Next hour load forecast in medium voltage electricity distribution

Purpose – The purpose of this paper is to assess next hour load forecast in medium voltage electricity distribution. Design/methodology/approach – The methodological approach used in this paper, is based on a regressive method – artificial neural network. A real life case study is used for illustrating the defined steps and to discuss the results. Findings – The presence of a de‐regulated environment reinforces the need of short‐term forecast algorithms (STLF). Actions like network management, load dispatch and network reconfiguration under quality of service constraints, require reliable next hour load forecasts. Methodological approaches based on regressive methods such as artificial neural networks are widely used in STLF, with satisfactory results. The construction of an “efficient” artificial neural networks goes through, among other factors, the construction of an “efficient” input vector (IV), in order to avoid over fitting problems and keeping the global simplicity of the model. The explanatory variables normally used, are grouped in two major classes, endogenous and exogenous. The endogenous variables are load values in past instants, and the exogenous variables are normally climatic. The main findings with this kind of vector presents satisfactory results compared to other proposals in the literature. Originality/value – This paper makes use of a procedural sequence for the pre‐processing phase that allows capturing some predominant relations among certain different sets of the available data, providing a more solid basis to decisions regarding the composition of the IV. To deal with load increasing during the winter period, the forecast average daily temperature was used in order to produce an indicator of the daily load average for the forecast day. This information brings more accuracy to the model.