Pal Shuvo Shuvendu, Pal Shibazee Shirshendu, Paul Goutam, Paul Mita Mitaly, Das Chaitee, Malakar Konika
Department of Civil Engineering, Khulna University of Engineering and Technology, Khulna, Bangladesh.
Department of Electrical and Electronic Engineering, Bangladesh University of Engineering and Technology, Dhaka, Bangladesh.
Sci Rep. 2025 Apr 25;15(1):14410. doi: 10.1038/s41598-024-82950-2.
Previous research has shown that predicting solar radiation is a challenging issue due to highly nonlinear and noisy climate data. Various hybrid approaches have been applied earlier for solar radiation prediction, which integrates the Wavelet Transform with various Machine Learning models. This research, therefore, intends to further improve the performance of these existing hybrid models. To address the limitations in handling nonlinear and noisy climate patterns, this study proposes a multi-hybrid model for accurately predicting solar radiation that incorporates the Hodrick-Prescott Filter (HP-Filter), Discrete Wavelet Transform (DWT), and Support Vector Machine (SVM). The collected data from the Bangladesh Meteorological Department for two different geological locations in Bangladesh, namely Dhaka and Chittagong, is divided into three categories for modeling: 70% for training, 15% for validation, and 15% for testing, whereas the model hyper-parameters of the SVM were optimized using the Particle Swarm Optimization algorithm. The proposed approach applies the Hodrick-Prescott Filter before analyzing DWT to strengthen the SVM model's ability to capture complicated climate patterns in great detail and also make the model more precise and reliable. Several performance metrics, such as Mean Squared Error (MSE), Root Mean Squared Error, Mean Absolute Error, Mean Absolute Percentage Error, and Coefficient of Determination (R), were considered for model evaluation. The results showed that it improves upon traditional SVM by 99.76% and 99.77% and hybrid DWT-SVM by 39% and 57% in terms of MSE reduction at Dhaka and Chittagong, respectively. R also improved by 49% and 54% over traditional SVM and by 4.40% and 3.16% over hybrid DWT-SVM model. The model well captures the complex nonlinear trend existing in solar radiation; thus, it shows its potential to be applied to other regions for efficient prediction of solar radiation.
先前的研究表明，由于气候数据具有高度非线性和噪声，预测太阳辐射是一个具有挑战性的问题。早期已经应用了各种混合方法来进行太阳辐射预测，这些方法将小波变换与各种机器学习模型相结合。因此，本研究旨在进一步提高这些现有混合模型的性能。为了解决处理非线性和噪声气候模式方面的局限性，本研究提出了一种用于准确预测太阳辐射的多混合模型，该模型结合了霍德里克 - 普雷斯科特滤波器（HP滤波器）、离散小波变换（DWT）和支持向量机（SVM）。从孟加拉国气象部门收集的孟加拉国两个不同地理位置（即达卡和吉大港）的数据被分为三类用于建模：70%用于训练，15%用于验证，15%用于测试，而支持向量机的模型超参数使用粒子群优化算法进行了优化。所提出的方法在分析离散小波变换之前应用霍德里克 - 普雷斯科特滤波器，以增强支持向量机模型捕捉复杂气候模式细节的能力，并使模型更加精确和可靠。模型评估考虑了几个性能指标，如均方误差（MSE）、均方根误差、平均绝对误差、平均绝对百分比误差和决定系数（R）。结果表明，在达卡和吉大港，就均方误差降低而言，该模型分别比传统支持向量机提高了99.76%和99.77%，比混合离散小波变换 - 支持向量机提高了39%和57%。决定系数R也比传统支持向量机提高了49%和54%，比混合离散小波变换 - 支持向量机模型提高了4.40%和3.16%。该模型很好地捕捉了太阳辐射中存在的复杂非线性趋势；因此，它显示出可应用于其他地区进行高效太阳辐射预测的潜力。
