1.算法概述
伪谱法，又称为正交配置法，主要利用Lagrange 插值多项式近似离散最优控制问题中的状态变量和控制变量，将连续型最优控制问题转化成离散形式的非线性规划(NLP) 问题，然后利用相应的NLP 算法求解。根据配置点的不同，伪谱法主要分为Legendre伪谱法、Gauss伪谱法 和Radau伪谱法 3 种。
在本课题中，飞行器的运动方程取为：
2.部分程序
`............................................................................
%约束（3个）
YY1(k) = 0.5pa4(k)^2 - qmax;%动压
YY2(k) = sqrt(L^2 + D^2)/G - nmax;%过载
YY3(k) = 3.0078*sqrt(p)*a4(k)^3.08 - Qmax;%驻点热流
end
%根据映射结果计算L值
for i = 1:K
for kk=1:Ns
if kk~=i
LL(i)=(t-Tao(kk))/(Tao(i)-Tao(kk));
end
end
end
for i = 1:K
Ls(i)=int(LL(i),t,-1,1);
end
Ls = double(Ls);
%目标方程（1个）
for i = 1:K
Tmp7(i) = C/sqrt(R)p^0.5a4(k)^3.08/(Ls(i)^2);
end
J = -(tf-t0)/(K*(K+1))*sum(Tmp7);
%%
%六状态
r_line = zeros(1,K);
Theta_line = zeros(1,K);
Fai_line = zeros(1,K);
V_line = zeros(1,K);
Gamma_line = zeros(1,K);
Si_line = zeros(1,K);
k_ = 0;
CNT = 0;
for k = 1:K
CNT = CNT + 1;
k
if k == 1
Dkl(k,:) = func_D(t0,tf,ts,K,Ns,k);
%离散状态变量定义为ak
a1(k) = r0;
a2(k) = Theta0;
a3(k) = Fai0;
a4(k) = V0;
a5(k) = Gamma0;
a6(k) = Si0;
%离散控制变量定义为bk
b1(k) = delta0;
b2(k) = alpha0;
x = [a1(k) a2(k) a3(k) a4(k) a5(k) a6(k) b1(k) b2(k)];
%将每个网络的最优解方程到过程变量数据中
%六状态
r_line(k) = x(1);
Theta_line(k) = x(2);
Fai_line(k) = x(3);
V_line(k) = x(4);
Gamma_line(k) = x(5);
Si_line(k) = x(6);
else
%注意，采用radau离散化之后的非线性方程组，没法直接使用fmincon进行求解，这里，我们自己编写了一个优化函数进行计算最优值
%对控制状态进行循环(fmincon的原理，也是基于如下过程进行的)
nn = 0;
mm = 0;
alphass = [ 10:Steps:20]/180*pi;
deltass = [-90:3Steps:90]/180pi;
for alphas = alphass
mm = 0;
nn = nn+1;
for deltas = deltass
mm=mm+1;
for NN = 1:N
k_= k-1;
Dkl(k_,:) = func_D(t0,tf,ts,K,Ns,k_);
g = u/a1(k_)^2;
p = Pexp(-0.00015(a1(k_)-Rs));
%部分由状态变量决定的参数
D = 0.5pa4(k_)^2Sref(bb0 + bb1alphas + bb2alphas^2);
L = 0.5pa4(k_)^2Sref(aa0 + aa1alphas + aa2alphas^2);
G = m;
%状态方程（6个）
%公式1
a1(k_+1) = a1(k_)+dtf0*((tf-t0)/2 * a4(k_)sin(a5(k_)) + a4(k_)gsin(1e3a5(k_)));
%公式2
a2(k_+1) = a2(k_)+dtf1*((tf-t0)/2 * a4(k_)*cos(a5(k_))*cos(a6(k_)) / (a1(k_)*cos(a3(k_))));
%公式3
a3(k_+1) = a3(k_)+dtf2*((tf-t0)/2 * a4(k_)*cos(a5(k_))*sin(a6(k_)) / (a1(k_)));
%公式4
a4(k_+1) = a4(k_)+dtf3*((tf-t0)/2 * (D/m + g*sin(a5(k_))));
%公式5
a5(k_+1) = a5(k_)+dtf4*((tf-t0)/2 * (L/m * cos(deltas) -g*cos(a5(k_)) + a4(k_)^2/a1(k_)*cos(a5(k_))))/a4(k_);
%公式6
a6(k_+1) = a6(k_)+dtf5*((tf-t0)/2 * (L/m * sin(deltas)/cos(a5(k_)) - a4(k_)^2/a1(k_)*cos(a5(k_))*cos(a6(k_))*tan(a3(k_))))/a4(k_);
end
D = 0.5pa4(k_+1)^2Sref(bb0 + bb1alphas + bb2alphas^2);
L = 0.5pa4(k_+1)^2Sref(aa0 + aa1alphas + aa2alphas^2);
if (0.5pa4(k_+1)^2 <= qmax) & (sqrt(L^2 + D^2)/G <= nmax) & (3.0078*sqrt(p)*a4(k_+1)^3.08 <= Qmax) &...
(0.5pa4(k_+1)^2 > 0) & (sqrt(L^2 + D^2)/G > 0) & (3.0078*sqrt(p)*a4(k_+1)^3.08 > 0)&...
a4(k_+1) >= 1.508 & a4(k_+1) <= V0 & a1(k_+1) <= Rs+80 & a1(k_+1) >= Rs+20;
..............................................
02-007m`
3.算法部分仿真结果图
