Welcome to F97.BE!

The other day, I had an interesting conversation about one of the phrases I use on this site — things like “AI Energy Production Forecast” or “Solar energy forecast powered by two AI models.” The question came up: is it really AI doing the work? Or is that just a buzzword?

That question stuck with me — and it’s a fair one. So I decided to write this down.

My forecasting system is built upon machine learning, a powerful branch of what we today call artificial intelligence (AI). At its core, machine learning allows computers to do something remarkably human: to learn from experience. By analyzing historical data — in my case, years of solar energy production and weather patterns — the system begins to understand relationships, detect patterns, and make informed predictions about what tomorrow may bring.

What makes this technology so compelling is that it doesn’t rely on a rigid set of rules. I don’t tell the computer exactly what to do when the sky is partly cloudy or the temperature drops below 10°C. Instead, it draws its own conclusions by learning from reality itself. The result is a system that evolves, improves, and adapts — much like we do.

On a deeper level, there’s something profoundly philosophical here: this is a machine engaging in a primitive form of understanding — not awareness, not thought in the human sense, but a functional intuition based on evidence. It's not guessing; it's learning. And though the intelligence is narrow, its application is transformative.

So when I say that my forecasts are powered by AI, I don’t mean some futuristic science fiction concept. I mean that I’ve harnessed a form of logic that learns from the past to illuminate the future — a tool that stands quietly behind the scenes, helping my solar system become a little smarter, day by day.

What I Do

Every day, my system quietly gathers data from the environment around it — and from the technology on my roof. At the center of it all is my solar installation, generating clean electricity as the sun rises and sets. But just as important as the energy it produces is the story it tells: a stream of numbers that, over time, becomes a living record of light, weather, and performance.

I continuously collect and store this data, including:

• How much solar energy was generated each day (in kWh) — collected directly from my Fronius inverter using a home-made Python program that polls the device every 15 minutes
• Weather conditions — temperature, cloud cover, wind speed, solar radiation, and precipitation — retrieved from the Open-Meteo API using another custom-built Python script that fetches and processes daily forecasts and historical data

These datasets don’t exist in isolation. By combining them, I can begin to teach the system something meaningful: how different weather conditions — a hot sunny day, a cold clear morning, or an overcast afternoon — shape the daily output of my solar panels.

I use this real-world data to train machine learning models, algorithms designed to detect patterns that aren’t always obvious to the human eye. Over time, the system learns to correlate specific weather profiles with production outcomes, building a statistical understanding of how my solar panels behave under different conditions.

Once trained, the model can take tomorrow’s weather forecast — a prediction in itself — and apply its learned experience to forecast how much energy I can expect to produce. It’s a kind of digital intuition, built not on guesswork but on a growing understanding of cause and effect.

Machine Learning: The Core of My Forecasting

Machine learning (ML) is the science — and the art — of enabling computers to learn from data. Unlike traditional programming, where every possible condition must be accounted for manually, ML systems don’t follow fixed rules like “if it's sunny, then expect high production.” Instead, they look at vast amounts of data and discover patterns on their own — often uncovering relationships that are too complex or subtle for a human to define explicitly.

In my case, I use supervised learning, a branch of ML where I feed the system with both the inputs (like weather conditions) and the outputs (actual energy production). The model learns to associate the two — understanding, for example, how a cloudy day with low radiation might lead to a 40% drop in output. Once trained, the system is able to generalize: it can make accurate predictions for new, unseen days based on its learned experience.

To do this, I experiment with several different machine learning algorithms. Each has its strengths and assumptions, and I test them all to see which performs best for my specific use case. The models I regularly use include:

• Random Forest — an ensemble method that builds many decision trees and combines their results for a more robust forecast
• Gradient Boosting — a powerful technique that builds decision trees sequentially, each one learning from the errors of the last

I automatically evaluate each model’s performance using statistical metrics like R² and mean absolute error (MAE), and I select the one that performs best — not just based on accuracy, but also on consistency and generalization ability. It’s an ongoing process of training, testing, and refining, and it ensures that the forecasts I generate are grounded in real-world data, not guesswork.

Is This Artificial Intelligence?

Yes — what I’ve built is a real-world application of what’s known as narrow AI: a system designed to perform a specific task — in this case, predicting solar energy production — with increasing accuracy and autonomy. It doesn’t think, feel, or understand like a human, but it does something surprisingly close: it adapts, learns from experience, and makes informed decisions based on real-world data.

Some might argue that calling it “AI” stretches the term, and in a way, they’re right — this isn’t a sentient machine. But if we define intelligence not as consciousness, but as the ability to interpret inputs, learn from them, and act with purpose, then this system, humble though it may be, fits the bill.

There’s something quietly profound about that. I’ve taken a machine — a computer running some Python scripts — and given it the ability to anticipate tomorrow. Not because I programmed it with rules, but because it built its own understanding from the past. In that sense, this is more than just a tool — it’s a kind of evolving knowledge system, one that lives in the space between sunlight, silicon, and statistics.

So no, it’s not science fiction. It’s applied artificial intelligence, powering my rooftop solar system, learning with each new sunrise.

Why It Matters

For me, forecasting isn’t just about numbers — it’s about taking control of my energy future. Every kilowatt-hour I generate, store, or send to the grid is part of a larger ecosystem, and having a reliable forecast turns that ecosystem into something I can interact with — intelligently and deliberately.

Living in Belgium, I’m subject to the prosumer tariff: a fee structure that charges solar panel owners for using the grid, regardless of how much energy they produce or consume. It's a system that assumes the grid is my battery — absorbing excess production during the day and delivering energy back when I need it.

But that model is already changing. By 2030, my goal is to no longer rely on the grid as a buffer at all. I want my system — panels, smart meters, electric vehicles, and storage — to be intelligent and autonomous enough to manage itself completely. No virtual battery, no guesswork — just true, local optimization.

That’s why forecasting matters. It’s not just about prediction — it’s about preparation. I’m building a system that listens, learns, and reacts — so that when the sun shines tomorrow, it does so with purpose.

This page isn’t just a technical overview — it’s a statement of intent. Forecasting my solar energy production with AI isn’t a gimmick or a trend; it’s a conscious choice to align technology with sustainability, autonomy, and trust.

I built this system not to chase perfection, but to create understanding — of my energy flow, of my impact, and of what’s possible when data meets daylight. Every forecast is a quiet act of preparation, a small but meaningful way to live more in sync with the world around me.

If you're reading this and thinking about your own solar setup, your grid, or your role in this energy transition — I hope this gives you a glimpse of what’s already possible, and what lies just ahead.

Erik Schmidt, April 2025.