Paper: arXiv 2503.12305
Authors: Thiha Aung, Mike Ludkovski
Abstract
We develop a mathematical model for intraday dispatch of co-located wind-battery energy assets. Focusing on the primary objective of firming grid-side actual production vis-a-vis the preset day-ahead hourly generation targets, we conduct a comprehensive study of the resulting stochastic control problem across different firming formulations and wind generation dynamics. Among others, we provide a closed-form solution in the special case of a quadratic objective and linear dynamics, as well as design a novel adaptation of a Gaussian Process-based Regression Monte Carlo algorithm for our setting. Extensions studied include an asymmetric loss function for peak shaving, capturing the cost of battery cycling, and the role of battery duration. In the applied portion of our work, we calibrate our model to a collection of 140+ wind-battery assets in Texas, benchmarking the economic benefits of firming based on outputs of a realistic unit commitment and economic dispatch solver.
Complexity vs Empirical Score
- Math Complexity: 8.5/10
- Empirical Rigor: 7.5/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced stochastic control, SDEs, Gaussian Processes, and derives closed-form solutions under specific conditions, indicating high mathematical density. It also demonstrates high empirical rigor through calibration to 140+ real wind-battery assets in Texas and benchmarking against a realistic unit commitment solver.
Research Flowchart
flowchart TD
A["Research Goal: Optimal Intraday Battery Dispatch<br>for Co-located Wind-Battery Assets"] --> B{"Methodology: Stochastic Control Formulations"};
B --> C["Primary: Firming w.r.t. Day-Ahead Targets"];
B --> D["Extensions: Peak Shaving & Battery Cycling"];
C & D --> E{"Solution Approaches"};
E --> F["Closed-Form Solution: Quadratic Objective"];
E --> G["Algorithm: GP-Adapted Regret Monte Carlo"];
H["Data: 140+ Texas Wind-Battery Assets"] --> I["Calibration & Simulation"];
F & G & I --> J{"Benchmarking"};
J --> K["Economic Analysis: Unit Commitment & Economic Dispatch Solver"];
K --> L["Key Findings: Economic Benefits of Firming & Optimal Dispatch Strategies"];
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes a mathematical model for intraday dispatch of co‑located wind‑battery assets, provides a closed‑form solution for a quadratic objective with linear dynamics, and introduces a novel Gaussian‑Process based Regression Monte Carlo algorithm (SHADOw‑GP) for solving the stochastic contr
Source sentence
We develop a mathematical model for intraday dispatch of co-located wind-battery energy assets. Among others, we provide a closed-form solution in the special case of a quadratic objective and linear dynamics, as well as design a novel adaptation of a Gaussian Process-based Regression Monte Carlo algorithm for our setting.
Assumes 6
- Quadratic penalty on state and control variables is used to relax hard constraints in the admissible charge/discharge set. model
Source sentence
To solve our problem analyti-cally, we relax the hard constraints in Bt using a quadratic penalty on the state and control variables.
- Linear dynamics for the state of charge (SoC) process are assumed with 100 % efficiency (η = 1). model
Source sentence
Furthermore, we take a quadratic terminal cost and linearize the dynamics of It: 6 • Linear dynamics for the SoC process by considering η “ 1 (100% efficient) BESS: dIt “ Bt dt, I0 “ ι, (10) • Quadratic stepwise cost ¯f, with the firming criterion modified to approximate the hard state and control constraints with penalties c1, c2 ą 0.
- Quadratic terminal cost centered around a target SoC level is imposed. model
Source sentence
(11) • A quadratic terminal penalty centered around a target SoC level Itarget: ¯gpIT q “ P ¨ pIT ´ Itargetq2 with P ą 0.
- The control set is taken as unconstrained real‑valued after relaxation. model
Source sentence
(12) Under assumptions (1), (10)-(12), we obtain an HJB PDE with unconstrained real-valued control set: Bt ¯V ` inf ¯bPR ! αtpmt ´ xqBX ¯V ` ¯bBI ¯V ` 1 2σ2 t xpXmax ´ xqBXX ¯V ` px ´ ¯b ´ Mtq2 ` c1¯b2 ` c2pι ´ Imq2) “ 0 (13) with terminal condition V pT, x, ιq “ ¯gpιq. Remark 3.
Show 2 more
- Wind generation dynamics are calibrated from hourly actual and forecasted NREL data for 149 Texas wind units. data
Source sentence
16 4.1 Calibration of Wind Dynamics We utilize the reanalyzed NREL data set (Xu et al., 2017) for calibration of wind assets. The data set comprises hourly (∆t “ 1) actual and forecasted (as of noon on the day prior) wind generation profiles in 2018 for the 149 wind units in Texas-7k, with generation capacities ranging from 30MW to 352MW.
- Shocks in the wind dynamics are non‑Gaussian; a binned bootstrap and mixed distribution are used to model them. statistical
Source sentence
It remains to specify the exogenous i.i.d. shocks ϵk in the calibrated dynamics: Xℓ,d k`1 “ Xℓ,d k ` αℓ RpF ℓ,d k qpF ℓ,d k ´ Xℓ,d k q ` σℓ RpF ℓ,d k q ¨ ϵℓ k for k “ 0, 1, . . . , 23, (38) After calibrating αℓ ¨ and σℓ ¨ , the binned residuals Eℓ ¨ reveal that the shocks ϵℓ k are non-Gaussian (Supplementary Figure S3). First, we replace the Gaussian assumption with a binned bootstrap: ϵℓ k ” ϵℓ RpF ℓ,d k q, where
Data
- NREL reanalyzed wind generation data set synthetic public
Source sentence
16 4.1 Calibration of Wind Dynamics We utilize the reanalyzed NREL data set (Xu et al., 2017) for calibration of wind assets. The data set comprises hourly (∆t “ 1) actual and forecasted (as of noon on the day prior) wind generation profiles in 2018 for the 149 wind units in Texas-7k, with generation capacities ranging from 30MW to 352MW.
- Synthetic Texas‑7k transmission grid (digital twin of ERCOT) synthetic unknown
Source sentence
To provide empirical validation, we carry out a comprehensive case study, calibrating our model to realistic wind power production data from the synthetic Texas-7k transmission grid, a digital 2 twin of the Electric Reliability Council of Texas (ERCOT) grid.
Costs
Not applicable
Finds 5
- Hybridizing wind assets with a three‑hour BESS yields average percentage deviation reduction (DR) ranging from 40 % to 67 % across 149 assets, with higher reductions in southern Texas. performance figure from elsewhere in the paper
Source sentence
We observe DR ℓas low as 40% and as high as 67%, with a clear spatial pattern whereby higher reductions are observed in the southern and southeastern Texas, as indicated by the brighter yellow bubbles, which also correspond to larger hybrid assets. We observe lowest DR ℓ(%) in the Far North region of ERCOT.
- Increasing the degradation‑aware penalty λBL from 0 to 0.2 more than doubles the expected BESS lifetime (from ≈5 to ≈12 years) while reducing firming performance by about 25 %. performance
Source sentence
Increasing λBL from 0 to 0.2 more than doubles the expected BESS lifetime (from „5 to „12 years) due to shallower cycles, at the cost of about a 25% reduction in firming performance.
- The SHADOw‑GP algorithm can be trained on 24 randomly selected test days per asset, with each run taking less than 5 minutes. other
Source sentence
Since we have 149 hybrid assets, this totals to 149 ˆ 24 runs of SHADOw-GP, each run taking « 5 minutes.
- A controller that jointly minimizes firming and an asymmetric curtailment violation metric reduces expected cumulative violation by ~40 % with only a 6 % loss in deviation reduction. performance
Source sentence
The ECV response to λCM is highly non-linear, with a sharp decline between λCM “ 0 and 0.2. Setting λCM “ 1 reduces ECV by about 40% relative to λCM “ 0, while incurring only a 6% loss in EDR, indicating that the fCM criterion effectively mitigates curtailment with minimal firming performance decrease.
- Calibration reveals that wind forecast residuals are non‑Gaussian, exhibiting skewness and heavy tails, justifying the use of flexible noise models. statistical
Source sentence
Figure S3 displays histograms of bin-wise residuals Eℓ r after calibration for the Foard City wind unit. While residuals are centered around zero, their shapes deviate from Gaussianity in several bins, with heavier tails and skewness, particularly at intermediate and high-output regimes.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 15 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (94)
- Regression M onte C arlo for Microgrid Management (2019) — link
- Model-Free reinforcement learning with continuous action in practice (2012) — link
- Partial differential equation methods for stochastic dynamic optimization: an application to wind power generation with energy storage (2017) — link
- Integrating Sequential Hypothesis Testing into Adversarial Games: A Sun Zi-Inspired Framework (2025) — link
- 2023 Special Report on Battery Storage (2024)
- A model predictive control approach to the problem of wind power smoothing with controlled battery storage (2010) — link
- Valuing American Options by Simulation: A Simple Least-Squares Approach (2001)
- Simulation Methods for Stochastic Storage Problems: A Statistical Learning Perspective (2020) — link
and 86 more
- A toolbox on the distribution of the maximum of G aussian process (2013)
- Regress-Later M onte C arlo for Optimal Inventory Control with applications in energy (2018) — link
- Deep Learning Approximation for Stochastic Control Problems (2016)
- Solving high-dimensional partial differential equations using deep learning (2018) — link
- Deep backward schemes for high-dimensional nonlinear PDEs (2020)
- Deep Neural Networks Algorithms for Stochastic Control Problems on Finite Horizon: Numerical Applications ( 2022) — link
- A Multi-Factor Battery Cycle Life Prediction Methodology for Optimal Battery Management (2015) — link
- Optimal Battery Aging: An Adaptive Weights Dynamic Programming Algorithm (2018) — link
- Optimal operation of energy storage systems considering forecasts and battery degradation (2016)
- Defining a degradation cost function for optimal control of a battery energy storage system (2013) — link
- Trading with propagators and constraints: applications to optimal execution and battery storage (2024)
- Novel battery degradation cost formulation for optimal scheduling of battery energy storage systems (2022) — link
- Optimal Battery Control Under Cycle Aging Mechanisms in Pay for Performance Settings (2019) — link
- Equivalence of four-point and three-point rainflow cycle counting algorithms (2008)
- DGM: A deep learning algorithm for solving partial differential equations (2018)
- Extensions of the deep Galerkin method (2022) — link
- Deep Neural Networks Algorithms for Stochastic Control Problems on Finite Horizon: Numerical Applications (2021)
- Valuation of energy storage: an optimal switching approach (2010) — link
- Regression Monte Carlo for Impulse Control (2022) — link
- An alternative optimal strategy for stochastic model predictive control of a residential battery energy management system with solar photovoltaic (2021) — link
- Probabilistic spatiotemporal modeling of day-ahead wind power generation with input-warped Gaussian processes (2025) — link
- Self-Supervised Primal-Dual Learning for Constrained Optimization (2022)
- End-to-End Constrained Optimization Learning: A Survey (2021)
- Learning Constrained Optimization with Deep Augmented Lagrangian Methods (2024)
- DC3 : A learning method for optimization with hard constraints (2021)
- Enhancing Battery Storage Energy Arbitrage With Deep Reinforcement Learning and Time-Series Forecasting (2024)
- Stochastic predictive control of battery energy storage for wind farm dispatching: Using probabilistic wind power forecasts (2015) — link
- A stochastic continuous time model for microgrid energy management (2016) — link
- Optimal Dispatch of Hybrid Renewable–Battery Storage Resources: A Stochastic Control Approach (2024) — link
- Comparison of least squares M onte C arlo methods with applications to energy real options (2017) — link
- Optimal operation of a grid-connected battery energy storage system over its lifetime (2023) — link
- Scenario-based stochastic optimal operation of wind, photovoltaic, pump-storage hybrid system in frequency- based pricing (2015) — link
- Optimal Management of a Wind Power Plant with Storage Capacity (2018)
- Price impact and long-term profitability of energy storage — → 2410.12495
- Statistical learning for probability-constrained stochastic optimal control (2021) — link
- Hybrid Resources: Challenges, Implications, Opportunities, and Innovation (2021) — link
- Cooperation of Wind Power and Battery Storage to Provide Frequency Regulation in Power Markets (2017) — link
- Stochastic optimization of the daily operation of wind farm and pumped-hydro-storage plant (2012) — link
- A Price-Maker/Price-Taker Model for the Operation of Battery Storage Systems in Electricity Markets (2019) — link
- An energy storage assessment: Using optimal control strategies to capture multiple services (2015) — link
- Optimal Bidding Strategy of Battery Storage in Power Markets Considering Performance-Based Regulation and Battery Cycle Life (2016) — link
- Energy Storage Arbitrage Under Day-Ahead and Real-Time Price Uncertainty (2018) — link
- Control method evaluation for battery energy storage system utilized in renewable smoothing (2013) — link
- Using Battery Storage for Peak Shaving and Frequency Regulation: Joint Optimization for Superlinear Gains (2018) — link
- Gaussian processes for machine learning. (2006)
- A class of stochastic optimal control problems with state constraint (2002)
- Viscosity Solutions of Second Order Fully Nonlinear Elliptic Equations with State Constraints (1994)
- Hybrid Distributed Wind and Battery Energy Storage Systems (2022) — link
- Energy storage systems: a review (2022) — link
- Stochastic control with state constraints via the F okker– P lanck equation. Application to renewable energy plants with batteries (2024) — link
- Nonlinear Elliptic Equations with Singular Boundary Conditions and Stochastic Control with State Constraints. I. The Model Problem. (1989)
- Continuous-time Stochastic Control and Optimization with Financial Applications (2009)
- Deep Fictitious Play for Finding M arkovian N ash Equilibrium in Multi-Agent Games (2020) — link
- Unlocking the Flexibility of Hybrid Resources (2022)
- Battery Energy Storage Scheduling for Optimal Load Variance Minimization (2018)
- KrigHedge: Gaussian Process Surrogates for Delta Hedging (2021) — link
- Dynamic Programming (2010)
- Probabilistic Forecasts of Wind Power Generation by Stochastic Differential Equation Models (2016) — link
- Quantifying uncertainty with a derivative tracking SDE model and application to wind power forecast data (2021) — link
- Approximate Dynamic Programming: Solving the curses of dimensionality (2007)
- Energy Storage Price Arbitrage via Opportunity Value Function Prediction (2023) — link
- Reinforcement Learning Based Demand Charge Minimization Using Energy Storage (2023) — link
- Optimal Energy Storage Scheduling for Wind Curtailment Reduction and Energy Arbitrage: A Deep Reinforcement Learning Approach (2023)
- Multi-service battery energy storage system optimization and control (2022)
- Co-optimizing battery storage for the frequency regulation and energy arbitrage using multi-scale dynamic programming (2016)
- 2H 2023 Energy Storage Market Outlook (2023)
- A stochastic optimal control solution to the energy management of a microgrid with storage and renewables (2016) — link
- Today in Energy: U.S . battery storage capacity expected to nearly double in 2024 (2024)
- Creation of synthetic electric grid models for transient stability studies (2017)
- Educational Applications of Large Synthetic Power Grids (2019) — link
- Policy iteration for Hamilton - Jacobi - Bellman equations with control constraints (2020)
- A consumption-investment model with state-dependent lower bound constraint on consumption (2021)
- Solving the Hamilton - Jacobi - Bellman equation for a stochastic system with state constraints (2014) — link
- Viscosity Solutions (2006) — link
- Learning HJB Viscosity Solutions with PINNs for Continuous - Time Reinforcement Learning
- HJB Optimal Feedback Control with Deep Differential Value Functions and Action Constraints (2020)
- Grid Forming Technology: Bulk Power System Reliability Considerations (2021) — link
- Solvability of Differential R iccati Equations and Applications to Algorithmic Trading with Signals (2022)
- Optimal strategies for wind turbine environmental curtailment (2020) — link
- Maximum Hands-Off Control: A Paradigm of Control Effort Minimization (2016) — link
- Stochastic $L_1$ -optimal control via forward and backward sampling (2018) — link
- Learning Deep Stochastic Optimal Control Policies using Forward-Backward SDEs (2021)
- State-Constrained Stochastic Optimal Control Problems via Reachability Approach (2016) — link
- State Constrained Stochastic Optimal Control for Continuous and Hybrid Dynamical Systems Using DFBSDE (2023)
- Hybrid power plants: Status of operating and proposed plants, 2024 edition (2024) — link
- Utilities report batteries are mos commonly used for arbitrage and grid stability (2024)
1 of 94 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.