In this challenge, we provide you with a dataset containing consumption/production profiles. Your task is to develop a predictive model that can accurately forecast energy consumption/production patterns. The model needs to be deployed on a resource-constrained edge device, simulating real-world constraints. The targeted edge device is CLEMAP. For sake of simplicity, a virtual machine (VM) could be used to emulate the edge device with limited resources.

To make the deployment process simple, we ask you to use NuvlaEdge technology. The main goal is to deploy the developed forecasting module using NuvlaEdge, a powerful edge computing platform. The objective is to optimize and efficiently use resources while achieving accurate predictions.

The evaluation of your solution will be based on the following criteria:

Model Performance: The accuracy of your predictive model will be assessed using the Mean Squared Error (MSE) score. We are looking for models that can effectively capture the energy consumption/production patterns and generate accurate forecasts.

Overhead: Your solution must be deployed on a limited resources device. It must also “cohabit” with NuvlaEdge. Avoiding overload and efficiently using the available resources will be crucial. It is therefore paramount that the footprint (CPU and memory) of your solution is as low as possible.

Optimisation: We encourage you to take into account optimization concerns. This includes strategies to minimize resource contention, reduce computational complexity, and optimize memory usage, all while maintaining accurate predictions.

Summary steps for the challenge:

1. Develop the forecasting module as a Docker container: Create a predictive model for energy consumption/production based on the provided dataset, and encapsulate it along with its dependencies into a Docker container.
2. Deploy the forecasting container: Use the NuvlaEdge platform to deploy the developed forecasting module.
3. Monitor/Optimise resource usage
4. Evaluate model performance: Assess the performance of your deployed model using evaluation metrics such as the Mean Squared Error (MSE) score.

By participating in this challenge, you will gain valuable experience in addressing the resource constraints and optimization challenges inherent in edge computing environments. Your contributions will drive advancements in energy forecasting and pave the way for efficient deployment of predictive models on edge devices.

Challenge Owner: HES-SO

• Mohamad Moussa
• Nabil Abdennadher
• Philippe Glass