Machine learning – Predicting renewable power generation
Our entire energy system has to become more efficient. Consequently, technologies like blockchain, artificial intelligence, machine learning algorithms, and data software are of uttermost necessity. Promising technologies and applications show immense potential in helping to diminish the intermittency aspect of some renewable energy sources. A company called Deep Mind is able to predict wind power generation already 36 hours ahead of time.
Is this the solution to help combat a major downside of wind energy?
The problem with wind energy
Along with solar PV, wind energy is amongst the cheapest energy sources to date. However, there is one problem with wind energy. Because of its nature, gusts of wind are inconsistent, which makes it nearly impossible to generate electricity at a constant output.
With an increased share of renewables in our electricity mix and an ongoing energy transition, grid stabilization is more complicated than ever before. One possible solution to stabilize our grid is through energy storage applications. Another is through controlling or predicting renewable power generation and the amount of energy fed into the grid.
Generally speaking, the prices achieved for a predictable and scheduled power delivery are higher than for unpredicted and unstable supplies. Therefore, machine learning will increase the value of wind energy through providing predictability. Ultimately, wind will become an even more attractive power source and investors can increase profitability.
Machine learning to boost wind energy’s value
With plummeting levelized wind electricity costs over the past decade, wind energy has contributed to decarbonizing the energy sector. Now it is time to increase its resilience to contribute even further and serve as a reliable power source.
Machine learning is the tool we need to achieve coherence for wind power! An artificial intelligence company called Deep Mind has developed machine learning algorithms to predict wind power generation and boost its energy value. Applied to 700 Megawatts of wind power capacity across the United States, this technology already shows immense prosperity.
A neural network trained on weather forecasts and historical wind power generation data makes an output prediction in a timely and precise manner. On top of that, the program suggests optimal hourly delivery commitments to the power grid 24 hours in advance. As a result, compared to a scenario without time-based commitments, this machine learning tool has helped boost wind power’s value by 20% to date.
What is machine learning?
Most of you have already heard of machine learning and artificial intelligence, but do you know exactly what it is and does?
Machine learning is a tool that enables systems of all kind to learn automatically. Further, improvements in the system itself are possible the more experience the program gains. This can take place without having to program the system over and over again explicitly. As part of the super-ordinated term ‘artificial intelligence’, machine learning focuses on computer programs to access data and learn for themselves.
In other words, no human interaction is needed to enable the system to teach itself and apply the gained knowledge.
Deep learning shares similarities with machine learning and AI but is a more human like application.
Specifically, it imitates the functioning of the human brain in data processing and decision making. Even unstructured or unlabelled data can be processed and used to learn from. It is also called a neural network or neural learning.
Deep Mind makes use of this specific kind of machine learning, trains it on weather and production data and has it predict power output.
To see how well this technology already works, the company has published a graph that compares both predicted and actual wind power output.
Predicted vs Actual wind power output, Source: DeepMind
What value is a prediction that a high pressure system has moved in and the wind will be near zero for 5 days? The answer to the question is “none”, except perhaps it does provide an opportunity for everyone to rush out and buy some dry ice.