使用 RBFInterpolator 进行二维样条插值及外推

心靈之曲

发布时间：2025-09-26 18:58:01

604人浏览过

来源于php中文网

原创

使用 rbfinterpolator 进行二维样条插值及外推

本文介绍了如何使用 scipy.interpolate.RBFInterpolator 对二维数据进行样条插值，并实现超出原始数据范围的外推。通过示例代码演示了如何创建插值器，并利用它计算任意点的插值结果，包括原始数据范围之外的点。同时，强调了 RBFInterpolator 相对于 Rbf 的优势，以及外推可能带来的不确定性。

在科学计算和数据分析中，二维样条插值是一种常用的技术，用于在离散数据点之间估计函数值。scipy.interpolate 模块提供了多种插值方法，其中 RBFInterpolator 是一种强大的工具，尤其适用于处理不规则分布的数据，并且能够进行外推。本文将详细介绍如何使用 RBFInterpolator 进行二维样条插值，并实现超出原始数据范围的外推。

RBFInterpolator 简介

RBFInterpolator 基于径向基函数 (Radial Basis Function, RBF) 实现插值。RBF 的基本思想是，每个数据点都会对周围区域产生影响，影响程度随着距离的增加而减小。RBFInterpolator 通过组合这些影响来估计任意点的函数值。

RBFInterpolator 相对于旧版本的 Rbf 具有显著的优势：

悦灵犀AI

一个集AI绘画、问答、创作于一体的一站式AI工具平台

下载

更快的计算速度： 尤其是在处理大量数据点时，RBFInterpolator 的性能更优。
更好的内存管理： RBFInterpolator 在内存使用方面更加高效。
直接支持外推： 无需额外设置，RBFInterpolator 可以直接用于计算原始数据范围之外的点。

使用 RBFInterpolator 进行插值和外推

以下示例代码演示了如何使用 RBFInterpolator 对给定的二维数据进行插值和外推。

import io
import numpy as np
import pandas as pd
from scipy.interpolate import RBFInterpolator
import matplotlib.pyplot as plt
from numpy import ma

# 模拟数据，替换成你的数据来源
data_str = """
dte,4185,4215,4245,4275,4305,4335,4365,4395,4425,4455,4485,4515,4545,4575,4605,4635,4665,4695,4725,4755,4785,4815,4845,4875,4905,4935,4965,4995,5025
0.015,0.14936,0.13411,0.11997,0.10711,0.09569,0.08569,0.07699,0.06949,0.06305,0.05754,0.05283,0.04882,0.0454,0.04248,0.03998,0.03784,0.03599,0.03438,0.03297,0.03174,0.03065,0.02969,0.02883,0.02806,0.02737,0.02675,0.02618,0.02567,0.0252
0.046,0.15398,0.13742,0.12183,0.10799,0.09574,0.08499,0.07564,0.06758,0.06069,0.05487,0.04998,0.04588,0.04246,0.03959,0.03718,0.03516,0.03347,0.03205,0.03084,0.02981,0.02893,0.02817,0.02751,0.02694,0.02643,0.02598,0.02558,0.02523,0.02491
0.076,0.15647,0.13904,0.12276,0.10828,0.09557,0.08452,0.07495,0.0667,0.05972,0.05382,0.04885,0.04467,0.04118,0.03824,0.03578,0.0337,0.03196,0.03049,0.02924,0.02818,0.02728,0.02652,0.02587,0.02532,0.02485,0.02445,0.0241,0.0238,0.02354
0.162,0.16199,0.14311,0.12574,0.11024,0.09687,0.08527,0.07525,0.06673,0.05948,0.05343,0.04831,0.04403,0.04047,0.0375,0.03504,0.03294,0.03116,0.02964,0.02835,0.02724,0.0263,0.02549,0.02479,0.02418,0.02366,0.02321,0.02282,0.02248,0.02218
0.251,0.16667,0.14654,0.12797,0.11141,0.09726,0.08516,0.07479,0.06601,0.05862,0.05246,0.04723,0.04285,0.03922,0.03618,0.03363,0.03146,0.0296,0.02801,0.02665,0.02548,0.02447,0.02359,0.02283,0.02216,0.02158,0.02107,0.02062,0.02023,0.01988
0.339,0.17044,0.14925,0.13002,0.11275,0.09803,0.08559,0.07497,0.06602,0.05851,0.05226,0.04695,0.0425,0.03881,0.03573,0.03315,0.03095,0.02907,0.02746,0.02607,0.02487,0.02382,0.0229,0.02209,0.02138,0.02076,0.02021,0.01973,0.0193,0.01891
0.426,0.17361,0.15147,0.1317,0.11396,0.09889,0.08621,0.0754,0.06633,0.05874,0.05243,0.04706,0.04256,0.03883,0.03572,0.03312,0.0309,0.02901,0.02738,0.02598,0.02477,0.02371,0.02278,0.02196,0.02124124,0.02061,0.02005,0.01956,0.01913,0.01874
0.512,0.17637,0.15337,0.13311,0.11501,0.09961,0.08673,0.07577,0.06658,0.05891,0.05255,0.04714,0.0426,0.03885,0.03572,0.0331,0.03087,0.02896,0.02733,0.02592,0.0247,0.02363,0.02269,0.02186,0.02114,0.0205,0.01994,0.01945,0.01901,0.01862
0.598,0.17884,0.15504,0.13435,0.11593,0.10024,0.0872,0.07613,0.06685,0.05911,0.0527,0.04725,0.04268,0.03891,0.03577,0.03314,0.0309,0.02898,0.02734,0.02593,0.0247,0.02363,0.02269,0.02186,0.02113,0.02049,0.01993,0.01944,0.019,0.01861
0.684,0.18106,0.15655,0.13546,0.11676,0.10079,0.08762,0.07644,0.06709,0.0593,0.05285,0.04737,0.04278,0.03899,0.03584,0.0332,0.03095,0.02902,0.02737,0.02595,0.02472,0.02364,0.02269,0.02186,0.02113,0.02048,0.01992,0.01942,0.01898,0.01859
0.769,0.18308,0.15794,0.13646,0.1175,0.10128,0.08801,0.07674,0.06733,0.05949,0.05301,0.0475,0.04289,0.04044,0.0359,0.03325,0.031,0.02906,0.02741,0.02598,0.02474,0.02366,0.02271,0.02187,0.02114,0.02049,0.01992,0.01942,0.01898,0.01858
"""

vol = pd.read_csv(io.StringIO(data_str))
vol.set_index('dte', inplace=True)

valid_vol = ma.masked_invalid(vol).T

Ti = np.linspace(float((vol.index).min()), float((vol.index).max()), len(vol.index))
Ki = np.linspace(float((vol.columns).min()), float((vol.columns).max()), len(vol.columns))

Ti, Ki = np.meshgrid(Ti, Ki)

valid_Ti = Ti[~valid_vol.mask]
valid_Ki = Ki[~valid_vol.mask]
valid_vol = valid_vol[~valid_vol.mask]

points = np.column_stack((valid_Ti.ravel(), valid_Ki.ravel()))
values = valid_vol.ravel()

# 创建 RBFInterpolator 对象
rbf = RBFInterpolator(points, values, kernel='linear')  # 可选 kernel: 'linear', 'thin_plate_spline', 'gaussian', 'multiquadric', 'inverse_quadratic', 'inverse_multiquadric'

# 在原始数据范围内进行插值
Ti_flat = Ti.flatten()
Ki_flat = Ki.flatten()
interp_values = rbf(np.column_stack((Ti_flat, Ki_flat))).reshape(Ti.shape)

# 进行外推 (Ti=0, Ki=4500)
extrapolated_value = rbf(0, 4500)
print(f"Extrapolated value at (0, 4500): {extrapolated_value}")

# 可视化结果
fig = plt.figure(figsize=(12, 6))
ax = fig.add_subplot(111, projection='3d')

x = np.linspace(Ti.min(), Ti.max(), 100)
y = np.linspace(Ki.min(), Ki.max(), 100)
x, y = np.meshgrid(x, y)
z = rbf(x, y)

ax.plot_surface(x, y, z, cmap='viridis')
ax.set_xlabel('Ti')
ax.set_ylabel('Ki')
ax.set_zlabel('Interpolated Value')
ax.set_title('RBF Interpolation with Extrapolation')

plt.show()

代码解释：

数据准备： 首先，加载数据并将其转换为适合插值的格式。示例中使用了 pandas 读取CSV格式的字符串数据，并使用 numpy 处理数据。
创建 RBFInterpolator 对象： 使用 RBFInterpolator(points, values) 创建插值器对象。points 是一个二维数组，每一行代表一个数据点的坐标，values 是对应的数据值。 kernel 参数指定了使用的径向基函数类型。常用的 kernel 包括 'linear'（线性）, 'thin_plate_spline' (薄板样条), 'gaussian'（高斯）, 'multiquadric'（多二次）等。选择合适的 kernel 可以影响插值结果的平滑度和准确性。
进行插值： 使用 rbf(x, y) 对任意点进行插值。x 和 y 是要插值点的坐标。
进行外推： RBFInterpolator 可以直接用于计算原始数据范围之外的点。只需将超出范围的坐标传递给 rbf() 函数即可。
可视化： 使用 matplotlib 将插值结果可视化，以便直观地了解插值效果。

注意事项

外推的风险： 外推本质上是基于现有数据对未知区域进行预测。外推结果的准确性取决于数据的分布和模型的选择。在进行外推时，需要谨慎评估结果的可靠性。通常情况下，离原始数据越远，外推结果的不确定性越高。
Kernel选择： RBFInterpolator 提供了多种 kernel 函数。不同的 kernel 函数适用于不同的数据特征。选择合适的 kernel 函数可以提高插值和外推的准确性。
数据预处理： 在进行插值之前，对数据进行预处理可以提高插值效果。例如，可以对数据进行标准化或归一化处理，以消除量纲的影响。
参数调优： RBFInterpolator 提供了一些参数可以进行调整，例如 epsilon (对于某些 kernel 函数) 和 smoothing。通过调整这些参数，可以优化插值效果。