Privacy-Preserving Detection and Localisation of Behind-the-Meter Low-Carbon Technologies in Low-Voltage Networks

The increasing adoption of behind-the-meter low carbon technologies (BTM LCTs), including photovoltaic (PV) systems, electric vehicles (EVs), and heat pumps (HPs) is transforming low-voltage distribution networks (LVDNs), necessitating real-time update and improved observability of the network. However, many BTM LCT installations remain unregistered, while privacy regulations limit access to individual customer demand data, reducing network observability. This paper proposes a privacy-preserving framework for detecting, classifying and localising newly installed BTM LCTs using UK General Data Protection Regulation (GDPR) compliant measurements, namely branch-level aggregated current and individual customer voltage measurements. The proposed framework comprises three stages: (i) BTM LCT event detection using a residual input-based multivariate Long Short-Term Memory Autoencoder (LSTM-AE); (ii) BTM LCT type classification using a statistical rule-based (SRB) classifier; and (iii) customer-level BTM LCT localisation and rating classification using a distribution network digital twin (DNDT). The framework is validated using 15-minute resolution smart meter data from the Fluvius’ dataset and simulations on the IEEE 33-bus distribution system. The residual input-based LSTM-AE achieves an F1-score of 0.873, outperforming the raw-input model with an F1-score of 0.741. The SRB classifier successfully identifies the tested BTM LCT types, while the DNDT-trained hybrid one-dimensional convolution neural network (1D-CNN) achieves 98% customer-level localisation accuracy under the evaluated network conditions. Furthermore, the proposed detection approach achieves a 100% detection success rate for the higher-rated BTM LCT cases (10 kW PV and 7.2 kW EV charger) across aggregation levels of 5-25 customers.