Multi-scale information fusion time series long-term forecasting model based on neural network

doi:10.11772/j.issn.1001-9081.2024070930

Journal of Computer Applications ›› 2025, Vol. 45 ›› Issue (6): 1776-1783.DOI: 10.11772/j.issn.1001-9081.2024070930

• CCF BigData 2024 • Previous Articles

Multi-scale information fusion time series long-term forecasting model based on neural network

Lanhao LI¹^,², Haojun YAN¹, Haoyi ZHOU¹^,³(), Qingyun SUN¹^,², Jianxin LI¹^,²

^1.School of Computer Science and Engineering，Beihang University，Beijing 100191，China
^2.Beijing Advanced Innovation Center for Big Data and Brain Computing （Beihang University），Beijing 100191，China
^3.School of Software，Beihang University，Beijing 100191，China

Received:2024-07-04 Revised:2024-10-21 Accepted:2024-10-22 Online:2024-12-04 Published:2025-06-10
Contact: Haoyi ZHOU
About author:LI Lanhao， born in 1997， Ph. D. candidate. His research interests include time series data analysis.
YAN Haojun， born in 2003. His research interests include time series data generation.
SUN Qingyun， born in 1996， Ph. D.， assistant professor. Her research interests include graph data mining.
LI Jianxin， born in 1979， Ph. D.， professor. His research interests include big data analysis.
Supported by:
National Natural Science Foundation of China for Distinguished Young Scholars(62225202);Young Scientists Fund of National Natural Science Foundation of China(62202029);Project of the 9th Young Elite Scientists Sponsorship Program by CAST(2023QNRC001)

基于神经网络的多尺度信息融合时间序列长期预测模型

李岚皓¹^,², 严皓钧¹, 周号益¹^,³(), 孙庆赟¹^,², 李建欣¹^,²

^1.北京航空航天大学计算机学院，北京 100191
^2.北京市大数据科学与脑机智能高精尖创新中心（北京航空航天大学），北京 100191
^3.北京航空航天大学软件学院，北京 100191

通讯作者: 周号益
作者简介:李岚皓（1997—），男，辽宁鞍山人，博士研究生，CCF会员，主要研究方向：时序数据分析
严皓钧（2003—），男，湖南永顺人，主要研究方向：时序数据生成
孙庆赟（1996—），女，山东聊城人，助理教授，博士，CCF会员，主要研究方向：图数据挖掘
李建欣（1979—），男，内蒙古赤峰人，教授，博士，CCF杰出会员，主要研究方向：大数据分析。
基金资助:
国家杰出青年科学基金资助项目(62225202);国家自然科学基金青年科学基金资助项目(62202029);第九届中国科协青年人才托举工程项目(2023QNRC001)

Abstract

Abstract:

Time series data come from a wide range of social fields， from meteorology to finance and to medicine. Accurate long-term prediction is a key issue in time series data analysis， processing， and research. Aiming at exploitation and utilization of the correlation of different scales in time series data， a multi-scale information fusion time series long-term forecasting model based on neural network — ScaleNN was proposed with the purpose of better handling multi-scale problem in time series data to achieve more accurate long-term forecast. Firstly， fully connected neural network and convolutional neural network were combined to extract both global and local information effectively， and the two were aggregated for prediction. Then， by introducing a compression mechanism in the global information representation module， longer sequence input was accepted with a lighter structure， which increased perceptual range of the model and improved the model’s performance. Numerous experimental results demonstrate that ScaleNN outperforms the current excellent model in this field — PatchTST （Patch Time Series Transformer） on multiple real-world datasets. In specific， the running time is shortened by 35% with only 19% parameters required. It can be seen that ScaleNN can be applied to time series prediction problems in various fields widely， providing a foundation for forecasting in areas such as traffic flow prediction and weather forecasting.

Key words: time series, big data, data mining, deep learning, series forecasting

摘要：

时间序列数据广泛来源于社会各个领域，从气象学到金融学再到医学，准确的长期预测是时间序列数据分析、处理与研究中的一个关键问题。针对时间序列数据中存在的不同尺度相关性的挖掘与利用，提出一种基于神经网络的多尺度信息融合时间序列长期预测模型ScaleNN，旨在更好地处理时间序列数据中的多尺度问题，从而实现更准确的长期预测。首先，结合全连接神经网络和卷积神经网络，有效提取全局信息与局部信息，并将2种信息聚合后进行预测；其次，通过在全局信息表征模块中引入压缩机制，以更轻量化的结构接受更长的序列输入，增大模型的感知范围并提高模型效能。大量实验结果表明，ScaleNN在多个真实世界数据集上的性能优于当前该领域的优秀模型PatchTST （Patch Time Series Transformer），在运行时间降低35%的同时仅需19%的参数量。可见，ScaleNN可广泛应用于不同领域的时间序列预测问题，为交通流量预测、天气预报等领域提供预测的基础。

关键词: 时间序列, 大数据, 数据挖掘, 深度学习, 序列预测

CLC Number:

TP183

Lanhao LI, Haojun YAN, Haoyi ZHOU, Qingyun SUN, Jianxin LI. Multi-scale information fusion time series long-term forecasting model based on neural network[J]. Journal of Computer Applications, 2025, 45(6): 1776-1783.

李岚皓, 严皓钧, 周号益, 孙庆赟, 李建欣. 基于神经网络的多尺度信息融合时间序列长期预测模型[J]. 《计算机应用》唯一官方网站, 2025, 45(6): 1776-1783.

Figures/Tables 6

References 30

1	BROCKWELL P J， DAVIS R A. Time series： theory and methods， SSS［M］. New York： Springer， 1991： 1-2.
2	ELMUNIM N， ABDULLAH M， HASBI A M， et al. Comparison of statistical Holt-Winter models for forecasting the ionospheric delay using GPS observations［J］. Indian Journal of Radio and Space Physics， 2015， 44（1）： 28-34.
3	RODRIGUES P M M， GOUVEIA P M D C B. An application of PAR models for tourism forecasting ［J］. Tourism Economics， 2004， 10（3）： 281-303.
4	KAUR J， PARMAR K S， SINGH S. Autoregressive models in environmental forecasting time series： a theoretical and application review ［J］. Environmental Science and Pollution Research， 2023， 30（8）： 19617-19641.
5	AGANA N A， HOMAIFAR A. EMD-based predictive deep belief network for time series prediction： an application to drought forecasting ［J］. Hydrology， 2018， 5（1）： No.18.
6	洪永淼，汪寿阳. 大数据、机器学习与统计学：挑战与机遇［J］. 计量经济学报， 2021， 1（1）：17-35.
	HONG Y M， WANG S Y. Big data， machine learning and statistics： challenges and opportunities［J］. China Journal of Econometrics， 2021， 1（1）：17-35.
7	ELMAN J L. Finding structure in time［J］. Cognitive Science， 1990， 14（2）： 179-211.
8	VASWANI A， SHAZEER N， PARMAR N， et al. Attention is all you need ［C］// Proceedings of the 31st International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2017： 6000-6010.
9	POPESCU M C， BALAS V E， PERESCU-POPESCU L， et al. Multilayer perceptron and neural networks［J］. WSEAS Transactions on Circuits and Systems， 2009， 8（7）： 579-588.
10	ZHOU H， ZHANG S， PENG J， et al. Informer： beyond efficient transformer for long sequence time-series forecasting［C］// Proceedings of the 35th AAAI Conference on Artificial Intelligence. Palo Alto： AAAI Press， 2021： 11106-11115.
11	WU H， XU J， WANG J， et al. Autoformer： decomposition transformers with auto-correlation for long-term series forecasting［C］// Proceedings of the 35th International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2021： 22419-22430.
12	ZENG A， CHEN M， ZHANG L， et al. Are Transformers effective for time series forecasting？［C］// Proceedings of the 37th AAAI Conference on Artificial Intelligence. Palo Alto： AAAI Press， 2023： 11121-11128.
13	NIE Y， NGUYEN N H， SINTHONG P， et al. A time series is worth 64 words： long-term forecasting with Transformers ［EB/OL］. ［2024-06-22］..
14	KIM T， KIM J， TAE Y， et al. Reversible instance normalization for accurate time-series forecasting against distribution shift［EB/OL］. ［2024-06-22］..
15	ZHOU T， MA Z， WEN Q， et al. FEDformer： frequency enhanced decomposed Transformer for long-term series forecasting［C］// Proceedings of the 39th International Conference on Machine Learning. New York： JMLR.org， 2022： 27268-27286.
16	TAYLOR S J， LETHAM B. Forecasting at scale［J］. The American Statistician， 2018， 72（1）： 37-45.
17	GRAVES A. Long short-term memory［M］// Supervised sequence labelling with recurrent neural networks， SCI 385. Berlin： Springer， 2012： 37-45.
18	CHUNG J， GULCEHRE C， CHO K， et al. Empirical evaluation of gated recurrent neural networks on sequence modeling［EB/OL］. ［2024-06-22］..
19	LI S， JIN X， XUAN Y， et al. Enhancing the locality and breaking the memory bottleneck of transformer on time series forecasting［C］// Proceedings of the 33rd International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2019： 5243-5253.
20	LIU S， YU H， LIAO C， et al. Pyraformer： low-complexity pyramidal attention for long-range time series modeling and forecasting ［EB/OL］. ［2024-06-22］..
21	CAI W， LIANG Y， LIU X， et al. MSGNet： learning multi-scale inter-series correlations for multivariate time series forecasting［C］// Proceedings of the 38th AAAI Conference on Artificial Intelligence. Palo Alto： AAAI Press， 2024： 11141-11149.
22	WANG H， PENG J， HUANG F， et al. MICN： multi-scale local and global context modeling for long-term series forecasting ［EB/OL］. ［2024-06-22］..
23	DOSOVITSKIY A， BEYER L， KOLESNIKOV A， et al. An image is worth 16x16 words： Transformers for image recognition at scale［EB/OL］. ［2024-06-22］..
24	LIU Z， LIN Y， CAO Y， et al. Swin Transformer： hierarchical vision transformer using shifted windows［C］// Proceedings of the 2021 IEEE/CVF International Conference on Computer Vision. Piscataway： IEEE， 2021： 9992-10002.
25	CIRSTEA R G， GUO C， YANG B， et al. Triformer： triangular， variable-specific attentions for long sequence multivariate time series forecasting［C］// Proceedings of the 31st International Joint Conference on Artificial Intelligence. California： ijcai.org， 2022： 1994-2001.
26	CHEN P， ZHANG Y， CHENG Y， et al. Pathformer： multi-scale Transformers with adaptive pathways for time series forecasting［EB/OL］. ［2024-06-22］..
27	ZHONG S， SONG S， ZHUO W， et al. A multi-scale decomposition MLP-mixer for time series analysis［EB/OL］. ［2024-06-22］..
28	HOU H， YU F R. RWKV-TS： beyond traditional recurrent neural network for time series tasks［EB/OL］. ［2024-06-22］..
29	GONG Z， TANG Y， LIANG J. PatchMixer： a patch-mixing architecture for long-term time series forecasting ［EB/OL］. ［2024-06-22］..
30	ZHOU T， NIU P， WANG X， et al. One fits all： universal time series analysis by pretrained LM and specially designed adaptors［EB/OL］. ［2024-06-22］..

数据集	特征数	时间步	细粒度/min
ETTm1	7	69 680	15
ETTm2	7	69 680	15
ETTh1	7	17 420	60
ETTh2	7	17 420	60
Weather	21	52 696	10
ECL	321	26 304	60

数据集	特征数	时间步	细粒度/min
ETTm1	7	69 680	15
ETTm2	7	69 680	15
ETTh1	7	17 420	60
ETTh2	7	17 420	60
Weather	21	52 696	10
ECL	321	26 304	60

数据集	预测长度	ScaleNN		PatchTST		DLinear		MICN		FEDformer		Autoformer		Pyraformer
数据集	预测长度	MSE	MAE	MSE	MAE	MSE	MAE	MSE	MAE	MSE	MAE	MSE	MAE	MSE	MAE
Weather	96	0.148	0.188	0.149	0.198	0.176	0.237	0.161	0.229	0.238	0.314	0.249	0.329	0.896	0.556
	192	0.190	0.231	0.194	0.241	0.220	0.282	0.220	0.281	0.275	0.329	0.325	0.370	0.622	0.624
	336	0.242	0.283	0.245	0.282	0.265	0.319	0.278	0.331	0.339	0.377	0.351	0.391	0.739	0.753
	720	0.311	0.323	0.314	0.334	0.323	0.362	0.311	0.356	0.389	0.409	0.415	0.426	1.004	0.934
ETTh1	96	0.365	0.395	0.370	0.400	0.375	0.399	0.398	0.427	0.376	0.415	0.435	0.446	0.664	0.612
	192	0.409	0.425	0.413	0.429	0.405	0.416	0.430	0.453	0.423	0.446	0.456	0.457	0.790	0.681
	336	0.423	0.445	0.422	0.440	0.439	0.443	0.440	0.460	0.444	0.462	0.486	0.487	0.891	0.738
	720	0.440	0.459	0.447	0.468	0.472	0.490	0.491	0.509	0.469	0.492	0.515	0.517	0.963	0.782
ETTh2	96	0.273	0.336	0.274	0.336	0.289	0.353	0.299	0.364	0.332	0.374	0.332	0.368	0.645	0.597
	192	0.339	0.378	0.339	0.379	0.383	0.418	0.422	0.441	0.407	0.446	0.426	0.434	0.788	0.683
	336	0.357	0.398	0.329	0.384	0.448	0.465	0.447	0.474	0.400	0.447	0.477	0.479	0.907	0.747
	720	0.393	0.427	0.379	0.422	0.605	0.551	0.442	0.467	0.412	0.469	0.453	0.490	0.963	0.783
ETTm1	96	0.285	0.333	0.290	0.342	0.299	0.343	0.316	0.362	0.326	0.390	0.510	0.492	0.543	0.510
	192	0.326	0.359	0.332	0.369	0.335	0.365	0.363	0.390	0.365	0.415	0.514	0.495	0.557	0.537
	336	0.359	0.381	0.366	0.392	0.369	0.386	0.408	0.426	0.392	0.425	0.510	0.492	0.754	0.655
	720	0.421	0.415	0.416	0.420	0.425	0.421	0.459	0.464	0.446	0.458	0.527	0.493	0.908	0.724
ETTm2	96	0.160	0.245	0.165	0.255	0.167	0.260	0.179	0.275	0.180	0.271	0.205	0.293	0.435	0.507
	192	0.216	0.285	0.220	0.292	0.224	0.303	0.262	0.326	0.252	0.318	0.278	0.336	0.730	0.673
	336	0.266	0.319	0.274	0.329	0.281	0.342	0.305	0.353	0.324	0.364	0.343	0.379	1.201	0.845
	720	0.352	0.375	0.362	0.385	0.397	0.421	0.389	0.407	0.410	0.420	0.414	0.419	3.625	1.451
ECL	96	0.130	0.224	0.129	0.222	0.140	0.237	0.164	0.269	0.186	0.302	0.196	0.313	0.386	0.449
	192	0.147	0.244	0.147	0.240	0.153	0.249	0.177	0.285	0.197	0.311	0.211	0.324	0.386	0.443
	336	0.163	0.263	0.163	0.259	0.169	0.267	0.193	0.304	0.213	0.328	0.214	0.327	0.378	0.443
	720	0.198	0.293	0.197	0.290	0.203	0.301	0.212	0.321	0.233	0.344	0.236	0.342	0.376	0.445

数据集	预测长度	ScaleNN		PatchTST		DLinear		MICN		FEDformer		Autoformer		Pyraformer
数据集	预测长度	MSE	MAE	MSE	MAE	MSE	MAE	MSE	MAE	MSE	MAE	MSE	MAE	MSE	MAE
Weather	96	0.148	0.188	0.149	0.198	0.176	0.237	0.161	0.229	0.238	0.314	0.249	0.329	0.896	0.556
	192	0.190	0.231	0.194	0.241	0.220	0.282	0.220	0.281	0.275	0.329	0.325	0.370	0.622	0.624
	336	0.242	0.283	0.245	0.282	0.265	0.319	0.278	0.331	0.339	0.377	0.351	0.391	0.739	0.753
	720	0.311	0.323	0.314	0.334	0.323	0.362	0.311	0.356	0.389	0.409	0.415	0.426	1.004	0.934
ETTh1	96	0.365	0.395	0.370	0.400	0.375	0.399	0.398	0.427	0.376	0.415	0.435	0.446	0.664	0.612
	192	0.409	0.425	0.413	0.429	0.405	0.416	0.430	0.453	0.423	0.446	0.456	0.457	0.790	0.681
	336	0.423	0.445	0.422	0.440	0.439	0.443	0.440	0.460	0.444	0.462	0.486	0.487	0.891	0.738
	720	0.440	0.459	0.447	0.468	0.472	0.490	0.491	0.509	0.469	0.492	0.515	0.517	0.963	0.782
ETTh2	96	0.273	0.336	0.274	0.336	0.289	0.353	0.299	0.364	0.332	0.374	0.332	0.368	0.645	0.597
	192	0.339	0.378	0.339	0.379	0.383	0.418	0.422	0.441	0.407	0.446	0.426	0.434	0.788	0.683
	336	0.357	0.398	0.329	0.384	0.448	0.465	0.447	0.474	0.400	0.447	0.477	0.479	0.907	0.747
	720	0.393	0.427	0.379	0.422	0.605	0.551	0.442	0.467	0.412	0.469	0.453	0.490	0.963	0.783
ETTm1	96	0.285	0.333	0.290	0.342	0.299	0.343	0.316	0.362	0.326	0.390	0.510	0.492	0.543	0.510
	192	0.326	0.359	0.332	0.369	0.335	0.365	0.363	0.390	0.365	0.415	0.514	0.495	0.557	0.537
	336	0.359	0.381	0.366	0.392	0.369	0.386	0.408	0.426	0.392	0.425	0.510	0.492	0.754	0.655
	720	0.421	0.415	0.416	0.420	0.425	0.421	0.459	0.464	0.446	0.458	0.527	0.493	0.908	0.724
ETTm2	96	0.160	0.245	0.165	0.255	0.167	0.260	0.179	0.275	0.180	0.271	0.205	0.293	0.435	0.507
	192	0.216	0.285	0.220	0.292	0.224	0.303	0.262	0.326	0.252	0.318	0.278	0.336	0.730	0.673
	336	0.266	0.319	0.274	0.329	0.281	0.342	0.305	0.353	0.324	0.364	0.343	0.379	1.201	0.845
	720	0.352	0.375	0.362	0.385	0.397	0.421	0.389	0.407	0.410	0.420	0.414	0.419	3.625	1.451
ECL	96	0.130	0.224	0.129	0.222	0.140	0.237	0.164	0.269	0.186	0.302	0.196	0.313	0.386	0.449
	192	0.147	0.244	0.147	0.240	0.153	0.249	0.177	0.285	0.197	0.311	0.211	0.324	0.386	0.443
	336	0.163	0.263	0.163	0.259	0.169	0.267	0.193	0.304	0.213	0.328	0.214	0.327	0.378	0.443
	720	0.198	0.293	0.197	0.290	0.203	0.301	0.212	0.321	0.233	0.344	0.236	0.342	0.376	0.445

数据集	预测长度	ScaleNN		去掉局部感知模块的ScaleNN		去掉全局感知模块的ScaleNN
数据集	预测长度	MSE	MAE	MSE	MAE	MSE	MAE
Weather	96	0.148	0.188	0.175	0.211	0.181	0.212
	192	0.190	0.231	0.225	0.254	0.229	0.254
	336	0.242	0.283	0.276	0.294	0.281	0.293
	720	0.311	0.323	0.352	0.345	0.354	0.342
ETTh1	96	0.365	0.395	0.392	0.402	0.378	0.393
	192	0.409	0.425	0.447	0.430	0.437	0.424
	336	0.423	0.445	0.487	0.448	0.478	0.440
	720	0.440	0.459	0.480	0.468	0.474	0.459
ETTh2	96	0.273	0.336	0.292	0.340	0.289	0.334
	192	0.339	0.378	0.364	0.385	0.364	0.382
	336	0.357	0.398	0.385	0.407	0.376	0.399
	720	0.393	0.427	0.418	0.436	0.404	0.424
ETTm1	96	0.285	0.333	0.316	0.346	0.311	0.340
	192	0.326	0.359	0.366	0.372	0.369	0.370
	336	0.359	0.381	0.402	0.396	0.397	0.392
	720	0.421	0.415	0.467	0.435	0.465	0.430
ETTm2	96	0.160	0.245	0.177	0.253	0.173	0.250
	192	0.216	0.285	0.243	0.296	0.239	0.294
	336	0.266	0.319	0.301	0.335	0.299	0.333
	720	0.352	0.375	0.396	0.391	0.394	0.390
ECL	96	0.131	0.226	0.170	0.258	0.184	0.263
	192	0.148	0.243	0.190	0.272	0.189	0.271
	336	0.163	0.263	0.206	0.288	0.205	0.286
	720	0.198	0.293	0.246	0.321	0.245	0.319

Multi-scale information fusion time series long-term forecasting model based on neural network

基于神经网络的多尺度信息融合时间序列长期预测模型

RichHTML

PDF

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 6

References 30

Related Articles 15

Recommended Articles

Metrics

[1]	Kai CHEN, Hailiang YE, Feilong CAO. Classification algorithm for point cloud based on local-global interaction and structural Transformer [J]. Journal of Computer Applications, 2025, 45(5): 1671-1676.
[2]	Dan WANG, Wenhao ZHANG, Lijuan PENG. Channel estimation of reconfigurable intelligent surface assisted communication system based on deep learning [J]. Journal of Computer Applications, 2025, 45(5): 1613-1618.
[3]	Wenpeng WANG, Yinchang QIN, Wenxuan SHI. Review of unsupervised deep learning methods for industrial defect detection [J]. Journal of Computer Applications, 2025, 45(5): 1658-1670.
[4]	Xueying LI, Kun YANG, Guoqing TU, Shubo LIU. Adversarial sample generation method for time-series data based on local augmentation [J]. Journal of Computer Applications, 2025, 45(5): 1573-1581.
[5]	Sijie NIU, Yuliang LIU. Auxiliary diagnostic method for retinopathy based on dual-branch structure with knowledge distillation [J]. Journal of Computer Applications, 2025, 45(5): 1410-1414.
[6]	Lihu PAN, Shouxin PENG, Rui ZHANG, Zhiyang XUE, Xuzhen MAO. Video anomaly detection for moving foreground regions [J]. Journal of Computer Applications, 2025, 45(4): 1300-1309.
[7]	Yiding WANG, Zehao WANG, Yaoli LI, Shaoqing CAI, Yuan YUAN. Multi-scale 2D-Adaboost microscopic image recognition algorithm of Chinese medicinal materials powder [J]. Journal of Computer Applications, 2025, 45(4): 1325-1332.
[8]	Yang ZHOU, Hui LI. Remote sensing image building extraction network based on dual promotion of semantic and detailed features [J]. Journal of Computer Applications, 2025, 45(4): 1310-1316.
[9]	Guangju YANG, Tianjian LUO, Kaijun WANG, Siqi YANG. Multi-branch multi-view based contextual contrastive representation learning method for time series [J]. Journal of Computer Applications, 2025, 45(4): 1042-1052.
[10]	Zhenhua XUE, Qiang LI, Chao HUANG. Vision foundation model-driven pixel-level image anomaly detection method [J]. Journal of Computer Applications, 2025, 45(3): 823-831.
[11]	Ruilong CHEN, Tao HU, Youjun BU, Peng YI, Xianjun HU, Wei QIAO. Stacking ensemble adversarial defense method for encrypted malicious traffic detection model [J]. Journal of Computer Applications, 2025, 45(3): 864-871.
[12]	Tianqi ZHANG, Shuang TAN, Xiwen SHEN, Juan TANG. Image watermarking method combining attention mechanism and multi-scale feature [J]. Journal of Computer Applications, 2025, 45(2): 616-623.
[13]	Qianting ZHANG, Liying HU, Lifei CHEN. Robust shapelet representation method for time series [J]. Journal of Computer Applications, 2025, 45(2): 436-443.
[14]	Zirong HONG, Guangqing BAO. Review of radar automatic target recognition based on ensemble learning [J]. Journal of Computer Applications, 2025, 45(2): 371-382.
[15]	Hanlin ZHANG, Junlu WANG, Baoyan SONG. Time series event classification method fused with derived features [J]. Journal of Computer Applications, 2025, 45(2): 428-435.