Wind-Integrated OPF: Real-Time & Risk-Aware Models

• Wind-integrated OPF comprises optimization models that explicitly represent wind generation as a controllable and uncertain resource affecting voltage profiles, line flows, and dispatch decisions.
• Formulations range from deterministic approaches with curtailment to real-time scenario-based and probabilistic risk-aware models, each addressing the challenge of wind variability in power systems.
• Advanced strategies integrate AC/DC, HVDC, and learning-based approximations to balance forecast uncertainty and computational efficiency, ensuring robust and economically efficient operation.

Searching arXiv for relevant papers on wind-integrated OPF, real-time OPF under wind penetration, chance-constrained OPF with wind, CVaR-based robust OPF, and offshore wind/AC-DC OPF. Wind-integrated optimal power flow (OPF) denotes the class of OPF formulations in which wind generation is represented explicitly within the optimization problem, either as a controllable or partially controllable injection, an uncertain renewable source, or a forecast-driven exogenous input whose variability materially affects dispatch, network feasibility, and operating cost. In the literature, the term spans several distinct modeling paradigms: deterministic AC or DC OPF with wind curtailment, real-time prediction-updating OPF, chance-constrained formulations with affine recourse, CVaR-regularized dispatch under wind shortfall risk, forecast-integrated AC/DC OPF for offshore wind systems, and market-coupled OPF for multi-terminal HVDC grids with large offshore wind injections (Mohagheghi et al., 2018, Mohagheghi et al., 2018, Bent et al., 2013, Zhang et al., 2013, Ruan et al., 17 Aug 2025, Du et al., 29 Oct 2025, Valerio et al., 26 May 2025). Across these formulations, the common problem is that wind power is variable, forecast-dependent, and operationally consequential for voltage profiles, line flows, slack-bus support, curtailment, and economic dispatch, so conventional single-snapshot deterministic OPF is often insufficient (Mohagheghi et al., 2018, Chamanbaz et al., 2018).

1. Deterministic formulations and the role of curtailment

A central deterministic formulation appears in the real-time distribution-level work on wind penetration, where OPF is posed as a nonlinear AC optimization with wind curtailment factors as explicit control variables (Mohagheghi et al., 2018, Mohagheghi et al., 2018). In that framework, the generic OPF is written as

$\max f(x,u)$

subject to

$g(x,u)=0,\qquad h(x,u)\le h^{\max},$

together with state and control bounds, where the control vector is the set of wind curtailment factors $\beta_{c.w}$ and the state vector includes the real and imaginary parts of bus voltages and the slack-bus active and reactive injections (Mohagheghi et al., 2018). Wind is therefore modeled as a controllable renewable injection with actual dispatched power $P_w \beta_{c.w}$, with

$0 \le \beta_{c.w} \le 1,$

so $\beta_{c.w}=1$ means no curtailment and smaller values imply increasing curtailment (Mohagheghi et al., 2018, Mohagheghi et al., 2018).

In the implementation paper, the objective is made explicit as a technical-economic trade-off: $\max f = f_1 - f_2 - f_3 - f_4,$ where $f_1$ is revenue from utilized wind power, $f_2$ is the cost of active losses, $f_3$ is the cost of active import at the slack bus, and $g(x,u)=0,\qquad h(x,u)\le h^{\max},$0 is the cost of reactive import at the slack bus (Mohagheghi et al., 2018). The active and reactive power balance constraints are written as

$g(x,u)=0,\qquad h(x,u)\le h^{\max},$1

$g(x,u)=0,\qquad h(x,u)\le h^{\max},$2

with voltage, line apparent-power, slack-bus, and curtailment-factor limits added as inequality constraints (Mohagheghi et al., 2018). This formulation is a nonlinear AC OPF solved as an NLP in GAMS/CONOPT3 on a 41-bus, 27.6 kV rural medium-voltage system with two 10 MW wind stations at buses 2 and 16, both operating at unity power factor (Mohagheghi et al., 2018).

This deterministic curtailment-based formulation clarifies a recurring property of wind-integrated OPF: curtailment often becomes the principal control action when wind units do not provide reactive support and reverse active/reactive power flow is not allowed (Mohagheghi et al., 2018). In such settings, wind is not merely an uncontrollable negative load. It is a schedulable resource whose utilization factor is optimized to preserve feasibility under voltage, feeder, and substation constraints (Mohagheghi et al., 2018, Mohagheghi et al., 2018). A plausible implication is that deterministic wind-integrated OPF is often shaped less by fuel substitution alone than by local network bottlenecks and the availability of fast curtailment actuators.

2. Real-time and moving-horizon wind-integrated OPF

Wind-integrated OPF becomes operationally distinctive when the optimization must be executed faster than wind conditions evolve. The real-time approach developed in the paired 2018 papers addresses precisely the conflict between fast wind fluctuations and slower OPF computation by using a prediction-updating architecture rather than solving one deterministic OPF per control step (Mohagheghi et al., 2018, Mohagheghi et al., 2018).

The method uses a moving prediction horizon

$g(x,u)=0,\qquad h(x,u)\le h^{\max},$3

and an update interval

$g(x,u)=0,\qquad h(x,u)\le h^{\max},$4

with reserved computation time

$g(x,u)=0,\qquad h(x,u)\le h^{\max},$5

for solving a bank of scenario OPFs (Mohagheghi et al., 2018). Around each forecast mean wind power $g(x,u)=0,\qquad h(x,u)\le h^{\max},$6, the framework generates higher and lower wind levels: $g(x,u)=0,\qquad h(x,u)\le h^{\max},$7

$g(x,u)=0,\qquad h(x,u)\le h^{\max},$8

with $g(x,u)=0,\qquad h(x,u)\le h^{\max},$9 and

$\beta_{c.w}$0

For two wind stations, this yields $\beta_{c.w}$1 scenarios (Mohagheghi et al., 2018, Mohagheghi et al., 2018).

All scenario OPFs are solved independently and in parallel, and the resulting optimal curtailment factors and network variables are stored in a lookup table (Mohagheghi et al., 2018). Every 20 seconds, the actual wind $\beta_{c.w}$2 is measured, the corresponding scenario is selected using a lookup-table and selection algorithm, and the associated curtailment is applied (Mohagheghi et al., 2018). In Part I, the scenario selection rule is explicitly conservative: if actual wind lies between precomputed levels, the nearest higher level is chosen to preserve feasibility (Mohagheghi et al., 2018).

The implementation shows that the 49 nonlinear OPFs can be solved within the reserved 112-second window on a multi-processor server with 2 physical processors, Intel Xeon X5690, 3.47 GHz, 6 cores, 12 threads, and 64 GB RAM, with 7 processors used in runtime plots (Mohagheghi et al., 2018). On the first horizon, all 49 scenarios yielded $\beta_{c.w}$3, with load about 6.65 MW and strong curtailment required, while reactive support remained imported at about $\beta_{c.w}$4 because the wind stations operated at unity PF (Mohagheghi et al., 2018). The curtailment factors differed markedly by location; for example, in the high-high scenario $\beta_{c.w}$5, the paper reports

$\beta_{c.w}$6

showing location-specific sensitivity to network constraints (Mohagheghi et al., 2018).

This real-time strand is best interpreted as a scenario-based precomputation architecture with lookup-table dispatch refreshed on a moving horizon, rather than as classical stochastic programming or standard model predictive control, though the paper notes the structure is strongly MPC-like (Mohagheghi et al., 2018, Mohagheghi et al., 2018). It is especially relevant to wind-integrated OPF in distribution systems where detailed AC models are needed but full online reoptimization at sub-minute timescales is impractical.

3. Probabilistic and risk-aware formulations

A second major branch of wind-integrated OPF models wind uncertainty explicitly through probabilistic or risk measures. One influential transmission-level formulation is the synchronization-aware chance-constrained OPF, which starts from a lossless, fixed-voltage active-power model retaining the sine nonlinearity

$\beta_{c.w}$7

rather than using a pure DC approximation (Bent et al., 2013). The deterministic baseline is

$\beta_{c.w}$8

subject to

$\beta_{c.w}$9

$P_w \beta_{c.w}$0

$P_w \beta_{c.w}$1

Wind is decomposed into mean and fluctuation components, $P_w \beta_{c.w}$2, and balancing is handled by an affine recourse policy

$P_w \beta_{c.w}$3

(Bent et al., 2013).

The paper’s distinctive contribution is to add probabilistic protection not only against thermal overloads and generator violations, but also against loss of synchrony. In its tractable formulation, the nonlinear power-flow-feasibility chance constraint is replaced by linewise phase-difference conditions such as

$P_w \beta_{c.w}$4

alongside thermal chance constraints

$P_w \beta_{c.w}$5

and generator upper and lower chance constraints (Bent et al., 2013). Under independent Gaussian wind deviations, these become deterministic convex constraints with safety margins proportional to $P_w \beta_{c.w}$6 and the standard deviation of the angle-difference fluctuation (Bent et al., 2013). The final Sync CC-OPF is a convex conic program solved by a cutting-plane algorithm and tested on IEEE 9-bus, IEEE 118-bus, and Polish systems, with the IEEE 9-bus example requiring 13 iterations and the Polish grid 11 iterations (Bent et al., 2013).

A different risk-aware paradigm appears in the CVaR-based robust DC OPF for wind integration (Zhang et al., 2013). Here the uncertain quantity is actual wind realization $P_w \beta_{c.w}$7, while the decision variable $P_w \beta_{c.w}$8 is day-ahead committed wind injection. The network is modeled by the DC nodal balance

$P_w \beta_{c.w}$9

with line limits

$0 \le \beta_{c.w} \le 1,$0

(Zhang et al., 2013). Wind shortfall creates transaction cost

$0 \le \beta_{c.w} \le 1,$1

and the core risk term is

$0 \le \beta_{c.w} \le 1,$2

The resulting OPF is

$0 \le \beta_{c.w} \le 1,$3

subject to DC power flow and generator constraints (Zhang et al., 2013).

This CVaR-regularized formulation is not worst-case robust and not chance-constrained; it is a convex risk-aware stochastic/robustified DC-OPF that penalizes the tail of wind shortfall transaction cost (Zhang et al., 2013). The expectation is approximated by a distribution-free sample average approximation,

$0 \le \beta_{c.w} \le 1,$4

and auxiliary variables yield a finite-dimensional LP/QP solved in CVX with SeDuMi (Zhang et al., 2013). On the IEEE 30-bus system with seven wind farms at buses 1, 3, 7, 15, 19, 24, and 26, real data from the Kaggle 2012 wind forecasting competition, $0 \le \beta_{c.w} \le 1,$5, and $0 \le \beta_{c.w} \le 1,$6, the CVaR-based dispatch achieved a mean total cost of $0 \le \beta_{c.w} \le 1,$7 versus $0 \le \beta_{c.w} \le 1,$8 for the no-risk scheme, and a variance of $0 \le \beta_{c.w} \le 1,$9 versus $\beta_{c.w}=1$0 (Zhang et al., 2013).

A more recent deterministic-but-risk-priced direction internalizes wind uncertainty by adding an expected wind deviation cost to AC OPF using a Gaussian mixture model fitted to historical wind data (Ruan et al., 17 Aug 2025). The total wind cost is

$\beta_{c.w}=1$1

with

$\beta_{c.w}=1$2

$\beta_{c.w}=1$3

where shortage and surplus expectations are evaluated analytically using the integral identity of the Gaussian mixture model (Ruan et al., 17 Aug 2025). This cost is then embedded in the objective

$\beta_{c.w}=1$4

with piecewise linear epigraph constraints

$\beta_{c.w}=1$5

The method is solved through an enhanced SOCR with second-order Taylor expansion and rolling cutting planes, and on a modified PEGASE 1354-bus case with a 225 MW wind farm it reported scheduled wind output of 40.5 MW, total generation cost $\beta_{c.w}=1$6, fossil fuel cost share 97.04%, wind generation cost share 2.96%, maximum voltage error $\beta_{c.w}=1$7, maximum angle error $\beta_{c.w}=1$8, and solution time 10.94 s (Ruan et al., 17 Aug 2025).

4. AC/DC, offshore wind, and HVDC-coupled formulations

Wind-integrated OPF is not limited to conventional AC transmission or distribution networks. Offshore wind and hybrid AC/DC systems have motivated formulations in which wind forecasts, converter setpoints, and DC states are co-optimized.

One such formulation is the forecast-integrated AC/DC OPF for hybrid AC–MTDC systems under uncertainty (Du et al., 29 Oct 2025). The network consists of two IEEE 9-bus AC systems, two offshore wind farms, and a 4-terminal VSC-MTDC grid (Du et al., 29 Oct 2025). The OPF is described as a nonlinear AC/DC OPF with objective

$\beta_{c.w}=1$9

subject to full AC power flow equations

$\max f = f_1 - f_2 - f_3 - f_4,$0

$\max f = f_1 - f_2 - f_3 - f_4,$1

AC balances

$\max f = f_1 - f_2 - f_3 - f_4,$2

DC equations

$\max f = f_1 - f_2 - f_3 - f_4,$3

and converter loss/current relations

$\max f = f_1 - f_2 - f_3 - f_4,$4

(Du et al., 29 Oct 2025).

Wind enters not as scenarios or chance constraints, but as forecasted wind farm generation $\max f = f_1 - f_2 - f_3 - f_4,$5 derived from a Random Forest wind speed model with $\max f = f_1 - f_2 - f_3 - f_4,$6 trees, history length $\max f = f_1 - f_2 - f_3 - f_4,$7, and prediction horizon $\max f = f_1 - f_2 - f_3 - f_4,$8 (Du et al., 29 Oct 2025). The OPF computes baseline setpoints, while real-time correction is handled by adaptive droop: $\max f = f_1 - f_2 - f_3 - f_4,$9 with

$f_1$0

This is therefore a forecast-driven deterministic OPF coupled to real-time adaptive control rather than a stochastic or robust OPF (Du et al., 29 Oct 2025). In hardware-in-the-loop validation using RTDS, RSCAD, Julia/PowerModelsACDC.jl, IPOPT, and GTNETx2, the proposed method gave the smallest active power deviations from OPF references under wind forecast uncertainty and better frequency/DC voltage performance than active power control, DC voltage control, and adaptive voltage droop baselines (Du et al., 29 Oct 2025).

A more market-oriented offshore-wind formulation appears in the North Sea interconnector study (Valerio et al., 26 May 2025). There, the grid is a hybrid AC/DC network with offshore wind hubs and MTDC links connecting price zones. The OPF objective is built from zonal generation costs

$f_1$1

with zonal price

$f_1$2

Renewable injections are controlled by a utilization factor $f_1$3, so offshore wind enters as $f_1$4, and curtailment is $f_1$5 (Valerio et al., 26 May 2025). Price-zone exchange bounds

$f_1$6

are added because large offshore wind hubs are modeled as price-making rather than price-taking (Valerio et al., 26 May 2025).

The paper’s case study considers Belgium, Denmark, Germany, Great Britain, Netherlands, and Norway, with offshore developments such as Princess Elizabeth Island at 3.5 GW, NL Energy Hub at 2 GW, German Offshore Interconnection Cluster at 4 GW, Danish Energy Island at 3.5 GW, and Norwegian Energy Hub at 2 GW (Valerio et al., 26 May 2025). It reports that in endogenous-price Scenario II, curtailment rises significantly because import capacities of price zones can bind before physical HVDC line capacities, showing that market absorption limits as well as cable limits can drive offshore wind curtailment (Valerio et al., 26 May 2025). This suggests that, for very large offshore wind systems, wind-integrated OPF may need to model not only physical feasibility but also price-zone exchange behavior.

5. Computational architectures and convexification strategies

Because wind-integrated OPF is often repeatedly solved and frequently nonconvex, computational methodology has become a major research topic. Several structural and decomposition approaches are relevant.

In radial and distribution networks, distributed solution methods for convexified OPF are motivated explicitly by renewable variability and the need for real-time or near-real-time computation (Lam et al., 2011, Peng et al., 2014, Peng et al., 2015). The balanced radial distribution algorithm based on SOCP relaxation and ADMM solves branch-flow OPF with closed-form $f_1$7- and $f_1$8-updates, and on a 2,065-bus feeder it converged in 1,114 iterations with total runtime 1,153 s on one machine and estimated distributed convergence time about 0.56 s, with local subproblem speedups of about $f_1$9 relative to generic iterative solvers (Peng et al., 2014). The unbalanced multiphase distribution algorithm based on ADMM and SDP relaxation reduces local updates to either closed form or eigendecomposition of a $f_2$0 Hermitian matrix, with estimated distributed convergence around 2.5 s across IEEE 13-, 34-, 37-, and 123-bus feeders (Peng et al., 2015). Although these papers do not model wind uncertainty explicitly, they are directly relevant algorithmically to wind-integrated OPF in radial feeders because wind injections can be represented as controllable or bounded nodal injections within the same OPF machinery (Peng et al., 2014, Peng et al., 2015).

For meshed systems, an earlier distributed algorithm for convexified OPF exploits cases with zero duality gap and decomposes the SDP relaxation over maximal cliques of a chordal graph (Lam et al., 2011). The paper is motivated by the observation that renewable energy calls for very efficient OPF solutions usable in real time (Lam et al., 2011). It reports, for example, on radial distribution feeders that distributed dual solution times can remain nearly size-independent relative to centralized solves, with normalized times on 8-, 34-, and 123-bus radial systems of 1.00, 1.70, and 1.64 for distributed dual versus 1.85, 298.68, and out-of-memory for centralized solving (Lam et al., 2011).

A distinct structural approach is the hybrid AC/HVDC transmission architecture enabling exact semidefinite relaxation of AC OPF (Hotz et al., 2016). The idea is to keep a spanning tree of the transmission grid as AC and convert every other branch to HVDC, making the AC subgrid radial while preserving a physically meshed hybrid network (Hotz et al., 2016). The OPF is then solved through an SDP relaxation that becomes exact under the paper’s conditions, enabling globally optimal solution in polynomial time (Hotz et al., 2016). Although this paper is not a wind-uncertainty paper, it is structurally relevant because high renewable generation increases the need for controllable transfer capacity, and the case study shows generation cost reduction of 14.13% relative to DC OPF + AC PF and 6.10% for AC OTS on an adapted PJM 5-bus system, with continued benefit even at 7.0% HVDC loss factor (Hotz et al., 2016). A plausible implication is that network architecture itself can be an enabling variable for tractable wind-integrated OPF.

The enhanced SOCR with rolling cutting planes (Ruan et al., 17 Aug 2025) belongs to a newer family of scalable AC approximations that aim to preserve much of AC fidelity while avoiding full nonlinear nonconvex solution. Its key innovation is not only the second-order Taylor expansion and lifted variables $f_2$1, $f_2$2, but the iterative rolling cuts around the current operating point to reduce fictitious losses and tighten the conic relaxation (Ruan et al., 17 Aug 2025). Across systems up to PEGASE 2869, the method reported objective error within 0.3%, typically 2–3 iterations, and solution times such as 10.43 s on PEGASE 1354 and 13.11 s on PEGASE 2869 (Ruan et al., 17 Aug 2025). This positions it between full NLP and coarse linearization as an efficient large-scale AC engine for wind-integrated OPF.

6. Learning-based approximations and hybrid solver architectures

A more recent line of work treats wind-integrated OPF not as a stochastic program to be solved from scratch, but as a repeated mapping from load and weather conditions to dispatch, learned offline and evaluated rapidly online.

The attention-based method for highly renewable power systems is an explicit example (Li et al., 2023). The physical optimization behind the labels is a repeated DC-OPF on clustered European networks generated by PyPSA-Eur and solved with Gurobi, while wind and solar enter through weather-derived renewable capacity coefficients $f_2$3 from ERA5 (Li et al., 2023). Renewable limits are imposed as

$f_2$4

(Li et al., 2023). The neural model learns a mapping from demand and weather to line flows and nodal generation, with projection post-processing

$f_2$5

subject to nodal balance, total nodal generation bounds, and line-flow bounds, so that the final output is DC-feasible (Li et al., 2023).

In a 33-node and a 300-node clustered European system with wind, solar, OCGT, and coal, the proposed method outperformed several data-driven baselines in MAAPE and reduced average absolute nodal imbalance from 0.185 GW before projection to 0.454 kW after projection (Li et al., 2023). Runtime per 100 data points was 2.40 s versus 2.60 s for the interior-point baseline on 33 nodes, and 4.90 s versus 22.22 s on 300 nodes (Li et al., 2023). The paper’s interpretability analysis identifies nighttime strong-wind conditions in which Northern European nodes become especially influential, indicating that the model captures geography-specific wind-driven congestion patterns (Li et al., 2023).

More broadly, this literature suggests two distinct uses of learning in wind-integrated OPF. The first is surrogate replacement of repeated deterministic OPF solves conditioned on weather, as in the attention model (Li et al., 2023). The second, more general but not wind-specific in the supplied material, is label-free neural OPF based on energy gradient flow, which the paper itself notes could be extended to wind by treating wind injections as negative load or as explicit input variables, though it does not implement wind modeling (Liu, 1 Dec 2025). The former is directly wind-integrated; the latter is methodologically relevant but not yet a wind-integrated OPF formulation in the strict sense (Liu, 1 Dec 2025).

7. Conceptual divisions, limitations, and common misconceptions

Wind-integrated OPF is not a single model class. The literature in the supplied material supports at least five distinct interpretations.

First, there is deterministic AC or DC OPF with explicit wind variables, often curtailment factors or scheduled wind injections, where wind is optimized but uncertainty is not represented probabilistically (Mohagheghi et al., 2018, Mohagheghi et al., 2018, Ruan et al., 17 Aug 2025). Second, there are forecast-updating real-time architectures that discretize plausible wind deviations and precompute scenario OPFs over a short horizon (Mohagheghi et al., 2018, Mohagheghi et al., 2018). Third, there are risk-aware stochastic formulations such as chance-constrained OPF with affine recourse (Bent et al., 2013) and CVaR-regularized DC-OPF (Zhang et al., 2013). Fourth, there are deterministic forecast-driven AC/DC OPF layers embedded in larger control architectures for offshore wind systems (Du et al., 29 Oct 2025). Fifth, there are market-coupled formulations where large offshore wind hubs interact with price zones and MTDC exchanges (Valerio et al., 26 May 2025).

A common misconception is that “wind-integrated OPF” necessarily means scenario-based stochastic optimization. The CVaR paper, the GMM-cost paper, and the AC–MTDC forecast-control paper contradict this by showing formulations in which wind uncertainty is internalized through tail-risk cost, expected deviation cost, or forecast-conditioned setpoints rather than explicit scenario trees (Zhang et al., 2013, Ruan et al., 17 Aug 2025, Du et al., 29 Oct 2025). Another misconception is that wind can be represented adequately as a negative load in all settings. While this is sometimes possible in simplified deterministic formulations, the real-time distribution papers show that wind-specific curtailment variables, slack-bus restrictions, and no-reverse-flow assumptions fundamentally alter the optimization structure (Mohagheghi et al., 2018, Mohagheghi et al., 2018).

The literature also shows important limitations. Many methods are single-period and omit ramping, unit commitment, and storage (Zhang et al., 2013, Valerio et al., 26 May 2025, Li et al., 2023). Several papers rely on simplified network models: DC approximation in the CVaR and attention-based work (Zhang et al., 2013, Li et al., 2023), lossless fixed-voltage active-power models in the synchronization-aware chance-constrained work (Bent et al., 2013), or deterministic forecast trajectories without probabilistic reserve models in the AC–MTDC framework (Du et al., 29 Oct 2025). Conversely, the more AC-faithful methods often face computational burdens or require approximations such as SOCR and iterative cutting planes (Ruan et al., 17 Aug 2025).

A further controversy, implicit across the literature, concerns conservatism versus computational tractability. The prediction-updating real-time approach is conservative because it selects the nearest higher wind scenario (Mohagheghi et al., 2018, Mohagheghi et al., 2018). Chance constraints reduce conservatism relative to worst-case robust methods but require distributional assumptions or scenario bounds (Bent et al., 2013, Chamanbaz et al., 2018). CVaR captures tail risk without full worst-case pessimism but abstracts operational insecurity into transaction cost (Zhang et al., 2013). The GMM-cost method incorporates expected mismatch cost but does not give explicit probabilistic feasibility guarantees (Ruan et al., 17 Aug 2025). These are not contradictions so much as different answers to the same question: how to price, hedge, or enforce security against wind uncertainty while keeping OPF solvable.

8. Research directions indicated by the literature

Several directions emerge directly from the supplied papers. One is the continued convergence of wind-integrated OPF with moving-horizon and real-time control architectures, especially when forecast error evolves faster than a full AC OPF can be solved (Mohagheghi et al., 2018, Mohagheghi et al., 2018, Du et al., 29 Oct 2025). Another is richer treatment of offshore wind systems in hybrid AC/DC grids, where converter control, DC voltages, market coupling, and zonal exchanges become inseparable from the OPF layer (Du et al., 29 Oct 2025, Valerio et al., 26 May 2025).

A third direction is more expressive risk modeling beyond Gaussian assumptions or simple scenario envelopes. The GMM-based wind cost paper explicitly uses historical wind data and Gaussian mixtures to avoid crude unimodal distributions (Ruan et al., 17 Aug 2025), while the scenario-with-certificates chapter emphasizes that wind power is not Gaussian and motivates sample-based probabilistic guarantees (Chamanbaz et al., 2018). This suggests a broader trend toward data-driven uncertainty representations paired with convex or convexified OPF back ends.

A fourth direction is computational. Parallel scenario solution (Mohagheghi et al., 2018), chordal and ADMM decompositions (Lam et al., 2011, Peng et al., 2014, Peng et al., 2015), structural HVDC-enabled convexification (Hotz et al., 2016), enhanced SOCR with rolling cuts (Ruan et al., 17 Aug 2025), and weather-conditioned learning surrogates (Li et al., 2023) all respond to the same practical fact: wind-integrated OPF is often not a one-off optimization but a repeatedly solved operational problem. This suggests that future work will continue to blur the boundaries between optimization, control, forecasting, and learned approximations.

In sum, wind-integrated OPF is best understood not as a single canonical formulation but as a family of network-constrained optimization models in which wind generation is an economically valuable, physically constrained, and uncertainty-bearing resource. The specific formulation—deterministic, real-time scenario-based, chance-constrained, CVaR-regularized, forecast-coupled AC/DC, or market-integrated—depends on which aspect of wind integration is being prioritized: feasibility under rapid fluctuations, tail-risk management, synchronization security, converter coordination, or market absorption of large offshore injections (Mohagheghi et al., 2018, Bent et al., 2013, Zhang et al., 2013, Du et al., 29 Oct 2025, Valerio et al., 26 May 2025, Ruan et al., 17 Aug 2025).