5.1. Centralized Control
For MMG operation and control, an MG shares the information of available power generation capacity (dispatchable and non-dispatchable), power demand, the SOC of ESSs, the types of loads (controllable, uncontrollable, elastic, and inelastic), and the load priorities with a central entity that collects and stores this information in a central database through designated communication channels. In the centralized control structure (as shown in
Figure 2), the central controller executes the operation management algorithms and the control commands are communicated to each MG. One of the main advantages of a centralized controller is that, since a single entity is responsible for the control decisions, it is straightforward to implement the coordination between different MGs. However, this might increase the risk of a single point of failure for the whole MMG system [
28]. Moreover, these types of controllers generally require more communication and computational resources, which restricts their scalability [
3,
28].
Minimizing the operational cost and maximizing the global benefits are the two main and primary objectives of a centralized EMS in MMGs. The authors of [
29,
31,
33,
35,
36,
38,
39,
84,
106,
117] mainly focus on the economic operation of a MMG system. A MPC-based control was implemented in [
29] to maximize the global benefits of the MMG system, taking into account the uncertainties in RESs’ power generation and load demand. Their proposed algorithm schedules the power exchange between the MGs and distributed network operator (DNO). In [
106], cooperation among two MGs is encouraged by reducing the energy drawn from the main grid as the controller has the information of the energy generation, storage, and load of both MGs and can schedule the sharing of available resources. The offline optimization Lagrange duality method is implemented considering the transmission losses and the distance between MGs. A similar study is presented in [
31], where a three-stage multi-agent system (MAS)-based coalition game strategy is applied to form coalitions among MGs for cooperative and direct power exchange to achieve economical operation of MMGs. In [
33], cooperative game theory is deployed to increase the energy efficiency of the MMGs while considering the interaction with distributed network (DN). In [
117], the MGs with higher RES and ESS capacity are given higher priority and incentive for engaging in MMG energy exchange. A collaborative energy exchange scheduling scheme for residential MGs is discussed in [
35] to minimize the operating cost of the MMG system using a method named column-and-constraint generation (C&CG) and to mitigate the disturbance caused by solar power uncertainty. The adaptive robust optimization (ARO) technique with a budget parameter is used in this study. Considering the intermittent nature of RESs, the EMS in [
36] tries to minimize the operation cost by forming an MG coalition. In [
36], a centralized macro station searches for the pair of MGs with lower power losses until demands of all MGs are met. To mitigate the uncertainty of RESs, reserve support among MGs is implemented in [
38]. A real-time cooperative EMS is established in [
39] namely “store-then-cooperate” and “cooperate-then-store”, and it is shown that the performance is close to the optimal offline solution. In the store-then-cooperate strategy, the MGs charge the ESSs when there is a surplus generation and then transmit energy to other MGs and vice versa for the cooperate-then-store approach [
39]. With increasing penetration of EVs, they are now considered as mobile ESSs. A cooperative operation of RESs and EVs is achieved in [
84] to reduce the cost of operation of MMGs using the particle swarm optimization (PSO) algorithm. It is shown that the participation and cooperation of EVs in the operation of MMGs reduce the need for the installation of more RESs [
84]. Finally, wild goats algorithm (WGA) is used in [
41] for the optimal operation of an MMG system to maximize the profits of MG owners. Along with the economical operation of the MMG system, the main objective of [
30] is the implementation of optimal power dispatch considering the probability distribution function (PDF) of the energy generation, consumption, power purchasing/selling of the MGs, operation and maintenance cost, and the cost of power transaction using the PSO algorithm. Similarly, the authors of [
86] aim at minimizing load shedding when the MG cluster is operating in the islanded mode. A mixed integer linear programming (MILP)-based robust optimization (RO) technique is developed for the scheduling of MMG’s operation considering the uncertainties in RESs power and load forecasts. In [
85], to minimize the load curtailment, the optimal scheduling of energy resources is implemented in both grid-connected and islanded modes of operation. The resulting large-scale optimization problem is decomposed into multiple sub-problems using the alternating direction methods of multipliers (ADMM) decomposition method and solved by parallel computing techniques. Information is exchanged between the master and distributed computing nodes to find an optimal solution. In [
32], an algorithm based on second-order cone programming (SOCP), point estimation method (PEM), and MPC is proposed to maximize the robustness of power exchange between MMGs and the DN taking into account the constraints and spinning reserve of the DN. A risk-constrained energy management strategy is proposed in [
37] considering the variation of the MG’s expected profit and the arising volatility in power exchange. Profit variation occurs due to the fluctuation in RESs power, resulting in energy exchange [
37]. The voltage, frequency, and output power of MGs’ converters will fluctuate in the case of the transition from grid-connected to islanded mode. MPC is used in [
90] to regulate the MG’s voltage, frequency, and power for safe islanding of the MG from the MMG system. A deep neural network (DNN) and model-free reinforcement learning (RL)-based technique are applied for intelligent MMG energy management in [
40]. Given the retail price, the DNN estimates the aggregated MMG power exchange with the main distribution grid. The RL-based technique at the DSO-level then optimizes the retail pricing to maximize the profit of selling power and reduce the load peak-to-average (PAR) of the distribution system [
40].
To increase the reliability of the MMG system, dynamic optimal planning is presented in [
119]. The proposed optimal planning framework minimizes the operating costs and the energy not supplied in both grid-connected and islanded mode considering the optimal size, site, type, and uncertainties of RESs. The DNO applies the multi-objective PSO algorithm to minimize the cost functions, and the best solution is selected based on three different risk attitudes, namely risk averse, risk neutral, and risk seeker. The authors of [
34] analyze the MMG reliability performance and outage management scheme by DSO. A reliability assessment framework is developed to divide the electricity grid into smaller MGs based on the location of protection devices. Different sections are evaluated using the reliability indices. A summary of the reviewed studies on centralized control and EMS of MMGs can be found in
Table A2.
5.3. Distributed Control
An intermediate of centralized and decentralized control is distributed control (as shown in
Figure 4), where there is communication between multiple controllers, which cooperate to achieve a common goal [
126]. These types of controllers feature a high degree of flexibility to support the integration of new MGs, which facilitates plug-and-play operation [
4,
126]. In [
91], a distributed control scheme is proposed to regulate the power exchange and the average voltage among MGs. The controllers are interconnected through low-bandwidth communication links, controlling the current of DERs. In [
92], a robust distributed control scheme is implemented to mitigate the poorly damped oscillations through proper regulation of ESSs and DGs among interconnected MGs. In [
100], active power sharing among interconnected MGs is studied while accounting for communication delay and communication loss in the interlinking converters. The converter control agents request information from neighboring controllers, and using an event-based control scheme and consensus protocol among multiple agents, the active power load of individual MGs is adjusted. Furthermore, a distributed algorithm is designed in [
48] for energy exchange with neighboring MGs to minimize the transmission power losses.
In [
51], the operation cost of DNO and MGs is minimized using a bi-level stochastic optimization algorithm. The algorithm uses a deterministic decomposition methodology, and the bi-level problem is solved using progressive hedging. ADMM is a widely used method for distributed operation management and control. In [
55], ADMM is used by the EMS to minimize the overall cost of the MMG system. The proposed algorithm reduces the communication overhead since only direct neighbors communicate with each other. In [
93], the optimization of operational cost is performed using ADMM in an AC–DC hybrid MMG system, where a DC network interconnects the MGs. Similarly, the authors of [
56] consider distribution system (DS) and MGs as distinct entities with the objective of minimizing operating costs. Power exchange among the DS and MGs is determined using ADMM considering the uncertainties in RESs production and load. In [
61], DNO performs economic energy management using the coordinated distributed model predictive control (MPC) (DMPC) technique and provides references to the MGs. An algorithm derived from ADMM named distributed adjustable robust optimal scheduling algorithm (DAROSA) is proposed in [
105] to optimize the MGs operating cost by trading energy among MGs and the main grid. The ARO deals with the uncertainties of RESs production, MGs’ load, and the buying/selling electricity prices of the main grid. In [
64], considering the existence of multiple MMG stakeholders, an analytical target cascading (ATC)-based optimization framework is developed to improve the operational performance of the MMG system and benefits of all stakeholders. The diagonal quadratic approximation (DQA) technique is employed for implementing the parallel operation. Using ATC and DQA, energy trading and energy dispatch are implemented considering the uncertainty in RESs and loads. Cooperative energy trading among MGs is discussed in [
47] using the ATC algorithm, which shows performance improvement compared to the independent operation. In [
46], the performance of integrated ATC and C&CG is compared with the C&CG performance for economic dispatch and coordinated operation of MMGs. In [
45], ATC is used to decouple the DN and MGs power dispatch so that the MGs and DNs can use their resources independently to achieve economical operation, which is called dynamic economic dispatch (DED). According to [
52], optimal scheduling of MMG systems using ADMM reduces the coalition operation cost of the MMG system while sharing only the information of the expected power to be exchanged, which preserves the MGs’ privacy.
In addition to the operating cost of MGs, energy exchange with the DS is reduced in [
53] to encourage a higher contribution of RESs. MPC and MILP are employed for energy management of the MMG system. The MPC technique accounts for the future behavior of the system in a receding horizon manner, making the solution strategy more robust against uncertainty [
53]. In [
94], DMPC is used to optimize RES utilization and coordinate renewable energy sharing among MGs. The EMS improves the RES utilization while reducing the deep discharge of the battery and the customer dissatisfaction. The authors of [
62] develop an optimal strategy for internal resource scheduling and energy trading of MGs using DMPC and a dimensionally distributed algorithm based on PSO. To benefit from the diversity that exists in the RESs production and load patterns of different MGs in an MMG system, a bargaining-based incentive mechanism is developed in [
54] for energy scheduling and trading among multiple MGs. Considering the presence of multiple MG owners and stakeholders with different attitudes in an MMG system, distributing the economical benefits of cooperative operation in a fair manner is a challenging issue. In this regard, Nash bargain theory is used in [
54] to encourage energy trading and having a fair benefit sharing among MGs. An online ADMM-based energy management strategy for the DNO is proposed in [
57] to schedule the power exchange of interconnected MGs. The objective is to minimize the power production cost of MGs, and the prediction of the uncertain RESs production and power demand is not required. In [
87], a decentralized–distributed adaptive RO (DD-ARO) methodology is proposed to address the distributed scheduling of AC–DC hybrid MMGs in case of accidental communication or source–load power line failures. Active and reactive power sharing among MGs in MMGs are discussed in [
100] considering the impact of communication delay and the loss of the controllers using MAS and event-based distributed consensus control. The proposed DD-ARO method consists of two optimization problems, which include a MILP-based minimization problem and a tri-level min–max–min problem solved using C&CG and the Benders decomposition algorithms. Besides, the privacy concerns of multiple stakeholders are taken into consideration for economic operation and uncertainty management.
Several studies design the voltage and frequency control of MMGs using distributed control. In [
49], voltage and frequency control and restoration techniques are designed using droop control for active power sharing. The voltage is restored to the desired value for all DGs regardless of frequency using finite-time voltage control, and a distributed consensus-based control restores the frequency to the desired value. In [
50], a proportional–integral droop control with distributed averaging algorithms, called the distributed-averaging proportional–integral (DAPI) method, is designed for voltage–frequency regulation. The neighboring DGs communicate with each other and share information for voltage–frequency regulation and active–reactive power sharing. The authors of [
58] propose an adaptive voltage and frequency control for DGs using distributed cooperative control and an adaptive neural network. The proposed control technique simplifies multiple layers of control and achieves good active–reactive power sharing with reduced dependency on system dynamics. In [
59], in case the voltage of one MG moves outside the limits, the MG controller tackles the voltage problem by instantly communicating with the neighboring MGs in the MMG system. MPC is employed for coordination and distributed sharing of resources considering the different load profiles of industrial, commercial, and residential MGs. Similarly, the authors of [
127] focus on frequency regulation while maintaining the voltage in all buses of the MMG system using DMPC. Using a consensus-based ADMM, the neighboring MGs operate in a limited communication framework to ensure privacy and scalability.
Along with voltage control, the authors of [
91] discuss the load sharing and power management of MMGs. A droop-based control scheme along with a combination of global positioning system (GPS) timing technology and non-linear voltage–current characteristics for fast dynamic response of load sharing is designed. The voltage and frequency control is performed using a distributed averaging technique according to the average voltage of the network, for regulating the tie-line current and improving the voltage profile. In [
101], a real-time control method for power sharing among MGs consisting of multiple DERs and loads is proposed using a novel distributed optimal tie-line power flow (DOTLPF) control technique. In [
65], a unified small-signal dynamic model of the MMG system facilitates load sharing among MGs and the plug-and-play operation. Droop control is implemented for the output power regulation of the MGs, and distributed control is used for the power sharing of DGs in each MG. In [
92], a robust distributed control scheme is proposed to mitigate the power oscillations caused by dynamic interactions among MGs in an MMG system. In [
99], a distributed real-time optimal power flow (RTOPF) is proposed, where the MGs are modeled and controlled using the MASs approach. In [
103], an MAS-based distributed secondary control is implemented for active–reactive power and droop control of DGs in MMG clusters. In the event of switching to the islanded mode, the voltage and frequency change due to the change in the power control references [
104]. In [
104], this problem is addressed by implementing differential game theory (DGT) for solving the multi-agent non-cooperative distributed coordination problem.
Control strategies can also ensure the resilience-oriented operation of MMGs. The authors of [
60] define a resilience index based on the ability of MGs to supply critical loads at the time of power interruption. A consensus algorithm in a distributed framework establishes the optimal load shedding of MGs. The proposed fault-tolerant control of MGs in [
63] maintains the power balance within and between the MGs to increase the reliability of the MMG system. The control system maintains the voltage and frequency of MGs within the permissible range in the case of the generation and load fluctuations. A summary of the reviewed studies on distributed control and EMS of MMGs is presented in
Table A3.
5.4. Hierarchical Control
The hierarchical control (as shown in
Figure 5) allows organizing the MMG control at several control levels. Communication is established among the controllers of different levels to share the control signals or information. The controllers may have different time scales in the order of minutes or hours. Generally, in hierarchical control, each control level is associated with the objective(s) of the MMG system. Furthermore, each control level can be divided into several sub-layers, where upper-level controllers can schedule references for the lower-level controllers. For instance, the EMS in an MMG system might have two levels. The upper-level controller could be responsible for optimizing the energy exchange of the entire MMG system, demand management, market prices, and so on, while the lower-level controllers might be responsible for controlling the power production and demand at the MG level, maintaining power balance, minimizing the MG operational cost, etc., following the references provided by the upper-level controllers. In [
66], the power scheduling and power trading are split into two levels, where the first tier accounts for the user utility function and grid load variance and the second tier controls the power generation and energy storage in the MGs.
In [
128], the hierarchical approach is adopted in an AC–DC hybrid MMG system for power exchange among MGs, maintaining the normal operation of the MMG system. A hierarchical EMS is implemented in [
95] to achieve power balance within each MG at the MG level and power exchange coordination between the DN and MGs at the MMG level using RO and MILP. The authors of [
69] design a two-level EMS to minimize the operating cost of the MMG system using MILP. A central autonomous controller executes the control signals from the DSO for MGs operation management and control, while the MG central controller (MGCC) controls the micro-sources, controllable loads, and ESSs of individual MGs. Similarly, distributed economic MPC (EMPC) is implemented in [
74] for economical operation of the whole DS. At the upper layer, the DNO maintains the supply–demand balance in the MMG system, schedules power exchanges, and provides references to the lower-level MGs controllers. At the lower layer, MG controllers maintain the power balance and minimize the operation cost of the MG, while tracking the references provided by DNO using DMPC and EMPC. A stochastic EMS in the chance constrained MPC (CCMPC) framework is designed in [
88] for the operation management of MMGs. After coordinating the operation of MGs and scheduling power exchange among MGs and between the MGs and the main grid at the upper level, the decisions are communicated to the lower-level EMSs. Taking into account the several sources of uncertainties, CCMPC ensures the satisfactory operation of each MG at the lower level. In [
76], to minimize the operation cost of the MMG system, the EMS is also informed about the MGs adjustable power in addition to their power shortage and surplus information. In [
70], a bi-level optimization based on multi-period imperialist competition algorithm (ICA) is implemented for economic and optimal operation, demand-side management, and scheduling the interactions among MGs in an MMG system. In [
42], the economic operation of an MMG system is discussed considering the DNO and MGs as distinct entities. The problem is modeled as a stochastic bi-level problem, where the upper level is related to DNO and deals with operational constraints such as power flows and voltage levels. The MGs controllers at the lower level minimize the operating costs of the individual systems. In [
73], EVs are considered as ESSs to reduce the peak energy exchange with the main grid and the operational cost of the MMG system. At the first level, the total power exchange between the MMG system and the main grid is kept within the permissible limits, and the total electricity cost of the MMG system is minimized. At the second level, the charging and discharging power are allocated to each EV by optimizing the transitions between charging and discharging modes to increase the lifetime of the EV batteries. Similarly, the authors of [
78] design an EV coordination mechanism to minimize the system cost. Using the gradient projection method, the EV charging/discharging coordination is accomplished based on the time of use (TOU) pricing and the power exchange capability between MGs in the MMG system to minimize the load PAR ratio.
To optimize both the performance of the MG and user satisfaction, the authors of [
66] design a two-tier power control problem, which is solved using an online algorithm. The first tier maximizes the user satisfaction while minimizing the transmission cost between the main grid and the MMG system and the grid load variance. The second tier control is responsible for minimizing the power generation and transmission cost for each MG and ESS utilization. In [
67], a bi-level optimization problem is discussed to maximize the DSO’s profit at the upper level and minimize the cost of MGs in the lower level. Bi-level optimization is also implemented in [
114] to reduce the power loss and improve the voltage profile at the upper level. The optimal operation of DGs is determined at the lower level using the self-adaptive genetic algorithm and non-linear programming. A day-ahead optimal power scheduling approach is discussed in [
75], where a two-level interactive model is developed using the Stackelberg game. The DNO acts as a leader, which facilitates the trading of power and announces the trading prices of each MG operator (MGO). The MGO schedules the energy exchange with the DNO considering the day-ahead trading prices. The authors of [
77] develop an agent-based cooperative power management using Nash bargain solution (NBS) for resource management of MGs in an MMG system. The upper-level controller maintains the power balance at all PCCs of the MMG system, and the lower-level controller coordinates the operation of the micro-sources (MSs)’ control agents using internal bargaining.
Along with the optimal operation and minimization of operational costs, the authors of [
72] design a bi-level multi-objective cooperative dynamic EMS in grid-connected MMGs. The DNO minimizes the power fluctuations, voltage deviations, and power loss at the MMG level, while the MGs controllers minimize the power loss and operational costs at the MG level. The interaction between MGs and DN is modeled as a bi-level optimization problem and solved using a hybrid algorithm of hierarchical genetic algorithm (HGA)–non-dominated sorting genetic Algorithm II (NSGA-II)–rough set theory (RST). The interaction between MGs is modeled using an interactive energy game matrix (IEGM). In [
122], a bi-level interactive energy management is implemented for enhancing power balance in an MMG system and handling uncertainties. At the upper level, the energy exchange among MGs is optimized considering the operational costs, user priorities, and power consumption behavior of each MG. At the lower level, the MG operation is optimized using the MPC technique to minimize the deviation from the planned strategies generated at the upper level. In [
44], a voltage regulation mechanism is proposed for an MMG system based on MAS. A bi-level game model is developed where the Stackelberg-game-based incentive mechanism and static-game-based voltage control strategies are implemented at the DNO level and MG level, respectively. In [
102], a hierarchical DMPC algorithm is designed to optimize the operation of MMGs taking into account the MGs’ privacy while exchanging information and the flexibility to adapt to changes in the MMG network. The optimization problem is solved in a distributed way using ADMM, and an MPC-based scheme is utilized at each time step by individual MGs. To coordinate different EMS of MGs in an MMG system, the authors of [
62] propose an MAS-based stochastic programming approach taking into account the complicated interactions of MGs, restrictions, and uncertainties. In [
79], a two-time-scale control scheme based on MPC is deployed in an MMG system when the demand exceeds the power generated by RESs and batteries. The control strategy is divided into three levels, where energy trading between the DNO and MGs is scheduled at the top level, the load balance of each MG and battery ESSs operation optimization are dealt with at the middle level, and load shedding decisions are made at the lowest level. In some scenarios, the power management optimization can become a challenging task for cooperative agents due to the limited or incomplete knowledge of the internal behavior of MGs beyond the PCC [
80]. A bi-level model-free RL-based algorithm is designed in [
80] to handle the relation between the power exchange among MGs and the retail price of power. At the upper level, a central agent maximizes the power exchange profit of MGs by predicting the behavior of their unknown internal entities. At the lower level, the MGCC optimizes the power-flow-constrained power management of the MG considering the pricing signal obtained from the cooperative agent.
To enhance the resiliency of MMGs, the authors of [
120] design an exhaustive framework for the optimal operation and self-healing of DGs. During normal operation, a two-stage stochastic optimization is implemented to optimize the operation of MGs considering the predicted outputs of non-dispatchable DGs and load consumption at the first level. At the second level, power generation is adjusted based on the variations of non-dispatchable DGs output power. During the normal operation, the goal is to minimize the operational costs and maximize the MGs’ profit. In faulty conditions, corrective measures are taken to restore the system by sectionalizing the faulty section into multiple self-sufficient MGs. In [
68], the DSO is the coordinator of the MMG system at the upper level, which minimizes load shedding in the whole system. In case of a disturbance, the system is managed by automatic power scheduling mechanisms to share available resources among MGs. At the lower level, the MGCC is responsible for the individual MGs operation management. A novel resiliency index is also defined in [
68] to evaluate the effectiveness of the designed structure in terms of the expected energy curtailment during emergency conditions. In [
71], a self-healing mechanism is designed for MMGs. In case of a fault in an MG, the resources of other MGs are shared to compensate for the generation deficiency in the faulty MG and support the critical functionalities of the MMG system. At the lower level, the local EMS of the self-adequate MGs controls the DGs, ESSs, and loads while communicating with the neighboring MGs. In case of an emergency, the damaged MG broadcasts its power support request, and using an average consensus algorithm, a solution strategy is devised. At the upper level, after receiving the operating schedules of local EMSs, the operation of the DGs, ESSs, and loads is optimized. In [
96], the MMG system’s reliability is maximized while considering the minimization of the operation cost. The reliability of the MMG system is evaluated considering uncertainties in MGs components in both grid-connected and islanded modes of operation. ICA is used in the proposed hierarchical EMS. Preserving the privacy of MGs customers, the authors of [
129] develop an EMS for day-ahead scheduling of an MMG system with the objectives of operation cost minimization in the grid-connected mode and maximizing resiliency and reliability in the islanded mode using MILP. At the lower level, the MG’s EMS optimizes the local resources operation and informs the upper-level controller, which optimizes the operation cost of the MMG network. In the islanded mode, the resiliency of the MMG system is increased by sub-grouping of MGs. In [
129], the resiliency index of an MG is defined as the ratio of the amount of load restored during a particular event to the total amount of the MG’s load. In [
81], the resiliency of an MMG system is enhanced by maximizing the energy sharing among MGs in case of an emergency, neglecting the financial benefits and without compromising the MGs’ privacy. In an emergency situation, if the MG operates in the islanded mode, the total operational cost is minimized, while in the case of being connected to the MMG system, power exchange is maximized to minimize load shedding. At the first level, the MG’s EMS schedules its resources and calculates the amount of import/export power from/to the MMG system. At the second level, the MG’s EMS communicates with its neighbors to share the scheduled power. A summary of the reviewed studies on hierarchical control and EMS of MMGs is presented in
Table A4.
The advantages and disadvantages of centralized, decentralized, distributed, and hierarchical control structures are listed in
Table A1.