只需调用scipy.integrate.solve_ivp即可加速脚本

Question

我正在尝试尽可能快地执行以下脚本：

1. 从文本文件中读取浮点列表
2. 求解微分方程
3. 将一些结果输出到stdout。

应该认为它是测试优化工具的“黑匣子”，该优化工具与脚本作为系统命令交互。因此，该工具调用python blackbox.py x.txt（其中x.txt是评估黑盒的点，此处有0到3.5之间的500个元素），并且输出是目标函数以及对违反约束的度量。 （例如-1.007 -0.067 -4.471）。

在这种情况下，我正在尝试解决Goddard rocket problem。

这是我所拥有的代码，以及我所知道的加快技巧的所有技巧：

import sys

from numpy import max, sqrt, exp, errstate # Avoid resolving NumPy methods (as np.max) within the code
from scipy.integrate import solve_ivp

TF = 0.2 # Final time

# Read point
with open(sys.argv[1], 'r') as f:
    x_u = [float(xi) for xi in f.read().split()]
i_last_x_u = len(x_u)-1 # Index of last element

# Function to solve
def f(t, y):
    h, v, m = y # Decompose by rows

    D_hv = 0.5*620*v**2*exp(500*(1-h)) # Drag
    u_t = x_u[int(t/TF*i_last_x_u)] # "Interpolation" on input vector

    return [
            v, # h'
            (u_t - D_hv)/m - 1/h**2, # v'
            -u_t/0.5, # m'
            ] 

# Solver call
with errstate(all='ignore'): # Errors are catched by looking at sol.success
    sol = solve_ivp(f, t_span=(0, TF), y0=[1,0,1], vectorized=True)

if sol.success:
    # Solution
    h_sol, v_sol, m_sol = sol.y # Decompose by rows

    # Objective function: maximum height
    # Minus sign since NOMAD tries to minimize obj
    obj_h_max = -max(h_sol)

    # Constraint: Final mass must be >0.6
    pb_m = 0.6 - m_sol[-1]

    # Constraint: Dynamic pressure
    pb_q = max( v_sol - sqrt(2*10/exp(500*(1-h_sol))) ) 

    # Output solution
    print(f'{obj_h_max} {pb_m} {pb_q}')
else:
    # Solver failed
    # Non-zero exit status is flagged as a failed evaluation
    print('0 0 0')
    sys.exit(1)

下面是Spyder探查器的输出。看来罪魁祸首是_find_and_load。

如果不使用C编写整个代码，我该怎么做才能加快速度？

谢谢！