가중치 반영하여 계산하도록 수정

lib/weekly_visitor_forecast_prophet.py

@@ -23,14 +23,22 @@ ga4 = db_schema.ga4_by_date
 weather = db_schema.weather
 air = db_schema.air
 
+# config 불러오기
 config = load_config()
 serviceKey = config['DATA_API']['serviceKey']
+weight_cfg = config.get('FORECAST_WEIGHT', {})
+
 VISITOR_CA = tuple(config['POS']['VISITOR_CA'])
 
-
+visitor_forecast_multiplier = weight_cfg.get('visitor_forecast_multiplier', 1.0)
+minTa_weight = weight_cfg.get('minTa', 1.0)
+maxTa_weight = weight_cfg.get('maxTa', 1.0)
+sumRn_weight = weight_cfg.get('sumRn', 1.0)
+avgRhm_weight = weight_cfg.get('avgRhm', 1.0)
+pm25_weight = weight_cfg.get('pm25', 1.0)
+is_holiday_weight = weight_cfg.get('is_holiday', 1.0)
 
 # --- 데이터 로딩 및 전처리 ---
-
 def get_date_range(start_date, end_date):
     return pd.date_range(start_date, end_date).to_pydatetime().tolist()
 
@@ -140,7 +148,19 @@ def prepare_prophet_df(df):
     return prophet_df
 
 def train_and_predict_prophet(prophet_df, forecast_days=7):
+    # 가중치 적용 - 훈련 데이터의 기상/환경 변수 컬럼별 곱하기
+    prophet_df = prophet_df.copy()
+    prophet_df['minTa'] *= minTa_weight
+    prophet_df['maxTa'] *= maxTa_weight
+    prophet_df['sumRn'] *= sumRn_weight
+    prophet_df['avgRhm'] *= avgRhm_weight
+    prophet_df['pm25'] *= pm25_weight
+    prophet_df['is_holiday'] *= is_holiday_weight
+
+    # 기존 fix_zero_visitors_weighted 함수 호출 (필요 시)
     prophet_df = fix_zero_visitors_weighted(prophet_df)
+
+    # 결측치 처리
     prophet_df.fillna({
         'minTa': 0,
         'maxTa': 0,
@@ -160,22 +180,22 @@ def train_and_predict_prophet(prophet_df, forecast_days=7):
 
     m.fit(prophet_df)
     future = m.make_future_dataframe(periods=forecast_days)
-    future_dates = future['ds'].dt.strftime('%Y%m%d').tolist()
 
-    weekly_precip = get_weekly_precip(serviceKey)  # {'YYYYMMDD': {'sumRn': x, 'minTa': y, 'maxTa': z, 'avgRhm': w}, ...}
+    # 미래 데이터에 날씨 예보 값 가져와서 가중치 적용
+    weekly_precip = get_weekly_precip(serviceKey)
 
-    # 미래 데이터에 강수량 및 기온/습도 반영
     sumRn_list = []
     minTa_list = []
     maxTa_list = []
     avgRhm_list = []
-    for dt_str in future_dates:
+    for dt in future['ds']:
+        dt_str = dt.strftime('%Y%m%d')
         day_forecast = weekly_precip.get(dt_str, None)
         if day_forecast:
-            sumRn_list.append(float(day_forecast.get('sumRn', 0)))
-            minTa_list.append(float(day_forecast.get('minTa', 0)))
-            maxTa_list.append(float(day_forecast.get('maxTa', 0)))
-            avgRhm_list.append(float(day_forecast.get('avgRhm', 0)))
+            sumRn_list.append(float(day_forecast.get('sumRn', 0)) * sumRn_weight)
+            minTa_list.append(float(day_forecast.get('minTa', 0)) * minTa_weight)
+            maxTa_list.append(float(day_forecast.get('maxTa', 0)) * maxTa_weight)
+            avgRhm_list.append(float(day_forecast.get('avgRhm', 0)) * avgRhm_weight)
         else:
             sumRn_list.append(0)
             minTa_list.append(0)
@@ -187,16 +207,21 @@ def train_and_predict_prophet(prophet_df, forecast_days=7):
     future['maxTa'] = maxTa_list
     future['avgRhm'] = avgRhm_list
 
-    # pm25는 과거 마지막 데이터 복사
+    # pm25는 과거 마지막 데이터 * 가중치 적용
     last_known = prophet_df.iloc[-1]
-    future['pm25'] = last_known['pm25']
+    future['pm25'] = last_known['pm25'] * pm25_weight
 
-    # is_holiday 계산
-    future['is_holiday'] = future['ds'].apply(lambda d: 1 if is_korean_holiday(d.date()) else 0)
+    # 휴일 여부도 가중치 곱해서 적용
+    future['is_holiday'] = future['ds'].apply(lambda d: 1 if is_korean_holiday(d.date()) else 0) * is_holiday_weight
 
     forecast = m.predict(future)
 
-    # 예측 결과 저장
+    # 최종 방문객 예측에 multiplier 곱하기
+    forecast['yhat'] = (forecast['yhat'] * visitor_forecast_multiplier).round().astype(int)
+    forecast['yhat_lower'] = (forecast['yhat_lower'] * visitor_forecast_multiplier).round().astype(int)
+    forecast['yhat_upper'] = (forecast['yhat_upper'] * visitor_forecast_multiplier).round().astype(int)
+
+    # csv 저장 및 반환
     output_path = os.path.abspath(os.path.join(os.path.dirname(__file__), '..', 'data', 'prophet_result.csv'))
     os.makedirs(os.path.dirname(output_path), exist_ok=True)
 
@@ -204,13 +229,9 @@ def train_and_predict_prophet(prophet_df, forecast_days=7):
     df_to_save.columns = ['date', 'visitor_forecast']
     df_to_save['date'] = df_to_save['date'].dt.strftime("%Y-%m-%d")
 
-    # 오늘 날짜 이후 데이터만 필터링
     today_str = date.today().strftime("%Y-%m-%d")
     df_to_save = df_to_save[df_to_save['date'] >= today_str]
 
-    # visitor_forecast를 정수로 변환
-    df_to_save['visitor_forecast'] = df_to_save['visitor_forecast'].round().astype(int)
-
     df_to_save.to_csv(output_path, index=False)
 
     return forecast