**4. Brazilian and Uruguayan case study: Mucuri municipality, tropical region, and Colonia Eulacio, subtropical region**

The figures below show the time series used in the models which consist of 744 h for Mucuri in Brazil (**Figure 4**) and 8760 h for Colonia Eulacio in Uruguay (**Figure 5**), corresponding to hourly mean wind speed data. As can be observed in the figures, there is noticeable data randomness, and it is difficult to find a series tendency or seasonality.

In this sense, a descriptive statistic regarding wind speed at different sites is shown in **Table 2**. It can be noted that wind speed data measured at Colonia Eulacio has a higher variability.

**Tables 3–6** show the evaluation metrics of the prediction results obtained by the proposed model. **Table 3** presents simulation results referring to errors in wind speed forecasting 1, 3, and 6 h ahead for Mucuri using the RNN model. In **Table 4**, we present the results for regression for Mucuri. In **Table 5**, we can see errors in wind speed forecasting 1, 3, and 6 h ahead for Colonia Eulacio using the RNN model. **Table 6** shows the regression results to Colonia Eulacio. The percentage of the data of a factor of two is a fraction of data [%] for 0.5 ≤ wind predicted/wind anemometer ≤ 2.0. **Table 4** shows that the percentage of the data of a factor of two [%] to Mucuri is bigger than the percentage of the data of a factor of two [%] to Colonia Eulacio (see **Table 6**).

**Figures 6–8** show the results of six-step predictions of the wind speed series to Colonia Eulacio and to Mucuri. In **Figure 6**, we observe wind speed forecasting results of the model with RNN in six-step ahead for Colonia Eulacio, Uruguay. **Figure 7** presents wind speed forecasting 6 h ahead in a period of 744 ho of

**227**

**Table 2.**

**Table 3.**

**Table 4.**

**Prediction horizon [h]**

**Figure 5.**

*Study of the Wind Speed Forecasting Applying Computational Intelligence*

anemometric tower measurements. In **Figure 8**, we can see wind speed forecasting

*Simulation results (regression): wind speed forecasting 1, 3, and 6 h ahead for Mucuri using the RNN model.*

**Figures 9** and **10** show the multistep root mean square error (RMSE) evaluation of RNN and MLP for Mucuri, Brazil, and Colonia Eulacio, Uruguay. It is observed in **Figure 9** that as the prediction horizon increases, RNN were more efficient when compared to those employed in the study which applies MLP referenced by [36]. These results indicate that if we want a higher accuracy in the result to Mucuri, we must use a

results of the model with RNN in six-step ahead for Mucuri, Brazil.

**Pearson correlation coefficient**

*DOI: http://dx.doi.org/10.5772/intechopen.89758*

**Site Arithmetic mean of wind speed** 

*The experimental wind speed time series—Colonia Eulacio (Uruguay).*

*Statistics (Mucuri and Colonia Eulacio).*

**[m/s]**

Mucuri 7.91 8.53 2.92 Colonia Eulacio 7.21 9.02 3.00

**Prediction horizon [h] MAE MSE RMSE MAPE [%]** 0.839 1.111 1.054 11.07 1.385 3.154 1.775 17.63 1.779 5.108 2.260 21.26

*Simulation results (errors): wind speed forecasting 1, 3, and 6 h ahead for Mucuri using the RNN model.*

1 0.940 0.885 99.48 3 0.850 0.723 98.95 6 0.742 0.550 98.94

**Coefficient R2**

**Variance [m2 /s2 ]**

**Standard deviation [m/s]**

**Percentage of the data of a factor of two [%]**

**Figure 4.** *The experimental wind speed time series—Mucuri (Brazil).*

### *Study of the Wind Speed Forecasting Applying Computational Intelligence DOI: http://dx.doi.org/10.5772/intechopen.89758*

#### **Figure 5.**

*Aerodynamics*

tendency or seasonality.

has a higher variability.

Colonia Eulacio (see **Table 6**).

some agencies [33]. Nowcasting is critical when managing operations of the smart grid, such as system integration, ensuring power continuity and managing ramp rates [34]. In this chapter, nowcasting refers to short-term wind speed forecasting 6 h ahead. In [35] they described and evaluated a proposal for nowcasting wind speed for wind farm locations from historical time series, based on the method of regression by support vectors (in short, nowcasting of wind speed using support

**4. Brazilian and Uruguayan case study: Mucuri municipality, tropical** 

The figures below show the time series used in the models which consist of 744 h for Mucuri in Brazil (**Figure 4**) and 8760 h for Colonia Eulacio in Uruguay (**Figure 5**), corresponding to hourly mean wind speed data. As can be observed in the figures, there is noticeable data randomness, and it is difficult to find a series

In this sense, a descriptive statistic regarding wind speed at different sites is shown in **Table 2**. It can be noted that wind speed data measured at Colonia Eulacio

**Tables 3–6** show the evaluation metrics of the prediction results obtained by the proposed model. **Table 3** presents simulation results referring to errors in wind speed forecasting 1, 3, and 6 h ahead for Mucuri using the RNN model. In **Table 4**, we present the results for regression for Mucuri. In **Table 5**, we can see errors in wind speed forecasting 1, 3, and 6 h ahead for Colonia Eulacio using the RNN model. **Table 6** shows the regression results to Colonia Eulacio. The percentage of the data of a factor of two is a fraction of data [%] for 0.5 ≤ wind predicted/wind anemometer ≤ 2.0. **Table 4** shows that the percentage of the data of a factor of two [%] to Mucuri is bigger than the percentage of the data of a factor of two [%] to

**Figures 6–8** show the results of six-step predictions of the wind speed series to Colonia Eulacio and to Mucuri. In **Figure 6**, we observe wind speed forecasting results of the model with RNN in six-step ahead for Colonia Eulacio, Uruguay. **Figure 7** presents wind speed forecasting 6 h ahead in a period of 744 ho of

vector regression: experiments with time series from Gran Canaria).

**region, and Colonia Eulacio, subtropical region**

**226**

**Figure 4.**

*The experimental wind speed time series—Mucuri (Brazil).*

*The experimental wind speed time series—Colonia Eulacio (Uruguay).*


#### **Table 2.**

*Statistics (Mucuri and Colonia Eulacio).*


#### **Table 3.**

*Simulation results (errors): wind speed forecasting 1, 3, and 6 h ahead for Mucuri using the RNN model.*


#### **Table 4.**

*Simulation results (regression): wind speed forecasting 1, 3, and 6 h ahead for Mucuri using the RNN model.*

anemometric tower measurements. In **Figure 8**, we can see wind speed forecasting results of the model with RNN in six-step ahead for Mucuri, Brazil.

**Figures 9** and **10** show the multistep root mean square error (RMSE) evaluation of RNN and MLP for Mucuri, Brazil, and Colonia Eulacio, Uruguay. It is observed in **Figure 9** that as the prediction horizon increases, RNN were more efficient when compared to those employed in the study which applies MLP referenced by [36]. These results indicate that if we want a higher accuracy in the result to Mucuri, we must use a

#### *Aerodynamics*


**Table 5.**

*Simulation results (errors): wind speed forecasting 1, 3, and 6 h ahead for Colonia Eulacio using the RNN model.*


#### **Table 6.**

*Simulation results (regression): wind speed forecasting 1, 3, and 6 h ahead for Colonia Eulacio using the RNN model.*

**Figure 6.** *Wind speed forecasting results of the model with RNN in six-step ahead (Colonia Eulacio, Uruguay).*

recurrent neural network. A recurrent neural network allows self-loops and backward connections between all neurons in the network. That enables the networks to do temporal processing and learn sequences, e.g., temporal association/prediction.

Unlike the previous comparison, for Colonia Eulacio's anemometers data, the study which applies MLP referenced by [13] was more appropriate (**Figure 10**) than this study that applies RNN.

The multistep Pearson correlation coefficient evaluation of different architecture and site is shown in **Figure 11**.

The computational cost employed to simulate wind speed prediction through RNN for Colonia Eulacio is not viable when compared to the application of MLP. In

**229**

**Figure 8.**

**Figure 7.**

*2015, at 13:00 to April 28, 2015, at 14:00.*

*Study of the Wind Speed Forecasting Applying Computational Intelligence*

contrast, for Mucuri, it is considerably more viable to apply RNN, as can be seen from the figure above. Lastly, nowadays, adopting renewable energy has become a national energy policy for many countries due to concerns with pollution from fossil fuel consumption and climate change. Regarding wind energy, the accurate prediction of wind is crucial in managing the power load. Thus, this work presented the short-term wind speed forecasting for two representative sites in South America, Brazil, and Uruguay, which are the most important countries in terms of

*Wind speed forecasting results of the model with RNN in six-step ahead (Mucuri, Brazil).*

*Wind speed forecasting results of the model with RNN in six-step ahead (Colonia Eulacio, Uruguay), April 20,* 

Each scientific study on wind speed prediction has its own characteristics, such as the height of the anemometer that records atmospheric data (e.g., wind speed,

renewable energy production in Latin America.

*DOI: http://dx.doi.org/10.5772/intechopen.89758*

#### *Study of the Wind Speed Forecasting Applying Computational Intelligence DOI: http://dx.doi.org/10.5772/intechopen.89758*

**Figure 7.**

*Aerodynamics*

**Prediction horizon [h]**

**Table 6.**

**Table 5.**

*model.*

*model.*

**Pearson correlation coefficient**

1 0.922 0.849 98.44 3 0.729 0.531 93.60 6 0.543 0.295 88.52

*Simulation results (errors): wind speed forecasting 1, 3, and 6 h ahead for Colonia Eulacio using the RNN* 

*Simulation results (regression): wind speed forecasting 1, 3, and 6 h ahead for Colonia Eulacio using the RNN* 

**Prediction horizon [h] MAE MSE RMSE MAPE [%]** 0.895 1.408 1.187 16.08 1.687 4.710 2.170 30.65 2.266 8.051 2.837 39.86

> **Coefficient R2**

**Percentage of the data of a factor of two [%]**

**228**

**Figure 6.**

this study that applies RNN.

ture and site is shown in **Figure 11**.

recurrent neural network. A recurrent neural network allows self-loops and backward connections between all neurons in the network. That enables the networks to do temporal processing and learn sequences, e.g., temporal association/prediction. Unlike the previous comparison, for Colonia Eulacio's anemometers data, the study which applies MLP referenced by [13] was more appropriate (**Figure 10**) than

*Wind speed forecasting results of the model with RNN in six-step ahead (Colonia Eulacio, Uruguay).*

The multistep Pearson correlation coefficient evaluation of different architec-

The computational cost employed to simulate wind speed prediction through RNN for Colonia Eulacio is not viable when compared to the application of MLP. In

*Wind speed forecasting results of the model with RNN in six-step ahead (Colonia Eulacio, Uruguay), April 20, 2015, at 13:00 to April 28, 2015, at 14:00.*

**Figure 8.** *Wind speed forecasting results of the model with RNN in six-step ahead (Mucuri, Brazil).*

contrast, for Mucuri, it is considerably more viable to apply RNN, as can be seen from the figure above. Lastly, nowadays, adopting renewable energy has become a national energy policy for many countries due to concerns with pollution from fossil fuel consumption and climate change. Regarding wind energy, the accurate prediction of wind is crucial in managing the power load. Thus, this work presented the short-term wind speed forecasting for two representative sites in South America, Brazil, and Uruguay, which are the most important countries in terms of renewable energy production in Latin America.

Each scientific study on wind speed prediction has its own characteristics, such as the height of the anemometer that records atmospheric data (e.g., wind speed,

#### **Figure 10.**

*The multistep root mean square error evaluation of different architecture for Colonia Eulacio, Uruguay.*

direction, temperature, humidity, and atmospheric pressure) and the time series of this atmospheric data. These data are applied to train and test the efficiency of the ANN. On the accuracy of the use of ANN in the estimation of short-term wind speed and wind power forecasting, we can mention these earlier studies (see **Table 7**). The MAE average value for 1 h ahead was 0.847 m/s and for 3 h ahead was 1.420 m/s.

Other scientific research on wind speed has been developed. We can cite the works in [39, 40]. In [39] they analyzed the time series of wind speeds in Mucuri, Mucugê, and Esplanada, cities of the state of Bahia, and the Abrolhos Archipelago, Brazil, through the use of the detrended fluctuation analysis technique to verify the existence of long-range correlations and associated power laws. Already [40] describes a short-term wind energy forecasting tool based on a run set forecasting system of the WRF-GFS model that has been operationally implemented in the electricity system in Uruguay with estimates for Brazil wind energy production.

**231**

**5. Conclusions**

**Figure 11.**

**Table 7.**

**Prediction horizon [h]**

**Mucuri: this study**

worst in Colonia Eulacio.

Brazil and Uruguay.

*Study of the Wind Speed Forecasting Applying Computational Intelligence*

*The multistep Pearson correlation coefficient evaluation of different architecture and site.*

**Colonia Eulacio: this study**

*Accuracy of the use of ANN in the estimation of wind speed and wind power.*

The present chapter aimed to define the most efficient RNN configuration to predict the wind speed for 1 h and, after that, to apply it for 6 h ahead, using as reference observational data collected from two anemometric towers, with anemometers installed at 100.0 and 101.8 m height, located, respectively, in a tropical region in Mucuri, Bahia, northeastern Brazil, and in a subtropical region in Colonia Eulacio, Soriano Department, Uruguay. It has been shown graphically and verified through numerical simulations that the RNN was better than MLP in Mucuri and

**Reference [13]**

1 0.839 0.895 0.892 0.720 1.050 0.684 3 1.385 1.687 1.678 1.370 1.230 1.168

Mean absolute error

**Reference [36]**

**Reference [37]**

**Reference [38]**

In the light of the statistical results recorded in this work, the application of computational intelligence is a viable alternative for the predictability of wind speed and, in this way, wind power generation, mainly due to the low computational cost; however, one must choose the ANN architecture that best suits the project, as well as quantitatively and qualitatively analyzes the available data that will feed the network, since these variables directly impact the results of the forecast. The results of the short-term wind speed forecasting showed good accuracy at all the anemometer heights tested. Therefore, the proposed short-term wind speed forecasting method is an important scientific contribution for reliable large-scale wind power forecasting and integration in tropical and subtropical regions, like in

*DOI: http://dx.doi.org/10.5772/intechopen.89758*

*Study of the Wind Speed Forecasting Applying Computational Intelligence DOI: http://dx.doi.org/10.5772/intechopen.89758*

#### **Figure 11.**

*Aerodynamics*

**Figure 9.**

**Figure 10.**

**230**

energy production.

direction, temperature, humidity, and atmospheric pressure) and the time series of this atmospheric data. These data are applied to train and test the efficiency of the ANN. On the accuracy of the use of ANN in the estimation of short-term wind speed and wind power forecasting, we can mention these earlier studies (see **Table 7**). The MAE average value for 1 h ahead was 0.847 m/s and for 3 h ahead was 1.420 m/s. Other scientific research on wind speed has been developed. We can cite the works in [39, 40]. In [39] they analyzed the time series of wind speeds in Mucuri, Mucugê, and Esplanada, cities of the state of Bahia, and the Abrolhos Archipelago, Brazil, through the use of the detrended fluctuation analysis technique to verify the existence of long-range correlations and associated power laws. Already [40] describes a short-term wind energy forecasting tool based on a run set forecasting system of the WRF-GFS model that has been operationally implemented in the electricity system in Uruguay with estimates for Brazil wind

*The multistep root mean square error evaluation of different architecture for Colonia Eulacio, Uruguay.*

*The multistep root mean square error evaluation of different architecture for Mucuri, Brazil.*

*The multistep Pearson correlation coefficient evaluation of different architecture and site.*


#### **Table 7.**

*Accuracy of the use of ANN in the estimation of wind speed and wind power.*
