JOC / data process /generate_RF.py
BTBUzhy's picture
Upload generate_RF.py
63c4f57 verified
Raw History Blame Contribute Delete
8.43 kB
import os
import pandas as pd
from datetime import datetime
import random
import numpy as np
from sklearn.model_selection import cross_val_score
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestRegressor
from sklearn.model_selection import GridSearchCV
from sklearn.cluster import KMeans
from sklearn.cluster import AgglomerativeClustering
import matplotlib.pyplot as plt
import joblib
from gurobipy import *
import warnings
warnings.filterwarnings('ignore')
# 系统参数设定
station_num=5
random.seed(10)
soc=[0,1,2]
trip_data = pd.read_csv(r"ebike_data/trip.csv")
trip_data = trip_data.drop_duplicates(keep='last', inplace=False, ignore_index=True)
station_data = pd.read_csv(r"ebike_data/station.csv")
weather_data = pd.read_csv(r"ebike_data/weather.csv")
weather_data = weather_data.loc[:,
['Date', 'Mean_Temperature_F', 'Mean_Humidity', 'Mean_Visibility_Miles', 'Mean_Wind_Speed_MPH',
'Precipitation_In']]
weather_data['Date'] = pd.to_datetime(weather_data["Date"]).dt.strftime("%Y-%m-%d")
# out_label = trip_data['from_station_name']
# in_label = trip_data['to_station_name']
# label = pd.concat([out_label, in_label])
# label_count = label.value_counts(sort=True)
def clustering(n_clusters,station_data,plot):
data=station_data.copy()
data=data.drop(labels=["name"],axis=1)
data=data.drop(labels=["install_date"], axis=1)
data=data.drop(labels=["install_dockcount"],axis=1)
data=data.drop(labels=["modification_date"],axis=1)
data=data.drop(labels=["decommission_date"],axis=1)
# data_2=data[(data['lat']<47.63) & (data['long']<-122.31)]
# data_1=data[(data['lat']>=47.63) | (data['long']>=-122.31)]
# i=0
# processed_data=[]
# for data_i in [data_1,data_2]:
coords = data[['long', 'lat']].values
# 使用层次聚类方法进行初始聚类,为KMeans算法确定聚类中心
agglo = AgglomerativeClustering(n_clusters=n_clusters)
initial_clusters = agglo.fit_predict(coords)
# 计算每个簇的中心点
initial_cluster_centers = np.array([coords[initial_clusters == i].mean(axis=0) for i in range(n_clusters)])
# 使用计算得到的中心点作为KMeans算法的初始中心
kmeans = KMeans(n_clusters=n_clusters, init=initial_cluster_centers, n_init=1)
data['cluster'] = kmeans.fit_predict(coords)
# processed_data.append(data_i)
# i=i+1
# data=pd.concat(processed_data)
data.reset_index(drop=True, inplace=True)
cluster_dict = data.set_index('station_id')['cluster'].to_dict()
unique_clusters = np.unique(data['cluster'])
if plot:
plt.figure(figsize=(10, 8))
for cluster in unique_clusters:
subset = data[data['cluster'] == cluster]
plt.scatter(subset['long'], subset['lat'], label=f"Cluster {cluster}", s=100)
# 标记聚类中心
plt.scatter(kmeans.cluster_centers_[:, 0], kmeans.cluster_centers_[:, 1], s=200, c='red', label='Centers', marker='*')
plt.xlabel('Longitude')
plt.ylabel('Latitude')
# plt.title('KMeans 聚类结果')
# plt.legend(title='Cluster')
plt.grid(True)
plt.savefig('cluster_5.png')
plt.show()
# 可视化
dock_dict={}
for s in unique_clusters:
dock=data.loc[(data['cluster']==s)]['current_dockcount'].sum()
dock_dict[s]=dock
long_dict={i:kmeans.cluster_centers_[i,0] for i in range(n_clusters) }
lat_dict={i:kmeans.cluster_centers_[i,1] for i in range(n_clusters)}
return cluster_dict,dock_dict,long_dict,lat_dict
# station_list = label_count.keys()[:station_num].to_list()
# station_dict = dict(zip(station_list, list(range(station_num))))
# trip_data = trip_data[
# (trip_data['from_station_name'].isin(station_list)) & (trip_data['to_station_name'].isin(station_list))]
n_clusters=station_num
cluster_dict,dock_dict,long_dict,lat_dict=clustering(n_clusters,station_data,plot=True)
travel_data = trip_data.loc[:, ['starttime', 'stoptime', 'from_station_id', 'to_station_id']]
travel_data['from_cluster'] = travel_data['from_station_id'].map(cluster_dict)
travel_data['to_cluster'] = travel_data['to_station_id'].map(cluster_dict)
# travel_data['day_of_week'] = trip_data["starttime"].apply(lambda x: datetime.strptime(x, "%m/%d/%Y %H:%M").isoweekday())
travel_data['start_hour'] = trip_data["starttime"].apply(lambda x: datetime.strptime(x, "%m/%d/%Y %H:%M").hour)
travel_data['end_hour'] = trip_data["stoptime"].apply(lambda x: datetime.strptime(x, "%m/%d/%Y %H:%M").hour)
travel_data['Date'] = pd.to_datetime(trip_data["starttime"]).dt.strftime("%Y-%m-%d")
travel_data.to_csv(r"ebike_data/travel04.csv",index=False)
rent_counts = pd.DataFrame(travel_data.groupby(["Date", 'start_hour', 'from_cluster'], as_index=False)["from_station_id"].count())
rent_counts = rent_counts.rename(columns={"from_station_id": "count","from_cluster":"cluster"})
rent_counts['cluster']=rent_counts['cluster'].astype(int)
return_counts = pd.DataFrame(travel_data.groupby(["Date", "end_hour", "to_cluster"], as_index=False)["to_station_id"].count())
return_counts = return_counts.rename(columns={"to_station_id": "count","to_cluster":"cluster","end_hour":"start_hour"})
return_counts['cluster']=return_counts['cluster'].astype(int)
data_1=rent_counts.loc[:,["Date", "start_hour"]]
data_2=return_counts.loc[:,["Date", "start_hour"]]
data=pd.concat([data_1,data_2])
# data['Date']=pd.to_datetime(data['Date'])
data = data.drop_duplicates(keep='first', inplace=False, ignore_index=True)
data = data.sort_values(axis=0, by=['Date','start_hour'], ascending=True)
data.reset_index(drop=True, inplace=True)
data['day_of_week'] = data["Date"].apply(lambda x: datetime.strptime(x, "%Y-%m-%d").isoweekday())
data = pd.merge(data, weather_data, on="Date")
for i in range(station_num):
data['{}_rent'.format(i)]=0
data['{}_return'.format(i)]=0
for e in soc:
data['{}_dispatch_{}'.format(i,e)]=0
if e!=2:
data['{}_exchange_{}'.format(i,e)]=0
data['{}_I_{}'.format(i,e)]=-1
from simulation_f import *
data=fill_demand_data(data,rent_counts,'rent')
data=fill_demand_data(data,return_counts,'return')
data=data[(data['start_hour']>=8) &(data['start_hour']<=20)]
data.reset_index(drop=True, inplace=True)
length=len(data)
for i in range(station_num):
rent_array=np.random.randint(0,10,size=length)
return_array=np.random.randint(0,10,size=length)
data['{}_rent'.format(i)]=data['{}_rent'.format(i)]*10+rent_array.tolist()
data['{}_return'.format(i)]=data['{}_return'.format(i)]*10+return_array.tolist()
data.to_csv(r"ebike_data/origin_data_4.csv", index=False)
# 运行simulation_f.py文件,生成训练数据 #如果参数改了,要改文件名
regressor = RandomForestRegressor(n_estimators=30,max_depth=60)
def dataset_generate(dataset,test_size=0.2,random_state=0):
data=dataset.copy()
data=data.drop(labels=['Date','penalty_0','penalty_1'],axis=1)
for s in range(station_num):
data=data.drop(['{}_rent'.format(s),'{}_return'.format(s)], axis=1)
# data=data.drop(['{}_pick'.format(s),'{}_put'.format(s),'{}_I'.format(s)], axis=1)
# data=data.drop(['{}_dispatch'.format(s),'{}_I'.format(s)], axis=1)
data.to_csv(r"ebike_data/train_data_4.csv",index=False)
X = np.asarray(data.iloc[:, :-1])
y = np.asarray(data.iloc[:, -1])
x_train, x_test, y_train, y_test = train_test_split(X, y, test_size=test_size, random_state=random_state)
return x_train, x_test, y_train, y_test
dataset=pd.read_csv(r"ebike_data/data04.csv") #跟simulation_f里的文件名对应
x_train, x_test, y_train, y_test=dataset_generate(dataset)
regressor.fit(x_train, y_train)
y_pred=regressor.predict(x_test)
print("Training Score:%f" % regressor.score(x_train, y_train))
print("Testing Score:%f" % regressor.score(x_test, y_test))
# K折检验
param_grid = {"n_estimators":range(5,11,1)} #,"max_depth":[8,9,10]
grid_search = GridSearchCV(RandomForestRegressor(),param_grid,cv = 3)
grid_search.fit(x_train, y_train)
# y_pred=grid_search.predict(x_test)
# 拟合优度R2
print("r2:", grid_search.score(x_test, y_test))
# 用Scikit_learn计算MSE
grid_search.set_params()
print(grid_search.best_params_)
path = r"ebike_data/regressor.model"
joblib.dump(regressor, path)
import gurobipy as gp
model = gp.read('ebike_data/model04.mps')
model.write('ebike_data/model04.lp')