1. Introduction
The increasing issues of the environment and climate change are becoming increasingly vital constraints on power industry development. In the past few years, interests in the use of renewable energy have been growing steadily [
1,
2]. However, the inherent intermittency and variability of renewable resource (e.g., wind and solar) increase the fluctuation of net load significantly and require additional flexibility resources to smooth the load curve [
3,
4]. Besides, in cases where wind and solar generation are more than load minus must-run generation, the excess of wind and solar power has to be curtailed to ensure the power balance between the demand side and supply side. Thereby, it poses significant challenges to the safe and economic operation of standalone microgrid with a high penetration level of wind and solar power. Both battery storage and hydro storage are a topic of great importance for addressing the uncertain renewable energy. However, the authors in [
5] stated that hydro storage is more cost-competitive than battery storage, and presents practical potential and technically feasible opportunities for power supply in remote areas. Thus, coordinating the traditional controllable hydropower with the uncontrollable wind and solar power to form a hybrid multi-source microgrid is a promising solution for promoting renewable energy penetration [
6,
7].
There have been many works in the literature to investigate the coordinated operation of wind-solar-thermal-hydro (WSTH) from various aspects. For instance, Reference [
8] proposed a complementary operation of WSTH system to address the problem of renewable energy curtailment, considering minimizing the fluctuation of thermal output and maximizing the penetration of renewable energy. Reference [
9] developed a new strategy for the day-ahead operation of the WSTH system with storage, which aims to provide the “best-fit” scheduling by minimizing day-ahead and real-time operation costs. The authors in [
10] explored the principle of wind-hydro compensation, and the principle was also quantitatively analyzed in a power grid. In [
11], a multi-time-scale dispatching mechanism for WSTH and battery was proposed to handle renewable energy fluctuations. Note that unlike conventional controllable units, the outputs of wind power and solar power are highly uncertain and unpredictable. Even the best commercially available methods applied in wind/solar precasting are with the 15–20% error [
12]. It should be noted that even with 10–15% error, it may result in great uncertainties for the balance of supply and demand, which affects the optimal power system operation [
13,
14,
15]. For example, Reference [
16] studied the output shortage, spilled water, and power curtailment risk in WSTH system to enhance resource utilization efficiency. In [
17], the authors investigated the effectiveness of locational marginal pricing in WSTH to reduce the wind power curtailment. In [
18], a model predictive control method was developed to address the frequency issue with considering the intermittent generation. Hence, the coordinated operation strategy of WSTH system is still needed, especially in a scenario of large scale penetration of uncertain renewable energy.
The risk management methods for uncertain wind and solar power have been studied in many papers. From existing literature, the stochastic optimization method has been successfully used in stochastic microgrid scheduling for addressing various uncertainties [
19,
20,
21]. This method is inspired by portfolio optimization, and aims to study how to get the optimal penetration level of uncertain renewable energy. In [
22], authors proposed to use the mean-variance (MV) model to deal with the wind power integrated stochastic power flow problem. Note that the MV is based on the assumption that the utility distribution functions are quadratic, or the return of each asset is with a norm distribution [
23]. The study carried out by Chen et al. [
24] pointed out that the MV is not suitable to be employed for addressing uncertain wind power due to the fact that the returns brought by wind power are not symmetrically distributed around the mean. Noting that using variance as the risk measure may sacrifice the higher return in stochastic optimization [
25], semi-variance was employed in [
26] in an uncertain power system environment, which tries to get the undesirable deviation of uncertain wind power. Comparison results illustrated that semi-variance is more convincing and reliable for stochastic power system scheduling under risk aversion [
26,
27]. However, the variance and semi-variance rely on the distribution of cost expectation, and they are not sufficient to describe the uncertain risk when the cost expectation does not follow normal distribution [
28]. In addition, the authors in [
24] asserted that in non-normal cost expectation, the higher moments should be considered in decision making of power system optimization.
All aforementioned risk evaluation methods were formulated based on the probability theory by characterizing the uncertainty as a random variable. Besides, stochastic optimization relies on the repeated samplings, and is thought of as occupying much time. Some researchers in the literature stated that many fuzzy characteristics exist commonly in stochastic power system operation, so that the returns brought by intermittent and variable renewable energy contain other aspects of uncertainty, such as ambiguity and vagueness [
13,
29]. Abdul-Rahman and Shahidehpour [
30] stated that the uncertainty of bus loads can be addressed using a fuzzy set based optimization model. Subsequently, in renewable energy integrated optimization problems, the uncertainty can also be addressed by fuzzy optimization [
31,
32,
33]. However, the study carried out by Simoneli [
34] indicated that taking the variance as a risk index is less convincing than entropy in asset allocation. For example, Armando et al. employed entropy to evaluate the loss of load risk in generating systems [
35] and overloads risk in transmission equipment [
36]. However, entropy evaluates the uncertainty of cost expectation with low and high extreme situations, and it will inevitably sacrifice the higher assets brought by renewable energy and limit its penetration level. Besides, in the situation of high energy consumption, the available dispatchable resources to address the fluctuation are already used, and a sudden reduction of wind or solar power output can have critical consequences on the system reliability [
37,
38]. In view of the above discussion, the main contributions of this paper can be summarized as below:
This paper presents a WSTH-coupled multi-source standalone microgrid (WSTHcMSSM) considering pumped hydro storage and demand response (DR) to mitigate the challenge of supply and demand imbalance, resulting in the effect of promoting wind and solar power consumption.
A conditional value-at-credibility (CVaC)-based quantitative risk-averse model is first developed to address the uncertainty of wind and solar power in WSTHcMSSM system, making the system reach a trade-off option.
In consideration that the most severe issues caused by wind and solar power fluctuation happen during the peak load, this paper proposes a load partitioning method to determine the time-of-use (TOU) price in DR to explore the potential flexibility of WSTHcMSSM microgrid scheduling.
The rest of the paper is organized as follows.
Section 2 models the energy systems of WSTHcMSSM.
Section 3 formulates the CVaC-based risk evaluation for WSTHcMSSM.
Section 4 gives the numerical studies. Finally, a conclusion and interesting future work are drawn in
Section 5.
3. CVaC-Based Risk Evaluation for WSTHcMSSM
Given the system load demand, hydraulic continuity limits, water discharge limits, reservoir storage volume limits, unit generation limits, and available wind and solar power at a specified time horizon, this paper presents a conditional value-at-credibility (CVaC) model to quantitatively evaluate the uncertain risk associated with wind and solar forecasting error, and optimally dispatch of micro-turbine, wind turbine, solar station, and hydro source in WSTHcMSSM system, which amounts to taking the advantage of multi-source complementarity. To this end, the optimization objective and operation constraints of the CVaC-based WSTHcMSSM with DR and pumped hydro storage are given as follows:
where (
8a) represents the operation profit of WSTHcMSSM by selling power to end users,
denotes the fixed price at time
t,
represents the time-of-use (TOU) price at time
t,
denotes the number of hydro turbines,
denotes risk measures with respect to wind or solar power and it will be given in the following subsection. In view of this, the objective function defined in this paper is the profit of WSTHcMSSM. (8c) represents the system power balance constraint. (8d) represents the TOU price constraints, where
and
are, respectively, load sets of valley period, flat period and peak period, and
and
are, respectively, the lower and upper values of price deviation
during time
t.
3.1. CVaC for Uncertain Wind and Solar Power
In this subsection, to facilitate the analysis, we take one wind turbine and one solar station for example. At time
t, the joint probability density function of wind and solar power
, where
, can be expressed by:
where
denotes the covariance matrix, and
. Then, the expected values
and
associated with the materialized outputs
and
not exceeding the forecasting values
and
can be expressed by Equations (
10) and (
11), respectively.
Similarly, the expected values
and
associated with the materialized outputs exceeding the forecasting values can be expressed by Equations (
12) and (
13), respectively.
Besides the random characteristics, the outputs of wind and solar power also have fuzzy characteristics, considering the knowledge is incomplete due to the environment and operational conditions. Here, at time
t, the fuzzy measures denoted as
and
are derived from the Cauchy distribution by studying the expected value related to the forecasting value and materialized value, which is expressed as
where
represents a weighting factor, and
,
.
The credibility function shown by (
16), which is defined as the average of possibility and necessity measures [
42], has been shown to satisfy the properties of dual, normality, monotonicity, and nonnegativity [
28,
43]. Taking uncertain wind power for example, based on credibility function, the credibility measure developed by (
17) is applied to evaluate uncertain wind power, considering the random measure and fuzzy measure.
where
A represents a nonempty set, and
denotes the credibility measure associated with uncertain wind power.
Proof. According to (
16), at time
t, for
, we have
If
, we have
and
Based on (
19) and (
20), the credibility measure is
If
, we have
and
Based on (
22) and (
23), the credibility measure is
This completes the proof. □
3.2. Load Partition Method
Define
, and set
,
, and
,
,
represents the initial load demand during time
t. Based on the shortest distance method, the net load curve of WSTHcMEEM considering wind and solar power integration can be partitioned into peak period, flat period, and valley period using (
25)–(
27):
The main steps of load partition method are given as follows:
Step 1: Set . According to the forecasted load, and forecasted outputs of wind and solar power, rank , and we have . Then, obtain .
Step 2: If
, based on (
25) and (
26), we can obtain sets
and
, respectively, i.e.,
,
; otherwise,
.
Step 3: If
, based on (
26) and (
27), we can determine sets
and
, respectively, i.e.,
,
; otherwise,
.
Step 4: . Continue 2 to 4 until .
By (
25)–(
27), the demand sensitivity
E can be written by:
where
is the load demand after implementing DR during time
t.
Therefore, according to
E and the TOU price, the load demand with DR can be written by (
29). Note that the participation percentage of load in DR is considered to be 20%.
4. Simulation Studies
We consider the WSTHcMSSM system with one micro-turbine, one pumped hydro storage with two reservoirs, one wind turbine, and one solar turbine. The entire scheduling period is 24 h, and thus the installed capacities of wind, solar, micro-turbine, and hydro storage are fixed in this paper. In pumped hydro storage, two-chain cascade hydro turbines are considered on one stream to represent the complex hydro network. This subsection describes the main features of the operation characteristics of micro-turbine, pumped hydro storage, wind, and solar turbine under study. The hourly load demand, forecasted wind, and solar power are illustrated by
Figure 2, and the standard deviations of wind and solar power outputs are set to 5% of the forecasted values.
In this paper, the optimization algorithm proposed in [
44] is employed to solve the proposed model. The algorithm is inspired by studying the behavior of group animal searching and living theory. The practicability and feasibility of the algorithm have been proved in optimizing multi-modal engineering problems and benchmarks. For a detailed description of the algorithm, please refer to [
44].
The initial electricity price is 220
$/MW. The maximum outputs of wind and solar unit are, respectively, 1.0 MW and 0.8 MW. The cost coefficients of the micro-turbine are
, and the operation limits of the micro-turbine are set to 0 MW and 0.8 MW. The ramp rate of the thermal unit is set to 0.55 MW. The initial and final reservoir storage volumes of the two reservoirs are, respectively, 1.0
, 1.2
, and 1.2
, 1.4
. The minimum volume values of the two reservoirs are, respectively, 0.8
and 0.7
, and the maximum volume values of the two reservoirs are, respectively, 1.5
and 1.6
. The pumped hydro storage ranges from 0 MW to 0.6 MW, and the lower and upper boundaries of the water release are, respectively, 1
and 8
. The generation coefficients of hydro units are
,
. The water transport delay time from reservoir 1 to reservoir 2 is set to 1 h. Suppose the spillage discharge rate of the reservoir is zero, and the external inflow to the reservoir at each time is given by
Table 1.
Table 2 gives the optimal wind power, solar power, water discharge, TOU price, and load demand with DR over 30 independent runs. From the table, we can find that the optimal outputs meet the operation constraints. In addition, the reservoir volumes of the two reservoirs are illustrated by
Figure 3. From the figure, we can see that the reservoir volumes are within their requirements. Moreover, note that the optimal wind and solar outputs are not their upper bounds. This is because the fact that a large penetration of renewable energy does not correspond to a large credibility measure.
Figure 4 provides the distribution characteristics of credibility measure, operation profit, and wind and solar outputs. The correlation coefficient between credibility and wind and solar power is −0.0852, which means that these two aspects show a negative, but not significant, relationship. Additionally, we find that the correlation coefficient between profit and renewable energy is −0.0074. This phenomenon shows that a large penetration of renewable energy is not always a benefit for WSTHcMSSM considering the risk brought by uncertainty. On the other hand, we can conclude that the proposed CVaC-based quantitative risk evaluation method can make the WSTHcMSSM system a trade-off solution considering the uncertainty of wind and solar power.
The load curves with and without DR and pumped hydro storage are depicted by
Figure 5. In
Figure 5, the load characteristics with DR are much more smooth than these without considering DR. In order to show the load characteristics more intuitively, some factors used in [
13] are employed, and these factors are given by
Table 3. From the table, we can clearly see that DR is benefit for improving load curve. Specifically, compared with the load characteristics without DR, the load factors considering DR increase 3.01% and 5.50%, the peak to valley factors decrease 3.36% and 4.69%, and the peak compensate factors decrease 2.69% and 2.92%. On the other hand, the proposed WSTHcMSSM can explore the potional flexibility in multi-resource complementarity for promoting the penetration of renewable energy.
Moreover, to validate the coupling characteristic of the uncertainty, the hourly wind, solar, and operation profit of WSTHcMSSM are given by
Table 4. Considering the coupling characteristic of multiple sources uncertainty, the penetration levels of wind and solar power have increased 1.58% and 9.33%, respectively. The operation profit considering coupling characteristic has increased 63.6362
$. The comparison illustrates that in WSTHcMSSM scheduling, the coupling of uncertain wind and solar power can mitigate the impact of uncertainty on system operation, and in future research, the virtual aggregation mechanism should be considered in WSTHcMSSM for addressing the uncertain renewable energy.
Additionally, the comparison results between the proposed CVaC model and interval optimization (IO) model [
14,
45] are shown in
Table 5, where
denotes the total outputs of wind and solar power. It can be observed from the obtained results that the proposed model performs better than IO model in terms of the profit and renewable energy penetration, which increases 13.8352% and 6.0152%, respectively. Note that a lower penetration level of renewable energy does not mean a lower level of risk. From the above analysis, it can be concluded that the proposed CVaC is more suitable than IO to provide a reliable guidance for WSTHcMSSM in day-ahead perspective, because it provides a flexible framework in realizing profit–risk trade-off and presents a reliable viewpoint for the dispatcher in choosing the best strategy for the optimal operation of WSTHcMSSM.
Furthermore, the optimal results obtained by CVaC considering different forecasting errors are illustrated in
Figure 6. From the figure, we can find that a larger forecasting error corresponds to a lower penetration level of uncertain renewable energy and a smaller operation profit. This demonstrates that a larger forecasting error brings a higher operation risk, and thus the penetration level of uncertain renewable energy should be limited in WSTHcMSSM scheduling. Therefore, the proposed CVaC-based risk evaluation method can effectively address uncertain renewable energy in WSTHcMSSM scheduling. Specifically, for every 5% increase in forecasting error, the output of wind is reduced by the maximum of 0.73 MW, and the output of solar is reduced by the maximum of 0.24 MW, and the operation profit is reduced by the maximum of 759
$.
In order to further demonstrate the effectiveness of the proposed model for addressing the uncertain renewable energy, the stochastic model using (
9) and the fuzzy model using (
14) or (
15) are employed for comparison. The credibility measures and profit values (
H) obtained by different models are given by
Table 6. From the table, we can see that the proposed model can obtain a high profit with a large credibility in most cases.
The hourly wind and solar power of each model are provided in
Figure 7, and the total wind and solar power integrated with fuzzy model, stochastic model, and the proposed model are, respectively, 23.2275 MW, 17.9877 MW, and 29.4453 MW. The results show that compared with the proposed model, both the fuzzy and stochastic models are conservative in assessing uncertain renewable energy. Additionally, these results also illustrate that a higher wind or solar power output does not always correspond to a higher risk.