ABSTRACT

This study focuses on the effect of environmental factors on the power generation in a solar photovoltaic (PV) panel and the estimation of the power generated from this panel. Artificial Neural Network (ANN) and Multiple Linear Regression (MLR), as two different data-driven models, were used to assess the power of solar PV panels. Six factors are determined as input parameters in these models such as irradiance (W/m²), panel temperature (°C), ambient temperature (°C), wind speed (m/s), wind chill (°C), and humidity data (RH%). The output parameter of these models is the electrical power (W) generated instantly by the PV panel. In the development of both models, 550 experimental data sets (6x550 input data and 1 × 550 output data) were used, which were obtained from the position of the PV panel at different times throughout the year. In the light of these data, eight different training algorithms of ANN algorithms were tested and their success results were compared. According to the study results, the R² value was found 98.9% in ANN trained with Levenberg-Marquardt algorithm (trainlm), which showed the highest success. In the MLR analysis, the solar panel’s power estimation was found to be 94.8% (R²). Also, the weight of each environmental factor on the output power was determined by MLR analysis. The results indicate that the ANN model can successfully estimate the power generated under variable environmental conditions without extract a solar PV panel’s model-physical parameters. Also, the effects of environmental factors on power were determined by correlation and MLR analysis. When the presented methodology is generalized for PV plants, accurate energy planning can be made and a relationship can be established between demand and installed power capacity that prevents unrequited investment.

摘要

本研究侧重于环境因素对太阳能光伏 (PV) 面板发电的影响以及对该面板产生的功率的估计。人工神经网络 (ANN) 和多元线性回归 (MLR) 作为两种不同的数据驱动模型，用于评估太阳能光伏电池板的功率。六个因素被确定为这些模型中的输入参数，例如辐照度 (W/m ²)、面板温度 (°C)、环境温度 (°C)、风速 (m/s)、风寒 (°C) 和湿度数据 (RH%)。这些模型的输出参数是光伏板瞬间产生的电功率（W）。在两种模型的开发中，使用了 550 个实验数据集（6x550 输入数据和 1×550 输出数据），这些数据是从全年不同时间的光伏板位置获得的。根据这些数据，对 ANN 算法的八种不同训练算法进行了测试，并比较了它们的成功结果。根据研究结果，在使用 Levenberg-Marquardt 算法 (trainlm) 训练的 ANN 中发现R ²值达到 98.9%，显示出最高的成功率。在 MLR 分析中，发现太阳能电池板的功率估计为 94.8% (R ²）。此外，每个环境因素对输出功率的权重由 MLR 分析确定。结果表明，ANN 模型可以成功地估计在可变环境条件下产生的功率，而无需提取太阳能光伏电池板的模型物理参数。此外，环境因素对功率的影响由相关性和 MLR 分析确定。当所提出的方法适用于光伏电站时，可以进行准确的能源规划，并可以在需求和装机容量之间建立关系，以防止单方面投资。