2nd Internatonal Conference on Computer Engneerng, Informaton Scence & Applcaton Technology (ICCIA 207) Research of Dspatchng Method n Elevator Group Control System Based on Fuzzy Neural Network Yufeng Da a, Yun Du b School of electrcal engneerng, Hebe Unversty of Scence and Technology, Shjazhuang 0008, Chna a da39387@63.com, b37926909@qq.com Keywords: Elevator group control system, Comprehensve evaluaton functon, Fuzzy neural network, Dspatchng method. Abstract. Elevator group control system (EGCS) s a complex optmzaton system to wth multobjectve, stochastc and nonlnear characterstcs. It s hard to descrbe EGCS wth exact mathematc model and to ncrease the capablty of the system wth tradtonal control method. The fuzzy control technology and neural network technology are combned n ths paper and a dspatchng method appled to varous passenger traffc condtons s proposed. The comprehensve evaluaton functon of traffc sgnal s establshed and the rght heavy of every evaluaton factor (watng tme, rdng tme, energy consume) s studed by the neural network, so the elevator s dspatched optmally. The result of smulaton shows that ths method realzes reasonable elevator dspatchng under varous passenger traffc condtons and ndcates the valdty of ths method.. Introducton Wth the contnuous development and progress of human socety, the number of hgh-rse buldngs has become a symbol of a cty's development and prosperty, and the elevator s an ndspensable part of the hgh-rse buldng, so the elevator technology rapdly developed. Elevator n the control technology gradually developed from a lft to the centralzed control of elevators. Frst of all, through the analyss of buldng passenger flow, the traffc pattern of elevator group s classfed. Fuzzy neural network s used to dentfy the traffc pattern of elevator group. Accordng to the dentfcaton of the system to determne the elevator group s currently n the traffc model. Then, the fuzzy neural network s used to calculate the credblty of the elevator n response to the call sgnal of the elevator, and the elevator wth the hghest relablty s selected to fnally complete the servce [6]. 2. Classfcaton and dentfcaton of traffc patterns 2. The classfcaton of the traffc patterns. Accordng to the dfferent needs of the buldng and the people who work n the buldng are also dfferent, so there are a varety of traffc patterns correspondng to them [2].In ths paper, the hgh-rse offce buldng as the research object, ac-cordng to the fxed tme traffc flow changes, the elevator traffc model s dvded nto 6 categores: up peak traffc pattern, down peak traffc pattern, dle traffc mode, the mddle layer of busy mode, sngle mode and double layer busy mode. 2.2 Fuzzy neural network. After the classfcaton of traffc patterns, we need to dentfy the current traffc pattern of elevator group. In ths paper, fuzzy neural network s used to dentfy. In the fuzzy neural network structure, the number of layers s fxed, whch are nput layer, fuzzy layer, rule layer, ntegrated layer and output layer []. () Input layer: the nput layer s the frst layer of the fuzzy neural network topology, each node n the nput layer represents an nput varable, and the number of neurons s equal to the number of nput varables. Copyrght 207, the Authors. Publshed by Atlants Press. Ths s an open access artcle under the CC BY-NC lcense (http://creatvecommons.org/lcenses/by-nc/4.0/). 247
(2) Fuzzy level: each nput varable should be defned by ther fuzzy subsets, and the membershp functons of all fuzzy subsets are calculated by ths layer. (3) Rule layer: the node s a rule node, whch represents the logc rules, and each node has the logc operaton functon. (4) Synthess layer: ths layer node performs a fuzzy "or" operaton to synthesze rules wth the same result. () Output layer: Ths layer s also called the ant-fuzzy layer. 2.3 Traffc pattern recognton. The hgh-rse offce buldng as an example, n mnutes as a unt of tme, characterstc values are: unt tme total passenger traffc, passenger flow, passenger flow nto the left, the largest passenger flow, the mddle floor large mddle floor. After normalzaton, the nput varables belong to [0, ]. The fuzzy neural network needs to be traned before pattern recognton, and the recognton can be carred out after the tranng []. Because the tranng nput data need to be normalzed, the nput value can only be n the [0, ], so the nput sample nterval value s 0.2. The structure of fuzzy neural network n traffc pattern recognton s 3-8-26-24-4 and 2-8-6-2-3. Accordng to the prevous learnng methods to tran the network, the fuzzy rule extracton threshold s =0.0. Network tranng results s as Table 2.. Table Network tranng results Learnng Error back Error precson Network type Rule base effcency propagaton Frst step 0.0 26 23 0.000997 Second step 0.0 2 37 0.00099 It can be seen from the table that the rules are deleted and merged by compettve learnng. Now, the traned fuzzy neural network s used to dentfy the traffc pattern of the offce buldng. The smulaton results are as Fg -4. Fg. Traffc statstcs of the hall Fg.2 The total traffc volume of the offce buldng and ts specal floor traffc characterstcs 248
Fg. 3 The proporton of the up peak pattern Fg. 4 The proporton of the downward peak mode Fgure and Fgure 2 s a traffc statstcal feature curve of a workng day n an offce buldng, and the statstcal tme nterval s mnutes. Usng fuzzy neural network to dentfy traffc n Fg and Fg 2, t dentfes the proporton of up peak pattern and down peak pattern as shown n Fg 3 and Fg 4. 3. Elevator group control system ladder algorthm The elevator group dspatch algorthm s a mult-objectve optmzaton. It needs to consder the entre populaton and the average watng tme, average operaton tme of all people n the elevator, the tme of watng and the proporton of the overall energy consumpton of elevator group and other factors. The elevator group control system needs to control multple elevators to respond to the varous callng sgnals at the same tme, so the general mathematcal model cannot be acheved [4]. We wll contnue to use fuzzy neural network to acheve mult-objectve optmzaton. Total passenger flow Passenger flow out of the buldng Maxmum floor passenger flow Second floor passenger flow dle traffc pattern Interlayer traffc patterns When the layer rato s large, the second network s dentfed Second network dentfcaton Total passenger flow Frst network dentfcaton Passenger flow nto the buldng AWT up peak traffc pattern down peak traffc pattern Sngle mddle layer pattern ART The weghts of AWT, ART and RNC are determned accordng to the proporton of each traffc pattern RNC Double mddle layer pattern HCWT Callng sgnal Car nformaton S AWT Collect nformaton, calculate and judge CV maxhcwt GD fuzzy neural network S S ART Weghted average output S RNC Fg. Schematc dagram of elevator group control 249
Ths elevator performance of multple ndcators, the most mportance s the passengers average watng tme AWT, the average passenger elevator tme ART and energy consumpton RNC these three ndcators. Objectve functon: () S AWT S AWT ART S ART RNC S RNC Where S s the confdence of the average watng tme of the elevator, S s the confdence of the average rde tme, and S s the confdence of the energy loss. Indcates the relablty of the -th elevator response call. AWT ART RNC s the weghtng factor, and the sum s.the weghtng factor s determned by the pattern of traffc flow dentfed n the prevous chapter. Wth the formula (), the relablty of the n elevators n response to the call sgnal can be obtaned, and then the elevator wth the hghest degree of confdence s selected to respond to the call sgnal. (2) se max( s, s2,..., sn) When a call sgnal generated, the system wll mmedately through the HCWT, maxhcwt, CV and GD these four parameters to calculate the average watng tme for each elevator elevator AWT, the average passenger elevator tme ART and energy consumpton RNC ndcators [3]. The HCWT generates a call sgnal to respond to the elevator's arrval tme and watng tme for the layer. maxhcwt s the maxmum watng tme for all call sgnals that a elevator responds to.cv rep-resents the ablty to respond to future calls. GD represents the shortest dstance between the newly generated call sgnal floor and all sgnal floors that the elevator responds to [7]. The ntra-vector components are the weghts of AWT, ART, and RNC, respectvely. The traffc rato of the up peak traffc pattern, down peak traffc pattern, nterlayer traffc pattern, dle traffc pattern, sngle mddle layer pattern and double mddle layer pattern are A, A2, A3, A4, A, A6 respectvely. Also there A A2 A3 A4 A A6. Then the fnal calculaton of the AWT, ART, RNC weghts are as follows: (3) A A2 2 A3 3 A4 4 A A6 6 The fuzzy neural network model s establshed. The nput varables are HCWT, max HCWT, CV and GD, and the output varables are AWT, ART and RNC. Accordng to dfferent modes of transport, t can adjust the weght of the three confdence, calculate the fnal credblty. By comparng the relablty of the elevator n response to the call sgnal, select the most relable elevator to complete the escalator. AWT ART RNC 4. Elevator group control system smulaton In ths paper, accordng to the actual stuaton of a certan offce buldng statstcs tranng data, s got consderng the tme perod and the mode of traffc flow. The sample data were collected n 0 groups. The data of the 40 groups were tranng data, and the remanng 0 groups of data were test data. The error precson s controlled wthn the preset threshold by repettve learnng tranng, e <0.00. Fgure 6 shows that n the tranng samples, the center and wdth of the membershp functon of the nput vector of the fuzzy neural network by the error back propagaton algorthm s optmzed. If you follow the checklst your paper wll conform to the requrements of the publsher and facltate a problem-free publcaton process. Tranng network error curve error 0.8 0.6 0.4 0.2 0 0 20 2 30 Number of tranng steps 3 40 4 0 Fg. 6 Varaton of network error performance durng tranng 20
It can be seen that after 30 teratons, the gradent curve has stablzed, ndcatng that the tranng of the fuzzy neural network has been completed. After the end of the network tranng, use the test sample to test the output error range, as shown n Fgure 7. -3 Test sample error curve x 0 error 4 3 2 0 2 3 4 6 Test samples 7 8 9 0 Fg. 7 Error result of test sample data It can be seen from Fgure 7 that the errors are wthn the set range, ndcatng that the fuzzy neural network tranng s completed.. Concluson In ths paper, a 20-story buldng elevator group as an example of the smulaton, the buldng has sx elevators, each elevator rated people. AWT, ART and RNC weghts are calculated accordng to the current traffc pattern. The parameters of AWT, ART and RNC are calculated by HCWT, max HCWT, CV and GD. AWT, ART and RNC are calculated accordng to the current traffc pattern., RNC three of the credblty, accordng to the traffc model to calculate the fnal credblty. By comparng the sze of the fnal relablty of the sx elevators, select the most relable elevator to send ladder. In ths paper, the smulaton experment s carred out by usng MATLAB software, and the expermental results are obtaned. How to adjust the weghtng system reasonably and make the algorthm more optmzed. References []. Bao H: Research on Fuzzy Neural Network Based recognton of traffc patterns of Elevator Group Control System and mult-target dspatchng algorthm.[d] Tongj Unversty, Chna 2007. [2]. Guo JL: Research on Elevator Group Control System Based on Fuzzy Neural Network. [D]Northeastern Unversty, Chna 203. [3]. Lu XY,He P,Chang JH : Research of Dspatchng Method n Elevator Group Control System Based on Fuzzy Neural Network. Computer Technology and Development. (2008)8():220-222 [4]. L XL: Studes of Elevator Group Control System Based on Fuzzy Neural Network. [D] Soochow Unversty, Chna 2006. []. Tang HY: Research on optmal Control of Elevator Group Based on Fuzzy Neural Network. Harbn Insttute of Technology, [D] Chna 2006. [6]. Wang QX,Jn X :Research on Elevator Group Control Algorthm based on Fuzzy Neural Network. Chna New Technologes and Products20,9(2):4- [7]. Zhang SQ Applcaton Research of Fuzzy Neural Network n Elevator Group Control system. Northeastern Unversty, [D] Chna2009. 2