优化DataFrame中矩阵乘法以提升飞行器迎角与侧滑角计算效率

你的代码核心性能瓶颈在于**iterrows()逐行迭代+手动三重循环**，这两种操作在处理大数据量时效率极低。以下是具体优化方案，全部基于NumPy的向量化运算（底层为C实现，性能比纯Python循环高几个数量级）：

方案1：直接展开矩阵运算（最优性能） #

直接把矩阵乘法的计算式展开，避免中间数组的创建，用元素级向量化运算完成计算：

# 提前计算所有三角函数值，避免重复计算
cos_pitch = np.cos(df['Pitch'])
sin_pitch = np.sin(df['Pitch'])
cos_yaw = np.cos(df['Yaw'])
sin_yaw = np.sin(df['Yaw'])
cos_roll = np.cos(df['Roll'])
sin_roll = np.sin(df['Roll'])

# 提取全局速度向量为NumPy数组
V_global = df[['Vx', 'Vy', 'Vz']].values

# 直接展开旋转矩阵与速度向量的点乘计算
df['Vxx'] = cos_pitch * cos_yaw * V_global[:,0] + cos_pitch * sin_yaw * V_global[:,1] - sin_pitch * V_global[:,2]
df['Vyy'] = (sin_pitch * sin_roll * cos_yaw - cos_roll * sin_yaw) * V_global[:,0] + \
            (sin_pitch * sin_roll * sin_yaw + cos_roll * cos_yaw) * V_global[:,1] + \
            sin_roll * cos_pitch * V_global[:,2]
df['Vzz'] = (cos_yaw * cos_roll * sin_pitch + sin_roll * sin_yaw) * V_global[:,0] + \
            (sin_yaw * cos_roll * sin_pitch - sin_roll * cos_yaw) * V_global[:,1] + \
            cos_roll * cos_pitch * V_global[:,2]

方案2：批量矩阵乘法（更直观） #

如果想保留矩阵运算的逻辑清晰性，可以构造三维旋转矩阵数组，批量完成所有行的矩阵乘法：

# 提前计算三角函数值
cos_pitch = np.cos(df['Pitch'])
sin_pitch = np.sin(df['Pitch'])
cos_yaw = np.cos(df['Yaw'])
sin_yaw = np.sin(df['Yaw'])
cos_roll = np.cos(df['Roll'])
sin_roll = np.sin(df['Roll'])

# 构造三维旋转矩阵数组：shape为(数据行数, 3, 3)
rot_matrices = np.stack([
    np.stack([cos_pitch*cos_yaw, cos_pitch*sin_yaw, -sin_pitch], axis=1),
    np.stack([
        sin_pitch*sin_roll*cos_yaw - cos_roll*sin_yaw,
        sin_pitch*sin_roll*sin_yaw + cos_roll*cos_yaw,
        sin_roll*cos_pitch
    ], axis=1),
    np.stack([
        cos_yaw*cos_roll*sin_pitch + sin_roll*sin_yaw,
        sin_yaw*cos_roll*sin_pitch - sin_roll*cos_yaw,
        cos_roll*cos_pitch
    ], axis=1)
], axis=1)

# 全局速度向量reshape为(数据行数, 3, 1)
V_global = df[['Vx', 'Vy', 'Vz']].values.reshape(-1, 3, 1)

# 批量矩阵乘法计算机体坐标系速度
V_body = np.matmul(rot_matrices, V_global)

# 赋值回DataFrame
df['Vxx'] = V_body[:, 0, 0]
df['Vyy'] = V_body[:, 1, 0]
df['Vzz'] = V_body[:, 2, 0]

额外优化建议 #

1. 避免不必要的DataFrame列：原代码中把a11到a33都存在DataFrame中，如果后续不需要这些中间值，直接用变量计算即可，减少内存占用和DataFrame读写开销。
2. 提前缓存三角函数值：原代码多次重复计算np.cos(df['Pitch'])等，提前缓存这些值能节省重复计算的时间。
3. 用df.assign()链式操作：如果喜欢简洁代码，可以用df.assign()一次性创建所有新列，避免多次修改DataFrame。

内容的提问来源于stack exchange，提问作者Decaff_42