October 26, 2026
Integrated forecasting and optimization model for field operations in the electric power sector
Rondinele Araujo Carvalho; Patrícia Belfiore Fávero
DOI: 10.22167/2675-6528-202603429
Article derived from a Course Conclusion Work (TCC), with content based on the student’s original work and adapted to the editorial format of the E&S Magazine with the support of the ResumeAI tool, an artificial intelligence solution developed by Instituto Pecege for textual synthesis and organization.
Summary
Field operations management in the electric sector faces daily challenges such as a high volume of service orders, strict regulatory constraints, and variable demand, frequently resulting in unproductivity and deadline violations. To address this issue, an integrated forecasting and optimization model was developed and evaluated for these operations. 321,781 real service orders, collected between December 2024 and April 2026, with an average of 639.7 orders per day, were used, treated in an aggregated and anonymized manner. The methodology combined exploratory data analysis, predictive modeling with SARIMA, and prescriptive prioritization and allocation strategies. The SARIMA model reduced the mean squared error (RMSE) by 48.6% compared to a moving average baseline (177.27 versus 344.66). Prospective scenario simulation demonstrated that the hybrid approach, which prioritized by regulatory deadline and optimized spatially, increased Service Level Agreement (SLA) compliance by 0.83 percentage points over the baseline scenario and reduced estimated average travel by 20%. It was concluded that the model generated simultaneous gains in regulatory compliance and logistical efficiency, replacing reactive decisions with a structured, data-driven process.
Keywords: Resource allocation; Field management; Operations research; SARIMA; Time series.
1. Introduction
Field operations management in the electric power sector represents a complex challenge, characterized by the high daily volume of service orders, the rigidity of regulatory restrictions, and highly variable demand. Utilities rely on a service delivery chain to ensure that disconnections, reconnections, new connections, and emergency services are performed with quality and within the deadlines established for millions of consumers. This dynamic is particularly demanding during the Summer Plan, a period between October and March, when intense rains increase emergency call-outs.
Historically, the decisions for allocation and prioritization of services are made reactively, based on empirical rules or the individual experience of managers. This model, although functional in smaller-scale operations, reveals fragilities in large-scale environments, with geographically dispersed teams and multiple simultaneous restrictions (Gil, 2017). The direct consequence of this approach is low productivity, unnecessary team displacement, and the risk of violating regulatory deadlines, known as Service Level Agreement (SLA), in addition to an increased risk of backlog formation and unproductivity, especially in the critical months of the Summer Plan.
To mitigate these challenges, specialized literature points to the effectiveness of quantitative techniques, such as data analysis, predictive modeling, and operational research, which demonstrate consistent results in improving efficiency and complying with regulatory requirements (Fávero; Belfiore, 2024; Hillier; Lieberman, 2015). The growing maturity of Data Science and Operational Research allows the integration of demand forecasts, geographic information, and operational constraints into more robust decision support systems. Such systems enable the anticipation of critical scenarios, structured prioritization of services, and logical organization of execution sequences, promoting proactive rather than reactive management (Santos et al., 2023). The problem of allocating and sequencing field service orders bears a close analogy to the Vehicle Routing Problems with Time Windows (VRPTW), whose combinatorial nature makes the search for optimal solutions computationally unfeasible in real operational contexts, justifying the use of heuristic approaches and structured prescriptions (Hillier; Lieberman, 2015; Arenales et al., 2007).
In this context, the research proposes an integrated forecasting and optimization model for field operations in the electricity sector. The model combines exploratory data analysis, predictive modeling using Seasonal Autoregressive Integrated Moving Average (SARIMA), and prescriptive prioritization and allocation strategies. 321,781 actual service orders were used, covering the period from December 2024 to April 2026, with an average of 639.7 service orders per day, treated in an aggregated and anonymized manner.
Given the need to improve regulatory compliance and logistical efficiency, the present work aimed to develop and evaluate an integrated forecasting and optimization model for field operations in the electric sector.
2. Material and Methods
The research adopted a quantitative and applied design, according to Creswell’s (2016) guidelines, to investigate field operations management in the electric sector. The study was based on a database that gathered historical records of service orders from real operations. In compliance with confidentiality and ethical standards, the data were treated in an aggregated and anonymized manner, without any identification of institutions, people, or individual values, following the Manual of Instructions and Norms for Course Conclusion Works of USP/Esalq.
The unit of analysis consisted of 321,781 actual service orders, collected over the period from December 2024 to April 2026, with a daily average of 639.7 service orders. For each record, information was available such as the opening and execution date and time of the service order, geographic coordinates (latitude and longitude), service type, execution status, and the regulatory service level agreement (SLA).
Data preprocessing was performed in a Python environment, using the Pandas library for the steps of reading, cleaning, filtering, and transformation. Records with a null `ordem_servico` field were excluded from the analysis. The operational time variables, `duracao` (activity execution time in the field) and `tempo_de_deslocamento` (travel time between consecutive service calls), were processed to allow independent analyses of execution efficiency and logistics.
The exploratory analysis aimed to map operational behavior over the studied period. Patterns of daily volume, the distribution of orders by service type, and seasonal variations were investigated. Aggregating records by date allowed for the calculation of daily service volume and the identification of weekly seasonality, as well as the annual variation associated with the Summer Plan.
For the predictive modeling of the daily volume of service orders, the Seasonal Autoregressive Integrated Moving Average (SARIMA) model was employed, given the presence of weekly seasonality identified in the exploratory analysis (Fávero; Belfiore, 2024). SARIMA combined autoregressive, moving average, and seasonal components, suitable for series with regular cyclical patterns. A seven-day moving average was used as a baseline for performance comparison.
The performance of the SARIMA model was evaluated by the mean absolute percentage error (MAPE) and the root mean squared error (RMSE), after partitioning the data into training (80%) and testing (20%) sets, a procedure recommended for evaluating the generalization capability of predictive models (Hastie et al., 2009). The selection of SARIMA model parameters (p, d, q)(P, D, Q)[s] followed the Akaike Information Criterion (AIC), which penalizes model complexity and minimizes the risk of overfitting.
The stationarity of the series was verified by the Augmented Dickey-Fuller (ADF) test, where a p-value lower than 0.05 confirmed the absence of a unit root, dispensing with additional differencing beyond the order d = 1 already incorporated. The adequacy of the fit was validated by the Ljung-Box test on the residuals, which did not reject the hypothesis of absence of residual autocorrelation, a necessary condition for the validity of the predictions generated by the model.
The operational risk estimate was performed to capture differences in the criticality level of service orders. A risk indicator was calculated based on the historical rate of unproductivity by service type, assigning a higher weight to orders with a higher frequency of non-completion on the first attempt, influencing the prioritization process.
The prescriptive stage focused on structuring an optimization process for the allocation and prioritization of services, seeking three simultaneous objectives: maximizing compliance with regulatory deadlines (SLA), reducing geographic travel of teams, and increasing overall operational efficiency. The strategy combined explicit prioritization rules with a nearest neighbor type spatial sequencing heuristic, where orders with critical deadlines were allocated first, and the others sequenced by geographic proximity, calculated by the Euclidean distance between coordinates.
From the perspective of Operations Research, the prescriptive model was structured with decision variables (the execution sequence by team), an objective function (simultaneous maximization of SLA compliance and minimization of total travel time), and constraints (maximum service capacity per team per day, ANEEL deadlines by service type, and geographical availability of teams) (Hillier; Lieberman, 2015). Due to the combinatorial nature of the problem, which makes exact resolution unfeasible for the daily volume of service orders, a constructive heuristic approach was adopted, a consolidated practice in the operational optimization literature (Arenales et al., 2007).
To evaluate the impact of the proposed strategies in relation to current practice, three operational scenarios were simulated: (i) Base Scenario, with sequencing according to current practice and without differentiation by regulatory deadline; (ii) Efficiency Scenario, with sequencing exclusively by geographic proximity; and (iii) SLA-first (Hybrid) Scenario, where service orders with critical deadlines were allocated first, and the remaining ones sequenced by proximity. The simulation was prospective, modeling the sequential execution of up to 50 service orders per day per team, with an average time of 30 minutes per service order and 15 minutes of base travel, using parameters calibrated by real operational data, including the historical execution rate and ANEEL’s regulatory deadlines by type of service.
3. Results and Discussion
The research revealed an integrated forecasting and optimization model capable of transforming field operations management in the electric sector, replacing reactive approaches with a structured, data-driven process. The findings demonstrate significant gains in regulatory compliance and logistical efficiency, validating the application of data analysis and operations research techniques. Initial exploratory analysis provided an in-depth understanding of demand behavior, while predictive modeling with SARIMA offered robust anticipation capabilities. Finally, scenario simulation confirmed the effectiveness of prescriptive prioritization and allocation strategies, highlighting the superiority of the proposed hybrid approach.
Exploratory data analysis
The exploratory data analysis revealed significant variability in the daily volume of service calls throughout the studied period. The database comprised a total of 321,781 service order records, collected between December 2024 and April 2026, with an overall average of 639.7 orders per day. The majority of these service calls were concentrated in essential services such as cutting, reconnection, and new connections, reflecting the critical nature of field operations in the electricity sector. This initial characterization was fundamental to understanding the operational dynamics and the challenges inherent in daily management.
A relevant seasonal variation was identified, especially between October and March, a period known as the Summer Plan. During this interval, which coincides with a higher incidence of rain and emergency activations, the daily average of service calls decreased to 550.4 service orders per day, in contrast to the average of 689.1 orders per day observed in the regular period, from April to September. This decrease in service capacity during the Summer Plan indicates a significant operational restriction and a high risk of backlog formation, particularly in activities with shorter regulatory deadlines, which underscores the need for proactive strategies.
The distribution of operational statuses of service orders also pointed to a considerable portion of services that were not concluded on the first contact. This unproductive rate, which varied according to the type of service and the time of year, showed higher rates during the Summer Plan. This finding reinforced the importance of incorporating the historical non-completion rate as a fundamental criterion in the prioritization process of services, aiming to mitigate the risks of non-compliance with regulatory deadlines and optimize resource allocation.
Demand Forecasting Results
To predict the daily volume of service orders, the Seasonal Autoregressive Integrated Moving Average (SARIMA) model was developed and compared to a seven-day moving average baseline. After daily aggregation of service orders, the data were split into 402 days for training and 101 days for testing. The Mean squared error (RMSE) was the primary evaluation metric, as the Mean Absolute Percentage Error (MAPE) was discarded due to the presence of days with zero volume, which would distort the results by division by zero.
The SARIMA model demonstrated superior performance, achieving an RMSE of 177.27, compared to the 344.66 RMSE of the moving average baseline. This performance represents a 48.6% reduction in error, indicating a significantly greater ability to anticipate the daily volume of service orders. The selection of SARIMA model parameters was guided by the Akaike Information Criterion (AIC), and the stationarity of the series was confirmed by the Augmented Dickey-Fuller test, with a p-value less than 0.05, validating the absence of a unit root.
The adequacy of the model fit was further validated by the Ljung-Box test on the residuals, which did not reject the hypothesis of absence of residual autocorrelation, an essential condition for the reliability of forecasts. SARIMA proved particularly effective in capturing the seasonal drop in demand characteristic of the December to February quarter, a period when the volume of emergencies tends to increase due to rainfall. This forecasting accuracy is crucial for management, allowing for proactive adjustment of team allocation and reduction of the risk of non-compliance with regulatory deadlines, as highlighted by Silver (2012).
Prescriptive optimization
The prescriptive optimization stage was structured to achieve three simultaneous objectives: maximizing regulatory deadline compliance (SLA), reducing teams’ geographical displacements, and increasing overall operational efficiency. To this end, a set of explicit prioritization rules was combined with a *nearest neighbor* type spatial sequencing heuristic. Service orders with critical deadlines were prioritized and allocated first, while the others were sequenced based on geographical proximity, calculated by the Euclidean distance between coordinates.
The criticality of the service orders was determined by a risk indicator based on the historical rate of unproductivity by service type. Orders with a higher frequency of non-completion on the first attempt received a higher weight in the prioritization process, ensuring that resources were directed primarily to the most sensitive services. This systematic approach contrasts with the reactive decisions historically adopted, which often resulted in unproductivity and inefficient dispatch.
From the perspective of Operations Research, the prescriptive model was formally structured with decision variables that define the execution sequence of orders by each field team. The objective function sought the simultaneous maximization of SLA compliance and the minimization of total travel time. The constraints included the maximum service capacity per team per day, ANEEL’s regulatory deadlines by service type, and the geographic availability of the teams (Hillier; Lieberman, 2015). The combinatorial complexity of solving this model exactly for a daily volume of approximately 640 service orders and dozens of teams justified the adoption of a constructive heuristic approach, which sacrifices global optimality in favor of viable and computationally efficient solutions (Arenales et al., 2007).
SLA constraint scenario simulation
To evaluate the impact of the proposed strategies in relation to current practice, three distinct operational scenarios were simulated. The Base Scenario represented the current practice, without differentiation by regulatory deadline. The Efficiency Scenario focused on sequencing exclusively by geographic proximity. Finally, the SLA-first, or Hybrid, Scenario prioritized service orders with critical deadlines and sequenced the remaining ones by proximity. The simulation parameters were calibrated with real operational data, including a historical execution rate of 50.93% and ANEEL’s regulatory deadlines for different types of service, such as 4 hours for reconnection/emergency, 8 hours for disconnection, 48 hours for new connection, and 72 hours for metering.
The prospective simulation modeled the sequential execution of up to 50 service orders per team per day, considering an average time of 30 minutes per service order and 15 minutes of base travel time. The resulting indicators revealed that the SLA-first Scenario presented the highest deadline compliance, reaching 39.80% of SLA met, which represents a gain of 0.83 percentage points over the Base Scenario, which recorded 38.97%. The Efficiency Scenario, in turn, achieved 39.70% of SLA met, with a gain of 0.73 percentage points compared to the base scenario.
In terms of logistical efficiency, both optimization scenarios, Efficiency and SLA-first, resulted in a significant reduction in the teams’ average daily travel. Total travel was reduced by 20%, from 750 minutes per day in the Base Scenario to 600 minutes per day in the optimized scenarios. This 20% reduction in travel time has direct and substantial economic implications, as in operations with multiple active teams, each minute saved per team per day translates into accumulated gains in fuel costs, fleet depreciation, and team hours freed up for additional productive service calls (Lima, 2023).
Although the absolute gain in SLA compliance is modest, reflecting the operational capacity bottleneck imposed by the combination of short regulatory deadlines (4 to 8 hours for 55% of service orders) and the time required for sequential execution, the hybrid strategy proved to be the most effective. The comparison between the Efficiency Scenario and the SLA-first Scenario confirmed that prioritization by regulatory urgency should precede spatial sequencing, as the hybrid approach achieved slightly superior SLA performance. Structured SLA prioritization also contributes to reducing contractual exposure to regulatory penalties, mitigating operational risks arising from non-compliance with ANEEL deadlines.
The results obtained are in line with the specialized literature, which points to the effectiveness of combining structured prioritization with logistical criteria in promoting simultaneous improvements in compliance and efficiency (Hillier; Lieberman, 2015; Santos et al., 2023). The research also showed that the main limitation to SLA compliance is team service capacity, suggesting that future extensions of the model could incorporate workforce sizing decisions. From a managerial perspective, even an initial version of the integrated model, with relatively simple heuristics, is capable of generating immediate operational value, allowing for anticipation of workforce adjustments and increasing regulatory compliance and team productivity.
In summary, the integrated forecasting and optimization model demonstrated its ability to generate simultaneous gains in regulatory compliance and logistical efficiency in the field operations of the electric sector. Exploratory analysis detailed seasonality and operational challenges, the SARIMA model provided demand forecasts with a 48.6% reduction in RMSE, and scenario simulation confirmed that the hybrid approach of prioritizing by regulatory deadline and spatial optimization increased SLA compliance by 0.83 percentage points and reduced average travel by 20%. These findings validate the replacement of reactive decisions with a structured, data-driven process, promoting more proactive and efficient management.
4. Conclusion
This work sought to develop and evaluate an integrated forecasting and optimization model for field operations in the electricity sector, with the objective of improving regulatory compliance and logistical efficiency. It was found that the SARIMA model demonstrated superior predictive capability, reducing the mean squared error (RMSE) by 48.6% compared to the moving average baseline, which allowed for robust anticipation of the daily volume of service orders. Prospective scenario simulation showed that the hybrid approach, which combined prioritization by regulatory deadline with spatial optimization, increased Service Level Agreement (SLA) compliance by 0.83 percentage points over the base scenario and reduced the estimated average team travel by 20%. These findings confirm the feasibility of replacing reactive decisions with a structured, data-driven process, generating simultaneous gains in regulatory compliance and logistical efficiency, in addition to offering immediate managerial value by allowing proactive workforce adjustments and mitigating risks of regulatory penalties.
Despite the observed gains, it was identified that the main limitation to SLA compliance lies in the operational capacity of team service, influenced by the combination of short regulatory deadlines and the time required for sequential order execution. As future extensions, it is suggested to incorporate workforce sizing decisions into the model, explore more sophisticated predictive algorithms, such as Random Forest and XGBoost, to capture non-linear relationships, and include ANEEL’s financial and regulatory indicators in the scenario evaluation. It is also recommended to evolve towards a real-time decision support system, integrated with operational databases, which would broaden the model’s applicability to other distributors and regions.
Bibliographic References
Arenales, M.; Armentano, V.; Morabito Neto, R.; Yanasse, H. 2007. Pesquisa operacional. Elsevier, Rio de Janeiro, RJ, Brasil.
Creswell, J.W. 2016. Projeto de pesquisa: métodos qualitativo, quantitativo e misto. 5ed. Penso, Porto Alegre, RS, Brasil.
Fávero, L.P.; Belfiore, P. 2024. Análise de dados: estatística e modelagem multivariada com Excel, SPSS, Stata, R e Python. 2ed. GEN | LTC, Rio de Janeiro, RJ, Brasil.
Gil, A.C. 2017. Métodos e técnicas de pesquisa social. 7ed. Atlas, São Paulo, SP, Brasil.
Hastie, T.; Tibshirani, R.; Friedman, J. 2009. The elements of statistical learning: data mining, inference, and prediction. 2ed. Springer, New York, NY, USA.
Hillier, F.S.; Lieberman, G.J. 2015. Introduction to operations research. 10ed. McGraw-Hill Education, New York, NY, USA.
Lima, F.G. 2023. Análise de riscos. 3ed. Atlas, São Paulo, SP, Brasil.
Santos, M.; Gomes, C.F.S.; Moreira, M.A.L.; Costa, I.P.A. 2023. Ferramentas computacionais de apoio à tomada de decisão. 1ed. Atlas, São Paulo, SP, Brasil.
Silver, N. 2012. The signal and the noise: why so many predictions fail but some don’t. Penguin Press, New York, NY, USA.
Article originating from the Final Course Project of the Specialization in Data Science and Analytics of the MBA USP/Esalq
To learn more about the course, click here and access the MBX Academy platform