摘要
对持续法、ARIMA和BP网络三种方法在提前1 h风速预测中的应用进行了研究和比较。为消除季节对预测结果的影响,针对一年12个月分别建立预测模型;认为风速具有不平稳性,应该对原始风速序列进行差分处理;通过对样本数据进行相关性分析来确定神经网络的输入神经元个数;结果表明:大多数情况下,ARIMA模型和BP网络模型的预测结果都好于持续法,并且BP一般都好于ARIMA;但也有持续法好于ARIMA和BP网络模型的情况。不能笼统地说某个方法优于另外一个方法,应该根据具体情况进行分析和判断,选择合适的模型种类,以取得最优预测效果。
The comparison of persistence, ARIMA and BP neural network methods in wind speed prediction application of lh in advance was discussed. To avoid the seasonality influence, models were built respectively for every calendar month. Wind speed was a non - stationary sequence and difference algorithm should be used for original wind speed sequence. The input nerve cells number was decided through relativity analysis for sample data. The result shows that ARIMA model and BP network model are better than persistence method and BP better than ARIMA under most circumstances. But there are examples that persistence method is better than ARIMA and BP models and ARIMA better than BP model. It is improper to say that one method is always superior to another; to choose proper prediction method for the best prediction, detailed analysis should be made.
出处
《华北电力大学学报（自然科学版）》
CAS
北大核心
2008年第3期57-61,共5页
Journal of North China Electric Power University：Natural Science Edition
基金
国家高技术研究发展计划(863计划)资助项目(2007AA05Z428)
