凌曉,徐魯帥,高甲程,馬娟娟,馬賀清,付小華
基于IFA-BPNN的長(zhǎng)輸管道外腐蝕速率預(yù)測(cè)
凌曉1a,徐魯帥1a,高甲程2,馬娟娟1a,馬賀清1a,付小華1b
(1.蘭州理工大學(xué) a.石油化工學(xué)院 b.理學(xué)院,蘭州 730050;2.中國(guó)石油天然氣股份有限公司 甘肅蘭州銷售分公司,蘭州 730050)
構(gòu)建陸地長(zhǎng)輸管道外腐蝕速率的預(yù)測(cè)模型,提升管道外腐蝕速率預(yù)測(cè)的精度,對(duì)長(zhǎng)輸管道外腐蝕狀態(tài)進(jìn)行準(zhǔn)確把控。深入解析了螢火蟲算法(FA)的工作原理,針對(duì)FA易出現(xiàn)陷入局部最優(yōu)或因控制參數(shù)設(shè)置不合適而導(dǎo)致函數(shù)無法收斂等問題,提出了FA的改進(jìn)方案:采用Logistics混沌映射的方法初始化螢火蟲的位置,提升螢火蟲種群的所養(yǎng)性;引入一種新的慣性權(quán)重計(jì)算方法來改進(jìn)螢火蟲位置移動(dòng)公式,提升FA全局尋優(yōu)能力。利用改進(jìn)的螢火蟲算法(IFA)對(duì)誤差反向傳播神經(jīng)網(wǎng)絡(luò)(BPNN)初始權(quán)值和閾值進(jìn)行優(yōu)化,建立基于IFA-BPNN的長(zhǎng)輸管道外腐蝕速率預(yù)測(cè)模型。以111組長(zhǎng)輸管道外腐蝕檢測(cè)數(shù)據(jù)為例,在MATLAB中進(jìn)行模擬仿真計(jì)算,使用粒子群算法優(yōu)化的BPNN(PSO-BPNN)、遺傳算法優(yōu)化的BPNN(GA-BPNN)以及未進(jìn)行優(yōu)化的BPNN作為對(duì)比模型進(jìn)行對(duì)比分析。使用IFA優(yōu)化BPNN,大幅提升了BPNN模型的預(yù)測(cè)精度。使用IFA-BPNN模型預(yù)測(cè)12組管道腐蝕速率,平均相對(duì)誤差僅為5.94%,預(yù)測(cè)結(jié)果的2為0.995 95,均優(yōu)于BPNN、PSO-BPNN以及GA-BPNN模型的預(yù)測(cè)結(jié)果。IFA-BPNN作為預(yù)測(cè)管道腐蝕速率工具具有較好的預(yù)測(cè)精度和魯棒性。
螢火蟲算法;BP神經(jīng)網(wǎng)絡(luò);混沌初始化;慣性權(quán)重;管道;腐蝕速率預(yù)測(cè)
管道作為最快速、最安全的石油和天然氣運(yùn)輸方式,建設(shè)規(guī)模不斷擴(kuò)大。2019年末,我國(guó)長(zhǎng)輸管道總里程已達(dá)13.9×104km[1],管道失效將直接影響企業(yè)的經(jīng)濟(jì)效益和居民的安全[2-3]。近年來,國(guó)內(nèi)外研究人員對(duì)管道失效模式進(jìn)行了大量研究,發(fā)現(xiàn)腐蝕是造成油氣長(zhǎng)輸管道失效的主要原因之一[4-6]。管道開挖檢測(cè)費(fèi)時(shí)費(fèi)力,易造成資源浪費(fèi)。因此,基于算法模型以及影響管道腐蝕的各項(xiàng)數(shù)據(jù),對(duì)管道腐蝕狀態(tài)進(jìn)行準(zhǔn)確預(yù)測(cè),可為管道檢維修提供決策支持,對(duì)保障管道安全運(yùn)行具有重大意義[7-11]。
張河葦?shù)萚12]使用互信息理論確定了管道腐蝕的主要影響因素,為選取預(yù)測(cè)管道腐蝕速率的數(shù)據(jù)提供了決策支持。畢傲睿等[13]利用主成分分析法對(duì)輸油管道內(nèi)腐蝕因素進(jìn)行優(yōu)選,選出了影響管道腐蝕的主要因素,并結(jié)合改進(jìn)的支持向量機(jī)模型對(duì)管道內(nèi)腐蝕情況進(jìn)行預(yù)測(cè),且結(jié)果較為理想,但支持向量機(jī)由于算法本體缺陷,針對(duì)大規(guī)模訓(xùn)練樣本難以實(shí)施。陳迪等[14]研究發(fā)現(xiàn)了影響含硫管道腐蝕情況的四大因素,并結(jié)合實(shí)驗(yàn)結(jié)論建立了一套預(yù)測(cè)含硫管道腐蝕狀態(tài)模型,預(yù)測(cè)效果較優(yōu),但該模型僅針對(duì)管道內(nèi)腐蝕有效,不適用于管道外腐蝕速率預(yù)測(cè)。章玉婷等[15]使用單一的BP神經(jīng)網(wǎng)絡(luò)(BP Neural Network,BPNN)對(duì)管道腐蝕速率進(jìn)行分析預(yù)測(cè),由于未經(jīng)優(yōu)化的BPNN易陷入局部最優(yōu),因此預(yù)測(cè)值相對(duì)真實(shí)數(shù)據(jù)誤差較大。董紹華等提出中國(guó)管道的發(fā)展目標(biāo)是建立基于管道全生命周期大數(shù)據(jù)的智慧管網(wǎng)[16],而基于機(jī)器學(xué)習(xí)的管道大數(shù)據(jù)分析處理是構(gòu)建智慧管網(wǎng)的重要內(nèi)容之一。近年來,Yang等[17-19]提出了螢火蟲算法(Firefly Algorithm,F(xiàn)A),并通過仿真證實(shí)了FA要優(yōu)于粒子群優(yōu)化算法(Particle Swarm Optimization,PSO)和遺傳算法(Genetic Algorithm,GA)。NANDY等[20]使用FA優(yōu)化BPNN的初始權(quán)值和閾值,并通過仿真證實(shí)了其可行性。
綜上所述,目前研究工作著重于管道腐蝕小樣本數(shù)據(jù)預(yù)測(cè)研究,或因算法本體缺陷而導(dǎo)致腐蝕速率的預(yù)測(cè)誤差較大。隨著智慧管網(wǎng)的發(fā)展,管道數(shù)據(jù)采集量將會(huì)增大,而BPNN可應(yīng)對(duì)數(shù)據(jù)量較大的問題。因此,本文對(duì)FA算法進(jìn)行了改進(jìn),增強(qiáng)其全局尋優(yōu)能力,并利用IFA對(duì)BPNN初始權(quán)值和閾值進(jìn)行優(yōu)化,建立了基于IFA-BPNN的長(zhǎng)輸管道外腐蝕速率預(yù)測(cè)模型,并對(duì)該模型進(jìn)行實(shí)例應(yīng)用,驗(yàn)證了該模型的適應(yīng)性和魯棒性。
BPNN模型具有較強(qiáng)的非線性函數(shù)擬合能力,但該模型常因初始權(quán)值和閾值的設(shè)置不當(dāng)而陷入局部最優(yōu),從而使模型預(yù)測(cè)結(jié)果不理想。為改進(jìn)BPNN本體存在的不足,本文提出使用IFA對(duì)BPNN初始值進(jìn)行優(yōu)化,并以GA和PSO作為對(duì)比優(yōu)化模型進(jìn)行測(cè)試分析,以驗(yàn)證IFA-BPNN模型的使用效果。
BPNN于1986年提出,是現(xiàn)階段應(yīng)用較廣泛的神經(jīng)網(wǎng)絡(luò)模型之一。BPNN一般采用3層網(wǎng)絡(luò)結(jié)構(gòu)便可達(dá)到良好的非線性逼近效果[21-22]。BPNN輸入和輸出的節(jié)點(diǎn)數(shù)量分別根據(jù)數(shù)據(jù)輸入類別和預(yù)期輸出類別確定,隱藏層中的節(jié)點(diǎn)數(shù)量可根據(jù)經(jīng)驗(yàn)公式確定,各層通過權(quán)值相連,隱藏層和輸出層各節(jié)點(diǎn)設(shè)有閾值。BPNN網(wǎng)絡(luò)拓?fù)鋱D如圖1所示。
BPNN模型正向傳遞過程按式(1)—(4)進(jìn)行運(yùn)算[9]。
圖1 BPNN模型示意圖
誤差反向傳播的誤差函數(shù)公式及權(quán)值閾值修正系數(shù)的公式為[9]:
GA由Holland及其學(xué)生于1975年創(chuàng)建,其思想是基于達(dá)爾文的進(jìn)化論和Mendel的遺傳學(xué)說,其主體分為選擇、交叉和變異三部分[23-24]:
1)選擇。文內(nèi)采用輪盤賭方法進(jìn)行選擇操作,其公式為式(10),其中f計(jì)算公式與式(5)相同。
2)交叉。文中采用實(shí)數(shù)交叉法,公式見式(11)。
3)變異。變異操作的公式見式(12)。
PSO算法原理是隨機(jī)初始化一組粒子,通過跟蹤個(gè)體極值和群體極值來更新粒子群,粒子的速度和位置分別根據(jù)式(13)和式(14)更新[25]。
慣性權(quán)重采用線性遞減的方式[26],其計(jì)算公式見式(15)。
1.4.1 FA模型
4)計(jì)算更新位置后螢火蟲的亮度。
5)滿足結(jié)束條件后輸出全局極值和最優(yōu)個(gè)體值;若不滿足,轉(zhuǎn)步驟2繼續(xù)迭代搜索,迭代次數(shù)加1。
1.4.2 FA模型改進(jìn)策略
FA作為新型優(yōu)化算法,易出現(xiàn)陷入局部最優(yōu)或因控制參數(shù)設(shè)置不合適而導(dǎo)致函數(shù)無法收斂等問題[27-29]。鑒于上述問題,對(duì)FA進(jìn)行改進(jìn),表1為IFA的偽代碼。
表1 IFA偽代碼
Tab.1 IFA pseudocode
處理流程如下所述。
1)對(duì)螢火蟲位置進(jìn)行Logistics混沌初始化,提升了螢火蟲初始種群的多樣性和螢火蟲搜索的全局遍歷性,又與螢火蟲隨機(jī)初始化位置本質(zhì)相匹配。Logistics混沌映射公式見式(20)[30]。
分別使用GA、PSO以及IFA 3種優(yōu)化算法對(duì)BPNN的初始權(quán)值和閾值進(jìn)行優(yōu)化,結(jié)合第1節(jié)各算法的理論基礎(chǔ),構(gòu)建長(zhǎng)輸管道外腐蝕速率預(yù)測(cè)模型GA-BPNN、PSO-BPNN和IFA-BPNN,具體的模型構(gòu)建流程如圖2所示。
圖2 混合模型流程圖
選用文獻(xiàn)[32]的111組管道外腐蝕數(shù)據(jù),這111組數(shù)據(jù)的管道防腐層均為煤焦油瓷漆涂層。因數(shù)據(jù)組數(shù)較多,文內(nèi)僅展示20組檢測(cè)數(shù)據(jù),如表2所示。每組數(shù)據(jù)包含11項(xiàng)檢測(cè)數(shù)據(jù),其中Corrosion rate為管道外腐蝕速率,TT為管道運(yùn)行總時(shí)長(zhǎng),PP為氧化還原電位,pH為管道外界土壤的pH值,RP為管地電位,RE為土壤電阻率。此數(shù)據(jù)集包括對(duì)現(xiàn)場(chǎng)開挖點(diǎn)的土壤成分分析數(shù)據(jù),該數(shù)據(jù)集采用標(biāo)準(zhǔn)實(shí)驗(yàn)室方法進(jìn)行分析檢測(cè),所測(cè)數(shù)據(jù)類型包括含水量(WC)、容重(BD)及溶解氯化物(CC)、碳酸氫鹽(BC)、硫酸鹽(SC)的離子濃度。隨機(jī)選取99組數(shù)據(jù)讓各模型進(jìn)行學(xué)習(xí)訓(xùn)練,利用剩余的12組數(shù)據(jù)對(duì)各模型進(jìn)行測(cè)試分析。
分別使用未經(jīng)優(yōu)化的BPNN模型、GA-BPNN模型、PSO-BPNN模型、IFA-BPNN模型進(jìn)行管道數(shù)據(jù)的學(xué)習(xí)預(yù)測(cè)。
4.2.1 BPNN網(wǎng)絡(luò)設(shè)置
BPNN采用3層網(wǎng)絡(luò)結(jié)構(gòu),各層節(jié)點(diǎn)數(shù)的設(shè)置方法如下所述。
BPNN輸入數(shù)據(jù)類型包括管道運(yùn)行總時(shí)長(zhǎng),氧化還原電位,管道外界土壤的pH值,管地電位,土壤電阻率,土壤含水量,容重,土壤中溶解氯化物、碳酸氫鹽以及硫酸鹽離子濃度。BPNN輸出數(shù)據(jù)為管道外腐蝕速率。因此,將BPNN輸入層節(jié)點(diǎn)數(shù)設(shè)為10,輸出層節(jié)點(diǎn)數(shù)設(shè)為1。隱含層神經(jīng)元節(jié)點(diǎn)數(shù)根據(jù)經(jīng)驗(yàn)公式(27)計(jì)算選取[33],因?yàn)閘og2(99)≈7,所以BPNN隱含層神經(jīng)元節(jié)點(diǎn)數(shù)設(shè)置為7。
隱含層選用logsig函數(shù)作為傳遞函數(shù),其表達(dá)式見式(28)。選用pureline型函數(shù)作為輸出層的傳遞函數(shù),表達(dá)式見式(29)。
4.2.2 模型初始化設(shè)置
表2 長(zhǎng)輸管道外腐蝕數(shù)據(jù)集
Tab.2 External corrosion data set of long-distance pipeline
表3 GA初始化參數(shù)
Tab.3 GA initialization parameters
表4 PSO初始化參數(shù)
Tab.4 PSO initialization parameters
表5 IFA初始化參數(shù)
Tab.5 IFA initialization parameters
為避免數(shù)值問題,加快BPNN的收斂速度,在進(jìn)行訓(xùn)練之前,對(duì)所有數(shù)據(jù)進(jìn)行歸一化操作,歸一化公式見式(30)。
模型訓(xùn)練后的測(cè)試結(jié)果如表6、圖3和圖4所示。由表6可知,使用未經(jīng)優(yōu)化的BPNN預(yù)測(cè)管道腐蝕速率的效果最差,最大相對(duì)誤差(Max RE)達(dá)到了30.88%,最小相對(duì)誤差(Min RE)為14.17%;預(yù)測(cè)效果最好的是IFA-BPNN模型,其預(yù)測(cè)結(jié)果的Max RE為8.82%,Min RE僅為1.47%,也就是其誤差區(qū)間為[1.47%,8.82%];相較于PSO-BPNN和GA-BPNN模型的誤差區(qū)間,IFA-BPNN模型的[Min RE,Max RE]的取值最小,跨度最小,證明了IFA-BPNN的預(yù)測(cè)精度要優(yōu)于BPNN、PSO-BPNN以及GA-BPNN。為進(jìn)一步分析驗(yàn)證IFA-BPNN的使用效果,對(duì)BPNN、PSO-BPNN、GA-BPNN、IFA-BPNN的預(yù)測(cè)結(jié)果進(jìn)行MAE計(jì)算,結(jié)果分別為22.26%、15.03%、10.74%、5.94%。相較于BPNN、PSO-BPNN、GA-BPNN預(yù)測(cè)結(jié)果的平均相對(duì)誤差,IFA-BPNN分別提升了16.32%、9.09%、4.8%。使用訓(xùn)練好的模型BPNN、PSO-BPNN、GA-BPNN、IFA-BPNN對(duì)訓(xùn)練集的99組數(shù)據(jù)進(jìn)行測(cè)試,其平均相對(duì)誤差分別為18.11%、12.54%、9.12%、5.66%,進(jìn)一步驗(yàn)證了利用IFA優(yōu)化BPNN后可有效提升其預(yù)測(cè)精度。
表6 模型預(yù)測(cè)誤差統(tǒng)計(jì)表
Tab.6 Model prediction error statistics table
圖3 模型預(yù)測(cè)結(jié)果對(duì)比圖
圖4 模型預(yù)測(cè)誤差曲線圖
由圖3可見,IFA-BPNN預(yù)測(cè)的管道外腐蝕速率與實(shí)測(cè)值最接近,相較于未經(jīng)優(yōu)化的BPNN,其預(yù)測(cè)精度有了較大幅度的提升,且IFA-BPNN的預(yù)測(cè)精度也優(yōu)于PSO-BPNN和GA-BPNN。圖4為預(yù)測(cè)結(jié)果的誤差對(duì)比圖。由圖4可見,BPNN預(yù)測(cè)結(jié)果的誤差最大,PSO-BPNN次之,GA-BPNN優(yōu)于BPNN和PSO-BPNN,IFA-BPNN預(yù)測(cè)結(jié)果的相對(duì)誤差最小,且相對(duì)誤差曲線最為平緩,這不僅體現(xiàn)出IFA-BPNN模型預(yù)測(cè)結(jié)果的準(zhǔn)確率較高,也體現(xiàn)出該模型具有較強(qiáng)的魯棒性。
分別把BPNN、PSO-BPNN、GA-BPNN、IFA-BPNN預(yù)測(cè)的管道腐蝕速率與實(shí)際檢測(cè)的管道外腐蝕速率進(jìn)行相關(guān)性分析,其結(jié)果如圖5—8所示,圖中黑色線公式為=,紅色線為預(yù)測(cè)結(jié)果擬合線??芍狟PNN模型的2為0.88037,PSO-BPNN的2為0.95876,GA-BPNN的2為0.97888,IFA-BPNN的2為0.99595。IFA-BPNN的2最接近1,這進(jìn)一步驗(yàn)證了IFA-BPNN作為預(yù)測(cè)管道腐蝕速率工具的準(zhǔn)確性和魯棒性。
圖5 BPNN預(yù)測(cè)結(jié)果線性擬合圖
圖6 PSO-BPNN預(yù)測(cè)結(jié)果線性擬合圖
圖7 GA-BPNN預(yù)測(cè)結(jié)果線性擬合圖
圖8 IFA-BPNN預(yù)測(cè)結(jié)果線性擬合圖
1)對(duì)FA進(jìn)行改進(jìn),一是對(duì)螢火蟲初始位置進(jìn)行Logistics混沌初始化,二是引入了新的慣性權(quán)重計(jì)算公式,有利于函數(shù)跳出局部最優(yōu)尋找全局最優(yōu)。利用改進(jìn)的螢火蟲算法優(yōu)化BPNN的初始權(quán)值和閾值,建立了IFA-BPNN管道外腐蝕速率預(yù)測(cè)模型。
2)分別使用BPNN、PSO-BPNN、GA-BPNN以及IFA-BPNN模型對(duì)長(zhǎng)輸管道外腐蝕速率數(shù)據(jù)進(jìn)行訓(xùn)練、預(yù)測(cè)。IFA-BPNN模型預(yù)測(cè)結(jié)果的MRE為5.94%,2為0.99595,均優(yōu)于BPNN、PSO-BPNN、GA-BPNN的預(yù)測(cè)結(jié)果,驗(yàn)證了IFA-BPNN作為預(yù)測(cè)管道腐蝕速率工具的準(zhǔn)確性和魯棒性。應(yīng)用IFA-BPNN模型預(yù)測(cè)管道外腐蝕速率可為長(zhǎng)輸管道的檢維修提供決策支持。
3)由于管道外腐蝕因素較多,且工程上難以獲取較為整齊的數(shù)據(jù),后期研究可在數(shù)據(jù)集中添加隨機(jī)變量以及噪音數(shù)據(jù)進(jìn)行深入研究。
[1] 高鵬, 高振宇, 王峰, 等. 2018年中國(guó)油氣管道建設(shè)新進(jìn)展[J]. 國(guó)際石油經(jīng)濟(jì), 2019, 27(3): 62-67. GAO Peng, GAO Zhen-yu, WANG Feng. New progress in China’s oil and gas pipeline construction in 2019[J]. International petroleum economics, 2019, 27(3): 62-67.
[2] 王大慶, 張鵬. 油氣輸送管線失效后果模糊事件樹分析[J]. 安全與環(huán)境學(xué)報(bào), 2014, 14(3): 88-92. WANG Da-qing, ZHANG Peng. Analysis of failure cons-e-quences of oil and gas transmission pipelines by using a fuzzy event tree[J]. Journal of safety and environment, 2014, 14(3): 88-92.
[3] 黃坤, 吳佳麗, 全愷, 等. 含腐蝕缺陷油氣管道評(píng)價(jià)技術(shù)研究進(jìn)展[J]. 表面技術(shù), 2018, 47(1): 116-122. HUANG Kun, WU Jia-li, QUAN Kai, et al. A review on evaluation technology of oil-gas pipelines with corrosion defect[J]. Surface technology, 2018, 47(1): 116-122.
[4] 單克, 帥健, 楊光, 等. 美國(guó)油氣管道基本失效概率評(píng)估方法及啟示[J]. 油氣儲(chǔ)運(yùn), 2020, 39(5): 530-535. SHAN Ke, SHUAI Jian, YANG Guang, et al. Evaluation method for basic failure probability of oil and gas pipeline based on accident statistics in the United States and enli-ghtenments[J]. Oil & gas storage and transportation, 2020, 39(5): 530-535.
[5] CHIO Lam, ZHOU Wen-xing. Statistical analyses of inci-dents on onshore gas transmission pipelines based on PHMSA database[J]. International journal of pressure ve-ssels and piping, 2016, 145: 29-40.
[6] 張英, 呂佳慧, 王衛(wèi)澤, 等. 失效數(shù)據(jù)庫的建立和近年國(guó)內(nèi)失效案例的統(tǒng)計(jì)分析[J]. 金屬熱處理, 2019, 44(1): 668-675. ZHANG Ying, LYU Jia-hui, WANG Wei-ze, et al. Estab-lishment of failure database and statistical analysis of do-m-estic failure cases based in recent years[J]. Heat treatment of metals, 2019, 44(1): 668-675.
[7] MOHAMMED S, ABBASY E, AHMED S, et al. Artificial neural network models for predicting condition of offsh-ore oil and gas pipelines[J]. Automation in construction, 2014, 45: 50-65.
[8] 曲志豪, 唐德志, 胡麗華, 等. 基于優(yōu)化隨機(jī)森林的H2S腐蝕產(chǎn)物類型及腐蝕速率預(yù)測(cè)[J]. 表面技術(shù), 2020, 49(3): 42-49. QU Zhi-hao, TANG De-zhi, HU Li-hua, et al. Prediction of H2S corrosion products and corrosion rate based on optimized random forest[J]. Surface technology, 2020, 49(3): 42-49.
[9] 凌曉, 徐魯帥, 梁瑞, 等. 基于改進(jìn)PSO-BPNN的輸油管道內(nèi)腐蝕速率研究[J]. 中國(guó)安全生產(chǎn)科學(xué)技術(shù), 2019, 15(10): 63-68. LING Xiao, XU Lu-shuai, LIANG Rui, et al. Study on internal corrosion rate of oil pipeline based on improved PSO-BPNN[J]. Journal of safety science and technology, 2019, 15(10): 63-68.
[10] 馬鋼, 李俊飛, 白瑞, 等. 基于PSO-SVM模型的油氣管道內(nèi)腐蝕速率預(yù)測(cè)[J]. 表面技術(shù), 2019, 48(5): 43-48. MA Gang, LI Jun-fei, BAI Rui, et al. Prediction of corr-osion rate in oil and gas pipelines based on PSO-SVM model[J]. Surface technology, 2019, 48(5): 43-48.
[11] 王文輝, 駱正山, 張新生. 基于PSO-GRNN模型的埋地管道腐蝕剩余壽命預(yù)測(cè)[J]. 表面技術(shù), 2019, 48(10): 267-275. WANG Wen-hui, LUO Zheng-shan, ZHANG Xin-sheng. Prediction on remaining service life of buried pipeline after corrosion based on PSO-GRNN model[J]. Surface technology, 2019, 48(10): 267-275.
[12] 張河葦, 金劍, 董紹華, 等. 一種基于管道大數(shù)據(jù)的腐蝕因素相關(guān)性分析方法[J]. 科學(xué)通報(bào), 2018, 63(8): 777-783. ZHAN He-wei, JIN Jian, DONG Shao-hua, et al. A corr-osion correlation analysis method based on pipeline big data[J]. Chinese science bulletin, 2018, 63(8): 777-783.
[13] 畢傲睿, 駱正山, 喬偉, 等. 基于主成分和粒子群優(yōu)化支持向量機(jī)的管道內(nèi)腐蝕預(yù)測(cè)[J]. 表面技術(shù), 2018, 47(9): 133-140. BI Ao-rui, LUO Zheng-shan, QIAO Wei, et al. Prediction of pipeline inner-corrosion based on principal component analysis and particle swarm optimization-support vector machine[J]. Surface technology, 2018, 47(9): 133-140.
[14] 陳迪, 廖柯熹, 何國(guó)璽, 等. 基于EWM-GRA的腐蝕主控因素分析與腐蝕模型建立[J]. 表面技術(shù), 2019, 48(6): 268-273. CHEN Di, LIAO Ke-xi, HE Guo-xi, et al. Analysis of corrosion main influencing factors based on EWM-GRA and corrosion model establishment[J]. Surface technology, 2019, 48(6): 268-273.
[15] 章玉婷, 楊劍鋒. 基于BP神經(jīng)網(wǎng)絡(luò)的管道腐蝕速率預(yù)測(cè)[J]. 全面腐蝕控制, 2013, 27(9): 67-71. ZHANG Yu-ting, YANG Jian-feng. Corrosion rate predi-c-tion of pipeline based on BP artificial neural network[J]. Total corrosion control, 2013, 27(9): 67-71.
[16] 董紹華. 中國(guó)油氣管道完整性管理20年回顧與發(fā)展建議[J]. 油氣儲(chǔ)運(yùn), 2020, 39(3): 241-261. DONG Shao-hua. Review of China’s oil and gas pipeline integrity management in the past 20 years and developm-ent suggestions[J]. Oil & gas storage and transportation, 2020, 39(3): 241-261.
[17] YANG X S. Firefly algorithms for multimodal optimi-zation[J]. Lecture notes in computer science, 2009, 5792: 169-178.
[18] YANG, X S. Firefly algorithm, stochastic test functions and design optimisation[J]. International journal of bio inspired computation, 2010, 2(2): 78-84.
[19] YANG X S, HE X. Firefly algorithm: recent advances andapplications[J]. International journal of swarm intelligence, 2013, 1(1): 36-50.
[20] NANDY S, SARKAR P P, DAS A. Analysis of a nature inspired firefly algorithm based back-propagation neural network training[J]. Computer science, 2012, 8(22): 207- 220.
[21] HORNIK K, STINCHCOMBE M, WHITE H. Multilayer feedforward networks are universal approximators[J]. Ne-ural networks, 1989, 2(5): 359-366.
[22] HOLLAND J H. Adaptation in natural and artificial sys-tems[M]. Ann Arbor: University of Michigan Press, 1975.
[23] 許宏良, 殷蘇民. 基于改進(jìn)BP神經(jīng)網(wǎng)絡(luò)優(yōu)化的管道腐蝕速率預(yù)測(cè)模型研究[J]. 表面技術(shù), 2018, 47(2): 177- 181.XU Hong-liang, YIN Su-min. Prediction model of pipeline corrosion rate based on improved BP neural network[J]. Surface technology, 2018, 47(2): 177-181.
[24] 吉根林. 遺傳算法研究綜述[J]. 計(jì)算機(jī)應(yīng)用與軟件, 2004, 21(2): 69-73. JI Gen-lin. Survey on genetic algorithm[J]. Computer ap-p-lications and software, 2004, 21(2): 69-73.
[25] SHI Y H, EBERHART R C. A modified particle swarm optimizer[C]//Proc of the IEEE international conference on evolutionary computation. Piscataway: IEEE Service Center, 1998: 69-73.
[26] CHATTERJEE A, SIARRY P. Nonlinear inertia weight variation for dynamic adaptation in particle swarm opti-mi-zation[J]. Computers and operations research, 2006, 33(3): 859-871.
[27] 程美英, 倪志偉, 朱旭輝. 螢火蟲優(yōu)化算法理論研究綜述[J]. 計(jì)算機(jī)科學(xué), 2015(4): 24-29. CHENG Mei-ying, NI Zhi-wei, ZHU Xu-hui. Overview on glowworm swarm optimization or firefly algorithm[J]. Computer science, 2015(4): 24-29.
[28] 楊雁瑩, 徐仙偉, 曹霽. 基于仿生理論的新型優(yōu)化算法綜述[J]. 計(jì)算機(jī)仿真, 2016, 33(6): 233-237. YANG Yan-ying, XU Xian-wei, CAO Ji. Overview of new optimization algorithms based on bionic theory[J]. Computer simulation, 2016, 33(6): 233-237.
[29] 李巍華, 翁勝龍, 張紹輝. 一種螢火蟲神經(jīng)網(wǎng)絡(luò)及在軸承故障診斷中的應(yīng)用[J]. 機(jī)械工程學(xué)報(bào), 2015(7): 99-106. LI Wei-hua WENG Sheng-long, ZHANG Shao-hui. A firefly neural network and its application in bearing fault diagnosis[J]. Journal of mechanical engineering, 2015(7): 99-106.
[30] ROBERT M. On the derivation of a “Chaotic” encryption algorithm[J]. Cryptologia, 1989, 13(1): 29-42.
[31] 李丹丹. 基于改進(jìn)粒子群算法的混沌時(shí)間序列預(yù)測(cè)[D]. 秦皇島: 燕山大學(xué), 2014. LI Dan-dan. Chaotic time serise prediction based on imp-roved particle swarm algorithm[D]. Qinhuangdao: Yanshan University, 2014.
[32] VELAZQUEZ J C, CALEYO F, VALOR A, et al. Tech-nical note: Field study-pitting corrosion of underground pipelines related to local soil and pipe characteristics[J]. Corrosion, 2010, 66(1): 0160011-0160015.
[33] WANAS N, AUDA G, KAMEL M S, et al. On the optimal number of hidden nodes in a neural network[C]// Procee-ding of 1998 IEEE canadian conference on electrical and computer engineering. Toronto: IEEE, 2002: 918-921.
Prediction of External Corrosion Rate of Oil Pipeline Based on Improved IFA-BPNN
1a,1a,2,1a,1a,1b
(1.a.College of Petroleum and Chemical Engineering, b.College of Sciences, Lanzhou University of Technology, Lanzhou 730050, China; 2.PetroChina Gansu Lanzhou Marketing Company, Lanzhou 730050, China)
In order to establish a machine learning model for predicting the external corrosion rate of long land transport pipelines, improve the prediction accuracy of the external corrosion rate of the pipeline, and accurately grasp the external corrosion status of the long-distance pipeline, this paper analyzes the working principle of FA, to solve the problems of FA, such as local optimization or function convergence failure due to initial parameter setting, and an improved FA algorithm is proposed: This paper uses the method of Logistics chaotic mapping to initialize the position of the firefly, and improve the cultivability of the firefly population; this paper introduces a new inertia weight calculation method to improve the formula of the firefly position movement and enhance the FA global optimization ability. The improved FA (IFA) was used to optimize the initial weights and thresholds of BPNN, and a long-distance pipeline external corrosion rate prediction model based on IFA-BPNN was established. Taking 111 sets of long-distance pipeline external corrosion detection data as an example, the simulation calculation is carried out in MATLAB, and PSO-BPNN, GA-BPNN and unoptimized BPNN are used as comparative models for comparative analysis. The IFA model is used to initialize the BPNN model, which greatly improves the prediction accuracy of the BPNN model. The IFA-BPNN model was used to predict and analyze the external corrosion rates of 12 groups of pipelines, the average relative error was only 5.94%, and the2of the prediction results was 0.995 95. The prediction results of IFA-BPNN model are superior to those of BPNN model, PSO-BPNN model and GA-BPNN model in all aspects. IFA-BPNN has good accuracy and robustness as a tool to predict pipeline corrosion rate.
firefly algorithm; BP neural network; chaos initialization; inertia weight; oil pipelines; corrosion rate prediction
2020-07-30;
2020-11-27
LING Xiao (1982—), Male, Doctor, Associate professor, Research focus: oil and gas pipeline integrity management. E-mail: lingxiao_ lut@163.com
凌曉, 徐魯帥, 高甲程, 等.基于IFA-BPNN的長(zhǎng)輸管道外腐蝕速率預(yù)測(cè)[J]. 表面技術(shù), 2021, 50(4): 285-293.
TG172
A
1001-3660(2021)04-0285-09
10.16490/j.cnki.issn.1001-3660.2021.04.029
2020-07-30;
2020-11-27
國(guó)家自然科學(xué)基金青年項(xiàng)目(51904138);甘肅省自然科學(xué)基金(20JR5RA451);甘肅省高等學(xué)校創(chuàng)新能力提升項(xiàng)目(2020A-019)
Fund:Supported by the Youth Program of National Natural Science Foundation of China (51904138); the Natural Science Foundation of Gansu Province (20JR5RA451); Innovation Ability Improvement Project of Colleges and Universities in Gansu Province (2020A-019)
凌曉(1982—),男,博士,副教授,主要研究方向?yàn)橛蜌夤艿劳暾怨芾?。郵箱:lingxiao_lut@163.com
LING Xiao, XU Lu-shuai, GAO Jia-cheng, et al. Prediction of external corrosion rate of oil pipeline based on improved IFA-BPNN[J]. Surface technology, 2021, 50(4): 285-293.