1 简介
针对樽海鞘群算法(Salp Swarm Algorithm,SSA)在寻优过程中存在的收敛速度较慢,容易陷入局部最优的缺点,提出了一种改进的采用莱维飞行策略的条件化更新的樽海鞘群算法(Levy Flight-based Conditional Updating Salp Swarm Algorithm,LECUSSA),并将其运用于分类算法的特征子集选择过程.首先,利用莱维飞行策略的长短跳跃特点对领导者位置进行随机更新,以增强全局最优的搜索能力;其次,增加对追随者位置的更新条件,让追随者不再盲目地跟随,从而加快收敛速度.在23个优化基准函数上对LECUSSA算法与其他算法进行了性能比较实验;并把算法运用到支持向量机(Support Vector Machine,SVM)算法的分类特征子集选择中,采用8个UCI数据集对特征选择后的分类结果进行了性能比较实验.实验结果表明,LECUSSA具有良好的全局最优搜索能力和较快的收敛速度,利用LECUSSA算法进行特征选择后,能够找到最佳分类准确率的特征子集.
2 部分代码
function [sFeat,Sf,Nf,curve] = jSSA2(feat,label,N,max_Iter,HO)
lb = 0;
ub = 1;
thres = 0.5;
fun = @jFitnessFunction;
dim = size(feat,2);
X = zeros(N,dim);
for i = 1:N
for d = 1:dim
X(i,d) = lb + (ub - lb) * rand();
end
end
% Pre
fit = zeros(1,N);
fitF = inf;
curve = inf;
t = 1;
%---Iteration start----------------------------------------------------
while t <= max_Iter
for i = 1:N
fit(i) = fun(feat,label,(X(i,:) > thres),HO);
if fit(i) < fitF
Xf = X(i,:);
fitF = fit(i);
end
end
% Additional sort in the first iteration to improve the
% initial behavior by divide salps into leader and followers
if t == 1
[fit, idx] = sort(fit,'ascend');
X = X(idx,:);
end
c1 = 2 * exp(-(4 * t / max_Iter) ^ 2);
for i = 1:N
if i == 1
for d = 1:dim
c2 = rand();
c3 = rand();
if c3 >= 0.5
X(i,d) = Xf(d) + c1 * ((ub - lb) * c2 + lb);
else
X(i,d) = Xf(d) - c1 * ((ub - lb) * c2 + lb);
end
end
else
for d = 1:dim
X(i,d) = (X(i,d) + X(i-1,d)) / 2;
end
end
XB = X(i,:); XB(XB > ub) = ub; XB(XB < lb) = lb;
X(i,:) = XB;
end
curve(t) = fitF;
fprintf('\nIteration %d Best (SSA)= %f',t,curve(t))
t = t + 1;
end
Pos = 1:dim;
Sf = Pos((Xf > thres) == 1);
Nf = length(Sf);
sFeat = feat(:,Sf);
end
3 仿真结果
4 参考文献
[1]张严, 秦亮曦. 基于Levy飞行策略的改进樽海鞘群算法[J]. 计算机科学, 2020, 47(7):
博主简介：擅长智能优化算法、神经网络预测、信号处理、元胞自动机、图像处理、路径规划、无人机等多种领域的Matlab仿真，相关matlab代码问题可私信交流。
