Inferring Daily Gas Consumption from Multiple Nonuniformly Sampled Billing Cycles with Hierarchical Constraints

Local natural gas demand forecasting is essential for local distribution companies (LDCs) to optimize resources, maintain efficiency, and meet regulatory requirements. However, inconsistencies arising from nonuniform billing cycles, hierarchical data structures, and aggregated forecasts often degrade accuracy. This dissertation introduces three methodological advancements to address these issues: the Multi-Source Iterative Shifting Disaggregation (ISD) algorithm, Single-Dimension Hierarchical Reconciliation, and Cross-Temporal Hierarchical Forecast Reconciliation. The ISD algorithm disaggregates multiple, overlapping, nonuniformly sampled time series into a coherent high-frequency signal. Unlike existing methods, ISD iteratively refines estimates through constrained load shifting, improving forecast accuracy. Applied to billing cycle data, ISD reduces the Weighted Mean Absolute Percentage Error (WMAPE) by 1.4–4.3%, while residential consumption disaggregation sees a 4.6–10.4% improvement. To enforce coherence in hierarchical forecasting, Single-Dimension Hierarchical Reconciliation corrects inconsistencies within either spatial or temporal hierarchies, enhancing forecast reliability and reducing bias. Cross-Temporal Hierarchical Forecast Reconciliation extends this approach by integrating both spatial and temporal constraints, ensuring consistency across all levels. This method improves hourly forecast accuracy by 10% and daily accuracy by 3%, outperforming Variance Scaling by 7% and 9%, respectively. This dissertation demonstrates that maintaining hierarchical coherence enhances gas demand forecasting. The proposed methods improve accuracy, reduce error, and provide a scalable framework applicable to broader energy forecasting problems. These advancements offer LDCs practical tools to optimize operations and inform decision-making.