【摘要】：隨著現(xiàn)代經(jīng)濟(jì)的發(fā)展,越來(lái)越多的人選擇飛機(jī)作為出行工具。為使乘客平安到達(dá)目的地,安全是民航空管中的重要主題,其中,危險(xiǎn)源的識(shí)別和分析是安全的重要保證。由于飛機(jī)在飛行中涉及環(huán)境和設(shè)備的種類繁多并且參數(shù)復(fù)雜,導(dǎo)致危險(xiǎn)源特征數(shù)據(jù)量大,深層特征較多并且具有較強(qiáng)的關(guān)聯(lián)性,這對(duì)危險(xiǎn)源識(shí)別和分析提出了挑戰(zhàn)。如何處理海量的危險(xiǎn)源數(shù)據(jù)并分析危險(xiǎn)源的深層特征是危險(xiǎn)源識(shí)別和分析的研究重點(diǎn)。本文以民航安全風(fēng)險(xiǎn)管理為背景,設(shè)計(jì)了民航危險(xiǎn)源管理系統(tǒng),并利用改進(jìn)的深層極限學(xué)習(xí)機(jī)和粒子群優(yōu)化算法完成危險(xiǎn)源的識(shí)別和分析。本文的主要研究?jī)?nèi)容如下:(1)給出了民航危險(xiǎn)源管理系統(tǒng)的總體框架和主要功能結(jié)構(gòu),并給出系統(tǒng)中關(guān)鍵技術(shù)的詳細(xì)介紹。(2)設(shè)計(jì)了一種基于深層極限學(xué)習(xí)機(jī)的危險(xiǎn)源識(shí)別算法,算法由多個(gè)深層棧式極限學(xué)習(xí)機(jī)(S-ELM)和一個(gè)單隱藏層極限學(xué)習(xí)機(jī)(ELM)構(gòu)成的深層網(wǎng)絡(luò)結(jié)構(gòu)組成。多個(gè)S-ELM采用平行的結(jié)構(gòu),各自擁有不同的隱藏結(jié)點(diǎn)個(gè)數(shù),按照危險(xiǎn)源領(lǐng)域接受危險(xiǎn)源狀態(tài)信息,并將其最上一層的隱層輸出作為ELM的輸入。在單隱藏層ELM中引入反向傳播算法并對(duì)其進(jìn)行改進(jìn),提高算法的識(shí)別準(zhǔn)確率。同時(shí),改進(jìn)S-ELM的輸入權(quán)重分配方式并采用分別訓(xùn)練深層S-ELM的方法,緩解了高維數(shù)據(jù)訓(xùn)練的內(nèi)存壓力和節(jié)點(diǎn)過(guò)多產(chǎn)生的過(guò)擬合現(xiàn)象。在某民航危險(xiǎn)源管理系統(tǒng)的數(shù)據(jù)庫(kù)上對(duì)算法進(jìn)行驗(yàn)證,結(jié)果表明該算法能夠提高設(shè)層神經(jīng)網(wǎng)絡(luò)的訓(xùn)練效率和對(duì)危險(xiǎn)源的識(shí)別精確度。(3)設(shè)計(jì)了一種基于加權(quán)的多種群粒子群優(yōu)化的危險(xiǎn)源原因分析算法。算法分為數(shù)據(jù)預(yù)處理階段和危險(xiǎn)源原因分析階段。在數(shù)據(jù)預(yù)處理階段,為危險(xiǎn)源事務(wù)數(shù)據(jù)庫(kù)中的項(xiàng)目分配權(quán)重,并定義加權(quán)的項(xiàng)目集范圍產(chǎn)生有意義的危險(xiǎn)源候選關(guān)聯(lián)規(guī)則。在危險(xiǎn)源原因分析階段,算法使用加權(quán)的多種群粒子群優(yōu)化產(chǎn)生危險(xiǎn)源關(guān)聯(lián)規(guī)則,并回溯產(chǎn)生的關(guān)聯(lián)規(guī)則得到危險(xiǎn)源原因。在算法中,采取多種群平行的搜索模式,并提供了種群間的交互機(jī)制,為了使得產(chǎn)生更有意義的危險(xiǎn)源關(guān)聯(lián)規(guī)則,為種群中的每個(gè)粒子引入權(quán)重的概念,同時(shí)在粒子速度更新公式中,引入粒子權(quán)重和全局局部最優(yōu)解,增強(qiáng)粒子間的交互。(4)完成了民航危險(xiǎn)源管理系統(tǒng)主要功能的開(kāi)發(fā),并將研究成果初步應(yīng)用于民航危險(xiǎn)源管理系統(tǒng)中,給出了設(shè)計(jì)算法及主要功能模塊的詳細(xì)技術(shù)實(shí)現(xiàn)及典型運(yùn)行界面。
文內(nèi)圖片：
圖片說(shuō)明：算法精確度與參數(shù)C的關(guān)系
[Abstract]:With the development of modern economy, more and more people choose aircraft as a travel tool. In order to make passengers reach their destination safely, safety is an important theme in civil aviation air traffic control, in which the identification and analysis of dangerous sources is an important guarantee of safety. Because there are many kinds of environment and equipment involved in aircraft flight and the parameters are complex, the data of hazard source characteristics is large, the deep features are more and have strong correlation, which challenges the identification and analysis of hazard sources. How to deal with massive hazard data and analyze the deep characteristics of dangerous sources is the research focus of hazard identification and analysis. In this paper, based on civil aviation safety risk management, a civil aviation risk source management system is designed, and the improved deep limit learning machine and particle swarm optimization algorithm are used to identify and analyze the risk sources. The main research contents of this paper are as follows: (1) the overall framework and main functional structure of civil aviation hazard source management system are given, and the key technologies in the system are introduced in detail. (2) A hazard source identification algorithm based on deep limit learning machine is designed, which is composed of multiple deep stack limit learning machines (S-ELM) and a single hidden layer limit learning machine (ELM). Multiple S-ELM adopt parallel structure, each with different number of hidden nodes, accept the status information of hazard source according to the domain of hazard source, and use the hidden layer output of the top layer as the input of ELM. The back propagation algorithm is introduced and improved in single hidden layer ELM to improve the recognition accuracy of the algorithm. At the same time, the input weight allocation method of S-ELM is improved and the deep S-ELM training method is adopted respectively, which alleviates the memory pressure of high dimensional data training and the overfitting phenomenon caused by too many nodes. The algorithm is verified on the database of a civil aviation hazard source management system. The results show that the algorithm can improve the training efficiency of layered neural network and the accuracy of hazard source recognition. (3) A hazard source cause analysis algorithm based on weighted multi-particle swarm optimization is designed. The algorithm is divided into data preprocessing stage and hazard source cause analysis stage. In the stage of data preprocessing, the project weight is assigned to the dangerous source transaction database, and the weighted item set scope is defined to generate meaningful hazard source candidate association rules. In the stage of hazard source cause analysis, the algorithm uses weighted multi-swarm particle swarm optimization to generate hazard source association rules, and backtracking the association rules to obtain the risk source causes. In the algorithm, a variety of parallel search patterns are adopted, and the interaction mechanism between populations is provided. In order to generate more meaningful risk source association rules, the concept of weight is introduced for each particle in the population. At the same time, in the particle velocity updating formula, particle weight and global local optimal solution are introduced to enhance the interaction between particles. (4) the development of the main functions of civil aviation hazard source management system is completed. The research results are preliminarily applied to the civil aviation hazard source management system, and the design algorithm, the detailed technical implementation of the main functional modules and the typical operation interface are given.
【學(xué)位授予單位】：南京航空航天大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2017
【分類號(hào)】：V328;TP315

【參考文獻(xiàn)】

相關(guān)期刊論文 前1條

1 鄧萬(wàn)宇;鄭慶華;陳琳;許學(xué)斌;;神經(jīng)網(wǎng)絡(luò)極速學(xué)習(xí)方法研究[J];計(jì)算機(jī)學(xué)報(bào);2010年02期

，

本文編號(hào)：2511596