Computational System for Sizing Wind Energy Generation Systems Using Artificial Neural Networks

The objective of this work was to develop a computational application for the design of wind power generation systems in small-scale On-Grid and Off-Grid installations, using a user friendly and interactive process. Using artificial intelligence concepts in conjunction with genetic algorithms, to verify the technical and economic viability of the implementation of the wind power generation system. The application coding was done using the languages Java, C, C++ and the database in MySQL language, containing technical specifications and costs of components of a wind system (of this type of system). For the development of neural networks and genetic algorithms, the Encog library was used. The application has proven effective in designing and economic analysis of small wind systems, allowing fast and simple simulation of On-Grid systems and Off-Grid systems. In addition, it proved effective in storing and accessing the information regarding the simulations performed and in the comparison between them, in order to perform a new simulation. Also, it was reliable in the accomplishment of the economic analysis, returning in a clear form the feasibility or not of the implantation of the project.


Introduction
Wind energy is basically the kinetic energy (movement) generated by air masses caused by the temperature differences on the planet's surface, which rotates a set of propellers connected to an electricity generator.In Lopes (2011), the amount of energy produced is directly proportional to the size of its propellers, and to the local wind regime.Kingdom, France and Turkey were the countries that increased the most installed wind power capacity in 2014 (GWEC, 2015).
Currently, 141.6 GW are installed in the European Union, with a cumulative total capacity of 147.8 GW for the whole of Europe.Wind energy accounts for 44.2% of all energy installation capacity, exceeding hydroelectric dams as the third largest source of energy in the European Union, accounting for 15.6% of total energy capacity by the end of 2015 (GWEC, 2015).
In Brazil, there were 92 wind farms authorized by ANEEL in 2003, and its growth has been constant in recent years.In 2013, 4.7 GW of wind projects were contracted, and in 2014, 2.3 GW-all to be deployed by 2019, when Brazilian wind capacity is expected to reach 15.2 GW.This capacity increased by around 60% in 2015, from 6 GW to 9.8 GW (Abeeólica, 2016), representing 4.21% of installed capacity in the country and 5.54% of the national energy matrix, of which 73.08% are located in the northeast, and second is the southern region (ANEEL, 2015).In 2016, construction capacity in Brazil is 8.62 GW, representing CO 2 reduction of 16,864,332 T year-1, with 390 plants installed and installation capacity of 9.8 GW (Abeeólica, 2016).According to Lopes (2011), Brazil has one of the largest wind potential on the planet, although wind is responsible for 0.03% of 92 thousand MW installed in the country.
According to the EIA -International Energy Agency in European countries and Asia, there has been growth of renewable energy, especially wind power.The increase in wind power generation in 2015 was equal to almost half of global growth of electric energy.This was made possible by factors such as industrial restructuring, improved energy efficiency and the growth of renewable energy such as wind power (GWEC, 2015).
According to ANEEL (2016), Brazil is among the countries with the highest percentage of wind energy in its energy matrix.This representation consists of about 6.15% of the total matrix.There is an estimated increase in the use of wind energy, from 1,700 MW in 2016 to 5,959 MW in 2018, and the states of the Brazilian Northeast are more representative in this generation, as detailed in Table 1.Renewable energy technologies are capital intensive because most of the investment is concentrated in the initial phase of the project, corresponding to 75% of the total wind farm investment (Tourkolias & Mirasgedis, 2011).Socioeconomic characteristics of many regions, such as unemployment, lack of economic development alternatives and high migration rates of the active population, make the investment in these technologies advantageous.In Nguyen (2007), due to these characteristics, the construction of power plants demands labour, generating potential for the training and employment of rural populations in several localities.
In 2004, the Federal Government created as bases for the model of the Brazilian energy sector, based on Laws 10,847 and 10,848, dated March 15, 2004and Decree 5,163, dated July 30, 2004(Melo, Santos, & Yamamoto, 2016).
Considering that there are several regions with great potential for wind power generation in Brazil, and to ensure the growth of this energy source, the development of new technologies and investments are essential, and it is of great interest to create means to facilitate the deployment of significantly to achieve maximum efficiency in the use and transfer of energy from the wind.Having it in mind, the process sizing a wind power system is of great interest for projects where, in addition to the search for cost reduction in energy acquisition, there is a concern for greater efficiency in energy transformation.
To this end, two commercial energy markets were created as components of the new electric sector model, a regulated contracting environment (ACR), in which a set of distributors purchase electricity from generators in public auctions, with defined prices; and a free contracting environment (ACL), in which consumers and electricity producers freely negotiate their bilateral contracts.This official electric power market, operating through public auctions with long-term contracts (15 and 30 years), is an important instrument for consolidating the process of liberalization of the energy supply industry in Brazil, reducing investor risks, and stimulating efficiency with correct signs of the system's cost of expansion through competition, fuelled by studies of government-made planning (Abeeólica, 2016).
The ACL includes free market consumers who are entitled to choose their energy supplier by paying a fee for the use of the distribution or transmission system, with bilateral contracts and definitions of price, quantity, duration and coverage clauses.This market was created in Brazil for more than ten years.The right to be a free consumer was specified in Law 9,074/1995, amended by Law 9,648/1998 and complemented by ANEEL Resolution 264/1998 and it was determined by this law that the market should be gradually liberated so that large consumers could become free.
The electricity system is responsible for the distribution of energy, whose operation is coordinated by the National System Operator (ONS), thus considering contracts in Brazil as financial instruments (Melo, Santos, & Yamamoto, 2016).
The differences between production and consumption and the quantity contracted are settled in the spot market, defined according to the local price called the Settlement Price of Differences (PLD).The cash market transaction and the financial settlement process are the responsibility of the Electric Energy Trading Chamber.
The goal of this study was to develop a multiplatform computing system using artificial neural networks under the terms of the GNU General Public License (Note 2) for small-scale Off-Grid and On-Grid wind power systems, as well as to quantify wind turbines and the other components of the system, using an integrated database with the technical specifications and costs of the components of a wind system.

Software Development
For the development of this study, the programming languages Java, C, C++ and MySQL database were used, which allowed a platform independence to execute the application, allowing it to run on any operating system.
To manipulate these languages together, the Microsoft® Visual Studio 2015 Enterprise integrated development environment was used.
For the analysis of requirements and the development of this system, an object-oriented methodology was used with the modelling through the Unified Modelling Language (UML) and the fountain-type life cycle model (Sommerville, 2016;Pressman, 2014).
In relation to artificial neural networks and genetic algorithms, these were developed using Encog (Heaton, 2008), which is a Java programming library focused on the development of neural networks (Haykin, 2000).
All parts related to the Autorregressive and moving average models, were implemented using Matlab® software, using as a Toolbox System Identification (Ljung, 2016).
In order to integrate the Java and Matlab® development platforms, in order to organize the loading of files and analysis of the results, it was used the Matlabcontrol library (Note 3), which is a programming interface for controlling and interacting with Matlab® software sessions, from a Java application.
In order to facilitate the evaluation of the implemented models of artificial neural networks and genetic algorithms, several classes were constructed in languages C and Java for cost and sizing functions, in order to present the best scenario for this one.

RNA Modelling
In an RNA application project, it is necessary to specify several parameters and related factors in the network architecture such as the number of neurons in each layer, the number of hidden layers, the type of the activation function, the format of the input matrix of the data, such as data pre-processing information and learning algorithm.
In this study, the network that was used had multiple inputs and one output.The entries vary according to the specified template.It was also chosen by the multi-layer perception widely evaluated in other works Hamzaçebi (2008), Shamseldin (1997), Dawson (1998), Riad et al. (2004), Minnis (1996), with 3 layers, consisting of 1 input layer, 1 hidden layer and 1 output layer.
According to Gallant (1995), Thierens and Goldberg (1994), Sedki et al. (2009), Anders et al. (1999), Minns (1996) Wang et al. (2009) the results showed that more than one hidden layer does not play a decisive role in the generalization impact of the network, which may even increase the consumption of processing.

Economic Aspects of Wind Energy Projects
According to Dutra and Tolmasquim (2002), detailing the economic aspects of a project is as important as the analysis of technical feasibility and can be divided into two steps: initial project costs and annual costs for operation and maintenance.The initial costs of a wind project include charges such as: technical feasibility study, negotiations and development, engineering projects, equipment costs, infrastructure and miscellaneous expenses.
For the annual costs of operation and maintenance, there are equipment costs (replacement and prevention), rental and use of the land, insurance, among others.The calculations for economic evaluation (cash flow, present value of money, economic indicators and sensitivity analysis) were done according (Luzio, 2014;Blasques, 2005;Melek, 2013).The calculations related to economic indicators (net present value (NPV), internal rate of return (IRR), and payback) and serve as a basis for comparative economic analysis in electric power generation projects were according to Newnan et al. (2013), Hoji, (2014), Rebellato (2004).

Method of Designing Wind Power Systems
To meet the objectives of the study, it is necessary to set a critical method, an end project or an oversized one.
Another goal is to make this process simple to the user.According to Pinto (2013), the following stand out: wind data collection, energy demand to be met (for later delimitation of the number of wind generators and maximum current generated), quantity of load controllers and inverters, when necessary and quantification of the storage system for the Off-grid systems, with the number of batteries needed for a preference and a period of autonomy for which there is no power generation.Wind energy systems were also designed to detail the economic costs, such as costs, project initiatives and operating and maintenance costs.
The point that delimits the beginning of the sizing is the choice of the type of wind system, On-grid or Off-grid, which will influence the complexity of the project and the final cost.The next step is to obtain the wind data.
For this purpose, an access medium was created between the application and the information bank of the Brazilian Wind Potential Atlas of Cresesb (Note 4), which is a web platform that aims to provide information such as: average seasonal wind speed, predominant directions and Weibull statistical parameters, thematic maps and wind power flows for the whole country.This query was performed by inserting the geographical coordinates of the point of interest (Cresesb, 2014).If you prefer, you can manually enter the wind speed data collected to generate the average wind speed value and subsequent calculations.
Then, for the verification of the energy demand, the user had two conditions for defining the reference values and performing the calculations: i) Insertion of a value referring to average daily consumption, in kWh or; ii) Output of the simulator of average daily consumption, which allows the simulation of consumption based on the electrical equipment to be used, returning the average daily consumption in kWh/month.
With the definition of the type of wind system and possession of the values of the energy demand and average wind velocity, the wind energy design was started.It is extremely important to carry out simulations with several wind turbines of different powers and manufacturers, as well as with the other equipment, in order to verify the ones best suited to the characteristics of the region to which the power generation system is to be implemented, guaranteeing the minimization in the costs arising from this generation method.From that point on, the use of genetic algorithms was started, which aims to verify which will provide the best scenario for the implementation, making a comparison between all the equipment, analyzing all possible configurations with the equipment previously registered in the database of the platform, having as initial parameters of choice the lowest cost together with the least amount of equipment.Another purpose of the use of this algorithm is that, at the time of initialization of the simulation, it searches in its history whether or not there was similar simulation, thus minimizing the processing and optimizing the software performance.The calculations used to size were according to Pinho et al. (2008), Cresesb (2014), and Albano (2009).

Weibull Distribution
Probability density records are important when described by analytical expressions, that is, the probability of the wind speed being equal to a given value.The distribution that best fits the wind parameters is called the Weibull distribution (Willis & Scott, 2000).
The mathematical expression of the Weibull probability density function is: Where, k = form factor (dimensionless); c = scale factor (m/s).
When little is known about the wind regime, a good starting point is to assume k = 2; in this case the probability function is called the Rayleigh distribution, used for preliminary studies, when only the average wind speed is known to find its frequency of distribution (Pinto, 2013).For calculation of c and k of the Weibull parameters, we use: Where, v = average wind speeds (m/s); σ = standard deviation; Γ = gamma function.
Wind speed varies on diurnal, monthly and annual scales.Most of the time the wind speed variation reaches 10% between its annual average value and the long term average, ie, the wind tends to a certain speed profile.For the analysis of wind speed data, such data are usually divided into 1m/s intervals, which are measured in anemometric stations, which record the wind speed in predefined bands in minutes or hours (Pinto, 2013).The probability density function of the wind speed can be calculated for each hour of the typical day in the month, using time series data.

Correction Wind Speed According to Height
For Jangamshetti; Rau (2009) the wind speed varies according to the height, since as the height increases, the wind speed also increases in magnitude.Thus, the relationship between velocities 1 and 2 at heights ℎ1 and ℎ2 can be approximated by the law of power: Where α is the (dimensionless) power exponent, estimated from the averages of the specific heights ℎ1 and ℎ2, through Equation 6: By obtaining a new wind speed from the previous expression, it is possible to delimit the wind speed according to the height of the wind turbine.Through the relation between wind speed and height, the following expression can be used to delimit the wind speed at a certain reference height: Where, v 3 is the wind speed at height h 3 , m/s; v 2 is the wind velocity at height h 2 , obtained by direct measurement, m/s; α is a coefficient relative to the roughness of the soil surface in that region, usually determined by local experimental measurements.

Total Daily Load Capacity
According to Pinho et al. (2008) the calculation of the energy produced by a wind turbine allows the evaluation of any project that aims to use wind energy for the generation of electricity.One way of estimating the electric energy produced by the measured wind data series using the wind speed frequency distribution over a period of time and applying this distribution to the output power curve of the wind turbine.The power curve of the wind turbines indicates the electrical output of the wind turbine to a load, a storage system, or a power grid, depending on the wind speed at the rotor height.The result of the application of the frequency distribution to the power curve of the wind turbines provides the amount of energy produced in the considered period.
Equation 8 relates the total daily load capacity to the specifications of the wind turbine, to later delimit the number of wind turbines required for the project.

Number of Wind Turbines
Equation 9 calculates the number of wind turbines that will be needed to meet the daily demand of the project.
Where, NA = number of wind turbines (un.);CCDT = daily capacity of total load (Ah/day); CDAVM = daily load of wind turbine, calculated as a function of the wind speed distribution and the power curve of the equipment (Ah/day).

Maximum Wind Turbine Current
Equation 10 calculates the maximum current produced by the wind turbines that will be necessary for the project.
Where, MCA = maximum current supplied by wind turbine (A); CAPN = wind turbine current at nominal power (A); NA = number of wind turbines (un.).

Number of Load Controllers
Load controllers have the function of controlling the flow of energy between the generator and the batteries, protecting them from being overloaded or discharged deeply, which will influence the increase in the life of the battery banks.The load controllers must be selected based on the voltage and current characteristics involved in the wind system (Ackermann, 2012).
With the current data of the equipment and the load controller, it was possible to determine the number of controllers, according to Equation 11.
Where, NCC = number of load controllers (un.);MCA = maximum current supplied by wind turbine (A); CCC = load controller current (A).

Battery Bank Capacity
After obtaining the maximum current data and the load capacity it was possible to establish the capacity of the energy storage system.The goal is to ensure that, on days of wind shortage, the energy demand is met, and the maximum discharge level allowed for the battery (ies) is not exceeded.
The type of battery that is used was initially chosen taking into account the specifications of each manufacturer.After choosing it, the depth of discharge to be worked was defined.The deeper the charge and discharge cycles, the shorter the battery life.For this definition, the life-time curves shall be used according to the depth of the discharge provided by the manufacturers.
With the data of windless days, as well as other parameters of the battery, wind turbine and demand, it was possible to calculate the capacity of the battery bank, using Equation 12: Where, CBB = storage capacity of battery bank, in Ah; CCDT = total daily charge capacity, in Ah/day; PDFA = discharge depth at end of range, decimal; DSV = autonomy-days without wind in the critical month, in days.

Number of Batteries
With Equation 13, it was possible to calculate the number of batteries needed to compose the storage system in order to support the energy demand of the project.
Where, NB = number of batteries (un.);CBB = storage capacity of battery bank (Ah); CB = storage capacity of a battery (Ah).

Number of Inverters
Inverters have the main function of converting direct current and generated by the wind turbines in alternating current available to the grid, adjusting the frequency and the voltage level of the grid to be connected.In On-grid systems they also have the function of emitting information about the energy production and the interaction with the electric network (Ackermann, 2012).
The parameters necessary sizing the voltage inverter are: i) input and output voltages and; ii) nominal power of continuous use and of short duration.Also, it is necessary to check the total power of the alternating current loads and select an inverter with minimum safety capacity above the requirement (value to be delimited).The input voltage must be equal to the voltage of the batteries and the output equal to the voltage of the alternating current loads.If the summary of loads has resulted in the sum of the loads installed in alternating current is of 100 W, the inverter must have a minimum capacity for continuous operation of the same proportion plus the safety value.Equation 14 calculates the amount of inverters that will be needed to service the system: Where, NI = number of inverters; DMP = maximum power demand (W); PI = inverter power (W).

General Remarks
At first, the calculations for the design of On-grid and Off-grid wind systems, through direct or alternating current generators were the same, what effectively changed was the use of equipment for current conversion and when this need was identified, the application will return the required amount of these.
For the On-grid and Off-grid systems, the initial calculations regarding the energy demand to be supplied to the consumer were similar.However, in the On-grid system, the quantities and technical specifications of wind turbines and inverters are established according to the ANEEL resolutions 482/2012 and 687/2015.These resolutions, among other aspects, define the method of connection to the distribution network.On the other hand, Off-grid systems, in addition to the aforementioned equipment, load controllers and batteries, with their respective technical specifications, will also be sized.
It will be up to the system to verify all the necessary equipment for the project, either for isolated or connected to the network, besides the comparison between them presenting the best deployment situation to the user.

DIMEE System
The DIMEE (Computational System for Wind Scaling) was developed to assist in the designing and subsequent verification of the feasibility of a power generation project through the use of wind turbines, in an easy way and with the m perspectiv do this, it w The DIME user inform the user p system.Th    u select the Ad information to so will select t ours (Capacity ese settings.

Figure
Figure 9.A

Table 1 .
Participation of states in the production of wind energy(ANEEL, 2016)