気象データの実績値と予報値の比較をしてみました

import pandas as pd
import datetime
from sklearn.metrics import r2_score

#実績値を読み込む
jisseki =  pd.read_csv('jisseki-data/data_202112.csv')
jisseki['timestamp'] = pd.to_datetime(jisseki['timestamp'])
#実績値の前1時間を後1時間に整形
jisseki['timestamp'] = jisseki['timestamp'] + datetime.timedelta(hours=-1)

#予報値を読み込む
forecast_ = pd.read_csv('jisseki-data/WEATHER_FORCAST_RESULT_202112.csv')
#項目を和名から英名に変更する
forecast_ = forecast_.rename(columns={'日時': 'timestamp', '日射量（W/m^2）': 'isolation', '場所名': 'station'})
forecast_ = forecast_[['station','timestamp' ,'isolation']]

#予報値の整形、地点ごとにカラムを作る
forecast_tokyo = forecast_[forecast_['station'] == 'tokyo'].reset_index(drop=True)
forecast_tokyo = forecast_tokyo.rename(columns={'isolation': 'forecast_tokyo'})
forecast_tokyo = forecast_tokyo[['timestamp' ,'forecast_tokyo']]
                                                
forecast_nigata = forecast_[forecast_['station'] == 'nigata'].reset_index(drop=True)
forecast_nigata = forecast_nigata.rename(columns={'isolation': 'forecast_nigata'})
forecast_nigata = forecast_nigata[['timestamp' ,'forecast_nigata']]                                                
                                                
forecast_kagoshima = forecast_[forecast_['station'] == 'kagoshima'].reset_index(drop=True)
forecast_kagoshima = forecast_kagoshima.rename(columns={'isolation': 'forecast_kagoshima'})
forecast_kagoshima = forecast_kagoshima[['timestamp' ,'forecast_kagoshima']]                                                    
                                                                                                   
forecast = pd.merge(forecast_tokyo, forecast_nigata, on='timestamp', how='inner')
forecast = pd.merge(forecast, forecast_kagoshima, on='timestamp', how='inner')

jisseki = jisseki.rename(columns={'nigata-isolation': 'jisseki_nigata', 'tokyo-isolation': 'jisseki_tokyo', 'kagoshima-isolation': 'jisseki_kagoshima'})
#実績値の単位をW/m2に変更
jisseki['jisseki_nigata'] = jisseki['jisseki_nigata']*1000000/3600
jisseki['jisseki_tokyo'] = jisseki['jisseki_tokyo']*1000000/3600
jisseki['jisseki_kagoshima'] = jisseki['jisseki_kagoshima']*1000000/3600

jisseki['timestamp'] = jisseki['timestamp'].dt.strftime('%Y-%m-%d %H:%M:%S')

#timestampをキーとして、予報と実績値をマージする
total = pd.merge(forecast, jisseki, on='timestamp', how='inner')
total = total[['timestamp', 'forecast_nigata', 'jisseki_nigata', 'forecast_tokyo', 'jisseki_tokyo', 'forecast_kagoshima' , 'jisseki_kagoshima']]

total = total.dropna().reset_index(drop=True)

#地点新潟の決定係数を算出
forecast_nigata = total['forecast_nigata'].to_list()
jisseki_nigata = total['jisseki_nigata'].to_list()

r2 = r2_score(forecast_nigata, jisseki_nigata)
print(r2)

#地点東京の決定係数を算出
forecast_tokyo = total['forecast_tokyo'].to_list()
jisseki_tokyo = total['jisseki_tokyo'].to_list()

r2 = r2_score(forecast_tokyo, jisseki_tokyo)
print(r2)

#地点鹿児島の決定係数を算出
forecast_kagoshima = total['forecast_kagoshima'].to_list()
jisseki_kagoshima = total['jisseki_kagoshima'].to_list()

r2 = r2_score(forecast_kagoshima, jisseki_kagoshima)
print(r2)