文档处理自动化——AI 批量处理网络配置、巡检报告

"用正则提取字段" 和 "让大模型直接读文档抽字段" 有什么区别？各自适合什么场景？给我一个对比表，并各举一个运维里的例子。

json{  "hostname": "SW-Core-01",  "model": "WS-C2960X-48TS-L",  "management_ip": "10.10.1.1",  "vlan_list": [1],  "ios_version": "15.2(7)E3",  "port_count": 48,  "interfaces": [    {"port": "GigabitEthernet0/1", "desc": "to-Firewall"},    {"port": "GigabitEthernet0/2", "desc": "to-Core-02"}  ]}

pythonimport re, os, globfrom openpyxl import WorkbookCONFIG_DIR = "./configs"rows = []pattern_host = re.compile(r"^hostname\s+(\S+)", re.M)pattern_model = re.compile(r"^model\s+(\S+)", re.M)pattern_ip = re.compile(r"ip address\s+([\d.]+)")pattern_ios = re.compile(r"^version\s+([\d().]+E]+)", re.M)for path in glob.glob(os.path.join(CONFIG_DIR, "*.conf")):    text = open(path, encoding="utf-8", errors="ignore").read()    host = pattern_host.search(text)    model = pattern_model.search(text)    ip = pattern_ip.search(text)    ios = pattern_ios.search(text)    rows.append({        "file": os.path.basename(path),        "hostname": host.group(1) if host else "",        "model": model.group(1) if model else "",        "mgmt_ip": ip.group(1) if ip else "",        "ios": ios.group(1) if ios else "",    })    print(f"已解析 {path}")wb = Workbook()ws = wb.activews.append(["文件名", "hostname", "型号", "管理IP", "IOS版本"])for r in rows:    ws.append([r["file"], r["hostname"], r["model"], r["mgmt_ip"], r["ios"]])wb.save("configs_report.xlsx")print(f"完成，共 {len(rows)} 台设备，已保存 configs_report.xlsx")

pythonimport os, glob, jsonfrom ollama import chatPROMPT = """你是网络配置解析器。只输出 JSON，字段：hostname, vendor, model, mgmt_ip, ios_version。不要任何解释。配置如下：%s"""results = []for path in glob.glob("./configs/*.conf"):    cfg = open(path, encoding="utf-8", errors="ignore").read()    resp = chat(model="deepseek-r1:7b",                messages=[{"role": "user", "content": PROMPT % cfg}])    results.append(json.loads(resp["message"]["content"]))print(json.dumps(results, ensure_ascii=False, indent=2))

pythonimport pdfplumberdef pdf_to_text(path):    with pdfplumber.open(path) as pdf:        return "\n".join(page.extract_text() or "" for page in pdf.pages)text = pdf_to_text("./inspection/sw01.pdf")print(text[:500])  # 看看抽出来的文本，再决定怎么抽字段

原始文档(配置/PDF/Excel)   ↓ ① 加载与清洗（去噪、统一编码 UTF-8、拆页/分段）   ↓ ② 定位（找到含目标字段的段落，过滤无关内容）   ↓ ③ 抽取（正则 or 大模型 or 专用解析库）   ↓ ④ 结构化校验（字段类型、必填项、范围、格式）   ↓ ⑤ 合并与去重   ↓ ⑥ 生成报告（Excel / HTML / 对比表）   ↓ ⑦ 人工复核 & 归档

pythonimport jsonschemaschema = {    "type": "object",    "properties": {        "hostname": {"type": "string"},        "mgmt_ip": {"type": "string", "pattern": r"^\d{1,3}(\.\d{1,3}){3}$"}    },    "required": ["hostname"]}# 抽完用 jsonschema.validate 校验，失败就重试一次或标记人工复核

pythonimport glob, osfrom openpyxl import Workbookfrom openpyxl.styles import Font, PatternFillREPORT_DIR = "./inspection/"rows = []pat_dev = re.compile(r"设备[:：]\s*(\S+)")pat_cpu = re.compile(r"CPU[利用率使用率]*[:：]?\s*(\d+)%")pat_mem = re.compile(r"内存[利用率使用率]*[:：]?\s*(\d+)%")pat_alarm = re.compile(r"告警[数条]*[:：]?\s*(\d+)")pat_ping = re.compile(r"连通性[:：]\s*(正常|异常|通|不通)")for path in glob.glob(os.path.join(REPORT_DIR, "*.txt")):    t = open(path, encoding="utf-8", errors="ignore").read()    dev = pat_dev.search(t)    cpu = pat_cpu.search(t)    mem = pat_mem.search(t)    al = pat_alarm.search(t)    pk = pat_ping.search(t)    rows.append([        dev.group(1) if dev else os.path.basename(path),        cpu.group(1) + "%" if cpu else "N/A",        mem.group(1) + "%" if mem else "N/A",        al.group(1) if al else "0",        pk.group(1) if pk else "未知",    ])wb = Workbook()ws = wb.activews.append(["设备名", "CPU利用率", "内存利用率", "告警数", "连通性"])for c in ws[1]:    c.font = Font(bold=True, color="FFFFFF")    c.fill = PatternFill("solid", fgColor="667EEA")for r in rows:    ws.append(r)red = PatternFill("solid", fgColor="FFE0E0")for row in ws.iter_rows(min_row=2):    v = str(row[3].value)    if v.isdigit() and int(v) > 0:        for c in row:            c.fill = redwb.save("weekly_inspection.xlsx")print("周报已生成：weekly_inspection.xlsx，共", len(rows), "台")

pythonimport difflibdef config_diff(old, new):    a = open(old, encoding="utf-8", errors="ignore").readlines()    b = open(new, encoding="utf-8", errors="ignore").readlines()    return list(difflib.unified_diff(a, b,                fromfile="变更前", tofile="变更后", lineterm=""))changes = config_diff("./backup/sw01_before.conf", "./backup/sw01_after.conf")with open("sw01_diff.txt", "w", encoding="utf-8") as f:    f.writelines(changes)print("变更对比已生成，共", len([c for c in changes if c.startswith(('+','-')) and not c.startswith(('+++','---'))]), "处改动")