Smart Meter Data-Driven Voltage Forecasting Model for a Real Distribution Network Based on SCO-MLP

Advanced metering infrastructure like smart meter technology has enabled the collection of high-resolution data on voltage, active, and reactive power consumption from end-users in real-time. This paper introduces a new machine learning model, named Single Candidate Optimizer (SCO) - Multi-layer perceptron (MLP), for accurate node voltage forecasting in low voltage (LV) distribution networks with high penetrations of low-carbon technologies. The proposed model utilizes historical active and reactive power measurements in one-minute resolution from smart meters to predict node voltage time series values without requiring the network’s electrical model topology and parameters. The computational performance of the MLP framework is improved with the SCO algorithm, which reduces the number of required iterations while maintaining accuracy. The model’s performance is evaluated with numerical metrics and compared against Particle Swarm optimization (PSO) and Differential Evolution (DE)-based models, revealing that the proposed model outperforms both, exhibiting a promising voltage forecasting capability with an average deviation of 1.296 volts relative to the measured values. Overall, this study demonstrates the potential of machine learning and smart meter data for enhancing the stability and efficiency of LV distribution networks.