给定正弦拟合的预测值

Question

我正在使用 Python 来拟合具有正弦函数的时间序列。我找到了一个很好的匹配项，现在我希望能够预测未来的值。我在这里迷路了。

这是我得到的：

timeSeries = [0.01146, 0.00724, 0.00460, 0.00192, 0.00145, 0.01559, 0.02585, 0.04118, 0.05073, 0.01966, 0.01486, 0.02784]

import numpy as np
from scipy.optimize import curve_fit

def createSinFromFit(x, freq, amplitude, phase, offset):
    return np.sin(x * freq + phase) * amplitude + offset

def sinRegr(series):
    t = np.linspace(0, 4*np.pi, len(series))
    guess_freq = 1
    guess_amplitude = 3*np.std(series)/(2**0.5)
    guess_phase = 0
    guess_offset = np.mean(series)
    p0=[guess_freq, guess_amplitude, guess_phase, guess_offset]
    fit = curve_fit(createSinFromFit, t, series, p0=p0)
    results = createSinFromFit(t,*fit[0])
    return results

plotThis = sinRegr(timeSeries)

此代码生成您在此图片中看到的配件：

如何扩展 sin 函数以便它预测系列的未来点？即我怎样才能让正弦图跨越到右侧，超出 'known' 数据点覆盖的区域？

Answer 1

您需要区分数据时间线（输入）和拟合时间线（输出）。一旦你这样做了，方法就很清楚了。下面我称它们为 tdata 和 tfit:

import numpy as np
from scipy.optimize import curve_fit
import matplotlib.pyplot as plt

tdata = np.linspace(0, 10)
timeSeries = np.sin(tdata) + .4*np.random.random(tdata.shape)

def createSinFromFit(x, freq, amplitude, phase, offset):
    return np.sin(x * freq + phase) * amplitude + offset

def sinRegr(tdata, series):
    tfit = np.linspace(0, 6*np.pi, len(series))
    guess_freq = .2
    guess_amplitude = 3*np.std(series)/(2**0.5)
    guess_phase = 0
    guess_offset = np.mean(series)
    p0=[guess_freq, guess_amplitude, guess_phase, guess_offset]
    fit = curve_fit(createSinFromFit, tdata, series, p0=p0)   # use tdata to create the fit
    results = createSinFromFit(tfit,*fit[0])                  # use tfit to generate a new curve 
    return tfit, results

tfit, plotThis = sinRegr(tdata, timeSeries)

plt.plot(tfit, plotThis)
plt.plot(tdata, timeSeries, "ro")
plt.show()

给定正弦拟合的预测值

predicting values given a sinusouidal fit

python

regression

numpy

time-series

scipy