马来亚大学

This research introduces and optimizes a novel multi-generation power system integrating a steam Rankine cycle (SRC), a gas turbine (GT), an absorption refrigeration cycle (ARC), a proton exchange membrane (PEM) electrolyzer, and a CO2 separation unit. This system is designed to improve energy efficiency while simultaneously capturing CO2 and producing hydrogen through electrolysis. Two configurations-with and without ARC-are evaluated using a genetic algorithm-based multi-objective optimization framework, which considers exergetic efficiency, CO2 emission reduction, and total cost rate. The findings demonstrate that the proposed system improves exergetic efficiency by up to 71% and reduces CO2 emissions by up to 3.9% compared to a standalone GT system. Furthermore, the system without ARC achieves higher hydrogen production, while the system with ARC provides valuable cooling. These findings demonstrate the feasibility and environmental advantages of integrated power, CO2 capture, and H 2 blending systems for sustainable energy generation.

Partial shading fragments the power-voltage characteristic of series-connected photovoltaic (PV) arrays into multiple local peaks, rendering conventional trackers ineffective and exposing the reliance of metaheuristic global maximum power point tracking (GMPPT) methods on random initialization and empirical tuning coefficients. This paper proposes TSB-GMPPT, a deterministic tanh-surrogate branch-and-bound GMPPT algorithm that eliminates both dependencies. A control-oriented hyperbolic-tangent surrogate of the PV I-V characteristic admits a non-iterative Lambert W asymptotic approximation for the maximum power point voltage. During discovery, a single saturation-region measurement per candidate is sufficient to construct a calibrated optimistic bound, safely pruning the search space before final peak localization. An offline polynomial calibration recovers the remaining parameters from short-circuit current. During discovery, an optimistic branch-and-bound rule maintains an upper power bound for each unconfirmed layer and permanently discards candidates that cannot exceed the incumbent confirmed power, contracting the admissible search space without exhaustive probing. A five-state finite-state machine governs initialization, shading-topology detection, discovery, winner refitting, convergence, and adaptive reinitialization. Experimental validation using a Chroma 62150H solar array simulator and a dSPACE DS1104 platform on five-module benchmark cases and additional eight-module complex multi-peak shading cases demonstrates average tracking efficiencies of 99.94% under partial shading and 99.98% under uniform irradiance, with corresponding average convergence times of 0.25 s and 0.21 s, respectively, for the five-module test set. Further validation under eight-module complex multi-peak shading cases and dynamic irradiance transitions confirms rapid re-tracking, accurate GMPP identification, and robust operation under realistic operating conditions.

Smart-building load forecasters are often trained offline on dense, multivariate, high-frequency data, but deployment may provide only hourly, feature-limited inputs. Missing features must then be reconstructed, and their errors can propagate through the model. If this input uncertainty is not reflected, prediction intervals may become miscalibrated, affecting demand-response scheduling. Our work examines where uncertainty should be placed once inference inputs are reconstructed. We develop a unified one-day-ahead probabilistic forecasting framework that aligns temporal resolution, reconstructs the unavailable inputs, and derives causal features, and we compare a modular post-hoc residual-quantile scheme with an integrated in-model quantile-learning scheme. The comparison uses three mid-scale Deep Learning (DL) backbones: recurrent, hybrid recurrent, and attention-based Temporal Fusion Transformer (TFT) models, under identical inputs, forecasting horizon, preprocessing rules, and training budgets. Results show that uncertainty placement is backbone-dependent. Integrated quantile learning is most reliable with the TFT, yielding 2.2-3.6% MAPE and 28-83 W RMSE on the labeled test window, while producing intervals about 5 & times; narrower than the modular intervals at the closest-to-nominal coverage level. Diebold-Mariano tests support the TFT ranking and the mixed behavior of the recurrent backbones. A reconstruction-sensitivity test shows that reconstructed inputs increase the Quantile Score (QS) by 106% while interval width remains nearly unchanged, indicating that the model does not automatically absorb reconstruction-induced uncertainty. Robustness checks against non-DL baselines and seasonal hold-out weeks support this ranking. Our results expose the limits of post-hoc residual quantiles when inference depends on reconstructed inputs.

Time series clustering is critical for analyzing noisy and non-stationary signals such as electroencephalogram (EEG), where conventional methods assume equal credibility across all trials, often degrading cluster quality in artifact-prone data. This paper proposes the Rough Set-based Credibility Adjusted Data-Conscious Clustering Method (RCADCCM), which incorporates trial-wise credibility into the clustering process. Employing fuzzy membership and rough-set lower/upper approximations concepts, RCADCCM emphasizes credible trials during centroid updates while selectively pruning ambiguous trials, thereby improving robustness without requiring labels or deep representation learning. The method is evaluated on three motor imagery EEG datasets (BCI Competition IV Datasets 1, 2a, and 2b) using CSP, OvR-CSP and FBCSP, and further validated on the non-EEG Cylinder-Bell-Funnel (CBF) time series benchmark to provide representative evidence of applicability beyond EEG data. RCADCCM is compared with classical clustering baselines and a lightweight deep clustering approach (autoencoder-based K-means). Across all evaluated datasets, RCADCCM consistently demonstrates superior clustering accuracy and structural compactness. Parameter sensitivity analysis reveals stable performance for moderate retention ratios, confirming an effective balance between data credibility and utilization. Runtime analysis indicates only modest overhead relative to classical clustering, while remaining substantially more efficient than deep clustering approaches. UMAP visualizations further illustrate improved cluster separability and the removal of ambiguous trials near overlap regions. RCADCCM achieves performance comparable to supervised EEG classifiers while operating in a fully unsupervised setting, highlighting its robustness, interpretability, and computational efficiency for noisy time series analysis beyond EEG.

Reliable carbon price forecasts support market monitoring, policy assessment, and operational planning in smart energy systems. Yet carbon prices are volatile, non-stationary, and driven by interacting regulatory, energy, commodity, and financial conditions. This study proposes an integrated Decomposition-Representation-Integration (DRI) framework that combines complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), a lightweight decoder-only causal Transformer, and XGBoost regression. CEEMDAN separates multi-scale dynamics; causal masked self-attention jointly reconstructs cross-scale dependencies without access to future observations; and XGBoost integrates the learned representation with exogenous drivers. The framework is evaluated using a strictly causal five-fold walk-forward protocol in which preprocessing, decomposition, validation, tuning, and model fitting are confined to each training window. Its CEEMDAN-CausalTransformer-XGBoost implementation achieves a minimum MAE of 0.6583, RMSE of 0.8071, and out-of-sample R2 of 0.7470. It consistently outperforms decomposition-only and standalone benchmarks across five tuning strategies, while Diebold-Mariano tests provide statistical evidence of improved predictive accuracy over key baselines. Prediction-path analysis suggests that gains are more visible during high-volatility periods. These results indicate that reconstructing cross-scale information improves carbon-market monitoring and provides decision support for smart energy system operators, market participants, and policymakers.