Abstract:Time series analysis is relevant in various disciplines such as physics, biology, chemistry, and finance. In this paper, we present a novel neural network architecture that integrates elements from ResNet structures, while introducing the innovative incorporation of the Taylor series framework. This approach demonstrates notable enhancements in test accuracy across many of the baseline datasets investigated. Furthermore, we extend our method to incorporate a recursive step, which leads to even further improvements in test accuracy. Our findings underscore the potential of our proposed model to significantly advance time series analysis methodologies, offering promising avenues for future research and application.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to improve the accuracy of time - series prediction by introducing a new neural network architecture - TaylorNet. Specifically, the paper attempts to solve the following problems: 1. **Limitations of existing models in handling complex time - series data**: - Traditional time - series models (such as ARIMA and exponential smoothing techniques) have difficulty in capturing nonlinear and dynamic spatio - temporal patterns. - Existing deep - learning models (such as RNN and LSTM), although performing well in many applications, still face problems such as vanishing gradients and difficulty in capturing seasonality. 2. **Improving the accuracy of time - series prediction**: - The paper proposes a neural network architecture that combines Taylor series expansion and recursive structure to enhance the modeling ability of time - series data. - By introducing higher - order derivative information (such as second - and third - order derivatives), TaylorNet can more accurately capture complex dynamic changes in time series. 3. **Expanding the application range of the model**: - TaylorNet is not only suitable for traditional univariate time - series analysis but also has broad application potential, including finance, medical, meteorology and other fields. - Through the introduction of recursive steps, the prediction performance of the model is further improved, making it perform excellently on more types of datasets. ### Specific methods - **TaylorNet architecture**: It combines the advantages of the ResNet structure and introduces the idea of Taylor series expansion. By transforming the time - series prediction problem into an approximation of derivatives, the prediction accuracy of the model is improved. \[ \hat{x}_{i + d}=x_{i + d - 1}+\Delta t N_1(\{x_j\}_{j = i}^{i + d - 1})+\frac{\Delta t^2}{2}N_2(\{x_j\}_{j = i}^{i + d - 1}) \] - **Recursive TaylorNet**: By introducing recursive steps, the time step is divided into multiple small steps to gradually predict future values. This is similar to the multi - step method for solving ordinary differential equations in numerical methods, further improving the prediction accuracy. \[ \hat{x}_{j + k + 1}=\hat{x}_{j + k}+\frac{\Delta t}{m}N(\hat{x}_{j + k}) \] ### Experimental results - The experimental results on multiple benchmark datasets show that TaylorNet2 and Recursive TaylorNet2 are superior to other traditional and advanced models (such as ResNet and LSTM) in prediction accuracy. - In particular, Recursive TaylorNet2 performs best in terms of average ranking and median percentage deviation, indicating that it has significant advantages in time - series prediction. ### Conclusion This paper provides a novel and effective time - series prediction method by introducing TaylorNet and its recursive version, which can improve the prediction accuracy in various application scenarios.

Incorporating Taylor Series and Recursive Structure in Neural Networks for Time Series Prediction

TPRNN: A Top-Down Pyramidal Recurrent Neural Network for Time Series Forecasting

Recurrent Neural Networks for Time Series Forecasting: Current Status and Future Directions

Time Series Representation Models

Deep State Space Recurrent Neural Networks for Time Series Forecasting

Neural Decomposition of Time-Series Data for Effective Generalization

Mind the Remainder: Taylor's Theorem View on Recurrent Neural Networks

AdaRNN: Adaptive Learning and Forecasting of Time Series

Mind the Remainder: Taylor’s Theorem View on Recurrent Neural Networks

Time series prediction using deep echo state networks

Network of Tensor Time Series

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

Taylorformer: Probabilistic Modelling for Random Processes including Time Series

A Dual-Stage Attention-Based Recurrent Neural Network for Time Series Prediction

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

Temporal pattern attention for multivariate time series forecasting

Stochastic Recurrent Neural Network for Multistep Time Series Forecasting

Taylor Neural Network for Real-World Image Super-Resolution

Neural Differential Recurrent Neural Network with Adaptive Time Steps

Time Elastic Neural Networks

RWKV-TS: Beyond Traditional Recurrent Neural Network for Time Series Tasks