基于CNN-GRU-Attention混合神经网络的负荷预测方法（Python代码实现）

 👨‍🎓个人主页：研学社的博客 

💥💥💞💞欢迎来到本博客❤️❤️💥💥

🏆博主优势：🌞🌞🌞博客内容尽量做到思维缜密，逻辑清晰，为了方便读者。

⛳️座右铭：行百里者，半于九十。

📋📋📋本文目录如下：🎁🎁🎁

目录

💥1 概述

📚2 运行结果

🎉3 参考文献

🌈4 Matlab代码实现

💥1 概述

   基于CNN-GRU(convolutional neural networks-gate recurrent unit)神经网络的电力系统短期负荷预测方法。首先使用卷积神经网络(CNN)对负荷及气象数据进行卷积处理,以更好地提取数据新特征,增强输入数据与输出数据间的相关性。然后使用门控循环单元(GRU)实现短期负荷预测。使用某地区的负荷数据结合当地的气象数据,对CNN-GRU-Attention方法进行了测试。结果表明:与单独的CNN网络或GRU网络相比,CNN-GRU-Attention网络对电力系统短期负荷的预测误差更小,预测精度更高。

📚2 运行结果

 

  

To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 reshape (Reshape)           (None, 1, 7, 1)           0         
                                                                 
 conv2d (Conv2D)             (None, 1, 7, 32)          320       
                                                                 
 max_pooling2d (MaxPooling2D  (None, 1, 7, 32)         0         
 )                                                               
                                                                 
 dropout (Dropout)           (None, 1, 7, 32)          0         
                                                                 
 reshape_1 (Reshape)         (None, 1, 224)            0         
                                                                 
 gru (GRU)                   (None, 1, 10)             7080      
                                                                 
 gru_1 (GRU)                 (None, 1, 20)             1920      
                                                                 
 attention (Attention)       (None, 50)                2400      
                                                                 
 dense (Dense)               (None, 10)                510       
                                                                 
 dense_1 (Dense)             (None, 1)                 11        
                                                                 
=================================================================
Total params: 12,241
Trainable params: 12,241
Non-trainable params: 0
_________________________________________________________________
None
Epoch 1/50
2022-11-15 20:40:42.293543: W tensorflow/core/grappler/costs/op_level_cost_estimator.cc:690] Error in PredictCost() for the op: op: "Softmax" attr { key: "T" value { type: DT_FLOAT } } inputs { dtype: DT_FLOAT shape { unknown_rank: true } } device { type: "CPU" vendor: "AuthenticAMD" model: "240" frequency: 3194 num_cores: 16 environment { key: "cpu_instruction_set" value: "SSE, SSE2" } environment { key: "eigen" value: "3.4.90" } l1_cache_size: 32768 l2_cache_size: 524288 l3_cache_size: 16777216 memory_size: 268435456 } outputs { dtype: DT_FLOAT shape { unknown_rank: true } }
2022-11-15 20:40:43.604200: W tensorflow/core/grappler/costs/op_level_cost_estimator.cc:690] Error in PredictCost() for the op: op: "Softmax" attr { key: "T" value { type: DT_FLOAT } } inputs { dtype: DT_FLOAT shape { unknown_rank: true } } device { type: "CPU" vendor: "AuthenticAMD" model: "240" frequency: 3194 num_cores: 16 environment { key: "cpu_instruction_set" value: "SSE, SSE2" } environment { key: "eigen" value: "3.4.90" } l1_cache_size: 32768 l2_cache_size: 524288 l3_cache_size: 16777216 memory_size: 268435456 } outputs { dtype: DT_FLOAT shape { unknown_rank: true } }
103/103 - 3s - loss: 0.0332 - mse: 0.0332 - val_loss: 0.0068 - val_mse: 0.0068 - 3s/epoch - 29ms/step
Epoch 2/50
103/103 - 0s - loss: 0.0038 - mse: 0.0038 - val_loss: 0.0011 - val_mse: 0.0011 - 380ms/epoch - 4ms/step
Epoch 3/50
103/103 - 0s - loss: 0.0021 - mse: 0.0021 - val_loss: 8.9063e-04 - val_mse: 8.9063e-04 - 380ms/epoch - 4ms/step
Epoch 4/50
103/103 - 0s - loss: 0.0017 - mse: 0.0017 - val_loss: 7.7933e-04 - val_mse: 7.7933e-04 - 390ms/epoch - 4ms/step
Epoch 5/50
103/103 - 0s - loss: 0.0015 - mse: 0.0015 - val_loss: 7.5322e-04 - val_mse: 7.5322e-04 - 380ms/epoch - 4ms/step
Epoch 6/50
103/103 - 0s - loss: 0.0014 - mse: 0.0014 - val_loss: 7.0502e-04 - val_mse: 7.0502e-04 - 370ms/epoch - 4ms/step
Epoch 7/50
103/103 - 0s - loss: 0.0013 - mse: 0.0013 - val_loss: 6.4338e-04 - val_mse: 6.4338e-04 - 380ms/epoch - 4ms/step
Epoch 8/50
103/103 - 0s - loss: 0.0012 - mse: 0.0012 - val_loss: 6.1803e-04 - val_mse: 6.1803e-04 - 370ms/epoch - 4ms/step
Epoch 9/50
103/103 - 0s - loss: 0.0011 - mse: 0.0011 - val_loss: 5.5700e-04 - val_mse: 5.5700e-04 - 380ms/epoch - 4ms/step
Epoch 10/50
103/103 - 0s - loss: 0.0011 - mse: 0.0011 - val_loss: 5.2479e-04 - val_mse: 5.2479e-04 - 370ms/epoch - 4ms/step
Epoch 11/50
103/103 - 0s - loss: 0.0010 - mse: 0.0010 - val_loss: 5.1472e-04 - val_mse: 5.1472e-04 - 407ms/epoch - 4ms/step
Epoch 12/50
103/103 - 0s - loss: 0.0010 - mse: 0.0010 - val_loss: 4.8988e-04 - val_mse: 4.8988e-04 - 404ms/epoch - 4ms/step
Epoch 13/50
103/103 - 0s - loss: 9.7950e-04 - mse: 9.7950e-04 - val_loss: 5.0468e-04 - val_mse: 5.0468e-04 - 390ms/epoch - 4ms/step
Epoch 14/50
103/103 - 0s - loss: 9.5678e-04 - mse: 9.5678e-04 - val_loss: 4.9064e-04 - val_mse: 4.9064e-04 - 448ms/epoch - 4ms/step
Epoch 15/50
103/103 - 0s - loss: 9.1728e-04 - mse: 9.1728e-04 - val_loss: 5.0738e-04 - val_mse: 5.0738e-04 - 374ms/epoch - 4ms/step
Epoch 16/50
103/103 - 0s - loss: 9.0497e-04 - mse: 9.0497e-04 - val_loss: 4.8773e-04 - val_mse: 4.8773e-04 - 386ms/epoch - 4ms/step
Epoch 17/50
103/103 - 0s - loss: 8.8163e-04 - mse: 8.8163e-04 - val_loss: 4.8866e-04 - val_mse: 4.8866e-04 - 391ms/epoch - 4ms/step
Epoch 18/50
103/103 - 0s - loss: 8.8587e-04 - mse: 8.8587e-04 - val_loss: 4.9860e-04 - val_mse: 4.9860e-04 - 378ms/epoch - 4ms/step
Epoch 19/50
103/103 - 0s - loss: 8.7227e-04 - mse: 8.7227e-04 - val_loss: 5.0278e-04 - val_mse: 5.0278e-04 - 417ms/epoch - 4ms/step
Epoch 20/50
103/103 - 0s - loss: 8.5672e-04 - mse: 8.5672e-04 - val_loss: 5.1483e-04 - val_mse: 5.1483e-04 - 427ms/epoch - 4ms/step
Epoch 21/50
103/103 - 0s - loss: 8.6189e-04 - mse: 8.6189e-04 - val_loss: 4.9968e-04 - val_mse: 4.9968e-04 - 380ms/epoch - 4ms/step
Epoch 22/50
103/103 - 0s - loss: 8.5941e-04 - mse: 8.5941e-04 - val_loss: 4.9605e-04 - val_mse: 4.9605e-04 - 375ms/epoch - 4ms/step
Epoch 23/50
103/103 - 0s - loss: 8.3976e-04 - mse: 8.3976e-04 - val_loss: 4.9938e-04 - val_mse: 4.9938e-04 - 372ms/epoch - 4ms/step
Epoch 24/50
103/103 - 0s - loss: 8.2392e-04 - mse: 8.2392e-04 - val_loss: 4.8102e-04 - val_mse: 4.8102e-04 - 380ms/epoch - 4ms/step
Epoch 25/50
103/103 - 0s - loss: 8.3040e-04 - mse: 8.3040e-04 - val_loss: 5.3633e-04 - val_mse: 5.3633e-04 - 376ms/epoch - 4ms/step
Epoch 26/50
103/103 - 0s - loss: 8.2367e-04 - mse: 8.2367e-04 - val_loss: 5.1988e-04 - val_mse: 5.1988e-04 - 381ms/epoch - 4ms/step
Epoch 27/50
103/103 - 0s - loss: 8.1775e-04 - mse: 8.1775e-04 - val_loss: 5.0360e-04 - val_mse: 5.0360e-04 - 375ms/epoch - 4ms/step
Epoch 28/50
103/103 - 0s - loss: 8.0505e-04 - mse: 8.0505e-04 - val_loss: 4.6806e-04 - val_mse: 4.6806e-04 - 373ms/epoch - 4ms/step
Epoch 29/50
103/103 - 0s - loss: 8.0057e-04 - mse: 8.0057e-04 - val_loss: 4.7476e-04 - val_mse: 4.7476e-04 - 369ms/epoch - 4ms/step
Epoch 30/50
103/103 - 0s - loss: 8.0176e-04 - mse: 8.0176e-04 - val_loss: 4.7907e-04 - val_mse: 4.7907e-04 - 370ms/epoch - 4ms/step
Epoch 31/50
103/103 - 0s - loss: 7.9466e-04 - mse: 7.9466e-04 - val_loss: 4.8413e-04 - val_mse: 4.8413e-04 - 380ms/epoch - 4ms/step
Epoch 32/50
103/103 - 0s - loss: 7.9054e-04 - mse: 7.9054e-04 - val_loss: 4.9138e-04 - val_mse: 4.9138e-04 - 376ms/epoch - 4ms/step
Epoch 33/50
103/103 - 0s - loss: 7.8007e-04 - mse: 7.8007e-04 - val_loss: 4.9920e-04 - val_mse: 4.9920e-04 - 376ms/epoch - 4ms/step
Epoch 34/50
103/103 - 0s - loss: 7.7809e-04 - mse: 7.7809e-04 - val_loss: 4.8480e-04 - val_mse: 4.8480e-04 - 380ms/epoch - 4ms/step
Epoch 35/50
103/103 - 0s - loss: 7.7086e-04 - mse: 7.7086e-04 - val_loss: 4.6018e-04 - val_mse: 4.6018e-04 - 389ms/epoch - 4ms/step
Epoch 36/50
103/103 - 0s - loss: 7.6884e-04 - mse: 7.6884e-04 - val_loss: 4.5810e-04 - val_mse: 4.5810e-04 - 374ms/epoch - 4ms/step
Epoch 37/50
103/103 - 0s - loss: 7.5240e-04 - mse: 7.5240e-04 - val_loss: 4.6063e-04 - val_mse: 4.6063e-04 - 381ms/epoch - 4ms/step
Epoch 38/50
103/103 - 0s - loss: 7.4910e-04 - mse: 7.4910e-04 - val_loss: 4.5822e-04 - val_mse: 4.5822e-04 - 368ms/epoch - 4ms/step
Epoch 39/50
103/103 - 0s - loss: 7.4235e-04 - mse: 7.4235e-04 - val_loss: 4.4666e-04 - val_mse: 4.4666e-04 - 385ms/epoch - 4ms/step
Epoch 40/50
103/103 - 0s - loss: 7.5361e-04 - mse: 7.5361e-04 - val_loss: 4.4947e-04 - val_mse: 4.4947e-04 - 371ms/epoch - 4ms/step
Epoch 41/50
103/103 - 0s - loss: 7.3891e-04 - mse: 7.3891e-04 - val_loss: 4.5411e-04 - val_mse: 4.5411e-04 - 380ms/epoch - 4ms/step
Epoch 42/50
103/103 - 0s - loss: 7.1854e-04 - mse: 7.1854e-04 - val_loss: 4.3896e-04 - val_mse: 4.3896e-04 - 379ms/epoch - 4ms/step
Epoch 43/50
103/103 - 0s - loss: 7.2564e-04 - mse: 7.2564e-04 - val_loss: 4.3702e-04 - val_mse: 4.3702e-04 - 378ms/epoch - 4ms/step
Epoch 44/50
103/103 - 0s - loss: 7.1769e-04 - mse: 7.1769e-04 - val_loss: 4.7473e-04 - val_mse: 4.7473e-04 - 382ms/epoch - 4ms/step
Epoch 45/50
103/103 - 0s - loss: 7.1356e-04 - mse: 7.1356e-04 - val_loss: 4.4865e-04 - val_mse: 4.4865e-04 - 378ms/epoch - 4ms/step
Epoch 46/50
103/103 - 0s - loss: 7.0446e-04 - mse: 7.0446e-04 - val_loss: 4.2275e-04 - val_mse: 4.2275e-04 - 384ms/epoch - 4ms/step
Epoch 47/50
103/103 - 0s - loss: 6.9736e-04 - mse: 6.9736e-04 - val_loss: 4.2445e-04 - val_mse: 4.2445e-04 - 380ms/epoch - 4ms/step
Epoch 48/50
103/103 - 0s - loss: 6.9928e-04 - mse: 6.9928e-04 - val_loss: 4.1869e-04 - val_mse: 4.1869e-04 - 374ms/epoch - 4ms/step
Epoch 49/50
103/103 - 0s - loss: 6.8223e-04 - mse: 6.8223e-04 - val_loss: 4.1971e-04 - val_mse: 4.1971e-04 - 370ms/epoch - 4ms/step
Epoch 50/50
103/103 - 0s - loss: 6.7743e-04 - mse: 6.7743e-04 - val_loss: 4.8369e-04 - val_mse: 4.8369e-04 - 383ms/epoch - 4ms/step
2022-11-15 20:41:02.847965: W tensorflow/core/grappler/costs/op_level_cost_estimator.cc:690] Error in PredictCost() for the op: op: "Softmax" attr { key: "T" value { type: DT_FLOAT } } inputs { dtype: DT_FLOAT shape { unknown_rank: true } } device { type: "CPU" vendor: "AuthenticAMD" model: "240" frequency: 3194 num_cores: 16 environment { key: "cpu_instruction_set" value: "SSE, SSE2" } environment { key: "eigen" value: "3.4.90" } l1_cache_size: 32768 l2_cache_size: 524288 l3_cache_size: 16777216 memory_size: 268435456 } outputs { dtype: DT_FLOAT shape { unknown_rank: true } }
Test RMSE: 189.232
Test MAPE: 1.764
Test R2: 0.981
Test MAE: 152.840

Process finished with exit code 0
 

🎉3 参考文献

部分理论来源于网络，如有侵权请联系删除。

[1]张立峰,刘旭.基于CNN-GRU神经网络的短期负荷预测[J].电力科学与工程,2020,36(11):53-57

[2]姚程文,杨苹,刘泽健.基于CNN-GRU混合神经网络的负荷预测方法[J].电网技术,2020,44(9):3416-3423

[3]高翱,李国玉,撖奥洋,周生奇,魏振,张智晟.基于Adam算法优化GRU神经网络的短期负荷预测模型[J].电子设计工程,2022,30(09):180-183+188.DOI:10.14022/j.issn1674-6236.2022.09.038.

[4]姚程文,杨苹,刘泽健.基于CNN-GRU混合神经网络的负荷预测方法[J].电网技术,2020,44(09):3416-3424.DOI:10.13335/j.1000-3673.pst.2019.2058.

🌈4 Python代码实现及数据