generate_dataset.py

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "ed71965b",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "\n",
    "def simulate_corsia_mrv_fuel_burn(\n",
    "    n_airlines: int = 600,\n",
    "    years=range(2019, 2026),\n",
    "    seed: int = 42,\n",
    "\n",
    "    # Fuel burn 분포(heavy-tail): lognormal(mean=ln_mu, sigma=ln_sigma)\n",
    "    # exp(ln_mu)는 2019년 중앙값 근처의 연료소모량(ton fuel)\n",
    "    ln_mu: float = 12.8,            # exp(12.8) ≈ 363k ton fuel (중앙값 느낌)\n",
    "    ln_sigma: float = 1.20,         # 쏠림 정도 (1.0~1.5면 강한 상위 집중)\n",
    "\n",
    "    # MRV 대상처럼 \"너무 작은 사업자\" 제거(연간 최소 연료소모량)\n",
    "    mrv_min_annual_tfuel: float = 80_000,\n",
    "\n",
    "    # 항공사별 연평균 성장률(연료소모 성장)\n",
    "    g_mean: float = 0.03,\n",
    "    g_sd: float = 0.04,\n",
    "\n",
    "    # 연도 공통 충격(글로벌 수요/유가/운항환경)\n",
    "    shock_sd: float = 0.08,\n",
    "\n",
    "    # 항공사-연도 특이 변동(소규모 noise)\n",
    "    idio_sd: float = 0.05,\n",
    "):\n",
    "    rng = np.random.default_rng(seed)\n",
    "\n",
    "    # 간단 국가/지역 더미(항공사 국적 = AOC 발급국)\n",
    "    countries = [\n",
    "        (\"United States\", \"NAM\"), (\"China\", \"APAC\"), (\"United Kingdom\", \"EUR\"),\n",
    "        (\"Germany\", \"EUR\"), (\"France\", \"EUR\"), (\"United Arab Emirates\", \"MEA\"),\n",
    "        (\"Qatar\", \"MEA\"), (\"Turkey\", \"EUR\"), (\"Singapore\", \"APAC\"),\n",
    "        (\"Japan\", \"APAC\"), (\"Korea, Rep.\", \"APAC\"), (\"Australia\", \"APAC\"),\n",
    "        (\"India\", \"APAC\"), (\"Ethiopia\", \"AFR\"), (\"South Africa\", \"AFR\"),\n",
    "        (\"Brazil\", \"LAM\"), (\"Mexico\", \"LAM\"), (\"Chile\", \"LAM\"),\n",
    "        (\"Kenya\", \"AFR\"), (\"Canada\", \"NAM\")\n",
    "    ]\n",
    "    country_names = np.array([c[0] for c in countries])\n",
    "    country_regions = {c[0]: c[1] for c in countries}\n",
    "\n",
    "    # 국가 가중치(대형 시장에 항공사 더 배치되는 느낌)\n",
    "    weights = np.array([\n",
    "        0.16, 0.14, 0.06, 0.06, 0.05, 0.05,\n",
    "        0.04, 0.04, 0.04, 0.04, 0.04, 0.03,\n",
    "        0.06, 0.02, 0.02, 0.04, 0.03, 0.02,\n",
    "        0.02, 0.04\n",
    "    ])\n",
    "    weights = weights / weights.sum()\n",
    "\n",
    "    # 항공사 식별자 + 국적 부여\n",
    "    airline_ids = [f\"AL{str(i).zfill(4)}\" for i in range(1, n_airlines + 1)]\n",
    "    operator_state = rng.choice(country_names, size=n_airlines, p=weights)\n",
    "    region = np.array([country_regions[s] for s in operator_state])\n",
    "\n",
    "    # 2019 기준 \"기본 연료소모\" (heavy-tail)\n",
    "    base_fuel_2019 = rng.lognormal(mean=ln_mu, sigma=ln_sigma, size=n_airlines)\n",
    "\n",
    "    # MRV 대상처럼 최소 컷 적용 (부족하면 추가 샘플로 채움)\n",
    "    mask = base_fuel_2019 >= mrv_min_annual_tfuel\n",
    "    while mask.sum() < n_airlines:\n",
    "        extra = rng.lognormal(mean=ln_mu, sigma=ln_sigma, size=n_airlines)\n",
    "        base_fuel_2019 = np.where(mask, base_fuel_2019, extra)\n",
    "        mask = base_fuel_2019 >= mrv_min_annual_tfuel\n",
    "\n",
    "    # 항공사별 성장률\n",
    "    g_i = rng.normal(loc=g_mean, scale=g_sd, size=n_airlines)\n",
    "    g_i = np.clip(g_i, -0.10, 0.20)\n",
    "\n",
    "    # 연도별 공통 충격(로그 스케일)\n",
    "    years = list(years)\n",
    "    year_shocks = {y: rng.normal(loc=0.0, scale=shock_sd) for y in years}\n",
    "\n",
    "    rows = []\n",
    "    for idx in range(n_airlines):\n",
    "        al_id = airline_ids[idx]\n",
    "        st = operator_state[idx]\n",
    "        reg = region[idx]\n",
    "        f0 = base_fuel_2019[idx]\n",
    "        gi = g_i[idx]\n",
    "\n",
    "        for y in years:\n",
    "            t = y - 2019\n",
    "            eps = rng.normal(0.0, idio_sd)\n",
    "\n",
    "            # FuelBurn = f0 * (1+gi)^t * exp(year_shock) * exp(idio_noise)\n",
    "            fuel = f0 * ((1.0 + gi) ** t) * np.exp(year_shocks[y]) * np.exp(eps)\n",
    "\n",
    "            rows.append({\n",
    "                \"airline_id\": al_id,\n",
    "                \"operator_state\": st,\n",
    "                \"region\": reg,\n",
    "                \"year\": y,\n",
    "                \"fuel_burn_tonnes\": float(fuel),\n",
    "                \"g_i\": float(gi),\n",
    "                \"year_shock\": float(year_shocks[y])\n",
    "            })\n",
    "\n",
    "    df = pd.DataFrame(rows).sort_values([\"airline_id\", \"year\"]).reset_index(drop=True)\n",
    "\n",
    "    # 연도별 점유율(상위 집중 확인용)\n",
    "    df[\"fuel_share_in_year\"] = df.groupby(\"year\")[\"fuel_burn_tonnes\"].apply(lambda s: s / s.sum()).reset_index(level=0, drop=True)\n",
    "\n",
    "    return df\n",
    "\n",
    "\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "    df_fuel = simulate_corsia_mrv_fuel_burn(\n",
    "        n_airlines=600,\n",
    "        years=range(2019, 2026),\n",
    "        seed=7,\n",
    "        ln_mu=12.8,\n",
    "        ln_sigma=1.25,\n",
    "        mrv_min_annual_tfuel=100_000\n",
    "    )\n",
    "\n",
    "    print(df_fuel.head(10))\n",
    "\n",
    "    print(\"\\nYearly totals (million tonnes fuel):\")\n",
    "    print((df_fuel.groupby(\"year\")[\"fuel_burn_tonnes\"].sum() / 1e6).round(2))\n",
    "\n",
    "    top10_share = (\n",
    "        df_fuel.sort_values([\"year\", \"fuel_burn_tonnes\"], ascending=[True, False])\n",
    "               .groupby(\"year\").head(10)\n",
    "               .groupby(\"year\")[\"fuel_burn_tonnes\"].sum()\n",
    "        / df_fuel.groupby(\"year\")[\"fuel_burn_tonnes\"].sum()\n",
    "    )\n",
    "    print(\"\\nTop-10 airline fuel share by year:\")\n",
    "    print(top10_share.round(3))\n",
    "\n",
    "# df_fuel[df_fuel['year'] == 2019].to_csv('./dataset/fuel_burn_2019_랜덤생성.csv', index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "fe95fd9c",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "bibiml",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}