diff --git a/src/handlers.py b/src/handlers.py
index e1dc92c..c768e0d 100644
--- a/src/handlers.py
+++ b/src/handlers.py
@@ -32,6 +32,7 @@ from src.models import (
     upsert_fixture_param,
     insert_device_log,
     update_device_status,
+    get_all_device_serials,
 )
 from src.dg430 import (
     parse_b2_status,
@@ -147,6 +148,7 @@ async def handle_count_off(data: dict, addr: tuple):
     _registry[serial] = dnt_id
     _heartbeat[serial] = time.time()
     record_interaction(serial)
+    _device_status[serial] = 1  # 登录即视为在线状态
 
     await ensure_collect_table(serial)
 
@@ -717,13 +719,14 @@ MONITOR_INTERVAL = 5           # 状态检查间隔 (秒)
 
 
 async def device_status_monitor():
-    """后台轮询：监控所有已注册设备在线/离线状态
+    """后台轮询：监控所有设备在线/离线状态
 
-    判定规则：
-      1. > OFFLINE_IDLE_SEC 无交互 → 离线
-      2. 最近 OFFLINE_IDLE_SEC 内交互次数 < POOR_MIN_INTERACTIONS → 通信不良
-      3. 最近 3 次交互间隔均 < INTERACTION_TIMEOUT → 在线
-      4. 状态变化时写入 tb_device_log + 更新 dnt_info.state
+    两个阶段：
+      Phase 1 — 遍历 _registry 中活跃设备，根据交互记录判定状态
+      Phase 2 — 扫描 dnt_info 全表，修正与实际交互不符的状态：
+        - state=1(在线) 但 >60s 无交互 → 更新为离线
+        - state=0(离线) 但有交互记录 → 根据交互模式更新为在线/通信不良
+        - state 与计算值不一致 → 同步修正
     """
     logger.info("设备状态监控服务启动")
     await asyncio.sleep(5)  # 等其它服务就绪
@@ -731,39 +734,39 @@ async def device_status_monitor():
     while True:
         try:
             now = time.time()
+
+            # ── Phase 1: 活跃设备（在 _registry 中）────
             for device_id, dnt_id in list(_registry.items()):
                 interactions = _interactions.get(device_id)
-                if not interactions:
+                actual_state = _calc_device_state(interactions, now) if interactions else 0
+                old_state = _device_status.get(device_id, 1)  # 注册时默认在线
+
+                if actual_state != old_state:
+                    await _apply_state_change(device_id, actual_state, old_state)
+
+            # ── Phase 2: 扫描 dnt_info 全表，修正不一致 ──
+            try:
+                rows = await get_all_device_serials()
+            except Exception as e:
+                logger.error(f"查询 dnt_info 失败: {e}")
+                rows = []
+
+            for serial, db_state, ip in rows:
+                if not serial:
                     continue
 
-                # 计算新状态
-                new_state = _calc_device_state(interactions, now)
-                old_state = _device_status.get(device_id, 1)  # 默认在线
+                interactions = _interactions.get(serial)
+                actual_state = _calc_device_state(interactions, now) if interactions else 0
+                memory_state = _device_status.get(serial)
 
-                if new_state != old_state:
-                    _device_status[device_id] = new_state
-                    logger.info(
-                        f"设备 {device_id} 状态变更: "
-                        f"{_state_name(old_state)} → {_state_name(new_state)}"
-                    )
+                # 优先用内存状态做基准（更实时）；无内存状态则用 DB 状态
+                tracked_state = memory_state if memory_state is not None else db_state
 
-                    # 更新 dnt_info
-                    try:
-                        await update_device_status(device_id, new_state)
-                    except Exception as e:
-                        logger.error(f"更新设备状态失败: {e}")
+                if actual_state != tracked_state:
+                    await _apply_state_change(serial, actual_state, tracked_state, ip=ip)
 
-                    # 写入事件日志
-                    dnt = await get_dnt_by_serial(device_id)
-                    dev_ip = dnt.get("ip", "") if dnt else ""
-                    event_type, content = _state_event(new_state, old_state)
-                    try:
-                        await insert_device_log(
-                            serial=device_id, ip=dev_ip,
-                            event_type=event_type, content=content,
-                        )
-                    except Exception as e:
-                        logger.error(f"写入设备事件日志失败: {e}")
+            # ── Phase 3: 清理 _registry 中已经不存在的设备 ──
+            # （已注销 / 长期无交互的设备不清除，继续监控）
 
         except Exception as e:
             logger.error(f"设备状态监控异常: {e}")
@@ -771,6 +774,40 @@ async def device_status_monitor():
         await asyncio.sleep(MONITOR_INTERVAL)
 
 
+async def _apply_state_change(device_id: str, new_state: int, old_state: int,
+                              ip: str = ""):
+    """应用状态变更：更新内存、dnt_info、写入日志"""
+    _device_status[device_id] = new_state
+    logger.info(
+        f"设备 {device_id} 状态变更: "
+        f"{_state_name(old_state)} → {_state_name(new_state)}"
+    )
+
+    # 更新 dnt_info
+    try:
+        await update_device_status(device_id, new_state)
+    except Exception as e:
+        logger.error(f"更新设备状态失败: {e}")
+
+    # 获取设备 IP（未传入时从 DB 查）
+    if not ip:
+        try:
+            dnt = await get_dnt_by_serial(device_id)
+            ip = dnt.get("ip", "") if dnt else ""
+        except Exception:
+            pass
+
+    # 写入事件日志
+    event_type, content = _state_event(new_state, old_state)
+    try:
+        await insert_device_log(
+            serial=device_id, ip=ip,
+            event_type=event_type, content=content,
+        )
+    except Exception as e:
+        logger.error(f"写入设备事件日志失败: {e}")
+
+
 def _calc_device_state(interactions: deque, now: float) -> int:
     """根据交互记录计算设备状态 (0=离线 1=在线 2=通信不良)"""
     # 清理过期记录（仅保留 60s 内）
diff --git a/src/models.py b/src/models.py
index 877f915..b854b55 100644
--- a/src/models.py
+++ b/src/models.py
@@ -604,6 +604,17 @@ async def update_device_status(serial: str, state: int):
                 )
 
 
+async def get_all_device_serials() -> list[tuple[str, int, str]]:
+    """获取所有设备 (serial, state, ip)，用于状态扫描"""
+    pool = await get_pool()
+    async with pool.acquire() as conn:
+        async with conn.cursor() as cur:
+            await cur.execute(
+                "SELECT serial, state, ip FROM dnt_info ORDER BY id"
+            )
+            return await cur.fetchall()
+
+
 # ─── tb_serialnet CRUD ─────────────────────────────────────────────
 
 async def get_pending_serialnet(dnt_id: int) -> dict | None: