含储能新型电力系统的现货市场出清模型研究

咨目

》储能在新型电力系统中的角色

》储能参与电力市场与电力调度的挑战

》考虑不确定性和非预期性的多阶段优化模型与算法

》结果与讨论

储能在新型电力系统中的角色

储能在新型电力系统中的重要性

»The duck curve »Net demand is reduced during the day (the duck's t demand is re »Net demand sharply increases at sunset (the duck's neck) » The need for flexibility significantly increases »The canyon curve » Recently, CAISO's net load has reached zero or gone negative »Dispatchable resources are sharply reduced during the daytime »The need for flexibility conversely increases as the sun sets » Solutions: » Charging/discharging of batteries »Additional forms of energy storage » …

新型电力系统中的储能技术

California Energy Storage Alliance (CESA)

山东东营津辉795兆瓦/1600兆瓦时集中式储能项目

国内的大规模储能项目（抽水蓄能）

河北丰宁3600兆瓦水整能电站 http://he.people.com.cn/n2/2022/0124/c192235-35109507.html

我国储能参与电力现货市场的现状

》电力现货市场建设情况

第一批电力现货试点区域（南方（从广东起步）、蒙西、浙江、山西、山东、福建、四川、甘肃8个地区）中，山西、广东的电力现货市场已分别于2023年12月22日、2023年12月28日转入正式运行

》第二批6个试点地区，江苏、安徽数、辽宁、湖北、河南这5个地区全年共完成9次结算试运行

》相关政策

2023年9月，国家发改委、国家能源局印发《电力现货市场基本规则(试行)》（发改能源规[2023)1217号1，明确提出推动分布式发电、负荷聚合商、储能和拟电厂等新型经营主体参与交易。

》储能如何参与电力现货市场？

独立储能参与电力现货市场的报价方式主要有报量报价、报量不接价两种方式

山东、山西、甘肃、青海、广东等省份明确了独立储骼蓉与现货市场的规则

东盟产置组（电网络：力伙立主参与联指市场，充效目价格均案用于在节点的分钙电络：电募能服联合效电机8，在联想车场以限量最价的力式学与文

山西省我立效月自主选择以托量据“款“提量不提诊方式参与联发场，良新错企业用究的客量不究实能作为登体参与现资市场

》山东省独立储能在现货市场电能量交易中按照报呈不报价原则出清，上网电量价格按照市场出清价格结算，并享受容量补偿费

美国ISOs/RTOs如何管理电力现货市场中的储能？

》NYISOandMISO

»Energy storage must offer in as a generator or load separately in the day-ahead market

» But may allow different offers in the real-time market

> PJM and NYISO (Phase 2)

»The ISO/RTO software will optimize the operational mode of PSH (pumped storage hydro) based on production cost minimization

>ISO-NE

»Includes features to more realistically capture the parameters of PSH in pumping modes, like minimum pumping levels, minimum run and down times, etc

》CAISO

»Has plans for its storage resources to provide cost offers from withdrawal mode to injection/generation mode if the resources have a continuous operation between maximum withdrawal injection/generation mode if the rescources have a continuous operation between maximum withdrawal

澳洲电力市场中的电化学储能

HornsdalePowerReserve近期盈利情况

储能参与电力市场与电力调度的挑战

储能在现货市场中的成本回收问题

系统参数及市场申报情况 <table><tr><td rowspan=1 colspan=1></td><td rowspan=1 colspan=1>负荷(MW)</td><td rowspan=1 colspan=1>新能源发电出力(MW)</td><td rowspan=1 colspan=1>火力发电机组报价(¥/MWh)</td><td rowspan=1 colspan=1>储能充电/放电报价（¥/MWh)</td></tr><tr><td rowspan=1 colspan=1>时段1</td><td rowspan=1 colspan=1>100</td><td rowspan=1 colspan=1>160</td><td rowspan=1 colspan=1>300</td><td rowspan=1 colspan=1>200/400</td></tr><tr><td rowspan=1 colspan=1>时段2</td><td rowspan=1 colspan=1>90</td><td rowspan=1 colspan=1>10</td><td rowspan=1 colspan=1>300</td><td rowspan=1 colspan=1>100/200</td></tr><tr><td rowspan=1 colspan=1>时段3</td><td rowspan=1 colspan=1>120</td><td rowspan=1 colspan=1>80</td><td rowspan=1 colspan=1>300</td><td rowspan=1 colspan=1>200/400</td></tr></table>

口火力发电机组出力上下限：10MW\~100MW 口储能充、放电功率上限：100MW 口储能充、放电效率：95% 口储能SOC初始值、最大值：100MWh、200MWh 口储能约束：末时段SOC与初始值一致

出清及结算结果

口以社会据利最大化为日标，使用混合整数规划进行出清计

口基于拉格朗日乘子得到每个时段的边际出清价格： ¥276.71、¥300.00、¥300.00

口根据电价和储能各时段的充放电功率进行结算，可知储能的收益为￥-1,365.00

储能在现货市场中的模型非凸性

储能一般模型的非凸性

口充放电工况下的能量-功率特性不同

口储能的充放电工况互斥，不允许能量往下松弛

口储能的能量一功率可行域是非凸的

口在电力现货市场出清模型中需引入二进制决策变量（0-1）来刻画非凸性

口非凸优化问款求辑的算法复杂是指数级别的

非可变抽蓄的功率特性

模拟结果

口挂水蕾能机组（变速推水蕾能机组除外）从发电工况转变为拍水（泵）工况后，必须满功率运行

口向电网注入的功率从寒操然变成一个较大的、固定的负功率，功率无法光滑调节

口功率特性存在显苦的离散性

新能源随机性？

随机性

不确定性

模糊性

口随机性：<sub>具备概率特性，例如，用伯努利分布刻间抛硬币实验</sub> 口不确定性：一般意义上的结果不明确。例如，鲁摔优化中的不确定集（所有可能的结果均会落在集合之内）口模糊性：<sub>模型不明确。例如，不如道新能源发电出力的具体概率分布，使用模糊集刻画新能够源发电出力</sub>

考虑新能源发电不确定性的出清模型

<table><tr><td>随机规划（随机性）</td><td>鲁棒优化（不确定性）</td><td>分布鲁棒优化（模糊性）</td></tr><tr><td></td><td> $min_{x \in (A, (F XZ)} c_{1}^{⊤} x + Q (x, ξ)$ </td><td> $min_{x \in {x / x \geq β}} c_{1}^{-} x + max_{P \in P} E_{ε P} [Q (x, ξ)]$ </td></tr><tr><td> $min_{x \equiv ⟨ x, F x \geq f ⟩} c_{1}^{'} x + E_{ξ - x} [Q (x, ξ)]$ $with Q (x, ξ) = min_{min (x + δ y - ζ ξ = \partial_{1} ξ)} c_{2} y$ </td><td> $\mathrm { w i t h } \underbrace { \mathcal { Q } ( x , \overline { { \xi } } ) } _ \substack { \lvert x ( 1 , i ) \rvert \mathcal { A } + B y + C \overline { { \xi } } = \mathcal { A } \} } \begin{array} { r } { c _ { 2 } ^ { \top } y } \end{array}$ $and {y ∣ Ax + By + C ξ = d} uncmpty \forall ξ \in Ξ$ </td><td> $with Q (x, ξ) = min_{min (x + ℏ ν - Γ c^= d)} c_{2} y$ $and P bcing an ambiguity set$ </td></tr></table>

□x日前决策变量，如开机计划； $Q (x, ξ)$ 日内最优成本函数；ξ不确定量或随机向量，如新能源发电出力；P随机向量的概率分布

<table><tr><td colspan="2">样本平均估计+Benders分解对偶转化、极限场景辨识+迭代</td><td colspan="2">对偶转化、极限分布辨识+迭代</td></tr><tr><td colspan="2"> $min_{x \in {x \in f}} c_{1}^{⊤} x + \sum_{x \in S} ω, Q (x, ξ_{s})$ $\begin{array}{c} \left[ Q ( x , \xi _ { 1 } ) = \operatorname* { m i n } _ { y _ { 1 } \in \{ 1 , | \mathcal { A } ( x + \beta ) + C _ { 3 1 } ^ { \circ } = \mathcal { A } | 1 } & { } \end{array} \right. c _ { 2 } ^ { \top } y$ s.t. ...</td><td> $min_{x \in {x, x ˉ \geq f}} c_{1} x + \supseteq (x, ξ ˉ)$ </td><td> $min_{x \in {x, \cdot; x \geq β}} c_{1}^{-} x + η$ $⌈ η > \sum_{s_{1} \in S_{1}} ω_{s_{1}} Q (x, ξ_{s_{1}}^{Fxunemal})$ </td></tr></table>

非确定型决策模型中的非预期性（non-anticipativity）约束

并行的场景（忽略了非预期性）

场景树（考虑了非预期性）

口求解非确定型决策问题需要对未来场晨进行采样（样本平均估计、极限场景、极端瞬率分布等）基于采样进行模拟决策，制定最优策略

口如果不确定量的样本是并行的时序轨遗，并基于这些轨迹进行模拟决策，则相当于利用未来的观测伯来制定当前策略，是偏乐观的

口以场最树的形式采样并构建确定型的决策模型，可以满足非预期性约束保证任意时刻的决策仅依赖于该时刻以前的观测值家正在司用的历安这之下决能方完是一的

口场票数量呈指数增长

非预期性的实例

A multistage game investors play with the uncertainty of price:

One owes a stock, and now he/she is in 05/2022. In determining whether to sell the stock, he/she needs to account for the uncertainty of the future price. Depending on his/her uncertainty model and profit maximization model, he/she may or may not sell the stock in 05/2022.

非预期性的实例（2）

A multistage game investors play with the uncertainty of price:

If his/she model does not respect the nonanticipativity, i.e., assuming that the investment decision can be made with all future price signals (in a scenario), then the optimal here-and-now decision is to sell the stock right now, since he/she can profit from any short-term growth in the future.

考虑不确定性和非预期性的多阶段优化模型与算法

考虑非预期性的非确定型优化模型

Optimization models

» Single-stage optimization model

» Determines beforehand how the resources linearly react temporally independently to the deviation of power output of variable renewable generation (VRE)

»Two-stage optimization model

» Determines beforehand the operational range of temporally coupled resources (e.g., energy-limited, ramping-limited) so that they will be dispatched temporally independently

»Multistage optimization model

» Precisely model the sequential decision making forward through time

How does it work?

ceactionfunction via static robust optimization. Then dispatch following the simple rule

Restrict the decision space + Temporally decomposition

Find a robust operational boundaries a <sub>adis</sub> Then implement the here-and-now decision

The multistage optimization model works simply because it is precise

用于电力系统优化调度的储能一般模型

\begin{array} { r } { \mathrm { 0 } \leq p _ { t } ^ { \mathrm { D C } } \leq P ^ { \mathrm { D C , m a x } } \qquad \mathsf { L o w e r / u p p e r / i m i t ~ o f ~ d i s c h a r g i n g ~ p o w e r } } \end{array} \begin{array} { r } \mathrm { { \left[ \begin{array} { l l l l l l l l l l l } { 0 \leq p _ { \ell } ^ { \mathrm { C H } } \leq { P ^ { \mathrm { C H } } , } { m a x } \end{array} } } \qqua\right.d} & { \mathrm { L o w e r / u p p e r ~ l i m i t ~ o f ~ c h a r g i n g ~ p o w e r } } \end{array} \begin{array} { r l } { \mathrm { ~ ~ \lambda ~ } } & { { } 0 \leq e _ { t - 1 } - ( p _ { l } ^ { \mathrm { { D C } } } / \eta ^ { \mathrm { { D C } } } ) \Delta } \end{array}

Upper limit of the state of energy (SOE)

The operation of energy

e_{t - 1} + η^{CH} p_{t}^{CH} Δ \leq E^{max}

Lower limit of the SOE

coupled due to the SOE

constraints

pPCpCH = 0 Cannot discharge and charge simultaneously

The discharging/charging power deployed at time t steers the SOE to a new state, which will become the initial state of the next period and limit the usability of the storage at time t+1.

When working on a temporally coupled system, the two-stage uncertainty-aware optimization model does not respect the nonanticipativity

基于储能最优鲁棒能量边界的多阶段优化模型

E_{s}^{min} \leq ϵ_{s, t}^{S} \leq ϵ_{s, t}^{S} \leq E_{s}^{max} \forall s \in S, t \in T e_{s, t}^{S} - e_{s, t - 1}^{S} \leq η_{s}^{Ch} P_{s}^{Ch, max} Δ t \forall s \in S, t \in T e_{s, t}^{S} - e_{s, t - 1}^{S} \geq - \frac{1}{η _{s}^{Dis}} P_{s}^{Dis, max} Δ l \forall s \in S, l \in T e_{s, t}^{S} \leq e_{s, t}^{S} \leq e_{s, t}^{S} \forall s \in S, t \in T 0 \leq p_{s, t}^{Ch} \leq x_{s, t}^{S} P_{s}^{Ch, max} \forall s \in S, t \in T 0 \leq p_{s, t}^{Dis} \leq (1 - x_{s, t}^{S}) P_{s}^{Dis, max} \forall s \in S, t \in T x_{s, t}^{S} \in {0, 1} \forall s \in S, t \in T r^{e_{s, t}^{S}} = ϵ_{s, t - 1}^{S} + η_{s}^{Ch} p_{s, t}^{Ch} Δ t - \frac{1}{η _{s}^{Dis}} p_{s, t}^{Dis} Δ t \forall s \in S, t \in T C_{s, t}^{S} = c_{s, t}^{S} = c_{s, 0}^{S} \forall s \in S, t = ∣ T

将储能SOE决策变量转化成不确定量，实现时序解揭！

∙ : e_{s, t}^{S} = u_{s, t}^{S} + η_{s}^{Ch} p_{s, t}^{Ch} Δ t - \frac{1}{η _{s}^{Dis}} p_{s, t}^{Dis} Δ t \forall s \in S, t \in T

口基于鲁摔优化原理，在日前优化谢度中确定储能在日内各时段的能量边界

口日内、实时运行时，储能只需要保证每个时段的能量水平在边界之内

口可以使用两阶段（分布）鲁棒优化的成熟算法求解含储能的非确定型多阶段优化调度问题

储能调度在日前、日内优化中的衔接

口日前电力现货市场确定的储能最优鲁摔能量边界仅在每个滚动时域的最后一个时间口对储能能量边界进行限定口日内电力现货市场采取前瞻（look-ahead）调度模式，基于模型预测控制原理进行滚动出清 <sub>口联能发挥储能在日内电力现货市场的灵活性，</sub> 文能兼顾日前较长周期市场的蛋优决策，避免短视问题

前瞻调度策略1可行域示意图

日内动态优化策略#1：寻虑落动时域内的多时段的美量边界的秉

储能调度在日前、日内优化中的衔接（2)

口在日前得到的储能显优鲁棒能量边界（以及其他可调资源的鲁棒运行域）的基础上，采用改进的日前-日内衔接方法进行前瞻调度，更加充分地发挥其灵活资源的调节和互济能力

口日内前瞻调度基于日前确定的完整时域上的鲁摔运行域来开展。能够克服能量受限型灵活资源在日内调度中的短视问题，保证全天调度运行的安全性

考虑单个时段的能量边界约束，独立决策

p_{g, t}^{G *} \leq p_{g, t}^{G} \leq p_{g, t}^{G *} \forall g \in G p_{g, τ}^{G *} \leq p_{g, τ}^{G} \leq p_{g, τ}^{G *} \forall g \in G, τ \in T_{t} e_{s, τ}^{S *} \leq e_{s, τ}^{S} \leq e_{s, τ}^{S *} \forall s \in S, τ = ∣ T_{t} ∣ e_{s, t}^{S *} \leq e_{s, t}^{S} \leq e_{s, t}^{S *} \forall s \in S \forall s \in S, τ \in T_{t} e_{s, τ}^{S *} \leq e_{s, τ}^{S} \leq e ˉ_{s, τ}^{S *} E_{s}^{min} \leq e_{s, τ}^{S} \leq E_{s}^{max} \forall s \in S, τ = T_{t} ∖ ∣ T_{t} ∣

前瞻调度策略2可行域示意图

处理新能源不确定性的分布鲁棒优化方法

Why distributionally robust optimization (DRO) is good?

(e) Ex[(,] x is the first-stage decision P is the ambiguity set of distributions of the random vector ξ f(x), Q(x, ξ) are the here-and-now cost, second-stage cost, respectively

1) The value function is bounded (robustness) 2) The worst-case cost expectation is minimized (out-of-sample performance guaranteed)

Stochastic programming (SP): 1) The value function may be unbounded (no robustness) 2) The cost expectation w.r.t. an empirical dist. is minimized

Robust optimization (RO) Robust optmizatilon (ROI: 1) The vslus function is bcurdsd [robustness] minimized

基于值函数估计的高效求解算法

结果与讨论

各类型资源的优化结果

A modified IEEE118-bus system with 38% renewable generation capacity <table><tr><td>Resources</td><td>Thermal generators</td><td>ESS</td><td>Wind farms</td></tr><tr><td>» Number</td><td>39</td><td>15</td><td>4</td></tr><tr><td>» Total capacity</td><td>5540 MW</td><td>424 MW, 1463 MWh</td><td>3395 MW</td></tr></table>

»Simulation of chronological operations of intraday markets are performed on 7 typical days. (The day-ahead market clearing model that models the optimal robust energy bounds of energy storage as decisiondependent uncertainty is being studied, and will be presented in the near future!)

»ESS function well to hedge against the variability and uncertainty of VRE

Dispatch results on a day with lower VRE output (Day 2)

Dispatch results on a day with higher VRE output (Day 6)

经济性对比和分析

»Model 2, a distributionally robust intraday market clearing model that does not respect nonanticipativity, induces a slightly higher (0.22%) average cost.The costs of Model 2 can be slower, but the cost on Day 6 is significantly higher.

»Model 3, a deterministic-optimization-based model, incurs load shedding of amount 107.27MWh and 485.64 MWh on Day 1 and Day 2, respectively.

Comparison of dispatch costs <table><tr><td rowspan="2">Day</td><td rowspan="2">Model 1 (ours) Cost ( $) < / t d >< t d co l s p an = "2" > M o d e l 2 < / t d >< t d co l s p an = "2" > M o d e l 3 < / t d >< / t r >< t r >< t d > C os t ($ )</td><td>Difference (%)</td><td>Cost ($)</td><td>Difference (%)</td></tr><tr><td>» Day 1</td><td>1,108,205</td><td>1,111,460</td><td>0.29%</td><td>1,163,699</td><td>5.01%</td></tr><tr><td>» Day 2</td><td>878,541</td><td>878,265</td><td>-0.03%</td><td>1,179,757</td><td>34.29%</td></tr><tr><td>» Day 3</td><td>761,912</td><td>760,541</td><td>-0.18%</td><td>760,459</td><td>-0.19%</td></tr><tr><td>» Day 4</td><td>677,320</td><td>676,143</td><td>-0.17%</td><td>675,594</td><td>-0.25%</td></tr><tr><td>» Day 5</td><td>499,806</td><td>503,264</td><td>0.69%</td><td>494,861</td><td>-0.99%</td></tr><tr><td>» Day 6</td><td>350,460</td><td>356,771</td><td>1.77%</td><td>346,120</td><td>-1.24%</td></tr><tr><td>» Day 7</td><td>303,198</td><td>303,167</td><td>-0.01%</td><td>304,041</td><td>0.28%</td></tr></table>

所提算法的效率

»Algorithm 1: our algorithm, which makes use of the value function of the relaxed (convex) ESS mode

»Algorithm 2: another algorithm that uses the nested column-and-constraint generation method to refine a discrete set of scenarios

» Improvements:

» less runtime (49.39s vs. 153.03s on Day 2; 112.46s vs. 525.55s on Day 6)

» less scenarios/events (13 vs. 22 on Day 2; 15 vs. 22 on Day 6)

Performance of two algorithms (Day 2)

Performance of two algorithms (Day 6)

Thanks!

Contact me at +86 15914392043,

or eezhengxd@scut.edu.cn

Read more: X. Zheng. M. E. Khodayar, J. Wang, M. Yue and A. Zhou, "Distributionally Robust Multistage Dispatch With Discrete Recourse of Energy Storage Systems." in IEEE Transactions on Power Systems. doi:10.1109/TPWRS.2024.3369664

含储能新型电力系统的现货市场出清模型研究