The Saudi Electricity Company (SEC) has partnered with King Abdullah University of Science and Technology (KAUST) to leverage the University's expertise in machine learning to slash non-technical losses.
The SEC's incubator has been working with the KAUST Visualization Core Lab (KVL) since the fall of 2018, through its Innovation Energy Incubator, as well as through DistriMod, its innovative and incubation project, which targets all areas where energy loss might occur.
The collaboration aims to apply machine learning and visualization tools to reduce fraud, waste and abuse in the Saudi power sector.
Non-technical losses are losses due to fraud, waste and abuse. Technical losses stem from the physics of generating, transmitting and distributing electricity through the grid. The partnership aims to target non-technical losses in the Saudi power sector as they are significantly higher than those experienced in many European and North American countries, and they represent a significant amount of lost revenue for the SEC.
"Together with KVL, our analysts estimated that the SEC could recover at least SR73 million ($19.4 million) in lost revenue by correcting anomalies identified by KAUST machine learning models," said Khalid Aldossary, SEC's Innovation Energy Incubator manager.
Using five years of SEC billing data from the Riyadh area, KVL scientists also developed minimum viable product models that used machine learning to predict SEC customer electricity usage and to detect anomalous billing transactions.
KVL scientists met with SEC analysts and leadership in the spring of 2019 to discuss the preliminary results of the machine learning models to develop more focused research questions for further collaboration.
"After receiving feedback from SEC on our prototype machine learning models, we used our models to predict customer electricity consumption patterns," said Dr. David R. Pugh, KVL staff scientist. "We then applied advanced data-science techniques to identify customers whose predicted electricity consumption patterns were inconsistent with the physical characteristics of their electricity meters and circuit breakers."
"We were able to validate the model predictions by comparing them with the results of the field survey carried out," stated Yazeed Al-Dligan, DistriMod project leader. "KAUST's machine learning models successfully identified anomalous electricity consumption patterns over 70 percent of the time—compared with an expected 3 percent success rate from randomly surveying customers."