重采样后pandas Series执行scipy FFT触发索引错误的解决方案问询

你触发ValueError: Can only tuple-index with a MultiIndex的核心原因是resampled_signal是一维pandas Series结构，仅包含数值列和时间索引，不存在第二列，[:,1]是二维数据（DataFrame/多索引Series）的索引写法，应用在普通Series上就会触发索引错误。

仅需要调整FFT调用的传参即可，不需要破坏原有Series结构：
将原有代码中的

yf = fft(resampled_signal[:,1])

修改为

yf = fft(resampled_signal.values)

resampled_signal.values会直接提取Series的数值数组传入FFT接口，原有resampled_signal的Series结构、时间索引完全保留不受影响。同时你需要把测试代码中print(resampled_signal.head())后的exit()注释或删除，否则程序会提前终止不会运行FFT相关逻辑。

要实现两个数据集重采样后点数一致，按以下步骤操作即可：

• 先将两个原始信号都转为带时间索引的Series结构
• 计算两个信号时间范围的交集作为公共计算区间：

common_start = max(signal1.index.min(), signal2.index.min())
common_end = min(signal1.index.max(), signal2.index.max())

• 用完全相同的重采样规则对两个信号做裁剪、重采样、插值处理：

RESAMPLE_RULE = '1S'
# 裁剪到公共时间区间
signal1_cut = signal1.loc[common_start:common_end]
signal2_cut = signal2.loc[common_start:common_end]
# 统一重采样+插值补全空值
signal1_resampled = signal1_cut.resample(RESAMPLE_RULE).mean().interpolate(method='time')
signal2_resampled = signal2_cut.resample(RESAMPLE_RULE).mean().interpolate(method='time')

• 处理完成后校验长度一致即可：assert len(signal1_resampled) == len(signal2_resampled)，之后分别调用FFT计算结果再做差值运算即可。

import numpy as np
import pandas as pd 
from datetime import datetime
from datetime import timedelta
import matplotlib
import matplotlib.pyplot as plt
from scipy.fft import fft, fftfreq


datathick = "20210728_rig_thick.csv" 

with open(datathick) as f:
        lines = f.readlines()
        dates = [str(line.split(',')[0]) for line in lines]
        thick = [float(line.split(',')[1]) for line in lines]
        z = [float(line.split(',')[2]) for line in lines]

        date_thick = [datetime.strptime(x,'%Y-%m-%dT%H:%M:%S.%f').time() for x in dates]

time_list_thick = []
for i in np.arange(0, len(date_thick)):
    q = date_thick[i]
    t = timedelta(hours= q.hour, minutes=q.minute,seconds=q.second, microseconds = q.microsecond).total_seconds()
    time_list_thick.append(float(t))
    

#---RESCALE-----
signal = pd.Series(thick, index = pd.TimedeltaIndex(time_list_thick,unit = 's'))
resampled_signal = signal.resample('1S').mean()
resampled_signal = resampled_signal.interpolate(method='time')

print(resampled_signal.head())

# 注释掉提前退出的代码
# exit()
#----FFT Transform of Output and Noise ----
# Number of samples in normalized_tone

SAMPLE_RATE = 1
DURATION = len(resampled_signal)
N = SAMPLE_RATE * DURATION

# 修改FFT传参
yf = fft(resampled_signal.values)
print(yf)
xf = fftfreq(N, 1 / SAMPLE_RATE)

#------------------------------------------------
fig=plt.figure(figsize=(7.,7.))

ax=fig.add_subplot(1,1,1)
ax.set_zorder(1)
ax.patch.set_visible(False)
ax.spines['right'].set_visible(False)
ax.spines['top'].set_visible(False)
ax.spines['left'].set_visible(False)
ax.spines['bottom'].set_visible(False)
ax.set_xlabel('Frequency (Hz)')
ax.set_ylabel('Amplitude (a.u)')
ax.minorticks_on() # enable minor ticks
ax.xaxis.set_ticks_position('bottom')
ax.spines['left'].set_color('black')
ax.yaxis.label.set_color('black')

plt.yscale('log')
ax.tick_params(direction='out', axis='y', which='both', pad=4, colors='black')
ax.grid(b=True, which='major', color='#eeeeee', linestyle='-', zorder=1, linewidth=0.4) # turn on major grid
ax.grid(b=True, which='minor', color='#eeeeee', linestyle='-', zorder=1, linewidth=0.4) # turn on minor grid

ax.plot(np.abs(xf), np.abs(yf))

plt.savefig('fft.pdf', dpi=300, bbox_inches='tight', format='pdf')
plt.savefig('fft.png', dpi=300, bbox_inches='tight', format='png')

内容的提问来源于stack exchange，提问作者tjsmert44