The purpose of the literature review is to gain insight into the concepts of smart grids and optimization, with a special focus on energy management systems for microgrids. Therefore, the main topic of the present literature review is the classification energy management systems and optimization strategies for microgrids.
Section 2.1 presents a description of microgrids within the smart grid concept. A brief explanation of distributed generation, demand-side management and supply-side management of microgrids is provided. Moreover, different approaches for energy management systems of microgrids will be presented, as well as some strategies for weather and load demand prediction. Finally, a review and comparison of previous studies in the field of energy management and optimization of microgrids is provided.
Section 2.2 provides an introduction to optimization, focused in linear programming (LP) and mixed-integer linear programming (MILP), together with an example of MILP formulation. Moreover, this section provides an overview of some control strategies including weather and load demand prediction.
Microgrids will play a critical role in the transformation of the existing electricity grid towards the smart grid of the future. They can be defined as low voltage distribution networks consisting of renewable energy sources (RESs), backup controllable energy sources such as diesel generators, storage systems and controllable loads. They can help to integrate the previous components into the grid, enhancing its reliability and reducing the reliance on carbon emitting fossil fuels for power generation. Moreover, they are fundamental in supporting deep penetration of plug-in electric vehicles (PEVs) and plug-in hybrid electric (PHEV) vehicles.
Power and energy management systems (EMS) are important parts of the control of a microgrid and they have been a subject of significant ongoing research, addressed in Section 2.1.6. EMS are able to optimally operate controllable generators, dispatch the RESs active power and control the consumption of controllable loads following an economic criterion.
Microgrids can operate connected to the main grid or in stand-alone mode. In grid-connected mode, voltage and frequency regulation is performed by the main grid. Moreover, since the main grid in western countries has a very high availability, the main task of the MG will not be to guarantee power supply to the end-users, but to apply its EMS to inject energy to the main grid when the prices are high and buy energy from the grid when the prices are low. Moreover, grid-connected microgrids provide a better way of interfacing RESs with the main grid, potentially maximizing their energy yield 21.
On the other hand, if the microgrid operates in stand-alone mode, it is necessary to control the distributed energy sources to ensure voltage and frequency stability of the microgrid. Moreover, energy balance must be strictly respected at any time. The microgrid concept is vital to manage the rising energy demand via the local production of energy, as well as to reduce the GHG emissions by promoting RESs 21 41.
Optimal Control and Energy Management Systems for Microgrids
In order to guarantee microgrid stability and reliable operation, and to consider environmental and economic issues, optimality depends on the minimization of fossil fuel consumption, the management of storage units and loads. Control strategies for microgrids need to consider a cost function and specific system constraints. Figure 2.1 presents a classification of optimal control techniques for microgrid control 21.
Figure 2.1: Classification of optimal control techniques for microgrid control 21.
Authors in 2 performed a review of the optimization objectives, constraints, tools and algorithms used in energy management systems for microgrids. Figure 2.2 presents a schematic of the classification of objective functions considered in this study.
Moreover, 2 provided a brief review of the main optimization types used in energy management problems, together with several examples of previous studies that used the described optimization strategies. These types are linear programming, non-linear programming, stochastic programming, dynamic programming and non-differential programming.
Furthermore, a classification of the solution approaches for EMS was provided, including heuristic, agent based, evolutionary, model predictive control, neural network, round robin, Gauss Seidel and SD Riccati equation approach.
A summary of the control objectives and development methodologies in the field of advanced microgrid supervisory controllers (MGSC) and energy management systems (EMS) was carried out in 20. First, a classification of the control objectives was proposed according to the definition of hierarchical control layers in MGs. As a continuation, a detailed methodology review was performed with a focus on the MGSC/EMS related studies.
Optimal control of MGs includes basic functions such as local generation and consumption control and energy storage system’s management. The top level of a MG control system, known as EMS or microgrid supervisory controller (MGSC), performs the functions of Supervisory Control And Data Acquisition (SCADA). The aim of the EMS is to provide vital functions such as power quality control, ancillary services, participation in the energy market and optimization of the operation of a microgrid by enhancing its intelligence level. Some of the challenges that the increasing implementation of MGs brings are a smooth transition between islanded and grid-connected modes under either intentional or unintentional conditions, the integration of demand-side strategies and the management of the increasing penetration of RESs by the development of advanced scheduling and dispatching strategies considering prediction errors 20.
Figure 2.2: Classification of objective functions for EMS ofMG 2.
Hierarchical control has been proposed and widely accepted as a standardized solution for efficient MGs management 20. Figure 2.3 shows a hierarchical control scheme for large power systems. This strategy can be adapted for the control of MGs and it requires a telecommunication infrastructure to communicate between the different layers. The primary control is designed to manage distributed generation (DG) units adding virtual inertia or controlling the output impedances. Moreover, it performs the control of local power, voltage and current. It is the first one to react, applying a droop speed control in order to stabilize the grid frequency and prevent further deviations. Secondary control is a slower measure that corrects steady-state errors in frequency and voltage magnitudes produced by the primary control loop. It also deals with power quality control issues such as voltage unbalance and harmonic compensation. The tertiary control introduces intelligence to the whole system and is focused on the energy management and optimization of the system according to MG stability, environmental issues, economics, etc. This layer is developed at an entity called microgrid central controller (MGCC). The MGCC determines the control actions for power management based on the distributed energy resources (DERs) active power, load demand and storage requirements. The communication infrastructure allows the MGCC to communicate the power references (set points) to the DERs and loads. Moreover, each distributed controller (DC) ensures that the power reference from the central control level is reached 20 21.
Figure 2.4 depicts a centralized hierarchical control structure applied to isolated microgrids, which shows the control actions and variables related to each layer of the control structure 28. The main difference between this schematic and the one presented in Figure 2.3 is that in this case the EMS is located in the secondary control level, instead of in the tertiary control level.
MGCC are applied in MG power and energy management systems with a special focus on optimal dispatching and scheduling of energy units. Among these optimization-based EMS strategies one can find power management, economic dispatch (ED) and unit commitment (UC). Both ED and UC are essential in a decision making problem. According to the time scale of the EMS, day-ahead scheduler and short-term/real-time dispatcher can be differentiated, as depicted in Figure 2.5 20.
Figure 2.3: Hierarchical control scheme for power systems 21.
Figure 2.4: Hierarchical centralized approach to control of microgrids 28.
Moreover, genetic algorithms and particle swarm algorithms are largely employed in power systems due to their flexibility and the possibility to use complex formulations of the problem at hand. However, the user has no guarantee to reach optimality and they usually have long calculation times.
Mixed-integer linear and non-linear optimizers are commonly used for decision making processes for EMS in MGs, as they are commercially available in efficient software packages such as GAMS and CPLEX. Finally, rule-based systems and machine learning systems have an intrinsic ability to produce results in limited calculation times. Rule base systems can also use fuzzy systems. However, the main disadvantage is that the optimal strategy cannot be guaranteed 20.
2.1.3. Distributed Generation
Distributed renewable energy sources (DRESs) are intended to be the main suppliers of energy of a microgrid. Among DRESs one can find solar panels, wind turbines, biomass energy, geothermal energy, hydroelectric energy, etc. However, due to their uncontrollable character, they are supported by controllable generators such as diesel generators, and energy storage sources such as batteries or flywheels. Moreover, supply-side management strategies can be applied in order to control the power output of the DRESs, as it is explained in Section 2.1.5. As a continuation, a brief explanation of the RESs used in this work is provided.
OECD gross electricity production from renewable products (excluding generation from pumped storage plants) reached 2,471.1 TWh in 2015, a 3.8% increase from the 2014 level of 2,381.6 TWh. As it can be seen from Figure 2.6, this represents 23.0% of total OECD electricity production in 2015,
Figure 2.5: Schematic of day-ahead scheduler and short-term/real-time dispatcher for EMS inMGs 20.
which is the largest share of renewables in gross electricity production for any year in the renewables time series beginning from 1990 11. The two renewable sources that experimented the largest increase in electricity production from 2014 to 2015 were wind and PV energy. In the case of wind, electricity production increased by 78.1 TWh, mostly coming from Germany and followed by US and UK, reaching a 5.3% of the total. Solar PV increased by 27.3 TWh mainly driven by Japan and followed by UK, Italy and Germany, generating 1.6% of the total share 11.
These two renewable sources, wind and solar PV, are the most common ones in hybrid microgrids, together with diesel generators and energy storage. Moreover, they are the renewable sources considered in this study.
Figure 2.6: Renewable shares in OECD electricity production in 2015 11.
Solar PV Energy
A typical photovoltaic system employs solar panels, each comprising a number of solar cells. The working principle of solar cells is based on the photovoltaic effect, which describes the generation of a potential difference at the junction of two different materials in response to electromagnetic radiation. Electrons from the semi-conductor material are excited by photons and electron-hole pairs are consequently created. Then, the charge carries are collected at either end of the semi-conductor material by electrical contacts connected to an external circuit. In this way, a current can flow through the circuit and power is generated 40.
Wind turbines make use of aerodynamics to convert the kinetic energy stored in the wind into electricity. They are composed by blades, a hub and a rotor-nacelle assembly supported by a tower and a foundation. The elements contained in the nacelle vary with each technology, but it commonly includes the hub, low and high speed shafts, rotor bearings, a gearbox, a brake, a generator and a converter.
2.1.4. Demand Side Management
Demand-side management (DSM) consists of two main points: energy efficiency (EE) and demand response (DR). The focus of this work is on DR, which from now on will be referred to DSM, and it is based on changing the energy usage by the end-use customers from their normal consumption patterns in order to respond to changes in the price of electricity over time or in the availability of RESs energy. A DSM strategy can be benefit-driven or penalties-driven, both based on price incentives. There are two main ways of performing;
DSM: load shedding and load shifting. The first way consists on reducing the power consumption of a certain load, while the second is based on rescheduling the energy consumption of a load to a moment of the day where the energy is cheaper or there is an excess of RESs that needs to be consumed. Examples of controllable loads used for load shifting are cooling devices, air conditioning systems or heat pumps. Moreover, other types of loads can act as a buffering mechanism, consuming excess energy at a certain moment of the day and producing energy it is scarce. An example of such loads could be a water pump with an irrigation tower. Besides reducing customers’ electricity costs, a DSM mechanism can help to regulate the security and efficiency of the grid through peak shaving, direct load control (DLC) and capacity market programs 52.
It is important to note that if the microgrid is connected to the main grid, the DSM strategy needs to meet certain regulations and standards, such as the UFLS scheme of system protection. More information about DSM can be found in 47.
2.1.5. Supply SideManagement
Supply-side management (SSM) is applied in this thesis through the control of the power output from RESs, also called RESs curtailment. This strategy will be studied and applied in a stand-alone microgrid whenever there is overproduction of energy from RESs that cannot be stored in the BESS or consumed by the loads.
However, it is also common to control the output of RESs in grid-connected microgrids with high penetration of RESs, in order to ensure grid stability. The SSM strategy is based on a grid standard developed by the German association VDE, which copes with preventing stability issues due to an overproduction of renewable energy sources 45 46.
2.1.6. Previous Studies
Energy management and optimal operation of a microgrid (MG) is a widely researched topic that can be solved with various optimization techniques and different objective functions. Moreover, the architecture of the MG can vary, as well as the renewable energy sources and the controllable generation systems included. Finally, the MGs analyzed in the previous studies can operate in stand-alone mode, grid-connected or both, and can be interconnected to create a bigger and stronger grid.
The authors in 31 developed an EMS using unit commitment with a Rolling Horizon (RH) strategy for a renewable-based MG. The problem was formulated as Mixed-Integer Linear Programming (MILP) and it was based on forecasting models in order to implement a prediction horizon of 2 days. The MG considered for the study included two Photovoltaic (PV) systems, one wind turbine, one diesel generator, one Energy Storage System (ESS) and several loads, including a controllable water supply system. The fuel consumption of the diesel generator was represented by a non-convex function and approximated as piecewise linear segments for the MILP formulation. The EMS was designed to provide optimal online points for each generation unit, as well as signals for consumers based on a Demand-Side Management (DSM) mechanism. The authors performed an economic comparison between the RH and the standard unit commitment (UC). The objective of the EMS is to minimize operational costs. In order to do that, several goals had to be achieved: minimize the use of diesel; deliver active generation points for the diesel generator, ESS inerter and PV plant; turn on or off the water pump in order to keep the elevated water tank level within predefined limits and send signals to consumers in order to promote behavior changes following the DSM mechanism. All the communications of the MG where based on SCADA and several measurement devices.
In this study thee scenarios were analyzed. The first one considered PV panels, an ESS, a diesel generator and no DSM mechanism. It was proven that the UC-RH strategy reduced the expected total costs by 18% in summer and 27% in winter due to lower start-up and diesel operation costs. In the second scenario, wind energy was added to the system. Results showed a lower total cost with respect to scenario 1 for both UC and UC-RH. In the case of UC-RH, the cost reduction of scenario 2 with respect to 1 is 20%. Finally, scenario 3 implemented DSM mechanism and proved further cost reductions by implementing demand shifting coefficients and optimizing the water pump activation as a flexible load during periods with energy surplus.
A MILP formulation to manage the energy production and demand of a grid-connected microgrid within a reactive scheduling approach was proposed in 39. The objective of the EMS was to maximize the profit of the system while implementing DSM trough flexible demand profiles and with penalty terms. Moreover, a RH approach was applied to deal with the uncertainty associated with renewable energy production and consumption. The optimization program had to determine the amount of power to be produced and by which system generators (economic dispatch and unit commitment), the optimal storage level, the load-shifting schedule and the amount of power purchased from or sold to the utility grid. The total operational costs considered the production costs, storage costs and penalty costs. The latter was determined as a function of the delay in satisfying each energy demand. The revenue or income of the system was provided by the energy sold to the utility grid.
The RH approach was enhanced by introducing some features of the UC problem such as ramping constraints and minimum up and down time constraints, as well as non-convex production costs and time-dependent startup costs, which may introduce non-linearities in the model. The authors did not consider the fixed costs associated with the investment and installation of energy generators, as the design of the energy network was not taken into account and only short-term decisions were addressed. Moreover, the variable costs related to the energy production of PV and wind energy were considered to be zero. The case study was focused on a grid-connected MG with PV panels, a micro-wind turbine, energy storage and 30 different energy consumers. The EMS considered a control horizon of 15 min and a scheduling horizon of 24 hours. The MILP model was implemented in GAMS 24.1 and solved using CPLEX 12 to zero optimality.
The authors concluded that longer prediction horizons led to a significant reduction in the use of grid power to satisfy the demand under the assumption of accurate demand predictions, since more future information is received to solve the optimization problem. Therefore, the profit increased. On the contrary, if load-shifting was not considered, the profit of the system decreased. Moreover, managing energy demand allowed enhancing the flexibility and autonomy of the MG. As future work, the authors proposed the consideration of different factors, like environmental impact, through the implementation of multi-objective optimization approaches and the incorporation of rescheduling action penalties to avoid major changes in the initial schedule after the occurrence of an unexpected event.
In 19 the authors developed an EMS prototype for an isolated renewable-based MG, which included management of energy sources and flexible timing of energy consumption by modeling controllable and uncontrollable loads. A deterministic management model was formulated and then integrated into a Rolling Horizon (RH) control strategy in order to reduce the uncertainty in both production capacity and energy demand by considering updatable forecasts. The objective of the EMS was to maximize the economic performance of the system. DSM was applied by means of load-shifting to controllable loads in order to reduce the operational costs. Moreover, penalties were applied for not meeting the load demand or if the consumption started before its earliest bound of after its latest bound. The fuel consumption of the diesel generator was modeled as MILP by transforming the non-linear characteristic into line segments.
The authors considered a case study with a PV generation system, a diesel generator, an Energy Storage System (ESS) and sets of loads, of which 5 were controllable. The MG behavior was compared with and without DSM for a prediction horizon of 24 hours with time intervals of 1 hour for a scheduling horizon of 30 days.
For the scenario with DMS the operating cost of diesel was reduced by 11.3% due to a better exploitation of RES, which led to 27% less use of non-RES energy. Moreover, the environmental impact was reduced by preventing the emission of 1.62 tons of CO2 pear year, linked to the diesel fuel that was not consumed. Finally, the authors proposed as future work the consideration of different types of energy sources, the connection to the utility grid and the incorporation of weather and demand uncertainty.
Research conducted in 32 applied Model Predictive Control (MPC) to efficiently optimize the operation of a microgrid. The mathematical problem was formulated as MILP in order to ensure solution quality, reduce the computational burden and guarantee non-simultaneous charging and discharging of the BESS or buying and selling power to or from the utility grid. Moreover, the branch-and-bound technique was applied to the optimization problem because if a problem is reached, the solution is known to be globally optimal. The uncertainties related with renewable generation, time-varying load and time-varying energy prices were compensated with the feedback control law implemented by MPC. The objective function of the EMS was to minimize the overall MG operating costs while meeting the predicted demand of a certain period for controllable and non-controllable loads and satisfying complex operational constraints. The operational costs were modeled as a quadratic function including energy production costs, startup and shutdown decisions, possible earnings and curtailment penalties. The fuel consumption was approximated as a linear function.
The EMS was implemented in a centralized, high level controller in order to perform unit commitment and economic dispatch for all generating units and storages, RES generation curtailment, load forecasting, demand-side optimization and energy exchange with the utility grid. This study was focused on a grid connected MG, therefore the frequency was maintained within a tight range by the utility grid. One important assumption was that the state-space power quality was preserved. The effectiveness of the EMS was proved in a simulation setup which included PV panels with 16 kW, four distributed generators (DGs) and energy storage between 25 and 250 kWh. The prediction horizon was established for 24 hours with a sampling time of 1h.
A simulation setup with a grid-connected microgrid was established in order to compare different control strategies: heuristic, MILP, MPC-MILP and a benchmark strategy with a 24h prediction horizon and no forecast errors as a base case for the comparison. The results showed that a MPC-MILP strategy without storage led to a more efficient utilization of the DGs and a larger amount of energy sold to the utility grid. However, considering storage made the MG more economically efficient. An experimental validation of the control algorithm was performed at the Center for Renewable Energy Sources and Saving (CRES) in Greece, and savings went up to 34.7% for the MPC-MILP experiment with a prediction horizon of 72 steps of 15 min.
A MPC framework for reliable microgrid energy management was presented in 34. The grid-connected microgrid incorporated a local consumer, a wind turbine and a battery. The MPC strategy allowed considering cost values, power consumption, generation profiles, specific constraints and uncertainty due to variations in the generator model parameters. The EMS relied on a multi-criteria decision making for battery scheduling with the objectives of increasing the utilization rate of the battery during high electricity demand, therefore decreasing the electricity purchase from the external grid, and increasing the utilization rate of the wind turbine for local use, consequently increasing the consumer independence from the external grid. The MPC technique aimed at minimizing a cost function over a finite prediction horizon and provided as an output a sequence of optimal set points for the system components based on the given constraints. The mathematical problem was formulated as MILP due to the need of implementing binary variables in order to account for the physical limitation of the battery, which cannot charge and discharge at the same moment in time. The efficiency of the proposed approach was validated through simulation results and comparisons using real numerical data for a test system usually applied in bulk power system reliability evaluation studies. Wind velocity was assumed to be constant along a sampling interval and the system included time-varying electricity prices. The authors concluded that the increase of the prediction horizon led to a decrease in the electricity cost from the external grid. Moreover, the scheduler considered a longer period of time to determine an optimal plan, discarding “shortsighted” decisions. The EMS implemented a cost function with time-varying weights in order to assign less importance to the cost values which are further in the future.
Authors in 48 proposed an operational optimization and demand response for a hybrid renewable energy system consisting of solar and wind energy, a diesel generator and a battery storage system. Energy management strategies were applied to both demand and generation sides to realize the objectives of meeting the electricity demand while minimizing the overall operating and environmental costs. The EMS integrated day-ahead and real-time weather forecasting, as well as demand response and model updating through the implementation of a receding horizon optimization strategy. The demand response strategy was performed in two stages, first a day-ahead scheduling stage and then a real-time dispatch and adjustment stage. In the first stage, the consumers planned the electricity usage of the household for the next day based on the available generation information and electricity price. On the second stage, the on-line generation forecasting and the dispatch by the electricity providers are carried out. The environmental impact was accounted for by indicators of the equivalent cost in the objective function and the amount of load demand allowed to be unmet was regulated via a penalizing factor, which also represented the unwillingness of the user to shift the load demand of controllable loads. The EMS was validated through the application in a single-family residential home. Among the contributions of this research one can find the RES curtailment in order to avoid problems caused by over-production. The length of the prediction horizon was one of the most influential parameters on the optimal allocation of electricity generation among the available resources of the microgrid. With longer moving horizons, more information about the future can be obtained in order to tackle uncertainties and improve the economic performance. On the contrary, smaller prediction horizons lead to a reduced computational time. In order to find a trade-off between the previous characteristics, a prediction horizon of 6 hours was selected as the preferred solution. Simulation results showed that the optimal electricity cost provided by an EMS with a receding horizon of 6h and demand response strategy was 5% lower than the one of a DSM strategy without demand response. As a conclusion, the proposed approach was able to attain global optimization and improve the overall system efficiency by taking better advantage of the available resources.
A unit commitment problem modeled as generic mixed integer linear program (MILP) was proposed by 15 to minimize the operating cost of a grid-connected residential microgrid. Photovoltaic panels, thermal solar panels, distributed generators such as combined heat and power units, thermal and electrical storages, controllable home appliances and electric vehicles composed the studied residential microgrid. The optimization problem was designed to consider the supply and demand of both electrical and thermal loads.
A demand-side management strategy was implemented through shifting of controllable loads. The charging process of electric vehicles was modeled as a preemptive and reversible load job between release date and due date. A model predictive control (MPC) scheme produced the control scheme for the microgrid in an iterative manner. The operating costs of the system included electricity and natural costs. Moreover, revenues were considered from selling electricity back to the grid. The degradation of cooled goods in the refrigerator and loss of comfort due to temperature fluctuations were penalized in the objective function.
A control sequence of 7.5 hours was calculated every 15 minutes, based on a deterministic forecast of 22.5 hours. The simulation time was on average 28.5 seconds long, with an average MIP optimality gap of 0.03%. The objective of the simulations was to compare the performance of the MPC scheme with benchmark control policies. Three price scenarios where considered, including two market-based scenarios. Results indicated that the proposed approach (MILP-MPC) with demand side management and storage control achieved a reduction of 4.7% up 7.6% in the annual total operating cost compared to operation in heat-led mode with no storages or demand side management. Moreover, scheduling home appliances and electric vehicle charging jobs subject to technical constraints and user preferences reduced annual electricity cost of these loads by up to 30.4%. Therefore, the study confirmed that the proposed approach achieves remarkable relative savings for the residential microgrid. However, nominal savings seemed to be rather small compared with the expected investments required for the automated control.
Research conducted in 53 developed an MPC based optimization method for residential microgrid scheduling considering storage impacts and uncertainties in weather forecast. The microgrid under study was composed by renewable energy sources, a combined heat and power (CHP) unit, a boiler, electrical energy storage (EES), thermal energy storage (TES), an electrical vehicle with vehicle to home (V2H) option, and consumer loads. Among consumer loads one could find delay flexible loads, controllable power loads and critical loads. The Home Energy Management System (HEMS) included, among other characteristics, the on/off generator status constraints, the grid interaction and the feasible behavior of storage units and the electrical vehicle. In this research, the conditions of a deregulated electricity market and high RESs penetration level were considered. Moreover, a demand side management strategy was implemented through load shedding and shifting. The objective of the HEMS was to minimize the overall operation cost by controlling both supply and demand side units while taking into account the owners’ preferences.
The MILP optimization problem was executed at each decision time, considering short-term forecasts of RES generation, load demand and electricity price. Moreover, it was integrated into a MPC framework in order to reduce the negative impacts of forecasting errors and update the forecast model in a rolling way. The control horizon was one day, with updates every 30 minutes. The superiority of the proposed approach was proven by comparing it with a traditional day-ahead programming based residential microgrid scheduling optimization method. Under perfect forecasting conditions, the MPC approach showed a total cost of 62.39% of the day-ahead programming strategy. Moreover, a case study with forecast uncertainties was implemented to both the MPC and day-ahead approaches. The cost increment of MPC with respect to its perfect forecast was of 2.03%. However, a day-ahead showed a much higher cost increment, 19%. Furthermore, the impacts of peak power price mechanism were evaluated, which consisted on adding an extra penalty whenever the power purchased or sold to the grid exceeded a certain value. Simulations showed that the MG operation cost increased by 15.02% with respect to the case with no peak power mechanism. However, the mechanism effectively reduced the peak power frequencies and decreased the fluctuant of power exchanged with the external grid. Finally, a sensitivity analysis was carried out to discuss the impacts of energy storage systems (ESS) on the MG operation. It was proven that the costs of the MG operation without ESS increased by 13.45%. Moreover, the cost of MPC without EES is lower than for the day-ahead programming strategy.
Authors in 9 developed an energy management system (EMS) for a stand-alone microgrid composed by a wind turbine (WT), a diesel generator (DG), an energy storage system (ESS) and a sea water desalination system.
The EMS controlled the distributed generators (WT and DG) and the ESS and a hierarchical control was applied, with a coordination control layer and an energy management layer. It was to maximize the utilization of wind energy and minimize the use of the DG ensuring a stable operation. The coordination control of the MG was performed in two operation modes, where either the ESS or the diesel generator worked as the main power source to provide reference values of the system voltage and frequency. The EMS based the required remote control, regulation, signaling and metering of the system components in a SCADA design program. Moreover, rolling horizon (RH) strategy was applied to the EMS to control the intermittency and randomness of the WT power output and suppress the power fluctuations. Therefore, wind power was forecasted via a genetic algorithm-BP neural network-based wind speed forecasting method. The efficiency of the proposed method was proven via tests on a real-time digital simulator system. The tests considered a microgrid with 500 kWh ESS, a 1250 kVA diesel generator and 2MWWT generator.
The authors concluded that the EMS could recognize the system operation mode online and successfully achieve the switch between operation modes according to the forecasted wind speed and SoC of the ESS.
Research conducted in 52 proposed an optimal energy management for a grid-connected photovoltaic battery hybrid system with controllable loads linked to a demand-side management mechanism. The objective of the EMS was to minimize the electricity cost considering constraints such as power balance, solar output and battery capacity. The DSM mechanism was applied in an open loop method to schedule the power flow of the hybrid system over 24h. The DSM was based on a time of use (TOU) program, in which the electricity price was high during peak load periods and low during off-peak periods. The objective function considered the cost of buying electricity from the grid, the cost of selling electricity to the grid and the wearing cost of the hybrid system. In order to dispatch the power flow in real-time and overcome uncertain disturbances, a model predictive control was applied. It is important to note that the scope of the model was restricted to the system operation. Therefore, the installation costs were not considered in the model.
The hybrid system under evaluation consisted of a 28.8 kWh battery and a 7kW PV system. In order to account for the demand and weather changes between summer and winter, as well as between weekdays and weekends, four cases were evaluated: weekdays and weekends for summer and winter. In the simulations for the optimal control, without applying MPC, the forecast for both load demand and PV output was assumed to be the average value for winter or summer. When applying MPC, a linear state-space model was created with a distribution fitting method to implement uncertainty. In that case, the sampling period was one hour with a prediction period of 24h in order to optimally schedule a period of five workdays in winter.
In the case of optimal control, the operation of the hybrid system achieved maximal use of solar energy and battery storage. Moreover, consumers required a minimal amount of power from the grid for their consumption and they reduced their monthly expenses. The different scenarios showed a reduction in daily electricity cost from 58% to 65% and the costumers earned between 1.23$ to 1.83$ per day.
When MPC was applied, two scenarios were evaluated. In the case of positive disturbances of PV output, earnings increased by 27% compared to optimal control. When considering positive disturbances in load demand, earnings increased by 31%. Therefore, authors concluded that MPC provided great control performance in terms of accuracy and robustness, achieving higher cost savings than the optimal control method.
A stochastic-predictive EMS for isolated microgrids was developed in 28, which addressed uncertainty using a two-stage decision process combined with a receding horizon approach. A stochastic unit commitment (SUC) of the system variables was determined as a first stage using a stochastic mixed-integer linear programming (MILP) formulation with a 1-h time-step. On a second stage, authors proposed a non-linear shrinking horizon optimal power flow (SHOPF) to optimally accommodate intra hour dispatch variations while keeping a fixed boundary condition for the SoC of the ESSs at the end of the horizon of the SHOPF. The forecast used for the SHOPF had a 5-min resolution. The proposed EMS used a 24-h look-ahead window.
In order to prove the efficiency of the proposed approach, simulations were carried out in a modified CIGRE test system under different configurations and compared with a deterministic approach. Results showed that SUC provided lower total operation costs compared with the deterministic approach, which was 13% more expensive. Moreover, a higher ESS capacity proved to positively impact the operational cost of the microgrid.
Research in 30 proposed a model predictive control (MPC) for the optimal power exchanges and the charge or discharge of each local storage system in a smart network of power microgrids (MGs). The objective was to design an innovative control strategy for a cluster of interconnected MGs in order to maximize global benefits and exploit the fluctuations of stochastic renewable sources and demands. Each microgrid was composed by wind and PV energy, ESS, household loads and an energy management unit (EMU). The MPC algorithm included information about power prices, power generation and load and RES forecasts.
The length of the prediction and control horizons was set to 24 h, with a control interval of 1 h. Simulations were carried out with and without prediction errors on loads and RES power production. The operation of a single MG was simulated to prove the advantage of the proposed cooperative framework related to the control of a single MG. Therefore, three scenarios were evaluated: no prediction errors, prediction errors and comparison between one single MG and the cooperative network of MGs. The optimal value for the cost function in case study 1 varied from 49.35$ to 261$ when the initial state of ESS was incremented from 5 kWh to 500 kWh. In the second case scenario, with an initial state of ESS of 5 kWh, the implementation of prediction errors lead to a small increase of 3% in the system benefits. In the third case study, the cooperation between MGs showed an increase of 8.5% compared with the single operation scenario. Therefore, results showed that the cooperation among MGs provided significant advantages and benefits with respect to the operation of each MG separately due to the flexibility provided by this cooperation through exchanges of power with neighboring MGs.
A switched MPC strategy was designed in 54 in order to dispatch energy of a hybrid power system composed by PV panels, a diesel generator and a battery, which covered the daily energy requirements of an off-grid area. Switching constraints were implemented to describe the charging and discharging mode of the battery, instead of designing a switched state-space model. A unified linear multiple-input, multiple-output (MIMO) state-space model was used to design a simple predictive model. The objective of the proposed strategy was to optimally schedule the power flows to cover the load, charge and discharge the battery while minimizing the usage of the diesel generator, penalizing the excessive use of the battery bank and promoting the use of the PV generator. Uncertain battery parameters were estimated online by using the adaptive updating law. Moreover, the control and predictive horizons varied according to switching times.
One important remark was that the proposed strategy for the dispatch of the hybrid power system was fundamentally an optimization problem, instead of a control design problem. Therefore, the stability of the closed-loop system was not assured although the states of the closed-loop system were guaranteed bounded due to the proposed constraints.
Simulations were carried out for a summer and winter weekday considering the load demand of a Zimbabwe rural community clinic. In order to test the performance of the closed-loop system with disturbances, the load demand was incremented by a 20% and the PV output was reduced by a 20%. The initial state of charge of the battery was set to 70% of its maximum capacity. The time spans of simulation cases were established as 4 days (96 h).
For summer days, the switching time from discharging to charging was established at 7:30, and from charging to discharging at 17:30. As the PV power was sufficient to cover the load, the diesel generator was only used to cover the imbalance caused by disturbances. In winter days, switching from discharging to charging occur at 8:30, whereas the switch from charging to discharging was established at 16:30. In this case, due to the lower PV output, the diesel generator had to cover part of the load demand, besides the imbalance due to disturbances. In order to prove the effectiveness of the proposed approach, diesel energy consumption was compared when using different strategies such as switched MPC without online estimation, intuitive strategy and open loop optimal control. In every case, adaptive switched MPC proved to lead to the smallest diesel energy consumption. Moreover, if the battery capacity increased, the diesel consumption decreased.
Authors in 36 developed an advanced control strategy for the optimal microgrid operation of a diesel-PV battery islanded microgrid for rural areas using a two-layer MPC method. The first optimization layer solved a nonlinear mixed integer optimization problem (MINLP) using discrete dynamic programming (DDP) based on real-time predictions on future power profiles for the calculation of the optimal energy dispatch. In order to improve the robustness of the control strategy regarding prediction errors, the use of the reference trajectories of the diesel generator power and the SoC of the battery from the first layer was used to adjust the diesel generator on/off time over a short horizon. This was performed solving a boundary value problem (BVP), which used continuous states, instead of the discrete states of DDP. Moreover, an MPC control scheme was implemented in order to dynamically update the weighting factors, increasing the adaptivity and reducing forecast errors. The optimal power dispatch was determined over a horizon of 24 hours, where the MPC control was updated every 10 minutes and the BVP problem every 2 minutes.
The objective of the optimization problem was to minimize the diesel generator fuel cost and its running time while optimizing the battery usage and maximizing the renewable energy usage. Moreover, the robustness of the system towards prediction errors of load demand and power from RESs was improved in order to provide uninterrupted power supply to the microgrid. The implemented prediction method consisted of a seasonal auto regression integrated moving average model (SARIMA) and exponential smoothing.
The proposed method was tested in order to obtain the optimal power dispatch over a horizon of five days using a load profile from a Philippine village and a PV power profile from New Delhi. Simulations showed that the average forecast error is reduced by using the adaptive forecast model. The comparison of the optimal diesel generator operation assuming perfect knowledge about PV power and load demand and the operation from the two layer MPC method showed a small difference in the power dispatch of 0.54 – 1.09 kW. The battery power showed slightly more deviations. However, it was able to supply the surplus energy to guarantee an uninterrupted power supply. The average time to solve the MINLP problem was less than 20 seconds using a one minute time step and optimization horizon of 24 hours. Finally, benefits of the proposed methods were proven with a 7% cost reduction of the leveled cost of electricity (LCOE) over a lifetime of 10 years compared with the state-of-the-art method.
An online optimal energy storage control strategy for grid-connected microgrids was developed in 18 based on a mixed-integer linear program (MILP) optimization formulated over a rolling horizon window, considering predicted future electricity usage and renewable energy generation. A robust counterpart formulation was proposed in order to handle uncertainty in energy demand and generation prediction through prediction error bounds in a computationally efficient way. Moreover, further reduction in the computations was achieved through variable time steps and relaxation of binary variables.
The proposed adaptive energy management system (A-EMS) operated as a high-level controller to make optimal power flow decisions by solving a multiple-objective robust MILP optimization problem over a rolling horizon prediction window. Integer variables were used in order to consider charging/discharging inefficiencies and asymmetric buying/selling prices, as well as discrete decision variables related to controllable loads.
The A-EMS could communicate with all the microgrid elements, as well as have access external information on weather forecasts and market electricity prices. Moreover, the A-EMS discouraged electricity usage at peak times through a time-of-use pricing scheme, therefore reducing the energy usage from the main grid and avoiding load curtailment by covering the load demand with energy from storage and RESs.
The concept of battery green zone power rates and red zone incremental power rates were included. Red zone power rates were used to reduce peak demand costs by allowing the battery to temporarily operate outside its normal ratings in order to reduce the peak demand.
Simulations were carried out for a grid-connected microgrid using real electricity data from a commercial and residential setting (peak usage below 50 kW), an industrial setting (peak usage above 50 kW) and a solar power generation station up to 30 kW. A uniform random noise of up to the square root of the usage data was added to evaluate robustness of the controller to uncertainty in load and generation. Simulations for the robust MILP-based rolling horizon controller increased the computational time by 24% compared with the non-robust MILP. Moreover, the use of variable time steps greatly reduced computational costs by reducing the size of the optimization problem. The authors concluded that the proposed robust formulation was computationally more efficient than many other robust optimization approaches.
Research in 35 designed a novel security constrained energy and reserve management system of an islanded microgrid considering the steady-state frequency and demand response programs. The proposed EMS used the grid frequency as the key control variable, which could be excursed of its nominal value due to unpredictable intermittencies of renewable sources or load consumptions and due to the use of inertia-less inverter-interfaced DRESs. Moreover, it included a linearized ancillary service demand response program and up and down scheduled demand response secondary control reserves were considered.
The optimization problem was based on the frequency dependent behavior of droop-controlled distributed generation and formulated as a mixed-integer linear program (MILP) and its main purpose was to optimize the microgrid frequency according to environmental and economic policies. In order to model the impacts of uncertainties in the microgrid, a two-stage stochastic optimization algorithm was employed. In the first stage, system uncertainties were modeled by generating some random scenario using Monte-Carlo Simulation (MCS) and Roulette Wheel Mechanism (RWM).
Steady-state frequency control performance of the inertia-less static voltage source inverter (VSI) was modeled in order to obtain the optimal microgrid day-ahead frequency profile considering expected frequency excursions in a cost-effective scheme. Simulations were carried out for an islanded microgrid during a scheduling horizon of 24 hours. Two fuel cells, two micro-turbines and a gas engine were controlled using the proposed P-f droop method. Moreover, three wind turbines and two photovoltaic panels were also considered. Two cases were analyzed, with and without the demand response program.
Simulations shown that use of DR programs reduced the amount of the MG day ahead operational cost and emission levels. Moreover, the use of DR programs caused a 20% cost reduction on the scheduled secondary control reserves.
A methodology for the optimal dispatch of energy sources in hybrid and isolated energy systems was proposed in 4. A nonlinear discrete optimization problem was formulated with the objective of optimizing input and output time trajectories for a set of combined generating and storage technologies, minimizing the overall operation costs. RESs curtailment was implemented, as well as load shedding of controllable loads, including interruptible ones. Load shedding was modeled in the optimization problem according to their interruption costs. Relaxation techniques were applied in order to avoid the use of integer variables and forecasts of load and RES generation were considered over a time window of six hours. The mathematical model was designed to be easily implemented to complex hybrid system architectures with several storage technologies, such as batteries, pumping and hydrogen.
The proposed methodology was based on an existing experimental hybrid installation comprising PV and wind generators, a pumping hydro storage unit, a BESS and an electrolyzer-fuel cell system meant to be used in military and civil defense applications or in remote isolated areas and small islands. Simulations were carried out with a time step of one hour to show the feasibility of the discrete optimal control (DOC) compared to a suboptimal system behavior given by a load-following strategy called “greedy”, which dispatches resources at a specific operating point without considering the forecasted demand and RES production. The DOC approach for a summer day implied a cost reduction of 36% with respect to the “greedy” solution, as storage capabilities where fully exploited during the PV peak and no load shedding was executed. In a winter day, the “greedy” approach performed load shedding to both interruptible and firm loads, causing large interruption costs. The DOC strategy reduced the overall cost by a 33%.