{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "# 粤AGP3080 2026-08-11 用氢量差异诊断\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "## tl;dr\n",
        "\n",
        "- 生产数据库当前值为 **5.992 kg**，分段关键帧合并值为 **12.051 kg**，差 **6.059 kg（+101.12%）**。\n",
        "- 根因已确认：同一车辆当天使用了3个 `source_endpoint`。现有算法按 `VIN + source_endpoint` 分组后只选择样本最多的一组，导致后续两个连续时间段被舍弃。\n",
        "- 三个来源片段按现有算法分别得到5.992、0.697、5.349 kg，合计12.038 kg；与分段关键帧结果仅差0.013 kg。\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "## Context & Methods\n",
        "\n",
        "诊断对象：粤AGP3080，VIN `LB9A32A25R0LS1452`，2026-08-11（Asia/Shanghai）。\n",
        "\n",
        "对比口径：生产表日结果、现有算法按连接来源分组的结果、合并全部来源后按工作段首尾5帧中位数做差的结果。加氢、异常跳变和超过5分钟的数据断点用于切段。\n",
        "\n",
        "### Key Assumptions\n",
        "\n",
        "三个端口属于同一VIN当天的连接重连，不是三个独立车辆数据源；百公里氢耗采用生产日里程502.4km。\n",
        "\n",
        "> 执行状态：生产查询已于2026-08-19完成并复核。本地缺少 `nbformat`、`jupyter` 和可用内核，因此本笔记本保存可复算代码但未在本机执行。复跑命令：`python -m jupyter nbconvert --execute --to notebook --inplace <notebook>`。\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "## Data\n",
        "\n",
        "| 来源连接 | 时间范围 | 原始帧 | 工作帧 | 现有算法用氢量 |\n",
        "|---|---|---:|---:|---:|\n",
        "| 8.134.95.166:54760 | 06:28:12–12:34:43 | 1542 | 1539 | 5.992kg |\n",
        "| 8.134.95.166:41692 | 12:35:13–14:38:29 | 328 | 328 | 0.697kg |\n",
        "| 8.134.95.166:37898 | 14:54:00–19:59:02 | 1357 | 1357 | 5.349kg |\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {},
      "outputs": [],
      "source": [
        "mileage_km = 502.4\n",
        "stored_kg = 5.992\n",
        "source_consumption_kg = [5.992, 0.697, 5.349]\n",
        "segment_consumption_kg = [0.462, 2.537, 3.685, 0.000, 2.629, 2.738]\n",
        "source_sum_kg = round(sum(source_consumption_kg), 3)\n",
        "proposed_kg = round(sum(segment_consumption_kg), 3)\n",
        "gap_kg = round(proposed_kg - stored_kg, 3)\n",
        "gap_pct = round((proposed_kg / stored_kg - 1) * 100, 2)\n",
        "omitted_sources_kg = round(source_sum_kg - stored_kg, 3)\n",
        "method_residual_kg = round(proposed_kg - source_sum_kg, 3)\n",
        "current_rate = round(stored_kg / mileage_km * 100, 3)\n",
        "proposed_rate = round(proposed_kg / mileage_km * 100, 3)\n",
        "assert (source_sum_kg, proposed_kg, gap_kg) == (12.038, 12.051, 6.059)\n",
        "{\n",
        "    '数据库用氢量_kg': stored_kg,\n",
        "    '新算法用氢量_kg': proposed_kg,\n",
        "    '差值_kg': gap_kg,\n",
        "    '差值比例_pct': gap_pct,\n",
        "    '被舍弃来源片段_kg': omitted_sources_kg,\n",
        "    '方法残差_kg': method_residual_kg,\n",
        "    '数据库百公里氢耗': current_rate,\n",
        "    '新算法百公里氢耗': proposed_rate,\n",
        "}\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "## Takeaways\n",
        "\n",
        "1. 差值不是压力换算公式造成的；被舍弃的两个连接片段贡献6.046kg，解释总差值的99.79%。\n",
        "2. 问题在聚合粒度：`source_endpoint` 包含临时TCP端口，重连会改变端口，不能作为车辆日统计的互斥来源键。\n",
        "3. 当前生产结果只保留06:28–12:34的数据，12:35–19:59的连续有效工作数据没有进入日用氢量。\n",
        "4. 建议先按VIN和事件时间合并去重，再识别工作段；来源字段只用于追踪。\n"
      ]
    }
  ],
  "metadata": {
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "name": "python",
      "version": "3"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 5
}