We propose a Relaxed Checkpoint algorithm (R-Checkpoint) to solve Multi-Event Expert Query Parametric Estimation (ME-EQPE) problems over multivariate time series. Our proposed algorithm combines the strengths of both domain-knowledge-based and formal-learning-based approaches to learn decision parameters for yielding a reasonable time utility over multivariate time series. More specifically, our approach solves the decision optimization problems to yield the time utility from multiple decision time points, as well as learns the multiple sets of decision parameters in their respective events during the computations at a lower cost. We show that our approach produces a reasonable forecasting result by using the learned multiple sets of decision parameters.