19、时间序列滤波建模与分析

时间序列滤波建模与分析

1. 稀疏最小二乘滤波输入估计

在Python代码中，有关于稀疏最小二乘滤波输入估计的实现。以下是相关代码：

# edapy_07_03: sparse, least sq estimation of filter input
. . .
eda_e = 10.0 * np.amax(G);  # note damping is a global variable
. . .
# set up right hand side, F'f = GT qobs + e HT h
FTf = GLSFilterMulRHS(edafilterg, qobs, eda_e, edaHsparse, h, M)
# define linear operator needed for conjugate gradient solver
LO = las.LinearOperator(shape=(M, M), matvec=GLSFilterMul,
                        rmatvec=GLSFilterMul);
# solve for model parameters via congugate gradient algorithm
mest = np.zeros((M, 1));
q = las.cg(LO, FTf, tol=1e-12, maxiter=(3 * (N + M) + 100));
# note that q is a tupple, the first element of which is the solution
mest = eda_cvec(q[0]);