The increasing concern for environmental conservation has spurred government initiatives towards energy efficiency. One of the key research areas in this regard is demand response, particularly focusing on differential pricing initiatives such as Time-of-Use (ToU). Differential tariffs are typically designed based on mathematical or statistical models analyzing historical electricity price and consumption data. This study proposes a methodology for identifying time intervals suitable for implementing ToU energy tariffs, achieved by analyzing electric power demand variability to estimate demand flexibility potential. The methodology transforms consumption data into variation via the coefficient of variation and, then, employs k-means data analysis techniques and the a priori algorithm. Tested with real data from smart meters in the Colombian electrical system, the methodology successfully identified time intervals with potential for establishing ToU tariffs. Additionally, no direct relationship was found between external variables such as socioeconomic level, user type, climate, and consumption variability. Finally, it was observed that user behavior concerning consumption variability could be categorized into two types of days: weekdays and non-working days.
In this paper, we propose a novel methodology for estimating missing data in energy consumption datasets. Conventional data imputation methods are not suitable for these datasets, because they are time series with special characteristics and because, for some applications, it is quite important to preserve the shape of the daily energy profile. Our answer to this need is the use of autoencoders. First, we split the problem into two subproblems: how to estimate the total amount of daily energy, and how to estimate the shape of the daily energy profile. We encode the shape as a new feature that can be modeled and predicted using autoencoders. In this way, the problem of imputation of profile data are reduced to two relatively simple problems on which conventional methods can be applied. However, the two predictions are related, so special care should be taken when reconstructing the profile. We show that, as a result, our data imputation methodology produces plausible profiles where other methods fail. We tested it on a highly corrupted dataset, outperforming conventional methods by a factor of 3.7.
This paper presents a methodology developed to perform the processing, analysis, and characterization of AMI measurement data from the substations of three network operators of the Colombian electrical grid. This methodology includes the analysis of the data, which presents the sources of information used by the model, along with the preprocessing and exploratory analysis of the substations data. It also includes the formulation of the data reconstruction method, which uses a constrained optimization model to characterize the substations, based on the different behaviors of the end users of the Colombian electrical grid. In addition to the proposed methodology, the results of its application to the data provided by the operators are provided. These results show the capacity of the proposed methodology to adequately identify the most common behaviors of the users in a given area and characterize most of the energy demand profiles of each substation.
Non-technical losses are difficult to identify from Advanced Metering Infrastructure (AMI) data because fraud has different physical signatures, communication failures can resemble manipulation, and field-confirmed labels are scarce. This paper proposes an evidence-fusion methodology that transforms hourly consumption into traceable inspection events. The method verifies reading provenance, estimates expected profiles from previous observations, evaluates mechanism-specific experts, and consolidates persistent detections under fixed inspection capacity. Because no open hourly AMI dataset with field-confirmed fraud dates was identified, evaluation used controlled episodes on real consumption backgrounds and TDD2022, an open benchmark with synthetic theft events. In a user-disjoint controlled cohort, the policy detected 108 of 150 episodes within seven days (recall 0.720), 17 more than the reference workflow (exact McNemar ). On TDD2022, evidence fusion obtained a recall of 0.689, a precision of 0.998, an F1 score of 0.816, and an area under the precision–recall curve (PR–AUC) of 0.928, with two false alerts among 2455 normal days. The results show that causal and mechanism-specific evidence can produce a capacity-aware inspection list across two controlled benchmarks. Field inspections remain necessary to estimate diagnostic accuracy.