Study of Momentum Optimisation Based on a Combination of Markov Chain and Neural Network Models

In this paper, we investigate the change of ‘momentum’ in complex dynamic systems and its impact on the results based on various machine learning and statistical modelling techniques. Firstly, Factor Analysis is used to identify the key factors, and a combination of Linear Regression Models (LRMs) and Generalised Weighted Score Regression Models (GSRFs) are used to assess the leverage and momentum in the system, thus generating a visual representation of the performance. Then, this paper introduces Markov Chain for probabilistic prediction, and proposes a ‘Momentum-based Markov Chain’ model, which significantly improves the prediction accuracy compared with the traditional Markov Chain, and the mean square error is reduced from 0.06247 to 0.04842, which verifies the importance of momentum in decision making.. Finally, this paper uses gated recurrent unit (GRU) neural network to model the system volatility, and gives weights to different influencing factors through factor analysis, and the root mean square error of the prediction result is 0.06, which demonstrates excellent prediction ability. The study in this paper demonstrates the effective application of factor analysis, regression models, Markov chains and GRU neural networks in complex dynamic systems, providing support for optimisation decisions in related fields.