Energy optimization in intelligent buildings is a multi-objective decision-making problem that requires the consideration of uncertainties on both supply and demand sides. However, existing optimization strategies often consider a single indicator in isolation and lack collaborative system optimization, resulting in poor overall performance. To address this issue, this study proposes an intelligent building energy optimization model based on deep reinforcement learning and multi-criteria decision-making. This model integrates Long Short-Term Memory Transformer network for multi-source temporal feature extraction and modeling, uses differential evolution algorithm to optimize Proximal Policy Optimization reinforcement learning strategy, and introduces analytic hierarchy process to allocate weights to competitive indicators such as energy efficiency, economy, comfort, to construct a comprehensive incentive function that guides the intelligent agent to generate optimization strategies that benefit both supply and demand sides in synergy. The results indicate that the proposed model achieves a 40% convergence rate and a 95.5% policy performance achievement rate. The proposed strategy achieves a maximum energy optimization efficiency of 15.8% and a maximum economic benefit improvement of 16.2%. In simulation tests, the proposed strategy achieved a maximum satisfaction rate of 92.1% on the supply side and 95.6% on the demand side. These results demonstrate that the proposed method can effectively address uncertainties from both the supply and demand sides to formulate optimal strategies that consider energy consumption, economic performance, and other key indicators. This method provides a new perspective for smart building energy optimization and promotes the application of deep learning algorithms in diverse decision-making tasks.
This paper proposes a model for optimizing enterprise office efficiency through multimodal employee behavior analysis. The model integrates computer vision, sensor data, workflow data, and text information in a four step process: data acquisition, feature extraction, behavior identification, and optimization strategy formulation, to identify task, collaborative, non-work, and organizational behaviors. It dynamically optimizes the office environment, workflow, and human resource allocation. Key innovations include: first, the “in-depth fusion of four-modal data” via the “temporal-semantic dual alignment” technique, to address data heterogeneity. Second, the “behavior-efficiency dynamic mapping” mechanism, which quantifies negative correlation (r=-0.72) between non-work behaviors and task efficiency, provides a quantitative basis for optimization. Experiments involving 50 technology company employees demonstrate the model’s superiority, achieving an office efficiency improvement rate of 0.85, which surpasses both traditional methods (0.65) and common multimodal methods (0.72). The model also attained a behavior identification accuracy rate of 0.90, outperforming comparison methods (0.80 and 0.85). The model also excels in optimizing the office environment, workflow, and human resources, with improvements of up to 30%. Core contributions include: 1) a multimodal data fusion framework that increases behavior recognition accuracy to 90%; 2) an integrated “three-in-one” optimization strategy for the environment, processes, and human resources, which increased employee satisfaction by 30% and reduced task time by 25%; and 3) empirical validation that demonstrates an 85% efficiency improvement rate, significantly outperforming traditional and standard multimodal methods. The study highlights the model’s practicality and potential for widespread adoption in enhancing office efficiency.
The integration of emerging technologies has profoundly reshaped interior architecture, transforming design processes, user experience, and sustainability practices. However, research remains fragmented across disciplines, highlighting the need for a consolidated synthesis that bridges theory and practice. This study systematically reviews the impact of digital technologies, particularly Virtual Reality (VR), Augmented Reality (AR), Artificial Intelligence (AI), and smart interactive systems, on interior architecture. Following PRISMA 2020 guidelines and the Cochrane Handbook, we conducted a systematic search of Scopus, Web of Science, IEEE Xplore, and ScienceDirect identified 1,246 records. After rigorous screening and quality appraisal using Joanna Briggs Institute (JBI) tools, we included 36 peer-reviewed studies (2018–2025). Four dominant themes emerged: (1) enhanced visualization and perception through VR/AR, enabling improved spatial cognition and client–designer collaboration. (2) user experience and interactivity facilitated by smart and adaptive environments that support multisensory engagement. (3) sustainability and resource optimization, driven by AI-assisted material and energy efficiency. (4) educational and professional applications, where immersive systems enhance digital literacy and design comprehension. Despite their transformative potential, limitations persist regarding scalability, methodological rigor, and cross-cultural generalizability. This review presents an integrated framework for understanding technology-enabled interior design, positioning VR, AI, and smart systems as core drivers of sustainable, human-centered, and innovation-oriented design practice, rather than peripheral tools. Future research should pursue longitudinal, large-scale, and cross-cultural studies, alongside theoretical refinements, to advance both academic knowledge and practical implementation.
Driven by an urgent need for systematic risk assessment of Build-Operate-Transfer (BOT) projects in Syria's post-war reconstruction context, this research addresses the critical risk factors specific to BOT projects in post-conflict environments, systematically classifies and prioritizes these risks, and develops a framework for effective risk assessment and management. Using a rigorous mixed-methods approach, we developed a comprehensive risk register that identifies and categorizes 48 distinct risks across five major categories: organization/political, financial/economic, environmental/natural, design/implementation/operation, and humanitarian/materials/equipment. The research methodology consisted of questionnaire responses from 161 industry professionals, including government officials and engineering consultants. The collected data were analyzed using SPSS for validation. Both the impact and the probability of each risk were measured on a five-point Likert scale, and a z-test was used to test the hypothesis. The results indicated that organizational and political risks, specifically the imposition of international sanctions and regional tensions, pose the primary threat to BOT project implementation in Syria, while environmental risks were found to be less critical. The developed risk register provides a structured tool for risk assessment and management, incorporating local contextual factors that are often absent in traditional frameworks. For policymakers, investors, and project managers involved in Syria’s post-conflict reconstruction, this research offers practical insights and contributes to a theoretical understanding of BOT projects more generally.
This research evaluates the efficacy of innovation and entrepreneurship in higher education institutions and proposes an assessment approach that uses an enhanced multivariate neural network to improve the scientific rigor and precision. Existing evaluation methods are plagued by simplistic index systems, an inability to capture nonlinear relationships between indicators, and low adaptability to high-dimensional data. This leads to insufficient evaluation accuracy and poor practical guidance. First, we construct an assessment index system covering five aspects: student’s entrepreneurial achievements, innovation ability, curriculum teaching, teacher construction, and practice platform. We then optimized the Multivariate Neural Network (MNN) model using the Heap-Based Optimizer (HBO) and experimentally verified on the basis of 15,328 data points from 56 “dual-innovation” demonstration institutions between 2015 to 2022. The results show that the HBO-MNN model outperforms the traditional methods, achieving an accuracy of 0.954 precision of 0.9322, recall of 0.941, and an F1 score 0.9585 and the evaluation time is only 0.801 seconds. Feature importance analysis revealed that the number of cooperative platforms, the size, the practice bases and course examination results had the greatest impact on the assessment results. The study demonstrates that this method can effectively improve assessment accuracy and provide data support for universities to optimize innovation and entrepreneurship programs. In the future, multimodal data can be integrated to further improve the method’s applicability.
With the growth of big data, the application of distributed storage is becoming increasingly widespread. Traditional Erasure Codes (ECs) are computationally intensive in distributed storage systems and inefficient for managing large-scale economic data. To address this issue, a Reed-Solomon code storage method based on the Reed-Muller transform is proposed. The encoding and decoding algorithm of Reed-Solomon (RS) codes is verified through the Reed-Muller (RM) transform, and a cyclic shift is added to the Reed-Muller transform to reduce the computational complexity of ECs. The distributed storage file system is expanded to ensure smooth operation across the system's modules. The experimental results showed that in the extended file system, the proposed Reed-Muller Reed-Solomon code achieved fast encoding and decoding speeds across different numbers of data blocks, with maximum encoding and decoding speeds of 981MB/s and 915MB/s, respectively. The EC proposed by the research was not affected by file size, and its encoding and decoding speed increased with increased file size. Moreover, the Reed-Muller and Reed-Solomon codes can efficiently handle economic data with lower memory usage than other ECs. Overall, the Reed-Muller Reed-Solomon code proposed by the research has good performance and can efficiently handle big data.
Archival resource management, as a critical foundation of the information society, faces challenges in efficient storage, accurate retrieval, and intelligent analysis of multimodal data. Traditional approaches relying on manual classification and keyword-based retrieval struggle with semantic understanding, predictive modeling, and real-time anomaly detection. To address these issues, this study proposes a novel archival resource management system that integrates Multi-Head Self-Attention (MHSA), Temporal Convolutional Network (TCN), and Long Short-Term Memory (LSTM). The novelty of this system lies in the combination of a global semantic representation of MHSA, multi-scale time-series modeling with TCN, and long-term dependency capture with LSTM, further enhanced by a phase mechanism and Gated Linear Unit (GLU) to optimize feature selection. Experimental evaluations on CMU and MIMIC-III datasets demonstrate that the system achieved superior performance in multiple tasks: feature similarity of 0.97, clustering accuracy of 98.25%, classification accuracy of 96.42%, F1-score of 95.28%, time-series forecasting RMSE of 0.31, MAE of 0.23, anomaly detection accuracy of 94.37%, and a low false alarm rate of 2.81%. Moreover, the system maintains robustness under noisy conditions and generalizes well across different archival datasets. These results highlight the effectiveness and originality of the proposed framework, providing a feasible solution for building intelligent archival resource management systems with high precision, predictive capability, and anomaly monitoring.
The construction industry is characterized by a high risk of accidents due to multiple hazards, dynamic work environments, weather, tight schedules, and a large workforce, all of which challenge the consistent enforcement of safety measures. Despite the issuance of Ministerial Regulation No. 10/2021 concerning the Construction Safety Management System (CSMS), its implementation in Indonesia remains suboptimal, as evidenced by the high frequency of workplace accidents recorded between 2018 and 2020. This study aims to identify the key factors supporting and hindering the successful implementation of CSMS in construction projects. A descriptive quantitative approach was used, involving a two-stage survey: a preliminary questionnaire for expert validation and a main questionnaire distributed to 147 respondents working on road and bridge projects managed by state-owned enterprises. Factor analysis was applied to simplify complex variable relationships and identify latent constructs. The results revealed five supporting factors: (1) strong management commitment to safety. (2) The availability of safety procedures, policies, and resources. (3) Organizational competence and effective managerial implementation. (4) A strong safety culture, and (5) comprehensive safety training for supervisors and workers. Conversely, three primary hindering factors were identified: (1) weak safety culture, (2) management shows deficiency in commitment and capability, and (3) low worker competence. These findings offer practical insights to help Indonesian construction companies to strengthen their safety culture and managerial commitment. They also provide government policy aimed at improving the monitoring and enforcing CSMS implementation.
In the digital economy, the demand for financial data sharing among enterprises is growing rapidly. Traditional financial sharing models face prominent issues, including data security risks, insufficient efficiency and transparency. To enable the encrypted storage and hierarchical access of sensitive financial data, this study proposes a blockchain-based scheme for secure sharing and encrypted computation of enterprise financial information. The blockchain component incorporates credibility evaluation, process optimization, and cross-organizational collaboration models. Its layered architecture optimizes sharing efficiency and supports applications in supply chain finance, among other scenarios. The encryption component is based on the Number Theory Research Unit (NTRU) encryption algorithm, which utilizes polynomial rings, homomorphic encoding, and secure two-party protocols to ensure data privacy and operational integrity. The synergy between these two components provides digital support for the integration of business and finance as well as ecological collaboration. Experimental results showed that the proposed method achieved a stable transaction throughput of 4,500 TPS and reduced audit time by 79.8%. In practical application testing, the proposed method achieved stable financial error and audit variance rates of 0.3% and 0.2%, respectively, with storage costs that stabilized at approximately 0.5 yuan/GB/month. The blockchain-based platform for enterprise financial information sharing and encrypted computation proposed in this study demonstrates high data quality, compliance, and security. This platform effectively solves the problems of low security, inefficiency, and non-compliance.
With the development of technology and economy, unmanned delivery vehicles are being applied across multiple industries. To more accurately design optimal delivery routes and reduce time and energy consumption, this paper builds a route optimization model by integrating an improved Q-learning algorithm with an enhanced, lightweight laser-based odometry and mapping algorithm. The innovation of this study lies in the integrating multiple algorithms for route optimization. By augmenting the Q-learning algorithm with a simulated annealing algorithm and an improved reward mechanism, the approach effectively avoids local optima while enhancing global search capabilities. This advancement overcomes key limitations of traditional reinforcement learning, enabling the improved algorithm to outperform deep Q-networks and random reward reinforcement learning methods demonstrating faster convergence, stronger learning capacity, and better adaptability. Furthermore, this study incorporates data from an inertial measurement unit and an extended Kalman filter to optimize the lightweight LiDAR-based ranging algorithm, significantly improving the system’s stability and positioning accuracy. Through these integrations, the study establishes a high-performance route optimization model that avoids local optima, maintains enhanced stability, and exhibits superior environmental adaptability, making it particularly suitable for logistics enterprises and urban planning departments. This study improves the original algorithm by incorporating inertial measurement unit data for better mapping and pose estimation. Subsequently, Simulated Annealing (SA)and a modified reward mechanism are then applied to optimize Q-learning for path generation. The proposed model generates routes that are 23 m and 61 m shorter than those of the two comparison models, respectively. In addition, the Root Mean Square Error (RMSE) of the proposed model is 0.32, which is lower than those of the baseline models, demonstrating higher accuracy and better fit. These results indicate that the proposed model performs well in predicting optimal routes. It offers significant advantages in route optimization and can reliably analyze and predict complex road conditions, supporting the safe operation of unmanned delivery vehicles.
Long-term water conservancy engineering projects face safety issues during operation, and effective assessment is key to ensuring stable operation. Therefore, this study proposes an intelligent digital evaluation technology for water conservancy engineering based on digital twin technology. The study uses a cloud model to quantify safety indicators and utilizes three parameters, namely expectation, entropy, and super entropy, to establish a two-way mapping between qualitative concepts and quantitative values. This mapping is used to construct a four-level safety status discrimination criterion. Second, the study combines an improved mutation series method with deep learning models, integrating physical interpretability with data-driven advantages to construct a hybrid evaluation framework. In a case study of a certain water conservancy hub, the cloud model achieved 100% consistency between its evaluation results for 32 indicators, including horizontal displacement and uplift pressure, and the actual conditions, outperforming comparable technologies. In water conservancy parameter identification, the research model had an error rate of 0.017, with an anomaly detection accuracy of 97.25%, demonstrating the best performance. Finally, an overall assessment of water conservancy engineering was conducted, with a total mutation level of 0.93 to 0.95, all of which were classified as “normal”. In summary, this research technology has good application results and provides more efficient and reliable technical support for digital twin security assessment for water conservancy engineering.
To address the limitations of traditional warehousing systems in coping with order fluctuations and unexpected events, resulting in an imbalance between inventory and cost control, this study proposes an intelligent warehousing management system that integrates the Ant Colony Genetic Algorithm (ACGA) and the Internet of Things (IoT). To enhance the intelligence of warehouse management, this study selected Ant Colony Optimization for path planning and Genetic Algorithm (GA) to enhance global search capability. The combination of the two forms ACGA to improve convergence performance. The system simultaneously integrates IoT to enable real-time data collection and dynamic feedback, compensating for the algorithm's shortcomings in sensing unexpected events. By deeply integrating ACGA with IoT, this study constructs an intelligent warehouse management system that enables integrated, closed-loop operation of perception and scheduling. Experimental results show that the root mean square error of path planning in scheduling tasks is 62.3m. In terms of accuracy, the proposed system improves from 36.9% to 96.7%, and the F1-score stabilizes at 0.97 by the end of iterations. In a simulated cross-border e-commerce warehousing management environment, the system availability increases to 94.2% after 100 iterations. Regarding overall performance, the response delay is 25ms, and the computational load rate reaches 68%, helping to avoid resource idleness and system overload. At the same time, the energy consumption is only 18 W·h, the task completion rate reaches 98.5%, and the system stability is 99.2%. These results indicate that the proposed system significantly outperforms the comparison system in comprehensive performance and meets the optimization needs in complex cross-border e-commerce task environments, offering a feasible technical solution for intelligent warehousing management.
The textile industry is a pillar of the national economy. Due to the fragmentation of production factors, insufficient capture of spatiotemporal characteristics, and slow response to dynamic adjustments, current clothing production faces problems such as inefficient process coordination and lagging dynamic scheduling. These issues manifest as a high proportion of waiting time for process connection and low equipment utilization. The study combines knowledge graph with graph spatiotemporal attention network to construct a production process model for textile and clothing industry and conducts research on production process prediction. The results indicated that the average accuracy of the proposed model kept rising and finally neared 1.00, which was notably superior to that of other comparative models. The average precision of the proposed model ultimately exceeded 0.95, accurately matching the process collaboration requirements for multi-variety production. For the proposed model, its mean absolute error remained stable within a range of 0.15. The prediction accuracy of different production processes ranged from 1.87 to 2.36, with a root mean square error of 2.87 to 3.52 and an R2 of 0.91 to 0.95. The prediction error was reduced by 33.8%-49.2%. The research method can effectively model and predict the clothing production process, providing accurate and efficient decision support for the textile industry's clothing production, and improving production efficiency and quality.
When investing in waste disposal projects, local government faces a critical dilemma: how to move beyond short-term cost evaluation to long-term environmental and economic benefits. Therefore, this research focuses on local government waste disposal projects, using a life cycle cost analysis to construct a comprehensive evaluation model that integrates energy, environment, and economy, and to analyze the system performance and comprehensive benefits of the proposed technological transformation plan. The results indicated that the total life cycle cost of processing one ton of tail vegetables was -97.05 Chinese Yuan (CNY), of which the internal cost was -85.04 CNY and the external environmental cost was -12.01 CNY. Sensitivity analysis revealed that government subsidies and processing benefits were the most sensitive factors affecting total costs. In terms of environmental impact, this technology route generated significant benefits across key indicators, including global warming potential and acidification potential, through biogas recovery and utilization. Its global warming potential reached -49.88 kg CO2 eq/t, significantly better than that of the original project. The energy analysis confirms that the system is a net energy producer, but the amount of biogas generated is the most sensitive parameter for its energy performance. This model effectively evaluates the long-term operational efficiency of waste disposal projects, provides a quantitative basis for local governments in project planning, process comparison, policy formulation, and improves the accuracy and sustainability of public investment
With the accelerating global transition toward green manufacturing and ecological sustainability, establishing an efficient environmental risk assessment framework has become a critical requirement for sustainable industrial transformation. The study develops an environmental risk assessment system for the manufacturing industry based on ecological efficiency. By combining the Analytic Hierarchy Process (AHP)and the Fuzzy Comprehensive Evaluation (FCE) method. Additionally, the Particle Swarm Optimization (PSO) algorithm with the Genetic Algorithm (GA) were integrated into a unified multi-objective optimization framework, enabling the coordinated enhancement of both economic and environmental performance. Experimental results demonstrate that the proposed system achieves the shortest running time under small-scale enterprise conditions, approximately 9 seconds initially and under 20 seconds at 30 enterprises, with stable growth across larger scales. For a scale of 90-enterprises, the runtime remains about 42 seconds. It also exhibited the fewest convergence iterations across industries at around 30 for electronics and new energy sectors, with as few as 28 for new energy manufacturing, and the highest risk classification accuracy, exceeding 0.95. These findings emphasize the novelty of integrating multiple decision-making and optimization techniques into a cohesive eco-efficiency optimization framework, demonstrating significant potential for green decision-making, environmental risk control, and sustainable industrial development.
To address the high energy consumption and emissions of Cold Chain Logistics (CCL), this study proposes a distribution path optimization model based on an improved ant colony optimization algorithm. The model considers transportation, vehicle usage, energy, and carbon emission costs to achieve a balance between cost, carbon reduction, and distribution efficiency. Results indicated that the enhanced algorithm reduces the mean square error and computation time by 48.39% and 14.73%, respectively. The optimal path was found to lower cargo damage by 31.82% and improve customer satisfaction by 7.52%. Additionally, the model decreases the transportation, vehicle, energy, and carbon emission costs by 14.53%, 12.99%, 13.85%, and 17.41%, respectively. The results demonstrate that the proposed CCL distribution optimization model effectively balances route connectivity, low-carbon objectives, and customer satisfaction. It provides quantitative evidence to help governments establish green logistics subsidy standards and enables enterprises to fulfill their carbon reduction responsibilities. These findings are of practical significance for advancing policy implementation and promoting sustainable development within the CCL industry
To effectively identify Financial Fraud (FF) in Listed Companies (LC), the study combined financial, non-financial, and unstructured data to develop an initial set of financial indicators. To further enhance the fraud detection model’s efficacy, the indications were screened using chi-square tests and correlation coefficients. Convolutional Neural Networks (CNN) and bidirectional long short-term memory networks were merged in the study’s model development, and an attention mechanism was added to emphasize important details. The outcome revealed that the average accuracy of the research design identification model was 97.48%, which was much higher than that of the comparison models (82.56%, 88.17%, 90.13%, and 91.17%). In addition, the maximum precision of this model was 98.49% and the average recall rate was 97.19%, both of which were superior to those of the comparison models. In summary, models that integrate multi-source data are better able to identify financial fraud in LC and provide strong technical support for maintaining the normal order of the securities market.
As educational reforms advance, optimizing teaching evaluation systems is crucial for improving educational quality. However, traditional evaluation systems typically rely on fixed weighting methods, which fail to adapt to the dynamic changes in evaluation indicators, leading to a lack of precision in results. To address this, this paper proposes an optimization method based on a combined weighting approach integrating the entropy weighting method with state variable weighting vectors. This method introduces state variable weighting theory to dynamically correct the initial entropy weights derived from data dispersion, constructing a new evaluation model using the Hadamard product. Experimental verification demonstrates that, compared to traditional methods, the proposed approach exhibits higher distinctiveness and stability when processing data across multiple disciplines and educational stages, effectively capturing dynamic characteristics in the teaching process. The evaluation system constructed in this study not only overcomes the limitations of static weighting but also provides a scientific reference for educational administrators to optimize resource allocation and enhance teaching quality.
In response to the problem that traditional classroom student behavior analysis relies on manual observation and is difficult to achieve real-time, precise and personalized intervention. This paper constructs an end-to-end intelligent classroom intervention system based on deep reinforcement learning. This system adopts the Multi-Modal Fusion Spatio-Temporal Graph Convolutional Network (MM-ST-GCN), integrating visual skeletons, seat pressure and classroom interaction data, to achieve fine-grained and high-precision recognition of students' classroom behaviors. It models the classroom environment as a Partially Observable Markov Decision Process (POMDP) and uses the improved Soft Actor Critic (SAC) algorithm to generate the optimal intervention strategy that takes into account the learning benefits of students and the intervention costs of teachers. Experiments on the MMAct dataset show that the proposed behavior recognition model achieves 93.8% accuracy and a macro-average F1 score of 0.925. Simulation experiments indicate that the system has the potential to improve students' concentration and reduce distraction behavior. However, the above results are derived from the simulated environment and need to be further verified in the real classroom.
The rapid advancement of generative artificial intelligence has enabled significant breakthroughs in the visual realism and artistic expressiveness of AI-generated artworks. However, challenges persist in objectively evaluating their visual quality and emotional impact. Existing evaluation methods either focus on underlying image quality or rely on subjective user surveys, lacking a unified computational framework. This paper proposes a bimodal, multi-dimensional, and interpretable computational evaluation method. The Visual Quality and Emotion (VQ-Emo) framework is designed to jointly optimize visual quality scores and multidimensional emotional predictions. The framework comprises three core modules: a visual quality assessment module (a multi-branch convolutional neural network incorporating style perception, quantifying composition, color, texture, and lighting); an emotional impact computation module (a 20-dimensional multi-label emotion classifier based on the VAWE emotion model), and a visual-emotion association module (using attention mechanisms to identify emotion-driven visual regions). This model was trained and validated on the AGIQA-1K, ArtEmis, and self-constructed VAWE-Art datasets. Experimental results demonstrate that VQ-Emo outperforms existing methods in both visual quality assessment and emotion recognition, achieving an SRCC of 0.912 and a mAP of 0.678. Key contributions include proposing a multi-task learning framework for jointly optimizing visual quality and emotion, establishing the inaugural VAWE-Art dataset comprising 5,000 AI-generated images with 20-dimensional emotional annotations. This reveals quantifiable correlations between visual features and emotional responses across diverse artistic styles and provides computational foundations for emotion-controllable generative art systems.
