Xin Li,Qunxi Zhu,Chengli Zhao,Xiaojun Duan,Bolin Zhao,Xue Zhang,Huanfei Ma,Jie Sun,Wei Lin
Abstract:Abstract Recently, machine learning methods, including reservoir computing (RC), have been tremendously successful in predicting complex dynamics in many fields. However, a present challenge lies in pushing for the limit of prediction accuracy while maintaining the low complexity of the model. Here, we design a data-driven, model-free framework named higher-order Granger reservoir computing (HoGRC), which owns two major missions: The first is to infer the higher-order structures incorporating the idea of Granger causality with the RC, and, simultaneously, the second is to realize multi-step prediction by feeding the time series and the inferred higher-order information into HoGRC. We demonstrate the efficacy and robustness of the HoGRC using several representative systems, including the classical chaotic systems, the network dynamical systems, and the UK power grid system. In the era of machine learning and complex systems, we anticipate a broad application of the HoGRC framework in structure inference and dynamics prediction.
What problem does this paper attempt to address?
### Problems the Paper Attempts to Solve
This paper aims to address the challenge of improving the accuracy of dynamic predictions for complex systems while maintaining low model complexity. Specifically, the researchers designed a data-driven, model-free framework called Higher-order Granger Reservoir Computing (HoGRC), which has two main tasks:
1. **Higher-order Structure Inference**: Inferring higher-order structures by combining the concept of Granger causality with Reservoir Computing (RC).
2. **Multi-step Prediction**: Inputting time series data and inferred higher-order information into HoGRC to achieve multi-step predictions.
### Background and Motivation
In recent years, machine learning methods, including Reservoir Computing, have achieved great success in dynamic predictions across various fields. However, the current challenge lies in further improving prediction accuracy while maintaining low model complexity. Traditional RC methods, although effective, still face difficulties when dealing with high-dimensional, nonlinear, and complex structured systems. Therefore, developing a sophisticated machine learning method that can reveal internal interactions and utilize these interactions for future evolution predictions has become crucial.
### Methods and Innovations
To address the above issues, the researchers proposed the HoGRC framework, with the main innovations including:
1. **Higher-order Structure Embedding**: By introducing higher-order Granger causality, HoGRC can embed higher-order structural information into Reservoir Computing. This not only enhances the model's predictive capability but also enables it to handle higher-dimensional and more complex dynamic systems.
2. **Iterative Optimization**: The HoGRC framework continuously optimizes higher-order neighbors through an iterative process, gradually improving prediction performance. In this process, the inference of Granger causality and dynamic prediction complement and enhance each other.
3. **Parallel Processing**: The HoGRC framework simultaneously achieves complex structure inference and accurate dynamic prediction at the node level, making it suitable for high-dimensional and more complex dynamic systems.
### Experimental Validation
The researchers used multiple representative systems to validate the effectiveness and robustness of the HoGRC framework, including classic chaotic systems, network dynamic systems, and the UK power grid system. Experimental results show that the HoGRC framework performs excellently in structure inference and dynamic prediction, especially when dealing with systems with strong nonlinearity and complex structures.
### Conclusion
The HoGRC framework proposed in this paper significantly improves the accuracy of dynamic predictions for complex systems while maintaining low model complexity. The dual functionality of structure inference and dynamic prediction makes this framework widely applicable in the fields of machine learning and complex systems.