Python爬取腾讯疫情实时数据并存储到mysql数据库的示例代码
思路:
在腾讯疫情数据网站F12解析网站结构,使用Python爬取当日疫情数据和历史疫情数据,分别存储到details和history两个mysql表。
①此方法用于爬取每日详细疫情数据
import requests import json import time def get_details(): url = 'https://view.inews.qq.com/g2/getOnsInfo?name=disease_h5&callback=jQuery34102848205531413024_1584924641755&_=1584924641756' headers ={ 'user-agent': 'Mozilla/5.0 (Windows NT 10.0; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/70.0.3538.25 Safari/537.36 Core/1.70.3741.400 QQBrowser/10.5.3863.400' } res = requests.get(url,headers=headers) #输出全部信息 # print(res.text) response_data = json.loads(res.text.replace('jQuery34102848205531413024_1584924641755(','')[:-1]) #输出这个字典的键值 dict_keys(['ret', 'data'])ret是响应值,0代表请求成功,data里是我们需要的数据 # print(response_data.keys()) """上面已经转化过一次字典,然后获取里面的data,因为data是字符串,所以需要再次转化字典 print(json.loads(reponse_data['data']).keys()) 结果: dict_keys(['lastUpdateTime', 'chinaTotal', 'chinaAdd', 'isShowAdd', 'showAddSwitch', 'areaTree', 'chinaDayList', 'chinaDayAddList', 'dailyNewAddHistory', 'dailyHistory', 'wuhanDayList', 'articleList']) lastUpdateTime是最新更新时间,chinaTotal是全国疫情总数,chinaAdd是全国新增数据, isShowAdd代表是否展示新增数据,showAddSwitch是显示哪些数据,areaTree中有全国疫情数据 """ areaTree_data = json.loads(response_data['data'])['areaTree'] temp=json.loads(response_data['data']) # print(temp.keys()) # print(areaTree_data[0].keys()) """ 获取上一级字典里的areaTree 然后查看里面中国键值 print(areaTree_data[0].keys()) dict_keys(['name', 'today', 'total', 'children']) name代表国家名称,today代表今日数据,total代表总数,children里有全国各地数据,我们需要获取全国各地数据,查看children数据 print(areaTree_data[0]['children']) 这里面是 name是地区名称,today是今日数据,total是总数,children是市级数据, 我们通过这个接口可以获取每个地区的总数据。我们遍历这个列表,取出name,这个是省级的数据,还需要获取市级数据, 需要取出name,children(市级数据)下的name、total(历史总数)下的confirm、heal、dead,today(今日数据)下的confirm(增加数), 这些就是我们需要的数据 """ # print(areaTree_data[0]['children']) # for province_data in areaTree_data[0]['children']: # print(province_data) ds= temp['lastUpdateTime'] details=[] for pro_infos in areaTree_data[0]['children']: province_name = pro_infos['name'] # 省名 for city_infos in pro_infos['children']: city_name = city_infos['name'] # 市名 confirm = city_infos['total']['confirm']#历史总数 confirm_add = city_infos['today']['confirm']#今日增加数 heal = city_infos['total']['heal']#治愈 dead = city_infos['total']['dead']#死亡 # print(ds,province_name,city_name,confirm,confirm_add,heal,dead) details.append([ds,province_name,city_name,confirm,confirm_add,heal,dead]) return details
单独测试方法:
# d=get_details() # print(d)
②此方法用于爬取历史详细数据
import requests import json import time def get_history(): url = 'https://view.inews.qq.com/g2/getOnsInfo?name=disease_other&callback=jQuery341026745307075030955_1584946267054&_=1584946267055' headers ={ 'user-agent': 'Mozilla/5.0 (Windows NT 10.0; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/70.0.3538.25 Safari/537.36 Core/1.70.3741.400 QQBrowser/10.5.3863.400' } res = requests.get(url,headers=headers) # print(res.text) response_data = json.loads(res.text.replace('jQuery341026745307075030955_1584946267054(','')[:-1]) # print(response_data) data = json.loads(response_data['data']) # print(data.keys()) chinaDayList = data['chinaDayList']#历史记录 chinaDayAddList = data['chinaDayAddList']#历史新增记录 history = {} for i in chinaDayList: ds = '2021.' + i['date']#时间 tup = time.strptime(ds,'%Y.%m.%d') ds = time.strftime('%Y-%m-%d',tup)#改变时间格式,插入数据库 confirm = i['confirm'] suspect = i['suspect'] heal = i['heal'] dead = i['dead'] history[ds] = {'confirm':confirm,'suspect':suspect,'heal':heal,'dead':dead} for i in chinaDayAddList: ds = '2021.' + i['date']#时间 tup = time.strptime(ds,'%Y.%m.%d') ds = time.strftime('%Y-%m-%d',tup)#改变时间格式,插入数据库 confirm_add = i['confirm'] suspect_add = i['suspect'] heal_add = i['heal'] dead_add = i['dead'] history[ds].update({'confirm_add':confirm_add,'suspect_add':suspect_add,'heal_add':heal_add,'dead_add':dead_add}) return history
单独测试此方法:
# h=get_history() # print(h)
③此方法用于数据库的连接与关闭:
import time import pymysql import traceback def get_conn(): """ :return: 连接,游标 """ # 创建连接 conn = pymysql.connect(host="127.0.0.1", user="root", password="000429", db="mydb", charset="utf8") # 创建游标 cursor = conn.cursor() # 执行完毕返回的结果集默认以元组显示 return conn, cursor def close_conn(conn, cursor): if cursor: cursor.close() if conn: conn.close()
④此方法用于更新并插入每日详细数据到数据库表:
def update_details(): """ 更新 details 表 :return: """ cursor = None conn = None try: li = get_details() conn, cursor = get_conn() sql = "insert into details(update_time,province,city,confirm,confirm_add,heal,dead) values(%s,%s,%s,%s,%s,%s,%s)" sql_query = 'select %s=(select update_time from details order by id desc limit 1)' #对比当前最大时间戳 cursor.execute(sql_query,li[0][0]) if not cursor.fetchone()[0]: print(f"{time.asctime()}开始更新最新数据") for item in li: cursor.execute(sql, item) conn.commit() # 提交事务 update delete insert操作 print(f"{time.asctime()}更新最新数据完毕") else: print(f"{time.asctime()}已是最新数据!") except: traceback.print_exc() finally: close_conn(conn, cursor)
单独测试能否插入数据到details表:
update_details()
⑤此方法用于插入历史数据到history表
def insert_history(): """ 插入历史数据 :return: """ cursor = None conn = None try: dic = get_history() print(f"{time.asctime()}开始插入历史数据") conn, cursor = get_conn() sql = "insert into history values(%s,%s,%s,%s,%s,%s,%s,%s,%s)" for k, v in dic.items(): # item 格式 {'2021-01-13': {'confirm': 41, 'suspect': 0, 'heal': 0, 'dead': 1} cursor.execute(sql, [k, v.get("confirm"), v.get("confirm_add"), v.get("suspect"), v.get("suspect_add"), v.get("heal"), v.get("heal_add"), v.get("dead"), v.get("dead_add")]) conn.commit() # 提交事务 update delete insert操作 print(f"{time.asctime()}插入历史数据完毕") except: traceback.print_exc() finally: close_conn(conn, cursor)
单独测试能否插入数据到history表:
# insert_history()
⑥此方法用于根据时间来更新历史数据表的内容:
def update_history(): """ 更新历史数据 :return: """ cursor = None conn = None try: dic = get_history() print(f"{time.asctime()}开始更新历史数据") conn, cursor = get_conn() sql = "insert into history values(%s,%s,%s,%s,%s,%s,%s,%s,%s)" sql_query = "select confirm from history where ds=%s" for k, v in dic.items(): # item 格式 {'2020-01-13': {'confirm': 41, 'suspect': 0, 'heal': 0, 'dead': 1} if not cursor.execute(sql_query, k): cursor.execute(sql, [k, v.get("confirm"), v.get("confirm_add"), v.get("suspect"), v.get("suspect_add"), v.get("heal"), v.get("heal_add"), v.get("dead"), v.get("dead_add")]) conn.commit() # 提交事务 update delete insert操作 print(f"{time.asctime()}历史数据更新完毕") except: traceback.print_exc() finally: close_conn(conn, cursor)
单独测试更新历史数据表的方法:
# update_history()
最后是两个数据表的详细建立代码(也可以使用mysql可视化工具直接建立):
create table history( ds datetime not null comment '日期', confirm int(11) default null comment '累计确诊', confirm_add int(11) default null comment '当日新增确诊', suspect int(11) default null comment '剩余疑似', suspect_add int(11) default null comment '当日新增疑似', heal int(11) default null comment '累计治愈', heal_add int(11) default null comment '当日新增治愈', dead int(11) default null comment '累计死亡', dead_add int(11) default null comment '当日新增死亡', primary key(ds) using btree )engine=InnoDB DEFAULT charset=utf8mb4; create table details( id int(11) not null auto_increment, update_time datetime default null comment '数据最后更新时间', province varchar(50) default null comment '省', city varchar(50) default null comment '市', confirm int(11) default null comment '累计确诊', confirm_add int(11) default null comment '新增确诊', heal int(11) default null comment '累计治愈', dead int(11) default null comment '累计死亡', primary key(id) )engine=InnoDB default charset=utf8mb4;
Tomorrowthe birds will singing.
到此这篇关于Python爬取腾讯疫情实时数据并存储到mysql数据库的文章就介绍到这了,更多相关Python爬取数据存储到mysql数据库内容请搜索我们以前的文章或继续浏览下面的相关文章希望大家以后多多支持我们!
赞 (0)