Abstract:We present a probabilistic forecasting framework based on convolutional neural network for multiple related time series forecasting. The framework can be applied to estimate probability density under both parametric and non-parametric settings. More specifically, stacked residual blocks based on dilated causal convolutional nets are constructed to capture the temporal dependencies of the series. Combined with representation learning, our approach is able to learn complex patterns such as seasonality, holiday effects within and across series, and to leverage those patterns for more accurate forecasts, especially when historical data is sparse or unavailable. Extensive empirical studies are performed on several real-world datasets, including datasets from <a class="link-external link-http" href="http://JD.com" rel="external noopener nofollow">this http URL</a>, China's largest online retailer. The results show that our framework outperforms other state-of-the-art methods in both accuracy and efficiency.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the **probabilistic prediction problem of multiple correlated time series**, especially when dealing with a large number of large - scale, complex and correlated time series, how to effectively capture the time - dependence and complex patterns (such as seasonality, holiday effects, etc.), and provide accurate predictions when historical data is scarce or non - existent. Specifically, the paper focuses on the following aspects: 1. **The importance of probabilistic prediction**: Different from traditional point prediction, probabilistic prediction can extract richer information from historical data and better capture future uncertainties. This is crucial for product supply - and - demand prediction in the retail business, logistics distribution planning and other fields. 2. **Challenges of large - scale time series**: In practical applications, it is often necessary to predict thousands or even more correlated time series simultaneously. In addition, the emergence of new products makes some time series lack historical data, which brings additional challenges to prediction. 3. **The influence of exogenous variables**: When predicting demand, it is also necessary to consider the influence of exogenous variables such as promotion plans and weather forecasts. These variables have a significant impact on future prediction results. 4. **Advantages of non - autoregressive models**: Traditional autoregressive models are prone to accumulate errors in multi - step prediction, while non - autoregressive models can avoid this problem and can be parallelized for training, improving computational efficiency. To solve these problems, the paper proposes a non - autoregressive probabilistic prediction framework based on convolutional neural network (CNN) - DeepTCN (Deep Temporal Convolutional Network). This framework captures the long - term dependencies of time series by stacking dilated causal convolutional layers, and combines representation learning to deal with the influence of complex patterns and exogenous variables. Experimental results show that this framework performs excellently on multiple real - world datasets and is superior to the existing state - of - the - art methods. ### Main contributions - Proposed a CNN - based probabilistic prediction framework that supports parametric and non - parametric probability density estimation. - The framework can learn the latent correlations between time series and deal with problems such as data sparsity and cold - start, and is highly scalable and flexible. - The model can flexibly incorporate exogenous covariates, such as promotion plans or weather forecasts. - Extensive experiments prove that this framework is superior to existing methods in both point prediction and probabilistic prediction tasks.

Probabilistic Forecasting with Temporal Convolutional Neural Network

Causal augmented ConvNet: A temporal memory dilated convolution model for long-sequence time series prediction

Biased Temporal Convolution Graph Network for Time Series Forecasting with Missing Values.

Spatial-Temporal Convolutional Transformer Network for Multivariate Time Series Forecasting

Daily Traffic Flow Forecasting Through a Contextual Convolutional Recurrent Neural Network Modeling Inter- and Intra-Day Traffic Patterns.

TSCND: Temporal Subsequence-Based Convolutional Network with Difference for Time Series Forecasting

Adaptive Convolutional Forecasting Network Based on Time Series Feature-Driven

Towards Better Forecasting by Fusing Near and Distant Future Visions

Wavelet-Temporal Neural Network for Multivariate Time Series Prediction

Enhancing Time Series Forecasting: A Hierarchical Transformer with Probabilistic Decomposition Representation

Conditional Local Convolution for Spatio-Temporal Meteorological Forecasting

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

DDTCN: Decomposed dimension time-domain convolutional neural network along spatial dimensions for multiple long-term series forecasting

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

EffiCANet: Efficient Time Series Forecasting with Convolutional Attention

Temporally Folded Convolutional Neural Networks for Sequence Forecasting

A hybrid temporal convolutional network and Prophet model for power load forecasting

Spectral Temporal Graph Neural Network for Multivariate Time-series Forecasting

A Novel Dual-Channel Temporal Convolutional Network for Photovoltaic Power Forecasting

A Multivariate Time Series Prediction Method Based on Convolution-Residual Gated Recurrent Neural Network and Double-Layer Attention

Multivariate Time Series Modeling and Forecasting with Parallelized Convolution and Decomposed Sparse-Transformer