陳鮮展 沈易成 洪飛揚 石紳
文章編號:1671?251X(2024)04?0128?05 ?DOI:10.13272/j.issn.1671-251x.18122
摘要:針對目前瓦斯?jié)舛阮A(yù)測方法存在數(shù)據(jù)處理不確定性、特征提取局限性及受主觀性因素影響產(chǎn)生預(yù)測偏差等問題,提出了一種用于煤礦掘進工作面的瓦斯?jié)舛阮A(yù)測方法。首先,在煤礦掘進工作面回風巷內(nèi)每隔1 m 布設(shè)激光瓦斯傳感器,形成傳感器網(wǎng)絡(luò),實時采集瓦斯?jié)舛葦?shù)據(jù)。然后,根據(jù)拉依達準則搜索并剔除瓦斯?jié)舛葦?shù)據(jù)中的異常值,并利用 Lagrange 插值多項式填補瓦斯?jié)舛葦?shù)據(jù)中的缺失值。最后,以剔除異常值及填補缺失值的瓦斯?jié)舛葦?shù)據(jù)為基礎(chǔ),采用經(jīng)驗?zāi)B(tài)分解算法將瓦斯?jié)舛葦?shù)據(jù)分解成本征模態(tài)函數(shù)和趨勢項,再利用 Hilbert 變換對本征模態(tài)函數(shù)進行處理以獲取數(shù)據(jù)的高頻項和低頻項,并將其輸入最小二乘支持向量機中進行加權(quán)處理,輸出瓦斯?jié)舛阮A(yù)測結(jié)果。通過掘進工作面模擬裝置進行瓦斯?jié)舛阮A(yù)測模擬試驗,并在某煤礦掘進工作面進行現(xiàn)場試驗,結(jié)果表明:該方法預(yù)測的瓦斯?jié)舛扰c實際測量值非常接近,均方誤差小,表明預(yù)測結(jié)果準確率高;均方誤差波動幅度小,表明適應(yīng)性好,預(yù)測結(jié)果的穩(wěn)定性強;預(yù)測用時短,表明預(yù)測效率高。
關(guān)鍵詞:掘進工作面;瓦斯?jié)舛阮A(yù)測;拉依達準則;Lagrange 插值;經(jīng)驗?zāi)B(tài)分解;最小二乘支持向量機
中圖分類號:TD712 ?文獻標志碼:A
Prediction of gas concentration in coal mine excavation working face
CHEN Xianzhan, SHEN Yicheng, HONG Feiyang, SHI Shen
(College of Civil Engineering, Anhui Jianzhu University, Hefei 230011, China)
Abstract: In current gas concentration prediction methods, there are problems of data processing uncertainty, feature extraction limitations, and prediction bias caused by subjective factors. In order to solve the above problems, a gas concentration prediction method for coal mine excavation working face is proposed. Firstly, laser gas sensors are installed every 1 meter in the return airway of the coal mine excavation working face, forming a sensor network to collect real-time gas concentration data. Secondly, the method searches and removes outliers in the gas concentration data according to the Laida criterion, and uses the Lagrange interpolation polynomial to fill in the missing values in the gas concentration data. Finally, based on removing outliers and filling in missing values in the gas concentration data, the empirical mode decomposition algorithm is used to decompose the gas concentration data into intrinsic mode functions and trend terms. The Hilbert transform is then used to process the intrinsic mode functions to obtain the high-frequency and low-frequency terms of the data, which are then input into the least squares support vector machine for weighted processing to output the gas concentration prediction results. The gas concentration prediction simulation experiment is conducted using a simulation device for the excavation working face, and an on-site test is conducted on a certain coal mine excavation working face. The results show that the predicted gas concentration by this method is very close to the actual measurement value, with a small mean square error, indicating a high accuracy of the prediction results. The small fluctuation of mean square error indicates good adaptability and strong stability of prediction results. Short prediction time indicateshigh prediction efficiency.
Key words: excavation working face; gas concentration prediction; Laida criteria; Lagrange interpolation; empirical mode decomposition; least squares support vector machine
0引言
煤炭開采過程中會產(chǎn)生大量瓦斯氣體。受巷道結(jié)構(gòu)、巖層特性、通風不暢等因素影響,在煤礦掘進工作面瓦斯會在特定區(qū)域積聚[1],造成瓦斯?jié)舛壬?,容易發(fā)生爆炸、中毒等事故[2]。因此,預(yù)測煤礦掘進工作面瓦斯?jié)舛?,有助于采取有效措施降低瓦斯?jié)舛?,保障礦工和設(shè)備的安全。
梁運培等[3]采用樣條插值法填補缺失的瓦斯?jié)舛葦?shù)據(jù),并對其進行無量綱化處理,通過布谷鳥搜索算法尋優(yōu)長短期記憶網(wǎng)絡(luò)中的超參數(shù),建立最優(yōu)瓦斯?jié)舛阮A(yù)測模型;但由于插值法填補缺失數(shù)據(jù)時會引入不確定性和偏差,導(dǎo)致預(yù)測結(jié)果缺乏準確性。賈澎濤等[4]首先通過隨機森林和 Hilbert-Huang 變換方法預(yù)處理瓦斯?jié)舛葦?shù)據(jù),然后利用卷積神經(jīng)網(wǎng)絡(luò)提取數(shù)據(jù)特征,最后基于雙向門控單元神經(jīng)網(wǎng)絡(luò)構(gòu)建瓦斯?jié)舛阮A(yù)測模型實現(xiàn)濃度預(yù)測;但該方法對于序列數(shù)據(jù)的時間特征提取存在一定的局限性,從而影響預(yù)測的準確性。劉瑩等[5]對瓦斯?jié)舛葦?shù)據(jù)進行融合和缺失值處理,利用特征衍生得出交叉項及高頻項特征,并確定隱藏層維度,通過長短期記憶網(wǎng)絡(luò)模型訓練完成瓦斯?jié)舛阮A(yù)測;然而隱藏層維度選擇錯誤會導(dǎo)致模型欠擬合或過擬合,限制預(yù)測性能的提升。 Wang Zhiming 等[6]在考慮時間滯后特性的情況下,提出了多變量灰色預(yù)測模型來預(yù)測礦井瓦斯?jié)舛龋⑼ㄟ^引入時間延遲參數(shù)來完善驅(qū)動項序列的分析;但在識別時間延遲參數(shù)和相關(guān)因素序列時,會存在主觀性因素的影響,導(dǎo)致預(yù)測結(jié)果的偏差。
本文提出了一種煤礦掘進工作面瓦斯?jié)舛阮A(yù)測方法。采用激光瓦斯傳感器采集瓦斯?jié)舛葦?shù)據(jù),并根據(jù)拉依達準則剔除異常數(shù)據(jù),再通過插值方式填補數(shù)據(jù)缺失部分,確保數(shù)據(jù)的連續(xù)性和一致性;結(jié)合經(jīng)驗?zāi)B(tài)分解(Empirical Mode Decomposition,EMD)和最小二乘支持向量機(Least Squares Support Vector Machine, LSSVM),實現(xiàn)對瓦斯?jié)舛鹊木珳暑A(yù)測。
1數(shù)據(jù)采集與處理
1.1數(shù)據(jù)采集
在煤礦掘進工作面回風巷每隔1 m距離布設(shè)1個激光瓦斯傳感器[7],確保能夠全面監(jiān)測瓦斯?jié)舛茸兓2荚O(shè)的傳感器可看成一個傳感器網(wǎng)絡(luò),不同傳感器之間能夠?qū)崟r傳輸數(shù)據(jù),迅速獲取巷道各處瓦斯?jié)舛葦?shù)據(jù)。采集的瓦斯?jié)舛葦?shù)據(jù)集合為
Q ={xi }l ?????(1)
式中:xi 為第 i(i=1,2,… , m,m 為瓦斯?jié)舛葦?shù)據(jù)總數(shù))個瓦斯?jié)舛葦?shù)據(jù);l 為單位采樣時間。
1.2數(shù)據(jù)處理
煤礦掘進工作面結(jié)構(gòu)復(fù)雜,存在瓦斯分布不均勻、不穩(wěn)定問題,導(dǎo)致傳感器采集的瓦斯?jié)舛葦?shù)據(jù)出現(xiàn)異?;蛉笔В瑫苯佑绊懲咚?jié)舛阮A(yù)測結(jié)果的準確性。
1)根據(jù)拉依達準則搜索并剔除瓦斯?jié)舛葦?shù)據(jù)中的異常值[8-9]。拉依達準則是指先假設(shè)1組檢測數(shù)據(jù)只含有隨機誤差,對其進行計算處理得到標準差,按一定概率確定一個區(qū)間,超過這個區(qū)間的誤差,則不屬于隨機誤差而是粗大誤差,含有該誤差的數(shù)據(jù)應(yīng)予以剔除。
標準差用來描述數(shù)據(jù)集合中數(shù)值的離散程度或分散程度,可評估瓦斯?jié)舛葦?shù)據(jù)的變異程度。瓦斯?jié)舛葧r間序列標準差為
β=(‘(xi ??(x))2/m ?(2)
式中?(x)為瓦斯?jié)舛葦?shù)據(jù)均值。
當瓦斯?jié)舛葧r間序列標準差β小于瓦斯?jié)舛葧r間序列殘差絕對值的1/3時,判斷為瓦斯?jié)舛葦?shù)據(jù)異常值。
β<|xi ??(x)|/3 ?(3)
2)通過 Lagrange 插值法填補瓦斯?jié)舛葦?shù)據(jù)缺失值。根據(jù)傳感器采集到的瓦斯?jié)舛葦?shù)據(jù)推導(dǎo)出 Lagrange 插值多項式[10-11],從而填充缺失值,獲取完整的瓦斯?jié)舛葦?shù)據(jù)。構(gòu)建 Lagrange 插值多項式:
式中 L(xi)為 Lagrange 基函數(shù)。
利用 Lagrange 插值多項式填補瓦斯?jié)舛葦?shù)據(jù)缺失值,插值填補后的瓦斯?jié)舛葦?shù)據(jù)集合為
2瓦斯?jié)舛阮A(yù)測
煤礦掘進工作面環(huán)境不斷受掘進、爆破等因素影響,導(dǎo)致瓦斯?jié)舛葦?shù)據(jù)呈現(xiàn)出非線性和非平穩(wěn)特點,傳統(tǒng)的線性回歸方法往往無法很好地擬合這類數(shù)據(jù),導(dǎo)致預(yù)測誤差較大。因此,針對瓦斯?jié)舛葦?shù)據(jù)的特點,以剔除異常值并填補缺失值后的瓦斯?jié)舛葦?shù)據(jù)為基礎(chǔ),結(jié)合 EMD 和 LSSVM 算法進行瓦斯?jié)舛阮A(yù)測。
利用 EMD 算法對處理后的瓦斯?jié)舛葦?shù)據(jù)進行多次迭代分解,獲取本征模態(tài)函數(shù)(Intrinsic Mode Function,IMF)和趨勢項[12-14]。
Q′= vh (xi )+ th (xi )Z(xi ) (6)
式中:vh (xi )為第 i 個瓦斯?jié)舛葦?shù)據(jù)的第 h(h=1,2,… , NIMF,NIMF 為 IMF 數(shù)量)個 IMF 分量;th (xi )為第 i 個瓦斯?jié)舛葦?shù)據(jù)的第 h 個趨勢項。
應(yīng)用 Hilbert 變換[15]對每個 IMF 分量進行處理,獲取瓦斯?jié)舛葦?shù)據(jù)的高頻項和低頻項[16-18]。其中,高頻項反映了瓦斯?jié)舛鹊目焖俨▌?,而低頻項則反映了瓦斯?jié)舛茸兓拈L期趨勢。通過將數(shù)據(jù)分解成這2個部分,可以更準確地識別瓦斯?jié)舛茸兓厔荩M而提高預(yù)測精度。
式中:Hxi 和Lxi 分別為瓦斯?jié)舛葦?shù)據(jù) xi 的高頻項和低頻項;N 為迭代總次數(shù)[19]。
LSSVM 能夠處理具有非線性特點的瓦斯?jié)舛葦?shù)據(jù),利用 LSSVM 對瓦斯?jié)舛葦?shù)據(jù)的高頻項、低頻項進行加權(quán)處理,實現(xiàn)瓦斯?jié)舛阮A(yù)測[20-21]。LSSVM輸出的瓦斯?jié)舛阮A(yù)測結(jié)果為
f (xi )=αpK(Hxi ; Lxi )+ bp ??(8)
式中:α為第p( p=1,2,… , ε , ε為支持向量機數(shù)量)個支持向量機的正則化系數(shù);K(·)為徑向基函數(shù);bp 為第p個支持向量機的偏置項。
3試驗與分析
3.1模擬試驗
為驗證本文方法的有效性,利用掘進工作面模擬裝置進行試驗。該裝置尺寸(長×寬×高)為10 m×2.5 m×2.5 m。為保證試驗環(huán)境接近礦井掘進工作面現(xiàn)場實際情況,實驗室設(shè)置通風系統(tǒng)并安裝恒溫恒濕設(shè)備,確保試驗環(huán)境中的空氣成分及溫濕度與礦井實際相吻合,使用通風機實現(xiàn)負壓通風。
設(shè)置局部通風機風速為0.6 m/s,巷道內(nèi)空氣初始溫度為25℃, 相對濕度為80%,氧氣含量為21%,初始瓦斯體積分數(shù)為0.5%,瓦斯排放速度為10 m3/min,排放時間為60 min,巷道圍巖初始溫度為20℃, 巷道圍巖和瓦斯在與空氣對流換熱時的熱導(dǎo)率、溫度均是可變的。試驗中開采煤層瓦斯壓力為0.5 MPa,使用激光瓦斯傳感器在不同時刻測定瓦斯?jié)舛?,同時記錄溫度、濕度等環(huán)境參數(shù)。利用流體動力學模擬軟件 Fluent 進行數(shù)值模擬計算,模擬瓦斯的流動和擴散過程。
分別采用本文方法、文獻[3]方法和文獻[4]方法進行瓦斯?jié)舛阮A(yù)測,結(jié)果如圖1所示??煽闯霰疚姆椒ǖ念A(yù)測值比文獻[3]和文獻[4]方法的預(yù)測值更接近瓦斯?jié)舛葘嶋H值,瓦斯?jié)舛阮A(yù)測效果最好。
3種方法的均方誤差如圖2所示??煽闯鑫墨I[3]方法和文獻[4]方法的均方誤差波動幅度較大,而本文方法的均方誤差波動幅度最小,表明本文方法的適應(yīng)性好,預(yù)測結(jié)果穩(wěn)定性強,且本文方法的均方誤差最小,即瓦斯?jié)舛阮A(yù)測結(jié)果準確率最高。
3種方法的瓦斯?jié)舛阮A(yù)測時間見表1。可看出本文方法預(yù)測瓦斯?jié)舛人脮r間低于文獻[3]和文獻[4]方法預(yù)測時間,具有較高的預(yù)測效率。
3.2現(xiàn)場試驗
根據(jù)某煤礦掘進工作面實際巷道尺寸和形狀,
煤礦巷道環(huán)境溫度保持在24℃ , 相對濕度為82%。使用6臺激光瓦斯傳感器測定瓦斯?jié)舛龋扛?0 min 記錄瓦斯?jié)舛?,每次記錄連續(xù)10次的測量數(shù)據(jù)。采用本文方法得到的瓦斯?jié)舛阮A(yù)測結(jié)果見表3,可看出本文方法預(yù)測的瓦斯?jié)舛扰c實際測量值非常接近。
4結(jié)論
1)提出了一種煤礦掘進工作面瓦斯?jié)舛阮A(yù)測方法。根據(jù)拉依達準則搜索并剔除瓦斯?jié)舛葦?shù)據(jù)中的異常值,并通過 Lagrange 插值法填補瓦斯?jié)舛葦?shù)據(jù)缺失值;采用 EMD 算法分解處理后的瓦斯?jié)舛葦?shù)據(jù),有效提取高頻項和低頻項特征,并將其作為LSSVM 的輸入,從而實現(xiàn)瓦斯?jié)舛阮A(yù)測。
2)試驗結(jié)果表明,該方法的預(yù)測值更接近瓦斯?jié)舛葘嶋H值,預(yù)測結(jié)果準確率高;均方誤差波動幅度小,預(yù)測結(jié)果穩(wěn)定性強;預(yù)測用時短,預(yù)測效率高。
3)在未來研究中,考慮將瓦斯?jié)舛葦?shù)據(jù)與其他環(huán)境參數(shù)(如溫度、濕度、氣壓等)融合分析,探索多源數(shù)據(jù)對瓦斯?jié)舛阮A(yù)測的影響,進一步提高預(yù)測精度。
參考文獻(References):
[1]王德忠,朱國宏,王禹,等.基于?GA?LSTM 的綜采面瓦斯?jié)舛阮A(yù)測模型[J].煤炭技術(shù),2023,42(1):219-221.
WANG Dezhong,ZHU Guohong,WANG Yu,et al. Prediction model of gas concentration in fully mechanized mining face based on GA-LSTM[J]. Coal Technology,2023,42(1):219-221.
[2]蘇培東,張睿,杜宇本,等.基于 Kriging 估值法的非煤隧道瓦斯預(yù)測[J].地下空間與工程學報,2021,17(3):953-960.
SU Peidong,ZHANG Rui,DU Yuben,et al. Prediction of gas spatial distribution in non-coal strata tunnel by Kriging method[J]. Chinese Journal of Underground Space and Engineering,2021,17(3):953-960.
[3]梁運培,栗小雨,李全貴,等.基于?CS?LSTM 的工作面瓦斯?jié)舛戎悄茴A(yù)測研究[J].礦業(yè)安全與環(huán)保,2022,49(4):80-86.
LIANG Yunpei,LI Xiaoyu,LI Quangui,et al. Research on intelligent prediction of gas concentration in working face based on CS-LSTM[J]. Mining Safety & Environmental Protection,2022,49(4):80-86.
[4]賈澎濤,張智遠,梁榮,等.基于?PSO?CNN?aBiGRU 的瓦斯?jié)舛阮A(yù)測方法[J].礦業(yè)研究與開發(fā),2021,41(12):76-81.
JIA Pengtao,ZHANG Zhiyuan,LIANG Rong,et al. Gas concentration prediction method based on PSO- CNN-aBiGRU[J]. Mining Research and Development,2021,41(12):76-81.
[5]劉瑩,楊超宇.基于多因素的 LSTM 瓦斯?jié)舛阮A(yù)測模型[J].中國安全生產(chǎn)科學技術(shù),2022,18(1):108-113.
LIU Ying,YANG Chaoyu. LSTM gas concentration prediction model based on multiple factors[J]. Journal of Safety Science and Technology,2022,18(1):108-113.
[6] WANG Zhiming,MIAO Yanzi,LI Shoujun,et al. Prediction of mine gas concentration based on multi- variable time-delayed DOGM(1,N) model[J]. The Journal of Grey System,2022,34(1):70-83.
[7]程浩東,韓萌,張妮,等.基于滑動窗口模型的數(shù)據(jù)流閉合高效用項集挖掘[J].計算機研究與發(fā)展,2021,58(11):2500-2514.
CHENG Haodong,HAN Meng,ZHANG Ni,et al. Closed high utility itemsets mining over data stream based on sliding window model[J]. Journal of Computer Research and Development,2021,58(11):2500-2514.
[8]劉丹丹.基于 EMD 的 GNSS 時間序列異常值探測算法[J].地球物理學進展,2021,36(5):1865-1873.
LIU Dandan. New method of outlier detection for GNSS coordinate time series based on EMD approach[J]. Progress in Geophysics,2021,36(5):1865-1873.
[9]鄭欣彤,邊婷婷,張德強,等. ARIMA 和 LSTM 方法長時間溫度觀測數(shù)據(jù)缺失值插補的比較[J].計算機應(yīng)用,2022,42(增刊1):130-135.
ZHENG Xintong,BIAN Tingting,ZHANG Deqiang,et al. Comparison of ARIMA and LSTM methods for interpolation of missing values of long-time temperature observations[J]. Journal of Computer Applications,2022,42(S1):130-135.
[10]吳哲,黃蓉,田朝薇.時間分數(shù)階?Black?Scholes 方程的重心?Lagrange 插值配點法[J].華僑大學學報(自然科學版),2023,44(2):269-276.
WU Zhe,HUANG Rong,TIAN Zhaowei. Barycentric Lagrange interpolation collocation method for time- fractional Black-Scholes ?equation[J]. Journal of Huaqiao University(Natural Science),2023,44(2):269-276.
[11]鐘凱強,周建平,薛瑞雷,等.一種采用 Lagrange 插值的相貫線簡化算法[J].熱加工工藝,2021,50(15):131-135,140.
ZHONG Kaiqiang,ZHOU Jianping,XUE Ruilei,et al. A intersecting line simplification algorithm using Lagrange interpolation[J]. Hot Working Technology,2021,50(15):131-135,140.
[12]吳虎,孔勇,王振偉,等.基于?EMD 分解與1?D CNN 算法的光纖振動信號的識別[J].激光與紅外,2021,51(8):1043-1049.
WU Hu, KONG Yong, WANG Zhenwei, et al. Recognition of optical fiber vibration signals based on VMD_CNN algorithm[J]. Laser & Infrared,2021,51(8):1043-1049.
[13]劉兵,鄭承利.基于 EMD 特征提取的高頻面板數(shù)據(jù)自適應(yīng)聚類方法[J].統(tǒng)計與決策,2022,38(10):16-20.
LIU Bing,ZHENG Chengli. Adaptive clustering method for high frequency panel data based on EMD feature extraction[J]. Statistics & Decision,2022,38(10):16-20.
[14]韓麗,李夢潔,喬妍.基于低頻波動挖掘和高頻校正的風電超短期預(yù)測[J].電網(wǎng)技術(shù),2022,46(7):2750-2758.
HAN Li,LI Mengjie,QIAO Yan. Ultra-short-termprediction of wind power based on low-frequency fluctuation mining and high-frequency correction[J]. Power System Technology,2022,46(7):2750-2758.
[15]蘇娟,方舒,邢廣進,等.考慮需求價格彈性的?CS? SVM 短期負荷預(yù)測方法[J].江蘇大學學報(自然科學版),2022,43(3):319-324.
SU Juan,F(xiàn)ANG Shu,XING Guangjin,et al. Short-term load forecasting method based on cuckoo search algorithm and support vector machine considering demand price ?elasticity[J]. Journal of Jiangsu University(Natural Science Edition),2022,43(3):319-324.
[16]付樂天,李鵬,高蓮.考慮樣本異常值的改進最小二乘支持向量機算法[J].儀器儀表學報,2021,42(6):179-190.
FU Letian,LI Peng,GAO Lian. Improved LSSVM algorithm considering sample outliers[J]. Chinese Journal of Scientific Instrument,2021,42(6):179-190.
[17]李森娟,張萍,岳大為,等.基于支持向量機的風電機組故障預(yù)測[J].計算機仿真,2022,39(5):84-88,180.
LI Senjuan,ZHANG Ping,YUE Dawei,et al. Fault prediction of wind turbine based on support vector machine[J]. Computer Simulation,2022,39(5):84-88,180.
[18]張皓,李東升.復(fù)分析 Hilbert變換計算理論及非線性檢測準則[J].振動工程學報,2022,35(6):1336-1345.
ZHANG Hao, LI Dongsheng. Hilbert transform calculated by complex analysis theory and its nonlinear detection ?criterion[J]. Journal ?of ?Vibration Engineering,2022,35(6):1336-1345.
[19]王媛彬,李媛媛,韓騫,等.基于?PCA?BO?XGBoost 的礦井回采工作面瓦斯涌出量預(yù)測[J].西安科技大學學報,2022,42(2):371-379.
WANG Yuanbin,LI Yuanyuan,HAN Qian,et al. Gas emission prediction of the stope in coal mine based on PCA-BO-XGBoost[J]. Journal of Xi'an University of Science and Technology,2022,42(2):371-379.
[20]趙偉,陳培紅,曹陽.基于?ACSOA?BP 神經(jīng)網(wǎng)絡(luò)的瓦斯含量預(yù)測模型[J].煤礦安全,2022,53(1):174-180.
ZHAO Wei,CHEN Peihong,CAO Yang. Prediction model of coal seam gas content based on ACSOA optimized BP neural network[J]. Safety in Coal Mines,2022,53(1):174-180.
[21]廖巍.復(fù)雜地質(zhì)條件下瓦斯含量精準預(yù)測研究與系統(tǒng)開發(fā)[J].煤炭工程,2022,54(5):142-145.
LIAO Wei. Accurate prediction of gas content and system ?development ?under ?complex ?geological conditions[J]. Coal Engineering,2022,54(5):142-145.