Non-Gaussian Residual Based Short Term Load Forecast Adjustment for Distribution Feeders

Abstract

The evolving role for electricity network operators means that load forecasting at the distribution level has become increasingly important, presenting the need for anticipation of the behavior of highly dynamic and diversely distributed loads. The commonly held assumption of Gaussian residuals in forecasting does not always hold for distribution network loads, increasing the uncertainty in balancing a system at this network level. To reduce the operational impact of forecast errors, this paper utilizes different multivariate joint probability distributions to capture the intra-day dependency structure of forecast residuals. Transforming these to the conditional form enables forecast corrections to be made at variable horizons even in the absence of the forecast model. Improvements in accuracy are demonstrated on benchmark load forecast models at distribution level low voltage substations. A practical distribution system application on scheduling embedded energy storage shows substantial reductions in grid imports and hence costs to distribution level customers from utilizing the proposed intraday correction approach.