DRL-STAF: A Deep Reinforcement Learning Framework for State-Aware Forecasting of Complex Multivariate Hidden Markov Processes

2026-05-14

AI Summary

This work addresses the challenge in multivariate hidden Markov models (HMMs) of balancing predictive accuracy and interpretable latent states, particularly when handling nonlinear, nonstationary observations and inter-sequence dependencies. To this end, the authors propose the DRL-STAF framework, which innovatively integrates deep reinforcement learning (DRL) into latent state estimation without requiring a predefined state transition structure, thereby mitigating state-space explosion. Concurrently, deep neural networks are employed to model complex observation emission mechanisms, enabling joint prediction with state awareness. Experimental results demonstrate that the proposed method significantly outperforms conventional HMM variants, pure deep learning models, and existing deep learning–HMM hybrid approaches across multiple tasks, achieving both high predictive performance and reliable interpretability of latent states.

Abstract

Forecasting multivariate hidden Markov processes is challenging due to nonlinear and nonstationary observations, latent state transitions, and cross-sequence dependencies. While deep learning methods achieve strong predictive accuracy, they typically lack explicit state modeling, whereas Hidden Markov Models (HMMs) provide interpretable latent states but struggle with complex nonlinear emissions and scalability. To address these limitations, we propose DRL-STAF, a Deep Reinforcement Learning based STate-Aware Forecasting framework that jointly predicts next-step observations and estimates the corresponding hidden states for complex multivariate hidden Markov processes. Specifically, DRL-STAF models complex nonlinear emissions using deep neural networks and estimates discrete hidden states using reinforcement learning, reducing the reliance on predefined transition structures and enabling flexible adaptation to diverse temporal dynamics. In particular, DRL-STAF mitigates the state-space explosion encountered by typical multivariate HMM-based methods. Extensive experiments demonstrate that DRL-STAF outperforms HMM variants, standalone deep learning models, and existing DL-HMM hybrids in most cases, while also providing reliable hidden-state estimates.

Problem

• multivariate hidden Markov processes
• state-aware forecasting
• nonlinear emissions
• latent state estimation
• temporal dynamics

Innovation

• Deep Reinforcement Learning
• State-Aware Forecasting
• Hidden Markov Processes
• Nonlinear Emissions
• State-Space Explosion