光伏电站发电功率短期预测研究

发布时间：2018-03-15 11:09

本文选题：光伏功率预测　切入点：PLS　出处：《东北电力大学》2017年硕士论文　论文类型：学位论文

【摘要】：光伏发电是新兴的产业发电技术,备受青睐。但是由于光伏发电系统的输出受到太阳辐照强度和天气因素的影响,使得光伏发电系统在输出的时候有较大的不稳定性,事实上光伏发电其实是一种非平稳的过程带有一定的随即性。正是这种性质,会造成光伏发电接入电网后对整个大电网产生冲击影响。何时做出何种电网调度,是减少冲击影响的关键。所以准确预测光伏发电量,成为许多国内外学者要研究的问题。课题就光伏电站发电功率短期的预测做了深入研究。首先,获取某光伏电站逆变器上的发电功率数据,进行数据分析,指出天气类型和温度对光伏发电功率有影响。并且,根据数据,对天气类型和温度做出相关性分析,得出各自的相关系数。天气类型与发电功率程正相关,且基本处于高度相关程度。由此使用把天气类型映射为天气类指数方法。温度与发电功率程负相关,且把最高温度、最低温度、平均温度相关系数做对比,最终确定最高温度与最低温度为温度的影响因素。建立了PLS、RF、SVM、退火优化SVM四种预测模型。四种模型对所有的样本进行了预测。PLS预测模型中,晴天、晴转多云、多云、多云转晴、雨五种天气类型预测平均准确率分别为94.2%、86.3%、81.6%、81.7%、73.6%。RF预测模型中,晴天、晴转多云、多云、多云转晴、雨五种天气类型预测平均准确率分别为93.5%、86.1%、82.3%、83.2%、75.3%。SVM预测模型中,晴天、晴转多云、多云、多云转晴、雨五种天气类型预测平均准确率分别为94.6%、87.9%、84.3%、85.7%、75.9%。退火优化SVM预测模型中,晴天、晴转多云、多云、多云转晴、雨五种天气类型预测平均准确率分别为94.8%、90.8%、86.7%、87.4%、79.1%。PLS模型属于多元回归模型,其模型结构相对简单,程序操作方便。RF模型、SVM模型、退火优化SVM模型具有机器学习能力,无论哪种天气类型,退火优化SVM模型相对比另外两种机器学习模型,都有较高的准确率,SVM模型次之。具有机器学习能力的预测模型,在晴转多云、多云、多云转晴等天气波动情况下,有一定的抗干扰能力,但需要一定的程序运行时间。在实际中,根据不同需求选择不同的预测模型。
[Abstract]:Photovoltaic power generation is a new industrial power generation technology, which is very popular. However, because the output of photovoltaic power generation system is affected by solar radiation intensity and weather factors, photovoltaic power generation system has greater instability in output. As a matter of fact, photovoltaic power generation is actually a non-stationary process with a certain degree of randomness. It is precisely this kind of property that will cause the impact of photovoltaic power generation on the whole large power grid. When and what kind of grid dispatch will be made. It is the key to reduce the impact of impact. Therefore, accurate prediction of photovoltaic power generation has become a problem to be studied by many scholars at home and abroad. The data of power generation on an inverter of a photovoltaic power plant are obtained, and the data are analyzed, and it is pointed out that the weather type and temperature have an effect on the power of photovoltaic power generation. In addition, according to the data, the correlation between weather type and temperature is analyzed. The correlation coefficient is obtained. The weather type is positively correlated with the generation power process, and is basically in a high degree of correlation. Therefore, the method of mapping weather type to synoptic index is used. The temperature is negatively correlated with the generation power range, and the maximum temperature, The correlation coefficient of minimum temperature and average temperature is compared, and the maximum temperature and the lowest temperature are determined as the influencing factors of temperature. Four prediction models of SVM are established, which are optimized by annealing, and all samples are predicted by four models. The average accuracy of forecasting the five weather types of sunny, sunny to cloudy, cloudy to cloudy, cloudy to sunny, and rain is 94.22 / 86.3s, respectively. The accuracy of forecasting the five weather types is respectively 94.22 / 86.3and 81.6 / 81.6 / 81.7/ 73.6. in the RF forecasting model, sunny, sunny to cloudy, cloudy to cloudy, cloudy to sunny, cloudy to sunny, The average accuracy of forecast for the five types of rain was 93.5, 86.1and 82.3s, respectively. In the prediction model of the five weather types, sunny, sunny to cloudy, cloudy to cloudy, cloudy to sunny, and rain, the average accuracy of forecast of five weather types was 94.60.84.35.70.The average accuracy of SVM model was optimized by annealing, sunny weather, sunny to cloudy, sunny to cloudy, and the average accuracy of forecast was 75.90.In the SVM prediction model, the average accuracy was 94.6% 84.37.70.In the SVM prediction model, the sunny weather, sunny weather, sunny to cloudy, sunny to cloudy, were 94.6%, 87.9% and 85.9%, respectively. The average prediction accuracy of the five weather types of cloudy, cloudy to sunny and rainy is 94.80.88 and 86.77.40.PLS models belong to the multivariate regression model. The model structure is relatively simple, the program operation is convenient, the RF model has the SVM model, and the annealing optimization SVM model has the ability of machine learning. Regardless of weather type, annealing optimized SVM model has higher accuracy than other two machine learning models. In the case of cloudy to sunny weather fluctuation, it has certain anti-interference ability, but it needs certain program running time. In practice, different prediction models are selected according to different demand.
【学位授予单位】：东北电力大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM615

【参考文献】