PhD Candidates
Olumide Adeleke
Olumide Adeleke develops federated machine learning models for non-technical loss detection — i.e., electricity theft — in Nigeria's distribution grid, where aggregate annual losses exceed 40% of billed energy in Lagos state alone. At the University of Lagos's Electrical Engineering department, supervised by Prof. Tunde Oladapo, he formulates the detection problem as temporal anomaly identification in 15-minute interval smart meter consumption patterns, using a long short-term memory autoencoder trained in a privacy-preserving federated learning configuration so that raw household data never leaves each distribution transformer's edge computing node. The federated averaging protocol across 12 distribution transformer nodes in a Lagos pilot area converges in 22 communication rounds to within 3% of a centralised training baseline, while the privacy guarantee prevents customer data exposure. The model achieves a precision of 0.82 and recall of 0.76 for flagging meter ID–substation association anomalies confirmed by follow-up field audits — corresponding to a 18% higher detection rate than the current visual inspection programme. A consulting engagement with Eko Electricity Distribution Company (EKEDC) provides both the meter data infrastructure and the field audit validation resources, and Olumide's model has been approved for a 2026 citywide rollout covering 400,000 smart meter endpoints across the Ikorodu and Ikeja distribution business units.
Details
- Advisor
- Prof. Tunde Oladapo
- Field
- Smart Grid Analytics
- Institution
- University of Lagos
- Skills
- federated averaging FedAvg implementationLSTM time-series anomaly detectionPython TensorFlow/PyTorchfield audit correlation analysis
- Thesis
- Anomaly Detection in Nigerian Distribution Grid Electricity Theft Using Federated Learning on Smart Meter Data
- Transition Signals
- consulting with Eko Electricity Distribution Companypresenting at IEEE ISGT-Africa 2026open to power utility analytics roles