复杂天气条件下光伏电站太阳辐射量短期预测  

Short-term Prediction of Solar Radiation for a Photovoltaic PowerPlant under Complex Weather Conditions

作　　者：宋晓通 卢艺玮 师芊芊 梅杨[1] SONG Xiao-tong;LU Yi-wei;SHI Qian-qian;MEI Yang(School of Electrical and Control Engineering,North China University of Technology,Beijing 100144,China)

机构地区：[1]北方工业大学电气与控制工程学院,北京100144

出　　处：《科学技术与工程》2024年第30期12985-12995,共11页Science Technology and Engineering

基　　金：北京市自然科学基金(3222051)。

摘　　要：复杂天气条件下,天气变化波动较大;光伏电站传统太阳辐射量预测模型无法很好地处理复杂的非线性关系,存在精度不足的缺陷,给电力系统的保护和并网安全带来了挑战。为了应对这一挑战,建立了一种基于自适应模糊神经网络(adaptive-network-based fuzzy inference systems,ANFIS)的太阳辐射量预测模型。该模型引入了卫星遥感数据作为输入量,以补充传统的气象数据。首先,使用样本熵计算法对复杂天气进行判定;其次,采用自回归移动平均(auto regression integrated moving average,ARIMA)模型,预测未来24 h的云团光学厚度和气溶胶光学厚度这两种关键的卫星遥感数据。结合大气层上界的太阳辐射量和大气平均温度,建立了基于ANFIS的太阳能辐射量预测模型,从而得到未来24 h的太阳能辐射量预测结果。在算例研究中,将ANFIS模型与多层前馈(back propagation,BP)神经网络预测模型、长短期记忆(long short-term memory,LSTM)神经网络预测模型在不同天气类型中的精度进行了对比。结果表明,在简单天气条件下,ANFIS模型、BP模型、LSTM模型的均方根误差分别为0.1122、0.3184、0.2534 W/m^(2),三者相对较小且相差不大;在复杂天气条件中,ANFIS模型的均方根误差为0.8606 W/m^(2),比BP模型和LSTM模型分别降低了4.0396、2.0252 W/m^(2),这说明ANFIS模型在复杂天气条件下表现较好,能够适应具有较强波动性的数据。研究同时表明,在考虑气象数据的基础上计及卫星遥感数据,可将预测的均方根误差降低0.132 W/m^(2),进一步改进了预测精度。Under complex weather conditions,weather fluctuations are significant,and traditional solar radiation prediction models used in photovoltaic power stations cannot effectively handle complex nonlinear relationships.This inaccuracy poses challenges for the protection and grid connection safety of power systems.To address this challenge,a solar radiation prediction model based on ANFIS(adaptive-network-based fuzzy inference systems)was established.This model incorporates satellite remote sensing data as input to complement traditional meteorological data.Firstly,the sample entropy calculation method was used to determine complex weather conditions.Secondly,ARIMA(autoregressive integrated moving average)model was adopted to predict the optical thickness of cloud cover and aerosols,two critical satellite remote sensing data points,for the next 24 hours.Combining solar radiation and temperature data from the upper atmosphere,an ANFIS-based solar radiation prediction model was established to obtain solar radiation predictions for the next 24 hours.In a case study,the accuracy of the ANFIS model was compared to that of BP(back propagation)neural network prediction model and LSTM(long short-term memory)neural network prediction model under different weather conditions.The results show that under simple weather conditions,the root mean square errors of the ANFIS,BP,and LSTM models are 0.1122,0.3184,and 0.2534 W/m^(2),respectively,which are relatively small and not significantly different.However,under complex weather conditions,the root mean square error of the ANFIS model is 0.8606 W/m^(2),which is 4.0396 W/m^(2) and 2.0252 W/m^(2) lower than that of the BP and LSTM models,indicating that the ANFIS model performs better under complex weather conditions and can adapt to data with strong volatility.The study also shows that considering satellite remote sensing data in addition to meteorological data can reduce the predicted root mean square error by 0.132 W/m^(2),further improving prediction accuracy.

关 键 词：复杂天气 太阳辐射量预测 气象卫星数据 自适应模糊神经网络 自回归移动平均模型 

分 类 号：TM615[电气工程—电力系统及自动化]