如何在ggplot2中连接历史数据与模型预测折线？

场景说明 #

使用R语言，拥有Historical_Data（历史数据）和Model_Predictions（三类预测数据：A、B、C）两个数据集，需求是用ggplot2绘制：

• 历史数据的折线图
• 三类预测的mean_value折线从历史数据最后一个点开始延伸
• 用low_value和high_value绘制预测的置信区间带

模拟数据集 #

先模拟符合结构的数据集：

library(tidyverse)

# 模拟历史数据：6个月的时间序列
set.seed(123)
Historical_Data <- tibble(
  date = seq.Date(as.Date("2023-01-01"), as.Date("2023-06-30"), by = "month"),
  value = rnorm(6, 50, 5)
)

# 模拟初始预测数据：无衔接历史的点
Model_Predictions <- tibble(
  category = rep(c("A", "B", "C"), each = 3),
  date = rep(seq.Date(as.Date("2023-07-01"), as.Date("2023-09-30"), by = "month"), 3),
  mean_value = c(rnorm(3, 52, 4), rnorm(3, 55, 3), rnorm(3, 48, 5)),
  low_value = mean_value - 3,
  high_value = mean_value + 3
)

初始问题：预测折线与历史数据断开 #

直接用初始数据绘图时，预测折线的起点是第一个预测日期，和历史数据最后一点没有连接：

# 初始绘图代码：折线无衔接
ggplot() +
  geom_line(data = Historical_Data, aes(x = date, y = value), color = "black", size = 1) +
  geom_line(data = Model_Predictions, aes(x = date, y = mean_value, color = category)) +
  geom_ribbon(data = Model_Predictions, aes(x = date, ymin = low_value, ymax = high_value, fill = category), alpha = 0.2) +
  theme_minimal()

解决方法：为预测数据集添加衔接点 #

核心思路是把历史数据的最后一个点，复制到每个预测类别下，作为预测折线的起始点：

# 提取历史数据最后一个点，并转换为预测数据的格式
last_hist_point <- Historical_Data %>% 
  slice(n()) %>% 
  mutate(
    category = NA,
    mean_value = value,
    low_value = value,
    high_value = value
  )

# 为每个预测类别添加衔接点，合并后按类别和日期排序
joined_predictions <- Model_Predictions %>%
  bind_rows(
    last_hist_point %>% mutate(category = "A"),
    last_hist_point %>% mutate(category = "B"),
    last_hist_point %>% mutate(category = "C")
  ) %>%
  arrange(category, date)

验证有效性：重新绘图 #

用修改后的数据集绘图，预测折线会从历史数据最后一点自然延伸，置信区间也同步衔接：

# 最终绘图代码：折线成功衔接
ggplot() +
  geom_line(data = Historical_Data, aes(x = date, y = value), color = "black", size = 1) +
  geom_line(data = joined_predictions, aes(x = date, y = mean_value, color = category)) +
  geom_ribbon(data = joined_predictions, aes(x = date, ymin = low_value, ymax = high_value, fill = category), alpha = 0.2) +
  theme_minimal()

内容的提问来源于stack exchange，提问作者stats_noob