Prediction of solar radiance to improve solar power network management
By Mila | Published on March 11, 2022
Problem and objectives
Problem
- Search for precise weather models for the company’s needs (short-term prediction);
- Having to predict the quantity of solar radiance with more recurring granularity.
Objectives
- Provide a detailed and precise forecast model.
Proposed solution
Solution
Mila is a Montréal-based research institute that specializes in artificial intelligence. As part of a partnership project with Hydro-Québec, Mila was given the mandate to create a model capable of predicting the quantity of solar radiance at a given point in northeastern North America. Solar radiance measurements will then be used to determine the quantity of electricity that could be generated through photovoltaic panels. This solar collector functions as a direct current electric generator in the presence of solar radiation.
Principle
Compared to standard weather models, the model developed by Mila is able to generate predictions every 15 minutes, within a time horizon of 0 to 6 hours. Using satellite images, the predictive model shows the solar energy potential for each pixel 4 km2 in size.
Role of AI
The algorithm developed by Mila uses AI to accurately predict the potential quantity of solar energy at a given point. The latter is found in the following categories:
- Predictive models;
- Algorithm;
- Decision support.
The techniques used to create machine learning and deep learning algorithms are:
- Deep neural networks;
- Convolutional neural networks;
- Computer vision.
Impacts
Main outcomes
Impact on Hydro-Québec:
- Obtaining a predictive tool for potential commercial use;
- Predictions every 15 minutes within a time horizon of 0 to 6 hours.
Impact on Mila:
- Development of remote sensing knowledge;
- Refinement of skills in data preprocessing and use of specific infrastructures.
Generation and dissemination of new knowledge
Knowledge sharing between Mila and Hydro-Québec researchers.
Challenges takled
- Data accessibility;
- Quality of weather station data;
- Assessment and cleansing of data;
- Limited computation capacity.
Conditions for success
Team mobilised
The project’s AI team was made up of Hydro-Québec and Mila. researchers who specialize in:
- Remote sensing and satellite image analysis;
- Machine learning;
- Deep learning.
Collaborations
A major AI player in Québec, Mila has a multitude of ecosystem partners. These include:
- Dialogue;
- ServiceNow;
- Horoma AI;
- Imagia;
- Valence;
- Keatext;
- And more.
The research institute also has international partners such as:
- Microsoft;
- Google;
- IBM;
- Samsung;
- And more.
Mila collaborates with research institutes and universities, including:
- McGill University;
- Université de Montréal;
- Polytechnique Montréal;
- École des hautes études commerciales de Montréal (HEC Montréal);
- Montreal Clinical Research Institute (IRCM);
- And more.
Funding received
Grant received from the Ministère de l’Économie et de l’Innovation du Québec (MEI).
Project stages
Data analysis and understanding;
Literature review;
Development of an experimental protocol;
Algorithm development phases;
Model testing and training;
Report provided to the client.
Context of the solution
Use cases
The solution developed by Mila can be used in the following situations:
- Use by the energy trading floor;
- Identifying potential locations for the installation of solar panels;
- Short-term solar energy predictions.
Customers
Mila’s services are available to both the private and public sectors. They can pertain to a variety of industries such as:
- Finance and insurance;
- Information industry and cultural industry;
- Professional, scientific and technical services;
- Management of companies and businesses;
- Health care and welfare;
- Public administration;
- Utilities.