Machine Learning can boost the value of wind energy, and fight climate change
Google’s DeepMind is famous for its creation of strategy games winning AI agents. However, they have been dealing with energy efficiency for a while now. The DeepMind team was successful in maintenance costs reduction of Google’s gigantic server farms. Now they have tackled AI driven weather forecast for optimisation of wind turbines.
Since the times farmers have logged weather conditions of their paddocks in diary on daily basis, weather prediction has been a big data game. Wind regimes are particularly difficult to predict. If your computer servers farms energy is generated by wind turbines, and you wish it to work 24/7, it’s a tough call to solely rely on the wind to be there at all times. Design for 100 percent solar, wind or hydro power is known to be rather expensive.
The DeepMind Machine Learning algorithm collects data from various sources. It attempts to provide a 36 hours prediction for the amount of wind generated energy per turbine. This way, a smart energy grid can adjust: store some of the generated energy in days of strong wind in anticipation for use when there is little or no wind; it can prepare for alternative power resources, such as getting more power from other remote sources. It is always cheaper to buy power in advance as opposed to spot rates. Google has been known to fulfil its commitment to buy green power: from wind, solar and hydro resources.
Bottom line: the AI driven prediction and action research presented in DeepMind blog post’s presents with 20% increased profitability of wind farm. When wind generated power cost is going down, it becomes a stronger competitor of fossil fuel power resources, and increases the attractiveness of carbon free, renewable energy investments.