1 簡介

BP神經(jīng)網(wǎng)絡(luò)算法使用非常廣泛,傳統(tǒng)的BP神經(jīng)網(wǎng)絡(luò)算法雖然具有不錯(cuò)的擬合非線性函數(shù)的能力,但是容易陷入局部的極小值,并且傳統(tǒng)的算法收斂的速度慢.本篇文章詳細(xì)地論述了如何使用帝企鵝算法優(yōu)化傳統(tǒng)的BP神經(jīng)網(wǎng)絡(luò)算法中初始的權(quán)值和閥值,通過相應(yīng)的驗(yàn)證和比較提出了該模型的有效性.

2 部分代碼

%% 2021.9.2 ?AFO 算法優(yōu)化神經(jīng)網(wǎng)絡(luò)% 2021.9.2 AFO algorithm optimizes neural network weights and thresholds%% 這是使用原始算法的直接求解結(jié)果，添加專用于本問題的更新方式可以進(jìn)一步提高精度% This is the direct result of using the original algorithm, % adding some specific update methods to this problem can further improve the accuracyclc;clear;close all;warning off%% 固定隨機(jī)數(shù)種子noRNG=1;rng('default')rng(noRNG)%% 載入數(shù)據(jù)%% 數(shù)據(jù)預(yù)處理load inputload output%%data.x=[input'];data.y=[output];num_Train=120; %訓(xùn)練集數(shù)量data.len_train=num_Train;num_Test=48; %測試集數(shù)量index=randperm(num_Train+num_Test); %隨機(jī)亂序data.train_x=data.x(:,index(1:num_Train))';data.train_y=data.y(:,index(1:num_Train))';data.test_x=data.x(:,index(num_Train+1:end))';data.test_y=data.y(:,index(num_Train+1:end))';% 歸一化[data.train_x0,option.ps_x]=mapminmax(data.train_x');data.train_x0=data.train_x0';[data.train_y0,option.ps_y]=mapminmax(data.train_y');data.train_y0=data.train_y0';data.test_x0 = mapminmax('apply',data.test_x',option.ps_x);data.test_x0=data.test_x0';data.test_y0 = mapminmax('apply',data.test_y',option.ps_y);data.test_y0=data.test_y0';%%data.weight=1;data=data;data.hiddenumber= [5];%隱含神經(jīng)元數(shù)%% 未優(yōu)化神經(jīng)網(wǎng)絡(luò)clear result[x(1,:),result(1)]=creat_x_1(option,data);%draw(result(1),'未優(yōu)化')data.m=result(1).m;data.n=result(1).n;data.m_lw=result(1).m_lw;data.n_lw=result(1).n_lw;data.m_iw=result(1).m_iw;data.n_iw=result(1).n_iw;data.m_b=result(1).m_b;data.n_b=result(1).n_b;data.len=result(1).len;[y(1),result(1)]=aimFcn_1(x(1,:),option,data,result(1).net);%% 參數(shù)設(shè)置lb=-5;ub=5;option.lb=lb;option.ub=ub;dim=length(x(1,:));%%option.dim=dim; %八個(gè)決策變量lb=0;ub=1;option.lb=lb;option.ub=ub;if length(option.lb)==1 ? ?option.lb=ones(1,option.dim)*option.lb; ? ?option.ub=ones(1,option.dim)*option.ub;endoption.fobj=@aimFcn_1;%option.fobj0=option.fobj;option.showIter=0;%% 算法參數(shù)設(shè)置 Parameters% 基本參數(shù)tic[bestY(1,:),bestX(1,:),best_result1,recording(1)]=AFO1(x,y,result,option,data);tt(1,1)=toc;%%figurehold onfor i=1:length(recording) ? ?plot((recording(i).bestFit),'LineWidth',2)endlegend(str_legend)title('fitness curve')%% 未優(yōu)化神經(jīng)網(wǎng)絡(luò)rng(7)[x(1,:),result(1)]=creat_x_1(option,data);[y(1),result(1)]=aimFcn_1(x(1,:),option,data,result(1).net);%%figurehold onplot(result(1).ygabptest(i,:));plot(best_result1.ygabptest(i,:));plot(data.test_y(:,i));legend(['未優(yōu)化神經(jīng)網(wǎng)絡(luò)(mse=',num2str(result(1).mse(i)),')'],['AFO優(yōu)化神經(jīng)網(wǎng)絡(luò)(mse=',num2str(best_result1.mse(i)),')'],['真實(shí)數(shù)據(jù)'])

3 仿真結(jié)果

4 參考文獻(xiàn)

[1]肖雄. PSO優(yōu)化BP神經(jīng)網(wǎng)絡(luò)巖爆預(yù)測的Matlab實(shí)現(xiàn)[J]. 中國房地產(chǎn)業(yè), 2018, 000(025):244.

博主簡介：擅長智能優(yōu)化算法、神經(jīng)網(wǎng)絡(luò)預(yù)測、信號(hào)處理、元胞自動(dòng)機(jī)、圖像處理、路徑規(guī)劃、無人機(jī)等多種領(lǐng)域的Matlab仿真，相關(guān)matlab代碼問題可私信交流。

部分理論引用網(wǎng)絡(luò)文獻(xiàn)，若有侵權(quán)聯(lián)系博主刪除。