CleanTech Client Develops Machine Learning Models to Predict Renewable Energy Power Output

Our client is an energy services provider and electricity producer focusing on sustainable and climate-friendly energy sources and operations.

Business Challenges

The share of renewable energy resources like solar photovoltaic (PV), geothermal, tidal waves, wind power, and biomass has been increasing rapidly, predicted to grow at a CAGR of 8.4% from 2021 to 2030. 

Currently most popular types of renewable energy sources like solar and wind are highly variable and the resulting fluctuations in the generation capacity cause instability to power grids. 

During the energy generation process the stability of the power grid relies on the equilibrium between demand and supply. This makes it necessary to generate power when resources are available and store it for later use in order to achieve consumption based renewable energies alone. 

Solution

Wind and solar PV energy are expensive to store so careful energy generation management is required. In this case machine learning resolves two main challenges:

Forecasting power output from a renewable energy power plant is crucial as this depends on various non-human-controllable factors such as environmental parameters. Depending on the energy source it uses, the power plant exhibits certain characteristics that enable machine learning techniques for prediction purposes.

Through the creation of the solution, the following AWS services were applied: AWS IoT Core, AWS IoT SiteWise, AWS Lambda, Amazon SageMaker.

Results

Before:

• The heuristic pipeline for estimation of the equilibrium between demand and supply wasn’t accurate enough
• More energy was consumed from the traditional grid due to the lack of an energy management system and forecasting for renewable sources, resulting in increased CO2 emissions.

After:

• Machine learning models to predict PV output and the consumption for 3 hours ahead were developed with 10% and 5% forecasting error respectively
• Eventually, combining these two models we developed a control strategy for battery charging and reduced grid consumption by 40%