Data-driven optimization and statistical modeling to improve meter reading for utility companies. (September 2022)

• Record Type: Journal Article
• Authors: Sinha Roy, Debdatta; Defryn, Christof; Golden, Bruce; Wasil, Edward
• Is Part Of: Computers & operations research. Volume 145(2022)
• Publication Date: 2022-09
• Subjects: Bayesian statistics -- Integer programming -- Vehicle routing -- Meter reading; Operations research -- Periodicals; Electronic digital computers -- Periodicals; 004.05
• DOI: 10.1016/j.cor.2022.105844
• Languages: English
• ISSNs: 0305-0548

Abstract

Abstract: Utility companies collect usage data from meters on a regular basis. The usage data are collected automatically using radio-frequency identification (RFID) technology. Each meter transmits signals from an RFID tag that are read by a vehicle-mounted reading device within a specified distance. Routing the vehicles can be modeled by a close-enough vehicle routing problem on a street network. In practice, there is uncertainty while reading meters. The signal transmitted by an RFID tag is discontinuous, and the range that each meter can be read is different and stochastic due to weather conditions, surrounding obstacles, interference, and decreasing battery life of the RFID tags. These factors can lead to meters not being read. A vehicle has to be sent at a later time to read the missed meters, and this leads to increased costs for a utility company due to additional operational costs and overtime payments to drivers. Our aim is to address the uncertainty issues of the RFID technology by generating routes that are both cost-effective and robust (we seek to minimize the number of missed reads). We use data analytics, optimization, and Bayesian statistical models to address the uncertainty. Simulation experiments using real data show that the hierarchical Bayesian statistical model gives better results compared to other Bayesian statistical models. Utility companies can potentially integrate the results from the hierarchical Bayesian statistical model into their route generating software as a decision-support tool to produce routes that are more cost-effective and robust than the routes that they currently generate. Highlights: Formulate the stochastic meter reading problem as a two-stage integer program (IP). The two-stage IP formulation is deterministic even though the problem is stochastic. Develop three Bayesian learning models to capture the inherent uncertainty in the data. The hierarchical Bayesian learning model performs better than non-hierarchical models. Develop an iterative algorithmic framework that the utility companies can directly use.