TSstudio时间序列包使用

TSstudio包提供了一组时间序列数据的描述性和预测性分析工具。这包括用于预处理时间序列数据的实用功能、基于plotly包引擎的交互式可视化功能，以及用于训练和评估预测的forecast、bsts等包中的时间序列预测模型的一组工具，下面小编对该包的使用进行介绍。

• ts_info对象属性函数

install.packages("TSstudio")
library(TSstudio)
> NGC
     Qtr1   Qtr2   Qtr3   Qtr4
2000 2050.6 1513.1 1475.0 2587.5
2001 2246.6 1444.4 1494.1 2120.2
2002 2258.4 1591.4 1542.2 2378.9
> ts_info(NGC)
 The NGC series is a ts object with 1 variable and 80 observations
 Frequency: 4 
 Start time: 2000 1 
 End time: 2019 4
ts_info给出了这个ts对象的一些基本属性，包括长度、频率和起止时间。周期/频率是规则的时间序列对象很重要的属性，周期是规则且重复的时间单元，它将序列划分成连续等长的子集。频率定义了周期中的单元个数或者单元长度。在NGC中，周期是1年，观测单元是季度，所以频率就是4。

• ts_lags可视化

ts_lags(USgas, lags = c(12,24,36,48))

• ts_plot可视化函数

ts_plot(tvs_ts, title = "US Monthly total vehicle sales",
        Ytitle = "Thousands of Vehicle",slider = TRUE)

• ts_cor函数

UK_ts <- ts(UKdaily$ND, 
            start = c(year(start_date), yday(start_date)),
            frequency = 365)
ts_cor(UK_ts, lag.max = 365 * 4)#分析该系列与过去四年的滞后之间的关系

• ts_ma函数

two_sided_ma <- ts_ma(ts.obj = USVSales,#ts类型
                      n = c(2,5),#设定5号和11号的平均线
                      n_left = 6, n_right = 5, #设置12号平均线
                      plot = TRUE, multiple = TRUE,
                      margin = 0.04)
two_sided_ma

• ts_quantile函数

ts_quantile(UKgrid, period = "monthly", n = 2)

• ccf_plot可视化函数

ccf_plot(x = USVSales, y = USUnRate, lags = 0:12)

• ts_split函数

#创建训练和测试集
USgas_partitions <- ts_split(USgas, sample.out = 12)
train <- USgas_partitions$train
test <- USgas_partitions$test

• test_forecast模型预测函数

md <- auto.arima(train)
checkresiduals(md)
fc <- forecast (md, h = 12)
accuracy(fc,test)
test_forecast(actual = USgas,
              forecast.obj = fc,
              test = test)
naive_model <- naive(train, h = 12)
test_forecast(actual = USgas,
              forecast.obj = naive_model,
              test = test)
accuracy(naive_model,test)
snaive_model <- snaive(train, h = 12)
test_forecast(actual = USgas,
              forecast.obj = snaive_model,
              test = test)
accuracy(snaive_model, test)

• plot_forecast模型预测可视化函数

md_final <- auto.arima(USgas)
fc_final <- forecast (md_final, h = 12)
plot_forecast(fc_final,
              title = "The US Natural Gas Consumption Forecast",
              Xtitle = "Year",Ytitle = "Billion Cubic Feet")
#Confidence interval
fc_final2 <- forecast(md_final, h = 60, level = c(80,90))
plot_forecast(fc_final2,
              title = "The US Natural Gas Consumption Forecast",
              Xtitle = "Year",Ytitle = "Billion Cubic Feet")

• train_model模型训练

library(TSstudio)
USgas_forecast <- train_model(ts.obj = USgas,
                                 periods = 6, models = "abehntw",
                                 error = "MAPE",
                                 window_size = 12,
                                 h = 60, plot = FALSE)

• ts_grid网格函数

deep_grid <-  ts_grid(train,
 			model = "HoltWinters",periods = 6,
                      window_space = 6,
                      window_test = 12,
                      hyper_params = list(alpha = seq(0.1,0.2,0.1),
                                          beta = seq(0,0.2,0.1),
                                          gamma = seq(0.2,0.3,0.1)),
                      parallel = TRUE, n.cores = 8)
plot_grid(deep_grid, type = "3D", top = 250)