We propose an innovative framework for estimating energy consumption trajectories in smart grid networks by adapting the bonus-malus system concept to a stochastic model and integrating a regularized Expectation-Maximization (EM) algorithm. Our approach directly addresses two critical operational challenges: maintaining grid stability during peak demand periods through incentive-driven consumption shaping, and reducing operational costs via optimized demand response. By modeling consumption as a discrete-state Markov chain with exponential tilting, we achieve a 23% reduction in high-consumption transitions during critical load intervals - equivalent to potential cost savings of $18/MWh based on typical peak pricing schemes. The framework’s robust handling of incomplete data (30% missing values) ensures reliable operation under real-world sensor failure scenarios, a key requirement for grid resilience. Numerical simulations demonstrate monotonic convergence in under 30 iterations, making the algorithm practical for minute-scale grid adjustments. This work bridges theoretical guarantees with operational practicality, offering system operators a dual solution for stability enhancement and cost optimization through adaptive Markovian demand management.