利用python，创建时序数据

@创建于：20210715
@修改于：20210715

最近在做时序预测，不想把精力放在各种时序数据的处理上，决定自己生成一些时间序列数据。供研究用。

# -*- coding:UTF-8 -*- 

# datetime:2021/7/15 10:22
# software: PyCharm

"""
 文件说明：
 创建时序数据：包括基线、趋势、周期（2种）
 """

import numpy as np
import pandas as pd
import random
import matplotlib.pyplot as plt
import datetime


class GenerateTimeSeries():

    def __init__(self,  base=100, trend=0.02, freq='H', n_freq=1000, period=7, time_end=None):
        self.base = base
        self.trend = trend
        self.freq = freq
        self.n_freq = n_freq
        self.period = period
        self.time_end = time_end

        self.use_rnd = True
        self.period_2nd = self.period * 4

        if ('24' in self.freq) or (freq=='H'):
            self.freq = 'H'
            self.period = 24
            self.period_2nd = self.period * 7
        elif ('96' in self.freq) or (freq=='15T'):
            self.freq = '15T'
            self.period = 96
            self.period_2nd = self.period * 7

    def __str__(self):
        return '制造具有基准线、趋势、周期和随机扰动的时序序列。单个变量。'

    def get_datetime(self):

        if self.time_end is None:
            self.time_end = datetime.date.today()
        return pd.date_range(end=self.time_end, periods=self.n_freq, freq=self.freq)

    def get_base(self):
        if self.use_rnd:
            # 线性随机 更好的模拟 基础数据的随机变动
            return random.uniform(self.base*(1-0.003), self.base*(1+0.003))
        else:
            return self.base

    def get_trend(self, number=1):
        if self.use_rnd:
            # 线性随机 vs 高斯随机
            rnd = random.uniform(-1 * self.base * 0.005, self.base * 0.005)
            rnd = random.gauss(mu=0, sigma=self.base * 0.002)
        else:
            rnd = 0
        return self.trend * number + rnd

    def get_period(self, number=1):
        if self.use_rnd:
            # 随机数加载乘数上，对任何周期会产生同等的随机效果。
            number = random.gauss(mu=number, sigma=self.period * 0.05)
            return self.base * self.trend * np.sin(number * 2 * np.pi / self.period)

            # # 改变周期形式，效果差：渐变式周期消失, 因为number是在变大的，这种方式不可取。
            # temp = random.gauss(mu=self.period, sigma=self.period * 0.002)
            # return self.base*self.trend*np.sin(number * 2 * np.pi / temp)
        else:
            return self.period

    def get_period_2nd(self, number=1):

        if self.use_rnd:
            # 改变乘数方式，效果好
            number = random.gauss(mu=number, sigma=self.period_2nd * 0.01)
            return self.base * self.trend * np.sin(number * 2 * np.pi / self.period_2nd)
        else:
            return self.period

    def get_rnd(self):
        if self.use_rnd:
            return random.gauss(mu=0, sigma=self.base * 0.002)
        else:
            return 0

    def get_value(self):
        data_list = []
        i = 1
        while i <= self.n_freq:
            data_point = self.get_base() + self.get_trend(i) + \
                         self.get_period_2nd(number=i) + self.get_rnd()
            data_list.append(data_point)
            i += 1
        return data_list

    def generate_time_series(self):
        df = self.get_datetime()
        df = pd.Series(self.get_value(), index=df)
        return df


gts = GenerateTimeSeries()
df_ts = gts.generate_time_series()
df_ts.plot()
plt.show()

结果图如下：

还需持续完善~