This study presents a computational intelligence-based method for robust detection and reconstruction of staircases in complex environments. The proposed framework combines point cloud processing with deep learning-based object detection to address challenges related to lighting variability, occlusion, and computational cost. A central component is the generation of Orthogonal Grayscale (OGS) images — 2.5D elevation maps encoded as grayscale images — which preserve spatial structure while enabling the application of efficient 2D convolutional neural networks. A customized YOLOv5-OBB model is trained on a newly collected dataset to localize and classify staircases under various conditions, including complete darkness. The detected regions are then analyzed using grayscale adaptive histograms to estimate stair height and count, followed by 3D reconstruction through bounding volume methods. Experiments demonstrate detection accuracy of 98.8% for upstairs and 94.5% for downstairs scenarios, with real-time inference performance. The integration of neural vision, elevation encoding, and spatial reconstruction exemplifies a hybrid computational intelligence system applicable to robotics, assistive navigation, and autonomous inspection tasks.
An emerging approach called cross-domain recommendation transfers knowledge across heterogeneous areas, including books and movies, to improve the user experience. Existing methods often suffer from sparse interactions, semantic gaps across domains, and insufficient explainability. To address these limitations, we propose a novel Cross-Domain Knowledge Graph Contrastive (X-DKGC) unified framework that unifies knowledge graph modeling, contrastive representation learning, and explainable reasoning. First, a unified heterogeneous knowledge graph is constructed by integrating users, books, movies, genres, authors, directors, similarity links, and cross-domain semantic relations. Next, X-DKGC introduces two new modules: Multi-Hop Semantic Sampler (MHSS) and Dual-Domain Contrastive Learning (DDCL) to refine initial KG embeddings using intra-domain and cross-domain contrastive signals. A structure-aware regularizer is also formulated to preserve graph topology during training. The resulting embeddings support both recommendation and transparent explanation generation via multi-hop reasoning, semantic overlaps, and collaborative neighbors. Extensive experiments were conducted on both the original Goodreads–MovieLens dataset and the publicly available Amazon Review Data benchmark (Books and Movies and TV). The proposed framework achieved higher observed performance than the representative state-of-the-art baselines across both datasets, with statistical significance varying across datasets, metrics, and baseline comparisons. The proposed framework was also evaluated against KG-NeuCMF under identical experimental settings and achieved higher observed performance across the reported evaluation metrics.
The global energy transformation promotes the development of large-scale grid-connected wind energy. The grid-connected operation performance of wind farms is related to wind power consumption and the safe and stable operation of the power system. A combined weighting system that integrates the improved analytic hierarchy process, entropy weight method, and multiplier synthesis normalization strategy is studied and constructed, and a static quantification model of the improved approximate ideal solution ranking method is established. The purpose of this system is to achieve comprehensive quantification and dynamic timing level assessment of wind farm grid-connected performance, and provide refined decision support for power grid dispatching and wind farm operation and maintenance. At the same time, it combines the improved gray wolf optimization algorithm and the multi-scale attention-bidirectional long short-term memory network to build a sequential level evaluation model. Experimental results showed that the subjective and objective consistency coefficient of the combined weighting method reached 0.9524, the standard deviation of the closeness of the improved approximate ideal solution ranking method was 0.0926, and the coefficient of variation in the small-sample scenario was 0.138. After optimization with the improved gray wolf optimization algorithm, the model’s F1-score increased to [Formula: see text] (validation-set optimum during IGWO search), and the accuracy of the fusion model reached [Formula: see text]%. Compared with the baseline long short-term memory network model, the accuracy increased by 10.6 percentage points, and all performance indicators were better than similar comparison models. The performance quantification of the proposed method can complete the precise prediction of the grid-connected performance of the wind farm, and the hierarchical evaluation method can realize the five-level sequential assessment of the grid-connected performance, providing an effective technical method for the wind farm grid-connected performance evaluation.
The rapid progress in information and communications technologies has made personal data a valuable resource, which can help data owners seamlessly meet the needs of their affiliates. Data owners are acquiring enormous and various amounts of personal data due to the widespread proliferation of digital tools and computing. Current efforts mostly concentrate on establishing Federated Learning (FL) methods that are optimized, ignoring the enduring and internal connections between users, like social relationships. This survey helps the researchers analyze several approaches to privacy preservation in social networks. 60 research articles that focus on different techniques for social clustering privacy preservation are analyzed. Moreover, this survey provides an analysis of the available privacy preservation techniques in social clustering on the basis of publication year, research techniques, tools used, dataset used, and performance assessment, along with the accomplishments of research techniques, paving the way for the establishment of innovative models in the future. Various researchers have developed models to address this issue, and the techniques can be categorized based on Deep Learning (DL), FL, and other techniques. The real-world datasets are most widely employed. Moreover, the research gaps and drawbacks of these techniques are discussed, highlighting the need for an effective approach to privacy preservation in social clustering using FL.
Accurate forecasting of indoor environmental variables is essential for occupancy-aware building energy management, yet the interplay between neural network architecture, forecasting strategy, and downstream occupancy inference remains poorly understood. This study presents a comprehensive controlled evaluation framework for long short-term memory (LSTM)-based multihorizon indoor CO 2 forecasting and downstream occupancy-proxy inference using two years of residential environmental sensor data. Established forecasting approaches are evaluated across six forecast horizons from 1 h to 48 h. A crossed strategy-by-hyperparameter experiment descriptively partitions observed mean absolute error (MAE) variation: hyperparameter configuration accounts for 80.7% at [Formula: see text] h, whereas strategy accounts for 39.1% at [Formula: see text] h, the larger of the two main-effect shares at that horizon. Temporal feature augmentation reduces MAE by 9.3% at [Formula: see text] h and 8.5% at [Formula: see text] h, while increasing MAE at [Formula: see text] h. Finally, we propose an unsupervised occupancy inference framework applying absolute threshold, rate-of-change, and adaptive threshold methods to both measured and LSTM-forecasted CO 2 . These findings provide operational guidance for deploying neural network forecasters in intelligent building management systems.
