The challenge was to minimise costs for single & multiple family households through self-devised energy management methods, and provide the tenants and landlord(s) with utility functions. We were provided real and/or realistic data to work with from industry experts. We won first place with our solution called "Bright Grid". To summarise, we formulated 2 mathematical optimization problems and used multiple forecasting-related AI solutions to tackle the challenge. My role in the team was full stack development, cloud deployment, and universal forecasting experimentation. Our main mathematical optimization was minimising the energy cost through distributing the load across battery, PV, and grid using forecasted data and under reasonable constraints. Our secondary optimization was to help the landlord increase their profits when selling electricity to the tenants. We mainly used interior-point methods for optimization, for we found them to be sufficient for PoC. We considered a later switch to ADMM. Our forecasts were made through both real and mocked data, using Moirai (universal multivariate time series forecasting), conventional neural networks, etc. We deployed on azure in two separate containers (BE, FE).
09/2024 · Thüga Solutions · Munich · Germany
1st Place at Thüga Solutions Hackathon
Cost minimization for tenants and landlords through energy management using AI (multivariate forecasters, neural networks, …), and Mathematical Optimization (interior-point methods) (Role: FullStack, Cloud, RL).