基于 Elman人工神经网络的电力系统短期负荷预测研究

%编制人：adwjt %编制时间：2020.9.19 %% 初始化 clear all; close all clc; format long; load_count=356; dian_count=96; load0=xlsread('zibo08.xls'); load=load0([1:load_count],[2:(dian_count+1)]);%因为quan.xls第一列增加了序号列 %对全体数据进行归一化处理 load_max=max(max(load)); load_min=min(min(load)); load=(load-load_min)./(load_max-load_min); wen_max=load0([1:load_count],dian_count+2); wen_min=load0([1:load_count],dian_count+3); wen_aver=load0([1:load_count],dian_count+4); wen_max=mapminmax(wen_max',0.1,0.9); wen_min=mapminmax(wen_min',0.1,0.9); wen_aver=mapminmax(wen_aver',0.1,0.9); wen_max=wen_max'; wen_min=wen_min'; wen_aver=wen_aver'; load1=zeros(load_count,dian_count+6); load1(1:load_count,1:dian_count)=load; load1(1:load_count,dian_count+1)=wen_max; load1(1:load_count,dian_count+2)=wen_min; load1(1:load_count,dian_count+3)=wen_aver; load=load1;%load已经进行归一化，而且数据很全，包括了温度和天气 %训练样本处理 %选择训练起始天d_start，终止天d_end d_start=108;d_end=117; p_train=zeros(96*(d_end-d_start+1),15);%【改】输入元素数量 t_train=zeros(96*(d_end-d_start+1),1); k=0;%k是训练样本数 for i=d_start:1:d_end for j=1:1:96 k=k+1; if j==1 p_train(k,:)=[load(i-2,96),load(i-1,j),load(i-1,j+1),load(i-3,96),load(i-2,j),load(i-2,j+1),load(i-4,96),load(i-3,j),load(i-3,j+1),load(i,97),load(i,98),load(i,99),load(i,100),load(i,101),load(i,102)]; elseif j==96 p_train(k,:)=[load(i-1,j-1),load(i-1,j),load(i,1),load(i-2,j-1),load(i-2,j),load(i-1,1),load(i-3,j-1),load(i-3,j),load(i-2,1),load(i,97),load(i,98),load(i,99),load(i,100),load(i,101),load(i,102)]; else p_train(k,:)=[load(i-1,j-1),load(i-1,j),load(i-1,j+1),load(i-2,j-1),load(i-2,j),load(i-2,j+1),load(i-3,j-1),load(i-3,j),load(i-3,j+1),load(i,97),load(i,98),load(i,99),load(i,100),load(i,101),load(i,102)]; end t_train(k)=load(i,j); end end k p_train=p_train'; t_train=t_train'; %预测处理,同时处理预测天的全天真实负荷值real_d %要预测第d_test天，第h_test个负荷点 p_test=zeros(96,15);%【改】输入元素数量 t_test=zeros(96,1); k1=0;%k是预测负荷点的数量，此处取一天中96点 d_test=121; for j1=1:1:96 k1=k1+1; if j1==1 p_test(k1,:)=[load(d_test-2,96),load(d_test-1,j1),load(d_test-1,j1+1),load(d_test-3,96),load(d_test-2,j1),load(d_test-2,j1+1),load(d_test-4,96),load(d_test-3,j1),load(d_test-3,j1+1),load(d_test,97),load(d_test,98),load(d_test,99),load(d_test,100),load(d_test,101),load(d_test,102)]; elseif j1==96 p_test(k1,:)=[load(d_test-1,j1-1),load(d_test-1,j1),load(d_test,1),load(d_test-2,j1-1),load(d_test-2,j1),load(d_test-1,1),load(d_test-3,j1-1),load(d_test-3,j1),load(d_test-2,1),load(d_test,97),load(d_test,98),load(d_test,99),load(d_test,100),load(d_test,101),load(d_test,102)]; elseif j1==96 else p_test(k1,:)=[load(d_test-1,j1-1),load(d_test-1,j1),load(d_test-1,j1+1),load(d_test-2,j1-1),load(d_test-2,j1),load(d_test-2,j1+1),load(d_test-3,j1-1),load(d_test-3,j1),load(d_test-3,j1+1),load(d_test,97),load(d_test,98),load(d_test,99),load(d_test,100),load(d_test,101),load(d_test,102)]; end t_test(k1)=load(d_test,j1); end p_test=p_test'; t_test=t_test'; k1 %net=newff(p_train,t_train,31,{'tansig','logsig'},'traingdm'); net=newelm(minmax(p_train),[34,1],{'tansig','purelin',});%不用purelin函数 %，因为结果点上下波动太大 %newff([每个输入向量的范围],[每层神经元的个数]，{'每一层传递函数(默认为tansig函数)'}，BTF,BLF,PF); %BTF:训练函数(默认为trainlm函数,此处使用的训练函数是traingdm函数); %BLF:权值/阈值学习函数(默认为learngdm函数); %PF:性能函数(默认为mse函数); net.trainParam.show=100; net.trainParam.lr=0.001; net.trainParam.epochs=3000; net.trainParam.goal=0.0001; %调用TRAINGDM算法训练BP网络 net=train(net,p_train,t_train); %对BP网络进行仿真 r=sim(net,p_test);%进行预测 m=r.*(load_max-load_min)+load_min; n=load0(120,2:97); cha=abs(m-n);%求绝对值 cha_xiangdui=cha./n; cha_jun=(sum(cha_xiangdui))/96 cha1=abs(cha_xiangdui); chamax=max(cha1) figure(1) plot(t_test,'r * -'); hold on; plot(r,'g o -'); hold off axis([0 100 0 0.8]); title('+为真实值，*为预测值'); legend('真实值','预测值'); box off AdAxesEdit=axes('Position',get(gca,'Position'),'XAxisLocation','top','YAxisLocation','right','Color','none','XColor','k','YColor','k'); set(AdAxesEdit,'XTick',[],'YTick',[]); box off