A Predictive Flexibility Aggregation Method for Low Voltage Distribution System Control

This paper proposes a predictive, privacy-preserving control scheme for low-voltage distribution systems that aggregates residential flexibility into real-time admissible power charts through offline multiparametric optimization, enabling efficient and cost-effective network management validated on a 43-bus test case.

The Gist

Imagine a neighborhood where every house has a solar roof, a battery in the garage, and an electric car. Now, imagine the power company (the DSO) needs to keep the electricity flowing smoothly without blowing fuses or causing blackouts, but they can't just walk into every house and flip switches. They need the houses to help out, but they don't want to know the family's private schedule or how much they hate paying for electricity.

This paper presents a smart "middleman" system that solves this problem. It's like a sophisticated traffic control system for electricity that lets houses volunteer their help without revealing their secrets.

Here is how it works, broken down into simple concepts:

1. The Problem: Too Many Variables, Too Little Time

In the old days, power companies just built bigger wires to handle more electricity. But now, with solar panels and electric cars, the flow of electricity changes every second.

• The Challenge: If the power company tries to calculate the perfect plan for 43 houses in real-time, their computers would crash. It's like trying to solve a giant Sudoku puzzle while driving a car at 100 mph.
• The Privacy Issue: The power company can't ask every homeowner, "How much does it cost you to turn off your AC?" That's too invasive.

2. The Solution: The "Menu" (Flexibility Chart)

Instead of asking for specific instructions, the system asks every house to create a "Menu of Options" (called a Flexibility Chart).

Think of this like a restaurant menu:

• The Dish: "I can give you 5kW of power right now."
• The Price: "It will cost me $0.10 to do that."
• The Catch: "But if I do that, I might run out of battery later, which will cost me $0.50."

The house calculates this menu offline (while they are sleeping or doing nothing). They pre-calculate every possible scenario: What if the sun is bright? What if the price of electricity is high? What if my battery is half-full?

They turn all these complex math problems into a simple map (a chart) that says: "Here is what I can do, and here is how much it costs me."

3. The Two-Step Brain: "The Planner" and "The Doer"

The magic of this paper is splitting the thinking into two parts:

• Step A: The Long-Term Planner (Operational Planning)
  Imagine a chess player looking 10 moves ahead. This step looks at tomorrow's weather forecast and electricity prices. It answers the question: "If I use my battery now, will I regret it when the sun goes down and prices spike?"
  It creates a "Future Cost Map." This map tells the house: "Having a full battery right now is valuable because tomorrow morning is expensive."

• Step B: The Real-Time Doer (Local Aggregation)
  This is the "Doer" who looks at the current moment. It takes the "Future Cost Map" and the current weather (is the sun shining right now?) and updates the Menu.

  • Result: The house sends a fresh, up-to-the-second menu to the power company.

4. The Central Conductor

The power company (the central controller) collects these menus from all 43 houses.

• It doesn't need to know how the house calculated the menu or what their private battery settings are.
• It just looks at the "Price" on the menu.
• It picks the cheapest options from all the houses to balance the grid, ensuring no one's voltage gets too high or too low.

5. Why This is a Game-Changer

• Speed: Because the hard math is done beforehand (offline), the system reacts instantly. It's like having a GPS that already calculated the route before you started driving.
• Privacy: The power company never sees your personal data. They only see the "Menu" (the aggregated result). It's like ordering food without telling the chef your allergies; the chef just sees what you are willing to eat.
• Fairness: The system ensures that if a house is asked to do something difficult (like draining a battery), they are compensated fairly based on the "price" they set on their menu.

The Analogy: The Neighborhood Potluck

Imagine a neighborhood potluck where everyone brings a dish.

• Old Way: The organizer calls every house, asks what they have, and tries to coordinate who brings what while everyone is cooking. It's chaotic and slow.
• This Paper's Way: Every house pre-makes a list of what they can bring and how much effort it takes them.
  • House A: "I can bring a salad (low effort) or a roast (high effort, costs me time)."
  • House B: "I can bring dessert, but only if I don't have to drive far."
    The organizer looks at all the lists, picks the best combination to feed everyone, and tells House A, "Bring the salad," and House B, "Bring the dessert."
    The houses didn't have to stop cooking to talk to the organizer; they just handed over their pre-made list.

The Bottom Line

This paper proves that by doing the heavy math before the crisis happens, we can manage a complex, modern power grid in real-time. It keeps the lights on, saves money, protects privacy, and gets us one step closer to a world powered entirely by renewable energy.

Technical Summary

Here is a detailed technical summary of the paper "A Predictive Flexibility Aggregation Method for Low Voltage Distribution System Control."

1. Problem Statement

The increasing penetration of residential renewable energy sources (e.g., PV) and electrification of heating/mobility (e.g., EVs, heat pumps) creates significant challenges for Low-Voltage Distribution Networks (LV-DNs), specifically regarding voltage violations (overvoltage and undervoltage) and network congestion.

• The Challenge: Distribution System Operators (DSOs) cannot directly control or measure flexible assets (batteries, inverters) due to unbundling regulations and privacy concerns. They rely on residential units to report flexibility.
• The Gap: Existing methods for aggregating flexibility often:
  • Neglect reactive power flexibility.
  • Ignore the economic cost of activating flexibility.
  • Fail to account for future forecasts (day-ahead tariffs, production) in real-time decisions, leading to suboptimal scheduling.
  • Require heavy online computation, making them unsuitable for real-time control.

2. Methodology

The authors propose a Predictive Flexibility Aggregation (PFA) method that combines Multiparametric Optimization (MPO) with a hierarchical control structure. The approach divides the optimization horizon into two stages: Operational Planning (OP) and Real-Time Control.

A. Core Architecture

1. Offline Pre-computation (Decentralized):

   • Each residential unit solves an MPO problem offline.
   • This generates an explicit map (polyhedral critical regions) of optimal control policies and value functions as a function of varying parameters (e.g., current State of Charge, load, generation potential, electricity prices).
   • This eliminates the need for repetitive online optimization.

2. Operational Planning Stage (Parametric):

   • Solved periodically (e.g., every 15 minutes) using day-ahead forecasts (PV production, load, tariffs).
   • It calculates a cost-to-go function ($\Pi_{SoC}$) representing the future cost associated with the battery's State of Charge (SoC) at the end of the current market period.
   • This function is modeled as a piecewise linear function of the SoC.

3. Real-Time Control Stage:

   • Local Projection: Local controllers update the offline MPO solution with real-time measurements (current SoC, load, irradiance) and the $\Pi_{SoC}$ function from the OP stage. This "projects" the offline solution onto the PQ plane, generating a Flexibility Chart (admissible active/reactive power vs. cost).
   • Central Optimization: The DSO collects flexibility charts from all nodes. Using a Branch Flow Model, it solves a centralized optimization to determine optimal nodal setpoints that minimize total system cost (local costs + network losses) while respecting voltage constraints.
   • Disaggregation: The central setpoints are sent back to local controllers, which use the explicit offline MPO solution to instantly disaggregate the setpoints among local assets (PV, Battery) without further optimization.

B. Mathematical Formulation

• Objective: Minimize daily energy costs, including import/export costs, reactive power penalties, and battery degradation.
• Constraints: Power balance, inverter apparent power limits (linearized), battery SoC dynamics, and voltage limits.
• Key Innovation: The integration of the $\Pi_{SoC}$ function allows the real-time decision to anticipate future costs, ensuring that current flexibility activation is economically optimal relative to future needs.

3. Key Contributions

1. Predictive Aggregation: Unlike previous methods that rely on fixed cost functions or current states only, this method integrates day-ahead forecasts into the real-time flexibility assessment, enabling proactive management of flexibility.
2. Active and Reactive Power: The method aggregates both active and reactive power flexibility, providing a comprehensive view of the node's capabilities.
3. Privacy Preservation: User preferences and specific asset details are hidden within the aggregated flexibility chart sent to the DSO.
4. Computational Efficiency: By shifting heavy computations (MPO solving) offline and using explicit solutions for real-time projection, the method meets real-time requirements (sub-second execution) while remaining decentralized.
5. Economic Fairness: The method triggers the most cost-effective flexibility first and compensates prosumers based on the marginal value of their flexibility, rather than applying arbitrary quadratic penalties.

4. Results

The method was validated on a 43-node low-voltage network (derived from SimBench) with realistic residential profiles (PV, batteries, loads) and compared against:

• OMNI: An omniscient controller with perfect foresight (ideal benchmark).
• FA: A standard flexibility aggregation method relying only on current system state (no predictive planning).

Key Findings:

• Cost Performance: The proposed PFA method achieved a total daily cost of €42.88, significantly outperforming the FA method (€74.11) and approaching the optimal OMNI benchmark (€38.79).
• Battery Utilization: PFA utilized battery storage more effectively (443.49 kWh charged vs. 396.23 kWh for FA), reducing reliance on grid imports during peak pricing.
• Voltage Control: The proposed controller successfully maintained voltage levels within the prescribed limits (0.95–1.05 p.u.) for all nodes.
• Computation Time:
  • Operational Planning: ~7.77 seconds (every 15 mins).
  • Real-time Projection/Disaggregation: < 0.5 seconds.
  • Central Optimization: ~0.80 seconds.
  • Conclusion: The method is fully compatible with real-time control loops.

5. Significance

This paper presents a scalable, privacy-preserving, and computationally efficient framework for managing future low-voltage grids. By bridging the gap between long-term operational planning and real-time control through multiparametric optimization, it allows DSOs to harness distributed flexibility effectively without needing direct access to user data or incurring prohibitive computational costs. The approach paves the way for the widespread integration of prosumers in local electricity markets and ancillary service provision.