基于GA-BP神经网络的逐时总辐射分组模型研究

通过分析影响太阳辐射的主要因素,提出以太阳高度角、季节和天气（晴空指数）作为数据划分依据的分组模型建立方法。以拉萨和西安地区的逐时气象数据和辐射数据为例,基于遗传算法（genetic algorithm,GA）优化的BP神经网络,建立太阳高度角、季节和天气类型的逐时总辐射分组模型。该研究揭示分组模型误差变化的规律,并将其估算误差与AllData模型比较。结果显示,相较于AllData模型,分组模型的估算误差均有降低。其中,天气分组模型误差最小,且西安的天气分组模型结果优于拉萨。西安天气分组模型平均绝对百分比误差（MAPE）和相对均方根误差（rRMSE）相较AllData模型结果分别下降3.96%和4.18%。研究结果表明分组模型能够降低逐时总辐射估算误差,可为估算逐时总辐射提供方法借鉴。

RESEARCH ON GROUPING MODEL OF HOURLY GLOBAL SOLAR RADIATION BASED ON GA-BP NEURAL NETWORK

By analyzing the main factors affecting solar radiation, this paper proposes a grouping modeling method, which uses the solar altitude angle, season and clearness index as the indicators of data classification. The hourly meteorological data and radiation data for Lhasa and Xi'an are selected as examples. The hourly global radiation grouping models for solar altitude, seasons and weather are established respectively through BP neural network optimized by genetic algorithm （GA）. The error variation rules of grouping model are revealed. The error of grouping models are compared with that of the AllData model further. The results show that the estimation error of the grouping model is reduced compared with the AllData model, and the estimation error of the weather grouping model is the smallest. The results of Xi'an's weather grouping model are better than those of Lhasa. Compared with the AllData model, the mean absolute percentage error （MAPE） and relative root mean square error （rRMSE） of Xi'an's weather grouping model is decreased by 3.96% and 4.18%, respectively. The results show that the grouping model can reduce the estimation error of hourly global solar radiation, and provide a reference for neural network as a method for estimating hourly global solar radiation.