Design of an Autonomous Intelligent Demand-Side Management System by using Electric Vehicles as Mobile Energy Storage Units by Means of Evolutionary Algorithms

Edgar Galván López, Marc Schoenauer, Constantinos PatsakisIJCCI (ECTA) 2015 · DOI

TL;DR

This paper proposes an intelligent system that uses electric vehicle (EV) batteries as mobile energy storage units to help manage electricity demand in homes. By employing evolutionary algorithms, the system dynamically optimizes when and how much energy to draw from EVs to reduce peak loads on the power grid, ensuring EVs are still ready for their owners' commutes.

What Problem Does This Paper Address?

The electricity grid faces a significant challenge in managing fluctuating demand, particularly during "peak hours" when everyone wants to use electricity simultaneously. This leads to increased costs, strain on infrastructure, and a greater reliance on less efficient power generation. Demand-Side Management (DSM) aims to mitigate these issues by influencing consumer behavior. While traditional DSM methods like smart pricing exist, they have limitations. The rise of electric vehicles presents a unique opportunity: their large batteries can act as distributed energy storage. However, effectively integrating EVs into the grid for DSM is complex due to their dynamic usage patterns (arrival/departure times, charging needs) and the need to ensure they are always ready for their owners. This paper tackles the problem of designing an autonomous system that can intelligently leverage EV batteries for DSM, balancing grid needs with user convenience.

Methods

The core of this research lies in the application of Evolutionary Algorithms (EAs), specifically Differential Evolution (DE), to solve this complex optimization problem. EAs are inspired by natural selection and are adept at finding solutions in dynamic and challenging environments where traditional analytical methods might fail.

Here's a breakdown of their approach:

  • Representation: Instead of simple on/off switches for appliances, the system uses a continuous representation where each element in a vector represents the specific amount of energy to be drawn from an EV's battery at a given time interval (15 minutes in this study). This allows for fine-grained control. The overall solution is represented as a matrix where rows correspond to individual EVs and columns to time slots.
  • Differential Evolution (DE): DE is chosen for its efficiency in optimizing continuous search spaces. It works by maintaining a population of potential solutions (individuals) and iteratively improving them through mutation and crossover operations. The paper explores various mutation strategies to find the most effective one.
  • Fitness Function: This is crucial for guiding the EA. The authors propose two fitness functions:
  • Initial Fitness Function (Equation 7): This function aims to maximize the energy drawn from EV batteries while ensuring a minimum state of charge (SoC) is maintained for each EV to guarantee it can complete its commute.
  • Enriched Fitness Function (Equation 8): This builds upon the first by adding a component that specifically prioritizes drawing more energy from EVs during high-peak periods, thereby actively reducing the load on the substation transformer. This function also incorporates a "critical time slot" threshold to identify these high-demand periods.

Datasets

The paper does not use pre-existing public datasets. Instead, it employs a simulated dynamic scenario to evaluate the proposed system. This scenario involves:

  • Household Units: Simulated consumption for 40 and 80 household units.
  • Appliances: Each household unit uses between 10 and 20 appliances.
  • EV Penetration: Approximately 20% of household units are assumed to have an EV.
  • Time Granularity: Decisions are made every 15 minutes.
  • Simulation Period: 30 simulated working days.
  • EV Parameters: Arrival and departure times, and initial State of Charge (SoC) for EVs are varied dynamically within defined ranges to mimic real-world usage.

Key Insights and Ideas

  • EVs as Distributed Batteries: The paper powerfully demonstrates that EV batteries can be more than just transportation; they can be a valuable, distributed energy storage resource for the grid.
  • Evolutionary Algorithms for Dynamic Optimization: The success of DE highlights its suitability for complex, real-world problems with dynamic constraints, like managing EV usage for grid stability.
  • Fine-Grained Control is Key: Moving beyond simple on/off controls to precisely managing energy draw from EVs allows for more nuanced and effective DSM.
  • The Power of an Enriched Objective: Simply maximizing EV energy usage isn't enough. The enriched fitness function, which explicitly targets high-peak periods, shows a significant improvement in reducing grid strain.
  • Balancing Grid Needs and User Convenience: The system successfully demonstrates the ability to draw significant energy from EVs while still ensuring they meet their minimum charge requirements for daily commutes, a critical factor for user acceptance.

Key Results

The study presents compelling results from simulations:

  • Maximizing EV Energy Usage: The initial fitness function (Equation 7) successfully maximized the energy drawn from EV batteries while ensuring a minimum SoC for each EV. However, the consumption pattern was somewhat random.
  • Reducing Peak Loads: The enriched fitness function (Equation 8) showed a marked improvement. It intelligently shifted EV energy consumption to high-peak periods, effectively reducing the load on the substation transformer. This is visually represented by a more patterned consumption of EV energy during peak times.
  • Lower Peak-to-Average Ratio (PAR): The PAR, a measure of grid load stability, was significantly lower when using the enriched fitness function (Equation 8). This indicates a smoother, more balanced load on the grid. For instance, Figure 4 shows a substantially lower PAR for the enriched function compared to the basic one.
  • Scalability: The approach was tested with both 40 and 80 household units, demonstrating its potential scalability.

Limitations and Future Work

  • Simulation-Based: The results are based on simulations. Real-world deployment would require validation with actual grid data and EV charging infrastructure.
  • EV Availability: The model assumes a certain percentage of EVs are available for grid interaction. The actual availability might fluctuate more significantly in practice.
  • User Behavior Modeling: While arrival and departure times are varied, more sophisticated models of user behavior and charging preferences could enhance realism.
  • Communication Infrastructure: The paper implicitly assumes a robust communication network for the DSM system to interact with EVs.
  • Future Work: The authors suggest exploring other research areas in DSM, particularly those that can leverage "new" available technologies. They also implicitly point towards the need for further refinement of the optimization algorithms and fitness functions to handle even more complex grid dynamics and user preferences.

Summary updated

BibTeX
@inproceedings{lopez2015design,
  author = {Edgar Galván López and Marc Schoenauer and Constantinos Patsakis},
  title = {{Design of an Autonomous Intelligent Demand-Side Management System by using Electric Vehicles as Mobile Energy Storage Units by Means of Evolutionary Algorithms}},
  booktitle = {IJCCI (ECTA)},
  year = {2015},
  doi = {10.5220/0005607401060115},
}
Figure