Performance Evaluation of Solar-PV Panels for Static and Sun Tracking System in Kirkuk-Iraq Using SVM

A significant achievement is the incorporation of the Maximum Power Point Tracking (MPPT) control algorithm into various converter architectures to improve the efficiency of PV systems. Katche et al. [13] introduced a photovoltaic grid-tied system that utilizes an MPPT control algorithm and a Single-Ended Primary-Inductor Converter (SEPIC) and Luo converter. This system achieved an impressive efficiency of 96% while also reducing total harmonic distortion. In Ganguly et al.’s study [14], the work primarily focuses on the development of MPPT algorithm design. The proposed system involves the utilization of a SEPIC converter to facilitate the charging of the battery using solar energy. To achieve maximum power under any environmental conditions, it needs to optimize the resistance (load) applied to the cells and accurately measure their output. MPPT devices are typically integrated into an electric power converter system that performs voltage or current conversion, regulation, and filtering. This system is used to drive various loads, such as power grids, batteries, or motors. This work focuses on the implementation of two methodologies.

Rehman and El-Amin [15] analyzed a 5.28 KW off-grid photovoltaic power plant in Dhahran, Saudi Arabia, for its efficiency. The research investigated how dust buildup and PV surface temperature can affect the panels' ability to generate electricity. Energy production decreased with increasing surface temperature and time, suggesting that dust buildup potentially be responsible, according to the study. This is particularly important for areas like Kirkuk, where similar environmental factors like dust and temperature play a significant role.

A study by Visser [16] evaluated solar power forecasting models for both intraday predictions and day-ahead collected data. It started by utilizing various methods such as physical, regression and AI (Artificial intelligence) based models, with focusing on deep learning. It demonstrates the importance of using high-quality data to create precise forecasting models. A notable feature of this research is the utilization of probabilistic forecasting models, which have shown superior performance compared to point forecasting models. Additionally for solar forecasting performance, many methods such as tree-based models, which are part of the field of artificial intelligence, are specifically recognized for their strong technical predicting. The study is crucial for comprehending the role of AI in improving the ability to forecast and ensure the consistency of solar power generation.

Another study has introduced an approach that assesses the technical and economic feasibility of a solar PV system. Where this approach combines machine learning and optimization techniques. The study utilizes site-specific data obtained from Moi University in Kenya, encompassing temperature, solar radiation, and power demand data. The simulation findings from this study emphasize the importance of accurately determining the appropriate dimensions for solar PV installations. Neglecting this crucial aspect often results in system malfunctions. Applying AI and machine learning in system design and optimization guarantees dependability, effectiveness, and cost-efficiency, rendering it a valuable point of reference for comparable research in diverse geographical areas like Kirkuk, Iraq [17]. Additionally, another work is concerned with designing an Adaptive Neuro-Fuzzy Inference System (ANFIS) based SEPIC Converter for MPPT of PV Modules. The method used for the system’s MPPT is Perturb and Observe. The system is controlled by an Adaptive Neuro Fuzzy Inference System. The work has a hardware prototype and the performance of the controller for MPPT is tested in simulation. A comparison between the fuzzy logic controller single-ended primary inductance converter for MPPT and the adaptive neuro-fuzzy inference system single-ended primary inductance converter for MPPT was discussed, and the results showed that the ANFIS has more efficiency [18].

Rahif [19] described MPPT system design for photovoltaic energy conversion by proposing the fuzzy logic controller and it details how to create fuzzy inference rules and fine-tune membership functions to regulate the duty cycle of a DC–DC converter so that photovoltaic array attained maximum power point despite fluctuating irradiance and temperature as well as it may offers an important control oriented alternative to data-driven ones such as SVM prediction that have been employed in recent studies and provides an alternate artificial intelligence method for maximizing photovoltaic output in consideration of variable environmental conditions.

The aim of apply machine learning techniques, and more specifically the SVM models, to predict the power output performance of PV systems accurately. This specific objective is set to compare static and sun-tracking solar panel efficiencies, so that the most efficient set-up concerning solar energy control and generation of electricity may be pinpointed. The contribution of this work lies in the development and validation of a prediction model to be applicable by both power producers and grid operators to give more exact forecasts on the generation of solar power. It introduces research on Support Vector Regression (SVR) in solar energy systems as follows:

• Helps to further understand the effect of panel orientation and different tracking mechanisms on the gain of energy.
• Is part of the complete knowledge framework in alternative energies which supports the transition towards more sustainable sources of energy generally and especially in Kirkuk Iraq.
• Offers insights that could lead to cost reductions and increased efficiency in the energy sector.

Solar PV systems are leading the way in renewable energy since they turn sunlight into electricity. Using the energy from the sun, a device called a solar cell, which is basically a p-n junction, can produce direct current (DC) electricity. Nevertheless, because each cell produces a relatively small voltage, solar panels consist of multiple cells working together to generate higher voltages. To comprehend and optimize solar panel systems, it is essential to know the properties of the current-voltage (I-V) and power-current (P-I) relationships. When it comes to improving the efficiency of PV energy conversion systems, the notion of the MPPT algorithm is crucial. Taking Özçelik and Yilmaz as an example, they investigated the use of wireless energy transmission in conventional and MPPT PV systems, and they searched how MPPT improves conversion efficiency [20].

Figure 1. Photovoltaic cell model

Figure 2. (a) I-V characteristics; (b) I-P characteristics current-voltage and current-power characteristics of a typical solar panel

The solar cell is a p-n junction that generates DC electricity power from the sun's radiation. Considering the resulting voltage for each single cell is small, the solar panel contains a number of these cells, that are linked together. The model has a diode, a current source, and a resistor in series, as shown in Figure 1. The effect of the resistor symbolizes the leakage resistance of the cell. The current source symbolizes the current produced by photons, the output of PV is stable under fixed temperature and fixed radiation of light. Figure 2 demonstrates the characteristics of the current voltage (I-V) and the current power (I-P) for a solar panel, respectively. MPP is the peak point of the power for a solar panel.

Earlier research has verified that the incorporation of single or double-axis tracking systems considerably enhances the energy output of photovoltaic systems; like Bazyari et al. [21] demonstrated that single-axis tracking and double-axis both has the capacity to enlarge the mean energy gathered in comparison to static panel installations on Qeshm Island in Iran. Carrying on in the same method this work assesses similar tracking systems under climatic conditions in Kirkuk in the Republic of Iraq using an Support Vector Machine (SVM) prediction model.

The next stage is the DC-AC inverter, which converts the DC electricity from the solar array into AC, suitable for distribution on the power grid. This conversion is necessary because the grid operates on AC, and most home and business appliances are designed to use AC power. Finally, the AC electricity is transmitted to the grid, where it is distributed to end-users.

In some systems, there may be additional steps or components, such as battery storage systems that store excess power for use when the sun is not shining, or transformers that adjust the voltage to the appropriate level for the grid.

After getting the current and voltage from the photovoltaic arrays, the MMPT system algorithm controls it by a method called hill climbing. The flow chart given in Figure 4 and the block diagram in Figure 5 shows the mentioned method. In the last cycle of the algorithm, the direction of the output voltage will continue in the same direction even if there are increments in the output power from the last measurement. If the output power has decreased since the last measurement, the voltage is inverted in the opposite direction. In each MPPT cycle, the voltage of PV will be set according to the algorithm. In addition, it will fluctuate around the voltage of MPP voltage when the MPP has reached, which will lead to the loss of power that depends on the step width calculated power of one adjustment. In case the width is large, any change in operating conditions that could slow or stabilize the algorithm of MPPT will quickly respond to that sudden change. In the other case when the width is very small; any change occurs slowly or stabling the system, it will respond very slowly to solve changes in insulation or temperature. For this, it is clear that all system works depending on the value of the width [24].

Figure 4. MPPT algorithm

Figure 5. A PV-based MPPT control and sun tracker equipment system

The data obtained from the designed system serve as inputs to the SVM model in our approach. The voltage and current measurements from the static and tracking panels are acquired through the SEPIC converter and MPPT configuration and were utilized to calculate power output. Those data points together with their related timestamps were subsequently input into the SVM for training and prediction. And an integrating hardware-collected data into the SVM ensures that the machine learning model is immediately informed by the actual performance of the system.

SVM is a supervised machine learning technique, and it identifies the decision boundaries to categorize data points based on prior classification. It thrives on complex data, transforming it into higher dimensions for clearer distinction. Focusing on key data points close to the boundary, it excels in prediction accuracy, making it valuable in domains like face recognition, bioinformatics, and image processing [25]. SVM stands out as a sophisticated algorithm designed for both classification and regression tasks, Figure 6 illustrates in detail how the SVM works in the realm of machine learning. The essence of SVM's functionality can be distilled into a sequential process that transforms raw input data into a predictive output, offering valuable insights across various domains. This paper delineates the operational framework of SVM, encapsulating its workflow into a cohesive narrative [26].

This study used a radial basis function as kernel in a SVR model. The given size of dataset and SVM hyperparameters were set to standard values like regularization parameters C = 1 and ε = 0.1 while the RBF kernel parameter γ using the default heuristic. And the hour of the day serves as the SVR model's input feature also the recorded PV power at that time is the model's intended output. Since the sun's position primarily affects solar irradiance and in turn power output the hour of the day was chosen as the predictor. Also, the data did not include any other climatic variables such as temperature or irradiance sensors. This work has made sure that scale differences wouldn't affect the SVM training by normalizing the input and output data and it was determined that this design and feature selection would adequately capture the daily power generation trend.

The initial data ingestion journey begins with the ingestion of input data into the SVM model. This foundational step involves collecting and feeding the dataset into the algorithm, setting the stage for the subsequent analytical processes. The data, comprising features and labels, serves as the raw material from which the model will extract patterns and relationships [27]. Feature space transformation is the heart of SVM's efficacy as its ability to project input data into a higher-dimensional feature space through a process known as feature mapping. This transformation is pivotal, as it enables the algorithm to discern complex patterns that are not readily apparent in the original input space. By elevating the data into a higher-dimensional realm, SVM facilitates the identification of a separable hyperplane, even in cases where the data is not linearly separable in its initial form. Optimization is the core algorithmic challenge that SVM addresses is the identification and optimization of a hyperplane that optimally separates the data into distinct classes for classification tasks or closely fits the data points for regression tasks. The optimal hyperplane is the one that maximizes the margin between the nearest points of the classes it divides, known as support vectors. This step is crucial, as the chosen hyperplane directly influences the model's generalization ability and its performance on unseen data [28].

Task-specific modeling depends on the nature of the problem at hand, classification, or regression. SVM adapts its strategy, in classification tasks, the model endeavors to categorize data into predefined groups, whereas, in regression tasks, it aims to predict continuous values. This versatility allows SVM to be applied across a wide spectrum of research areas and practical applications, from image recognition to financial forecasting [29].

Model Performance Evaluation where a critical phase in the SVM workflow is the evaluation of the model's performance. Through the application of various metrics such as accuracy, precision, recall, and mean squared error, researchers can assess the efficacy of the SVM model. This evaluation not only validates the model's predictive capabilities but also guides the fine-tuning of parameters and the selection of kernel functions to enhance model performance. Predictive outcome generation, a culmination of the SVM process, is the generation of predictions based on the input data. At this juncture, the model applies the learned patterns and the optimized hyperplane to make predictions on new, unseen data, providing actionable insights or decision support [30].

Figure 6. SVM algorithm workflow

The test is performed for both static panel and dual-axis tracker, the results are recorded from 7:00 a.m. to 7:00 p.m. The results are calculated and show the difference between the static panel and tracker system, as shown in Table 3. The second approach has shown more efficiency with an increase in the average power produced from both systems; the first static system produced 11.086 Watts on average per day while the system with sun tracking produced 17.037 Watts on average.

Table 3. Solar tracking for static panel and dual-axis

To predict the daily power output of both static and tracking solar panels, the offered graphs show the results of an evaluation of an SVM regression model. The tracking panel's graph reveals that the SVM model tracks the real power output trend very closely, capturing the midday production peaks. A sharp peak in the tracking system's power output lines up with the sun's path, allowing the panels to get the most sunlight possible. A reasonable correlation between the predicted and actual values indicates that the SVM model has learned the pattern of the tracking panel's energy production effectively. However, there are clear differences at specific times of day, especially in the morning and afternoon. Several factors, like changes in solar irradiance, weather, or shadows that the model doesn't fully account for, could be responsible for these deviations.

Using a paired t-test is statistically significant difference t-statistic = -4.381 and p-value = 0.0009 was obtained from the observed increase in average power output 17.037 W for tracking against 11.086 W for static panels. And the standard deviations were computed to show the variation of observed outputs as the tracking system produced 8.73 W while the stationary system obtained 8.33 W. These results are validating statistically significant advantages given by the dual-axis tracking system.

For the static panel, the SVM model was able to approximate the panel's behavior by producing predictions that were analogous to the actual data. Because these panels do not change their orientation in response to the sun's movement, the graph displays a smoother curve than the tracking panel. At the same time as the tracking panel anticipates that the sun will reach its highest point, the static panel also predicts that its power output will peak. However, the static panel's performance isn't as peaky as the tracking panels, so it produces less power in the end. The tracking system is better at capturing solar energy because its peak power output is lower than that of the static panels.

Assessing the efficiency of the model, one way to quantitatively evaluate the performance of the SVM model is to calculate metrics like MSE, which measures the prediction error. The more accurate the model, the lower the MSE value. The accuracy of a model can be assessed by comparing the predicted curve with the actual data points in plotted graphs, which offers a qualitative evaluation. Figures 8 and 9 demonstrate the predicted power output.

Figure 8. Predicted data of tracking panels

Figure 9. Predicted data of both static

The tracking panel's highest recorded output is about 31.8 W at noon time which fell short of its stated 40 W rating. The 40 W rating is determined under normal test conditions like STC: 1000 W/m² irradiation and 25℃ cell temperature and thus this variance is expected. While the panel's temperature much exceeds 25℃ and both factors reducing its instantaneous power output the actual irradiation may be somewhat below 1000 W/m² at any given point even in direct sunlight during the field experiments. That why during midday hot conditions in Kirkuk is reaching roughly 31.8 W about 79% of the specified power indicates normal performance of the panel and hardware configuration and under real conditions the panel was running close to its operational constraints.

This work has found that dual-axis tracking increases energy production by approximately 54%, which is comparable to but higher than the values reported in the literature but in mid-latitude areas the dual-axis trackers are generally said to produce between 33 and 41 percent more energy than stationary panels. Ideal circumstances improvements could be higher, for instance, Khan et al. [29] found improvements of 39% to 54% using a dual-axis tracker depending on daily and meteorological fluctuations. And this range includes a roughly 54% increase observed in Kirkuk. Which shows significant gain that can be attributed to the clear sky conditions that are had during the summer measurements and the fixed panel's constant orientation throughout the day also gave the tracker a significant edge in the early and late afternoon. Therefore, taking into account local factors and experimental configurations the results are in agreement with previous studies. So those results which found in current work is consistent with the performance improvements reported by Bazyari et al. [21] where single-axis tracking increased average energy by 35% and the double-axis has increased by 41%.