Abstract
The growing influence of digitalization and the widespread availability
of data from electrical systems have unlocked significant potential for
data-driven analysis. This has led to substantial opportunities for
enhancing power system performance, reliability, and overall operational
efficiency.
One concrete illustration of these data-driven analyses involves
assessing and predicting harmonic distortion using statistical learning
models, notably linear regression.
The implementation of linear regression models may present various
challenges, especially when dealing with time series data, which is the
case of harmonic measurements too. One of the main issues is the
presence of autocorrelation within error terms. In this situation, the
conventional linear regression approach (ordinary least square) tends to
lose its reliability due to the introduction of biased standard errors
impacting the accuracy of the analysis.
Experts in statistics have developed various methods to handle such
situations, where the error terms show autocorrelation. Interestingly,
these methods have been somehow ignored or perhaps sparsely discussed by
power system engineers.
The primary objective of this paper is to contribute to narrowing this
gap. It does so by presenting the application of the Cochrane-Orcutt
method in the harmonic data analysis context. The Cochrane-Orcutt method
represents a well-known econometric technique used to consider serial
correlations within the error terms of a linear model and can also be
applied in the harmonic analysis domain.
This paper is supported at first by an illustrative example based on
simulated data, designed to provide a clear insight into the subject
matter. Additionally, a case study with real-world measurement data is
also presented, to further enhance the understanding.