actually use computational modelling and mathematical optimisation, rather than machine learning and artificial intelligence. In this blog post we’re going to try and explain what that means.

What Gridcog does

Computational modelling is a method used to create simulations of real-world systems within a virtual environment. It’s a powerful tool for understanding complex systems by breaking them down into more manageable, simulated parts.

We have developed mathematical models to represent the behaviour and interactions of solar PV systems, wind turbines, thermal generators, battery and other energy storage systems, EV charging infrastructure and EV fleets, and flexible loads (pumping, heating, cooling, etc). This modelling allows us to take an analytical approach to understanding the behaviour of energy project options.

What about mathematical optimisation? This is also a direct analytical and deterministic approach, which is used to find the best solution to a problem based on a set of mathematical functions and constraints. The word “optimise” comes from the Latin word optimum, meaning “best.” We use mathematical optimisation to find the best option from a near-infinite number of possibilities: what combination of asset sizes and what asset controls will deliver the best project outcome given site constraints (like import limits), asset costs, and electricity supply and energy market pricing? More specifically, we use a technique called Mixed Integer Linear Programming to do this optimisation (see the explainer image below).

Where is AI used?

Modern AI, like the approach used to generate the 80s synth pop anthem above, is all based on statistical machine learning approaches.

These are approaches where we train a model based on past or example data to make predictions or decisions based on patterns and relationships embedded within the training data. Unlike numerical models, AI and machine learning (ML) models take an indirect and probabilistic approach to determining the output. This is great when the relationship between inputs and outputs is complex and too hard to represent numerically, and when we’re okay with ‘guessing’ at a good solution, rather than knowing we have found the optimal solution.

AI generated energy project (taking a guess at where that van should be parked to be charged)

There is definitely scope to apply AI/ML to energy project modelling. We have some clear ideas in this space that we’re going to be working on in the future. But these ideas depend on data that in many cases doesn’t exist today — for example, how many fleets of trucks are being deployed in the 'real world' today to provide frequency regulation services to energy markets via V2G? Analytical techniques allow us to explore these kinds of projects today.

One area where AI/ML definitely has a role is in the real-time operational control of the deployed assets.

Our job is to accurately represent the commercial and environmental outcomes that energy projects will achieve once they are deployed. This includes accurately representing the behaviour of the operational software systems driving the deployed assets within our simulations: orchestration platforms, battery optimisers, microgrid controllers, DERMS platforms, energy management systems, etc.

Many of these operational systems have “AI” prominently emblazoned on the side of the box, so what’s going on there?

The answer is they are typically using AI/ML specifically for forecasting. AI/ML models can be trained on historical data and can be used to make predictions .. temperatures are hot today, site load is likely to peak at 3pm, etc. Then, given these forecasts (e.g. future load, future prices), these systems will actually use numerical methods to work out the optimal way to control the assets (how they should be dispatched, deployed, auto-bid, etc to achieve the optimal outcome), i.e. they are typically using exactly the same kinds of mathematical optimisation methods Gridcog does during simulations to make optimal control decisions.

Via simulation, Gridcog is creating the ‘virtual world’ that these assets are being deployed into. This means we are creating the future rather than forecasting it in our models (i.e. we don’t need AI/ML to predict something we already know, like future solar irradiance; and, if you are interested in how we factor forecast error into this process, we’ve written about uncertainty previously). Once we have created these synthetic forecasts we then apply mathematical optimisation in the same way these operational systems do, providing good fidelity with the commercial and environmental outcomes that can actually be achieved in the ‘real world’.

Wrapping up

While AI might get the spotlight with its ability to generate catchy tunes (and more!), it's the rigorous, calculated approach of computational modelling and mathematical optimisation that sits at the heart of Gridcog. It’s how we deliver confidence in energy project decisions, and how we are accelerating clean energy investment around the world.

Fabian Le Gay Brereton
Chief Executive Officer & Co-Founder
February 13, 2024
view aLl articles
The Modelling behind Magic Mode

Figuring out the best configuration of energy assets for a given scenario is a tricky problem. The permutations are potentially infinite and every site and scenario is different.

For the energy transition to succeed, we have to say goodbye to spreadsheets

Excel spreadsheets are error prone, and lack the scalability and flexibility to model complex error projects. Here's why its time to say goodbye to Excel.

Nobody’s Fool: Weaving uncertainty into the fabric of our modelling

Our job at Gridcognition is help our customers plan and optimise distributed energy projects.

Subscribe to our newsletter
Thank you for subscribing to the Gridcog blog.
Oops! Something went wrong while submitting the form.
Related Articles