互联网大数据下基于改进FOA–ESN的需求预测研究——以电商和旅游行业为例

With the rapid development of E-business and tourism industries, a large amount of historical data was accumulated. Forecasting the demand precisely and fast is the recent research hotspot and difficulty. The current studies, including causal analysis, time series analysis and artificial intelligence-based approaches, cannot dig useful information from the existing big data with low value density, improve the forecast accuracy and support the quick decision-making. Taking the way to accurately excavate the feature variables related to the demand and the nonlinear mapping relation as the starting point, a new demand forecasting model named FOA-ESN is proposed based on the echo state network (ESN) which has powerful ability of coupling "time parameter” and fault tolerance with a simple training process, and embedded by the novel fruit fly optimization algorithm (FOA) which has the fast search speed and good global search ability. The main contents include: the optimization of ESN’s reservoir topology and neuronal types; the design of effective self-adaptive FOAs combined with backtrack search and differential evolution algorithms; the optimization of key parameters of ESN’s reservoir and readout using improved FOAs; and validation the scientific applicability of the proposed FOA-ESN by typical applications in E-business and tourism industries. This model can be used to mine the value of multiple sources data under big data environment, improve the prediction accuracy with the innovative integration of novel neural network and swarm intelligent optimization techniques, and support the rapid decision-making. This study can enrich the theory research of forecasting and intelligent optimization algorithms, and the research findings can be widely used in E-business and tourism industries, which will result in considerable economic and social benefits.

发展迅猛的电商和旅游行业积累了海量历史数据，准确和快速地预测需求是研究热点/难点。目前研究(因果分析、时间序列和人工智能方法)无法有效挖掘低价值密度大数据中的有用信息、精度欠佳且响应速度慢。本项目以从大数据中精准地挖掘与需求相关的特征变量和非线性映射关系为切入点，基于耦合时间参数能力强/容错性能佳/训练简单的回声状态网络(ESN)，嵌入搜索快和全局优化能力强的果蝇优化算法(FOA)，构建FOA-ESN需求预测新模型，包括：优化ESN储备池拓扑结构和神经元类型；设计融合回溯搜索/差分进化算法的自适应FOA，利用改进FOA优化储备池关键参数和读出网络；密切结合电商和旅游行业验证模型的科学实用性。该模型可挖掘大数据环境下多种来源数据价值，创新性集成新颖的神经网络和群体智能优化技术之优点，提升预测精度并支持快速决策，丰富预测和优化算法理论。成果可广泛应用于电商和旅游行业，并产生可观经济和社会效益。

发展迅猛的电商和旅游行业积累了海量历史数据，准确和快速地预测需求是研究难点。本项目从大数据中精准地挖掘与需求相关的特征变量和非线性映射关系，基于耦合时间参数能力强/容错性能佳/训练简单的回声状态网络(ESN)，嵌入搜索快和全局优化能力强的果蝇优化算法(FOA)，构建等多种FOA-ESN需求预测新模型，主要工作包括：对果蝇优化算法进行改进；明确ESN相关参数的优化机理，基于FOA对ESN储备池拓扑结构进行优化，搜索ESN储备池关键参数和优化读出网络；设计多种高质量的旅游需求预测模型；设计多种复杂数据环境下的基于改进神经网络的预测模型；密切结合典型企业和行业验证模型的科学实用性。该类模型可挖掘大数据环境下多种来源数据价值，创新性集成新颖的神经网络和群体智能优化技术之优点，提升预测精度，丰富预测和优化算法理论。研究成果可广泛应用于电商和旅游行业，并产生可观经济和社会效益。目前发表SCI和SSCI期刊论文18篇、中文核心论文1篇，依托项目毕业博士生4人和硕士生5人。