⚡ electrical engineering

Efficient Multi-Market Scheduling of Virtual Power Plants via Spectral Representation of Uncertainty

This paper proposes an efficient stochastic multi-market scheduling framework for Virtual Power Plants that utilizes intrusive Polynomial Chaos Expansion to represent uncertainty in the spectral domain, achieving solution quality comparable to scenario-based methods while reducing computational effort by up to 137 times and providing an open-source tool for automated spectral reformulation.

Original authors: Lorenzo Zapparoli, Blazhe Gjorgiev, Giovanni Sansavini

Published 2026-05-05

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Lorenzo Zapparoli, Blazhe Gjorgiev, Giovanni Sansavini

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a Virtual Power Plant (VPP) as a smart conductor leading an orchestra of small, scattered musicians. These musicians aren't traditional power plants; they are rooftop solar panels, home batteries, electric vehicles, and heat pumps scattered across a neighborhood. The conductor's job is to tell this orchestra exactly how much energy to sell to the grid and how much to save for emergencies, all while trying to make the most money.

The problem is that the musicians are unpredictable. The sun might hide behind a cloud, a neighbor might plug in their electric car unexpectedly, or the price of electricity might jump up or down. This uncertainty makes the conductor's job incredibly hard.

The Old Ways: Guessing vs. Being Too Cautious

Before this paper, conductors had two main ways to handle this uncertainty, and both had big flaws:

The "Scenario" Method (The Over-Prepared Chef):
Imagine a chef trying to cook for a party. To be safe, they try to simulate every single possible combination of guests: 10 people, 11 people, 12 people, with different allergies, different appetites, etc. They run thousands of simulations to find the perfect menu.
- The Flaw: This takes forever. As the number of guests (or energy sources) grows, the number of simulations explodes. The computer gets overwhelmed, and the chef runs out of time to cook.
The "Robust" Method (The Paranoid Chef):
This chef assumes the absolute worst-case scenario will happen every time. They assume the sun will never shine, the car will never charge, and prices will always be terrible. They cook a tiny, safe meal just in case.
- The Flaw: This is too conservative. They miss out on making money because they are too scared to take any risks.

The New Solution: The "Spectral" Recipe

This paper introduces a clever new way to solve the problem using something called Polynomial Chaos Expansion (PCE).

Think of the uncertainty (the weather, the prices, the demand) not as a chaotic mess of random numbers, but as a musical chord. Just as a complex sound wave can be broken down into a few simple, pure notes (frequencies), this method breaks down the complex uncertainty into a few simple mathematical "notes" (polynomials).

Instead of simulating thousands of different "what-if" scenarios (like the over-prepared chef), the new method calculates the amplitude of these few notes.

The Analogy: Instead of trying to predict the exact path of every single raindrop in a storm, the conductor just calculates the overall "shape" of the storm using a few key mathematical curves.
The Result: This turns a messy, impossible-to-solve problem into a clean, organized, and fast-to-solve math problem.

What the Paper Actually Found

The researchers tested this new "spectral" method on a real-world example: a virtual power plant in Switzerland with solar panels, batteries, and electric cars. They compared it against the best existing "scenario" method.

Here is what happened:

Speed: The new method was 137 times faster in terms of total computer effort. While the old method needed to run on 50 computer processors for hours, the new method solved the same problem on a single processor in less than an hour.
Accuracy: The new method made decisions (bids) that were almost identical to the super-accurate, slow method. In fact, it was even better at predicting the right amount of energy to bid than the old method when the old method didn't use enough simulations.
Memory: The old method needed a massive amount of computer memory (up to 1.5 Terabytes for the most accurate version). The new method needed only about 7 Gigabytes—roughly the size of a standard laptop's memory.

The "Magic Tool"

Finally, the authors didn't just solve this one problem; they built a free, open-source tool (like a universal translator).

The Analogy: Imagine they built a machine that can take any recipe for a complex problem (involving uncertainty) and automatically translate it into this efficient "spectral" language.
The Benefit: Other researchers and engineers don't need to be math geniuses to use this. They can just plug in their own problem, and the tool does the heavy lifting of converting it into the fast, efficient format.

Summary

In short, this paper found a way to help virtual power plants make smart, profitable decisions in an uncertain world without needing supercomputers or waiting days for results. It swaps the "brute force" of simulating thousands of scenarios for a "smart math" approach that breaks uncertainty down into simple, manageable pieces, delivering the same (or better) results in a fraction of the time.

Technical Summary: Efficient Multi-Market Scheduling of Virtual Power Plants via Spectral Representation of Uncertainty

Problem Statement

As distributed energy resources (DERs) such as solar PV, heat pumps, electric vehicles, and batteries increase in penetration, Virtual Power Plants (VPPs) are essential for aggregating these assets to provide grid flexibility. However, scheduling VPPs across sequential electricity markets (e.g., day-ahead energy and reserve capacity markets) under uncertainty presents a significant computational challenge.

Existing approaches face a tradeoff between economic optimality and tractability:

Stochastic Optimization: While probabilistically optimal, these methods rely on scenario approximation. As the number of scenarios increases to capture uncertainty accurately, the model size explodes, rendering large-scale problems computationally intractable.
Robust Optimization: These methods are more tractable but often yield overly conservative decisions that fail to capture the true probabilistic nature of the process, leading to suboptimal economic performance.

The paper addresses the gap of finding a scheduling framework that preserves the economic optimality of stochastic programming while achieving the computational efficiency of robust methods, specifically for multi-market VPP operations involving complex network constraints.

Methodology

The authors propose a novel framework that reformulates the two-stage stochastic VPP scheduling problem using Intrusive Polynomial Chaos Expansion (PCE).

1. Spectral Reformulation:
Instead of discretizing uncertainty into a finite set of scenarios, the method represents random variables (uncertain parameters like prices, weather, and demand) and second-stage decision variables as linear combinations of orthogonal polynomial basis functions.

First-Stage Decisions: Market bids (day-ahead energy and reserve capacity) are treated as deterministic variables (constant mode in the spectral domain).
Second-Stage Decisions: Operational decisions (DER dispatch, reserve activation, imbalance positions) are expanded into a truncated PCE series.
Galerkin Projection: The governing equations of the stochastic program are projected directly onto the polynomial basis using the stochastic Galerkin method. This transforms the original stochastic problem into a deterministic, low-dimensional spectral counterpart.

2. Constraint Handling:

Equality Constraints: Enforced exactly in the coefficient space via the Galerkin projection, preserving the algebraic structure of the original linear constraints.
Inequality Constraints: Reformulated as chance constraints and approximated using a moment-based Second-Order Cone (SOC) formulation. This ensures tractability while maintaining a specified probability of constraint satisfaction.

3. Resulting Model:
The reformulation yields a convex Second-Order Cone Program (SOCP) in the coefficient space. For linear VPP models, a first-degree expansion ( $N_d=1$ ) is sufficient, resulting in a problem size that scales linearly with the number of uncertainty sources rather than exponentially with the number of scenarios.

4. Open-Source Tool:
To address implementation complexity, the authors developed and released SpectralStochOpt, an application-agnostic projection tool. This tool automates the derivation of deterministic spectral counterparts for generic single- and two-stage stochastic programs, handling basis generation, variable expansion, and constraint projection.

Key Contributions

Spectral Reformulation of Multi-Market Scheduling: The paper presents the first application of intrusive PCE to multi-market VPP scheduling. It successfully models the joint optimization of day-ahead and reserve capacity market bids alongside real-time DER dispatch and network constraints.
Computational Efficiency with High Accuracy: The framework demonstrates that the spectral approach can achieve solution quality comparable to state-of-the-art scenario-based stochastic optimization while drastically reducing computational effort.
Generalized Projection Tool: The release of the open-source tool lowers the barrier to entry for intrusive PCE, enabling its application to a broad range of stochastic programs beyond the specific VPP case study.

Results

The framework was tested on a realistic Swiss low-voltage grid with a portfolio of 83 kW non-dispatchable demand, 375 kWp PV, 147 kW heat pumps, 305 kW battery storage, and EVs. It was benchmarked against a scenario-based approach using Latin hypercube sampling and Benders decomposition.

Computational Performance:
- The intrusive PCE approach solved the problem in 0.72 core-hours on a single core.
- The scenario-based benchmark required between 24.25 and 454.83 core-hours (using 50 parallel cores) to achieve comparable or slightly better accuracy, depending on the number of scenarios ( $N_s$ ).
- This represents a reduction in computational effort of up to 137 times compared to the scenario-based method.
- Memory usage was reduced from ~1.55 TB (for 2000 scenarios) to 7 GB for the PCE approach.
Solution Quality:
- Objective Value: The PCE solution deviated by only 2.8% from the high-accuracy scenario reference (2000 scenarios).
- First-Stage Decisions: The PCE method produced market bids (day-ahead energy and reserve capacity) with lower Root Mean Square Error (RMSE) than scenario-based solutions using fewer than 1000 scenarios. The PCE bid trajectories remained closely aligned with the reference mean across the 24-hour horizon.
Constraint Validation:
- Empirical validation on 10,000 uncertainty realizations confirmed that the moment-based SOC approximation for chance constraints resulted in negligible violation rates (maximum exceedance probability of 5.17%), validating the Gaussian assumption used for the safety factor.

Significance and Claims

The paper claims that intrusive PCE offers a viable alternative to the "curse of dimensionality" inherent in scenario-based stochastic optimization. By preserving the probabilistic structure of the problem without enumerating scenarios, the method enables VPP operators to:

Manage large-scale DER portfolios with limited computational resources.
Compute day-ahead offers closer to market gate-closure times, utilizing more recent forecast data.
Avoid the conservatism of robust optimization while maintaining computational tractability.

The authors modestly note limitations, including the approximate nature of the inequality constraint reformulation (dependent on residual distribution assumptions) and the current focus on linear, two-stage models with independent uncertainty sources. They identify future work in handling correlated inputs, nonlinear models, and developing decomposition schemes for the coefficient-space problem. The release of the projection tool is highlighted as a key enabler for the broader adoption of spectral methods in power systems research.