app.py

import streamlit as st
import pandas as pd
import numpy as np
import folium
from streamlit_folium import st_folium
import requests
import json
import os

st.set_page_config(page_title="심야 대중교통 최적화 대시보드", layout="wide")

@st.cache_data
def fetch_osm_data_v4():
    bbox = "37.491,127.020,37.505,127.035"
    endpoints = [
        "https://overpass-api.de/api/interpreter",
        "https://overpass.kumi.systems/api/interpreter",
        "https://overpass.osm.ch/api/interpreter"
    ]
    
    query = f"""
    [out:json][timeout:60];
    (
      relation["route"="bus"]["ref"~"N13|N15|N37|N75"]({bbox});
      node["highway"="bus_stop"]({bbox});
      node["amenity"~"pub|bar|nightclub|restaurant"]({bbox});
      node["highway"="street_lamp"]({bbox});
      node["man_made"="surveillance"]({bbox});
    );
    out body geom;
    """
    
    headers = {'User-Agent': 'SmartTransitMVP/2.0'}
    data = None
    cache_file = "osm_backup.json"
    
    for url in endpoints:
        try:
            response = requests.post(url, data=query, headers=headers, timeout=65)
            response.raise_for_status()
            data = response.json()
            with open(cache_file, "w", encoding="utf-8") as f:
                json.dump(data, f)
            break
        except Exception as e:
            print(f"Failed to fetch from {url}: {e}")
            continue
            
    if data is None:
        if os.path.exists(cache_file):
            st.warning("⚠️ 실시간 서버(Overpass API) 접속 지연으로 인해 기존 백업 데이터를 로드했습니다.")
            with open(cache_file, "r", encoding="utf-8") as f:
                data = json.load(f)
        else:
            st.error("데이터 서버 접속이 원활하지 않습니다. 잠시 후 다시 시도해주세요.")
            data = {'elements': []}
        
    bus_stops, amenities, safety_infra, real_bus_routes = [], [], [], []
    for element in data.get('elements', []):
        if element['type'] == 'node':
            lat = float(element['lat'])
            lon = float(element['lon'])
            tags = element.get('tags', {})
            if 'highway' in tags and tags['highway'] == 'bus_stop':
                bus_stops.append({'lat': lat, 'lon': lon, 'stop_id': str(element['id']), 'name': str(tags.get('name', '정류장'))})
            elif 'amenity' in tags:
                amenities.append({'lat': lat, 'lon': lon})
            elif 'highway' in tags or 'man_made' in tags:
                safety_infra.append({'lat': lat, 'lon': lon})
        elif element['type'] == 'relation':
            tags = element.get('tags', {})
            ref = str(tags.get('ref', 'N버스'))
            name = str(tags.get('name', ref))
            coords = []
            for member in element.get('members', []):
                if member['type'] == 'way' and 'geometry' in member:
                    for pt in member['geometry']:
                        coords.append([float(pt['lat']), float(pt['lon'])])
            if coords:
                real_bus_routes.append({'name': name, 'ref': ref, 'coords': coords})
                
    return pd.DataFrame(bus_stops), pd.DataFrame(amenities), pd.DataFrame(safety_infra), real_bus_routes

stops_df, amenities_df, safety_df, real_bus_routes = fetch_osm_data_v4()

@st.cache_data
def create_grid_features(_stops, _amenities, _safety):
    lats = np.linspace(37.492, 37.504, 20)
    lons = np.linspace(127.021, 127.034, 20)
    grid_data = []
    
    stop_coords = _stops[['lat', 'lon']].values if not _stops.empty else np.array([])
    amenity_coords = _amenities[['lat', 'lon']].values if not _amenities.empty else np.array([])
    safety_coords = _safety[['lat', 'lon']].values if not _safety.empty else np.array([])
    
    for lat in lats:
        for lon in lons:
            point = np.array([lat, lon])
            demand = int(np.sum(np.sqrt(np.sum((amenity_coords - point)**2, axis=1)) < 0.002)) if len(amenity_coords) > 0 else 0
            deficit = float(np.min(np.sqrt(np.sum((stop_coords - point)**2, axis=1)))) if len(stop_coords) > 0 else 0.0
            safety_count = int(np.sum(np.sqrt(np.sum((safety_coords - point)**2, axis=1)) < 0.002)) if len(safety_coords) > 0 else 0
            grid_data.append({'lat': float(lat), 'lon': float(lon), 'raw_demand': demand, 'raw_deficit': deficit, 'raw_safety_count': safety_count})
            
    df = pd.DataFrame(grid_data)
    df['base_demand'] = df['raw_demand'] / df['raw_demand'].max() if df['raw_demand'].max() > 0 else 0.0
    df['base_deficit'] = df['raw_deficit'] / df['raw_deficit'].max() if df['raw_deficit'].max() > 0 else 0.0
    df['base_risk'] = 1.0 - (df['raw_safety_count'] / df['raw_safety_count'].max()) if df['raw_safety_count'].max() > 0 else 1.0
    return df

grids_df = create_grid_features(stops_df, amenities_df, safety_df)

if not stops_df.empty and len(real_bus_routes) > 0:
    nbus_coords = np.array([pt for route in real_bus_routes for pt in route['coords']])
    stops_df['min_dist_to_nbus'] = stops_df.apply(lambda r: float(np.min(np.sqrt((nbus_coords[:,0]-r['lat'])**2 + (nbus_coords[:,1]-r['lon'])**2))), axis=1)
    stops_df['is_nbus_stop'] = stops_df['min_dist_to_nbus'] < 0.001
else:
    stops_df['is_nbus_stop'] = False

nbus_stops = stops_df[stops_df['is_nbus_stop']]
blind_stops = stops_df[~stops_df['is_nbus_stop']]

st.sidebar.header("⚙️ 시뮬레이션 설정")
selected_time = st.sidebar.select_slider("시간대 선택", options=[f"{str(h).zfill(2)}:{str(m).zfill(2)}" for h in [22, 23, 0, 1, 2, 3, 4] for m in [0, 30]] + ["05:00"], value="23:30")
time_idx = [f"{str(h).zfill(2)}:{str(m).zfill(2)}" for h in [22, 23, 0, 1, 2, 3, 4] for m in [0, 30]].index(selected_time) if selected_time != "05:00" else 14

st.sidebar.caption("수동으로 가중치를 조절하거나 시간대를 변경하여 노선을 시뮬레이션하세요.")
alpha = st.sidebar.slider("호출 수요(상업/유흥) 가중치 (α)", 0.0, 1.0, 0.8)
beta = st.sidebar.slider("정류장 결핍도 가중치 (β)", 0.0, 1.0, 0.2)
gamma = st.sidebar.slider("안전 취약도(어두운 골목) 가중치 (γ)", 0.0, 1.0, 0.4)

drt_budget = st.sidebar.number_input("투입 가능 DRT 차량 수", min_value=1, max_value=20, value=5)

current_grids = grids_df.copy()

# 시간대별 기반 수요/위험도 조절 (시간이 늦어질수록 유흥수요 감소, 주거지/안전 위험도 부각)
if time_idx <= 3:
    time_factor = 1.2
elif time_idx <= 7:
    time_factor = 0.8
else:
    time_factor = 0.3

current_grids['demand'] = current_grids['base_demand'] * time_factor
current_grids['deficit'] = current_grids['base_deficit']
current_grids['risk'] = current_grids['base_risk'] * (2.0 - time_factor)

current_grids['risk_score'] = alpha * current_grids['demand'] + beta * current_grids['deficit'] + gamma * current_grids['risk']

threshold = float(current_grids['risk_score'].quantile(0.85))
drt_targets = current_grids.nlargest(50, 'risk_score')

drt_assignments = []
for idx, grid_row in drt_targets.iterrows():
    if not blind_stops.empty:
        distances = np.sqrt((blind_stops['lat'] - grid_row['lat'])**2 + (blind_stops['lon'] - grid_row['lon'])**2)
        drt_assignments.append(blind_stops.loc[distances.idxmin()])

unique_blind_stops = pd.DataFrame(drt_assignments).drop_duplicates('stop_id')

loop_coords = []
transfer_coords = []
closest_hubs = pd.DataFrame()

def ccw(p1, p2, p3):
    return (p2[0] - p1[0]) * (p3[1] - p1[1]) - (p2[1] - p1[1]) * (p3[0] - p1[0])

def get_convex_hull(points):
    if len(points) <= 3: return points
    points = sorted(points, key=lambda p: (float(p[0]), float(p[1])))
    lower = []
    for p in points:
        while len(lower) >= 2 and ccw(lower[-2], lower[-1], p) <= 0: lower.pop()
        lower.append(p)
    upper = []
    for p in reversed(points):
        while len(upper) >= 2 and ccw(upper[-2], upper[-1], p) <= 0: upper.pop()
        upper.append(p)
    return lower[:-1] + upper[:-1]

if not unique_blind_stops.empty and not nbus_stops.empty:
    centroid_lat = float(unique_blind_stops['lat'].mean())
    centroid_lon = float(unique_blind_stops['lon'].mean())
    
    dist_to_hub = np.sqrt((nbus_stops['lat'] - centroid_lat)**2 + (nbus_stops['lon'] - centroid_lon)**2)
    closest_hubs = nbus_stops.loc[dist_to_hub.nsmallest(3).index]
    transfer_coords = sorted(closest_hubs[['lat', 'lon']].values.tolist(), key=lambda x: x[1])
    
    loop_stops = pd.concat([unique_blind_stops, closest_hubs]).drop_duplicates('stop_id')
    coords = loop_stops[['lat', 'lon']].values.tolist()
    hull_coords = get_convex_hull(coords)
    
    hull_lats = [float(pt[0]) for pt in hull_coords]
    hull_lons = [float(pt[1]) for pt in hull_coords]
    unique_blind_stops = unique_blind_stops[unique_blind_stops.apply(lambda r: any(abs(r['lat'] - hl) < 1e-6 and abs(r['lon'] - hlon) < 1e-6 for hl, hlon in zip(hull_lats, hull_lons)), axis=1)]
    
    hull_coords.append(hull_coords[0])
    loop_coords = [[float(pt[0]), float(pt[1])] for pt in hull_coords]

st.title("🚌 심야 대중교통 N버스-DRT 통합 대시보드")

col1, col2 = st.columns([2, 1])

with col1:
    m = folium.Map(location=[37.498, 127.027], zoom_start=15, tiles="CartoDB dark_matter")
    
    for idx, row in current_grids.iterrows():
        rs = float(row['risk_score'])
        if rs > 0.3:
            folium.CircleMarker(
                location=[float(row['lat']), float(row['lon'])], radius=4,
                color="red" if rs > threshold else "orange", weight=0, fill=True, 
                fill_color="red" if rs > threshold else "orange", fill_opacity=float(rs),
                popup=str(f"Risk: {rs:.2f}")
            ).add_to(m)
            
    for idx, row in blind_stops.iterrows():
        folium.CircleMarker(
            location=[float(row['lat']), float(row['lon'])], radius=2, color="gray", weight=0, fill=True, fill_color="gray", tooltip="일반 정류장"
        ).add_to(m)
        
    for idx, row in nbus_stops.iterrows():
        folium.CircleMarker(
            location=[float(row['lat']), float(row['lon'])], radius=3, color="cyan", weight=1, fill=True, fill_color="cyan",
            tooltip=str(f"환승: {row['name']}")
        ).add_to(m)
        
    colors = ['cyan', 'lime', 'yellow']
    for i, route in enumerate(real_bus_routes):
        if route['coords']:
            folium.PolyLine(
                locations=[[float(pt[0]), float(pt[1])] for pt in route['coords']], dash_array="15, 20",
                color=str(colors[i % len(colors)]), weight=4, opacity=0.6, tooltip=str(f"N버스: {route['name']}")
            ).add_to(m)

    if loop_coords:
        folium.PolyLine(
            locations=loop_coords, dash_array="10, 15",
            color='purple', weight=5, opacity=0.9, tooltip="DRT 루프"
        ).add_to(m)
        
        if len(transfer_coords) >= 2:
            folium.PolyLine(
                locations=[[float(pt[0]), float(pt[1])] for pt in transfer_coords], dash_array="1, 10",
                color='orange', weight=8, opacity=1.0, tooltip="환승 구역"
            ).add_to(m)
        
        for idx, row in unique_blind_stops.iterrows():
            folium.Marker(
                location=[float(row['lat']), float(row['lon'])], popup=str(f"승하차: {row['name']}"),
                icon=folium.Icon(color="purple", icon="bus", prefix="fa")
            ).add_to(m)
            
    if not closest_hubs.empty:
        for idx, row in closest_hubs.iterrows():
            folium.CircleMarker(
                location=[float(row['lat']), float(row['lon'])], radius=7, color="gold", weight=2, fill=True, fill_color="orange", fill_opacity=0.8,
                tooltip=str(f"환승: {row['name']}")
            ).add_to(m)

    # st_folium 호출 시 직렬화 에러를 방지하기 위해 returned_objects=[] 옵션 및 명시적 key 추가
    st_folium(m, width=800, height=500, returned_objects=[], key="main_map")

with col2:
    st.subheader("💡 다이내믹 루프 & 환승 시나리오")
    st.info(f"현재 선택된 시간: **{selected_time}**\n\n시간대에 따른 가중치 변화로 타겟 지역이 상업지구와 주거지구 사이를 자연스럽게 넘나듭니다.")
    
    if not closest_hubs.empty:
        hub_names = ", ".join(str(name) for name in closest_hubs['name'].tolist())
        st.warning(f"**🔥 환승 공유 구역 (Transfer Zone)**\n\nN버스 노선의 **'{hub_names}'** 정류장들을 DRT 루프가 그대로 따라 주행하며 겹칩니다. (지도 상 **오렌지색 선**)\n\n승객은 이 구역 내 아무 곳에서나 내려 N버스로 편하게 갈아탈 수 있습니다.")

st.divider()
st.subheader("📈 배차간격(대기 시간) 개선 효과")
col_b, col_a, col_diff = st.columns(3)

base_n_interval = 40
drt_interval = max(5, 30 - drt_budget * 4)

col_b.metric("사각지대 기존 배차간격", "운행 없음 (∞)", delta=None)
col_a.metric("DRT 도입 후 사각지대", f"약 {drt_interval} 분", delta="-∞ 분 (신규 서비스 창출)", delta_color="normal")
col_diff.metric("기존 N버스 배차간격 (간선)", f"약 {base_n_interval} 분", delta="유지", delta_color="off")

st.markdown("> **결과**: 지도 상 빛나는 오렌지색 선구간(환승 구역)에서 대기시간 없이 부드럽게 N버스로 갈아탈 수 있으며, 시간대에 맞춰 최적의 경로로 굽이치는 똑똑한 루프를 형성합니다.")