Eliseo Zarate-Perez

Optimal sizing and design of renewable microgrids and hybrid renewable energy systems require balancing renewable resource variability, demand uncertainty, storage operation, reliability, and techno-economic constraints. Artificial intelligence and metaheuristic optimization strategies have been increasingly used to address these challenges; however, the evidence remains methodologically heterogeneous. This scoping review maps the literature on artificial intelligence, learning-based, metaheuristic, heuristic, and optimization-based strategies for renewable microgrid sizing and design. The review followed PRISMA-ScR guidelines. Searches were conducted in Scopus and the Web of Science Core Collection for research articles published between 2009 and March 2026. A total of 69 studies were included. Metaheuristics dominated the field, appearing in 63 studies, with particle swarm optimization and genetic algorithm-based strategies as the most frequent methodological families. Artificial intelligence and learning-based strategies were mainly used to support forecasting, surrogate modeling, uncertainty handling, and energy management. The most recurrent configurations involved photovoltaic, wind, and battery storage systems, often with diesel backup in stand-alone or off-grid contexts. The literature is strongly oriented toward metaheuristic sizing of PV–wind–battery microgrids, with emerging integration of AI-assisted prediction and decision-support strategies. Future studies should address reproducibility, uncertainty modeling, real-world validation, degradation assessment, explainability, and scalability.

Background: Convenience stores face volatile demand and a direct trade-off between stock-outs and overstocking, both of which affect service levels and profitability. This study aims to optimize inventory management through a reproducible forecasting-and-optimization workflow, assessing its impact on return on investment (ROI) and operational metrics, such as fill rate and stockouts. Methods: The workflow integrates daily, store-level transactions with external covariates, constructs temporal and lag features, and trains a Random Forest (RF) model using chronological splitting and time-series validation. Daily forecasts are then aggregated to the monthly level and used as inputs to an inventory simulation and an ROI-based economic model. Building on this simulation, a Genetic Algorithm (GA) optimizes the parameters of a monthly replenishment policy, incorporating minimum-coverage constraints. Results: In testing, the forecasting model achieved a mean absolute percentage error (MAPE) below 13%, and the RF+GA scheme outperformed the 28-day moving average baseline (MA28) in ROI across all five stores, with an average improvement of 4.52 percentage points; statistical significance was confirmed using the Wilcoxon test. Conclusions: Overall, the RF+GA approach serves as a decision-support tool that generates monthly order quantities consistent with demand and operational constraints, delivering verifiable improvements in both economic and service metrics.

La transformación digital de los procesos industriales ha abierto una nueva etapa en la manera de comprender, controlar y optimizar los sistemas productivos. La incorporación de inteligencia artificial, aprendizaje automático y modelos predictivos permite superar progresivamente esquemas de control basados exclusivamente en mecanismos convencionales, haciendo posible que los datos generados durante la producción se conviertan en una fuente de conocimiento para anticipar desviaciones, reducir la variabilidad y mejorar la calidad. Esta evolución adquiere especial relevancia en industrias altamente reguladas, como la farmacéutica, donde pequeñas variaciones pueden repercutir directamente en las características del producto final. En este escenario se sitúa el presente libro, que aborda la aplicación de redes neuronales artificiales a uno de los problemas particularmente sensibles de la producción farmacéutica: la precisión del peso durante el envasado de polvos galénicos. La naturaleza física de estos productos, las condiciones del proceso y el funcionamiento de los sistemas de dosificación introducen múltiples factores de variabilidad. Por ello, alcanzar un peso uniforme no constituye únicamente una cuestión de eficiencia productiva, sino también un componente relacionado con la calidad, el cumplimiento de las especificaciones y la confiabilidad del producto farmacéutico. La obra parte de la convergencia entre dos campos que tradicionalmente han seguido trayectorias diferentes. Por un lado, se encuentra el desarrollo de las redes neuronales artificiales y de los modelos de aprendizaje capaces de identificar relaciones complejas a partir de grandes conjuntos de datos. Por otro, se encuentra la ingeniería de los procesos de dosificación y envasado, donde variables mecánicas, eléctricas, ambientales y operativas interactúan continuamente. La integración de ambas perspectivas permite plantear una cuestión de gran interés científico y tecnológico: hasta qué punto un sistema puede aprender del comportamiento histórico de un proceso industrial y utilizar ese conocimiento para mejorar su desempeño futuro. A lo largo de sus capítulos se construye progresivamente la respuesta a esta interrogante. Primero se examinan los fundamentos de las redes neuronales artificiales, el aprendizaje automático y los modelos predictivos, prestando especial atención a las arquitecturas recurrentes y a las redes LSTM, cuya capacidad para procesar información secuencial las convierte en alternativas especialmente interesantes para sistemas dinámicos. Posteriormente, el análisis se traslada hacia los polvos galénicos, los sistemas de dosificación, la precisión del peso, la variabilidad y la capacidad de los procesos productivos. El desarrollo de la obra sigue una secuencia deliberadamente progresiva. Primero se explica cómo una red neuronal transforma las variables de entrada en una salida y de qué manera se comprueba su aprendizaje. Después se examina el proceso físico de dosificación, considerando las propiedades del polvo, el tornillo sinfín, la instrumentación, la variabilidad y el control estadístico. Finalmente, ambos campos convergen en el caso de estudio, donde los registros reales se convierten en datos de entrenamiento y se comparan seis arquitecturas. Este recorrido permite avanzar de los conceptos hacia su aplicación sin presuponer que el lector domina previamente la inteligencia artificial o el control de calidad. La articulación de estos fundamentos encuentra su expresión práctica en el caso de estudio presentado en la parte aplicada de la obra. A partir de datos obtenidos durante el envasado de bicarbonato de sodio en presentación de 50 gramos, se analiza el comportamiento del proceso, se identifican sus principales variables operativas y se evalúan diferentes arquitecturas de redes neuronales. Este recorrido permite pasar de la observación de un proceso industrial real a la construcción y evaluación de una alternativa predictiva orientada a mejorar su precisión. Uno de los aportes más interesantes del trabajo reside precisamente en esta transición entre el dato y la decisión. Los registros históricos dejan de concebirse únicamente como evidencia de lo que ocurrió durante la producción y pasan a constituirse en insumos para aprender patrones, realizar predicciones y anticipar comportamientos. Bajo esta perspectiva, la inteligencia artificial no reemplaza los principios tradicionales del control de calidad ni de la ingeniería de procesos; los complementa mediante nuevas capacidades analíticas. Los resultados obtenidos permiten además reflexionar sobre las posibilidades que ofrecen las redes neuronales recurrentes para procesos industriales caracterizados por información secuencial y variabilidad operacional. La comparación entre diferentes modelos y la posterior selección de una arquitectura RNN-LSTM muestran que la elección de una herramienta de inteligencia artificial debe responder a las características particulares del fenómeno estudiado y no únicamente a la disponibilidad de tecnologías cada vez más sofisticadas. Conviene precisar desde el inicio que RNN, LSTM y RNN-LSTM no son expresiones equivalentes. Una RNN simple conserva un estado oculto que resume parte de la información previa; una LSTM es una unidad recurrente especializada que administra la memoria mediante compuertas; y una RNN-LSTM es una red recurrente cuya memoria se implementa mediante celdas o capas LSTM. La tesis compara RNN y LSTM como alternativas y, posteriormente, describe una arquitectura seleccionada con capas LSTM, por lo que...

The electrification of rural medical centers in high Andean areas represents a critical challenge for equitable development due to limited access to reliable energy. Hybrid Renewable Energy Systems (HRESs), which combine solar photovoltaic generation, Battery Energy Storage Systems (BESSs), and backup diesel generators, are emerging as viable solutions to ensure the supply of critical loads. However, their effective implementation requires optimal sizing methodologies that consider multiple technical and economic constraints and objectives. In this study, an optimization model based on metaheuristic algorithms is developed, specifically, Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Ant Colony Optimization (ACO), to identify optimal configurations of an HRES applied to a remote medical center in the Peruvian Andes. The results show that GA achieved the lowest Life Cycle Cost (LCC), with a high share of renewable energy (64.04%) and zero Energy Not Supplied (ENS) defined as the amount of load demand not met by the system, significantly outperforming PSO and ACO. GA was also found to offer greater stability and operational robustness. These findings confirm the effectiveness of metaheuristic methods for designing efficient and resilient energy solutions adapted to isolated rural contexts.

Purpose: This study develops a decision support model based on a genetic algorithm (GA) to select renewable energy technologies considering sustainability criteria. The model integrates environmental, technical, and economic indicators to support energy planning in complex and uncertain scenarios. Design/methodology/approach: The methodology implements a genetic algorithm to optimize the configuration of renewable technologies in a residential energy system. The evaluation incorporates real demand profiles, as well as standardized indicators such as carbon footprint, land use, operating and maintenance costs (LCOE), capacity factor, and reliability. The evaluation also includes a battery energy storage system (BESS). Sensitivity analyses were performed using the Morris and Sobol methods. The statistical significance and stability of the solutions were assessed using the non-parametric Friedman test and the Jaccard index, respectively. Findings: The findings of the study indicated that wind energy and hybrid systems with BESS achieved the most balanced performance under different weighting schemes. The model demonstrated convergence in fewer than 50 generations, yielding consistent solutions across multiple iterations. The findings revealed that environmental criteria exerted the greatest influence on the results, and statistical tests confirmed significant differences between the evaluation approaches. Originality/value: This study presents a robust and adaptable tool for the sustainable assessment of energy technologies, particularly useful in contexts where traditional decision-making methods are limited by complexity or a lack of structured information.

Purpose: This study evaluates the performance of genetic algorithms (GAs) in optimizing the sizing of wind photovoltaic systems with battery energy storage systems (BESS). The objective is to determine whether GAs can balance generation and storage, improve system autonomy, reduce operating costs and meet residential energy demand in real-world environments. Design/methodology/approach: A genetic algorithm was used as a multi-objective optimization tool to determine the optimal sizing of solar panels, wind turbines and BESS. The model considers energy demand, climate variability and resource intermittency. Metrics such as autonomy, total costs and total energy deficit (TED) were evaluated. The results were compared with the exhaustive search to validate the effectiveness of the GA. Findings: The GA identified an optimal configuration with a TED of 3.76%, an autonomy of 96.24% and an efficiency of 95% and a total cost of USD 7,500. In contrast, the exhaustive search achieved a TED of 4.3%, a range of 95.7% and an efficiency of 90% at a cost of USD 8,000. Although both methods ensure optimal performance, GA stands out for its computational efficiency and ability to balance multiple targets. Originality/value: This study not only highlights the usefulness of GAs for designing hybrid microgrids that address renewable resource intermittency and economic viability but also contributes to the sustainable development goal by promoting sustainable and affordable solutions for residential communities.

The variability of renewable energy sources, storage limitations, and fluctuations in residential demand affect the reliability of sustainable energy systems, resulting in energy deficits and the risk of service interruptions. Given this situation, the objective of this study is to diagnose and optimize the reliability of a residential microgrid based on photovoltaic and wind power generation and battery energy storage systems (BESSs). To this end, genetic algorithms (GAs) and ant colony optimization (ACO) are used to evaluate the performance of the system using metrics such as loss of load probability (LOLP), loss of supply probability (LPSP), and availability. The test system consists of a 3.25 kW photovoltaic (PV) system, a 1 kW wind turbine, and a 3 kWh battery. The evaluation is performed using Python-based simulations with real consumption, solar irradiation, and wind speed data to assess reliability under different optimization strategies. The initial diagnosis shows limitations in the reliability of the system with an availability of 77% and high values of LOLP (22.7%) and LPSP (26.6%). Optimization using metaheuristic algorithms significantly improves these indicators, reducing LOLP to 11% and LPSP to 16.4%, and increasing availability to 89%. Furthermore, optimization achieves a better balance between generation and consumption, especially in periods of low demand, and the ACO manages to distribute wind and photovoltaic generation more efficiently. In conclusion, the use of metaheuristics is an effective strategy for improving the reliability and efficiency of autonomous microgrids, optimizing the energy balance and operating costs.

This study proposes a sustainable, innovative approach to improving the structural resilience and energy performance of low-rise housing in seismic zones. It analyzes the structural and thermal behavior of residential buildings constructed with the two predominant masonry systems in Peru: structural and mixed masonry. Structural simulations were performed using software in accordance with national regulations and under high seismic hazard scenarios. A genetic algorithm (GA) was applied to optimize the design by simultaneously considering seismic stiffness, thermal comfort, and estimated cost criteria. The results show that both systems comply with regulatory drift limits. However, mixed masonry is more rigid and has lower total drift. The GA identified construction configurations that balance structural safety, energy efficiency, and economic viability. These findings contribute to sustainable housing development and support multi-criteria decision-making in emerging economies.

The objective of this study was to analyze the implementation of Design Thinking (DT) in higher education between 2013 and 2024. To understand its application and its coherence with the changing needs of higher education, 26 research articles were examined. The methodology included a systematic literature search in Scopus, followed by data selection and extraction according to the PRISMA-ScR methodology. The results revealed that DT has been widely adopted in higher education contexts, as it encourages creativity, problem solving and interdisciplinary collaboration. However, challenges were identified, such as the need for training for educators and resistance to change. In conclusion, DT has been effectively implemented in higher education, as it enriches pedagogy and promotes innovation, but attention must be paid to the challenges and sustainable implementation in this context must be guaranteed.

This systematic review analyzes the impact, trends, and challenges of AI adoption in higher education, following the PRISMA framework with a mixed-methods approach. A comprehensive search of Scopus and Web of Science (2019-2024) identified key AI applications, including personalized learning, automation, and learning analytics. The results show a growing interest in AI-driven education, with intelligent tutoring systems, chatbots, and adaptive learning platforms improving teaching efficiency and student engagement. Regional differences highlight Asia's focus on data analytics and Latin America's emphasis on digital literacy. Despite the benefits of accessibility, personalization, and administrative efficiency, challenges remain, including algorithmic bias, transparency, and ethical concerns. The study concludes that AI is reshaping education but requires responsible implementation to maximize its benefits while ensuring equity and ethical integrity.