Avaiabe onine www.jocpr.com Journa of Chemica and Pharmaceutica Research, 204, 6(7:93-936 Research Artice ISSN : 0975-7384 CODEN(USA : JCPRC5 BP neura networ-based sports performance prediction mode appied research Jian Wang Institute of Physica Education, Shihezi University, Shihezi, Xinjiang, China ABSTRACT Tae previous Oympic Games athetics performance as reference, use neura networ agorithms reative strong sef faut-toerant abiity and sef-training earning abiity as we as other good advantages to construct BP neura networ mode, carry out specific appication and verification on the mode, and research on No. performance by estabishing and appying neura networ. Research resut shows that BP neura networ can be used for sports performance predicting, so neura networ mode provides extremey wide deveopment space for sports performance prediction mode researching. Key words: Neura networ, prediction mode, BP agorithm, Matab simuation INTRODUCTION According to previous performance, it maes prediction on performance that to be generated, it is generay used to major sporting events, predict future sports competitive eves is particuar more important for athete, so the sports performance prediction becomes more and more important, but there are many inds of modern prediction methods, from which neura networ is more popuar in contemporary prediction and anaysis aspect [-5]. Regarding sports aspect each ind of events prediction research, ots of peope have made efforts, and got achievements, which provides beneficia conditions for schoars from a circes of society maing research and provides impetus for scientific prediction deveopment [6-9]. Such as: Zhong Wu and others constructed shot specia performance prediction in 2004, meanwhie they got that its accuracy is obvious higher than mutipe inear regression mode [0-2]; Wang Zong-Ping and others made prediction on men swimming by neura networ in 2006 and got higher accuracy; There were schoars had ever stated 2N + pieces of hidden ayers mode configuration with N as input nodes numbers after ANN function singe hidden ayer in 987; Cyrbeno as eary as 988 had ever proposed structura point adopting S type function, he pointed out a hidden ayer was used to sove artificia distribution probems, and two hidden ayers were using input graphs to output functions. After that, he mentioned any cosed intervas one continuous function coud use BP neura networ mode to approach in 989 [3, 4]. The paper on the basis of previous research achievements, it researches on sports performance infuence factors, and uses BP neura networ to predict sports performance, and combines with exampes to state the method impementation and appication, the resut shows it wi have important effects on estabishing neura networ prediction mode to sports aspect appied researches. BP NEURAL NETWORK THEORETICAL FORMING Regarding sports aspect performance, it can be divided into two inds, in genera tota performance is a ind of emotiona type that focuses on entirety phenotype, and sub performance is rationa that focuses on detais, but actuay tota performance and sub performance aways appear uneven status, and then it needs BP neura networ to 93
Jian Wang J. Chem. Pharm. Res., 204, 6(7:93-936 expore their mutua reations, so that it forms neura networ mode, after reative training, we ony input sports performance prediction into the neura networ mode then it wi cacuate mass tota performance, it improves performance accuracy in this way. Hierarchica neura networ is a feed forward mutipe-ayer networ mode and it is one ind of two main connection ways, its minimum unit is using nerve ce to connect and estabish output ayer, input ayer and hidden ayer three inds of modes BP neura networ mode, its structures is as Figure show: Figure : Neura networ theory process Though there are no any connections among them, their nerve ces are mutua correated. The agorithm earning process is composed of two directions that are respectivey forward direction process and reverse two propagation processes, from which, forward propagation is: net j ω o In above formua, represents number of ayers, is expressed by nodes, the input is the sampe, then: o, and when output j ( pieces of units Reverse propagation: o f ( net (2 If input unit node is j, then: Among them, use j o y as actua output unit which is expressed by y (3 2 If input unit node is not j, then: Revise weight: ωij ω IJ µ ω Here: ij, µ f 0 δ t δ m t δ ω ω ( y y f ( net f + + / t m mj ij δ o ( net / t (4 (5 (6 932
Jian Wang J. Chem. Pharm. Res., 204, 6(7:93-936 ω ij N K ωij (7 Among them, the process from input ayer to hidden ayer and then transfer to output ayer is information forward direction propagation, but once end cannot get corresponding output resut, it wi automaticay turn to reverse propagation, one nerve ce is expressed by foowing formua: u m t w i x t (8 ( u b y f + b u In above formua, nerve ce unit threshod vaue is, in inear combination, input signa output is, output signa y is w, protruded weight is i x, input signa is, and meanwhie activated function is F (, corresponding function formua is as foowing: f ( v + e Due to BP neura networ nerve ce does not change; corresponding mode is as Figure 3. v (0 (9 Figure 3:Neura networ operation process For BP nerve ce, its input end is: net x w + x w + L + x w n n 2 2 ( In above formua, connection weight vaue: w,, w2, L wn x, x,, x, input vaue: 2 n activated functions use S type function; the function not ony is continuous but aso can derive. L, these nerve ces a BP NEURAL NETWORK LEARNING Neura networ is mainy up to two aspects: mode parameters, features, from which parameters incude stopping, hidden ayer, earning rate and other criterions, and the earning process is as Figure 3 show: 933
Jian Wang J. Chem. Pharm. Res., 204, 6(7:93-936 Figure 3: Learning neura networ mode Neura networ earning process starts impementing form initiaized networ, and then inputs the input ayer into a training corresponding mode, after networ transitive signa recognition, it defines output vaue size and automaticay sets a matching minimum vaue, if error is out of the vaue, and then system wi automaticay circuate the function ti error reduces to range. Origina data standardization process Define that between 0 and is BP neura networ node vaue, if input information hasn t arrived at hidden ayer, then the node is 0, therefore to avoid the faut status, we adopt standardization handing with these origina data, adopt: Hidden point initia number vaues can be defined by formua (2, that is: m n + a (2 2 0.43nm + 0.2n + 2.54m + 0.77n + 0.35 + 0.5 (3 n, m Among them, in above two formuas, a is a constant, and is a number between and 0, are of output and input nodes. We wor out an initia vaue by formua (, and then sove it graduay[7]. the number Define error Assume when outputs networ, error vaue is: We assume that E E K E K ( y 2 o J 2 is the sum of the mode whoe process generated output errors, and in above formua, (4 actua output vaue is o j y, idea output vaue is. APPLY NEURAL NETWORK INTO SPORTS PERFORMANCE PREDICTION THEORETICAL RESEARCH MODEL The paper seects 24th to 30th Oympic Games 000 50000 500 800 400 200 00m men s competitions each event champions sports performances as training sampes, and testing sampes adopt 26th to 30th sports performance, checing sampes adopt 25th to 29th sports performance, we et matrix coumn to be every session different event performance vaue, and ine to be an event different number of sessions, so that fufi the matrix. 934
Jian Wang J. Chem. Pharm. Res., 204, 6(7:93-936 Parameters defining and data handing The paper defines output ayer activated function as purein( x x, from which networ earning precise is set as 0.00005, iteration times are 0000times, impicit function corresponding activated function is defined as hyperboic tangent (tan sigx S type transmission function: tan sig( x σ σ 2 2 σ + σ (7 In addition, it shoud ensure that input data is between 0-, by converting p, that: 9.98 9.96 9.97 9.84 9.87 9.86 969 9.80 9.75 20.0 9.30 20.09 9.79 9.30 44.27 43.87 43.50 43.49 43.84 44.00 43.75 p 03.00 03.45 403.66 02.58 05.08 04.45 04.46 22.53 25.96 220.2 25.78 22.07 24.8 23.94 783.59 79.70 792.52 787.96 822.49 749.39 777.82 667.57 64.46 647.72 627.34 638.20 625.0 62.7 p After that, divide by every coumn found corresponding maximum vaue and then get, that:.0000 0.9857.0000 0.9803 0.9655 0.9527.0000 0.980 0.983 0.990 0.9845 0.98 0.9626 0.9844 0.9970 0.9960 0.9826 0.9865.0000 0.9636 0.988 p 0.9850 0.9607 0.9824 0.9762 0.9803 0.9580 0.9759 0.9880.0000 0.9903.0000 0.9634.0000 0.9824 0.9860 0.985 0.9939 0.9940 0.9730 0.9658 0.9745 0.970 0.9607 0.9883 0.979 0.9457 0.9457 0.972 p ine three to ine seven as testing sampe p30 In the foowing tae, ine two to ine six as checing p20 sampe, top five ines as training sampe p 0, on the condition that hidden ayer node number meets that hidden ayer and input as we as output ayer number shoud be ess than N, sampes output is 3 number of nodes is 3, input is 5, from which N is the number of sampes. BP neura networ training process The paper using created BP neura networ to predict the 3st Toyo Oympic Games 000m 5000m 500m 800m 400m 200m 00m, prediction way is roing type aternate training unti prediction precise conforms to requirement, the performance is predicted performance. Neura networ about Matab appication program p Input matrix, and input: p p ; po max( p; p00 ones(7,* p0 p./ p00; p0 p(: 5,:; p p(6,:; p20 p(2 : 6,:; p2 p(7,:; p30 p(3: 7,:; net newff (min max( p0,[3],{ tan sig, purein,}; net. trainparam. epochs 0000; net. train Pr arm. goa 0.00005; net. trainparam. show 500; net train( net, p0, p; y sim( net, p0; y0 y.* p0; Trained the 29th performance E P y ; MASE mse( E. Training error rate y2 sim( net, p20 ; y20 y2.* p0; trained the 30th performance E2 P2 y2 MASE2 mse( E2 ; checing sampes error rate y3 sim( net, p30 ; y30 y3.* p0; mae prediction on the 3st Oympic Games performance. 935
Jian Wang J. Chem. Pharm. Res., 204, 6(7:93-936 Training resut anaysis MSE 4.998e 5 MSE2.74e 4 e + [ ] e + [ ] e + [ ] y 0.0 3* 0.0097 0.096 0.044 0.036 0.239 0.8044.6365 y 20.0 3* 0.0098 0.099 0.0440 0.049 0.239 0.7793.6289 y 30.0 3* 0.0097 0.095 0.0434 0.030 0.278 0.7998.6248 CONCLUSION The paper uses sports competition performance to mae BP neura networ prediction, it gets the mode has feasibiities; it maes indeibe contributions to the event future prediction deveopment. Athete performance prediction is affected by ots of factors, use BP neura networ method to evauate individua prediction that shows it has obvious superiorities. The paper not ony introduces performance prediction s BP neura networ agorithm, but aso appied specific exampes to verify, the resut shows the mode structura rationaity. According to sports features, it composes array matrix, so that gets BP neura networ agorithms good prediction efficiency. Due to appy previous athetics competitions performance quantity s imitations, adopt aternate training way, et its resut more reiabe, correct. REFERENCES [] ZHU Hong-bing, LIU Jian-tong, WANG Gang, etc,. Journa of Capita Coege of Physica Education, 2003, 5(, 8-2. [2] WANG Wei. Journa of Nanjing Institute of Physica Education, 2004, 8(6, 85-87. [3] ZHAO Yun hong, ZHOU Yao. China Sport Science and Technoogy, 2002, 38(2, 39-40. [4] CHEN Liang, SAI Qing-bin. Journa of Capita Coege of Physica Education, 2006, 8(5, 85-88. [5] SONG Ai-Ling, CHEN Kai. Journa of Capita Coege of Physica Education, 2003, 5(4, 68-69, 6. [6] CHEN Liang, TIAN De-bao. Journa of Capita Coege of Physica Education, 2006, 8(6, 27-28. [7] Tong Liping, Yuan Jianguo. Journa of Shanghai Physica Education Institute, 200, 25(2, 44-47. [8] LIU Jia-jin. Journa of Guangzhou Physica Education Institute, 2006, 26(2, 54-56. [9] LIU Jia-jin et a. Zhejiang Sport Science, 999, 2(, 60-64. [0] Zhang B.; Zhang S.; Lu G.. Journa of Chemica and Pharmaceutica Research, 203, 5(9, 256-262. [] Zhang B.; Internationa Journa of Appied Mathematics and Statistics, 203, 44(4, 422-430. [2] Zhang B.; Yue H.. Internationa Journa of Appied Mathematics and Statistics, 203, 40(0, 469-476. [3] Zhang B.; Feng Y.. Internationa Journa of Appied Mathematics and Statistics, 203, 40(0, 36-43. [4] Bing Zhang. Journa of Chemica and Pharmaceutica Research, 204, 5(2, 649-659. 936