Energy needs in the buildings sector accounts for 40 % of global energy demand. Therefore, the implementation of several renewable energy sources is necessary to reduce this demand. The design stage of a decentralized generation project requires quantifying the power to be installed and the energy forecast for each source throughout the useful life of the building. This study develops a novel optimization algorithm for a long-term economic function based on mixed-integer disciplined convex programming (MIDCP) which guarantees the sustainability of the building and its energy systems. The robust algorithm integrates risk management of intermittent sources, technical and economic parameters of selected technologies, and life cycle analysis (LCA) of different energy systems, including storage. Furthermore, the penetration of green hydrogen into the distributed generation mix is evaluated as an important contribution. Meteorological and energy demand variables of two antagonistic scenarios were also used as inputs to the algorithm. As a result, the optimal energy supply sizing for tertiary buildings in the two defined locations was obtained. The results of the simulations have achieved an optimal convergence of 100 % in the proposed scenarios, with a resolution time of 14 s and using a memory of about 183 MB. The simulations suggest a higher penetration of green hydrogen in scenarios where supply and investment costs decrease to gray hydrogen supply levels, reaching up to 81 % coverage of the thermal demand of the building. Hybrid energy systems under favorable conditions show a penetration of about 92 % within the distributed generation mix. The developed tool could enable decision-makers to optimally plan distributed generation projects in buildings based on economic, policy, and geographic conditions.

Estimating Photosynthetically Active Radiation (PAR) in the humid tropics is limited by prevailing cloud cover and a sparse network of weather stations. This study develops a five-phase framework to verify, reconstruct, and improve PAR estimates in a coastal region of Ecuador, using the CERES satellite product as a basis. The core is a benchmarking of six configurations (M0–M5), proposing a multivariate physical model (M3) that integrates seven predictors: GHI, zenith angle, relative humidity, temperature, precipitable water, cloud type, and AOD. Using only six months of in situ measurements, it proved feasible to reconstruct the complete annual series. M3 increased the coefficient of determination R² from 0.71 to 0.79 and reduced the overall RMSE to 32%, producing physically more consistent reconstructions than stochastic and quantile mapping methods. A marked satellite underestimation was detected at sunset (RMSE 68%, bias − 45%). Annually, M3 corrected the accumulated energy from 10,705 to 10,981 mol m⁻², approaching the 11,446 mol m⁻² measured in situ, a 37% reduction in annual bias. Because in situ data covered only July–December 2024, skill was assessed by internal cross-validation (leave-one-month-out R² ≈ 0.83), while January–June is reported as a reconstruction. Robustness to severe multicollinearity (VIF > 20) was confirmed with Ridge, LASSO, and PCR.

Reported accuracies for banana ripeness recognition are high but rarely comparable, because public releases mix source images with augmented exports and results are usually given as point estimates. This study proposes a reproducible, task-separated benchmark for visual banana postharvest quality assessment. Eight classifiers, including the compact MicroNet_Banana and lightweight and transformer-based models, were evaluated on the official 562-image test split of a classification release, and six detectors (five YOLO variants and RT-DETR-L) on the official 757-image test split of a detection release. Compared with previous work on the same release, the framework adds a provenance audit, hash-based split-independence checks, paired McNemar and equivalence tests with Holm adjustment, a standardized efficiency protocol with Pareto analysis, and external and cross-source validation. EfficientNet-B0 achieved the highest accuracy (99.47%; 95% CI: 98.44–99.82%) and was practically equivalent to three other lightweight models, whereas MicroNet_Banana reached 94.66% with 0.41 M parameters and 7.65 ms latency. The YOLO detectors (94.85–95.36% mAP@50) were practically equivalent, whereas RT-DETR-L was lower (93.36%). Removing near-duplicate test images changed the accuracy by less than 0.9 percentage points, but accuracy on an independent dataset fell to 32.8–61.7%. Model selection should therefore jointly consider accuracy, uncertainty, data provenance, generalization, and computational cost.

En los últimos 20 años los peines de frecuencias ópticas basados en lasers de femtosegundos representaron una revolución en el área de metrología y medidas de precisión, permitiendo medir directamente frecuencias de varias centenas de THz, así, como posibilitando el advenimiento de relojes atomicos ópticos. En este trabajo, estudiamos la generación de peines de frecuencias ópticas con base en moduladores electrópticos y fibras altamente no lineales, con el objetivo de implementar nuevas configuraciones y generar espectros con el mayor ancho de banda y alta coherencia entre las frecuencias generadas, aprovechando componentes comerciales disponibles en las comunicaciones ópticas. Nosotros implementamos un peine de frecuencias ópticas basado en efectos no lineales a través de una fibra óptica altamente no lineal combinándolo con modulación electroóptica. Los resultados experimentales muestran un peine de frecuencias ópticas con más de 100 nm de ancho de banda centrado en 1550 nm y con 25 GHz de separación entre las frecuencias, usando dos láseres continuos.