检测时间序列的快速增长

You could try differencing the series to get the amount the series changed at each step. This is a standard method in time-series analysis. Here is a good discussion of the theory behind the idea.

In practice, pandas gives us a simple way to difference the series via the .diff() method.

df = pd.DataFrame({'a': [10, 20, 30, 40, 50, 60],
                   'b': [1, 1000, 20000, 3, 50, 80],
                   'c': [1000, 4, 97, 16, 2500, 36]})
df.diff()

      a        b       c
0   NaN      NaN     NaN
1  10.0    999.0  -996.0
2  10.0  19000.0    93.0
3  10.0 -19997.0   -81.0
4  10.0     47.0  2484.0
5  10.0     30.0 -2464.0

This process subtracts the next item in the series from the prior (which is why the first observation is now NaN). You can add a feature to your data that is the first-differenced series:

df['a_first_difference'] = df['a'].diff()

然后，该系列中的值遵循它们自己的分布。也许找到差异序列中高于某个阈值的所有值，将为您提供一种方法，通过识别与前一个值相比变化最大的时期来衡量“快速增长”。还可以.diff()告诉您下一个值的更改：

df.diff(periods=-1)

      a        b       c
0 -10.0   -999.0   996.0
1 -10.0 -19000.0   -93.0
2 -10.0  19997.0    81.0
3 -10.0    -47.0 -2484.0
4 -10.0    -30.0  2464.0
5   NaN      NaN     NaN

这样，这将有助于确定下一步发生较大变化的时期。