ABSTRACT

Solar radiation represents the most important parameter for sizing and planning solar power systems. However, solar radiation depends significantly on meteorological conditions which are variable and uncontrollable. Therefore, forecasting global solar radiation can play a key role to integrate solar energy resources into the electric grid. This paper presents a new hybrid approach based on seasonal clustering technique and artificial neural network (ANN) for forecasting 1 h-ahead of global solar radiation. For this purpose, the fuzzy c-means algorithm (FCM) was used to cluster 3 years of monthly average experimental data into different seasons according to solar and meteorological parameters of Évora city. Subsequently, based on the seasonal clustering results, the meteorological dataset was divided into dfferent training subsets. Furthermore, for each subset, an ANN model has been designed to forecast hourly global solar radiation. In this study, hourly meteorological data from January 2012 to December 2016 have been used for forecasting. The hourly data were collected from Évora-city’s meteorological station in Portugal (38°34 N, 07°54 W). The results show the superiority of the hybrid approach compared to the individual ANN model according to statistical indicators.

摘要

太阳辐射是确定太阳能发电系统规模和规划的最重要参数。然而，太阳辐射在很大程度上取决于多变且不可控的气象条件。因此，预测全球太阳辐射可以在将太阳能资源整合到电网中发挥关键作用。本文提出了一种基于季节聚类技术和人工神经网络 (ANN) 的新混合方法，用于预测全球太阳辐射提前 1 小时。为此，根据埃武拉市的太阳和气象参数，使用模糊c-means算法（FCM）将3年的月平均实验数据聚类到不同的季节。随后，根据季节聚类结果，将气象数据集划分为不同的训练子集。此外，对于每个子集，设计了一个人工神经网络模型来预测每小时的全球太阳辐射。本研究采用 2012 年 1 月至 2016 年 12 月的每小时气象数据进行预报。每小时数据是从葡萄牙埃武拉市的气象站（38°34 N，07°54 W）收集的。结果表明，根据统计指标，混合方法与单个 ANN 模型相比具有优势。