# -*- coding: utf-8 -*- """ 第 18 题 · 基站退服数据生成器(Data-Generated-Core-Py/data.py 的改进版) ================================================================ 字段口径与 data.py 完全一致: data_outage.csv 退服记录 record_id, ne_id, cell_id, district, vendor, start_time, end_time, duration_min, cause_category, cause_detail, cause_code, ticket_id, closed, handle_time, close_time(后两列为本版新增:接单时间、闭环时间) data_alarm.csv 告警流 alarm_id, ne_id, alarm_code, alarm_name, alarm_level, category, start_time, clear_time, duration_s, cleared, parent_alarm_id sites.csv 站点库 ne_id, site_name, district, scene, site_type, vendor, cells, lng, lat 相对 data.py 的改进(对齐 _数据集总索引.md 第 18 题要求:站号、起止时间、原因码、厂家、片区): 1. 新增站点库:南宁 7 个城区 × 5 类通信场景 × 3 类站型,站号固定,退服可按站点汇聚; 2. 新增原因码 cause_code,编码体系参照《中兴 5G NR 基站日常故障处理指导手册》章节; 3. 退服与告警联动生成:每次退服产生“小区退服”告警及对应根因告警(含衍生关系 parent_alarm_id), 另混入背景告警与时间窗内的干扰告警,用于检验根因关联算法; 4. 约 1.5% 的上报时长存在人工计算错误,用于检验系统自动计算与核对; 5. 分布贴近现网:长尾时长(沿用 data.py 分段)、问题站点重复退服、夏季雷雨停电高发、割接集中在凌晨。 数据为合成数据,论文中须注明。仅依赖标准库,可单独运行:python data_gen.py --rows 6000 注意:固定随机种子保证每次生成的数据完全相同。修改代码时不要调整随机数的调用顺序,否则数据会变化。 """ import argparse import csv import datetime import os import random DISTRICTS = { # 片区 -> 中心经纬度 '青秀区': (108.385, 22.816), '兴宁区': (108.368, 22.853), '江南区': (108.273, 22.781), '西乡塘区': (108.306, 22.836), '良庆区': (108.393, 22.752), '邕宁区': (108.487, 22.758), '武鸣区': (108.275, 23.159), } VENDORS = ['厂家A', '厂家B', '厂家C'] # 场景 -> [(片区, 地标)] 地标名用于生成站名,均为南宁城区真实地名 LANDMARKS = { '商圈': [('青秀区', '万象城'), ('青秀区', '航洋国际'), ('青秀区', '东盟商务区'), ('兴宁区', '朝阳广场'), ('兴宁区', '三街两巷'), ('江南区', '沙井万达'), ('西乡塘区', '西大商圈'), ('良庆区', '五象万达')], '高校': [('西乡塘区', '广西大学'), ('西乡塘区', '广西民族大学'), ('青秀区', '广西医科大学'), ('西乡塘区', '南宁师范大学'), ('邕宁区', '南宁学院'), ('武鸣区', '武鸣大学城')], '交通枢纽': [('青秀区', '南宁东站'), ('西乡塘区', '南宁火车站'), ('江南区', '吴圩国际机场'), ('江南区', '江南客运站'), ('青秀区', '埌东客运站'), ('西乡塘区', '安吉客运站')], '城中村': [('江南区', '沙井村'), ('西乡塘区', '秀厢村'), ('西乡塘区', '石埠'), ('兴宁区', '三塘'), ('良庆区', '大沙田'), ('邕宁区', '蒲庙'), ('武鸣区', '城厢'), ('青秀区', '长塘')], '工业园区': [('西乡塘区', '高新区'), ('江南区', '经开区'), ('武鸣区', '东盟经开区'), ('邕宁区', '新兴产业园'), ('良庆区', '五象产业园')], } # 场景 -> (每个地标站点数, 站型权重[宏站, 室分, 分布式皮基站], 退服倾向系数) SCENE_CFG = { '商圈': (6, [3, 4, 3], 0.9), '高校': (6, [5, 3, 2], 0.8), '交通枢纽': (6, [4, 4, 2], 0.7), '城中村': (5, [9, 1, 0], 1.6), '工业园区': (5, [8, 2, 0], 1.2), } TYPE_CODE = {'宏站': '宏站', '室分': '室分', '分布式皮基站': '皮站'} # 站型 -> 站名中的简写 # 原因大类 / 细项 / 原因码(参照中兴手册章节)—— 大类与细项沿用 data.py CAUSES = { '传输中断': [('光缆被挖断', 'TR01'), ('光模块损坏', 'TR02'), ('传输误码超限', 'TR03'), ('上游设备掉电', 'TR04')], '电源故障': [('市电停电', 'PW01'), ('蓄电池失效', 'PW02'), ('开关电源模块故障', 'PW03')], '设备硬件': [('BBU 主控板故障', 'HW01'), ('RRU 功放故障', 'HW02'), ('GPS 失锁', 'HW03'), ('单板复位', 'HW04')], '天馈故障': [('驻波比超标', 'AN01'), ('馈线进水', 'AN02'), ('天线被破坏', 'AN03')], '机房环境': [('空调故障', 'EN01'), ('机房高温', 'EN02'), ('漏水', 'EN03')], '割接施工': [('计划割接', 'OP01'), ('施工误碰光缆', 'OP02'), ('设备搬迁', 'OP03')], '其他': [('原因待查', 'OT01'), ('业主断电', 'OT02')], } # 各场景原因大类权重,顺序同 CAUSES(城中村停电多、园区施工多、室分为主场景机房/业主问题多) CAUSE_W = { '商圈': [22, 20, 18, 6, 12, 10, 12], '高校': [24, 24, 20, 8, 8, 10, 6], '交通枢纽': [26, 18, 24, 8, 8, 12, 4], '城中村': [22, 38, 14, 10, 6, 4, 6], '工业园区': [30, 20, 14, 8, 6, 18, 4], } # 告警库:沿用 data.py ALARM_LIB,补充 ALM-113~116 以覆盖割接、主控板、开关电源、漏水 ALARM_LIB = { '小区退服': ('ALM-101', '小区退服', '严重'), '传输中断': [('ALM-102', 'BBU 链路断', '严重'), ('ALM-103', '光模块收发异常', '重要'), ('ALM-110', '传输误码超限', '次要')], '电源故障': [('ALM-105', '市电停电', '严重'), ('ALM-106', '蓄电池欠压', '重要'), ('ALM-115', '开关电源模块故障', '重要')], '设备硬件': [('ALM-114', 'BBU 主控板故障', '严重'), ('ALM-111', 'PRRU 离线', '重要'), ('ALM-108', 'GPS 失锁', '次要'), ('ALM-112', '时钟不同步', '次要')], '天馈故障': [('ALM-104', 'RRU 驻波比过高', '重要'), ('ALM-109', 'VSWR 告警', '重要')], '机房环境': [('ALM-107', '机房温度过高', '重要'), ('ALM-116', '机房水浸告警', '重要')], '割接施工': [('ALM-113', '网元人工闭塞/复位', '次要')], } # 根因告警相对退服开始的提前量(分钟):停电有蓄电池后备、高温有过热保护延时;未列出的类别为 (0, 8) LEAD = {'电源故障': (20, 240), '机房环境': (15, 150), '割接施工': (0, 20), '天馈故障': (0, 25)} T0 = datetime.datetime(2026, 1, 1) DAYS = 273 # 2026-01-01 ~ 2026-09-30 MONTH_W = [0.8, 0.75, 0.9, 1.0, 1.15, 1.35, 1.45, 1.4, 1.1] # 各月退服权重:夏季雷雨、高温停电高发 FMT = '%Y-%m-%d %H:%M:%S' OUTAGE_HEAD = ['record_id', 'ne_id', 'cell_id', 'district', 'vendor', 'start_time', 'end_time', 'duration_min', 'cause_category', 'cause_detail', 'cause_code', 'ticket_id', 'closed', 'handle_time', 'close_time'] ALARM_HEAD = ['alarm_id', 'ne_id', 'alarm_code', 'alarm_name', 'alarm_level', 'category', 'start_time', 'clear_time', 'duration_s', 'cleared', 'parent_alarm_id'] SITE_HEAD = ['ne_id', 'site_name', 'district', 'scene', 'site_type', 'vendor', 'cells', 'lng', 'lat'] def gen_sites(rng): """生成站点库:每个地标周围若干站点,同一地标同一厂家组网。""" sites = [] used_ids = set() for scene, marks in LANDMARKS.items(): per_mark, type_weights, _ = SCENE_CFG[scene] for district, mark in marks: center_lng, center_lat = DISTRICTS[district] # 地标位置在片区中心附近随机偏移 mark_lng = center_lng + rng.uniform(-.03, .03) mark_lat = center_lat + rng.uniform(-.025, .025) vendor = rng.choice(VENDORS) for k in range(per_mark): site_type = rng.choices(list(TYPE_CODE), type_weights)[0] number = rng.randint(10000, 99999) while number in used_ids: number = rng.randint(10000, 99999) used_ids.add(number) if site_type == '宏站': cells = 3 else: cells = rng.choice([1, 2, 4, 6]) lng = round(mark_lng + rng.uniform(-.006, .006), 6) lat = round(mark_lat + rng.uniform(-.005, .005), 6) sites.append({ 'ne_id': 'gNodeB-%05d' % number, 'site_name': '%s%s-%s%02d' % (district[:-1], mark, TYPE_CODE[site_type], k + 1), 'district': district, 'scene': scene, 'site_type': site_type, 'vendor': vendor, 'cells': cells, 'lng': lng, 'lat': lat, }) return sites def gen_duration(rng, category): """退服时长(分钟),沿用 data.py 的长尾分段:多数短退服,少数超长。割接受窗口限制偏短,停电、机房类偏长。""" factor = {'割接施工': .88, '电源故障': 1.08, '机房环境': 1.05}.get(category, 1) r = rng.random() * factor if r < 0.62: return rng.randint(5, 60) if r < 0.88: return rng.randint(61, 360) if r < 0.97: return rng.randint(361, 1440) return rng.randint(1441, 7200) def gen_start(rng, category): """退服开始时间:按月份权重抽日期;割接集中在凌晨,停电/机房一半发生在午后。""" while True: day = rng.randrange(DAYS) month = (T0 + datetime.timedelta(days=day)).month if rng.random() < MONTH_W[month - 1] / max(MONTH_W): break if category == '割接施工': minute = rng.randint(0, 6 * 60) # 夜间维护窗口 00:00-06:00 elif category in ('电源故障', '机房环境') and rng.random() < .5: minute = rng.randint(12 * 60, 20 * 60) # 午后雷雨、高温 else: minute = rng.randint(0, 24 * 60 - 1) return T0 + datetime.timedelta(days=day, minutes=minute) def add_alarm(rng, alarms, ne_id, category, start, end, parent=''): """生成一条告警追加到 alarms 列表,返回告警编号。end 为 None 表示未清除。""" if category == '小区退服': code, name, level = ALARM_LIB['小区退服'] else: code, name, level = rng.choice(ALARM_LIB[category]) alarm_id = 'ALM%08d' % (len(alarms) + 1) if end: clear_time = end.strftime(FMT) duration = int((end - start).total_seconds()) cleared = '是' else: clear_time = '' duration = 0 cleared = '否' alarms.append({'alarm_id': alarm_id, 'ne_id': ne_id, 'alarm_code': code, 'alarm_name': name, 'alarm_level': level, 'category': category, 'start_time': start.strftime(FMT), 'clear_time': clear_time, 'duration_s': duration, 'cleared': cleared, 'parent_alarm_id': parent}) return alarm_id def generate(n=6000, seed=2026, out='.'): rng = random.Random(seed + 18) sites = gen_sites(rng) # 站点退服倾向 = 场景系数 × 站点系数;约 12% 的站点是问题站点,系数明显更高(会被判为高频重复站点) weights = [] for site in sites: if rng.random() < .12: site_factor = rng.uniform(4, 8) else: site_factor = rng.uniform(.5, 1.5) weights.append(SCENE_CFG[site['scene']][2] * site_factor) categories = list(CAUSES) outages = [] alarms = [] for i in range(n): site = sites[rng.choices(range(len(sites)), weights)[0]] ne_id = site['ne_id'] category = rng.choices(categories, CAUSE_W[site['scene']])[0] detail, code = rng.choice(CAUSES[category]) start = gen_start(rng, category) duration = gen_duration(rng, category) end = start + datetime.timedelta(minutes=duration) # 约 1.5% 的流水为人工计算错误:跨天时漏算整天,或按整点取整后再差半小时 reported = duration if rng.random() < .015: if duration > 1440: reported = duration % 1440 else: reported = max(5, round(duration / 60) * 60 + rng.choice([-30, 30])) cell_id = '460-11-%s-%02d' % (ne_id[-5:], rng.randint(1, site['cells'])) # 最近 10 天的退服大多还没闭环,更早的 93% 已闭环 is_recent = (T0 + datetime.timedelta(days=DAYS) - start).days < 10 if is_recent: closed = rng.choice(['否', '否', '是']) else: closed = rng.choices(['是', '否'], [93, 7])[0] outages.append({ 'record_id': 'TF%07d' % (3000001 + i), 'ne_id': ne_id, 'cell_id': cell_id, 'district': site['district'], 'vendor': site['vendor'], 'start_time': start.strftime(FMT), 'end_time': end.strftime(FMT), 'duration_min': reported, 'cause_category': category, 'cause_detail': detail, 'cause_code': code, 'ticket_id': 'GD%06d' % (200001 + i), 'closed': closed, }) # 联动告警:根因告警先于退服发生 -> 小区退服告警(约 65% 记录了父告警 parent_alarm_id) root_alarm = '' if category in ALARM_LIB: lead_min, lead_max = LEAD.get(category, (0, 8)) alarm_start = start - datetime.timedelta(minutes=rng.randint(lead_min, lead_max)) root_alarm = add_alarm(rng, alarms, ne_id, category, alarm_start, end) if category == '电源故障' and rng.random() < .6: # 停电导致传输设备掉电,衍生传输告警 add_alarm(rng, alarms, ne_id, '传输中断', start, end, root_alarm) if rng.random() < .65: parent = root_alarm else: parent = '' add_alarm(rng, alarms, ne_id, '小区退服', start, end, parent) # 约 7% 的退服前出现一条无关的干扰告警,检验算法能否按优先级排除 if rng.random() < .07: noise_category = rng.choice(['传输中断', '设备硬件', '天馈故障', '机房环境']) noise_start = start - datetime.timedelta(minutes=rng.randint(0, 12)) noise_end = noise_start + datetime.timedelta(minutes=rng.randint(2, 30)) add_alarm(rng, alarms, ne_id, noise_category, noise_start, noise_end) # 闭环过程时间:用独立的随机数生成器,不改变上面已有数据的随机序列。 # 接单(响应)时间 = 退服开始 + 响应时延:75% 在 1.5 小时内,18% 在 1.5~5 小时,7% 超过 5 小时; # 已闭环的闭环时间 = max(恢复时间, 接单时间) + 10 分钟 ~ 10 小时(复核、回单); # 未闭环的一半已接单(处理中),一半未接单(待处理)。 flow_rng = random.Random(seed + 1018) for outage in outages: start = datetime.datetime.strptime(outage['start_time'], FMT) end = datetime.datetime.strptime(outage['end_time'], FMT) r = flow_rng.random() if r < .75: delay = flow_rng.randint(3, 90) elif r < .93: delay = flow_rng.randint(91, 300) else: delay = flow_rng.randint(301, 1440) handle = start + datetime.timedelta(minutes=delay) if outage['closed'] == '是': close = max(end, handle) + datetime.timedelta(minutes=flow_rng.randint(10, 600)) outage['handle_time'] = handle.strftime(FMT) outage['close_time'] = close.strftime(FMT) else: outage['handle_time'] = handle.strftime(FMT) if flow_rng.random() < .5 else '' outage['close_time'] = '' # 背景告警:与退服无关的日常闪断告警 for _ in range(n): site = rng.choice(sites) category = rng.choice(['传输中断', '设备硬件', '天馈故障', '机房环境', '电源故障']) alarm_start = T0 + datetime.timedelta(minutes=rng.randint(0, DAYS * 1440)) alarm_end = alarm_start + datetime.timedelta(minutes=rng.choice([1, 3, 8, 20, 45])) add_alarm(rng, alarms, site['ne_id'], category, alarm_start, alarm_end) alarms.sort(key=lambda alarm: alarm['start_time']) os.makedirs(out, exist_ok=True) outputs = [('sites.csv', sites, SITE_HEAD), ('data_outage.csv', outages, OUTAGE_HEAD), ('data_alarm.csv', alarms, ALARM_HEAD)] for filename, rows, header in outputs: with open(os.path.join(out, filename), 'w', newline='', encoding='utf-8-sig') as f: writer = csv.DictWriter(f, fieldnames=header) writer.writeheader() writer.writerows(rows) return len(sites), len(outages), len(alarms) if __name__ == '__main__': parser = argparse.ArgumentParser(description='第18题基站退服数据生成器') parser.add_argument('--rows', type=int, default=6000, help='退服记录条数,默认 6000') parser.add_argument('--seed', type=int, default=2026) parser.add_argument('--out', default=os.path.join(os.path.dirname(os.path.abspath(__file__)), 'data')) args = parser.parse_args() n_sites, n_outages, n_alarms = generate(args.rows, args.seed, args.out) print('站点 %d 个,退服记录 %d 条,告警 %d 条 -> %s' % (n_sites, n_outages, n_alarms, args.out))