06131024

models_processing/model_tf/tf_lstm.py

@@ -4,7 +4,15 @@
 # @Time      :2025/2/12 14:03
 # @Author    :David
 # @Company: shenyang JY
-
+# 固定NumPy随机种子
+import os
+os.environ['PYTHONHASHSEED'] = '42'
+import numpy as np
+np.random.seed(42)
+# 固定TensorFlow随机种子
+import tensorflow as tf
+tf.random.set_seed(42)
+from tensorflow.keras.initializers import glorot_uniform, orthogonal
 from tensorflow.keras.layers import Input, Dense, LSTM, concatenate, Conv1D, Conv2D, MaxPooling1D, Reshape, Flatten
 from tensorflow.keras.models import Model, load_model
 from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping, TensorBoard, ReduceLROnPlateau
@@ -12,11 +20,10 @@ from tensorflow.keras import optimizers, regularizers
 from models_processing.model_tf.losses import region_loss
 import numpy as np
 from common.database_dml import *
-from models_processing.model_tf.settings import set_deterministic
 from threading import Lock
 import argparse
 model_lock = Lock()
-set_deterministic(42)
+
 
 class TSHandler(object):
     def __init__(self, logger, args):
@@ -43,13 +50,17 @@ class TSHandler(object):
         l2_reg = regularizers.l2(opt.Model['lambda_value_2'])
         nwp_input = Input(shape=(opt.Model['time_step']*time_series, opt.Model['input_size']), name='nwp')
 
-        con1 = Conv1D(filters=64, kernel_size=5, strides=1, padding='valid', activation='relu', kernel_regularizer=l2_reg)(nwp_input)
+        con1 = Conv1D(filters=64, kernel_size=5, strides=1, padding='valid', activation='relu', kernel_regularizer=l2_reg, kernel_initializer=glorot_uniform(seed=42))(nwp_input)
         con1_p = MaxPooling1D(pool_size=5, strides=1, padding='valid', data_format='channels_last')(con1)
-        nwp_lstm = LSTM(units=opt.Model['hidden_size'], return_sequences=False, kernel_regularizer=l2_reg)(con1_p)
+        nwp_lstm = LSTM(units=opt.Model['hidden_size'], return_sequences=False, kernel_regularizer=l2_reg, kernel_initializer=glorot_uniform(seed=43),
+        recurrent_initializer=orthogonal(seed=44),  # LSTM特有的初始化
+        bias_initializer='zeros')(con1_p)
         if lstm_type == 2:
-            output = Dense(opt.Model['time_step'], name='cdq_output')(nwp_lstm)
+            output = Dense(opt.Model['time_step'], name='cdq_output', kernel_initializer=glorot_uniform(seed=45),
+            bias_initializer='zeros')(nwp_lstm)
         else:
-            output = Dense(opt.Model['time_step']*time_series, name='cdq_output')(nwp_lstm)
+            output = Dense(opt.Model['time_step']*time_series, name='cdq_output', kernel_initializer=glorot_uniform(seed=45),
+            bias_initializer='zeros')(nwp_lstm)
 
         model = Model(nwp_input, output)
         adam = optimizers.Adam(learning_rate=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-7, amsgrad=True)
@@ -64,13 +75,17 @@ class TSHandler(object):
         nwp_input = Input(shape=(opt.Model['time_step'] * time_series, opt.Model['input_size']), name='nwp')
 
         con1 = Conv1D(filters=64, kernel_size=1, strides=1, padding='valid', activation='relu',
-                      kernel_regularizer=l2_reg)(nwp_input)
+                      kernel_regularizer=l2_reg, kernel_initializer=glorot_uniform(seed=42))(nwp_input)
         con1_p = MaxPooling1D(pool_size=1, strides=1, padding='valid', data_format='channels_last')(con1)
-        nwp_lstm = LSTM(units=opt.Model['hidden_size'], return_sequences=False, kernel_regularizer=l2_reg)(con1_p)
+        nwp_lstm = LSTM(units=opt.Model['hidden_size'], return_sequences=False, kernel_regularizer=l2_reg, kernel_initializer=glorot_uniform(seed=43),
+        recurrent_initializer=orthogonal(seed=44),  # LSTM特有的初始化
+        bias_initializer='zeros')(con1_p)
         if lstm_type == 2:
-            output = Dense(opt.Model['time_step'], name='cdq_output')(nwp_lstm)
+            output = Dense(opt.Model['time_step'], name='cdq_output', kernel_initializer=glorot_uniform(seed=45),
+            bias_initializer='zeros')(nwp_lstm)
         else:
-            output = Dense(opt.Model['time_step'] * time_series, name='cdq_output')(nwp_lstm)
+            output = Dense(opt.Model['time_step'] * time_series, name='cdq_output', kernel_initializer=glorot_uniform(seed=45),
+            bias_initializer='zeros')(nwp_lstm)
 
         model = Model(nwp_input, output)
         adam = optimizers.Adam(learning_rate=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-7, amsgrad=True)