Problem Formulation:

A. Standard Chance-Constrained Model

The standard form of chance-constrained optimization problem is represented in (1). The model seeks to minimize
the decision variable, , subject to a set of equality and inequality constraints. The standard model for handling
the uncertainties due to uncertainty vector, 𝜉 , has been reorganized in (2), which is addressing the violations less
than a specific risk level, denoted by term .
𝜉
𝜉 (1)
𝜉) ≥ 0

(2)

Where indicates the risk level to be specified by the decision-maker. In this regard, the probability of the
satisfied scenarios shouldn’t be less than the specified threshold. The model can be solved by deploying Big M
method and converting the original problem into a standard MILP form, as explained in [1]:

∈ (3)

(4)

In this case, (2) has been converted into a bounded constraint with a set of a binary vector, . In the case of
violating the original form, i.e., ≠ , the associated binary term would be ‘1’ to satisfy (3). For any
violations, the total number of violated constraints will be counted and considered in (4). Thus, the term has been
chosen to determine the risk level and should be settled by the decision-maker. This chance-constrained model will
result in violations compared to the original stochastic form. It represents a form that can consider uncertainty
handling appropriately by determining the thresholds by the decision-makers.

B. TSO-DSO Transaction Model

The stochastic framework of the power transactions between TSO and DSO is represented in this section. The
objective function is represented deploying standard stochastic optimization model, as stated in (5). This framework
can be solved using stochastic programming, robust optimization, chance-constrained programming, and robust-
chance-constrained programming approaches with some modifications presented in the case studies simulated in
this paper.

(5)

The objective function covers five different terms. The first term indicates the TSO-DSO transaction costs
considering different trading power and associated hourly costs at the PCC. The second term represents the cost of
PV power production considering the uncertainties of the solar irradiance accepted in this study. The third term
indicates the costs associated with the DGs, including energy production and start-up and shutdown costs. The
degradation costs associated with the ES system, including both charging and discharging actions, have been
addressed in the fourth term. The last term in the objective function denotes the costs associated with load shedding
due to the lack of resources in the loadability analysis. It should be noted that no load shedding is allowed in normal
operating conditions. Thus, the cost of load shedding has been chosen to be sufficiently large to avoid load
curtailment in normal operating conditions.

(6)

(7)
The active and reactive power balance constraints are presented in (6) and (7), respectively. It should be highlighted
that, in this study, the inverted-based DERs, like PV and ES, only provide active power. However, the DGs can
contribute to reactive power provisions. The loadability index has been individually considered for both active and
reactive power terms, i.e., and , respectively. The nodal load shedding terms, and have been
considered to satisfy the load balance equations in the loadability analysis study.

(8)

(9)
(10)

(11)

(12)

(13)

The power transactions between TSO and DSO are subjected to (8)-(13). Equations (8) and (9) indicate the
maximum capacity of power exchange between TSO-DSO, considering associated binary decisions determining
the direction of power exchange, and . Equation (10) guarantees the unidirectional power exchange
at the TSO-DSO connection point. The reactive power exchange is modeled in (11) and (12), indicating that the
reactive power can be easily exchanged between the DSO and TSO regardless of the active power direction. The
active and reactive power exchange between the DSO and TSO is subject to the apparent power capacity at the
connection point, i.e., the power transformer.

(14)

(15)
The branch flow model has been applied for optimal power flow in the distribution network in this study. The model
is represented as a standard MIQCP optimization model. The voltage drop through the feeders connecting nodes
and is modeled in (14), while (15) deals with the power flow through the connecting feeder between nodes and
.
0 ̅ (16)

(17)

(18)

The current, active, and reactive power transmitted from node to are modeled in (16)-(18), respectively. It
should be highlighted that the current term is represented as a quadratic term. Hence, there is no need to define a
supplementary decision variable.

(19)

In a similar way, the nodal voltage limits should be between minimum and maximum acceptable range. This term
has also been represented by a quadratic term, satisfying the original limits for the voltage range, as stated in (19).

(20)

(21)
 (22)

The power generation limits for DG units are provided in (20)-(22). The apparent power limit is denoted in (20),
considering the binary decision variable, . In the case of power generation, , the corresponding
limits should be met. The active and reactive power generation limits are addressed in (21) and (22), respectively.

(23)
(24)
Considering the binary decision variable, the transitions from 0 to 1 considered as start-up and the transitions from
1 to 0 denoted as shut-down. The start-up and shut-down terms are modeled by binary decision variables,
and , respectively. The relationship between these two sets of decision variables and the status of DG unit,
, are provided in (23) and (24).

(25)

(26)

The ramp rates for increasing and decreasing power generations have been modeled in (25) and (26), respectively.
It should be highlighted that the ramp-up and ramp-down parameters are technical limits of the DG units.
  (27)

(28)

(29)

(30)

(31)

(32)

(33)

The ES constraints are modeled in (27)-(33). The dynamic behavior of the ES is subjected to the previously stored
energy, charging, and discharging power considering the efficiency of the ES, as stated in (27). Equation (28)
addresses the minimum and maximum level of energy at the ES system. The initial and final energy stored in the
ES are determined by (29) and (30), respectively. The charging and discharging power limits are provided in (31)-
(33), respectively. It should be noted that the and are the associated binary variables for charging
and discharging operation modes, respectively.
≤ (34)
≤ (35)

As it stated in the objective function and load balance equations, the load shedding terms have been considered to
model the violations in the loadability analysis. In this regard, two different binary decision variables have been
considered for active and reactive power. The amount of load curtailment is subject to the hourly active and
reactive loads, as stated in (34) and (35), respectively.

(36)

(37)

In the chance-constrained model, the number of scenarios with violations should be controlled by the
corresponding term, and , respectively. In this case, the decision-maker should set the risk measurement
index and the corresponding threshold for specific cases with chance-constrained and robust-chance-constrained
case studies.
Simulation Results

This section provides details for the simulation results obtained by deploying the models for different energy
transactions between TSO and DSO. The simulation studies conducted on the modified IEEE 33-bus test system,
as demonstrated in Figure 1. This network has one main feeder and three laterals and is fed by one power
transformer at the connection node. The standard topology of the network has been chosen and there is no
possibility for reconfiguring the original network. There are three energy storage units, two PV panels and two
DGs considered in this study as the DERs to support the network and leverage the self-sufficiency of the
distribution network.

EES

23 24 25

26 27 28 29 30 31 32 33

DG2

EES EES
TSO DSO
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Power Grid
PV1 DG1
19 20 21 22

PV2

Figure 1. Single line diagram of the modified IEEE 33-node test system

Two different sets of uncertainties have been considered in this paper, including the demand and PV power
generation. To demonstrate the performance of the proposed model, eight case studies have been introduced and
their features are provided in Table 1. In addition, Table 2 address the most important parameters of the mentioned
case studies in the simulation phase. The first three case studies deal with the deterministic version of the model
and the rest consider different uncertainties and different strategies for handling the uncertainties that have been
introduced. Cases 4 and 5 deal with the standard stochastic programming model and consider the expected value
for the objective function.

Table 1. Specifications of different case studies

Case Study Det. Stoch. PV DG EES Robust Chance

Case1 ⦿ ⦾ ⦾ ⦾ ⦾ ⦾ ⦾
Case2 ⦿ ⦾ ⦾ ⦾ ⦿ ⦾ ⦾
Case3 ⦿ ⦾ ⦾ ⦿ ⦿ ⦾ ⦾
Case4 ⦾ ⦿ ⦾ ⦿ ⦿ ⦾ ⦾
Case5 ⦾ ⦿ ⦿ ⦿ ⦿ ⦾ ⦾
Case6 ⦾ ⦿ ⦿ ⦿ ⦿ ⦿ ⦾
Case7 ⦾ ⦿ ⦿ ⦿ ⦿ ⦾ ⦿
Case8 ⦾ ⦿ ⦿ ⦿ ⦿ ⦿ ⦿

According to Table 2, | , means the number of scenarios that have been considered as the decision variables.
For instance, in the deterministic cases, i.e., Case 1 to Case 3, only one scenario has been considered, while in
Case 4 to Case 8, the number of scenarios is 10. In addition, deals with the possibility of injecting power
from DSO to TSO. In this regard, there is no power injection allowed from DSO to TSO in Case 1 and Case 2.
The loadability index for active and reactive powers, and , have been considered in Case 1 to Case 6, while
only active power loadability has been studied in Case 7 and Case 8. Another important part of the simulations is
| which introduces the number of decision variables in the robust optimization approach. This parameter
shows that for the deterministic case studies, i.e., Case 1 to Case 3, the power transactions from TSO to DSO has
only one dimension. In the Case 6 and Case 8, the model has been solved using robust optimization concept and
the power transactions from TSO to DSO is subjected to only one dimension. It means that the DSO is buying the
electricity from the grid as a predefined term in the day-ahead market and the DSO is responsible for handling the
inherent uncertainties internally with the local resources. The last two case studies, i.e., Case 7 and Case 8 deal with
the chance-constrained optimization, hence, , has been considered in these two case studies.

Table 2. Case’s studies parameters and features

Case Study | |

Case1 =1 =0 >0 >0 =1 =0

Case2 =1 =0 >0 >0 =1 =0
Case3 =1 >0 >0 >0 =1 =0
Case4 =10 >0 >0 >0 =10 =0
Case5 =10 >0 >0 >0 =10 =0
Case6 =10 >0 >0 >0 =1 =0
Case7 =10 >0 >0 =0 =10 >0
Case8 =10 >0 >0 =0 =1 >0

Case 1:
This case study deals with the analysis of the network deploying deterministic model for operational planning
assessment. In this case, there are no local DERs, and the model has been solved as a normal optimization problem.
The base case results for a demand of 3715 kW and 2300 kVAr at peak hour show an import of 3917.12 kW and
2434.617 kVAr from the TSO. In Case 1, it is important to highlight that no local power generation or energy
storage is available. Hence, the TSO is the only source for power provision. The power loss at peak hour is 202.12
kW, demonstrating the effectiveness of the MIQCP model in solving the OPF problem for the standard network
[2]. With the transformer at the TSO-DSO connection rated at 5000 kVA and maximum active and reactive power
of 4250 kW and 2634 kVAr, respectively, the transformer operates at 92% capacity under normal conditions. The
overall daily operational cost in the base scenario for Case 1 is $877.29. In the maximum loadability study for
Case 1, the loadability index for both active and reactive power is 0.076, resulting in a total loading of 4234.17
kW and 2632.56 kVAr, or 4985.84 kVA for the transformer. In this scenario, the new load reaches 3997.34 kW,
with peak power loss increasing to 236.83 kW. The reactive power provision capability constraint has been
activated, resulting in 99.95% loading of the transformer. The daily operational cost for this scenario rises to
$946.88. Therefore, a 7.6% increase in load results in a 7.93% increase in daily operational cost and a 17.17% rise
in peak power loss. The voltage profile for base and maximum loadability scenarios are illustrated in Figure 2. The
simulation result shows that the minimum voltage at node 18 reaches 0.913 and 0.905 pu in base and maximum
loadability scenarios, respectively.
 Figure 2. Voltage profile at peak hour in Case 1 for base and maximum loadability scenarios

Case 2:
This case study discusses the functionality of the energy storage system to enhance the flexibility of the distribution
network utilization considering the capability of the load shifting. Table 3 provides the technical parameters of
the energy storage units. In this case, the energy storage units’ contribution is to reduce the peak power
consumption and shifting the loads into off peak periods and providing more capacity for increasing the loadability
of the network. The maximum loadability index for active and reactive power is 8.2% which is higher than the
Case 1. Figure 3 provides the loadability index comparison for Case 1 and Case 2 in the presence of energy storage
units. As mentioned before, the DERs in this study only provide the active power provision, hence, the loadability
margin for reactive power provision remained constant. Considering the 8.2% loading factor, the new peak power
would be 4019.63 kW and by deploying the contribution of the energy storage units, the peak power will be
reduced to 3769.849 kW and operating more space for increasing the loading capacity of the network.

Table 3. Energy storage unit’s operating parameters

Node P_CH P_DCH Emax Emin Et_Ch Et_DCH E0

N08 150 150 1500 150 0.95 0.95 750
N15 50 50 500 50 0.95 0.95 250
N24 50 50 500 50 0.95 0.95 250

Figure 3. Loadability curves for Case 1 and Case 2

The hourly demand considering the contribution of the energy storage units are demonstrated in Figure 4. This
figure shows the percentage of the energy storage units in changing the load profile. The results show that the load
factor, as a key performance index, has been enhanced from 58.54% to 62.67%. It should be noted that the load
factor is defined as the average load divided by the peak load in a specified period, which is 24 hours in this case.
It should also be highlighted that the total energy consumption in the case without energy storage is 56.5 MWh
and by considering the energy storage units, the energy consumption will be increased to 56.7 MWh due to the
efficiency of the storage units in charging and discharging modes. Figure 5 illustrates the state of charge (SoC) of
the energy storage units in Case 2. The active and reactive power injections to the distribution grid are 3985.309
kW and 2631.831 kVAr, respectively. As expected, the reactive power limit is a barrier for loadability analysis in
this case. In this case, the reactive power almost reaches its maximum level, i.e., 99.92% of its rated reactive
power provision capability. It should be noted that the daily operation costs for base scenario and maximum
loadability are 857.809$ and 932.67$, respectively. This achievement has been obtained with the contribution of
the energy storage systems in this case study.

Figure 4. Hourly load profiles with and without energy storage units

Figure 5. State of charge of the energy storage units in Case 2

Case 3:
This case study deals with the implementation of DG units and leveraging the distribution network into an active
distribution network. This case study is running in a deterministic way and there is no uncertainty to be handled.
The DERs in this case include energy storage and DG units. The techno-economic features of the DG units are
provided in Table 4. The loadability of the active distribution network in the presence of DERs in Case 3 is
significantly increased. In this specific case, the maximum loadability of the network for increasing the active
power is 0.9 and the same index for reactive power is 1.0. The maximum loadability for both active and reactive
power increasing with the same ratio is 0.88. The daily operational cost for the base scenario in this case is
578.275$ while the operational cost for the maximum loadability scenario is 1318.031$. The power injection by
the external grid is demonstrated in Figure 6 . It should be noted that the cost of energy delivered by the DG units
is less than the external grid. The power transactions between TSO and DSO, considering the hourly energy tariff
adopted in this study show that the DSO is delivering power at off-peak hours, as much as it is possible. This
feature shows the self-sufficiency of the active distribution network and its capabilities to provide flexibility for
the TSO. In addition, this feature supports the DSO to leverage the limits for the loadability of the distribution
network for the given deterministic load profile.
In addition, the minimum voltage occurs at node 33, which is located at the end of the feeder and there is no DER
installed to support the voltage profile at this node. For the sake of comparison, the voltage at node 18 is also
depicted in Figure 7. The simulation results confirm that the voltage profile at node 18 has been considerably
enhanced due to the resourcefulness of the DERs and their contributions in voltage support.

Table 4. Techno-economic features of DG units

Node (kW) (kW) (kVAr) (kVA) (kW) (kW) ($/kWh) ($)
N14 250 2000 1500 2500 150 150 0.006 150
N27 100 800 600 1000 100 100 0.008 100

Figure 6. Power injection by external grid in Case 3

 Figure 7. Voltage profiles at nodes 18 and 33 for different loadability indexes in Case 3

Case 4:
This case study focuses on the stochastic optimization considering the load uncertainty. The model has only one
uncertainty in this case. Thus, the model has been upgraded to handle the load uncertainty using stochastic
programming approach. In this case, the expected cost of energy trading has been reported. Figure 8 provides the
chronological and boxplots for the uncertain demand in Case 4. The number of scenarios in this case is reduced to
10. The expected cost for the normal loading of the network is 607.622$. The maximum loadability index is 0.84,
which is lower than the deterministic scenario studied in Case 3. The expected cost in this scenario is 1345.373$
which is slightly higher than the deterministic scenario. The power transactions for the studied scenarios under
different loadings are demonstrated in Figure 9. In the normal loading, i.e., , the maximum power
injection by TSO is less than 1000 kW, and it confirms the self-sufficiency of the active distribution network under
normal operating conditions. In addition, in the maximum loading scenario, with , the peak
power will be 6835.6 kW, while the active distribution network will receive less than 4500 kW at the peak period.
The voltage profile for these two different loadability scenarios are illustrated in Figure 10. In normal loading, the
voltage profile is smooth and there is no violation from the reference voltage. However, in the maximum loading
scenario, during the peak hours, the minimum voltage is very close to the lower limit for the voltage, i.e., 0.9 pu.

Figure 8. Chronological and Boxplot of forecasted loads in Cases 4-8

 Figure 9. Power transactions between DSO and TSO in Case 4

Figure 10. Voltage profile for different loadability indexes in Case 4

Case 5:
In this case study, the second set of uncertainty has been added to the model, i.e., the power production by PV
panels. The PV panels are installed at nodes 8 and 22 and their rated capacity are 1200 kW and 1500 kW,
respectively. The PV production forecasts for the local power generation are illustrated in Figure 11. It should be
noted that the normalized scenarios are provided in this illustration and by considering the rated capacity of the
PV panels, the actual power production can be calculated accordingly. In the presence of local DERs and DGs,
the resourcefulness of the active distribution network is enhanced. Hence, comparing to the Case 4, the reactive
power provision capability is enhanced in this scenario. Figure 12 illustrates the loadability curves for Case 3,
Case 4, and Case 5. According to the simulation results, the reactive power provision capability has been improved
with the support of PV panels, as it opens more capabilities from local DGs and external grid. It should be
highlighted that the loadability index in this case is 0.87 and the expected cost is 1044.04$. The expected cost for
the normal loading condition is 313.668$ which is almost half of the expected cost achieved in Case 4 for the
same condition. The power transactions between TSO and DSO are depicted in Figure 13. As expected, the local
power generation and load management with the local resources has been effectively handled in the normal
loading conditions. The DSO is subject to import the power from the external grid during the peak hours, i.e.,
between 18-21h. However, the self-sufficiency in the maximum loadability case is very limited.
Figure 11. Chronological and Boxplot of forecasted PV power in Cases 5-8

Figure 12. Loadability curves in Cases 3-5

Figure 13. Power transactions between DSO and TSO in Case 5

The voltage profile, for Case 5 is depicted in Figure 14. The simulation results show that the voltage profile in the
normal loading condition is well supported by local power generation. In the maximum loading condition, the
minimum voltage met in the peak hours, and it is one of the barriers in increasing the loading capability of the
distribution network. The behaviors of the energy storage for the given 24 hours are demonstrated in Figure 15.
The state of charge of the set of batteries shows that the trend of charging and discharging are almost the same
with some fluctuations during the period with PV productions.

Figure 14. Voltage profile for different loadability indexes in Case 5

Figure 15. State of charge of the energy storages for different loadability indexes in Case 5

Case 6:
This case study provides a robust solution for the stochastic model addressed in Case 5. In this case study, the
DSO should buy the energy from the TSO as a firm and any violations from the contracted firm will be penalized.
In this case, the DSO will manage the fluctuations internally with all dispatchable resources. The computational
burden in this case is much higher than the previous case studied, as the decision variables associated with TSO
to DSO is limited to a deterministic firm. In the normal loading scenario, the operational cost is 318.451$. The
maximum loading index in this case is 0.87 resulted in the operational cost of 1204.908$. The power transactions
between TSO and DSO are demonstrated in Figure 16. According to the observations, the firm power injection by
the TSO is limited to four consecutive hours during the peak period, i.e., 18-21h for the normal loading condition.
For the rest, the DSO will sell power to the grid, and it depends on the power production capability and hourly
demand. In addition, in the maximum loadability scenario, the external grid injects power to the DSO and there
is no power injection from the DSO to TSO. It is mandatory for the DSO to decide how much power should be
imported and the DSO is responsible for handling the uncertainties internally. The state of charges of the batteries
are illustrated in Figure 17. As seen, the energy storage behavior for the maximum loadability scenario is
completely different from the normal loading conditions.

Figure 16. Power transactions between DSO and TSO in Case 6

Figure 17. State of charge of the energy storages for different loadability indexes in Case 6

The simulation results for case stochastic and robust optimization presented in Case 5 and Case 6 show that for
the maximum loadability index for both active and reactive power loadability, the overall costs for robust
optimization are higher than the stochastic, as it is expected. It should be noted that the maximum loadability for
active power is 90% of the normal load and this index for reactive power is 95%, identical for both stochastic and
robust optimization. This comparative analysis is depicted in Figure 18. It should be noted that this comparison is
different from the simultaneous loadability for both active and reactive powers. As indicated, the loadability for
both active and reactive power is 0.87.
 Figure 18. Comparative analysis for individual loadability index in Case 5 and Case 6

Case 7:

Case 7 deals with the assessment of the chance-constrained model for handling uncertainties. In this regard,
different risk measurement indexes have been considered. For instance, the model is solved for different values of
𝜎 to represent the probability of 80%, 85%, 90%, and 95% to represent the probability of the violation of the load
balance equality constraints. In this case, only maximum loadability index for active power has been investigated.
According to the number of scenarios and operational planning horizon, the number of specific scenarios is 240.
Thus, for the probability equal to 95%, the number of violation cases would be 𝜎 . Table 5 presents the report
for the loadability assessment under different conditions. It should be highlighted that the 𝜎 and 𝜎 address the
number of violated scenarios for each specific risk measurement studied. The simulation results confirm that by
increasing the risk level, the loading capability will be increased. For example, in the case with probability of
80%, the loading capability is 1.34 which is much higher than other risk indexes. This means that the DSO has
this opportunity to increase the loading capability of the network at a reasonable cost. According to the results,
the active power balance equation has been activated in most of the scenarios, reaching the maximum allowed
violations, except in the case with the risk index equal to 90%, in which 23 violations have been recorded. The
reactive power balance has not been activated in these scenarios. Figure 19 illustrates the power exchange and
state of charge of the batteries for normal and maximum loading conditions under risk level of 80%. According
to this figure, the power injection from the TSO reaches its maximum, i.e., 4250 kW, during the peak period. On
the other side, the state of charge of the batteries shows the fluctuations that have been handled with the support
of the energy systems under sever loading conditions and contribute to hedging the risk associated with internal
uncertainties.

Table 5. Loadability index for active power under different risk levels in Case 7

Probability 𝜎 𝜎 𝜎 𝜎 Cost ($)

95.00% 12 12 12 4 1.10 1223.540
90.00% 24 23 24 0 1.19 1293.108
85.00% 36 36 36 4 1.30 1371.331
80.00% 48 48 48 4 1.34 1398.573
 Figure 19. Power exchange and SoC of the batteries for normal and maximum loading under risk level of 80%

Case 8:
This case study provides the robust-chance-constrained optimization for the given test system. As mentioned
before, in the robust optimization approach, the firm injection power from the TSO is planned. In this case, the
DSO will request a firm from the TSO and the uncertainty handling will be organized by the local resources. In
addition, the chance-constrained model provides the flexibility for handling the fluctuation by considering a
certain number of violations for each planning scenario. In this case, the injected power from TSO supports only
one dimension, while the local resources have more freedom, i.e., 10 scenarios in this case study. For each risk
level, the associated measurements have been conducted and the results are reported in Table 6. The simulation
results for normal and robust chance constrained optimization model are provided in this section. The results show
that the overall operational cost in the robust optimization is higher than normal chance constrained model. In
addition, the loadability index for robust chance constrained model is saturated for less than 85%. It means that
the system couldn’t support more flexibility while the system allows more violations.

Table 6. Loadability index for active power under different risk levels in Case 8

Probability 𝜎 𝜎 𝜎 𝜎 Cost ($)

95.00% 12 12 12 0 1.08 1386.290
90.00% 24 24 24 0 1.17 1462.998
85.00% 36 36 36 0 1.29 1555.690
80.00% 48 48 48 0 1.29 1555.061

Moreover, the firm injected power from TSO for given scenarios have been illustrated in Figure 20. The simulation
results confirm that the firm injected power the external grid is completely expected to the main principles of the
robust-chance-constrained approach. In this case, the injected power reaches the maximum rated capacity of the
TSO-DSO transformer, i.e., 4250 kW during the peak period and thus, there is no possibility to import power from
the external grid. In addition, the energy storage units play a significant role in handling the load-generation
unbalances at the local network. The state of charge of the batteries are demonstrated in Figure 21.
 Figure 20. Firm power injection from TSO for different risk indexes

Figure 21. State of charge of the batteries for different risk indexes

[1] A. Esmaeel Nezhad, P. H. J. Nardelli, S. Sahoo, and F. Ghanavati, “Scheduling of Energy Hub Resources
Using Robust Chance-Constrained Optimization,” IEEE Access, vol. 10, pp. 129738–129753, 2022, doi:
10.1109/ACCESS.2022.3228388.
[2] M. S. Javadi, C. S. Gouveia, L. M. Carvalho, and R. Silva, “Optimal Power Flow Solution for Distribution
Networks using Quadratically Constrained Programming and McCormick Relaxation Technique,” 2021
IEEE International Conference on Environment and Electrical Engineering and 2021 IEEE Industrial
and Commercial Power Systems Europe (EEEIC / I&CPS Europe), pp. 1–6, Sep. 2021, doi:
10.1109/EEEIC/ICPSEUROPE51590.2021.9584627.