减少爆管定位代码中引入的不确定性

app/algorithms/burst_location/similarity_calculator.py

@@ -44,7 +44,7 @@ def cal_similarity_simple_return_dd(
             similarity_dis = 0
         else:"""
         none_flag = 0
-        sensor_for_cos = list(
+        sensor_for_cos = sorted(
             set(dpressure_s.index).intersection(set(act_dpressure.index))
         )
         """if len(dpressure_s) ==0 or len(dpressure) ==0:
@@ -74,7 +74,7 @@ def cal_similarity_simple_return_dd(
             none_flag = 1
         else:"""
         none_flag = 0
-        important_sensor = list(
+        important_sensor = sorted(
             set(important_sensor).intersection(set(act_dpressure.index))
         )
         part_dpressure = dpressure_s[important_sensor] - dpressure[important_sensor]
@@ -107,7 +107,7 @@ def cal_similarity_simple_return_dd(
 
     elif similarity_mode == "CAD_new_gy" or similarity_mode == "CDF":
         # cos
-        sensor_for_cos = list(
+        sensor_for_cos = sorted(
             set(dpressure_s.index).intersection(set(act_dpressure.index))
         )
         if len(sensor_for_cos) == 0 and len(dpressure_s) == 0:
@@ -141,7 +141,7 @@ def cal_similarity_simple_return_dd(
             similarity_cos = s1 / s2
 
         # DIS
-        important_sensor_new = list(
+        important_sensor_new = sorted(
             set(important_sensor).intersection(set(act_dpressure.index))
         )
         if len(important_sensor_new) == 0:
@@ -172,7 +172,7 @@ def cal_similarity_simple_return_dd(
         similarity_cos = 0
         none_flag = 0
         # DIS
-        important_sensor_new = list(
+        important_sensor_new = sorted(
             set(important_sensor).intersection(set(act_dpressure.index))
         )
         if len(important_sensor_new) == 0:
@@ -762,15 +762,24 @@ def update_similarity(leak_candidate_center, similarity, leak_center_dict):
     return similarity_new
 
 
-def extra_judge(similarity):
+def extra_judge(
+    similarity, min_candidates_to_prune: int = 200, std_relax_factor: float = 0.5
+):
+    if len(similarity.index) == 0:
+        return 1.0, similarity
+    if len(similarity.index) < int(min_candidates_to_prune):
+        return 1.0, similarity
+
     mean_similarity = float(similarity.mean())
-    record_sensor = []
-    record_value = []
-    for i in range(len(similarity.index)):
-        if similarity.iloc[i] >= mean_similarity - 1e-10:
-            record_value.append(similarity.iloc[i])
-            record_sensor.append(similarity.index[i])
-    out_put_similarity = pd.Series(record_value, index=record_sensor, dtype=float)
+    std_similarity = float(similarity.std())
+    if not math.isfinite(std_similarity):
+        std_similarity = 0.0
+
+    threshold = mean_similarity - float(std_relax_factor) * std_similarity
+    out_put_similarity = similarity[similarity >= threshold - 1e-10]
+    if len(out_put_similarity.index) == 0:
+        out_put_similarity = similarity.iloc[:1]
+
     cut_ratio = len(out_put_similarity.index) / len(similarity.index)
     return cut_ratio, out_put_similarity