Graph-Based Multivariate Multiscale Dispersion Entropy: Efficient Implementation and Applications to Real-World Network Data

John Stewart Fabila-Carrasco,Chao Tan,Javier Escudero

2024-05-01

Abstract:We introduce Multivariate Multiscale Graph-based Dispersion Entropy (mvDEG), a novel, computationally efficient method for analyzing multivariate time series data in graph and complex network frameworks, and demonstrate its application in real-world data. mvDEG effectively combines temporal dynamics with topological relationships, offering enhanced analysis compared to traditional nonlinear entropy methods. Its efficacy is established through testing on synthetic signals, such as uncorrelated and correlated noise, showcasing its adeptness in discerning various levels of dependency and complexity.

Combinatorics,Computational Engineering, Finance, and Science,Chaotic Dynamics

What problem does this paper attempt to address?

### Problems the paper attempts to solve

This paper aims to solve the problem of efficient analysis of multivariate time - series data in the framework of graphs and complex networks. Specifically, the paper introduces a new method - **Multivariate Multiscale Graph - based Dispersion Entropy (mvDE G)**, which combines temporal dynamics and topological relationships to provide more advanced analytical capabilities than traditional nonlinear entropy methods.

### Main contributions

1. **Combining temporal and topological dimensions**:
- mvDE G not only focuses on the dynamic characteristics of time series but also takes into account the topological structure in the data, thus providing a more comprehensive perspective for data analysis.

2. **Computational efficiency**:
- By optimizing matrix power operations, the computation time of mvDE G grows linearly with the increase in the number of nodes, while that of traditional classical methods grows exponentially. This gives mvDE G a significant advantage in handling large - scale data sets.

3. **Application verification**:
- The paper verifies the effectiveness and robustness of mvDE G through tests on synthetic signals and real - world data (such as two - phase flow systems and meteorological data). The results show that mvDE G can effectively distinguish different flow patterns and meteorological phenomena, revealing the complexity of data at different scales.

### Method overview

1. **Coarsening process**:
- Segment the original signal of each channel, calculate the average value of each segment, and generate a coarsened signal.

2. **Graph - based multivariate dispersion entropy calculation**:
- Construct an embedding matrix and use a mapping function to classify each element in the matrix into different categories.
- Calculate the relative frequency of each category, and finally calculate mvDE G through the normalized Shannon entropy formula:
\[
\text{mvDE G}(X, m, L, c)=-\frac{1}{\log(c^{m})}\sum_{\pi\in\Pi}p(\pi)\ln p(\pi)
\]

3. **Efficient algorithm implementation**:
- Utilize the properties of matrices and Kronecker products to optimize the calculation of large - matrix powers, significantly reducing the computational complexity.

### Experimental results

1. **Analysis of synthetic signals**:
- By generating time series of multivariate 1/f noise and white Gaussian noise (WGN), the performance of mvDE G under different scale factors is verified. The results show that mvDE G can effectively distinguish different types of noise signals and performs well when dealing with short - time series.

2. **Analysis of correlated noise**:
- Generate time series with different correlation structures to further verify the ability of mvDE G to handle correlated noise. The results show that mvDE G can clearly distinguish time series with different degrees of correlation, while the traditional mvDE method has difficulty doing so.

3. **Comparison of computation time**:
- Compared with the classical mvDE method, mvDE G has a significant computational advantage when dealing with large - scale data sets. Experimental results show that the computation time of mvDE G grows linearly with the increase in sample points and the number of channels, while the classical mvDE method may encounter memory overflow problems.

4. **Real - world applications**:
- Apply mvDE G to analyze meteorological data and industrial two - phase flow systems. The results show that mvDE G can effectively capture the complexity of meteorological parameters and distinguish different flow patterns, demonstrating its potential in practical applications.

### Conclusion

This paper provides an efficient multivariate time - series analysis method that combines temporal and topological dimensions by introducing mvDE G. The experimental results verify the effectiveness and robustness of this method, especially its outstanding performance in handling large - scale data sets and short - time series. mvDE G is expected to become an important tool for complex network data analysis, promoting further research and applications in related fields.