Modelling Energy Systems

Project description

This project focuses on developing adaptive EV charging algorithms using agent-based modelling (ABM) to minimize grid congestion and maximize the use of renewable energy. The simulation will model interactions between EVs, charging stations, and renewable energy sources like solar and wind, considering factors such as weather conditions and time of day. The project will compare scenarios with and without smart charging, using both individual charger and multi-charger algorithms managed by a Charging Point Operator (CPO). The simulation will use real-world data for weather and energy consumption and be containerized for scalability and easy deployment.

Context

The project addresses the increasing challenges of power grid management due to the global shift towards renewable energy and the rising adoption of electric vehicles (EVs). While the transition to renewable sources like solar and wind power offers a way to reduce carbon emissions, the intermittent nature of these sources complicates the balance between energy supply and demand. Current EV charging strategies, such as static and dynamic pricing, often fail to fully account for real-time fluctuations in renewable energy availability, leading to underutilized resources and potential grid instability.

Results

The agent-based simulations demonstrated the effectiveness of real-time adaptive charging algorithms in managing grid congestion and optimizing renewable energy use. Simulations were conducted over a week for each season, comparing scenarios with and without congestion-aware charging. A baseline was established by running the simulation without any congestion-aware charging, revealing typical daily energy consumption patterns with fluctuations influenced by weather, time, and user behavior. Power usage typically peaked during daytime hours and declined at night, with renewable energy sources fluctuating throughout the week, mirroring the variability in solar and wind energy generation.

The weather data, sourced from the Dutch PV Portal's Meteorological Data Portal, included hourly measurements of solar irradiance, ambient temperature, and wind speed, crucial for simulating renewable energy production. The simulations showed seasonal differences, highlighting how solar irradiance, temperature, and wind speed impact renewable energy production throughout the year. Congestion-aware algorithms were effective in smoothing energy demand curves, reducing peak energy demand by 33% during evening hours. The use of renewable energy increased during peak periods, demonstrating the system's ability to reduce reliance on non-renewable energy sources.

The comparison of the Individual Charging Algorithm (ICA) and the Charging Point Operator (CPO) Algorithm showed different performance characteristics. The ICA reached maximum power output during low congestion but struggled to balance loads efficiently under high demand. In contrast, the CPO algorithm offered better load balancing and mitigated congestion more effectively by adjusting charging speeds across multiple stations. However, the CPO algorithm sometimes struggled to maintain high charging speeds during periods of limited renewable energy, especially in winter. In contrast, the ICA does not scale back charging rates during periods of low renewable energy, so it can achieve higher charging speeds when renewable energy is abundant.

Scalability tests were performed on different machines, with varying hardware specifications. The results showed that the model was more performant on the server and desktop machines compared to the laptop. The startup time increased significantly when more complex network configurations were used, but the model maintained accuracy and real-time applicability across different scales. While larger networks demanded more computational power, the model could execute efficiently, with a minimal impact on real-time systems.