【摘要】：隨著互聯(lián)網(wǎng)的浪潮,越來(lái)越多的互聯(lián)網(wǎng)金融公司應(yīng)運(yùn)而生,互聯(lián)網(wǎng)金融風(fēng)險(xiǎn)預(yù)測(cè)也成了互聯(lián)網(wǎng)金融公司決策時(shí)的重要一環(huán)。如螞蟻金服通過(guò)對(duì)余額寶的歷史交易額進(jìn)行資金流入流出的預(yù)測(cè);融360根據(jù)用戶(hù)的歷史還款時(shí)間序列對(duì)用戶(hù)信貸進(jìn)行預(yù)測(cè)等。因此隨著互聯(lián)網(wǎng)金融交易激增,互聯(lián)網(wǎng)金融公司有必要提高系統(tǒng)風(fēng)險(xiǎn)預(yù)測(cè)能力,將金融風(fēng)險(xiǎn)降低到最小化。通過(guò)基于時(shí)間序列預(yù)測(cè)的方法可以為風(fēng)險(xiǎn)預(yù)測(cè)提供參考從而降低風(fēng)險(xiǎn)。本文一方面著眼于時(shí)間序列預(yù)測(cè)領(lǐng)域中一些模型,如基于傳統(tǒng)ARMA模型,基于神經(jīng)網(wǎng)絡(luò)模型等,并列出各模型在時(shí)間序列應(yīng)用中的優(yōu)缺點(diǎn),然后在基礎(chǔ)的Elman神經(jīng)網(wǎng)絡(luò)預(yù)測(cè)模型上進(jìn)行改進(jìn)。另一方面著眼于時(shí)序數(shù)據(jù)的特征學(xué)習(xí),分析了時(shí)序數(shù)據(jù)特征提取與特征選擇的常用算法,并由此提出了基于時(shí)序數(shù)據(jù)的特征學(xué)習(xí)框架。通過(guò)時(shí)序數(shù)據(jù)的特征學(xué)習(xí)框架與改進(jìn)的Elman神經(jīng)網(wǎng)絡(luò)預(yù)測(cè)模型相結(jié)合,提出了一個(gè)互聯(lián)網(wǎng)金融風(fēng)險(xiǎn)預(yù)測(cè)模型。本模型著眼于實(shí)際應(yīng)用。主要內(nèi)容包含如下:1、分析了神經(jīng)網(wǎng)絡(luò)模型在時(shí)間序列預(yù)測(cè)上的應(yīng)用。包含了前向神經(jīng)網(wǎng)絡(luò)和反饋神經(jīng)網(wǎng)絡(luò)。主要針對(duì)Elman神經(jīng)網(wǎng)絡(luò)模型的研究,分析了其模型結(jié)構(gòu),各層神經(jīng)元的特點(diǎn),并修改了Elman神經(jīng)網(wǎng)絡(luò)訓(xùn)練算法,在誤差計(jì)算中將歷史數(shù)據(jù)按照與當(dāng)前時(shí)間的遠(yuǎn)近賦予相應(yīng)的權(quán)值,以及加入時(shí)序數(shù)據(jù)隨機(jī)過(guò)程,提出了改進(jìn)的Elman神經(jīng)網(wǎng)絡(luò)時(shí)間序列預(yù)測(cè)模型(GT-Elman),從而增強(qiáng)了Elman神經(jīng)網(wǎng)絡(luò)對(duì)時(shí)間序列的預(yù)測(cè)性能。2、分析了時(shí)序數(shù)據(jù)常用的特征提取算法與特征選擇算法。通過(guò)將時(shí)域序列轉(zhuǎn)化為頻域序列如快速傅立葉變換,離散小波變換等特征提取算法,提取時(shí)序數(shù)據(jù)中的特征;在特征選擇算法里分析了Clamping Network的網(wǎng)絡(luò)結(jié)構(gòu),算法思想和缺陷,并根據(jù)該缺陷提出了一種改進(jìn)的Clamping Network(DS-Clamping),從而增強(qiáng)了Clamping Network在特征選擇上的性能。相比于直接使用原始時(shí)序數(shù)據(jù)作為輸入,通過(guò)這種對(duì)時(shí)序數(shù)據(jù)特征學(xué)習(xí)模型得到時(shí)序數(shù)據(jù)的特征作為時(shí)間序列預(yù)測(cè)模型的輸入,能更好的提高預(yù)測(cè)精度,提高系統(tǒng)的預(yù)測(cè)性能。3、針對(duì)互聯(lián)網(wǎng)金融風(fēng)險(xiǎn)預(yù)測(cè)系統(tǒng)的設(shè)計(jì)和開(kāi)發(fā)。本系統(tǒng)采用基于SpringMVC框架,結(jié)合Bootstrap、Echart、JQuery搭建了互聯(lián)網(wǎng)金融風(fēng)險(xiǎn)預(yù)測(cè)系統(tǒng)。并可視化展示了系統(tǒng)返回的結(jié)果,結(jié)果表明本系統(tǒng)有著較好的實(shí)用價(jià)值。
[Abstract]:With the tide of the Internet, more and more Internet financial companies emerge as the times require, and Internet financial risk prediction has become an important part of the decision-making of Internet financial companies. For example, Ant Financial Services Group forecasts the inflow and outflow of funds through the historical transaction volume of Yu'e Bao; Rong 360 forecasts the credit of users according to the historical repayment time series of users. Therefore, with the proliferation of Internet financial transactions, it is necessary for Internet financial companies to improve their ability to predict systemic risks and minimize financial risks. The method based on time series prediction can provide reference for risk prediction and reduce risk. On the one hand, this paper focuses on some models in the field of time series prediction, such as the traditional ARMA model and neural network model, and lists the advantages and disadvantages of each model in the application of time series. Then the prediction model based on Elman neural network is improved. On the other hand, based on the feature learning of temporal data, the common algorithms of feature extraction and feature selection for temporal data are analyzed, and a framework of feature learning based on temporal data is proposed. By combining the feature learning framework of time series data with the improved Elman neural network prediction model, an Internet financial risk forecasting model is proposed. This model focuses on practical application. The main contents are as follows: 1. The application of neural network model in time series prediction is analyzed. It includes forward neural network and feedback neural network. Aiming at the research of Elman neural network model, the structure of the model and the characteristics of each layer of neurons are analyzed, and the training algorithm of Elman neural network is modified. In the error calculation, the historical data is assigned the corresponding weight value according to the distance and near to the current time. An improved Elman neural network time series prediction model (GT-Elman) is proposed by adding time series random process, which enhances the performance of Elman neural network in time series prediction. The feature extraction algorithm and feature selection algorithm are analyzed. By transforming the time-domain sequence into frequency-domain sequences such as fast Fourier transform (FFT), discrete wavelet transform (DWT) and other feature extraction algorithms, the features of time-series data are extracted. In the feature selection algorithm, the network structure, algorithm idea and defect of Clamping Network are analyzed, and an improved Clamping Network (DS-Clamping is proposed to improve the performance of Clamping Network in feature selection. Compared with the direct use of the original time series data as the input, the feature of the time series data can be obtained by using the feature learning model of the time series data as the input of the time series prediction model, which can improve the prediction accuracy better. Improve the prediction performance of the system. 3, for the Internet financial risk forecasting system design and development. This system uses SpringMVC framework and Bootstrap,Echart,JQuery to build Internet financial risk forecasting system. The results show that the system has good practical value.
【學(xué)位授予單位】：電子科技大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2017
【分類(lèi)號(hào)】：F724.6;F832;TP183

【參考文獻(xiàn)】

相關(guān)期刊論文 前6條

1 余宇峰;朱躍龍;萬(wàn)定生;關(guān)興中;;基于滑動(dòng)窗口預(yù)測(cè)的水文時(shí)間序列異常檢測(cè)[J];計(jì)算機(jī)應(yīng)用;2014年08期

2 李敏;卡米力·木依丁;;特征選擇方法與算法的研究[J];計(jì)算機(jī)技術(shù)與發(fā)展;2013年12期

3 唐玉娜;李啟會(huì);;ARMA模型在預(yù)測(cè)問(wèn)題中的應(yīng)用[J];嘉興學(xué)院學(xué)報(bào);2006年S1期

4 翟宜峰,李鴻雁,劉寒冰,苑希民;用遺傳算法優(yōu)化神經(jīng)網(wǎng)絡(luò)初始權(quán)重的方法[J];吉林大學(xué)學(xué)報(bào)(工學(xué)版);2003年02期

5 范勁松,方廷健;特征選擇和提取要素的分析及其評(píng)價(jià)[J];計(jì)算機(jī)工程與應(yīng)用;2001年13期

6 安鴻志;General ARMA Models[J];數(shù)學(xué)進(jìn)展;1986年02期

相關(guān)博士學(xué)位論文 前1條

1 聶淑媛;時(shí)間序列分析的早期發(fā)展[D];西北大學(xué);2012年

，

本文編號(hào)：2340736