R Shiny价格预测应用数据过滤问题求助

核心修正方向 #

针对三个维度（曲线名称、峰值类型、交割期）的过滤逻辑进行精准约束，再基于过滤后的时序数据拟合ARIMA模型并生成预测图表。

具体实现步骤 #

1. 修正数据过滤逻辑 #

确保仅筛选与用户选择完全匹配的数据，同时保证时序数据的连续性：

filtered_data <- reactive({
  # 强制要求所有输入参数已选择
  req(input$curve_name, input$peak_type, input$delivery_period)
  # 多维度精准过滤
  df %>%
    filter(
      curve_name == input$curve_name,
      peak_type == input$peak_type,
      delivery_period == input$delivery_period
    ) %>%
    arrange(date) # 按日期排序，满足ARIMA模型对时序连续性的要求
})

2. ARIMA模型拟合与预测 #

基于过滤后的时序数据自动选择最优ARIMA参数，并生成指定周期的预测结果：

arima_forecast <- reactive({
  data <- filtered_data()
  # 转换为时间序列对象，frequency根据数据频率调整（月度用12，日度用365）
  ts_data <- ts(data$price, frequency = 12)
  # 自动拟合最优ARIMA模型
  optimal_arima <- auto.arima(ts_data)
  # 生成未来12个周期的预测，可按需修改h值
  forecast(optimal_arima, h = 12)
})

3. 生成预测可视化图表 #

整合原始数据与预测结果，绘制带置信区间的趋势图：

output$price_pred_plot <- renderPlot({
  pred_result <- arima_forecast()
  # 整理原始数据
  original_df <- data.frame(
    date = filtered_data()$date,
    price = filtered_data()$price,
    type = "原始价格"
  )
  # 整理预测数据（含置信区间）
  forecast_df <- data.frame(
    date = seq(max(filtered_data()$date), by = "month", length.out = 12),
    price = as.numeric(pred_result$mean),
    type = "预测价格",
    lower = as.numeric(pred_result$lower[,2]),
    upper = as.numeric(pred_result$upper[,2])
  )
  # 绘图
  ggplot() +
    geom_line(data = original_df, aes(x = date, y = price, color = type), size = 1) +
    geom_line(data = forecast_df, aes(x = date, y = price, color = type), size = 1, linetype = "dashed") +
    geom_ribbon(data = forecast_df, aes(x = date, ymin = lower, ymax = upper), alpha = 0.2, fill = "#2196F3") +
    scale_color_manual(values = c("原始价格" = "#333333", "预测价格" = "#2196F3")) +
    labs(
      title = paste(input$curve_name, "-", input$peak_type, "-", input$delivery_period, "价格预测"),
      x = "日期", y = "价格"
    ) +
    theme_minimal()
})

完整服务器端代码示例 #

server <- function(input, output) {
  # 数据过滤
  filtered_data <- reactive({
    req(input$curve_name, input$peak_type, input$delivery_period)
    df %>%
      filter(
        curve_name == input$curve_name,
        peak_type == input$peak_type,
        delivery_period == input$delivery_period
      ) %>%
      arrange(date)
  })

  # ARIMA预测
  arima_forecast <- reactive({
    data <- filtered_data()
    ts_data <- ts(data$price, frequency = 12)
    optimal_arima <- auto.arima(ts_data)
    forecast(optimal_arima, h = 12)
  })

  # 渲染预测图表
  output$price_pred_plot <- renderPlot({
    pred_result <- arima_forecast()
    original_df <- data.frame(
      date = filtered_data()$date,
      price = filtered_data()$price,
      type = "原始价格"
    )
    forecast_df <- data.frame(
      date = seq(max(filtered_data()$date), by = "month", length.out = 12),
      price = as.numeric(pred_result$mean),
      type = "预测价格",
      lower = as.numeric(pred_result$lower[,2]),
      upper = as.numeric(pred_result$upper[,2])
    )

    ggplot() +
      geom_line(data = original_df, aes(x = date, y = price, color = type), size = 1) +
      geom_line(data = forecast_df, aes(x = date, y = price, color = type), size = 1, linetype = "dashed") +
      geom_ribbon(data = forecast_df, aes(x = date, ymin = lower, ymax = upper), alpha = 0.2, fill = "#2196F3") +
      scale_color_manual(values = c("原始价格" = "#333333", "预测价格" = "#2196F3")) +
      labs(
        title = paste(input$curve_name, "-", input$peak_type, "-", input$delivery_period, "价格预测"),
        x = "日期", y = "价格"
      ) +
      theme_minimal()
  })
}

关键注意点 #

• 确保数据集包含date（日期列）、price（价格列）、curve_name、peak_type、delivery_period五个核心字段
• 若数据频率非月度，需修改ts()函数的frequency参数（如日度数据设为365）
• 如需支持多组数据同时预测，可将过滤逻辑改为分组循环处理

内容的提问来源于stack exchange，提问作者Miguel.H