Hierarchical Load Forecast Aggregation for Distribution Transformers Using Minimum Trace Optimal Reconciliation and AMI Data

Abstract

Overloading and load imbalance have a significant impact on the health of distribution transformers. The load of a distribution transformer can be considered in a hierarchical way: individual single-phase customers connected directly to the transformer (the bottom level), the load at each phase (the middle level), and the total load among three phases (the top level). Load at each hierarchical level can be predicted individually, known as “base forecast”, through a state-of-the-art forecasting method, but this practice often leads to incoherency and bias, i.e., forecasts at a lower hierarchical level are not aggregated correctly to the forecast at a higher-hierarchical-level. In this paper, a novel load aggregation technique based on minimum trace (MinT)-based optimal reconciliation is proposed to improve the accuracy of prediction models. Base forecasts at each hierarchical level are firstly determined using independent autoregressive integrated moving average (ARIMA) models; MinT is then used to optimally reconcile base forecasts to ensure higher accuracy. The proposed method is validated by case studies. Advanced metering infrastructure (AMI) data recorded by Saskatoon Light and Power in Saskatoon; Canada is used as historical data in this study.

Keywords

• Distribution transformer
• health monitoring
• hierarchical load forecast aggregation
• minimum trace optimal reconciliation