Solar radiation prediction using improved soft computing models for semi-arid, slightly-arid and humid climates

Alexandria Engineering Journal

Hojat Karami¹  Kwok-wing Chau¹  Hailong Huang²  Mohammad Ehteram³  Qian Zhang³  Shahab S. Band⁴ 

[1] Corresponding authors.;College of Computer Science and Artificial Intelligence, Wenzhou University, Wenzhou, China;Department of Water Engineering and Hydraulic Structures, Faculty of Civil Engineering, Semnan University, Semnan, Iran;Future Technology Research Center, College of Future, National Yunlin University of Science and Technology, 123 University Road, Section 3, Douliou, Yunlin 64002, Taiwan, ROC;

关键词: Solar radiation;    Grasshopper optimization algorithm;    Semi-dry climate;    Dry climate;   

DOI  :  

来源: DOAJ

【 摘 要 】

In this research, monthly solar radiation is predicted in semi-dry, dry, and wet climates. Adaptive neurofuzzy interface system (ANFIS), radial basis function neural network (RBFNN), and multi-layer perceptron (MLP) models are used for predicting solar radiation (SR). Grasshopper algorithm (GOA) is utilized to improve the performance of ANFIS, RBFNN, and MLP models. Three stations in Iran, namely Rasht (humid climate), Yazd (semi-arid) and Tehran (slightly arid), are considered as the case studies. The accuracy of GOA is benchmarked against particle swarm optimization (PSO) and salp swarm algorithm (SSA). The results reveal that the best-input combination is relative humidity, wind speed, rainfall, and temperature at these three stations. A comprehensive study is performed to select the best-input combination. The main contribution of paper is to create new hybrid ANFIS models for predicting monthly solar radiation in different climates. Besides, effects of different parameters are comprehensively investigated on solar radiation. This study indicates that temperature is the most effective parameter for estimating SR in dry and semi-dry climate. It is found that rainfall plays a key role for estimating SR in a wet region. The main finding of this paper is that the determination of the most suitable input scenario for predicting SR is an important issue because different input scenarios in different climates provide different performances. Besides, the use of a robust optimization algorithm as a training method is a significant step of the modeling process of SR. Results indicate that mean absolute error (MAE) of ANFIS-GOA is 3.8% and 8.9% less in comparison with that of MLP-GOA and RBFNN-GOA, respectively during the training stage at Yazd station. Besides, MAE of ANFIS-GOA is 26% and 31% less than that of MLP-GOA and RBFNN-GOA, respectively during the training stage at Tehran station. Results indicate that NSE values of ANFIS-GOA, ANFIS-SSA, ANFIS-PSO, and ANFIS models are 0.94, 0.88, 0.84, and 0.79, respectively during the testing stage at Rasht station. It is found that ANFIS-GOA attains higher accuracy in predicting SR under different climates.

【 授权许可】

Unknown