fix: 분석 파이프라인 정확도 개선 + 캐시 증분 갱신 + TTS 프록시

- MIN_TRAJ_POINTS 100→20 (16척→684척, 파이프라인 병목 해소)
- risk.py: SOG 급변 count를 위험도 점수에 반영 (+5/+10)
- spoofing.py: BD09 오프셋 중국 MMSI(412*) 예외 (좌표계 노이즈 제거)
- fishing_pattern.py: 마지막 조업 세그먼트 누락 버그 수정
- VesselAnalysisService: 인메모리 캐시 + 증분 갱신 (warmup 2h → incremental)
- nginx: /api/gtts 프록시 추가 (Google TTS CORS 우회)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

backend/src/main/java/gc/mda/kcg/domain/analysis/VesselAnalysisService.java

@@ -1,49 +1,74 @@
 package gc.mda.kcg.domain.analysis;
 
-import gc.mda.kcg.config.CacheConfig;
 import gc.mda.kcg.domain.fleet.GroupPolygonService;
 import lombok.RequiredArgsConstructor;
-import org.springframework.cache.Cache;
-import org.springframework.cache.CacheManager;
+import lombok.extern.slf4j.Slf4j;
 import org.springframework.stereotype.Service;
 
 import java.time.Instant;
 import java.time.temporal.ChronoUnit;
 import java.util.*;
+import java.util.concurrent.ConcurrentHashMap;
 
+@Slf4j
 @Service
 @RequiredArgsConstructor
 public class VesselAnalysisService {
 
     private final VesselAnalysisResultRepository repository;
-    private final CacheManager cacheManager;
     private final GroupPolygonService groupPolygonService;
 
-    /**
-     * 최근 2시간 내 분석 결과 + 집계 통계를 반환한다.
-     * mmsi별 최신 1건만. Caffeine 캐시(TTL 5분) 적용.
-     */
-    @SuppressWarnings("unchecked")
-    public Map<String, Object> getLatestResultsWithStats() {
-        Cache cache = cacheManager.getCache(CacheConfig.VESSEL_ANALYSIS);
-        if (cache != null) {
-            Cache.ValueWrapper wrapper = cache.get("data_with_stats");
-            if (wrapper != null) {
-                return (Map<String, Object>) wrapper.get();
-            }
-        }
+    private static final long CACHE_TTL_MS = 2 * 60 * 60_000L; // 2시간
 
-        Instant since = Instant.now().minus(2, ChronoUnit.HOURS);
-        Map<String, VesselAnalysisResult> latest = new LinkedHashMap<>();
-        for (VesselAnalysisResult r : repository.findByAnalyzedAtAfter(since)) {
-            latest.merge(r.getMmsi(), r, (old, cur) ->
+    /** mmsi → 최신 분석 결과 (인메모리 캐시) */
+    private final Map<String, VesselAnalysisResult> cache = new ConcurrentHashMap<>();
+    private volatile Instant lastFetchTime = null;
+    private volatile long lastUpdatedAt = 0;
+
+    /**
+     * 최근 2시간 분석 결과 + 집계 통계.
+     * - 첫 호출(warmup): 2시간 전체 조회 → 캐시 구축
+     * - 이후: lastFetchTime 이후 증분만 조회 → 캐시 병합
+     * - 2시간 초과 항목은 evict
+     * - 값 갱신 시 TTL 타이머 초기화
+     */
+    public Map<String, Object> getLatestResultsWithStats() {
+        Instant now = Instant.now();
+
+        if (lastFetchTime == null || (System.currentTimeMillis() - lastUpdatedAt) > CACHE_TTL_MS) {
+            // warmup: 2시간 전체 조회
+            Instant since = now.minus(2, ChronoUnit.HOURS);
+            List<VesselAnalysisResult> rows = repository.findByAnalyzedAtAfter(since);
+            cache.clear();
+            for (VesselAnalysisResult r : rows) {
+                cache.merge(r.getMmsi(), r, (old, cur) ->
                         cur.getAnalyzedAt().isAfter(old.getAnalyzedAt()) ? cur : old);
             }
+            lastFetchTime = now;
+            lastUpdatedAt = System.currentTimeMillis();
+            log.info("vessel analysis cache warmup: {} vessels from DB", cache.size());
+        } else {
+            // 증분: lastFetchTime 이후만 조회
+            List<VesselAnalysisResult> rows = repository.findByAnalyzedAtAfter(lastFetchTime);
+            if (!rows.isEmpty()) {
+                for (VesselAnalysisResult r : rows) {
+                    cache.merge(r.getMmsi(), r, (old, cur) ->
+                            cur.getAnalyzedAt().isAfter(old.getAnalyzedAt()) ? cur : old);
+                }
+                lastUpdatedAt = System.currentTimeMillis();
+                log.debug("vessel analysis incremental merge: {} new rows", rows.size());
+            }
+            lastFetchTime = now;
+        }
 
-        // 집계 통계 — 같은 루프에서 계산
+        // 2시간 초과 항목 evict
+        Instant cutoff = now.minus(2, ChronoUnit.HOURS);
+        cache.entrySet().removeIf(e -> e.getValue().getAnalyzedAt().isBefore(cutoff));
+
+        // 집계 통계
         int dark = 0, spoofing = 0, critical = 0, high = 0, medium = 0, low = 0;
         Set<Integer> clusterIds = new HashSet<>();
-        for (VesselAnalysisResult r : latest.values()) {
+        for (VesselAnalysisResult r : cache.values()) {
             if (Boolean.TRUE.equals(r.getIsDark())) dark++;
             if (r.getSpoofingScore() != null && r.getSpoofingScore() > 0.5) spoofing++;
             String level = r.getRiskLevel();
@@ -62,11 +87,10 @@ public class VesselAnalysisService {
             }
         }
 
-        // 어구 통계 — group_polygon_snapshots 기반
         Map<String, Integer> gearStats = groupPolygonService.getGearStats();
 
         Map<String, Object> stats = new LinkedHashMap<>();
-        stats.put("total", latest.size());
+        stats.put("total", cache.size());
         stats.put("dark", dark);
         stats.put("spoofing", spoofing);
         stats.put("critical", critical);
@@ -77,21 +101,15 @@ public class VesselAnalysisService {
         stats.put("gearGroups", gearStats.get("gearGroups"));
         stats.put("gearCount", gearStats.get("gearCount"));
 
-        List<VesselAnalysisDto> results = latest.values().stream()
+        List<VesselAnalysisDto> results = cache.values().stream()
                 .sorted(Comparator.comparingInt(VesselAnalysisResult::getRiskScore).reversed())
                 .map(VesselAnalysisDto::from)
                 .toList();
 
-        Map<String, Object> response = Map.of(
+        return Map.of(
                 "count", results.size(),
                 "items", results,
                 "stats", stats
         );
-
-        if (cache != null) {
-            cache.put("data_with_stats", response);
-        }
-
-        return response;
     }
 }

deploy/nginx-kcg.conf

@@ -94,6 +94,16 @@ server {
         proxy_ssl_server_name on;
     }
 
+    # ── Google TTS 프록시 (중국어 경고문 음성) ──
+    location /api/gtts {
+        rewrite ^/api/gtts(.*)$ /translate_tts$1 break;
+        proxy_pass https://translate.google.com;
+        proxy_set_header Host translate.google.com;
+        proxy_set_header Referer "https://translate.google.com/";
+        proxy_set_header User-Agent "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36";
+        proxy_ssl_server_name on;
+    }
+
     # gzip
     gzip on;
     gzip_types text/plain text/css application/json application/javascript text/xml application/xml;

prediction/algorithms/fishing_pattern.py

@@ -92,6 +92,26 @@ def detect_fishing_segments(df_vessel: pd.DataFrame,
                     })
             in_fishing = False
 
+    # 트랙 끝까지 조업 중이면 마지막 세그먼트 추가
+    if in_fishing and len(records) > seg_start_idx:
+        start_ts = records[seg_start_idx].get('timestamp')
+        end_ts = records[-1].get('timestamp')
+        if start_ts and end_ts:
+            dur_sec = (pd.Timestamp(end_ts) - pd.Timestamp(start_ts)).total_seconds()
+            dur_min = dur_sec / 60
+            if dur_min >= window_min:
+                zone_info = classify_zone(
+                    records[seg_start_idx].get('lat', 0),
+                    records[seg_start_idx].get('lon', 0),
+                )
+                segments.append({
+                    'start_idx': seg_start_idx,
+                    'end_idx': len(records) - 1,
+                    'duration_min': round(dur_min, 1),
+                    'zone': zone_info.get('zone', 'UNKNOWN'),
+                    'in_territorial_sea': zone_info.get('zone') == 'TERRITORIAL_SEA',
+                })
+
     return segments
 
 

prediction/algorithms/risk.py

@@ -54,6 +54,13 @@ def compute_vessel_risk_score(
     if teleports:
         score += 20
 
+    from algorithms.spoofing import count_speed_jumps
+    jumps = count_speed_jumps(df_vessel)
+    if jumps >= 3:
+        score += 10
+    elif jumps >= 1:
+        score += 5
+
     gaps = detect_ais_gaps(df_vessel)
     critical_gaps = [g for g in gaps if g['gap_min'] >= 60]
     if critical_gaps:

prediction/algorithms/spoofing.py

@@ -68,11 +68,13 @@ def compute_spoofing_score(df_vessel: pd.DataFrame) -> float:
     if jumps > 0:
         score += min(0.3, jumps / n * 5)
 
-    # BD09 오프셋 (중국 좌표 사용 의심)
+    # BD09 오프셋 — 중국 선박(412*)은 좌표계 차이로 항상 ~300m이므로 제외
+    mmsi_str = str(df_vessel.iloc[0].get('mmsi', '')) if 'mmsi' in df_vessel.columns else ''
+    if not mmsi_str.startswith('412'):
         mid_idx = len(df_vessel) // 2
         row = df_vessel.iloc[mid_idx]
         offset = compute_bd09_offset(row['lat'], row['lon'])
-    if offset > 300:  # 300m 이상
+        if offset > 300:
             score += 0.3
         elif offset > 100:
             score += 0.1

prediction/pipeline/constants.py

@@ -18,7 +18,7 @@ MIN_CLUSTER_SIZE = 5
 MMSI_DIGITS = 9
 MAX_VESSEL_LENGTH = 300
 MAX_SOG_KNOTS = 30.0
-MIN_TRAJ_POINTS = 100
+MIN_TRAJ_POINTS = 20
 
 KR_BOUNDS = {
     'lat_min': 32.0, 'lat_max': 39.0,