Cloud cover, changes on irradiance patterns, and power demand fluctuations encourage the use of forecasting techniques for microgrid management. In addition to estimated
\(N-1\) reserves to face generation outages under
\(N-1\) criterion, the microgrid still needs to deal with both, solar PV and demand forecasting errors. The inherent random nature of these errors increase system regulation reserve requirements in order to balance generation and demand. Regulation reserves and (
i.e., spinning and non spinning reserves) are in charge to balance the system by adjusting the power output of generating units [
35,
36]. In this work, a chance-constraint approach is proposed to allocate the regulation reserve to each generation based in the estimated regulation reserve requirements.
Assuming the existence of hourly forecasts for electric power demand,
\(D^{\text {el}}_{t}\) and solar PV generation,
\(p^{\text {av}}_{\text {pv},t}\), both actual random demand
\(\tilde{D}^{\text {el}}_{t}\) and random solar generation schedule
\(\tilde{p}_{\text {pv},t}\) can be expressed as (
28) and (
29) respectively:
$$\begin{aligned} \tilde{D}^{\text {el}}_{t}&= D^{\text {el}}_{t} + \varepsilon _{D,t},&\forall t \in \mathcal {T} \end{aligned}$$
(28)
$$\begin{aligned} \tilde{p}^{\text {av}}_{\text {pv},t}&= p_{\text {pv},t} + \varepsilon _{\text {pv},t},&\forall t \in \mathcal {T} \end{aligned}$$
(29)
where
\(\varepsilon _{D,t}\) and
\(\varepsilon _{\text {pv},t}\) represent hourly demand and solar PV estimation and forecast errors respectively, which are considered as independent random variables. The model so far has assumed that expected (or forecasted) demand can be met by both, forecasted solar PV and fuel-fired generation. However, random fluctuations in either demand or solar PV generation are not properly balanced in the model. Thus, it is essential to consider such random fluctuations on an hourly basis to properly scheduling regulation reserves.
Our approach is based on the idea of guaranteeing that electric demand can be satisfied with high probability. To do so, an amount of reserve is necessary to cover fluctuations of the net forecasting error
\(\varepsilon _{t}\) defined as:
$$\begin{aligned} \varepsilon _{t} = \varepsilon _{D,t} - \varepsilon _{\text {pv},t}. \end{aligned}$$
Net forecasting error can be either positive or negative, depending on the quality of forecasts. Net positive error requires additional generation; whereas net negative error requires a decrease in the generation to maintain demand-generation balance. As a result, in this approach, we distinguish the positive and negative parts of
\(\varepsilon _{t}\) in order to schedule positive and negative regulation reserves. That is,
$$\begin{aligned} \varepsilon _{t} = \varepsilon _{t}^+ - \varepsilon _{t}^- \end{aligned}$$
where
\(\varepsilon _{t}^+ = \max \left( \varepsilon _{t},0\right)\) and
\(\varepsilon _{t}^- = \max \left( -\varepsilon _{t},0\right)\). Therefore, we need to compute positive reserves
\(\text {SR}^+_{t}\) and negative reserves
\(\text {SR}^-_{t}\) such that:
$$\begin{aligned} \mathbb {P} (\varepsilon _{t}^+&\le \text {SR}^+_{t}) \ge 1 - \eta _t^+ ,\quad \forall t \in \mathcal {T} \end{aligned}$$
(30)
$$\begin{aligned} \mathbb {P} (\varepsilon _{t}^-&\le \text {SR}^-_{t}) \ge 1 - \eta _t^-,\quad \forall t \in \mathcal {T} \end{aligned}$$
(31)
According to the probabilistic criteria exposed in (
30) and (
31), deterministic regulation reserves
\(\text {SR}^+_{t}\) and
\(\text {SR}^-_{t}\) guarantee that both, positive and negative imbalances will be covered with probability
\(1-\eta ^+\) and
\(1-\eta ^-\) or higher, respectively. In this sense, reserve amounts
\(\text {SR}^+_{t}\) need to be at least the
\(\left( 1-\eta _t^+\right)\)-th quantile
\(q^+_{t,1-\eta ^+}\) of the probability distribution of
\(\varepsilon _t^+\). Likewise, and
\(\text {SR}^-_{t}\) should be at least the
\(\left( 1-\eta _t^-\right)\)-th quantile
\(q^-_{t,1-\eta ^-}\) of the probability distribution of
\(\varepsilon _t^-\):
$$\begin{aligned} \text {SR}^+_{t}&\ge q^+_{t,1-\eta ^+}, \quad \forall t \in T \\ \text {SR}^-_{t}&\ge q^-_{t,1-\eta ^-}, \quad \forall t \in T. \end{aligned}$$
Both quantiles, in general, can be computed from
$$\begin{aligned} q^+_{t,1-\eta ^+}= & {} {\left\{ \begin{array}{ll} q_{t,1-\eta ^+}^{\varepsilon }, &{} \text {if } 1 - \eta ^+ \ge \mathbb {P}(\varepsilon _t < 0)\\ 0, &{} \text {otherwise,} \end{array}\right. } \end{aligned}$$
(32)
$$\begin{aligned} q^-_{t,1-\eta ^-}= & {} {\left\{ \begin{array}{ll} -q_{t,\eta ^-}^{\varepsilon }, &{} \text {if } 1 - \eta ^- \ge \mathbb {P}(\varepsilon _t > 0)\\ 0, &{} \text {otherwise.} \end{array}\right. } \end{aligned}$$
(33)
where
\(q_{n}^{\varepsilon } = F^{-1}_\varepsilon \left( n\right) , \ n \in \left[ 0, 1\right]\) refers to the
n-th quantile of the distribution of the net forecast error
\(\varepsilon\); and
\(F^{-1}_\varepsilon \left( \cdot \right)\) is its corresponding inverse cumulative distribution function. An important feature of this formulation is that it can be employ with any probability distribution. Both
\(q^+_{t,1-\eta ^+}\) and
\(q^-_{t,1-\eta ^-}\) result in non-negative quantities.
When both errors are independent and normally distributed, i.e.,
\(\varepsilon _{\text {pv},t} \sim \mathcal {N}\left( \mu _{\text {pv,t}}, \sigma ^2_{\text {pv},t}\right)\) and
\(\varepsilon _{D,t} \sim \mathcal {N}\left( \mu _{D,t}, \sigma ^2_{D,t}\right)\), it is known for a fact that net error is also normally distributed. That is,
$$\begin{aligned} \varepsilon _t \sim \mathcal {N}\left( \mu _{D,t}-\mu _{\text {pv,t}}, \sigma ^2_{\text {pv},t} + \sigma ^2_{D,t}\right) , \quad \forall t \in T, \end{aligned}$$
therefore,
n-th quantile can be computed using (
34), where
\(\text {erf}^{-1}(n)\) is the inverse error function:
$$\begin{aligned} q^\varepsilon _{n} = \mu _{D,t} - \mu _{\text {pv,t}} + \text {erf}^{-1}\left( n\right) \sqrt{\sigma _{\text {pv},t}^2 + \sigma _{D,t}^2}. \end{aligned}$$
(34)
In the proposed approach, we assume that solar forecast error uncertainty is larger as long as solar production increases. In this sense, the hourly power output of the solar PV system ends up affecting parameters of the solar forecast error probability distribution. Solar PV mean forecast error
\(\mu _{\text {pv,t}}\) and standard deviation
\(\sigma _{\text {pv},t}\) are parameterized as percentages
\(\alpha _{\text {pv}}\) and
\(\vartheta _{\text {pv}}\) respectively of hourly scheduled PV power output
\(p_{\text {pv},t}\). That is,
$$\begin{aligned} \mu _{\text {pv,t}}&= \alpha _{\text {pv}} p_{\text {pv},t},\quad \forall t \in T \\ \sigma _{\text {pv},t}&= \vartheta _{\text {pv}} p_{\text {pv},t}, \quad \forall t \in T \end{aligned}$$
Quantiles
\(q_{t,1-\eta ^{+}}^{+}\) and
\(q_{t,1-\eta ^{-}}^{-}\) that define the amount of positive and negative regulation reserves respectively are computed as follows:
$$\begin{aligned} q_{t,1-\eta ^{+}}^{+}&= \mu _{D,t} - \alpha _{\text {pv}}p_{\text {pv},t} + \text {erf}^{-1}(1-\eta ^{+})\sqrt{\sigma _{D,t}^2 + \vartheta ^2_{\text {pv}}p_{\text {pv},t}^2},&\forall t \in \mathcal {T} \end{aligned}$$
(35)
$$\begin{aligned} q_{t,1-\eta ^{-}}^{-}&= \alpha _{\text {pv}}p_{\text {pv},t} - \mu _{D,t} - \text {erf}^{-1}(\eta ^{-})\sqrt{\sigma _{D,t}^2 + \vartheta ^2_{\text {pv}}p_{\text {pv},t}^2},&\forall t \in \mathcal {T} \end{aligned}$$
(36)
Quantiles defined by Eqs. (
35) and (
36) represent the minimum amount of regulation reserves to schedule on an hourly basis. This formulation leads to the second-order cone nature of the proposed model. Additionally we do not impose that
\(q_{t,1-\eta ^{+}}^{+},q_{t,1-\eta ^{-}}^{-} \ge 0\) in the model given the non-convexity of such constraints. Rather we verify that the optimal solution satisfies the non-negativity constraints on both quantiles. Finally, the positive and negative regulation reserve requirements are shown in (
37) and (
38) respectively, resulting in a set of conic constraints:
$$\begin{aligned} \sum _{{j\in \mathcal {J}: j\ne pv}}\Delta p_{j,t}^{\text {SR}+} + \Delta p^{\text {SR}+}_{\text {BESS},t}&\ge q^+_{1-\eta _t^+},&\forall t\in T, \end{aligned}$$
(37)
$$\begin{aligned} \sum _{j\in \mathcal {J}: j\ne pv}\Delta p_{j,t}^{\text {SR}-} + \Delta p^{\text {SR-}}_{\text {BESS},t}&\ge q^-_{1-\eta _t^-},&\forall t\in T. \end{aligned}$$
(38)
Both
\(N-1\) and positive regulation reserve requirements need to be high enough so as to meet the reliability constraints. However, reserves need to be tuned such that power output cannot exceed the nominal power of any generator as in (
39); and are also bounded by maximum ramp-up power as in (
40). For BESS,
\(N-1\) security reserve and up regulation reserves are limited by the current available energy stored and by its nominal power output as in (
41).
$$\begin{aligned}&p_{j,t} + \Delta p_{j,t}^{\text {SR}+} + \Delta p^{S}_{j,t} \le p^{\text {nom}}_j, \forall j\in \mathcal {J}: j\ne \text {pv}, t\in T \end{aligned}$$
(39)
$$\begin{aligned}&\frac{\Delta p_{j,t}^{\text {SR}+} + \Delta p^{S}_{j,t} + p_{j,t} - p_{j,t-1}}{\Delta t} \le r^{\text {up}}_{j} p^{\text {nom}}_{j}, \forall j\in \mathcal {J}: j\ne \text {pv}, t \in T. \end{aligned}$$
(40)
$$\begin{aligned}&\min \left( \frac{e^{\text {SOC}}_{\text {BESS},t} - \underline{\mu }^{\text {SOC}}_{\text {BESS}} e^{\text {nom}}_{\text {BESS}}}{\Delta t} , p^{\text {nom}}_{\text {BESS}} - p^{\text {dis}}_{\text {BESS},t} \right) \ge \Delta p^{\text {SR}+}_{\text {BESS},t} + \Delta p^{\text {S}}_{\text {BESS},t}, \forall t \in T \end{aligned}$$
(41)
where
\(\Delta p_{j,t}^{\text {SR}+} \ge 0\),
\(\Delta p^{S}_{j,t} \ge 0\),
\(\Delta p^{\text {SR}+}_{\text {BESS},t} \ge 0\), and
\(\Delta p^{\text {S}}_{\text {BESS},t} \ge 0\) correspond to positive generator regulation reserve, generators
\(N-1\) security reserve, BESS up regulation reserves, and BESS
\(N-1\) security reserve, respectively.
Likewise regulation reserve constraints of power generation and BESS required to balance negative shifts in net forecasting error
\(\varepsilon _t\) are shown in constraints (
42)–(
44):
$$\begin{aligned}&p_{j,t} - \Delta p_{j,t}^{\text {SR}-} \ge \underline{\mu }_{j} p_{j}^{\text {nom}}, \forall j\in \mathcal {J}: j\ne \text {pv}, t\in T \end{aligned}$$
(42)
$$\begin{aligned}&\frac{p_{j,t-1} - p_{j,t} - \Delta p_{j,t}^{\text {SR}-}}{\Delta t} \le r_{j}^{\text {down}} p_{j}^{\text {nom}}, \forall j\in \mathcal {J}: j\ne \text {pv}, t\in T \end{aligned}$$
(43)
$$\begin{aligned}&\Delta p^{\text {SR}-}_{\text {BESS},t} \le \min \left( \frac{ \bar{\mu }^{\text {SOC}}_{\text {BESS}} e^{\text {nom}}_{\text {BESS}} - e^{\text {SOC}}_{\text {BESS},t}}{\Delta t} , p^{\text {nom}}_{\text {BESS}} - p^{\text {ch}}_{\text {BESS},t} \right) , \forall t \in T \end{aligned}$$
(44)
where
\(\Delta p_{j,t}^{\text {SR}-} \ge 0\) and
\(\Delta p^{\text {SR}-}_{\text {BESS},t} \ge 0\) correspond to negative generator and BESS regulation reserves, respectively. Down regulation reserves are bounded by both the minimum operational limit and the ramp-down limit for the generators. For the BESS, down regulation reserves are bounded by the available storage capacity without exceeding its maximum SOC limit, and by its nominal power input.
This regulation reserve approach allows to differentiate probabilities according to the needs of the microgrid. Violation probabilities
\(\eta ^+\) and
\(\eta ^-\) must be small but can be different. Thus, it is possible to assign different small violation probabilities to either positive or negative imbalances. Additionally, probabilities can be different throughout the day. Also, any resulting probability distribution of forecast errors can be employed. In fact, even dependent random errors can be easily adapted to this formulation. Mitigating the proposed model conservativeness can be done by adjusting the maximum constraint violation probability
\(\eta _t^{(\cdot )}\). The larger this probability, the less conservative the solution. Constraints (
37) and (
38) allow to scheduling enough generation to face potential fluctuations (forecast errors) in demand and renewable generation as a function of the probability
\(\eta _t^{(\cdot )}\).