import pandas as pd
import xgboost as xgb
from sklearn.model_selection import train_test_split
from sklearn.metrics import r2_score
import numpy as np

df = pd.read_excel(r'./400v入模数据.xlsx')
df['stat_date'] = pd.to_datetime(df['stat_date'])

print(df.corr()['0.4kv及以下'])

X = df[(df['stat_date']>='2021-01-01')&(df['stat_date']<='2023-09-28')].drop(columns=['0.4kv及以下']).set_index('stat_date')

y = df[(df['stat_date']>='2021-01-01')&(df['stat_date']<='2023-09-28')]['0.4kv及以下']
x_eval = df[(df['stat_date']<='2023-09-30')&(df['stat_date']>='2023-09-01')].drop(columns=['0.4kv及以下']).set_index('stat_date')
print(x_eval)
y_eval = df[(df['stat_date']<='2023-09-30')&(df['stat_date']>='2023-09-01')][['0.4kv及以下','city']]


x_train,x_test,y_train,y_test = train_test_split(X,y,test_size=0.2,random_state=42)
model = xgb.XGBRegressor(max_depth=6,learning_rate=0.05,n_estimators=250)
model.fit(x_train,y_train)
y_pred = model.predict(x_test)
print(r2_score(y_test,y_pred))

predict = model.predict(x_eval)
result = pd.DataFrame({'real':y_eval.drop(columns='city').values.reshape(-1),'pred':predict},index=x_eval.index)
print(result.loc['2023-09-28':'2023-09-30'])


dict2 = {'杭州':0,'湖州':1,'嘉兴':2,'金华':3,'丽水':4,'宁波':5,'衢州':6,'绍兴':7,'台州':8,'温州':9,'舟山':10}
dict1 = {}
for city in x_eval['city'].drop_duplicates():
    eval_x = x_eval[x_eval['city']==city]
    eval_y = y_eval[y_eval['city']==city]['0.4kv及以下']
    pred = model.predict(eval_x)
    loss_rate = (np.sum(pred[-3:])-np.sum(eval_y[-3:]))/np.sum(eval_y)
    dict1[city] = loss_rate


for key in dict2.keys():
    dict2[key] = dict1[dict2[key]]
print(dict2)