基于加速项和优化解的增量型极限学习机算法及其应用研究

The project aims at the major of theories and methods of incremental extreme learning machine, traffic flow data of forecasting model under the new generation artificial intelligence. It mainly focuses on several key scientific problems existing in traffic flow data of forecasting model and parameter calculation of hidden layer nodes, network architecture optimization, output weights estimation for incremental extreme learning machine. In this project, the corresponding theories and methods of optimal parameter calculation of hidden layer nodes, network architecture optimization, output weights estimation for incremental extreme learning machine, and the method of traffic flow data of forecasting model based on variational mode decomposition and incremental extreme learning machine will be studied, and further a set of.theories, methods and technical for incremental extreme learning machine based on optimal parameter of hidden layer nodes, accelerating amount, optimization solution and least square robust estimation will be built, and the method of traffic flow data of forecasting model based on variational mode decomposition and incremental extreme learning machine will be proposed. On this basis, an intelligent prediction system for rail transit network will be develop, and then traffic flow intelligent forecasting will be verified so as to enhance the intelligence and efficiency of traffic flow based on rail transit network. The research results of this project are of great scientific and practical value for the development of intelligent transportation driven by new artificial intelligence.

本项目针对新一代人工智能背景下增量型极限学习机理论方法和交通流数据预测建模面临的挑战，主要围绕增量型极限学习机隐含层节点参数计算、网络结构优化、输出权值估计和交通流数据预测建模等关键科学问题，研究增量型极限学习机隐含层节点参数计算方法、增量型极限学习机网络结构优化理论与方法、增量型极限学习机输出权值估计理论与方法、基于变分模态分解与增量型极限学习机的交通流数据预测建模方法，建立基于最优隐含层节点参数、加速项、优化解、最小二乘稳健估计的增量型极限学习机理论方法与技术体系，提出基于变分模态分解与增量型极限学习机的交通流数据预测建模方法。在此基础上研制面向轨道交通网络的智能预测系统，面向轨道交通网络的交通流智能预测应用开展应用验证，提升面向轨道交通网络的交通流智能化程度和效率。该项目研究成果对于新一代人工智能驱动的智能交通的发展具有重要科学意义和应用价值。

申请人以增量型极限学习机算法与应用的关键问题为研究目标，深入开展增量型极限学习机基础理论与方法研究。在增量型极限学习机中，隐含层节点参数随机化、网络结构冗余、输出权值计算复杂，以及对交通流数据预测建模，具有很大难度。.针对这些问题，本项目开展了隐含层节点参数计算、网络结构优化、输出权值计算、交通流数据预测建模等问题的研究，取得了丰富的成果。在研究成果方面，针对增量型极限学习机算法与应用研究得到系列成果，以第一或通讯作者身份发表SCI期刊论文3篇，EI期刊论文2篇，EI会议论文1篇，申请发明专利5项(授权4项，受理1项)。这些发表的论文均和本项目紧密相关，都标注了得到国家青年科学基金的资助，研究成果见列表。在人才培养方面，培养博士研究生2名，硕士研究生2名。