Prediction of oxygen concentration and temperature distribution in loose coal based on B P neural ne Prediction of oxygen concentration and temperature distribution in loose coal based on B P neural ne

Prediction of oxygen concentration and temperature distribution in loose coal based on B P neural ne

  • 期刊名字:矿业科学技术(英文版)
  • 文件大小:844kb
  • 论文作者:ZHANG Yong-jian,WU Guo-guang,X
  • 作者单位:School of Chemical Engineering and Technology
  • 更新时间:2020-06-12
  • 下载次数:
论文简介

.e Science DirectMININGSCIENCE ANDTECHNOLOGYELSEVIERMining Science and Technology 19(2009)0216-0219. elsevier. com/locate/jcumtPrediction of oxygen concentration and temperature distributionin loose coal based on bp neural networkZHANG Yong-jian, WU Guo-guang, XU Hong-feng, MENG Xian-liang, WANG Guang-youSchool of Chemical Engineering and Technology, China University of Mining Technology, Xuzhou, Jiangsu 221008, ChinaAbstract: An effective method for preventing spontaneous combustion of coal stockpiles on the ground is to control the air- flow inloose coal. In order to determine and predict accurately oxygen concentrations and temperatures within coal stockpiles, it is vital toobtain information of self-heating conditions and tendencies of spontaneous coal combustion. For laboratory conditions, we deour own experimental equipment composed of a control-heating system, a coal column and an oxygen concentration andmperature monitoring system, for simulation of spontaneous combustion of block coal (13-25 mm)covered with fine coal (0-3nm). A BP artificial neural network(ANN)with 150 training samples was gradually established over the course of our experiment.network was applied to predict the trend on the untried experimental data. The results show that the oxygen concentration in thecoal column could be reduced below the minimum still able to induce spontaneous combustion of coal-6% by covering the coalpile with fine coal, which would meet the requirement to prevent spontaneous combustion of coal stockpiles. Based on the prediction of this anN, the average errors of oxygen concentration and temperature were respectively 0.5% and 7C, which meet actualtolerances. The implementation of the method would provide a practical guide in understanding the course of self-heating andspontaneous combustion of coal stockpilesKeywords: loose coal; neural network; spontaneous combustion of coal; oxygen concentration; temperature; prediction1 Introductionaffect its combustion characteristicgn-sand would notThe essential elements of spontaneous combustionIn our study, under laboratory conditions, appro.of coal consist of physical and chemical absorption priate testing equipment was designed to examinand chemical reactions between oxygen and variousoxygen concentrations and temperature distributionexternal active structures of coal molecules, ulti- of coal columns when different thicknesses of the finemately releasing heat. Self-heating and spontaneous coal covered lump coal and a corresponding BPcombustion of coal happen only on the condition that(backpropagation) artificial neural network was es-the amount of heat released is larger than the amount tablished. This net was obtained through repeatedof heat loss, causing the temperature of the loose coal training and became available to be called on for pre-stockpile to rise when the supply of oxygen is large diction purposes. The net was based on untried ex-enough. So, an effective method for preventing spon- perimental data, but has shown satisfying resultstaneous combustion of coal stockpileson the groundis to control the air-flow in loose coall-2. These days, 2 Testing equipment and processcoal stockpiles is a common method to prevent spon-taneous coal combustionwever, the large coa2.1 Design principle of equipmenstockpiles stored on the ground are market coal whichUncover a field fire and it will be discovered thatwould be used for power generation. The inorganic the size of its internal coal is large the amount ofibitors will affect the nature of coal(such as its powered coal is small and that some effective chancalorific value)and lower its economic efficiency. As nels中国煤化工mace, have beera result, using fine local coal to cover some inflam- forme-This is due to theable regions of coal stockpiles can reduceCNMHGthe accumulationCorrespondingauthorTel+86-516-83591053:e-mailaddressrentiandi@163.comZHANG Yong- jian et alPrediction of oxygen concentration and teremperature distributionof intermal heat in a vertical direction of large coal 2.3 Testing procedurestockpiles and intensifies the reaction between coaland oxygen. Therefore, covering the stockpile with anFirst of all, the sealed raw coal(three kinds of coal)appropriate layer of fine coal can effectively hinderwas sieved and portioned into two sizes: 0-3 mm and13-25 mm. The bottom of the coal column was cov-the flow of air and achieve the purpose of preventing ered respectively with a layer of 0 cm, 20 cm, 35 cmspontaneous combustion.he tendency of spontaneous coal combustion is and 50 cm deep fine coal and topped with lump coaldependent on the release of heat and oxidation. It ise 13-25 mm thick portion. Then the regulatingtransformer (testing voltage 120 V) began heatingvital to become informed about the self-heating con- and the system was programmed to record temperaditions. The tendency of spontaneous combustion ofture automatically every 10 seconds. The oxygencoal can be deduced by examining the oxygen con- concentration was determined by drawing gas at thecentration and temperature within coal stockpiles. Wedesigned our own experimental coal column to check measuring points every hour. When the maximumthe condition of both factors during the spontaneouscombustion of the coal column covered with fine coalxygen concentration at each point had not obviouslyIn the laboratory, the heat source in the coal stockpiledecreased, we stopped the heating. The entire simula-was simulated by a heat panel which was composedtion of spontaneous combustion of coal lasted aboutnine hoursof some heat resistancy units controlled by an adjust-able transformer to accelerate the process of spontneous combustion of coal3 Results and prediction2.2 Testing equipment3.1 AnalysisThis testing equipment was composed of a heatingThe testing process involved some complicatedsystem, a column for testing coal and a monitoring reactions between coal and oxygen, heat transfer andsystem,shown in Fig. 1. The heating system is made oxygen transmission Under our experimental condiup of AC power, an adjustable transformer and a tions, the heat from the constant power heating panelheating panel. Temperature sensors, conversion mod- heat transfer between coal and coal, coal and gas, theules, a computer and a gas chromatograph constitute heat released from coal oxygen reactions and a smallthe monitoring system. The main body of the coal amount of heat loss, ultimately formed the tempera-column is a 190 cm long, 25 cm diameter steel pipe, ture distribution in the coal column. The drop inwith 18 temperature measuring points set evenly on oxygen concentration at the measuring points of theboth sides and 9 oxygen concentration measuring coal column was due to the consumption of oxygen innts. The bottom of the coal column, pierced with the chemical reaction when coal was heated. In theholes, was covered by a screen which prevented high-temperature parts of the coal column, the air wasleakage of fine coal and guaranteed that air could liable to flow into it from the holes in the bottom offlow through the fine holes in the spontaneous com- the column. It is because the rapid speed of oxygenbustion of coal. It simulated the chimney effect dur- combustion and the low density in high-temperaturespontaneous combustion of large coal stock- parts, which led to the chimney effect. However, thehe head of the coal column was capped with a voidage(the ratio between the volumn of all gapspierced end. The internal wall of the coal column was during coal without containing inner holes and thepacked by asbestos material as an insulation layer to coal entire volumn)of fine coal is less than that of thereduce heat loss. The testing column was fixed in a bump and the permeability of fine coal is poor. So therotating bracket, facilitating the loading and unload- rate of oxygen consumption was lower than the rateing of coalof input, resulting in strong resistance of the fine coal,ultimately forming the oxygen concentration distribu-Thus, the effects of heat transfer and oxygentransmissioncoal oxygen reaction are so com-plicated, that general mathematical models could notComputer H metingsatisfy their requirements-.Under conditions oftesting the constant heating, and if the fundamentalHeat panelfactors, such as heating time, the location of measur-Ing中国煤化工 e determined, thedistrilC MH Galysis of test dataindicated non-liner relations between the basic factorsFig 1 Experimental system for measuring oxygenand the objective factors. Therefore, in order to avoidconcentration and temperaturthe various complex micro factors which affect theMining science and TechnoVol 19 No. 2transfer process, using an artificial neural network in Table 1. The input variables contained activationchecks the oxygen concentration and temperature energy(ae)4 and artificial density(AD)whichdistribution of the coal columnmeasured the tendency of spontaneous combustion of3. 2 BPnetwork predictioncoal, as well as heating time, relative position(RP)ofthe measurement points, voidage and the ratio of fine3.2.1 Factor selectioncoal thickness(RT), as experimental conditions. Con-In the vertical column, the obvious changes of sidering the impact of trial errors, the ratio of fineoxygen concentration and temperature distribution coal thickness was defined as the ratio between theshowed up at the top of the heating panel because of thickness of fine coal and the thickness of the coalthe chimney effect, Considering the actual situation column. The relative position of the measurementand neural network input samples, only eight heat points was the ratio between the coordinates of themeasuring points at the top of column were selected points(heating panel as origin) and the top length ofas network training samples. Then a total of 150 the column(110 cm). The output consisted of thegroups of representative data at every measuring measurements of oxygen concentration and temperaselected part of thesTable 1 Selected experimental results and training sample dataConcentration(%)T(CTime(minRPAD (g/m)AE(/mol)170.31200.181.2420905111976215126371515.121262.4448.342026026.8116.5124l0010681.2640.524026In the neural network training process, a BP net-ork of two-hidden layer 4 was selected in order tospeed up the convergence rate and improve the accu-racy of the network. The first and second hidden lay-9 vectors normalization, P as input matrix, T as outputmers contained fifteen neurons and the output layer twoLpn, minp, maxp, tn, mint, maxt]=premnmx(P,T);neurons. To avoid the negative impacts of the abso-create a new BP neural networklute size of the data on the results, we used the fol-net=newff(minmax(pn), 15, 15, 2], (tansig, tansiglowing method for the normalization treatment be-tween input data and output data:(actual value-thenet=init(net);minimum)/(the maximum- the minimum)=normalized%o weights and bias ofvalue5). After obtaining these results, the actual valinput Weights=net IW(1, 1); inputbias=net. b(1);ues were derived from anti-normalization. normali-%o weights and bias of present network layerzation and anti-normalization were respectivelyayer Weights=net Lw(2, 1 ); layerbias=net. b]:achieved by the premnmx and postmnmx functionVarious definitions and values are shown in table 29 set training parametersram. Ir= 0.053.2.2 Prediction results中国煤化工We selected arbitrarily 30 groups of untried dataand called the trained BP network to predict, thenCNMHGcompared the network output of oxygen concecall TRAINSCG arithmetic to train BP networktions and the experimentally obtained values, shown[net, tr]=train(net pn, tn);in Fig. 2a. The results of both are very close; theZHANG Yong- jian et allargest error is only 4%and the average error about average error about 7oC. Both errors are within per0.5%. We then compared the network output of the mitted range and we conclude that we obtained satis-temperature and the experimental values; these are fying prediction resultsshown in Fig 2b. The largest error is about 20'C and的00Prediction dataPrediction data(a)Oxygen concentration(b)TeFig 2 Prediction of oxygen concentration and temperature distribution4 ConclusionsRefe1)During the entire trial period, the oxygen[1] Schmal D Spontaneous Heating of Stored Coal: Chementration at each point not covered with fineistry of Coal Weathering. Amsterdam: Elsevier, 1989:133-215.was similar at the outside(atmosphere), but the oxy- [2] Carpenter D L, Sergeant G D. The initial stages of thegen concentration at the points covered with fine coaloxidation of coal with molecular oxygen Ill: effect ofwas clearly reduced to below 6%, which is the miniarticle size on rate of oxygen consumption. Fuel, 1966mum concentration to induce spontaneous coal com-45):311-327[3] Wang X S Mine Disaster Prevention Theories and Tech-2)Under the testing conditions of constant heating,niques. Xuzhou: China University of Mining and Tech-logy Press, 1986. (In Chinese)non-linear relations appeared between the outputs4 Fierro V, Miranda J L Romero C, Andres J M, Arriaga Aincluding oxygen concentration and temperature disSchmal D, visser G H. Prevention of spontaneous com-tribution and the inputs including heating time, posibustion in coal stockpiles. Fuel Processing Technology,tion of the measuring points, activation energy of coal,1999(59):23-34.etc. A satisfactory BP network which can meet the [5] Palmer A D, Cheng M, Goulet J C, Furimsky ERelationerror requirement was obtained through training andbetween particle size and properties of some bituminousoals.Fuel,199069):183-188.debugging of a large number of samples. When ex- [6]Liu J, Zhao F J. Effect of particle size on spontaneousperimental conditions change, it is possible to call thecombustion tendency characterization of coal. Journal ofnetwork, which had been trained, to save theLiaoning Technical University, 2006, 25(1): Imation on a disk in order to make a prediction.[7 Xu J C. Determination Theory of Coal Spontaneousmethod provides a theoretical guide for similar testsCombustion Zone. Beijing: China Coal Industry Pub-lishing House, 2001. (In Chineseto shorten the test cycle8] Wen H, Xu J C, Li L, Dai A P. Analysis of self-ignite3)The use of a neural network to examine the con-heat accumulating process and its effect factor, Jourmalditions of spontaneous coal combustion has providedof China Coal Society, 2003, 28(4): 370-374. (In Chi-a opportunities for preventing spontaneous combus-nese)tion of coal stockpiles by covering them with fine [9] Hull A, Lanthier JL, Agarwal P K. The role of the diffucoal In actual operation, it is necessary to accumulatesion of oxygen in the ignition of a coal stockpile in con-fined storage. Fuel, 1997(76): 975-982macro data of long term observations, affecting the (10 Rosema A, Guan H Veld H. Simulation of spontaneousprocess of spontaneous coal combustion in order tocombustion to study the causes of coal fires in thebuild neural networks based on actual conditions toRujigou Basin. FueL, 2001(80): 7-16.predict oxygen concentration and temperature of coal [11] Brooks K, Glasser D. A simplified model of spontaneousstockpiles and to deduce the course of self-heatingombustion in coal stockpiles. Fuel, 1986(65): 1035-and s[12] Lu W, Wang D M, Zhong XX, Zhou F B Tendency ofAcknowledgementsspontaneous combustion of coal based on activation en-ergy. Journal of China University of Mining Technol-y2006,35(2:201-205.( n Chinese)The authors thank the experimental team who [13) Chen X D, Stott JB. Oxidation rates of coals ashelped to load and unload a great deal coal during thetrials. Special thanks are due to Mr. Wu Guoguang中国煤化工585who provided invaluable guidance on the practical [14]aspects of coal stockpile handling and storage opera-LLD.CNMHS.on Chi sry laton.[15] Zhou K L, Kang Y H Neural Network Model and MAT-LAB Simulation Program Design. Beijing: TsinghuaUniversity Press, 2005. (In Chinese

论文截图
版权:如无特殊注明,文章转载自网络,侵权请联系cnmhg168#163.com删除!文件均为网友上传,仅供研究和学习使用,务必24小时内删除。