matlab與機(jī)械優(yōu)化設(shè)計(jì)三級(jí)項(xiàng)目

matlab與機(jī)械優(yōu)化設(shè)計(jì)三級(jí)項(xiàng)目--BP神經(jīng)網(wǎng)絡(luò)的建立與matlab實(shí)現(xiàn)組員：齊意媛指導(dǎo)教師：任素波PAGEPAGE6目錄一、BP二、BP三、應(yīng)用matlab進(jìn)行計(jì)算四、結(jié)論一、BP神經(jīng)網(wǎng)絡(luò)的概述BP（Back-ProPagationNetwork）又稱(chēng)反向傳播神經(jīng)網(wǎng)絡(luò)，通過(guò)樣本數(shù)數(shù)據(jù)壓縮和時(shí)間序列預(yù)測(cè)等。BP網(wǎng)絡(luò)由輸入層隱層和輸出層組成隱層可以有一層或多層圖2是的三層BP網(wǎng)絡(luò)模型，網(wǎng)絡(luò)選用S型傳遞函數(shù)， 通過(guò)反傳誤差函數(shù) （ （Ti為期望輸出、Oi為網(wǎng)絡(luò)的計(jì)算輸出），不斷調(diào)網(wǎng)絡(luò)權(quán)值和閾值使誤差函數(shù)E達(dá)到極小。BP網(wǎng)絡(luò)具有高度非線性和較強(qiáng)的泛化能力，但也存在收斂速度慢、迭代步數(shù)多、易于陷入局部極小和全局搜索能力差等缺點(diǎn)。可以先用遺傳算法對(duì)“BPBP間內(nèi)搜索最優(yōu)解。二、BP網(wǎng)絡(luò)的構(gòu)建下面選取一個(gè)例子作為研究對(duì)象，選取運(yùn)動(dòng)員測(cè)試各項(xiàng)素質(zhì)指標(biāo)下面選取一個(gè)例子作為研究對(duì)象，選取運(yùn)動(dòng)員測(cè)試各項(xiàng)素質(zhì)指標(biāo)序號(hào)跳高成績(jī)30行進(jìn)跑立定三級(jí)跳遠(yuǎn)助跑摸高助跑4—6步跳高負(fù)重深蹲杠鈴杠鈴半蹲系數(shù)100m抓舉12.243.29.63.452.151402.8115022.333.210.33.752.21203.410.97032.24393.52.21403.511.45042.323.210.33.652.21502.810.88052.23.210.13.52801.511.35062.273.4103.42.151303.211.56072.23.29.63.552.141303.511.86582.26393.52.11001.811.34092.23.29.63.552.11303.511.865102.243.29.23.52.11402.51150112.243.29.53.42.151152.811.950122.23.993.12802.21350132.23.19.53.62.1902.711.170142.353.29.73.452.151304.610.8570153.09.33.32.051002.811.250輸入輸出層的設(shè)計(jì)8，1。隱層設(shè)計(jì)有關(guān)研究表明有一個(gè)隱層的神經(jīng)網(wǎng)絡(luò)只要隱節(jié)點(diǎn)足夠多就可以以任意精網(wǎng)絡(luò)建立預(yù)測(cè)模型。在網(wǎng)絡(luò)設(shè)計(jì)過(guò)程中隱層神經(jīng)元數(shù)的確定十分重要。隱層神經(jīng)元個(gè)數(shù)過(guò)多會(huì)加大網(wǎng)絡(luò)計(jì)算量并容易產(chǎn)生過(guò)度擬合問(wèn)題神經(jīng)元個(gè)數(shù)過(guò)少則會(huì)影響網(wǎng)絡(luò)性能達(dá)不到預(yù)期效果。網(wǎng)絡(luò)中隱層神經(jīng)元的數(shù)目與實(shí)際問(wèn)題的復(fù)雜于隱層中神經(jīng)元數(shù)目的確定并沒(méi)有明確的公式只有一些經(jīng)驗(yàn)公式,神經(jīng)元個(gè)數(shù)問(wèn)題上參照了以下的經(jīng)驗(yàn)公式:其中n為輸入層神經(jīng)元個(gè)數(shù)m為輸出層神經(jīng)元個(gè)數(shù)a為1,10]之間的常數(shù)。4-136.網(wǎng)絡(luò)結(jié)構(gòu)示意圖如下：BPSigmoid可微函數(shù)和線性函數(shù)作為網(wǎng)絡(luò)的激勵(lì)函數(shù)。Stansig一到[-1,1]范圍內(nèi),因此預(yù)測(cè)模型選取Stansig作為輸出層神經(jīng)元的激勵(lì)函數(shù)。tansiglogsig函數(shù),網(wǎng)絡(luò)訓(xùn)練函數(shù)traingdx,mse,6epochs5000次,goal0.00000001lr0.01。設(shè)定完參數(shù)后開(kāi)始訓(xùn)練網(wǎng)絡(luò)。三、應(yīng)用matlab進(jìn)行計(jì)算代碼如下代碼如下P=[3.23.233.23.23.43.233.23.23.23.93.13.2;9.610.3910.310.1109.699.69.29.599.59.7;3.453.753.53.653.53.43.553.53.553.53.43.13.63.45;2.152.22.22.222.152.142.12.12.12.1522.12.15;140120140150801301301001301401158090130;2.83.43.52.81.53.23.51.83.52.52.82.22.74.6;1110.911.410.811.311.511.811.311.81111.91311.110.85;5070508050606540655050507070];T=[2.242.332.242.322.22.272.22.262.22.242.242.22.22.35];[p1,minp,maxp,t1,mint,maxt]=premnmx(P,T);%創(chuàng)建網(wǎng)絡(luò)net=newff(minmax(p1),[8,6,1],{'tansig','tansig','purelin'},'trainlm');%設(shè)置訓(xùn)練次數(shù)net.trainParam.epochs=5000;%設(shè)置收斂誤差net.trainParam.goal=0.0000001;%訓(xùn)練網(wǎng)絡(luò)[net,tr]=train(net,p1,t1);計(jì)算顯示如下%輸入數(shù)據(jù)a=[3.0;9.3;3.3;2.05;100;2.8;11.2;50];%a=premnmx(a);%b=sim(net,a);%將得到的數(shù)據(jù)反歸一化得到預(yù)測(cè)數(shù)據(jù)c=postmnmx(b,mint,maxt);再根據(jù)以上代碼輸入數(shù)據(jù)進(jìn)行模擬運(yùn)算的結(jié)果為c=2