betaFeedback

WorksMathematics

Second-Order Cone Valid Inequalities for Set-Point Tracking with Energy Storage

Abstract

Convex relaxations of energy storage models drop the complementarity constraints that forbid simultaneous charging and discharging. In the set-point tracking problem, the relaxation can then charge and discharge simultaneously at no cost, as the tracking error depends only on the net power of the storage. We show that the convex hull of the single-period epigraph of the tracking error, under complementarity, is the epigraph of a parabolic cylinder. The resulting second-order cone inequality adds to the tracking error a penalty on the product of the charging and discharging powers and is the tightest convex member of the pencil of quadrics through the two branches of the epigraph. With power limits and the mode variable, the hull is given by the perspective reformulation of the two modes, which dominates the cylinder. In 20000 instances, the cylinder reduces the hours with simultaneous charging and discharging by more than 99% and yields an optimal schedule in 98.8% of them.

Introduction

Storage models need complementarity constraints, written with bilinear equalities or binary variables, to prevent charging and discharging in the same period; their convex relaxations drop them and may return infeasible schedules [1]. Valid inequalities that strengthen the relaxation reduce this effect: Pozo [2] describes the convex hull of the single-period storage set, and Elgersma et al. [3] extend it to reserve provision. These inequalities act on the feasible set and leave the objective function untouched. Al Taha and Bitar [4] instead eliminate the power variables through a piecewise-affine bijection with the state-of-charge profiles, whose feasible set is convex; the nonconvexity remains in the mapping, so the power variables may appear only in the objective function, which stays convex if it is nondecreasing in the charging power. In the set-point tracking problem, the relaxation is weak precisely because of its objective: the tracking error depends only on the net power, so charging and discharging simultaneously is free [2]; it also violates the condition of [4] whenever the set-point asks the storage to charge.

This letter shows that the convex hull of the single-period epigraph of the tracking error under complementarity is the epigraph of a parabolic cylinder, which yields a second-order cone (SOC) inequality that penalizes simultaneous charging and discharging Section III. We identify this cylinder as the tightest convex member of the pencil of quadrics through the epigraph, extend the result to power limits through the perspective reformulation of the operating modes, and show that the cut nearly eliminates simultaneous charging and discharging in 20 000 instances (§ 3).

Set-Point Tracking Problem

Let ptdp_t^d, ptcp_t^c, and sts_t denote the discharging power, the charging power, and the state of charge of the storage in period t∈{1,…,T}t \in\{1,\ldots,T\}. Its operational set is

P={(pd,pc,s)∈R≥03T:pd≤P‾d, pc≤P‾c,\mathcal{P}=\{(p^d,p^c,s)\in\mathbb{R}_{\ge0}^{3T}:p^d\le\overline{P}^d,\ p^c\le\overline{P}^c,
1S‾≤s≤1S‾,1\underline{S}\le s\le1\overline{S},
st=st−1+Δ(ηcptc−ptd/ηd), ∀t,s_t=s_{t-1}+\Delta\left(\eta_cp_t^c-p_t^d/\eta_d\right),\ \forall t,
ptdptc=0, ∀t},(1)p_t^dp_t^c=0,\ \forall t\}, \tag*{(1)}

where P‾d\overline{P}^d and P‾c\overline{P}^c are the power limits, [S‾,S‾][\underline{S},\overline{S}] is the state-of-charge range, ηd,ηc∈(0,1]\eta_d,\eta_c\in(0,1] are the efficiencies, Δ\Delta is the period length, and s0s_0 is given. The complementarity constraints (1d) can be written with a binary variable utu_t per period and the power limits ptd≤P‾d(1−ut)p_t^d\le\overline{P}^d(1-u_t) and ptc≤P‾cutp_t^c\le\overline{P}^c u_t [2].

The set-point tracking problem (STP) seeks the operation of the storage that best follows a power signal ptsp_t^s [2]:

min⁡(pd,pc,s)∈P ∑t=1T(ptd−ptc−pts)2.(2)\min_{(p^d,p^c,s)\in\mathcal{P}}\ \sum_{t=1}^{T}(p_t^d-p_t^c-p_t^s)^2. \tag*{(2)}

Its natural relaxation drops (1d) (or relaxes utu_t to [0,1][0,1]). As the tracking error depends only on the net power, the relaxation can increase both ptdp_t^d and ptcp_t^c at no cost, and often does so to meet the state-of-charge limits [2]. To strengthen the objective, we write (2) in epigraph form, with one variable ztz_t per period:

min⁡(pd,pc,s)∈P, z{∑t=1Tzt:(zt,ptd,ptc)∈Et ∀t}.(3)\min_{(p^d,p^c,s)\in\mathcal{P},\ z}\left\{\sum_{t=1}^{T}z_t:(z_t,p_t^d,p_t^c)\in\mathcal{E}_t\ \forall t\right\}. \tag*{(3)}

where

Et:={(z,pd,pc)∈R×R≥02:z≥(pd−pc−pts)2, pdpc=0}.(4)\mathcal{E}_t:=\left\{(z,p^d,p^c)\in\mathbb{R}\times\mathbb{R}_{\ge0}^{2}:z\ge(p^d-p^c-p_t^s)^2,\ p^dp^c=0\right\}. \tag*{(4)}

As (3) is linear in ztz_t, the tightest convex relaxation that can be built from each set Et\mathcal{E}_t alone is its convex hull.

The Convex Hull of the Epigraph

We omit the subscript tt and write psp^s for ptsp_t^s. The set E\mathcal{E} is the union of two convex sets, the epigraphs of the pure-discharge parabola z=(pd−ps)2z=(p^d-p^s)^2 on the plane pc=0p^c=0 and of the pure-charge parabola z=(pc+ps)2z=(p^c+p^s)^2 on the plane pd=0p^d=0; these

J. M. Morales is with the research group OASYS, University of Málaga, Málaga 29071, Spain. E-mail: [email protected]. This work was supported by the Consejería de Universidad, Investigación e Innovación de la Junta de Andalucía through project DGP-PIDI-2024-00851. The author thankfully acknowledges the computer resources, technical expertise, and assistance provided by the SCBI (Supercomputing and Bioinformatics) center of the University of Málaga. parabolas are the two branches of E\mathcal{E} (thick curves in Figure 1a). Its convex hull is the epigraph of the parabolic cylinder

Tracking error for $p^s=1$

Figure 1. Tracking error for ps=1p^s=1. (a) Branches of E\mathcal{E} (thick), rulings of the cylinder (5) (solid), and natural relaxation along the segments pd+pc=rp^d+p^c=r (dashed). (b) Members λ=2\lambda=2 (cylinder), 00 (paraboloid), and −2-2 (natural relaxation) of the pencil (7) along pd+pc=1.5p^d+p^c=1.5, with θ=pd/(pd+pc)\theta=p^d/(p^d+p^c); the SOC cut removes the shaded area; dots: points of E\mathcal{E}. (c) Top view of the hull with power limits P‾d\overline{P}^d and P‾c\overline{P}^c: rulings of the cylinder and, where pd+pc>P‾dp^d+p^c>\overline{P}^d, of the cone (11) with apex at the full-discharge point (dot).

gC(pd,pc):=(pd+pc)2−2ps(pd−pc)+(ps)2.(5)g_C(p^d,p^c):=(p^d+p^c)^2-2p^s(p^d-p^c)+(p^s)^2. \tag*{(5)}

Theorem 1. conv⁡(E)=K:={(z,pd,pc)∈R×R≥02:z≥gC(pd,pc)}\operatorname{conv}(\mathcal{E})=\mathcal{K}:=\{(z,p^d,p^c)\in\mathbb{R}\times\mathbb{R}_{\geq0}^2:z\geq g_C(p^d,p^c)\}.

Proof. The set K\mathcal{K} is convex, because gCg_C is a convex quadratic function. On E\mathcal{E}, pdpc=0p^dp^c=0, so (pd+pc)2=(pd−pc)2(p^d+p^c)^2=(p^d-p^c)^2 and gC(pd,pc)=(pd−pc−ps)2≤zg_C(p^d,p^c)=(p^d-p^c-p^s)^2\leq z. Hence E⊆K\mathcal{E}\subseteq\mathcal{K} and conv⁡(E)⊆K\operatorname{conv}(\mathcal{E})\subseteq\mathcal{K}. Conversely, let (z,pd,pc)∈K(z,p^d,p^c)\in\mathcal{K}, with throughput r:=pd+pcr:=p^d+p^c and slack δ:=z−gC(pd,pc)≥0\delta:=z-g_C(p^d,p^c)\geq0. If r=0r=0, the point belongs to E\mathcal{E}. Otherwise, let θ:=pd/r∈[0,1]\theta:=p^d/r\in[0,1] and take the points that discharge or charge at power rr, with their tracking errors raised by δ\delta,

A:=((r−ps)2+δ,r,0),B:=((r+ps)2+δ,0,r).A:=((r-p^s)^2+\delta,r,0),\qquad B:=((r+p^s)^2+\delta,0,r).

Both belong to E\mathcal{E}, as δ≥0\delta\geq0, and θA+(1−θ)B=(ζ+δ,pd,pc)\theta A+(1-\theta)B=(\zeta+\delta,p^d,p^c) with

ζ=θ(r−ps)2+(1−θ)(r+ps)2=r2−2psr(2θ−1)+(ps)2=gC(pd,pc),\zeta=\theta(r-p^s)^2+(1-\theta)(r+p^s)^2 =r^2-2p^sr(2\theta-1)+(p^s)^2=g_C(p^d,p^c),

because r=pd+pcr=p^d+p^c and r(2θ−1)=pd−pcr(2\theta-1)=p^d-p^c. As ζ+δ=z\zeta+\delta=z, this convex combination of two points of E\mathcal{E} is (z,pd,pc)(z,p^d,p^c). □

Thus, along each segment pd+pc=rp^d+p^c=r, the hull is the chord between the pure-discharge and pure-charge points with throughput $r (Figure 1a). The SOC cut zt≥gC(ptd,ptc)z_t\geq g_C(p_t^d,p_t^c) can be written as

zt≥(ptd−ptc−pts)2+4ptdptc,(6)z_t\geq(p_t^d-p_t^c-p_t^s)^2+4p_t^dp_t^c, \tag*{(6)}

that is, the tracking error plus a penalty on simultaneous charging and discharging, which vanishes on Et\mathcal{E}_t. It is a convex quadratic, hence SOC-representable, constraint that does not involve the binary variables, so it can be added to any relaxation of P\mathcal{P}. No epigraph variables are needed: replacing the objective of (2) with ∑t[(ptd−ptc−pts)2+4ptdptc]\sum_t[(p_t^d-p_t^c-p_t^s)^2+4p_t^dp_t^c] yields the same relaxation as a convex quadratic program.

Is the cylinder unique?

The cylinder belongs to the pencil of quadrics that contain the two branches of E\mathcal{E}, which, following [5 Thm. 4.1], we write as z=gλ(pd,pc)z=g_\lambda(p^d,p^c) with

gλ(pd,pc):=(pd−pc−ps)2+(2+λ)pdpc,λ∈R.(7)g_\lambda(p^d,p^c):=(p^d-p^c-p^s)^2+(2+\lambda)p^dp^c,\qquad\lambda\in\mathbb{R}. \tag*{(7)}

All members coincide on the planes pd=0p^d=0 and pc=0p^c=0, and the Hessian of gλg_\lambda, with diagonal entries 2 and off-diagonal entries λ\lambda, is positive semidefinite if and only if ∣λ∣≤2|\lambda|\leq2 and singular if and only if λ=±2\lambda=\pm2, the two cylinders of the pencil. For λ=2\lambda=2, g2=gCg_2=g_C is the cylinder of Theorem 1, with rulings parallel to (−4ps,1,−1)(-4p^s,1,-1) in the space (z,pd,pc)(z,p^d,p^c); for λ=−2\lambda=-2, g−2g_{-2} is the tracking error itself, i.e., the objective of the natural relaxation, with rulings parallel to (0,1,1)(0,1,1). Since pdpc≥0p^dp^c\geq0, gλg_\lambda is non-decreasing in λ\lambda, so the cylinder λ=2\lambda=2 is the tightest convex member and the cylinder λ=−2\lambda=-2 the weakest (Figure 1b). Hence the cylinder is not unique, but it is the only one that yields the convex hull.

Power Limits: Perspective Reformulation

Theorem 1 ignores the power limits. With them, the cylinder remains valid but need not give the hull. In formulations that keep the mode variable uu, the single-period set is

M:={(z,pd,pc,u)∈R×R≥02×{0,1}:pd≤P‾d(1−u), pc≤P‾cu, z≥(pd−pc−ps)2},(8)\mathcal{M}:=\{(z,p^d,p^c,u)\in\mathbb{R}\times\mathbb{R}_{\geq0}^2\times\{0,1\}:p^d\leq\overline{P}^d(1-u),\ p^c\leq\overline{P}^cu,\ z\geq(p^d-p^c-p^s)^2\}, \tag*{(8)}

whose projection onto (z,pd,pc)(z,p^d,p^c) is E\mathcal{E} with the power limits. In each mode, the tracking error is a convex function of one power, (pd−ps)2(p^d-p^s)^2 or (pc+ps)2(p^c+p^s)^2, and the hull of such a disjunction follows from their perspectives [6]. Let

h(pd,pc,u):=(pd)21−u+(pc)2u−2ps(pd−pc)+(ps)2,(9)h(p^d,p^c,u):=\frac{(p^d)^2}{1-u}+\frac{(p^c)^2}{u}-2p^s(p^d-p^c)+(p^s)^2, \tag*{(9)}

with the convention 02/0:=00^2/0:=0.

Proposition 1. conv⁡(M)=F:={(z,pd,pc,u):0≤pd≤P‾d(1−u), 0≤pc≤P‾cu, z≥h(pd,pc,u)}\operatorname{conv}(\mathcal{M})=\mathcal{F}:=\{(z,p^d,p^c,u):0\leq p^d\leq\overline{P}^d(1-u),\ 0\leq p^c\leq\overline{P}^cu,\ z\geq h(p^d,p^c,u)\}. With two auxiliary variables wd,wc≥0w^d,w^c\geq0, the inequality z≥hz\geq h is equivalent to the SOC constraints

z≥wd+wc−2ps(pd−pc)+(ps)2,(10)z\geq w^d+w^c-2p^s(p^d-p^c)+(p^s)^2, \tag*{(10)}
∥(2pd,wd−1+u)∥≤wd+1−u,(11)\left\|(2p^d,w^d-1+u)\right\|\leq w^d+1-u, \tag*{(11)}
∥(2pc,wc−u)∥≤wc+u.(12)\left\|(2p^c,w^c-u)\right\|\leq w^c+u. \tag*{(12)}

where (10b) and (10c) are the rotated cones (pd)2≤wd(1−u)(p^d)^2\leq w^d(1-u) and (pc)2≤wcu(p^c)^2\leq w^cu, as ∥(2x,w−y)∥≤w+y⇔x2≤wy\|(2x,w-y)\|\leq w+y\Leftrightarrow x^2\leq wy, w,y≥0w,y\geq0.

Proof. The set F\mathcal{F} is convex, as the projection of a set defined by linear and SOC constraints, and contains M\mathcal{M}: if u=0u=0, then pc=0p^c=0 and h=(pd−ps)2≤zh=(p^d-p^s)^2\leq z; if u=1u=1, then pd=0p^d=0 and h=(pc+ps)2≤zh=(p^c+p^s)^2\leq z. Hence conv⁡(M)⊆F\operatorname{conv}(\mathcal{M})\subseteq\mathcal{F}. Conversely, let (z,pd,pc,u)∈F(z,p^d,p^c,u)\in\mathcal{F}, with slack δ:=z−h(pd,pc,u)≥0\delta:=z-h(p^d,p^c,u)\geq0. Its bounds give u∈[0,1]u\in[0,1]; if u∈{0,1}u\in\{0,1\}, it is in M\mathcal{M} by the cases above. Otherwise, a:=pd/(1−u)∈[0,P‾d]a:=p^d/(1-u)\in[0,\overline{P}^d] and b:=pc/u∈[0,P‾c]b:=p^c/u\in[0,\overline{P}^c], so A:=((a−ps)2+δ,a,0,0)A:=((a-p^s)^2+\delta,a,0,0) and B:=((b+ps)2+δ,0,b,1)B:=((b+p^s)^2+\delta,0,b,1) belong to M\mathcal{M}, and (1−u)A+uB=(ζ+δ,pd,pc,u)(1-u)A+uB=(\zeta+\delta,p^d,p^c,u) with ζ=(1−u)(a−ps)2+u(b+ps)2=\zeta=(1-u)(a-p^s)^2+u(b+p^s)^2= (1−u)a2+ub2−2ps((1−u)a−ub)+(ps)2=h(pd,pc,u)(1-u)a^2+ub^2-2p^s((1-u)a-ub)+(p^s)^2=h(p^d,p^c,u), because (1−u)a=pd(1-u)a=p^d and ub=pcub=p^c. As ζ+δ=z\zeta+\delta=z, this convex combination of two points of M\mathcal{M} is (z,pd,pc,u)(z,p^d,p^c,u).

Physically, a fractional uu splits the period: the storage discharges at power a=pd/(1−u)a=p^d/(1-u) during a fraction 1−u1-u and charges at power b=pc/ub=p^c/u during the rest, and hh is the time average of the two tracking errors, whereas the natural relaxation takes the error of the average net power.

The perspective inequality dominates the cylinder: by the Cauchy–Schwarz inequality, ((pd)2/(1−u)+(pc)2/u)((1−u)+u)≥(pd+pc)2\left((p^d)^2/(1-u)+(p^c)^2/u\right)((1-u)+u)\ge(p^d+p^c)^2, so h≥gCh\ge g_C, with equality iff a=ba=b, i.e., u=u∗≔pc/(pd+pc)u=u^*\coloneqq p^c/(p^d+p^c). The cylinder is thus the perspective inequality for the best time split, where both modes run at power r=pd+pcr=p^d+p^c, as AA and BB in the proof of Theorem 1.

The limits a≤P‾da\le\overline{P}^d and b≤P‾cb\le\overline{P}^c confine uu to the interval [pc/P‾c,1−pd/P‾d][p^c/\overline{P}^c,1-p^d/\overline{P}^d], which requires pd/P‾d+pc/P‾c≤1p^d/\overline{P}^d+p^c/\overline{P}^c\le1. As projection and convex hull commute, minimizing hh over this interval gives the hull of E\mathcal{E} with power limits (Figure 1c). Where r≤min⁡{P‾d,P‾c}r\le\min\{\overline{P}^d,\overline{P}^c\}, the split u∗u^* is feasible and the hull is the cylinder. Where, say, r>P‾dr>\overline{P}^d, it would discharge above P‾d\overline{P}^d, and, hh being convex in uu, the minimum is attained at the closest feasible split, 1−u=pd/P‾d1-u=p^d/\overline{P}^d, which discharges at full power, a=P‾da=\overline{P}^d. Substituting it in (9),

z≥P‾dpd+P‾d(pc)2P‾d−pd−2ps(pd−pc)+(ps)2.(13)z\ge\overline{P}^d p^d+\frac{\overline{P}^d(p^c)^2}{\overline{P}^d-p^d}-2p^s(p^d-p^c)+(p^s)^2. \tag*{(13)}

These points lie on segments that join the full-discharge point ((P‾d−ps)2,P‾d,0)((\overline{P}^d-p^s)^2,\overline{P}^d,0) with the pure-charge points, so (11) is a cone with apex at the former: with ℓ\ell denoting its affine terms, it reads (z−ℓ)(P‾d−pd)≥P‾d(pc)2(z-\ell)(\overline{P}^d-p^d)\ge\overline{P}^d(p^c)^2.

If utu_t appears only in the power limits, as in the natural relaxation, the optimizer chooses it to minimize hh, and (10) yields the hull of Figure 1c: it tightens the cylinder only where pd+pc>min⁡{P‾d,P‾c}p^d+p^c>\min\{\overline{P}^d,\overline{P}^c\}, and nowhere if P‾d=P‾c\overline{P}^d=\overline{P}^c, since then pd/P‾d+pc/P‾c≤1p^d/\overline{P}^d+p^c/\overline{P}^c\le1 implies pd+pc≤P‾dp^d+p^c\le\overline{P}^d. They pay off when utu_t is coupled with other constraints (e.g., mode-dependent costs or reserves), at the price of two rotated cones per period; the cylinder needs neither utu_t nor auxiliary variables.

Case Study

As in [2], a household with a 35-kW photovoltaic (PV) system uses a battery to follow its net demand over T=24T=24 hours; the first 200 daily PV profiles and the 100 battery configurations of [7] yield 20 000 instances. We solve the MIQP that models (1d) with binary variables; the quadratic relaxation based on the HCH model of [2] (HCH-QP); HCH-QP with the cylinder as convex QP objective (HCH-SOC); and HCH-QP with (10) (HCH-PER), with Gurobi 13.0.2 (MOSEK 10.2.0 for HCH-PER) through AMPL, on a 2.6 GHz CPU with 2 GB RAM, one thread and default settings; times are solver CPU times reported by AMPL. See [8] for data and code.

#[ptcptd>10−4]\# [p_t^c p_t^d>10^{-4}] (%)pc⋅pdp^c\cdot p^d (kW2^2)Gap (%)Time (ms)Time red. (%)
MIQP00—33.63—
HCH-QP15.87104.44−1.3414.8655.82
HCH-SOC0.080.04−0.000414.6656.40
HCH-PER0.080.04−0.000420.3939.36

Table 1. Results for the set-point tracking problem (20 000 instances).

Table I reports, as in [2], the hours with ptcptd>10−4p_t^c p_t^d>10^{-4} kW2^2, the average pc⋅pdp^c\cdot p^d and gap to the MIQP optimum, and the average time and its reduction with respect to the MIQP. The cylinder removes 99.5% of the hours with simultaneous charging and discharging and 99.96% of pc⋅pdp^c\cdot p^d, and nearly closes the gap of HCH-QP. In 19750 instances (98.8%), HCH-SOC returns a schedule without simultaneous charging and discharging. Such a schedule is feasible for (2) and, as the penalty in (6) vanishes, its tracking error equals the lower bound given by the relaxation, so it is optimal; the MIQP confirms it (relative tolerance 10−410^{-4}). The other 250 involve 11 small batteries with low efficiencies, which suggests a coupling of periods through the state-of-charge limits that a single-period hull cannot capture. HCH-PER gives the same bounds at a higher cost: P‾d≠P‾c\overline{P}^d\ne\overline{P}^c in 99 batteries, but no hour with simultaneous charging and discharging reaches ptd+ptc>min⁡{P‾d,P‾c}p_t^d+p_t^c>\min\{\overline{P}^d,\overline{P}^c\}, where the perspective is tighter (Section III-A). HCH-SOC is as fast as HCH-QP, whose optima are not unique: the tracking error fixes only the net power ptd−ptcp_t^d-p_t^c, so every feasible split of it into charging and discharging is optimal, and the solver picks one.

Conclusions

The convex hull of the single-period epigraph of the tracking error under complementarity is the epigraph of a parabolic cylinder, the tightest convex quadric through its two branches. The resulting SOC inequality penalizes simultaneous charging and discharging without binary variables and nearly eliminates it. With power limits and the mode variable, the perspective reformulation gives the single-period hull. Future work will address several units and multi-period hulls.

References

  1. [1]J. M. Arroyo, L. Baringo, A. Baringo, R. Bolaños, N. Alguacil, and N. G. Cobos, “On the use of a convex model for bulk storage in MIP-based power system operation and planning,” IEEE Trans. Power Syst., vol. 35, no. 6, pp. 4964–4967, 2020.DOI
  2. [2]D. Pozo, “Convex hull formulations for linear modeling of energy storage systems,” IEEE Trans. Power Syst., vol. 38, no. 6, pp. 5934–5936, 2023.DOI
  3. [3]M. Elgersma, M. Morales-España, K. I. Aardal, N. Helistó, J. Kiviluoma, and M. M. de Weerdt, “Tight MIP formulations for optimal operation and investment of storage including reserves,” IEEE Trans. Power Syst., vol. 41, no. 4, pp. 2428–2440, 2026.arxiv.org/abs/2411.17484
  4. [4]F. Al Taha and E. Bitar, “When are lossy energy storage optimization models convex?” in Proc. IEEE Power & Energy Soc. Gen. Meeting (PESGM), Austin, TX, USA, Jul. 2025, pp. 1–5.
  5. [5]P. Belotti, J. C. Gómez, I. Polik, T. K. Ralphs, and T. Terlaky, “On families of quadratic surfaces having fixed intersections with two hyperplanes,” Discrete Appl. Math., vol. 161, no. 16-17, pp. 2778–2793, 2013.DOI
  6. [6]S. Ceria and J. Soares, “Convex programming for disjunctive convex optimization,” Math. Program., vol. 86, no. 3, pp. 595–614, 1999.DOI
  7. [7]D. Pozo, “Appendix and dataset: Convex hull formulations for linear modeling of energy storage systems,” 2023, [Online]. Available: https://github.com/DavidPozo/ESS_data.DOI
  8. [8]OASYS, “Data and code for second-order cone valid inequalities for set-point tracking with energy storage,” 2026. [Online]. Available: https://github.com/groupoaysys/storage_valid_inequalities

Paper details

vlnc:W_qfrsffeu
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents