检索结果-南通市图书馆

Multi-stage Attention Mechanism based Deep Learning for Wind Power Forecasting

第43届中国控制会议

作者： Jingdi Wang Yundong Cao Huaikang Jin shouyuan fan Bin Wang Huaiping Jin Faculty of Information Engineering and Automation Kunming University of Science and Technology Yunnan Key Laboratory of Artificial Intelligence Kunming University of Science and Technology Huaneng Renewables Co. Ltd.Yunnan Branch China Energy Engineering Group Yunnan Electric Power Design Institute

With the continuous advancement of new energy technologies,wind power is becoming an increasingly crucial source of clean *** to the stochastic and fluctuating nature of wind power data,there is a significant challenge known as concept drift,where the data distribution changes over *** poses a substantial obstacle to accurate prediction,leading to performance degradation of *** address this issue,an adaptive deep-learning wind power prediction method combining transfer learning and concept drift detection is *** method incorporates the concept of online transfer learning,transferring offline-trained model parameters to adapt quickly to new data distributions during the online ***,a combination of an online moving window and Kullback-Leibler(KL) divergence is employed to capture and analyze changes in the online predictor's output layer network *** facilitates a more precise detection of concept drift in wind power prediction,determining whether model updates are *** dynamically updating model parameters and effectively handling concept drift,the approach can maintain high accuracy and stability in wind power *** effectiveness and superiority of the proposed method are validated through the analysis of real-world wind power field data.

关键词： Wind Power Forecasting Concept Drift Transfer Learning Online Updating Gated Recurrent Unit

来源： cnki会议评论

在线全文

cnki会议

学校读者我要写书评

暂无评论

Multi-stage Attention Mechanism based Deep Learning for Wind...

第43届中国控制会议

作者： Guanzhi Yang Yundong Cao Huaikang Jin shouyuan fan Bin Wang Huaiping Jin Faculty of Information Engineering and Automation Kunming University of Science and Technology Yunnan Key Laboratory of Artificial Intelligence Kunming University of Science and Technology Huaneng Renewables Co. Ltd.Yunnan Branch China Energy Engineering Group Yunnan Electric Power Design Institute

The accurate prediction of wind power is of great significance for the stable operation of the power ***,due to the inherent intermittency and randomness of wind speed,existing time series prediction models struggle to capture the complex nonlinear relationship between wind speed and wind power,resulting in low prediction ***,this study proposes a multi-stage attention long short-term memory network(MA-LSTM) prediction ***,convolution and attention mechanisms are fused to form the first stage of attention,extracting multi-dimensional features of dynamic changes in wind power ***,the feature vectors with spatial importance weights are input into the LSTM network for temporal feature ***,a second stage attention mechanism assigns different weights to hidden state features at various time ***,to counteract the low quality of the second stage attention extraction hidden state caused by the randomness of wind power data,a random difference structure is employed,forming multi-stage attention and LSTM units in parallel to develop the proposed *** proposed method is evaluated using a real wind farm dataset,and the experimental results show that the MA-LSTM model can better capture the spatiotemporal characteristics of historical wind power data and provide higher quality time step hidden features,thereby significantly improving the accuracy of wind power prediction.

关键词： Wind power prediction Long short-term memory network Attention mechanism Hidden state quality Spatial and temporal correlation

来源： cnki会议评论

在线全文

cnki会议

学校读者我要写书评

暂无评论

在线选择性集成即时学习风电功率自适应预测

太阳能学报 2024年第10期45卷 487-496页

作者：李运龙金怀平范守元金怀康王彬昆明理工大学信息工程与自动化学院昆明650500 中国能源建设集团云南省电力设计院有限公司昆明650051 华能新能源股份有限公司云南分公司昆明650000

风电功率预测可为风电的并网和优化调度提供有效的指导信息,在风能的开发利用中扮演着举足轻重的角色。然而,由于风电固有的间歇性和随机性给准确的风电功率预测带来巨大挑战。同时,由于受季节性、气候性、设备老化等因素影响,随着时间... 详细信息

风电功率预测可为风电的并网和优化调度提供有效的指导信息,在风能的开发利用中扮演着举足轻重的角色。然而,由于风电固有的间歇性和随机性给准确的风电功率预测带来巨大挑战。同时,由于受季节性、气候性、设备老化等因素影响,随着时间的推移,风电功率数据特征难免发生改变,这将直接导致离线的风电功率模型性能发生退化。为此,提出一种在线选择性集成即时学习(OSEJIT)自适应风电功率预测方法。首先,为了有效处理风电的非线性和时变性特征,通过相似度、学习器扰动以构建多样性JIT基模型库。其次,为了保证集成有效性,定义基于Friedman检验的多样性指标和基于预测精度的准确性指标以实现模型的在线选择。随后,在线预测阶段,根据模型近期的预测性能通过自适应加权集成的方式获得最终预测值。为了保证基模型库的更新,同时规避模型频繁重建导致计算资源耗费的问题,采用一种基于KL散度的过程状态识别方法以减少模型重建频率。所提方法的有效性和优越性在一个实际风电功率数据应用中获得了验证。

关键词：风电功率预测自适应算法过程状态识别统计假设检验在线选择性集成即时学习

同方期刊数据库博看期刊评论

在线全文

学校读者我要写书评

暂无评论

RFID标签中ACA固化的动态DSC研究

电子元件与材料 2013年第12期32卷 59-63页

作者：范守元陈建魁尹周平华中科技大学数字制造装备与技术国家重点实验室

采用动态差示扫描量热仪(DSC)对RFID标签制造工艺中所用各向异性导电胶(ACA)的固化反应动力学进行了研究。分别通过常用的n级反应模型法和自催化模型法求解了固化反应动力学方程。结果表明,n级反应模型和自催化模型忽略了ACA固化反应的... 详细信息

采用动态差示扫描量热仪(DSC)对RFID标签制造工艺中所用各向异性导电胶(ACA)的固化反应动力学进行了研究。分别通过常用的n级反应模型法和自催化模型法求解了固化反应动力学方程。结果表明,n级反应模型和自催化模型忽略了ACA固化反应的复杂性,计算曲线均与实验值存在较大偏差,不能很好地用于描述ACA的固化过程。采用非模型动力学计算该固化体系的反应活化能,结果表明反应活化能是固化度的函数,并不是一个定值。当较低的升温速率用于计算活化能曲线时,则在固化度高于80%时活化能有明显的增大。

关键词：各向异性导电胶固化反应 n级动力学模型自催化模型非模型动力学动态DSC

Cloning of HumCyr61 gene expressing down-regulatedly in Rhabdomyosarcoma

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Chinese Science Bulletin 1998年第10期43卷 878-879页

作者： Anding Bi Long Yu Ming Zhang Xiyuan Sun Yuxin fan shouyuan Zhao State Key Laboratory of Genetic Engineering Institute of Genetics Fudan University Shanghai China

Reports on the cloning of HumCyr61 gene expressing down-regulatedly in Rhabdomyosarcoma, a malignant tumor of the skeletal muscle. Histological categories; Use of homologous amplification by polymerase chain reaction;... 详细信息

关键词： gene expressing down-regulatedly in Rhabdomyosarcoma down Cloning of HumCyr61

Molecular cloning and expression analysis of a novel human cDNA fragment encoding a putative Ser/Thr protein kinase

在线全文

学校读者我要写书评

暂无评论

Chinese Science Bulletin 1999年第9期44卷 778-783页

作者： Qi Zhang fangyan Dai Min Zhang Peirong Hu Qiang Fu Yuxin fan Long Yu shouyuan Zhao Fudan Univ Inst Genet State Key Lab Genet Engn Shanghai 200433 Peoples R China

A novel full-length cDNA encoding a putative serine/threonine kinase has been isolated from a human testis cDNA library. A nucleotide sequence of 1 225 bp length has been determined containing an open reading frame of 1 044 nucleotides (encoding 348 amino acids). In view of its degree of homology to members of the Ser/Thr protein kinase family and the closest relationship to Mus musculus STK-1, the predicted product was designated by the name of HUMSTK-1. Its mRNA is present in large amounts in thymus, and small amounts in testis,small intestine and colon.

关键词： serine/threonine kinase cDNA cell cycle homology.

Isolation and identification of a novel STR(D17S2204)in the vicinity of NF1 locus

在线全文

学校读者我要写书评

暂无评论

Chinese Science Bulletin 1999年第7期44卷 627-631页

作者： Xiyuan Sun Long Yu Qiping Zheng Yurong Xin Liemin Zhou Yuxin fan Ying Jiang Jiahui Xia Zhuolin Liu shouyuan Zhao Fudan Univ Inst Genet State Key Lab Genet Engn Shanghai 200433 Peoples R China Sun Yatsen Univ Med Sci Affiliated Hosp 1 Dept Neurol Guangzhou 510089 Peoples R China Hunan Med Univ State Key Lab Med Genet Changsha 410078 Peoples R China

Using a synthetic oligonucleotide （dA-dC）15 as a probe to screen a probe pool made by microdissected human chromosome band 17q11-q12, a DNA fragment containing （CA）16 was isolated, which is a novel short tandem repeat （STR） determined by searching in the GenBank and GDB. It has 8 allelic types with a PIC value of 0.61. The novel STR is conformed with Mendelian segregation according to the linkage analysis of a three-generation neurofibromatosis type Ⅰ （NF1） pedigree. The STR was localized at chromosome band 17q11-q12 in the vicinity of NF1 gene by FISH analysis. The accession number of the STR in GenBank and GDB is G32112 and D17S2204 respectively.

关键词： chromosome 17q11-q12 short tandem repeats (STR) D17S2204 neurofibromatosis type Ⅰ (NF1).

在线全文

学校读者我要写书评

暂无评论

用PCR快速筛选递次截短的DNA克隆

中华医学遗传学杂志 1998年第1期15卷 46-48页

作者：胡培蓉余龙张民范玉新王小柯赵寿元复旦大学遗传学研究所

目的为了建立一种快速、简便而准确地筛选递次截短的ＤＮＡ克隆的技术，加速对基因全长ｃＤＮＡ或ＤＮＡ大片段的测序和鉴定。方法在载体多克隆位点两侧设计引物，直接ＰＣＲ扩增插入片段，以此为基础选择系列缺失亚克隆，并与常规酶切... 详细信息

目的为了建立一种快速、简便而准确地筛选递次截短的ＤＮＡ克隆的技术，加速对基因全长ｃＤＮＡ或ＤＮＡ大片段的测序和鉴定。方法在载体多克隆位点两侧设计引物，直接ＰＣＲ扩增插入片段，以此为基础选择系列缺失亚克隆，并与常规酶切法作比较。结果根据ＰＣＲ产物大小排序，准确地鉴定出系列缺失亚克隆，并可将ＰＣＲ产物直接用于ＰＣＲ循环测序，比酶切法分辨率高，操作步骤简化，筛选时间短。结果所建立的ＰＣＲ快速筛选法不仅是一种行之有效的鉴定方法，而且可以显著地节省时间和实验材料。该方法也可适用于其它各种质粒载体。

关键词： DNA克隆快速筛选递次截短分析聚合酶链反应

在线全文

学校读者我要写书评

暂无评论

北京地区早园竹竹鞭生长规律研究

世界竹藤通讯 2015年第3期13卷 21-25页

作者：范卓敏赵钰王寿远北京市紫竹院公园管理处北京100048

对北京地区早园竹竹鞭的生长规律进行了调查研究。结果表明：早园竹竹鞭生长量总体上随着竹鞭年龄的增大而减小;在一条完整的鞭系上,空芽数所占比例最大,笋芽所占比例最小,空芽所占比例随着鞭龄增大而增加;早园竹竹鞭主要分布在0~10 cm... 详细信息

对北京地区早园竹竹鞭的生长规律进行了调查研究。结果表明：早园竹竹鞭生长量总体上随着竹鞭年龄的增大而减小;在一条完整的鞭系上,空芽数所占比例最大,笋芽所占比例最小,空芽所占比例随着鞭龄增大而增加;早园竹竹鞭主要分布在0~10 cm的表土层中,随着土壤深度增加,竹鞭分布逐渐减少;竹林开花后,竹鞭生长量明显减小,且竹鞭仅分布于0~10 cm的土层中;开花竹林土壤中的有效磷含量明显降低。

关键词：早园竹竹鞭生长规律北京