Wind Energy Mapping in Al Batinah Region Using Data Mining Techniques

. Renewable energy sources are cheap to produce and have no upfront costs. It can take many different forms, including solar, wind, geothermal, hydroelectric, oceanic, and biological. There are numerous renewable energy resources in Oman, especially wind and solar. The kinetic energy of moving air is transformed into electrical energy to create wind power. This electrical energy has been used to pump water from the sea to tiny artificial ponds close to the wind farms, where the water is desalinated, used for agriculture, and then stored. The research attempts to identify the best locations for windmills that transfer seawater to small ponds. The various parameters are considered to select a suitable site for wind energy mapping, such as wind speed, elevation, distance to the main road, distance to an urban area, and land cover or land use. In order to choose appropriate locations for windmill mapping, data mining techniques are used. Based on parameters applied to classification and clustering techniques, a few wind mill sites are identified as suitable.


Introduction
Wind energy is one type of renewable energy that utilizes the wind's ability to generate electricity. Wind turbines are used to make use of kinetic energy and transform it into electricity. Energy from renewable sources is a clean and renewable source of power that does not release harmful emissions into the atmosphere. Wind energy is fuel-efficient, productive, and can be utilized to power homes, businesses, and store water in small ponds. Wind energy potential can be visualized in different ways. There is one strategy to use a wind energy map, which displays the estimated wind resources in a region. Wind energy maps can be created for broad geographic areas or for specific locations. A wind resource assessment, which calculates the amount of wind energy present at a specific location, is one strategy. Wind resource assessments can be used to identify potential wind energy sites and determine the viability of developing a wind energy project in a specific location.

Literature
The practical approach for choosing wind farm sites uses data mining techniques for wind mill installation. In [1], the parameters considered for wind farm site selection are wind speed, built-up areas, forests (if applicable), elevation, land type (if multiple land covers are available) and land inter-references, the structure of the earth's surface, and the location of the nearest energy installation. Since the data must be studied for the purpose of choosing a wind site, it must be reevaluated for the purpose of reducing dimensions and applying a prediction technique. For this, multiple regression analysis is used to forecast the suitability of the site for the installation of wind turbines, and the principal component analysis technique (PCA) is utilized to reduce the number of features. The primary components that are produced are subsequently used as multiple regression analysis' input parameters. While the data gathering is underway, information will be gathered from 35 distinct sites. In order to determine if a location is suitable for the installation of wind turbines, the data will be assessed using the two methods mentioned above. Site selection is undoubtedly the initial step for wind energy development. This study's primary restriction is to identify the components of specific qualities, which orders the attributes according to their significance in the selection of wind sites [1].
In [2], the author's discussed the accuracy and completeness of their predictions. It has been demonstrated that multilayer perceptrons, neural networks, and adaptive neuro-fuzzy inference systems produce better predictions for wind energy. In that paper, the models are specified for a few power projections, namely, for very short-term, short-term, medium-term, and long-term wind power projections. These models deliver superior performance on different time scales. As a result, models for wind power prediction are developed in an adaptable and efficient manner. In addition, dispatchers are provided with more palatable conclusions on wind power forecasts, which improve the accuracy of wind power predictions, cluster analysis, and testing analysis. Due to the spatial smoothing effect, the prediction error of a combined wind power system is lower than the prediction error of individual wind mill sites [2].
Estimate possible large-scale wind speeds in rural areas across Switzerland using multi-level techniques, which include geographic information systems (GIS), random forests (RF), and wind statistical models. Based on wind speed readings and various other factors, the rural areas are estimated on a monthly basis. The wind speed possibility across Switzerland's rural areas, the monthly wind power at turbine heights of 100 m for 200 m2 pixels encompassing Switzerland Given adequate data availability, the presented methodology is applicable to any sizable region. To train monthly RF models, 118 rural training points are used for feature extraction, and the wind power pixel values are labeled to estimate wind speed in the countryside. By taking a sample of the previously computed feature maps at the position of the training pixels, the training features are obtained. To create a rural yearly and monthly wind speed map at 10 m, wind maps are averaged throughout the year [3].
Countries with hot climates with low rainfall have to go for the option of generating drinking water from seawater. So, the availability of natural pools will be limited or negligible. There is a high demand to store water in artificial ponds, for which the water should be desalinated from seawater. In this paper, we have attempted to suggest artificial small ponds for storing desalinated sea water for irrigation and other regular purposes.
The minimum availability of fossil fuels, as well as their harmful and long-term negative effects on the environment, necessitates a focus on sustainable energy sources. Wind energy is cost-effective and produces less pollution [4]. Wind energy is environmentally friendly by reducing air pollution, thereby affecting climate change and maintaining energy equity [5]. Oman inaugurated a 0.5 GW solar power facility at Ibri for internal operations. Oman plans to generate 10% of renewable energy by 2025 and 30% of renewable energy by 2030 as a target using wind power [6]. With high wind flows along its extensive coastline, which stretches for over 2,000 kilometers, Oman also has a lot of wind power potential. Masdar completed a 50 MW (Dhofar Phase I) wind farm in the Sultanate's Thumrait, Dhofar region, in 2019. Muscat intends to construct more wind farms across the country [7]. A 150 MW wind farm, known as Dhofar Phase II, is anticipated to be finished in 2023, based on the report of IRENA [8]. In a different study, it was discovered that wind energy facilities have the ability to lower the cost of strategic development projects in the coastal city of Duqm. An enormous new port in the Indian Ocean called the Special Economic Zone at Duqm has drawn billions of dollars in foreign investment. The analysis discovered that, despite the region's decreased solar resource potential because of substantial mineral dust concentrations in the air, the project's 2,463 hours of packed wind each year may potentially deliver up to 75 GWh [9].3. Wind Speed Data.
A generic approach to express the wind blowing power is the airflow through the wind turbine moving at speed u, which may be used to calculate the kinetic energy of the wind: An equation can be used to determine how much energy the wind turbine produces: where, E-denotes energy in terms of Kwh, − Power factor, Q-Power in terms of Kw, ttime in terms of an hour.  The wind energy map of Sultanate of Oman is given in Figure 3a and 3b, shows high potential for generating wind energy. The various wind speed and wind direction data are available in [12], which is depicted in Figures 4a, 4b, 4c, and 4d. It is possible that the wind speed is higher between 1 p.m. and 4 p.m.

Classifier Measures:
The classifier measures are used to calculate the accuracy. The following values are calculated such as True Positive Rate(TP Rate), False Positive Rate(FP Rate), Precision, Recall, and F-measure [15].

Class Balancer
Reweights the data's instances so that each class has the same overall weight. All instances' weights will be maintained as total. In order to implement class balancing, which functions as a meta-learner and trains each algorithm using cross-validation technique,the class balancing (weight calculation) is only done on the training part. The class balancer is used to find the weights of each instance in the dataset [14].

Random Forest (RF) Algorithm
The RF algorithm [15] is given as follows: Step1: Each decision tree is constructed using a subgroup of data points and a subgroup of features. m features and n randomly selected records are obtained from a data set with k records.
Step 2: For every sample, an exclusive decision tree is built.
Step 3: An output will be generated by every decision tree.
Step 4: For supervised learning, the result is estimated using either the maximum number of votes or averaging.

Expectation -Maximization (EM) algorithm
EM algorithm: a technique to reach the maximum likelihood or maximum posteriori estimates of parameters using a statistical approach [15]. E-step: To allocate instances to clusters according to the current parameters of probabilistic clusters.

M-
Step: To identify the clustering or parameters that increase the sum of the squared error or expected likelihood.

Methodology
The methodology in obtaining wind energy map has been given in Figure 7. Finding nearby suitable sites that wouldn't be affected by the main road, urban area, farmland, and power grid is possible using the class balancer that was used in Phase 1 to identify the weights of each instance in the dataset. In Phase 2, the supervised technique was employed to identify the potential sites. Due to the above parameters based on distance, some areas are not suitable for windmill plants. Phase 3 involves identifying the sites using a semisupervised technique to separate location points based on parameters. Some location sites are identified as suitable by locating the center of each cluster. The clustering method can provide silhouettes of neighboring objects that are synchronized with their distance.

Study area:
Al Batinah region study area stretches over 554 km 2 in Oman. In Figure 8, the research area is the Al Batinah region in the coast of Oman, where the position points were chosen to be between (24°29′29.65" , 56°29′40.36" ) and (23°18′0.00" , 57°53′60.00" ). The dataset contains 56 sites that are used to find the wind energy mapping using the classification and clustering methods.    Table 2 reveals the categorization of suitable wind mill sites using the experimental parameters. This experiment reveals that 3 instances are related to class 1, 7 instances are related to class 2, 10 instances are related to class 3, and 36 instances are related to class 3.       Table 5 and Figure 10 show various clusters and their corresponding assignments. It reveals that 10 objects are silhouetted in Cluster 0, 7 objects are silhouetted in Cluster 1, 3 objects are silhouetted in Cluster 2, and 36 objects are silhouetted in Cluster 4.

Conclusion
This research work addresses site selection for wind energy mapping. In this work, classification and clustering techniques are used to find the sites in the Al Batina Region. 63 sites were selected for our study, and the parameters considered were wind speed, elevation, distance to the main road, distance to an urban area, and land cover or land use. Overall weights were calculated for each site based on the experimental variable values using a class balancer. Based on the weights assigned to each site, classification and clustering techniques have been implemented. In the Random Forest algorithm, the accuracy of the classifier was 98.2143%, and objects were classified into four classes. In the EM clustering technique, the four clusters are identified by the same classification method. It is evident that classes 1, 2, and 3 are suitable sites for wind energy mapping, as are clusters 0, 1, and 2.