Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Energy-Workload Coupled Migration Optimization Strategy for Virtual Power Plants with Data Centers Considering Fuzzy Chance Constraints

Jia-Kai Wu, Zhi-Wei Liu, Yong Zhao, Yan-Wu Wang, Fan-Rong Qu, Chaojie Li

TL;DR

The paper tackles precise demand response for CDL-based DR by jointly optimizing energy sharing and workload migration across a network of geo-distributed data centers within a virtual power plant. It develops an antisymmetric bidirectional migration framework, converts fuzzy chance constraints into deterministic equivalents to obtain a tractable SOCP formulation, and employs a distributed ADMM solver to handle high-dimensional coupling. An improved Shapley-value-based profit allocation balances fairness and scalability, with a VPPO intermediary reducing computation from exponential to linear in the number of VPPs. Simulations on Google's DC data demonstrate improved CDL tracking and reduced operational costs, validating the approach’s scalability and practical relevance for large-scale VPP-DC partnerships.

Abstract

This paper proposes an energy-workload coupled migration optimization strategy for virtual power plants (VPPs) with data centers (DCs) to enhance resource scheduling flexibility and achieve precise demand response (DR) curve tracking. A game-based coupled migration framework characterized by antisymmetric matrices is first established to facilitate the coordination of cross-regional resource allocation between VPPs. To address the challenge posed to conventional probabilistic modeling by the inherent data sparsity of DC workloads, deterministic equivalent transformations of fuzzy chance constraints are derived based on fuzzy set theory, and non-convex stochastic problems are transformed into a solvable second-order cone program. To address the multi-player interest coordination problem in cooperative games, an improved Shapley value profit allocation method with the VPP operator as intermediary is proposed to achieve a balance between theoretical fairness and computational feasibility. In addition, the alternating direction method of multipliers with consensus-based variable splitting is introduced to solve the high-dimensional non-convex optimization problem, transforming coupled antisymmetric constraints into separable subproblems with analytical solutions. Simulations based on real data from Google's multiple DCs demonstrate the effectiveness of the proposed method in improving DR curve tracking precision and reducing operational costs.

Energy-Workload Coupled Migration Optimization Strategy for Virtual Power Plants with Data Centers Considering Fuzzy Chance Constraints

TL;DR

The paper tackles precise demand response for CDL-based DR by jointly optimizing energy sharing and workload migration across a network of geo-distributed data centers within a virtual power plant. It develops an antisymmetric bidirectional migration framework, converts fuzzy chance constraints into deterministic equivalents to obtain a tractable SOCP formulation, and employs a distributed ADMM solver to handle high-dimensional coupling. An improved Shapley-value-based profit allocation balances fairness and scalability, with a VPPO intermediary reducing computation from exponential to linear in the number of VPPs. Simulations on Google's DC data demonstrate improved CDL tracking and reduced operational costs, validating the approach’s scalability and practical relevance for large-scale VPP-DC partnerships.

Abstract

This paper proposes an energy-workload coupled migration optimization strategy for virtual power plants (VPPs) with data centers (DCs) to enhance resource scheduling flexibility and achieve precise demand response (DR) curve tracking. A game-based coupled migration framework characterized by antisymmetric matrices is first established to facilitate the coordination of cross-regional resource allocation between VPPs. To address the challenge posed to conventional probabilistic modeling by the inherent data sparsity of DC workloads, deterministic equivalent transformations of fuzzy chance constraints are derived based on fuzzy set theory, and non-convex stochastic problems are transformed into a solvable second-order cone program. To address the multi-player interest coordination problem in cooperative games, an improved Shapley value profit allocation method with the VPP operator as intermediary is proposed to achieve a balance between theoretical fairness and computational feasibility. In addition, the alternating direction method of multipliers with consensus-based variable splitting is introduced to solve the high-dimensional non-convex optimization problem, transforming coupled antisymmetric constraints into separable subproblems with analytical solutions. Simulations based on real data from Google's multiple DCs demonstrate the effectiveness of the proposed method in improving DR curve tracking precision and reducing operational costs.

Paper Structure

This paper contains 30 sections, 1 theorem, 66 equations, 9 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. System Model and Problem Formulation
  3. System Description
  4. Basic Mathematical Bodel of Each Component
  5. DC Model
  6. PV and BESS Models
  7. CDL-based DR Mechanism
  8. Bidirectional Migration Strategies and Uncertainty Model Construction
  9. Deterministic Bidirectional Migration Model
  10. Independent Operation Costs
  11. Coupled Migration Strategy Based on Cooperative Game
  12. Improved Shapley Value-Based Benefit Sharing Strategy
  13. Uncertainty Modeling
  14. Solution Methods
  15. Solution Method for Independent Operation Scenario
  16. ...and 15 more sections

Key Result

Theorem 1

If the function takes the following form: where $\zeta_k$ is a trapezoidal fuzzy variable, $k=1,2,\cdots t, t\in R$. $r_{k1}-r_{k4}$ are trapezoidal membership parameters. When $\beta\geq 1/2$, the crisp equivalent of $\text{Cr}\left\{g\left(x,\varepsilon\right)\leq0\right\}\geq\beta$ is: where

Figures (9)

  • Figure 1: Framework for geo-distributed DCs participating in DR
  • Figure 2: PV power generation in different geographical locations
  • Figure 3: DC interactive workload requests
  • Figure 4: Convergence curve of the objective function
  • Figure 5: Scalability comparison of the proposed optimization and allocation algorithms with respect to the number of VPPs.
  • ...and 4 more figures

Theorems & Definitions (1)

  • Theorem 1