时序数据异常检测

library(tidyverse)
library(anomalize)
data("tidyverse_cran_downloads")

# time_decompose（）：将时间序列分为季节性，趋势和余数
# anomalize（）：将异常检测方法应用于余数组件。
# time_recompose（）：计算将“正常”数据与异常分开

tidyverse_cran_downloads %>%
  ggplot(aes(date, count)) +
  geom_point(color = "#2c3e50", alpha = 0.25) +
  facet_wrap(~ package, scale = "free_y", ncol = 3) +
  theme_minimal() +
  theme(axis.text.x = element_text(angle = 30, hjust = 1)) +
  labs(title = "Tidyverse Package Daily Download Counts",
       subtitle = "Data from CRAN by way of cranlogs package")

tidyverse_cran_downloads %>%
  # Data Manipulation / Anomaly Detection
  time_decompose(count, method = "stl") %>%
  anomalize(remainder, method = "iqr") %>%
  time_recompose() %>%
  # Anomaly Visualization
  plot_anomalies(time_recomposed = TRUE, ncol = 3, alpha_dots = 0.25) +
  labs(title = "Tidyverse Anomalies", subtitle = "STL + IQR Methods")

# Get only lubridate downloads
lubridate_dloads <- tidyverse_cran_downloads %>%
  filter(package == "lubridate") %>%
  ungroup()

# Anomalize!!
lubridate_dloads %>%
  # Twitter + GESD(Generalized Extreme Studentized Deviation)
  time_decompose(count, method = "twitter", trend = "2 months") %>%
  anomalize(remainder, method = "gesd") %>%
  time_recompose() %>%
  # Anomaly Visualziation
  plot_anomalies(time_recomposed = TRUE) +
  labs(title = "Lubridate Anomalies", subtitle = "Twitter + GESD Methods")

#iqr（）和gesd（）仅适用于数字数据。 请注意，对于时间序列数据，应已删除趋势和季节性。

lubridate_dloads %>%
  # STL + IQR Anomaly Detection
  time_decompose(count, method = "stl", trend = "2 months") %>%
  anomalize(remainder, method = "iqr") %>%
  time_recompose() %>%
  # Anomaly Visualization
  plot_anomalies(time_recomposed = TRUE) +
  labs(title = "Lubridate Anomalies", subtitle = "STL + IQR Methods")


# Time Frequency
time_frequency(lubridate_dloads, period = "auto")
# Time Trend
time_trend(lubridate_dloads, period = "auto")

tidyverse_cran_downloads %>%
  filter(package == "lubridate") %>%
  ungroup() %>%
  time_decompose(count) %>%
  anomalize(remainder) %>%
  plot_anomaly_decomposition() +
  labs(title = "Decomposition of Anomalized Lubridate Downloads")