Machine learning
The development of machine learning has enabled computers to function without being supervised. Through machine learning algorithms, various articles have been published analyzing climate quality. These articles use diverse ML models, logical techniques and approaches as the core components to anticipate the speed of the wind. Scientists have established that machine learning algorithms are essential for measuring wind speed. Some of the benefits of using machine learning algorithms are discussed below
The structures and networks in the human brain can be simulated by the use of an artificial neural network model. In most cases, the structure of the neural networks is composed of nodes that produce a signal or a signal as per the sigmoid activation function. ANNS are prepared with training inputs and known output information. For training, edge loads are manipulated to minimize errors that may occur during training. The modelling analyses a feedforward multi-perception network made up of 12 information hubs, ten concealed layers of hubs 6 and 4, and 1 output nodes, as demonstrated in figure 5.
Phase functions at concealed layer nodes are all Gaussian. Training Problem Error Back Campaign, where there are 5-6 working hours until the network performs well against the training set.
Various architectures have conducted system testing in which few attempts have been successful.
The structure of the ANN used for the investigation, alongside the previous methods, utilized the inductive top-down decision tree, notably, the Oblique Classifier (OCI), which is accounted for improved performance as compared to the standard decision tree algorithms.
The general idea of OCI is that at each node, the tree can be split as per algebraic sum of many features, rather than one similar to the standard properties.
Some scholars consider Square Support Vector Machines, as demonstrated in figure 6. Square Support Vector Machine is described as a type of machine learning technique which is grounded on statistical learning theory. It is used in time series prediction and regression, and it has been reported to show good results and overcome many shortcomings associated with MLP. In this paper, scholars say a prediction model grounded on Least Square Support Vector Machines for wind and weather data, which reveal good results.
A wind Rose is used to show the azimuthal distribution of wind speed and wind direction at a given area. Wind Rose is essential for showing anemometer data (wind speed and direction) for seated analysis. The most common form of a Wind Rose is made up of 12 equally spatial radial lines (each line representing a compass point) with spatial equal concentric circles, as shown in figure 7. As shown in the Legend, each wind speed range is denoted with a unique colour. In the middle part indicates calm conditions while long lines show the prevailing wind directions. In figure 8, the Wind Rose shows that at 10 m, the western direction contributes 6.3% of the overall time. The southwest indicated the most reliable and highest wind directions. As shown in figure 9, the southwest at 900 Mb contributes 11% of the overall time.
The free parameters (weight and network optimal values) are used in the training process. MLP learning models are characterized by weight connection changes and specific input patterns. In these models, input layers distribute the input signal to the optimal values of the network, as demonstrated in figure 10. The output layer and the concealed layers contain a vector of processing elements and an activation function. The connection weight passes through the signals. A network is considered to be ready for training if it has weight and bias.
The adjustment of the network is grounded on output and target comparison during training. The process of training needs appropriate network behaviour and a set of examples of target outputs. To minimize the performance function of the network during training, the weights and gases network is repeated. In the feedforward neural network, the Mean Square Error (MSE) was used during the training as the performance function, as shown in figure 11. Mean Square Error is the average paid error between the target output and the network output.
In the experiment, data from Switzerland and Bale city was used. Figure 12 shows a comparison of data from Basel city in Switzerland between 1983 and 2018. In figure 12, input variables of ANN such as humidity, temperature, cloud cover, sea level pressure, sunshine direction, and cloud cover were measured hourly to determine the daytime speed of the wind.
Regression is used in the verification of network performance. Two regression plots are used to show the network output during training. They include; validation, and testing the set objectives. A researcher must ensure that data is along the 45 degrees line so as the fit can be perfect. At 45 degrees, the outputs of the network are equal to the target. For this study, there is an ideal fit for all data sets. The R values in each case are 0.975 or higher, as shown in figure 13.
Data from Basel city is used in the training of neural network training time series response model. In the testing process data from figure 13 is utilized. To validate the neurons hidden layer number, the sensitivity test is carried out through calculating prediction error change (MAPE) when neurons hidden layer number is changed ±5 from concealed layer neurons calculated by equation.
To predict the speed of wind in an area, the Multilayer Perception (MLP) neural network architecture is used. The MLP must have a MAPE of at least 6.32%. The ANN model performance plot shows that an increase in the number of epochs minimizes the mean square error. An epoch is one complete sweep of validation, training, and testing. The validation set error and the test set error has similar characteristics; therefore, they do not overfit each other near the epoch. The R-value shows the relationship that exists among the target value and the outputs of an ANN model. The correlation coefficient (R-value) of 1 and 0 denotes a robust, random association, correspondingly.
When data falls along the 45-degree line, scholars call it a perfect fit (slope is close to 1), and this means that the network output is equal to targets and the correlation coefficient is 0.975.
The NF tool developed the ANN model, which predicts the speed of wind close to the values measured. The ANN model has been used to measure the speed of wind in Basel, and the data is shown in figure 15. From the measurement, the MAPE is 6.329%, and this shows the ANN high accuracy. From the same location, the R-value is 0.98, and during the validation process, the slope is 0.97, and this, therefore, makes the ANN model fit to measure the wind speed in Switzerland.
Meteorological Data from the database between 1983 to 2018 is used as input for ANN Model. The variables used include maximum temperature, average temperature, air pressure, minimum temperature, altitude and solar radiation. Table 2 shows the speed of wind variation in different years.
This paper has presented a system that can be used in predicting the speed of wind daily in the next few years in the Basel area. The average MAPE and R-value obtained from the daily wind speed prediction are found to be 6.35% and 97.5% respectively.
References
Arslan, T., Bulut, Y. M., & Yavuz, A. A. (2014). Comparative study of numerical methods for determining Weibull parameters for wind energy potential. Renewable and Sustainable Energy Reviews, 40, 820-825.
Chang, T. P. (2011). Performance comparison of six numerical methods in estimating Weibull parameters for wind energy application. Applied Energy, 88(1), 272-282.
Warudkar, V., Ahmed, S., & Chaurasiya, P. K. (2016, October). Application of remote sensing in wind resource assessment: a comparative analysis. In Proceedings of the 15th World Wind Energy Conference, Tokyo, Japan (Vol. 31).
Kang, G. K., Gao, J. Z., Chiao, S., Lu, S., & Xie, G. (2018). Air quality prediction: Big data and machine learning approaches. International Journal of Environmental Science and Development, 9(1), 8-16.
Sarkar, A., & Pandey, P. (2015). River water quality modelling using artificial neural network technique. Aquatic Procedia, 4, 1070-1077.
Ip, W. F., Vong, C. M., Yang, J. Y., & Wong, P. K. (2010, August). Least squares support vector prediction for daily atmospheric pollutant level. In 2010 IEEE/ACIS 9th International Conference on Computer and Information Science (pp. 23-28). IEEE.
Yetis, Y., Kaplan, H., & Jamshidi, M. (2014, August). Stock market prediction by using an artificial neural network. In World Automation Congress (WAC), 2014 (pp. 718-722). IEEE.
Yetis, Y., & Jamshidi, M. (2014, August). Forecasting of Turkey’s electricity consumption using Artificial Neural Network. In 2014 World Automation Congress (WAC) (pp. 723-728). IEEE.
Mbarak, A., Yetis, Y., & Jamshidi, M. (2018, June). Data-Based Pollution Forecasting via Machine Learning: Case of Northwest Texas. In 2018 World Automation Congress (WAC) (pp. 1-6). IEEE.
Jamshidi, M., Tannahill, B., Ezell, M., Yetis, Y., & Kaplan, H. (2016). Applications of big data analytics tools for data management. Big data optimization: Recent developments and challenges (pp. 177-199). Springer, Cham.
Manjili, Y. S., Vega, R., & Jamshidi, M. (2013). Cost-efficient environmentally-friendly control of micro-grids using intelligent decision-making for storage energy management. Intelligent Automation & Soft Computing, 19(4), 649-670.
Jamshidi, M., Tannahill, B., Yetis, Y., & Kaplan, H. (2015). Big data analytic via soft computing paradigms. In Frontiers of higher order fuzzy sets (pp. 229-258). Springer, New York, NY.
Tannahill, B. K., Maute, C. E., Yetis, Y., Ezell, M. N., Jaimes, A., Rosas, R., … & Jamshidi, M. (2013, June). Modelling of the system of systems via data analytics—Case for “Big Data” in SoS. In 2013 8th International Conference on System of Systems Engineering (pp. 177-183). IEEE.