Lightgbm 回归预测

#!/usr/bin/env python
#-*- coding:utf-8 -*- 


import pandas as pd

def loadData():

    mete_data = pd.DataFrame(pd.read_excel(meteorological_data))
    power_data = pd.DataFrame(pd.read_csv(power_real_predict,encoding='gbk'))
    mete_data.columns=['文档时间', '时间', '风速', '风向', '气压', '温度', '湿度', '空气密度']
    mete_data = time2datetime(mete_data)
    power_data = time2datetime(power_data)
    data = datamerge(power_data,mete_data)

    return data[ori_mete],data[Daxian_pre],data[Real_pow]

def time2datetime(df):

    df['时间'] = pd.to_datetime(df['时间'])
    return df

def datamerge(df1,df2):

    return pd.merge(df1,df2,on='时间',how='inner')

#!/usr/bin/env python
#-*- coding:utf-8 -*- 
# @Describe: Preprocess Data

from sklearn.preprocessing import Imputer
from sklearn.preprocessing import StandardScaler

def Nanimputer(df):

    imputer = Imputer(missing_values='NaN', strategy='mean', axis=)
    imputer.fit(df.drop(['实际功率'],axis=))

    return imputer.transform(df.drop(['实际功率'],axis=))

def train_test_data(df,label,persent):

    train = df[:int(-len(df)*persent)]
    y = label['实际功率'].values[:int(-len(df)*persent)]

    test = df[int(-len(df)*persent):]
    y_test = label['实际功率'].values[int(-len(df)*persent):]
    train,test = standardScaler(train,test)
    return train,y,test,y_test

def standardScaler(train,test):

    scaler = StandardScaler()
    train = scaler.fit_transform(train)
    test = scaler.transform(test)
    return train,test

def pre_processing(df,label,persent):

    data = Nanimputer(df)

    return train_test_data(data,label,persent)

import pandas as pd
from config import items,items1,shift_time

def feature_extrac(df):

    df['风速_湿度'] = df['风速'] * df['湿度']**
    df['风速_风向'] = df['风速'] * df['风向']
    df['风速_空气密度'] = df['风速'] * df['空气密度']**
    df['风速_空气密度1'] = df['风速'] * df['空气密度']


    for item in items:
        for t in shift_time:
            df[item + str(t)] = df[item].shift(t)
    window =[,,]
    for item in items1:
        for wind in window:
            df[item + '_std'] = df[item].rolling(wind,min_periods=).std()
            df[item + '_mean'] = df[item].rolling(wind,min_periods=).mean()
            df[item + '_sum'] = df[item].rolling(wind,min_periods=).sum()
            df[item + '_max'] = df[item].rolling(wind,min_periods=).max()
            df[item + '_min'] = df[item].rolling(wind,min_periods=).min()
            df[item + '_median'] = df[item].rolling(wind,min_periods=).median()

    return df

#!/usr/bin/env python
#-*- coding:utf-8 -*- 
# @Describe: config

meteorological_data = 'XXX.xls'
power_real_predict = 'YYYYYY.CSV'
ori_mete = ['实际功率','风速', '风向', '气压', '温度', '湿度', '空气密度']
Daxian_pre = ['预测功率']
Real_pow = ['实际功率']
shift_time = [-1,-2,-3,-4,-5,-6,-7,-8,-9,-10,-11,-12,,,,,,,,,,,,]
items = [ '风向','气压', '温度', '湿度', '空气密度','风速_湿度','风速_风向','风速_空气密度','风速_空气密度1']
items1 = ['风向','湿度', '空气密度','风速_湿度','风速_风向','风速_空气密度','风速_空气密度1']

params = {
    "objective": "regression",
    "metric": "mse",
    "num_leaves": ,
    "min_child_samples": ,
    "learning_rate": 0.01,
    "bagging_fraction": 0.8,
    "feature_fraction": 0.8,
    "bagging_frequency": ,
    "bagging_seed": ,
    "verbosity": -1
}

#!/usr/bin/env python
#-*- coding:utf-8 -*- 
# @Describe: Metrics
from math import sqrt
from sklearn.metrics import mean_squared_error
import pandas as pd
import numpy as np

class Metrics():
    def __init__(self, wepre, otpre, real):
        self.wepre = wepre
        self.otpre = otpre
        self.real = real

    def RMSE(self):
        rmse = sqrt(mean_squared_error(self.real,self.wepre))
        rmse_other = sqrt(mean_squared_error(self.real, self.otpre))
        print('Test RMSE: %.3f' % rmse)
        print('某同行公司Test RMSE: %.3f' % rmse_other)
    def MulHourEval(self):
        for metric_t in [,,]:
            zhibiao_list = []
            zhibiao_list_daxian = []
            for i in range(,len(self.real)):
                rmse = sqrt(mean_squared_error(self.real[i:i+metric_t-1], self.wepre[i:i+metric_t-1]))#
                rmse_other = sqrt(mean_squared_error(self.real[i:i+metric_t-1], self.otpre[i:i+metric_t-1]))
                zhibiao =  - rmse/
                zhibiao_other =  - rmse_other/
                zhibiao_list.append(zhibiao)
                zhibiao_list_other.append(zhibiao_other)
            zhibiao_list_pd = np.array(zhibiao_list)
            zhibiao_list_other_pd = np.array(zhibiao_list_other)
            print('Our model %s hour eval:%s'%(metric_t/,np.mean(zhibiao_list_pd)))
            print('Other model %s hour eval:%s'%(metric_t/,np.mean(zhibiao_list_other_pd)))

#定义模型
def lgbmodel(data_train,label_train,data_test):

    X_train, X_test, y_train, y_test = train_test_split(data_train,label_train,test_size=/,shuffle=True, random_state=)
    lgb_train = lgb.Dataset(X_train, label=y_train)
    lgb_eval = lgb.Dataset(X_test, label = y_test)
    lgbm = lgb.train(params,
                lgb_train,
                num_boost_round=,
                valid_sets=lgb_eval,
                early_stopping_rounds=,
                verbose_eval= )
    predict = lgbm.predict(data_test)

    return predict

 
import lightgbm as lgb
from sklearn.model_selection import train_test_split
from config import params # 从自创的congfig模块中 加载设定好的参数
from dataload import loadData #从自创的dataload 模块中加载 loadData以生产数据
from genFeature import feature_extrac  #从自创的genFeature加载 fetature_extrac以用于特征提取
from Mulmetics import Metrics   #从自创的多评估模块加载业内的评估方法
from DataPrepro import pre_processing #从自创的预处理模块加载数据预处理
 
def main():
 
    Percent = /  #设定训练集与验证集的比例
    data,pre,real = loadData()#加载数据
    data = feature_extrac(data)# 提取特征
 
    train,y,test,label_test = pre_processing(data,real,Percent) #分割数据
    print(train.shape,y.shape,test.shape,label_test.shape) #打印数据形状
    model_predict = lgbmodel(train,y,test) #训练
    metrics = Metrics(model_predict,pre['预测功率'].values[int(-len(data)*Percent):],label_test) #评估
    metrics.RMSE()
    metrics.MulHourEval()


if __name__ == '__main__':
    main()

(, ) (,) (, ) (,)
Training until validation scores don't improve for 200 rounds.
[100]    valid_0's l2: 77.9721
[]    valid_0's l2: 56.5769
[300]    valid_0's l2: 49.1198
[]    valid_0's l2: 45.1405
[500]    valid_0's l2: 42.4402
[]    valid_0's l2: 40.2658
[700]    valid_0's l2: 38.4283
[]    valid_0's l2: 36.812
[900]    valid_0's l2: 35.4044
[]    valid_0's l2: 34.1903
[1100]    valid_0's l2: 33.0945
[]    valid_0's l2: 32.1998
[1300]    valid_0's l2: 31.3365
[]    valid_0's l2: 30.5228
[1500]    valid_0's l2: 29.7129
Did not meet early stopping. Best iteration is:
[]    valid_0's l2: 29.7129
Test RMSE: 10.263
某同行公司Test RMSE: 11.768
Our model 4.0 hour eval:0.8319254785233036
Other model 4.0 hour eval:0.8141609206646702
Our model 24.0 hour eval:0.810845265380047
Other model 24.0 hour eval:0.7942520524266607
Our model 72.0 hour eval:0.8029606299934267
Other model 72.0 hour eval:0.7796889844798093