2.1. Microgrid Control System
One important component of microgrids is the energy source. There are different sources for the generation of energy that can be classified according to their predominance, called conventional and non-conventional sources, or according to the type of resources used, that is, renewable and non-renewable sources [
9]. Conventional sources refers to the most widely used energy sources worldwide, among which the energy supplied by means of hydroelectric plants or the burning of fossil fuels, such as diesel, coal, and gas, stand out. Less predominate sources in the market are called unconventional sources, where the primary source is produced by natural resources or it is an emerging energy source under development. In the same way, renewable energy sources are those whose potential is abundant and include solar energy, hydraulic energy, wind energy, biomass energy, and geothermal energy [
10]. Conversely, there are other sources of energy that are not renewable, that is, they are found in a limited quantity in the world and the rate of their consumption is higher than their regeneration time; some of these non-renewable energy sources are fossil fuels such as coal, oil, and natural gas, as well as uranium. A traditional microgrid system generally consists of a set of wind turbines, photo-voltaic panels, small hydro power, diesel micro-turbine engine, and battery storage.
The unpredictable nature of renewable energy sources such as solar and wind affects the performance and reliability of the microgrid, due to excess electricity generation or lack of generation, which is considered to be the main drawback for its adoption [
11]. However, some approaches such as those presented by [
12] argue that this problem can be solved by combining two or more power sources together with a backup unit to form a hybrid renewable energy system. In other words, the system is operated with a set of energy sources and storage devices to satisfy the demand, even when some of them are not available.
The microgrid control system is described in [
13] as a four-levels system: the fourth level has a system that assigns the production of active and reactive power to each generator element of the grid according to the demand. The third level sets the voltage and frequency references in the nodes, while the second level of the control system corrects the voltage and frequency deviations in the network. Finally, the primary control level executes actions locally on the generation sources, keeping the voltage and power at the reference values.
Within the fourth level of the control system mentioned above, setting the active and, sometimes, reactive power from each generator unit at each time period is a core element of the strategy adopted to manage the power generated by the microgrid. This strategy would depend of the system goals, minimizing costs and/or maximizing coverage. In this study, the management strategy is a Unit Commitment (UC) that defines an optimization problem that seeks the optimal scheduling of generation units in a specific time horizon (hourly, daily, or weekly). The objective of the UC strategy described in [
14] is to satisfy the demand at minimum cost, considering the on/off states of each generation unit, its ramps, reliability restrictions, system capacity, transmission, environmental impact, etc.
2.2. Literature Review
The UC has usually been formulated as a non-convex and nonlinear combinatorial optimization problem by some authors. However, in some cases, the UC strategy can be applied to models like the one described by [
15], where a lineal cost function is considered and only management constraints are taken into account. This research adopts a linear definition of the cost function based on the amount of energy produced by each of the generation technologies.
Optimal power flow and UC has been also divided by [
13] into two more specific problems: static and dynamic economic dispatch. The first one is a typical mode of power system planning and operation, which only studies the optimization scheme of a single time section rather than the connection between each time period. When modeling the static economic dispatch of the microgrid system, the objective function is usually to minimize the overall operating cost of the microgrid. Most of the constraints only consider the active power limits of the generating units in the microgrid and the power balance constraints within it, while ignoring the characteristics of battery storage units such as useful life, loading and unloading ramps, among other. On the other hand, dynamic economic dispatch is defined in [
16] as one that considers the relationships between the values of the optimization variables in subsequent periods; for example, the battery level in a defined period affects the battery charge level in future periods depending on the state of charge or discharge that is defined through optimization.
The dynamic economic dispatch model of the microgrid takes into account factors such as the ramp restriction of the controllable sources and the operation restriction [
17] of the energy storage units, etc., which is closer to the power system. The addition of energy storage units not only makes the microgrid more closely connected in time, but also makes the operation more economical and reliable. In [
18], it is shown that dynamic economic scheduling with energy storage units can save about 37% of the operating cost compared to static economic scheduling without an energy storage unit.
One example of the implementation of a UC strategy to manage microgrids is presented in [
15]. This research proposed a optimization model for planning an appropriate stand-alone, renewable-based electricity system for off-grid communities in Colombia. It used implicit stochastic optimization to make decisions regarding the sizing of renewable energy sources to meet energy demand during an average day. This research also considers the use of a unitary battery system for each zone in such a way that the energy generated from renewable sources during a certain period of time can be used later. This work concludes that the use of renewable energies must be adaptive according to the conditions associated with the environmental variables. In addition, the combination of these technologies provides a solution that is significantly cheaper for the community than typical diesel platforms because it is not necessary to buy or transport fuel.
Another important study related to power microgrid management is [
19]. In this study, a multi-objective economic-emission dispatch problem of combined heat and power is developed. In their model, multiple energy sources are also considered: renewable and non-renewable, but its objective function involves unit operating costs, emission level, emission tax, and the cost of power purchase from the main external grid. Their case is also important since it is possible to study the chance of managing microgrids in non-isolated areas, whose access to energy from the main grid is partial and, therefore, energy from it can be accessed to fully or partially satisfy the demand in certain periods of time.
An important approach to mention is the one developed in [
20], where the microgrid is approached from the basic consumer unit. In this case, it is considered that each client has an energy storage unit that can be used to store energy generated from renewable sources and that this is then used to reduce the total load on the general grid. In this case, the main objective is to reduce the load on the conventional network from the use of production and consumption forecasts. In this study, metaheuristics are used to solve the proposed model and the results and computation times for the methods used are compared.
In [
21], a mathematical optimization approach is proposed for the optimal operation focused on the economic dispatch of a DC microgrid using renewable energy generators and energy storage systems through semi-defined programming. The proposed mathematical approach contemplates the operation of a DC microgrid over a period of time with variable energy purchase prices. This characteristic makes it a practical methodology to be applied in real-time operating conditions. Further, a nonlinear auto regressive exogenous model (NARX) is used to train an artificial neural network (ANN) to forecast solar radiation and wind speed for the integration and dispatch of renewable generation considering prediction periods of 0.5 h and a time horizon 24 h.
Finally, the work developed in [
22] shows a more robust model in which a greater number of possible technologies (diesel, gas, fuel, solar, and wind power generators) can be used for power generation and are considered, as well as different costs associated with the power generation activity. Consideration of these multiple costs is then reflected in multiple objective functions. On the one hand, there is cost-effective operation, which is the minimization of the operation and the aging costs of the micro-grid components. On the other hand, there is the maximum islanding degree, which states either there is no physical connection with the macrogrid, or no power will be exchanged, or only a fixed and predefined power profile may be considered as exchange power; in this case, the formulation of the objective function is similar to the cost-effective operation objective function. Finally, there is eco-friendly operation, where the objective function is formulated as the minimization of the pollutant treatment costs. A genetic algorithm is used to solve instances and a case of study is presented.
2.3. Renewable Energy in Colombia
Colombia is one of the most privileged countries in Latin America thanks to its geographical location. It provides special characteristics like zones where wind speeds are twice the world average and there is also sunlight most days of the year [
23]. As a consequence, a great research scenario is open for new ideas to propose and develop different methodologies to take advantage of these characteristics and then to focus on the optimal use of renewable resources, specifically, in off-grid zones.
In 2020, 29 off-grid zones supplied their own energy demand using renewable energies; however, it is a small number in front of 1798 off-grid localities that are distributed in almost 51% of national territory [
24]. In most of the cases, off-grid energy demand is supplied with diesel-based generators as a single option and only 31.3% of them have electric service available 24 h a day. One of the reasons that could explain this limited use of renewable options to supply energy in those off-grid zones is that their installation cost used to be substantially high and made this solution financially impractical. However, their prices have come down and the renewable generator industry is more competitive than before [
25].
In the same way, in recent years, the Colombian government has developed some laws, such as “Ley, 1715” [
26], that have encouraged the development of projects that promote the generation of electrical energy solutions in off-grid areas [
27]. However, these laws do not present a clear framework on the benefits that the use of renewable generation sources can bring in specific contexts such as isolated areas. In this sense, and given the economic difficulties represented by the implementation of projects based on renewable energies in these contexts, it is necessary to develop alternatives that allow for the intelligent management of resources and that ensure their optimal operation.
The model we propose seeks to manage the energy microgrids considering uncertainty from different sources (environmental variables that affect the generation and demand of energy) at the same time as it implements specific strategies on some components of the microgrid, such as the battery system. This management is carried out through an optimization process by scenarios, which allows for adaptation to various situations whose conditions may affect the operation of the microgrid.
2.4. Proposed Optimization Approach
Mathematically, the problem is described as follows: Let be a set of available generation technologies (Solar: S, Wind: W, Diesel: D), be a set of electric generators, conventional and non-conventional, each of them with technical parameters and dimensions depending on the associated technology, and be the set of time periods within the planning horizon. In addition, consider a battery system with the capacity to charge when the total power of the microgrid exceeds the demand load and also supports the microgrid by serving as another energy source.
The goal is to satisfy the expected load demand for each time period (
) into an energy microgrid, determining the functional capacity connected to the network of each generation technology in each time frame, all while minimizing the total cost of operation.
Table A1 presents model variables.
According to the nature of the problem, the optimization model needs to calculate how much power needs to be produced by each generator at each time period (
). Therefore, based on the work of [
15], Equations (
1)–(
3) describe, respectively, the solar, wind, and diesel generation. Let
serve as the function that returns the type of generator technology
. Thereby, power generation
depends on whether the generator is switched on or off.
In the generation equations, the concept of installed capacity is used to measure how much energy can be generated from a specific source. In the case of renewable sources, the concept of peak power is used by the technology, which allows one to calculate how much electricity would be transformed according the amount of the primary source.
Considering the decisions and the associated generation functions, the mathematical model is defined as follows:
The objective function (
4) seeks to minimize the variable operational costs of the energy system. The variable operational cost,
, is associated with the maintenance of the equipment during its life-cycle and the fuel, if any is used by the generator.
The first group of constraints is related to the control of the use of the generated energy. Equation (
5) splits the generated energy into the portion that seeks to satisfy the demand, the portion that is used to load the battery, and a potential portion that is discarded. Meanwhile, constraints in Equation (
6) ensure that the demand is satisfied through the sum of the energy generated for that purpose, the energy coming from the battery, and the energy from a dummy generator representing the unmet demand.
The second group of constraints is related to the state of charge of the battery. The constraint in Equation (
7) calculates the state of charge of the battery system, at the end of the period
t, taking into account the energy dissipation
, and the charging and discharging efficiencies,
and
, respectively. Equation (
8) sets the limit to the capacity of the battery
. The remaining sets of constraints implement strategies for managing the charging and discharging processes in order to enhance the battery health and long term performance
Constraints in Equations (
9) and (
10) set the value of the binary variable
that indicates if the battery has been discharged (
) in period
t.
The first strategy is modeled through constraints (
11) and (
12), where they count the number of times the battery system enters a discharge process.
Constraints in Equations (
13) and (
14) implement the second strategy, which regulates the charging and discharging ramps. A third strategy to manage the battery ensures that the number of times that the battery reaches the deep discharge level (
), and the overcharge level (
), is less than
and
, respectively.
Hence, the equations from (
15) to (
18) determine when the discharge or overcharge levels are reached and constraints (
19) and (
20) set a limit to the number of times that it occurs. Finally, constraints (
21) and (
22) define the domain of the decision variables.
2.6. Instance Generation
This research considers energy microgrids in isolated communities of Colombia subsidized by the state. These areas are difficult to access and the energy solution currently implemented is 100% based on the use of diesel generators. This situation presents logistical challenges to ensure the correct and continuous operation of the microgrid.
Data was collected from three different regions in Colombia: San Andrés (SA), Providencia (P), and Puerto Nariño (PN). These locations already have diesel-based solutions to generate electricity and their installed capacities are presented in
Table A3.
The parameters considered in the model are classified in technical, demand, and environmental parameters. Technical parameters are defined in
Table A2. For batteries and generators, these parameters were drawn from existing bibliographic sources according to the used technology [
28]. The default values for each parameter within this category were established by experts in the field. The parameters within the environmental category are solar radiation, wind speed, and temperature. The web tool RenewableNinja [
29] was used to collect historical data for these parameters. This web tool provides a global meteorological database from the MERRA-2 system of the National Aeronautics and Space Administration [
30] that gives timely (per hour) information. The downloaded data is from January to December 2019.
Finally, the demand parameter represented by the demand loads of the NIZ in Colombia were obtained and analyzed from the “Instituto de Planificación y Promoción de Soluciones Energéticas para Zonas No Interconectadas” (IPSE) that shared all NIZ’s hourly energy supply reports from 2019. These data is supplied to the IPSE by the “Centro Nacional de Monitoreo” [
24] that measures the actual electrical energy consumption of different isolated communities through telemetry systems. Data for every day of the year is available. However, when data was analyzed, it was observed that there was no significant variation between the same days of the week throughout the year. For this reason, an annual average is used as the actual hourly load for every day of the week. Therefore, 21 instances of the problem were built, seven for each of the three NIZ considered.
2.7. Design of Scenarios
The experiments carried out with the model seek to generate information to validate the model’s sensitivity and to understand the impact that some selected parameters have on different management scenarios. For this, the following four research questions are formulated:
How sensitive is the model to changes in demand? (Scenario 1)
How sensitive is the model to changes in the availability of resources? (Scenario 2)
What impact do the technical characteristics of the battery have? (Scenario 3)
What is the impact of the penalty on the unserved demand? (Scenario 4)
For each of these questions, scenarios with the parameters of interest (called factors) will be identified and defined, and different levels will be experimented with.
Table 1 describes the four scenarios in terms of the parameters of interest in each case and the values that they will take. The experiments of each scenario correspond to a complete factorial design of the different levels for each factor.
The first scenario seeks to show the capacity of the model to properly manage the changes on demand under different conditions of availability of renewable resources. Therefore, it considers the actual demand, an increase of 25% and a decrease of 25% in the hourly demand, and a special case in which only the interval with the highest demand is increased by 25%. Additionally, the percentage of renewable energy generation varies from 0% to 30% with steps of 5%. The remaining factors are set to their default value according to the recommendations of experts in the field of study. The second scenario explores the model’s sensitivity to changes in the availability of renewable energy resources. This is achieved by keeping demand and other parameters constant and making variations in the percentage of renewable energy generation of 5%, starting from 0% and reaching 30%. The third scenario evaluates the impact of the technical characteristics of the battery in meeting the demand and the performance of the microgrid. For this, the demand is left constant, a level of renewable resource capacity is selected where the use of batteries is evidenced, and different levels of the parameters , , , and are tested. Finally, in the fourth scenario, the effect of the penalty to the unmet demand is studied. In this case, different levels of penalty cost are tested for the unmet demand to check the behavior of the latter as said cost increases. These different test values are based on percentages according to the cost of the most expensive generation source, generally diesel generation. In this way, it is possible to know how much the cost of unattended energy should be raised so that the management model minimizes the unmet demand.