bug似乎解决了

ga_calibration_by_one_factory.py

@@ -103,7 +103,7 @@ def update_production_line_csv(csv_path: Path, factory_name_cn: str, line_ids: L
         _release_lock(lock_path, fd)
 
 
-def evaluate(factory_code: str, line_ids: List[str], genes: List[float]) -> float:
+def evaluate(factory_code: str, factory_name_cn: str, line_ids: List[str], genes: List[float]) -> float:
     factory_factors = {}
     model = SimulationModel(
         factory_factors=factory_factors,
@@ -115,6 +115,13 @@ def evaluate(factory_code: str, line_ids: List[str], genes: List[float]) -> floa
         model.line_factor[lid] = float(val)
     while model.running:
         model.step()
+    # Prefer per-factory error ratio; fall back to aggregate model.error if missing.
+    if model.factory_error_df is not None and not model.factory_error_df.empty:
+        matched = model.factory_error_df[
+            model.factory_error_df["name"].astype(str).str.strip() == str(factory_name_cn).strip()
+        ]
+        if not matched.empty and "error_ratio" in matched.columns:
+            return float(matched.iloc[0]["error_ratio"])
     return model.error
 
 
@@ -235,7 +242,11 @@ def main():
     line_ids = [lid for lid, _, _, _ in seed_lines]
     seed_vals = [seed for _, seed, _, _ in seed_lines]
     bounds = [(mn, mx) for _, _, mn, mx in seed_lines]
-    prev_best_error = float(factory_row["最小误差"]) if pd.notna(factory_row["最小误差"]) else float("inf")
+    prev_best_error = (
+        float(factory_row[factory_cols["最小误差"]])
+        if pd.notna(factory_row[factory_cols["最小误差"]])
+        else float("inf")
+    )
 
     print(f"[START] 校准工厂 {args.factory} / {factory_name_cn} (产线数={len(line_ids)}, baseline_error={prev_best_error:.6f})")
 
@@ -248,7 +259,7 @@ def main():
         scored = []
         for indiv in population:
             indiv = apply_bounds(indiv, bounds)
-            score = evaluate(args.factory, line_ids, indiv)
+            score = evaluate(args.factory, factory_name_cn, line_ids, indiv)
             # print(f"[{args.factory}] Gen {gen+1} try factors={indiv} -> error={score:.6f}")
             scored.append((score, indiv))
             if score < best_score: