我正在尝试使用Eigen和Odeint实现状态空间模型的数值模拟。我的麻烦是我需要引用控制数据 U (在集成之前预定义)以正确解决状态中的 Ax + Bu 部分空间模型。我试图通过使用计数器来跟踪当前时间步长来实现这一点,但无论出于何种原因,每次Odeint调用系统函数时它都会重置为零。
我如何解决这个问题?我的状态空间系统建模方法是否存在缺陷?
我的系统
struct Eigen_SS_NLTIV_Model
{
Eigen_SS_NLTIV_Model(matrixXd &ssA, matrixXd &ssB, matrixXd &ssC,
matrixXd &ssD, matrixXd &ssU, matrixXd &ssY)
:A(ssA), B(ssB), C(ssC), D(ssD), U(ssU), Y(ssY)
{
Y.resizeLike(U);
Y.setZero();
observerStep = 0;
testPtr = &observerStep;
}
/* Observer Function:*/
void operator()(matrixXd &x, double t)
{
Y.col(observerStep) = C*x + D*U.col(observerStep);
observerStep += 1;
}
/* System Function:
* ONLY the mathematical description of the system dynamics may be placed
* here. Any data placed in here is destroyed after each iteration of the
* stepper.
*/
void operator()(matrixXd &x, matrixXd &dxdt, double t)
{
dxdt = A*x + B*U.col(*testPtr);
//Cannot reference the variable "observerStep" directly as it gets reset
//every time this is called. *testPtr doesn't work either.
}
int observerStep;
int *testPtr;
matrixXd &A, &B, &C, &D, &U, &Y; //Input Vectors
};
我的ODE解算器设置
const double t_end = 3.0;
const double dt = 0.5;
int steps = (int)std::ceil(t_end / dt) + 1;
matrixXd A(2, 2), B(2, 2), C(2, 2), D(2, 2), x(2, 1);
matrixXd U = matrixXd::Constant(2, steps, 1.0);
matrixXd Y;
A << -0.5572, -0.7814, 0.7814, 0.0000;
B << 1.0, -1.0, 0.0, 2.0;
C << 1.9691, 6.4493, 1.9691, 6.4493;
D << 0.0, 0.0, 0.0, 0.0;
x << 0, 0;
Eigen_SS_NLTIV_Model matrixTest(A, B, C, D, U, Y);
odeint::integrate_const(odeint::runge_kutta4<matrixXd, double, matrixXd, double,
odeint::vector_space_algebra>(),
matrixTest, x, 0.0, t_end, dt, matrixTest);
//Ignore these two functions. They are there mostly for debugging.
writeCSV<matrixXd>(Y, "Y_OUT.csv");
prettyPrint<matrixXd>(Y, "Out Full");
答案 0 :(得分:0)
使用经典的Runge-Kutta,您知道您的ODE模型函数每步调用4次,时间t, t+h/2, t+h/2, t+h
。对于实现自适应步长的其他求解器,您无法事先知道调用ODE模型函数的t
。
您应该通过某种插值函数实现U
,在最简单的情况下,作为步长函数计算来自t
的某个索引,并返回该索引的U
值。像
i = (int)(t/U_step)
dxdt = A*x + B*U.col(i);