基于二次分解的混合神经网络蜂窝流量预测

doi:10.12305/j.issn.1001-506X.2025.05.31

摘要/Abstract

摘要：

在移动通信网络快速发展的背景下, 蜂窝流量预测对于网络规划、优化和资源管理具有重大意义。针对蜂窝流量数据的复杂性和非线性特点, 提出一种基于二次分解的混合神经网络蜂窝流量预测方法。首先, 采用自适应噪声的完备集合经验模式分解(complete ensemble empirical mode decomposition with adaptive noise, CEEMDAN)方法将原始流量分解为多个子序列, 利用K-Shape聚类算法重构为频率序列和趋势序列。为了更细致地揭示数据的内在结构, 运用变分模态分解(variational mode decomposition, VMD)方法对频率序列进行二次分解, 生成多维频率序列。然后, 将一维趋势序列和多维频率序列分别输入至局部特征提取模块, 其中单通道特征提取层利用一维卷积神经网络(one-dimensional convolution neural network, 1DCNN)提取一维趋势序列的局部特征, 而多通道特征提取层则结合卷积块注意力模块(convolutional block attention module, CBAM)捕捉多维频率序列中的关键信息。紧接着将提取到的特征向量分别输入到时序信息学习模块中, 利用双向长短时记忆(bidirectional long short term memory, BiLSTM)网络和注意力机制学习时序变化规律, 完成预测流量的输出。最后, 通过对趋势序列和频率序列的预测结果求和, 实现对蜂窝流量的准确预测。为了验证所提方法的有效性, 利用公开数据集进行实验验证, 并与多种不同方法进行对比。实验结果表明, 所提预测方法展现出更优的预测性能, 为蜂窝网络的智能管理和优化提供了有力支持。

关键词: 蜂窝流量预测, 模态分解, 卷积神经网络, 双向长短时记忆网络, 卷积块注意力模块

Abstract:

In the context of the rapid development of mobile communication networks, cellular traffic prediction plays an important role in network planning, optimization and resource management. Taking the complexity and nonlinear characteristics of cellular traffic data into account, a hybrid neural network cellular traffic prediction method based on quadratic decomposition is proposed. The complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) method is first used to decompose the original traffic into multiple subsequences, and reconstruct them into frequency and trend sequences by using the K-Shape clustering algorithm. To reveal the underlying structure of data in more detail, the variational mode decomposition (VMD) method is used to perform secondary decomposition on the frequency sequence and generate multi-dimensional frequency sequences. Subsequently, the one-dimensional trend sequence and multi-dimensional frequency sequence are separately input into the local feature extraction module. The single-channel feature extraction layer uses the one-dimensional convolution neural network (1DCNN) method to extract local features of the one-dimensional trend sequence, while the multi-channel feature extraction layer combines the convolutional block attention module (CBAM) method to capture key information in the multi-dimensional frequency sequence. Then we input the extracted feature vectors into the temporal information learning module, and use the bidirectional long short term memory (BiLSTM) model and attention mechanism to complete the output prediction. Finally, we achieve accurate prediction of cellular traffic by summing the predicted results of frequency and trend sequences. To verify the effectiveness of the proposed method, this paper conducts experimental verification by using publicly available datasets and compares the predicted results with various methods. The experimental results show that the proposed method exhibits better predictive performance, which can provide strong support for the management and optimization of cellular networks.

Key words: cellular traffic prediction, modal decomposition, convolutional neural network (CNN), bidirectional long short term memory (BiLSTM) network, convolutional block attention module (CBAM)

中图分类号:

TN929.5

段阿敏, 张朝辉. 基于二次分解的混合神经网络蜂窝流量预测[J]. 系统工程与电子技术, 2025, 47(5): 1687-1697.

Amin DUAN, Zhaohui ZHANG. Quadratic decomposition-based cellular traffic prediction with hybrid neural network[J]. Systems Engineering and Electronics, 2025, 47(5): 1687-1697.

图/表 16

图1

图2

图3

图4

图5

图6

图7

图8

表1

表2

表3

图9

图10

表4

图11

表5

参考文献 31

1	Ericsson. Ericsson mobility report June 2024[R]. Stockholm: Ericsson, 2024.
2	Huawei. Communications networks 2030[R]. China: Huawei, 2024.
3	JIANG W W . Cellular traffic prediction with machine learning: a survey[J]. Expert Systems with Applications, 2022, 201, 117163.
4	YULE G U . Ⅶ. On a method of investigating periodicities disturbed series, with special reference to Wolfer's sunspot numbers[J]. Philosophical Transactions of the Royal Society A, 1927, 226 (636/646): 267- 298.
5	TIAN M , SUN C , WU S Z . An EMD and ARMA-based network traffic prediction approach in SDN-based internet of vehicles[J]. Wireless Networks, 2021, doi: 10.1007/s11276-021-02675-2
6	XU F L , LIN Y Y , HUANG J X , et al. Big data driven mobile traffic understanding and forecasting: a time series approach[J]. IEEE Trans.on Services Computing, 2016, 9 (5): 796- 805.
7	WANG J. A process level network traffic prediction algorithm based on ARIMA model in smart substation[C]//Proc. of the IEEE International Conference on Signal Processing, Communication and Computing, 2013.
8	MEHDIZADEH S , FATHIAN F , ADAMOWSKI J F . Hybrid artificial intelligence-time series models for monthly streamflow modeling[J]. Applied Soft Computing, 2019, 80, 873- 887.
9	韩驰, 熊伟. 基于改进灰狼算法优化SVR的航天侦察装备效能评估[J]. 系统工程与电子技术, 2021, 43 (10): 2902- 2910. doi: 10.12305/j.issn.1001-506X.2021.10.25
	HAN C , XIONG W . Operational effectiveness evaluation of space reconnaissance equipment based on SVR optimized by improved grey wolf optimizer[J]. Systems Engineering and Electronics, 2021, 43 (10): 2902- 2910. doi: 10.12305/j.issn.1001-506X.2021.10.25
10	陈静, 张昭冲, 王琳凯, 等. 基于卷积长短时记忆网络的短时公交客流量预测[J]. 系统仿真学报, 2024, 36 (2): 476- 486.
	CHEN J , ZHANG Z C , WANG L K , et al. Short-term bus passenger flow prediction based on convolutional long-short-term memory network[J]. Journal of System Simulation, 2024, 36 (2): 476- 486.
11	NIE L S, JIANG D D, YU S, et al. Network traffic prediction based on deep belief network in wireless mesh backbone networks[C]//Proc. of the IEEE Wireless Communications and Networking Conference, 2017.
12	ZHANG C T , ZHANG H X , YUAN D F , et al. Citywide cellular traffic prediction based on densely connected convolutional neural networks[J]. IEEE Communications Letters, 2018, 22 (8): 1656- 1659.
13	ZHANG C Y, PATRAS P. Long-term mobile traffic forecasting using deep spatio-temporal neural networks[C]//Proc. of the 18th ACM International Symposium on Mobile Ad Hoc Networking and Computing, 2018: 231-240.
14	XU Y , YIN F , XU W J , et al. Wireless traffic prediction with scalable Gaussian process: framework, algorithms, and verification[J]. IEEE Journal on Selected Areas in Communications, 2019, 37 (6): 1291- 1306.
15	闫啸家, 梁伟阁, 张钢, 等. 基于RCNN-ABiLSTM的机械设备剩余寿命预测方法[J]. 系统工程与电子技术, 2023, 45 (3): 931- 940. doi: 10.12305/j.issn.1001-506X.2023.03.35
	YAN X J , LIANG W G , ZHANG G , et al. Prediction method for mechanical equipment based on RCNN-ABiLSTM[J]. Systems Engineering and Electronics, 2023, 45 (3): 931- 940. doi: 10.12305/j.issn.1001-506X.2023.03.35
16	AGARAP A F M. A neural network architecture combining gated recurrent unit (GRU) and support vector machine (SVM) for intrusion detection in network traffic data[C]//Proc. of the 10th International Conference on Machine Learning and Computing, 2018: 26-30.
17	HUANG C W, CHIANG C T, LI Q H. A study of deep learning networks on mobile traffic forecasting[C]//Proc. of the IEEE 28th Annual International Symposium on Personal, Indoor, and Mobile Radio Communications, 2017.
18	BI J , ZHANG X , YUAN H T , et al. A hybrid prediction method for realistic network traffic with temporal convolutional network and LSTM[J]. IEEE Trans.on Automation Science and Engineering, 2021, 19 (3): 1869- 1879.
19	WANG D , BAO Y Y , WANG C M . A hybrid deep learning method based on CEEMDAN and attention mechanism for network traffic prediction[J]. IEEE Access, 2023, 11, 39651- 39663.
20	时维国, 国明. 基于MEEMD-PE与CS-WNN模型的网络时延预测[J]. 系统工程与电子技术, 2020, 42 (1): 184- 190. doi: 10.3969/j.issn.1001-506X.2020.01.25
	SHI W G , GUO M . Network delay prediction based on model of modified ensemble empirical mode decomposition-permutation entropy and cuckoo search-wavelet neural network[J]. Systems Engineering and Electronics, 2020, 42 (1): 184- 190. doi: 10.3969/j.issn.1001-506X.2020.01.25
21	薛锡瑞, 黄树彩, 韦道知, 等. 基于EMD-DESN的无人机集群航迹目的地预测[J]. 系统工程与电子技术, 2023, 46 (1): 290- 299. doi: 10.12305/j.issn.1001-506X.2024.01.33
	XUE X R , HUANG S C , WEI D Z , et al. Destination prediction of UAV cluster trajectory based on EMD-DESN[J]. Systems Engineering and Electronics, 2023, 46 (1): 290- 299. doi: 10.12305/j.issn.1001-506X.2024.01.33
22	CHANG T J , LEE T S , YANG C T , et al. A ternary-frequency cryptocurrency price prediction scheme by ensemble of clustering and reconstructing intrinsic mode functions based on CEEMDAN[J]. Expert Systems with Applications, 2023, 233, 121008.
23	WANG J Y , DAI B L , ZHANG T H , et al. A novel hybrid model of CEEMDAN-CNN-SAGU for Shanghai copper price prediction[J]. IEEE Access, 2024, 12, 25176- 25187.
24	ZHOU F T , HUANG Z H , ZHANG C H . Carbon price forecasting based on CEEMDAN and LSTM[J]. Applied Energy, 2022, 311, 118601.
25	吴宗收, 汪立新, 沈强, 等. 基于CEEMD半球谐振陀螺输出预测方法[J]. 系统工程与电子技术, 2023, 45 (10): 3259- 3264. doi: 10.12305/j.issn.1001-506X.2023.10.30
	WU Z S , WANG L X , SHEN Q , et al. Output prediction method of hemispherical resonator gyro based on CEEMD[J]. Systems Engineering and Electronics, 2023, 45 (10): 3259- 3264. doi: 10.12305/j.issn.1001-506X.2023.10.30
26	CHEN L Y , LIU X , ZENG C , et al. Temperature prediction of seasonal frozen subgrades based on CEEMDAN-LSTM hybrid model[J]. Sensors, 2022, 22 (15): 5742.
27	韩莹, 王乐豪, 王淑梅, 等. 宽度-深度融合时频分析的径流智能预测方法[J]. 系统仿真学报, 2024, 36 (2): 363- 372.
	HAN Y , WANG L H , WANG S M , et al. Runoff intelligent prediction method based on broad-deep fusion time-frequency analysis[J]. Journal of System Simulation, 2024, 36 (2): 363- 372.
28	WOO S H, PARK J C, LEE J Y, et al. Cbam: convolutional block attention module[C]//Proc. of the European Conference on Computer Vision, 2018: 3-19.
29	PAPARRIZOS J, GRAVANO L. K-shape: efficient and accur ate clustering of time series[C]//Proc. of the ACM SIGMOD International Conference on Management of Data, 2015: 1855-1870.
30	DRAGOMIRETSKIY K , ZOSSO D . Variational mode decomposition[J]. IEEE Trans.on Signal Processing, 2013, 62 (3): 531- 544.
31	GUO D , XIA X W , ZHU L , et al. Dynamic modification neural network model for short-term traffic prediction[J]. Procedia Computer Science, 2021, 187, 134- 139.

分量	A	B	C
IMF₁	3	1	0
IMF₂	3	0	4
IMF₃	1	0	4
IMF₄	1	4	1
IMF₅	1	4	1
IMF₆	2	4	2
IMF₇	2	3	3
IMF₈	0	3	3
IMF₉	0	2	3
IMF₁₀	0	2	3

序列	A	B	C
L₁	IMF₁+IMF₂	IMF₁	IMF₁
L₂	IMF₃+IMF₄+IMF₅	IMF₂+IMF₃	IMF₂+IMF₃
L₃	IMF₆+IMF₇	IMF₄+IMF₅+IMF₆	IMF₄+IMF₅
L₄	IMF₈+IMF₉+IMF₁₀	IMF₇+IMF₈	IMF₆
L₅	-	IMF₉+IMF₁₀	IMF₇+IMF₈+IMF₉+IMF₁₀

序列	A	B	C
L₁	0.91	0.93	0.90
L₂	0.87	0.65	0.65
L₃	0.15	0.25	0.24
L₄	0.12	0.12	0.20
L₅	-	0.10	0.12

模型参数	参数设置
学习率	0.001
批尺寸	48
召回率	0.3
优化器	Adam
步长	24
BiLSTM神经元数	96
BiLSTM层数	3
1DCNN层数	2
1DCNN卷积核	2
迭代次数	500
激活函数	ReLU

算法	A			B			C
算法	RMSE	MAPE	R2	RMSE	MAPE	R2	RMSE	MAPE	R2
ARIMA	0.271 48	0.272 13	0.862 46	0.162 15	0.132 82	0.914 56	0.167 62	0.160 28	0.919 65
SVR	0.123 14	0.250 48	0.901 67	0.113 58	0.114 15	0.920 41	0.104 35	0.142 77	0.923 68
LSTM	0.102 21	0.207 09	0.920 41	0.088 42	0.105 24	0.952 65	0.070 20	0.121 34	0.959 87
RCNN-ABiLSTM	0.092 43	0.162 14	0.937 04	0.075 40	0.101 78	0.965 28	0.063 12	0.132 87	0.965 18
ST-LSTM	0.081 35	0.081 18	0.962 19	0.065 86	0.030 10	0.975 38	0.053 93	0.031 24	0.980 18
CEEMDAN-TGA	0.080 47	0.076 74	0.970 62	0.059 57	0.026 64	0.981 04	0.054 52	0.025 18	0.977 53
所提方法	0.075 72	0.059 12	0.978 47	0.050 14	0.020 96	0.987 60	0.050 51	0.017 88	0.982 60