Probabilistic net load forecasting framework for application in distributed integrated renewable energy systems

Integrating various sectors enhances resilience in distributed sector-integrated energy systems. Forecasting is vital for unlocking full potential and enabling well-informed decisions in energy management. Given the inherent variability in generation and demand prediction, quantification of uncertainty is crucial. Therefore, probabilistic forecasting is becoming imperative compared to deterministic forecasting, as it ensures a more comprehensive depiction of uncertainty. This paper introduces probabilistic net load forecasting framework (PNLFF), a non-blackbox approach that is robust, non-parametric, computational and data inexpensive, and adaptable across sectors. It utilizes the personalized standard load profile for deterministic forecasts, and integrates quantile regression to generate probabilistic forecast. The cumulative distribution function is approximated from quantiles of probabilistic forecast using piecewise cubic hermite interpolating polynomial, and then it is derived to probability density function (PDF). Then the probabilistic net load was obtained by the convolution of PDFs for electricity demand, heat demand and PV generation. A case study demonstrates its application in operational optimization for a distributed energy system of the logistics facility. In the first stage of the PNLFF, the results of the personalized standard load profiles clearly show that they can be applied in all sectors and outperform their respective benchmarks. The second stage, the probabilistic expansion using quantile regression, also performs promisingly across all sectors, with the best results being achieved in particular with a small training data set of 30 days. With the extension of the quantiles and interpolation, it was demonstrated how a PDF can be approximated without prior knowledge of the distribution of the data. The result of the case study demonstrate that the PNL, as an aggregated PDF of the different sectors by convolution, can be used for decision making under uncertainty, e.g. for the planning of flexible loads.