Abstract:Multivariate time series forecasting is an important machine learning problem across many domains, including predictions of solar plant energy output, electricity consumption, and traffic jam situation. Temporal data arise in these real-world applications often involves a mixture of long-term and short-term patterns, for which traditional approaches such as Autoregressive models and Gaussian Process may fail. In this paper, we proposed a novel deep learning framework, namely Long- and Short-term Time-series network (LSTNet), to address this open challenge. LSTNet uses the Convolution Neural Network (CNN) and the Recurrent Neural Network (RNN) to extract short-term local dependency patterns among variables and to discover long-term patterns for time series trends. Furthermore, we leverage traditional autoregressive model to tackle the scale insensitive problem of the neural network model. In our evaluation on real-world data with complex mixtures of repetitive patterns, LSTNet achieved significant performance improvements over that of several state-of-the-art baseline methods. All the data and experiment codes are available online.

What problem does this paper attempt to address?

### Problems the paper attempts to solve This paper aims to solve a key challenge in multivariate time - series prediction, that is, how to capture and utilize long - term and short - term repetitive patterns in time - series data. Specifically: 1. **Multivariate time - series prediction**: In many practical applications, such as energy output prediction in solar power stations, power consumption prediction, and traffic congestion prediction, time - series data usually contains dynamic dependencies among multiple variables. The time - series data in these applications often mix long - term and short - term repetitive patterns. 2. **Limitations of existing methods**: - **Autoregressive models (AR) and Gaussian processes (GP)**: Traditional methods such as autoregressive models and Gaussian processes perform poorly when dealing with this mixed pattern because they cannot clearly distinguish and model these two patterns and their interactions. - **Deep learning methods**: Although existing deep learning methods (such as RNN and CNN) have made significant progress in some tasks, they still have some problems when dealing with time - series data, such as the problem of being insensitive to input scale. 3. **The proposed new framework**: To overcome these limitations, the authors propose a new deep - learning framework - Long - and Short - Term Time - Series Network (LSTNet). LSTNet combines Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) to extract short - term local - dependent patterns and discover long - term trend patterns. In addition, a traditional autoregressive model is introduced to solve the problem that neural network models are insensitive to input scale. ### Main contributions 1. **Multi - scale pattern capture**: LSTNet can capture both short - term and long - term patterns in time - series data simultaneously, thereby improving prediction accuracy. 2. **Model robustness**: By combining the traditional autoregressive model, LSTNet is more robust when dealing with cases where the input signal scale changes significantly. 3. **Experimental verification**: Experiments on multiple real - world data sets show that LSTNet significantly outperforms several existing state - of - the - art baseline methods in performance. ### Conclusion This paper proposes a new deep - learning framework LSTNet, which is specifically designed for multivariate time - series prediction tasks, especially when there are long - term and short - term repetitive patterns in the data. By combining CNN, RNN, and autoregressive models, the experimental results of LSTNet on multiple benchmark data sets show that it has significant performance advantages.

Modeling Long- and Short-Term Temporal Patterns with Deep Neural Networks

A novel hybrid deep learning time series forecasting model based on long-short-term patterns

LSTTN: A Long-Short Term Transformer-based Spatio-temporal Neural Network for Traffic Flow Forecasting

Effective LSTMs with Seasonal-Trend Decomposition and Adaptive Learning and Niching-Based Backtracking Search Algorithm for Time Series Forecasting

LSTTN: A Long-Short Term Transformer-based spatiotemporal neural network for traffic flow forecasting

Multivariate Temporal Convolutional Network: A Deep Neural Networks Approach for Multivariate Time Series Forecasting

A Neural Networks Based Method for Multivariate Time-Series Forecasting

Wavelet-Temporal Neural Network for Multivariate Time Series Prediction

A Deep Neural Network Model for Short-Term Load Forecast Based on Long Short-Term Memory Network and Convolutional Neural Network

TSARNet :Integrated Spatio-Temporal Attention Mechanism for Multivariate Time Series Prediction Deep Learning Framework

Spatiotemporal Self-Attention-Based LSTNet for Multivariate Time Series Prediction

A Multi-Granularity Hierarchical Network for Long- and Short-Term Forecasting on Multivariate Time Series Data

TLNets: Transformation Learning Networks for long-range time-series prediction

A TCN-Linear Hybrid Model for Chaotic Time Series Forecasting

Time Series Forecasting and Classification Models Based on Recurrent with Attention Mechanism and Generative Adversarial Networks

Short-Term Load Forecasting Using Multidimensional Long Time Series: A GMM- LSTNET Approach

Temporal pattern attention for multivariate time series forecasting

TCLN: A Transformer-based Conv-LSTM Network for Multivariate Time Series Forecasting

Temporal Attention Based TCN-BIGRU Model for Energy Time Series Forecasting

LSTM Network: a Deep Learning Approach for Short-Term Traffic Forecast

STDNet: A Spatio-Temporal Decomposition Neural Network for Multivariate Time Series Forecasting