Francisco Zayas-Gato, Antonio Díaz-Longueira, Paula Arcano-Bea, Álvaro Michelena, Jose Luis Calvo-Rolle, Esteban Jove
{"title":"A new deep learning-based approach for predicting the geothermal heat pump’s thermal power of a real bioclimatic house","authors":"Francisco Zayas-Gato, Antonio Díaz-Longueira, Paula Arcano-Bea, Álvaro Michelena, Jose Luis Calvo-Rolle, Esteban Jove","doi":"10.1007/s10489-025-06457-7","DOIUrl":"10.1007/s10489-025-06457-7","url":null,"abstract":"<div><p>In recent years, growing concern about climate change and the need to reduce greenhouse gas emissions have highlighted the role of energy efficiency and sustainability on the global agenda. Energy policies are decisive in establishing regulatory frameworks and incentives to address these challenges, leading to an inclusive and more resilient energy transition. In this context, geothermal energy is an essential source of renewable, low-emission energy, capable of providing heat and electricity sustainably. The present research focuses on a bioclimatic house’s geothermal energy system based on a heating pump and a horizontal heat exchanger. The main aim is to predict the generated thermal power of the heat pump using historical data from several sensors. In particular, two approaches were proposed with both uni-variate and multi-variate scenarios. Several deep learning techniques were applied: LSTM, GRU, 1D-CNN, CNN-LSTM, and CNN-GRU, obtaining satisfactory results over the whole dataset, which comprised one year of data acquisition. Specifically, promising results have been achieved using hybrid methods combining recurrent-based and convolutional neural networks.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10489-025-06457-7.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143667879","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Guochang Zhu, Jun Hu, Li Liu, Qinghua Zhang, Guoyin Wang
{"title":"Enhancing graph representation learning via type-aware decoupling and node influence allocation","authors":"Guochang Zhu, Jun Hu, Li Liu, Qinghua Zhang, Guoyin Wang","doi":"10.1007/s10489-025-06443-z","DOIUrl":"10.1007/s10489-025-06443-z","url":null,"abstract":"<div><p>The traditional graph representation methods can fit the information of graph with low-dimensional vectors, but they cannot interpret their composition, resulting in insufficient security. Graph decoupling, as a method of graph representation, can analyze the latent factors composing the graph representation vectors. However, in current graph decoupling methods, the number of factors is a hyperparameter, and enforce uniform decoupling vector dimensions which leads to information loss or redundancy. To address these issues, we propose a type-aware graph decoupling based on influence called Variational Graph Decoupling Auto-Encoder (VGDAE). It uses node labels as interpretable and objectively existing natural semantics for decoupling and allocates embedding space based on node influence, addressing the issues of manually setting the number of factors in traditional graph decoupling and the mismatch between node information size and embedding space. On the Cora, Citeseer, and fb-CMU datasets, VGDAE shows the impact of different node classes as decoupling targets on classification tasks. Furthermore, we perform visualization of the representations, VGDAE exhibits performance improvements of 2% in classification tasks and 12% in clustering tasks when compared with baseline models.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143668347","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"HRMG-EA: Heterogeneous graph neural network recommendation with multi-level guidance based on enhanced-attributes","authors":"Longtao Wang, Guiyuan Yuan, Chao Li, Yufei Zhao, Hua Duan, Qingtian Zeng","doi":"10.1007/s10489-025-06428-y","DOIUrl":"10.1007/s10489-025-06428-y","url":null,"abstract":"<div><p>Heterogeneous Graph Neural Networks are an efficient and powerful tool for modeling graph structure data in recommendation systems. However, existing heterogeneous graph neural networks often fail to model the dependencies between user and item attribute preferences, limiting graph structure optimization and consequently reducing the accuracy of recommendations. To overcome these issues, we propose a Heterogeneous graph neural network Recommendation with Multi-level Guidance based on Enhanced-Attributes (HRMG-EA). First, we design an attribute enhanced gated network to model user-item interaction attribute scenarios and obtain enhanced-attributes by capturing complex attribute dependencies. It effectively avoids the expansion of the graph scale in attribute graph scenarios and further covers personalized attribute relationship distribution characteristics of users and items. Then, we propose a novel multi-level graph structure guidance strategy based on enhanced-attributes. It guides graph structure learning from three optimization levels, optimizing from two perspectives: explicit (heterogeneity and homogeneity) and implicit (contrast enhancement). The former can screen higher-quality heterogeneous neighbor nodes in a direct interaction environment, and filter out redundant or erroneous edges under different similar semantic interest paths to improve the quality of the neighborhood environment. The latter aligns representation embeddings of enhanced-attributes and graph structure in a latent space, explores their potential commonalities, and obtains more comprehensive, fine-grained semantic and beneficial structural information. Finally, on two real-world datasets, HRMG-EA significantly outperforms the state-of-the-art baseline algorithms in both recall and normalized discounted cumulative gain. A large number of ablation experiments and analytical verifications also verify its effectiveness.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143668349","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Shijiang Li, Zhihai Wang, Xiaokang Wang, Zihao Yin, Muyun Yao
{"title":"Frequency-enhanced and decomposed transformer for multivariate time series anomaly detection","authors":"Shijiang Li, Zhihai Wang, Xiaokang Wang, Zihao Yin, Muyun Yao","doi":"10.1007/s10489-025-06441-1","DOIUrl":"10.1007/s10489-025-06441-1","url":null,"abstract":"<div><p>With the widespread adoption of the Internet of Things (IoT), vast amounts of multivariate time series data are generated, which reflect the operational status of systems. Accurate and efficient anomaly detection in these data is crucial for maintaining system stability. However, data from unstable environments often exhibit high volatility, data drift, and complex patterns of anomalies. Unsupervised anomaly detection models are typically designed for stable data and lack generalizability, leading to a high rate of false positives when applied to unstable data. This paper introduces the frequency-enhanced and decomposed transformer for anomaly detection (FDTAD), which is a novel anomaly detection model based on a transformer that is enhanced with frequency and time series decomposition. FDTAD addresses data drift by decomposing time series and leverages both time-domain and frequency-domain information to improve the generalization ability of the model. The model preserves major amplitudes in the frequency domain to extract primary periodic patterns, uses spectral residuals to capture detailed variations, and incorporates a frequency-domain correlation attention mechanism to extract dependencies in frequency-domain data in a sparse representation. Additionally, a spatiotemporal module is designed to extract the temporal correlations in the data and spatial correlations among the data with different attributes. FDTAD combines a data periodic pattern reconstructor and a data detailed pattern reconstructor through an adversarial mechanism to achieve maximum accuracy in reconstructing normal data. Extensive experiments on 10 public datasets demonstrate that FDTAD outperforms state-of-the-art baseline methods, with a 4.1% improvement in the F1 score and a 4.7% improvement in precision.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143668362","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"3DGCformer: 3-Dimensional Graph Convolutional transformer for multi-step origin–destination matrix forecasting","authors":"Yiou Huang, Hao Deng, Shengjie Zhao","doi":"10.1007/s10489-025-06371-y","DOIUrl":"10.1007/s10489-025-06371-y","url":null,"abstract":"<div><p>Forecasting Human mobility is of great significance in the simulation and control of infectious diseases like COVID-19. To get a clear picture of potential future outbreaks, it is necessary to forecast multi-step Origin–Destination (OD) matrices for a relatively long period in the future. However, multi-step Origin–Destination Matrix Forecasting (ODMF) is a non-trivial problem. First, previous ODMF models only forecast the OD matrix for the next time-step, and they cannot perform well on long-term multi-step forecasts due to error accumulation. Second, many ODMF methods capture spatial and temporal dependencies with separate modules, which is insufficient to model spatio-temporal correlations in the time-varying OD matrix sequence. To address the challenges in multi-step ODMF, we propose 3-Dimensional Graph Convolutional Transformer (3DGCformer). As an enhancement of the original 3DGCN, we propose a novel Origin–Destination Feature Propagation (ODFP) rule between 3DGCN layers and integrate 2 3DGCNs with different spatio-temporal graphs and corresponding feature propagation rules to model the formation of OD flows in a more comprehensive way. For multi-step forecasts, 3DGCformer uses Transformer to capture long-term global temporal dependency, and adapt its decoder using labeled tokens to avoid error accumulation and improve time efficiency. To avoid information loss as the number of regions increases, we propose a patch embedding approach to convert data from 3DGCNs to the Transformer module. We perform extensive experiments on 4 real-world human mobility datasets, and the results show that our proposed model outperforms the state-of-the-art methods.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143668350","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Leveraging CQT-VMD and pre-trained AlexNet architecture for accurate pulmonary disease classification from lung sound signals","authors":"Zakaria Neili, Kenneth Sundaraj","doi":"10.1007/s10489-025-06452-y","DOIUrl":"10.1007/s10489-025-06452-y","url":null,"abstract":"<p>This study presents a novel algorithm for classifying pulmonary diseases using lung sound signals by integrating Variational Mode Decomposition (VMD) and the Constant-Q Transform (CQT) within a pre-trained AlexNet convolutional neural network. Breathing sounds from the ICBHI and KAUHS databases are analyzed, where three key intrinsic mode functions (IMFs) are extracted using VMD and subsequently converted into CQT-based time-frequency representations. These images are then processed by the AlexNet model, achieving an impressive classification accuracy of 93.30%. This approach not only demonstrates the innovative synergy of CQT-VMD for lung sound analysis but also underscores its potential to enhance computerized decision support systems (CDSS) for pulmonary disease diagnosis. The results, showing high accuracy, a sensitivity of 91.21%, and a specificity of 94.9%, highlight the robustness and effectiveness of the proposed method, paving the way for its clinical adoption and the development of lightweight deep-learning algorithms for portable diagnostic tools.</p><p>Overview of the proposed methodology for pulmonary disease classification using CQT-VMD and pre-trained AlexNet architecture applied to lung sound signals</p>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143655321","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"OTMKGRL: a universal multimodal knowledge graph representation learning framework using optimal transport and cross-modal relation","authors":"Tao Wang, Bo Shen","doi":"10.1007/s10489-025-06459-5","DOIUrl":"10.1007/s10489-025-06459-5","url":null,"abstract":"<div><p>The demand for integrating multimodal information, such as text and images, has grown significantly as it enables richer and more comprehensive knowledge representations. Most existing multimodal knowledge graph representation learning (KGRL) methods focus primarily on fusing multimodal entity information, directly applying multimodal entities and single-modal relations to downstream tasks. However, these methods face challenges related to the heterogeneity of multi-source entity data, which amplifies the differences in feature distributions between entity and relation representations. To address these challenges, we propose a universal multimodal KGRL framework, OTMKGRL, which seamlessly incorporates multimodal information into three types of single-modal KGRL methods. First, OTMKGRL employs Tucker decomposition to project entity text and image data into a shared space, thereby generating multimodal entity representations. It then uses optimal transport to integrate multimodal entity information into the original single-modal entity representations. Second, OTMKGRL introduces a cross-modal relation attention mechanism that fuses effective multimodal entity features into the original single-modal relations, yielding cross-modal relation representations. Extensive experiments across three multimodal datasets demonstrate the effectiveness and versatility of our approach. The OTMKGRL framework significantly enhances the performance of existing single-modal KGRL models in multimodal settings.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143655322","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Simão P. Carvalho, Joana Figueiredo, João J. Cerqueira, Cristina P. Santos
{"title":"Locomotion mode prediction in real-life walking with and without ankle–foot exoskeleton assistance","authors":"Simão P. Carvalho, Joana Figueiredo, João J. Cerqueira, Cristina P. Santos","doi":"10.1007/s10489-025-06416-2","DOIUrl":"10.1007/s10489-025-06416-2","url":null,"abstract":"<p>Exoskeletons can assist human locomotion in real-life scenarios, but existing tools for decoding locomotion modes (LMs) focus on recognition rather than prediction, which can lead to delayed assistance. This study proposes a long short-term memory (LSTM) neural network to predict five LMs (level-walking, ramp ascent/descent, stair ascent/descent) with greater lead time compared to state-of-the-art methods. We examined the optimal sequence length (SL) for LSTM-based LM prediction, using data from inertial sensors placed on the lower limbs and the lower back, along with a waist-mounted infrared laser. Ten subjects walked in real-life scenarios, both with and without an ankle–foot exoskeleton. Results show that a 1-s SL provides the most advanced and accurate LM prediction, outperforming SLs of 0.6, 0.8, and 1.2 s. The proposed LSTM model achieved an accuracy of 98 ± 0.31%, predicting LMs 0.66 s in advance (for an average stride time of 1.98 ± 0.83 s). Level-walking presented more misclassifications, and the model primarily relied on inertial data over laser input. Overall, these findings demonstrate the LSTM’s strong predictive capability for both assisted and non-assisted walking and independent of which limb executes the transition, supporting its applicability for exoskeleton-assisted locomotion.</p>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10489-025-06416-2.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143645494","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Yuchuan Hu, Bitao Hu, Bing Guo, Cheng Dai, Yan Shen
{"title":"SDDP: sensitive data detection method for user-controlled data pricing","authors":"Yuchuan Hu, Bitao Hu, Bing Guo, Cheng Dai, Yan Shen","doi":"10.1007/s10489-025-06229-3","DOIUrl":"10.1007/s10489-025-06229-3","url":null,"abstract":"<p>In the era of big data, there is an urgent need for data sharing, in which data pricing is a crucial issue, because a reasonable price can not only enhance the willingness of users to share data but also promote the progress of data sharing. However, current research is mostly approached from the perspective of data sharing platforms, treating all data equally without sufficient evaluation of sensitive data within shared datasets and personalized perception of privacy from the users themselves. To address this problem, we detected sensitive data in each piece of data and then defined the pricing function based on information entropy and the user’s perception of sensitive information. To enhance the accuracy of sensitive data detection, we integrated an attention mechanism into a pre-trained model to comprehensively represent the samples. Subsequently, on the basis of automatically generating label correlation vectors to calculate the correlation matrix, a graph convolutional neural network was employed to mine the correlation between labels. Furthermore, based on the detection results, information entropy and user ratings are reasonably mapped to prices. Pricing based on user ratings is more suitable for pricing personal data rather than government or institutional data. The experimental results on the dataset of Twitter text sent by users have demonstrated that the average precision of our sensitive data detection model has improved by up to 9.26% compared to comparison models, and SDDP can provide reasonable pricing for samples containing sensitive data and fair compensation for users.</p>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143655367","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"HKGAT: heterogeneous knowledge graph attention network for explainable recommendation system","authors":"Yongchuan Zhang, Jiahong Tian, Jing Sun, Huirong Chan, Agen Qiu, Cailin Liu","doi":"10.1007/s10489-025-06446-w","DOIUrl":"10.1007/s10489-025-06446-w","url":null,"abstract":"<div><p>This paper presents the Heterogeneous Knowledge Graph Attention Network (HKGAT) for recommendation systems. As recommendation technology evolves, systems now emphasize diversity, fairness, and explainability alongside accuracy. Traditional methods encounter issues integrating knowledge graphs and lack explainability. HKGAT addresses these by leveraging heterogeneous knowledge graphs. It consists of a heterogeneous information aggregation layer, an attention-aware heterogeneous relation fusion layer, and a prediction layer. First, recommendation data forms a user-item knowledge graph. Then, the aggregation layer collects relation information, followed by the fusion layer integrating it for higher-order feature representations. The prediction layer combines link prediction and recommendation score prediction. Additionally, paths of top-ten results are analyzed and quantified for explainability to optimize ranking. Experiments on self-constructed and Amazon-book datasets show HKGAT outperforms baselines like HetGCN, with significant improvements in Precision, Recall, F1 score, and NDCG@10, and a notable 1.9% gain in NDCG@10 from explainable ranking optimization.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 6","pages":""},"PeriodicalIF":3.4,"publicationDate":"2025-03-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143655368","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}