兰州大学机构库 >数学与统计学院
一种新型动态人工鲸鱼算法的研究及在时间序列预测中的应用
Alternative TitleA Novel Dynamic Artificial Whale Algorithm Research and Application in Time Series Forecasting
董昀轩
Thesis Advisor王建州
2018-05-01
Degree Grantor兰州大学
Place of Conferral兰州
Degree Name硕士
Keyword特征识别 人工鲸鱼算法 时间序列预测 部分连接神经网络
Abstract

随着全球经济逐渐复苏和物联网的快速发展,计算机网络技术迎来了日新月异的高速进步,经济体中可供收集的信息呈现爆炸性增长,汹涌澎湃的大数据浪潮正迎面袭来。因此,如何通过优化算法挖掘海量数据中的有用信息,进而对数据进行准确预测以提升产业效率和增加经济收益,这正是众多研究者所关注和追求的。基于优化算法的时间序列预测是从预测指标的时间序列中找出演变模式或内在关联,然后对指标的未来发展趋势做出定量估计。准确有效的提前预测,不仅有助于合理安排生活生产计划,有效解决资源分布等实际问题,还有助于管理者做出正确决策,制定可持续发展计划政策,提高整个社会的经济效益。

近年来,国内外学者进行了大量混合预测方法的相关研究。基于传统群智能优化的方法虽简单且易实施,但却很难精确刻画数据的非线性特征,不能获得令人满意的预测精度。基于人工神经网络的预测方法,学习能力强且计算速度快,解决了很多传统统计方法难以处理的问题。然而,人工神经网络容易出现过拟合现象并且单个预测模型稳定性较差,使得这一方法在时间序列预测中具有一定的局限性。相比之下,将不同的预测方法有机融合后构成的组合预测方法,不仅综合了各种单一预测方法的优势,而且显著提高了整个模型的预测精度,其效果往往优于单一的预测模型。

但是,群智能优化策略与预测模型之间的结合一直是一个困难且具有挑战性的问题。随着实际问题趋于复杂化和规模化,基于严格数学原理的传统优化方法在实施预测时显得无能为力。本文在人工智能的框架内,将鲸鱼自治体引入到优化问题中,发展了一种混合优化问题与预测模型的新模式:人工鲸鱼模式,并由此产生了一种高效的智能优化算法:人工鲸鱼优化算法。与此同时,本文给出人工鲸鱼算法在多目标寻优与局部连接神经网络训练中的两个应用实例;最后,本文讨论了人工鲸鱼的行为模式和人工鲸鱼算法的发展方向。本文的主要贡献如下:MSI指标来度量模型的稳定性,实验结果表明,通过结合人工鲸鱼算法与传统算法的方式构造出的预测模型具有很好的稳定性和良好的预测精度。

Other Abstract

With the gradual recovery of the global economy and the rapid development of the Internet of Things, computer network technology has ushered in rapid progress of each passing day, and the information available for collection for the economy has shown an explosive growth. The surging wave of big data is facing the onset. Therefore, how to use the optimization algorithm to mine useful information about the massive data , and then accurately predict the data to improve industrial efficiency and increase economic returns , which is what many researchers are concerned about and pursue. Time series prediction based on the optimization algorithm is to find the evolution pattern or internal correlation from the time series of the forecasting indicators, and then make a quantitative estimation of the future development trend of the indicators. Accurate and effective forecasting in advance not only helps to rationalize life production plans , effectively solve practical problems such as resource distribution, but also helps managers make correct decisions, formulate sustainable development plans and policies , and improve the economic benefits of the entire society.

In recent years, domestic and foreign scholars have conducted a large number of related research methods for hybrid forecasting methods. Although the traditional group intelligence optimization method is simple and easy to implement, it is difficult to accurately characterize the nonlinear characteristics of the data, and satisfactory prediction accuracy cannot be obtained. Based on the artificial neural network prediction method, the learning ability is strong and the calculation speed is fast, which solves the problem that many traditional statistical methods are difficult to handle. However, artificial neural networks are prone to overfitting and the stability of single prediction model is poor, which makes this method has some limitations in time series prediction. In contrast, the combination of different prediction methods formed by the combination of the prediction method, not only combines the advantages of a variety of single prediction methods, but also significantly improves the accuracy of the entire model prediction, its effect is often better than a single prediction model.

However, the combination of smart intelligence optimization strategies and predictive models has always been a difficult and challenging problem. As the actual problems tend to be complicated and large-scale, traditional optimization methods based on strict mathematical principles are powerless in implementing predictions. In the framework of artificial intelligence, this paper introduces the whales' local government into the optimization problem and develops a new model of hybrid optimization problems and forecasting models: the artificial whale model and the resulting efficient intelligent optimization algorithm: artificial whale optimization.At the same time, two application examples of artificial whale algorithm in multi-objective optimization and local connection neural network training are given in this paper. Finally, this paper discusses the development patterns of artificial whale behavior model and artificial whale algorithm. The main contributions to this article are as follows: First, this paper gives the principle and detailed description of the artificial whale optimization algorithm, and analyzes the convergence performance of the algorithm and the influence of various parameters on the convergence of the algorithm. Based on the theory, this paper describes the distance for the artificial whale optimization algorithm. Concepts such as neighborhood and vision, and based on the concept of artificial autonomy, the original concept was developed and improved. Then the algorithm was applied into the combinatorial optimization problem. The experimental results proved the effectiveness of the improved algorithm. In addition, the artificial whale algorithm used in this dissertation solves the optimization problem for large-scale systems.Second, this paper proposes a hybrid optimization algorithm and a neural network-based time series prediction model. This paper conbines a hybrid prediction model which can apply parallel computing based on partial connection neural network. This paper presents the theoretical description and algorithm pseudo code of the new hybrid model, and discusses the superiority of the new model.Third, based on the new hybrid strategy, this paper developed four kinds of training algorithms and set up four new hybrid models to study the balance between the predictive model and the fitted model. We apply different training algorithms to partially connect neural networks based on deep learning methods to evaluate the model's value by examining the performance of the model on time series problems. At the same time, we developed MSI indicators to measure the stability of the model.Fourth, the other major contribution of this paper is to explore how the effect of feature learning on partially connected neural networks is reflected in the accuracy of time series prediction and further influence the stability of the model. This article explores the optimal characteristic learning strategies that are applicable to predictive issues based on comparative experiments by introducing the 'pooling' strategy and the 'freeway gate' approach. Combined with the above-mentioned development of artificial whale algorithm, this paper presents a highly efficient feature learning strategy that can be used in hybrid models.Fifth, in particular, in the field of wind energy and financial redeem, the stability of the predictive model plays an important role, because the stability and accuracy of the predictive model in these fields determine both economic benefits and social stability. The empirical results show that the proposed model not only has good prediction accuracy, but also has reliable stability and can be used to solve the forecasting problems in the field of wind energy and financial lending.

URL查看原文
Language中文
Document Type学位论文
Identifierhttps://ir.lzu.edu.cn/handle/262010/224409
Collection数学与统计学院
Recommended Citation
GB/T 7714
董昀轩. 一种新型动态人工鲸鱼算法的研究及在时间序列预测中的应用[D]. 兰州. 兰州大学,2018.
Files in This Item:
There are no files associated with this item.
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Altmetrics Score
Google Scholar
Similar articles in Google Scholar
[董昀轩]'s Articles
Baidu academic
Similar articles in Baidu academic
[董昀轩]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[董昀轩]'s Articles
Terms of Use
No data!
Social Bookmark/Share
No comment.
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.