Mechanism Interpretation Of Key Parameters In Reservoir Computing

Reservoir Computing is a machine learning tool that has been successfully used in chaotic system prediction and hidden variable observation.RC uses a reservoir as a hidden layer,which is a dynamical system that will react to changes in input over time.Since RC is a dynamical system with a specific time scale,it is very suitable for dealing with time series prediction problems.This paper studies the mechanism of some key parameters in RC which affect the final result,and introduces a new measurement method to measure the performance of the reservoir which aims to emphasize learning the overall prediction of the system,not just the short-term prediction.This paper proposes that one of the essential factors that affect the prediction effect of the reservoir is the dynamics of its internal nodes,and it is found that the level of connectivity in the reservoir network has a small effect on its performance.In the end,a mechanicsm interpretation of the impact of key parameters on the performance of the system is obtained from a dynamical perspective.This paper uses a commonly used benchmark reservoir system to perform numerical simulation on the Lorenz system,traverses the parameter space of some key parameters,and gives the trend of the performance changing with the key parameters.The original system can be improved by changing the activation function.Finally,this paper compares the performance of the improved reservoir with the original system and it is found that with other parameters unchanged,the performance of the reservoir can be improved by changing the activation function.