🍁🥬🕒摘要🕒🥬🍁
传统的测井解释需要建立精确的数学模型,并常伴有严格的条件限制,因此很难得到真实反映储层特性的结果。采用遗传算法与BP神经网络相结合,利用遗传算法的全局寻优特点,优化神经网络的连接权值和阀值,提高网络的训练精度和预测精度,避免了BP算法易陷入局部极小的缺点,提高运算速度。将相似度的概念引入到测井中,定义相似度在测井中的计算公式,提出相似度与遗传神经网络相结合的方法。实例研究表明,预测准确性较高,可以有效控制预测精度,避免因储层差别大而造成的预测精度降低的现象。
✨🔎⚡运行结果⚡🔎✨
💂♨️👨🎓Matlab代码👨🎓♨️💂
clear
clc
close all
data_reference=importdata('ikpikpuk Ivishak fm_reference.txt')
%sesuaikan dengan nama formasi /Kingak/Shublik/Ivishak
ft_reference=data_reference(:,1)
DEPT_reference=0.3048.*ft_reference
GR_reference=data_reference(:,6)
SP_reference=data_reference(:,2)
CALI_reference=data_reference(:,7)
ILD_reference=data_reference(:,3)
ILM_reference=data_reference(:,4)
LL8_reference=data_reference(:,5)
DT_reference=data_reference(:,11)
RHOB_reference=data_reference(:,8)
NPHI_reference=data_reference(:,10)
DRHO_reference=data_reference(:,9)
vp_reference=304800./DT_reference
data=importdata('ikpikpuk Ivishak fm_test.txt')
%sesuaikan nama formasi /Kingak/Shublik/Ivishak
ft=data(:,1)
DEPT=0.3048.*ft
GR=data(:,6)
SP=data(:,2)
CALI=data(:,7)
ILD=data(:,3)
ILM=data(:,4)
LL8=data(:,5)
DT=data(:,11)
RHOB=data(:,8)
NPHI=data(:,10)
DRHO=data(:,9)
vp=304800./DT
y=vp_reference
nv=8; %jumlah unknown parameter
x=DEPT_reference';
np=150; % ukuran populasi awal
ns=75; %jumlah individu survive
cm=0.5; %peluang mutasi
maxiter=100; %maximum generasi/iterasi
average=0; % average distribusi normal
std=4; %standar deviasi distribusi normal
%membuat populasi dari kumpulan parameter
a1_pop=normrnd(average,std,[1,np]);
b1_pop=normrnd(average,std,[1,np]);
a2_pop=normrnd(average,std,[1,np]);
b2_pop=normrnd(average,std,[1,np]);
a3_pop=normrnd(average,std,[1,np]);
b3_pop=normrnd(average,std,[1,np]);
a4_pop=normrnd(average,std,[1,np]);
b4_pop=normrnd(average,std,[1,np]);
a1_popawal=a1_pop'
b1_popawal=b1_pop'
a2_popawal=a2_pop'
b2_popawal=b2_pop'
a3_popawal=a3_pop'
b3_popawal=b3_pop'
a4_popawal=a4_pop'
b4_popawal=b4_pop'
fx=zeros(2,np);
fx(2,:)=1:1:np;
prob=zeros(1,ns);
errplot=zeros(1,maxiter);
for iter=1:maxiter
%hitung fitness tiap populasi
for i=1:np
a1=a1_pop(i);
b1=b1_pop(i);
a2=a2_pop(i);
b2=b2_pop(i);
a3=a3_pop(i);
b3=b3_pop(i);
a4=a4_pop(i);
b4=b4_pop(i);
y1=a1.*DEPT_reference.^b1+a2.*GR_reference.^b2+a3.*RHOB_reference.^b3+a4.*ILD_reference.^b4;
fx(1,i)=immse(y,y1); %fitness MSE
end
[Y,I]=sort(fx(1,:),2,'ascend'); %memilih best candidate yg bakal survive
best=fx(:,I);
prob=zeros(1,ns);
%clearing variable
a1_pop0=a1_pop;
b1_pop0=b1_pop;
a2_pop0=a2_pop;
b2_pop0=b2_pop;
a3_pop0=a3_pop;
b3_pop0=b3_pop;
a4_pop0=a4_pop;
b4_pop0=b4_pop;
*b2_pop(pick1);
beta=rand(1);
b2_pop(i+1)=(1-beta)*b2_pop(pick1)+beta*b2_pop(pick2);
beta=rand(1);
a3_pop(i)=(1-beta)*a3_pop(pick2)+beta*a3_pop(pick1);
beta=rand(1);
a3_pop(i+1)=(1-beta)*a3_pop(pick1)+beta*a3_pop(pick2);
beta=rand(1);
b3_pop(i)=(1-beta)*b3_pop(pick2)+beta*b3_pop(pick1);
beta=rand(1);
b3_pop(i+1)=(1-beta)*b3_pop(pick1)+beta*b3_pop(pick2);
beta=rand(1);
a4_pop(i)=(1-beta)*a4_pop(pick2)+beta*a4_pop(pick1);
beta=rand(1);
a4_pop(i+1)=(1-beta)*a4_pop(pick1)+beta*a4_pop(pick2);
beta=rand(1);
b4_pop(i)=(1-beta)*b4_pop(pick2)+beta*b4_pop(pick1);
beta=rand(1);
b4_pop(i+1)=(1-beta)*b4_pop(pick1)+beta*b4_pop(pick2);
📜📢🌈参考文献🌈📢📜
[1]杜宇,潘遥.基于GA-BP神经网络岩石可钻性预测模型[J].科学技术创新,2020(25):57-59.