Generation of synthetic benchmark electrical load profiles using publicly available load and weather data

•Synthetic electrical load profiles generation for power networks is proposed.•ANN training using publicly available weather and electrical demand data.•ANN learning from limited datasets is improved by using Bayesian regularization.•ANN prediction extrapolation by incorporation of domain knowledge is proposed.•Analysis presented shows close agreement with actual real load profiles. Electrical load profiles of a particular region are usually required in order to study the performance of renewable energy technologies and the impact of different operational strategies on the power grid. Load profiles are generally constructed based on measurements and load research surveys which are capital and labour-intensive. In the absence of true load profiles, synthetically generated load profiles can be a viable alternative to be used as benchmarks for research or renewable energy investment planning. In this paper, the feasibility of using publicly available load and weather data to generate synthetic load profiles is investigated. An artificial neural network (ANN) based method is proposed to synthesize load profiles for a target region using its typical meteorological year 2 (TMY2) weather data as the input. To achieve this, the proposed ANN models are first trained using TMY2 weather data and load profile data of neighbouring regions as the input and targeted output. The limited number of data points in the load profile dataset and the consequent averaging of TMY2 weather data to match its period resulted in limited data availability for training. This challenge was tackled by incorporating generalization using Bayesian regularization into training. The other major challenge was facilitating ANN extrapolation and this was accomplished by the incorporation of domain knowledge into the input weather data for training. The performance of the proposed technique has been evaluated by simulation studies and tested on three real datasets. Results indicate that the generated synthetic load profiles closely resemble the real ones and therefore can be used as benchmarks.