Abstract:Learning and forecasting stochastic time series is essential in various scientific fields. However, despite the proposals of nonlinear filters and deep-learning methods, it remains challenging to capture nonlinear dynamics from a few noisy samples and predict future trajectories with uncertainty estimates while maintaining computational efficiency. Here, we propose a fast algorithm to learn and forecast nonlinear dynamics from noisy time series data. A key feature of the proposed model is kernel functions applied to projected lines, enabling fast and efficient capture of nonlinearities in the latent dynamics. Through empirical case studies and benchmarking, the model demonstrates its effectiveness in learning and forecasting complex nonlinear dynamics, offering a valuable tool for researchers and practitioners in time series analysis.
What problem does this paper attempt to address?
### What problem does this paper attempt to solve?
This paper aims to address the challenges of learning and predicting nonlinear dynamical systems from noisy time - series data. Specifically, the authors propose a fast algorithm for learning and predicting nonlinear dynamics from noisy time - series data. The key to this method lies in using kernel functions to process projection lines, thereby being able to efficiently capture the nonlinear features in the underlying dynamics. Through empirical case studies and benchmark tests, this model demonstrates its effectiveness in learning and predicting complex nonlinear dynamics, providing a valuable tool for time - series analysis researchers and practitioners.
### Background and motivation of the paper
Time - series analysis is of great significance in multiple scientific fields, such as the return on investment in the stock market or the synchronization analysis of neuronal activities. Although various nonlinear filters and deep - learning methods have been proposed, it is still difficult to capture nonlinear dynamics from a small number of noisy samples, predict future trajectories and estimate uncertainty while maintaining computational efficiency. Therefore, developing a method that can efficiently learn and predict nonlinear dynamics in noisy data has important practical significance.
### Main contributions
1. **Model innovation**: A new nonlinear state - space model (PNL - SS) is proposed. This model uses kernel functions to process projection lines and can efficiently capture the nonlinear features in the underlying dynamics.
2. **Computational efficiency**: By using projection kernel functions, the model is more computationally efficient, especially in high - dimensional data.
3. **Empirical verification**: Through multiple empirical cases and benchmark tests, the effectiveness of the model in learning and predicting complex nonlinear dynamics is verified.
4. **Comparative analysis**: A comparison is made with existing classical methods (such as ARIMA, random forest) and deep - learning methods (such as LSTM, Transformer, NBEATS), demonstrating the superior performance of the model.
### Technical details
- **Model structure**: The PNL - SS model consists of an observation model and a state model. The observation model assumes a linear relationship, while the state model uses a kernel function to describe nonlinear dynamics.
- **Kernel function selection**: The projection kernel function is used to project high - dimensional data onto a one - dimensional line, simplifying distance calculation and improving computational efficiency.
- **Parameter learning**: The expectation - maximization (EM) algorithm is used to optimize model parameters, including the state - transition matrix, bias term, kernel function parameters, etc.
- **Inference method**: Forward and backward iterations are carried out through the moment - matching method to infer the posterior distribution of latent variables.
### Experimental results
- **Nonlinear oscillator**: Experiments were carried out on the Van der Pol oscillator, demonstrating the superior performance of the PNL - SS model in capturing and predicting nonlinear dynamics.
- **Lorenz system**: A comparison was made with the model based on the radial basis function (RBF). The PNL - SS model showed advantages in both computational time and fitting effect.
- **Benchmark test**: In the time - series prediction tasks of multiple chaotic systems, the performance of the PNL - SS model is better than that of various classical and deep - learning methods.
In summary, this paper addresses the challenges of learning and predicting nonlinear dynamics from noisy time - series data by proposing an efficient nonlinear state - space model, providing new tools and methods for time - series analysis.