基于数据挖掘的长短期记忆网络模型油井产量预测方法
|
摘要
传统的BP神经网络及其改进算法广泛应用于产量预测,但并不适宜时间序列预测问题。基于产油量变化的时间序列特征,提出利用长短期记忆网络(LSTM)深度学习模型实现具有长期记忆能力的时间序列预测,在描述LSTM神经网络的基本结构和算法原理基础上,阐述了样本数据处理,输入层、隐藏层和输出层节点数选择及表征方式,形成产量预测模型。实例应用表明,LSTM模型可以准确预测油井产量,整体平均误差约为1. 46%,并指出无预兆停产、特殊情况以及部分数据量缺失是影响预测准确性的主要原因。该模型的提出对于大数据和深度学习在石油方面的应用研究具有重要意义。
Traditional BP neural networks and their improved algorithms have been widely used in production forecast,which are certainly limited in the time series forecast. Based on the time series performance of oil production history,LSTM( long-short term memory) deep learning model is proposed to forecast the time series with long-term memory. According to the basic structure and algorithm principle of LSTM neural network,the sample data processing,input layer,hidden layer and output layer node number optimization and characterization were provided to establish this production forecast model. Field application indicates that this LSTM model can accurately forecast the oil well production with an integral average error of about 1. 46%. It also pointed out that the absence of omissions,special circumstances and partial data lack are considered as the major factors for production forecast accuracy. This proposed model could provide certain significance for the application of big data and deep learning in petroleum industry.
Traditional BP neural networks and their improved algorithms have been widely used in production forecast,which are certainly limited in the time series forecast. Based on the time series performance of oil production history,LSTM( long-short term memory) deep learning model is proposed to forecast the time series with long-term memory. According to the basic structure and algorithm principle of LSTM neural network,the sample data processing,input layer,hidden layer and output layer node number optimization and characterization were provided to establish this production forecast model. Field application indicates that this LSTM model can accurately forecast the oil well production with an integral average error of about 1. 46%. It also pointed out that the absence of omissions,special circumstances and partial data lack are considered as the major factors for production forecast accuracy. This proposed model could provide certain significance for the application of big data and deep learning in petroleum industry.
引文
[1]姚军.油藏工程原理与方法[M].东营:中国石油大学出版社,2016:200-257.
[2]樊灵,赵孟孟,殷川,等.基于BP神经网络的油田生产动态分析方法[J].断块油气田,2013,20(2):204-206.
[3]计秉玉.油田开发指标预测方法综述[J].大庆石油地质与开发,1999,18(2):19-22.
[4]周彩兰,刘敏.BP神经网络在石油产量预测中的应用[J].武汉理工大学学报,2009,31(3):125-129.
[5]程宇也.基于人工神经网络的短期电力负荷预测研究[D].杭州:浙江大学,2017.
[6]马林茂,李德富,郭海湘,等.基于遗传算法优化BP神经网络在原油产量预测中的应用:以大庆油田BED试验区为例[J].数学的实践与认识,2015,45(24):117-128.
[7]董玉安,赵蕊,叶欢,等.遗传算法与神经网络结合预测低渗透油藏产油量---以宝浪油田为例[J].石油地质与工程,2011,25(5):60-62,66.
[8]陈巧红,孙超红,余仕敏,等.基于递归神经网络的广告点击率预估研究[J].浙江理工大学学报,2016,35(6):880-885.
[9]李春生,谭民浠,张可佳.基于改进型BP神经网络的油井产量预测研究[J].科学技术与工程,2011,11(31):7766-7769.
[10]李达.基于时间序列分析方法的油田产量预测与应用[D].兰州:兰州理工大学,2018.
[11]宋成坤.油田海量数据挖掘技术研究及应用[D].大庆:东北石油大学,2017.
[12]陈亮,王震,王刚.深度学习框架下LSTM网络在短期电力负荷预测中的应用[J].电力信息与通信技术,2017,15(5):8-11.
[13]薛茹.数据挖掘技术在油田中的应用[J].微型电脑应用,2018,34(5):26-28.
[14]谈苗苗.基于LSTM和灰色模型集成的短期交通流预测[D].南京:南京邮电大学,2017.
[15]孙瑞奇.基于LSTM神经网络的美股股指价格趋势预测模型的研究[D].北京:首都经济贸易大学,2016.
[16]周瑞,魏正英,张育斌,等.基于LSTM递归神经网络的番茄目标产量时间序列预测[J].节水灌溉,2018,32(8):66-70.
[17]陈卓,孙龙祥.基于深度学习LSTM网络的短期电力负荷预测方法[J].电子技术,2018,47(1):39-41.
[18]徐鹏飞,李炜,郑华,等.神经网络在时间序列预测中的应用研究[J].电子技术,2010,47(8):5-7.
[19]王鑫,吴际,刘超,等.基于LSTM循环神经网络的故障时间序列预测[J].北京航空航天大学学报,2018,44(4):772-784.
[2]樊灵,赵孟孟,殷川,等.基于BP神经网络的油田生产动态分析方法[J].断块油气田,2013,20(2):204-206.
[3]计秉玉.油田开发指标预测方法综述[J].大庆石油地质与开发,1999,18(2):19-22.
[4]周彩兰,刘敏.BP神经网络在石油产量预测中的应用[J].武汉理工大学学报,2009,31(3):125-129.
[5]程宇也.基于人工神经网络的短期电力负荷预测研究[D].杭州:浙江大学,2017.
[6]马林茂,李德富,郭海湘,等.基于遗传算法优化BP神经网络在原油产量预测中的应用:以大庆油田BED试验区为例[J].数学的实践与认识,2015,45(24):117-128.
[7]董玉安,赵蕊,叶欢,等.遗传算法与神经网络结合预测低渗透油藏产油量---以宝浪油田为例[J].石油地质与工程,2011,25(5):60-62,66.
[8]陈巧红,孙超红,余仕敏,等.基于递归神经网络的广告点击率预估研究[J].浙江理工大学学报,2016,35(6):880-885.
[9]李春生,谭民浠,张可佳.基于改进型BP神经网络的油井产量预测研究[J].科学技术与工程,2011,11(31):7766-7769.
[10]李达.基于时间序列分析方法的油田产量预测与应用[D].兰州:兰州理工大学,2018.
[11]宋成坤.油田海量数据挖掘技术研究及应用[D].大庆:东北石油大学,2017.
[12]陈亮,王震,王刚.深度学习框架下LSTM网络在短期电力负荷预测中的应用[J].电力信息与通信技术,2017,15(5):8-11.
[13]薛茹.数据挖掘技术在油田中的应用[J].微型电脑应用,2018,34(5):26-28.
[14]谈苗苗.基于LSTM和灰色模型集成的短期交通流预测[D].南京:南京邮电大学,2017.
[15]孙瑞奇.基于LSTM神经网络的美股股指价格趋势预测模型的研究[D].北京:首都经济贸易大学,2016.
[16]周瑞,魏正英,张育斌,等.基于LSTM递归神经网络的番茄目标产量时间序列预测[J].节水灌溉,2018,32(8):66-70.
[17]陈卓,孙龙祥.基于深度学习LSTM网络的短期电力负荷预测方法[J].电子技术,2018,47(1):39-41.
[18]徐鹏飞,李炜,郑华,等.神经网络在时间序列预测中的应用研究[J].电子技术,2010,47(8):5-7.
[19]王鑫,吴际,刘超,等.基于LSTM循环神经网络的故障时间序列预测[J].北京航空航天大学学报,2018,44(4):772-784.