Generation Process Research Articles

The Investigation of the Application of RAG Technology in the Field of EHR

Electronic Health Records (EHR) play a crucial role in contemporary medical information systems, capturing extensive data in multiple scenarios. This review explores the impact of Retrieval Augmented Generation (RAG) on managing EHR. By integrating retrieval and generation processes, RAG significantly enhances data handling, clinical decision support, and patient management. EHRs contain extensive data like patient histories and diagnostic information, which are becoming increasingly complex. The RAG improves the accuracy and richness of this data processing by using a retrieval module to extract relevant text fragments and a generation module to create precise outputs. Despite its potential, RAG faces challenges such as data inconsistency, privacy concerns, and the need for efficient training. Recent studies highlight the RAG’s effectiveness in summarizing clinical notes and enhancing prediction accuracy, suggesting a promising future in healthcare. However, ongoing research is necessary to optimize the RAG’s application and address these challenges, aiming to transform healthcare delivery effectively.

A Novel Dictionary Learning Algorithm Based on Prior Knowledge for fMRI Data Analysis

ABSTRACTTask‐based functional magnetic resonance imaging (fMRI) has been widely utilized for brain activation detection and functional network analysis. In recent years, the K‐singular value decomposition (K‐SVD) algorithm has gained increasing attention in the research of fMRI data analysis methods. In this study, we propose a novel temporal feature region‐growing constrained K‐SVD algorithm that incorporates task‐based fMRI temporal prior knowledge and utilizes a region‐growing algorithm to infer potential activation locations. The algorithm incorporates temporal and spatial constraints to enhance the detection of brain activation. Specifically, this paper improves the three stages of the traditional K‐SVD algorithm. First, in the dictionary initialization stage, the automatic target generation process with an independent component analysis algorithm is utilized in conjunction with prior knowledge to enhance the accuracy of initialization. Second, in the sparse coding stage, the region‐growing algorithm is employed to infer potential activation locations based on temporal prior knowledge, thereby imposing spatial constraints to limit the extent of activation regions. Finally, in the dictionary learning stage, soft constraints and low correlation constraints are applied to reinforce the consistency with prior knowledge and enhance the robustness of learning for task‐related atoms. The proposed method was validated on simulated and real fMRI data, showing superior performance in detecting brain activation compared with traditional methods. The results indicate that the algorithm accurately identifies activated brain regions, providing an effective approach for studying brain function in clinical applications.

Modeling of Liquefied Natural Gas Cold Power Generation for Access to the Distribution Grid

Cold energy generation is an important part of liquefied natural gas (LNG) cold energy cascade utilization, and existing studies lack a specific descriptive model for LNG cold energy transmission to the AC subgrid. Therefore, this paper proposes a descriptive model for the grid-connected process of cold energy generation at LNG stations. First, the expansion kinetic energy transfer of the intermediate work mass is derived and analyzed in the LNG unipolar Rankine cycle structure, the mathematical relationship between the turbine output mechanical power and the variation in the work mass flow rate and pressure is established, and the variations in the LNG heat exchanger temperature difference, seawater flow rate, and the turbine temperature difference in the cycle system are investigated. Secondly, based on the fifth-order equation of state of the synchronous generator, the expressions of its electromagnetic power, output AC frequency, and voltage were analyzed. Finally, the average equivalent models of the machine-side and grid-side converters are established using a direct-fed grid-connected structure, thus forming a descriptive model of the overall drive process. The ORC model is built in Aspen HYSIS to obtain the time series expression of the torque output of the turbine; based on the ORC output torque, the permanent magnet synchronous generator (PMGSG) as well as the direct-fed grid-connected structure are built in MATLAB/Simulink, and the active power and current outputs of the grid-following-type voltage vector control method and the grid-forming-type power-angle synchronous control method are also verified.

Creating an intermediate energy band to boost the photoelectrochemical efficiency of TiO2 for solar-driven hydrogen production

The utilization of photoelectrochemical processes for hydrogen generation from water and solar energy offers a promising avenue to replace non-renewable energy sources. Nevertheless, achieving high-performance photoanode electrodes poses a significant challenge. In this study, we present the synthesis cerium and chromium-doped TiO2 (CeCrTiO2) is produced through a simple, scalable, and cost-effective hydrothermal method on a fluorine-doped tin oxide (FTO) substrate. the introduction of Ce and Cr impurities is highlighted for its role in creating an impurity band within CeCrTiO2. This impurity band is instrumental in enhancing the separation and movement of electrons and holes, contributing to the improved performance of CeCrTiO2 as a photoanode in photoelectrochemical applications. Hence, the photocurrent density value of CeCrTiO2 is 4.5 times higher than that of bare TiO2. This suggests that CeCrTiO2 exhibits improved efficiency in generating a photocurrent when exposed to light, indicating enhanced photoelectrochemical performance. The amount of hydrogen produced by CeCrTiO2 is noted to be 4.88 times greater than that of bare TiO2. This highlights the material's effectiveness in harnessing solar energy to drive the water-splitting reaction, leading to a higher yield of hydrogen gas. The synthesis of CeCrTiO2 through a hydrothermal method results in a photoanode with significantly enhanced performance, positioning it as a promising candidate for advancing photoelectrochemical processes aimed at replacing non-renewable energy sources.

Idea generation and planning time in second language academic writing: An empirical investigation at Howard College Campus, University of KwaZulu-Natal, Durban, South Africa

Idea generation is a cognitive process which underlies the production of coherent writing. However, little is known about the nature of this process and how it is affected by different preparatory conditions. The current study examines the effects of three planning time conditions: ‘planning time’ (10 minutes), ‘extended planning time’ (20 minutes) and ‘no planning time’ and two task conditions, namely ‘topic given’ and ‘topic and ideas given’, and their effect on the quality and the quantity of idea units produced in the planning notes and essays of 30 English Second Language (ESL) learners at a South African university. The study aims to replicate an earlier study by Ong, and tests four hypotheses: hypotheses 1 and 2 state that an extended planning time has a positive effect on both the quantity and the quality of ideas generated in planning notes (hypothesis 1) and essays (hypothesis 2). Hypotheses 3 and 4 state that additional ideas alongside a topic enhance the idea generation process in both planning notes (hypothesis 3) and essays (hypothesis 4). Our findings do not verify hypothesis (1) as neither in the planning notes nor the essays was the quantity of ideas affected by the planning time conditions. Hypothesis 2 was partially verified as the quality of ideas in the essays — but not in the planning notes — improved as an effect of an extended planning time. These results differ from Ong whose data fully support both hypotheses 1 and 2. Our data falsify hypotheses 3 and 4 as the topic-given condition consistently produced both a better quality and a larger quantity of ideas in the planning notes and in the essays of the participants. This finding concurs with Ong. In conclusion, the attempt at a replication of Ong only partially yielded the same results. Interestingly, the data indicate that the idea-generation process differed between the production of the planning notes and the production of the essays.

Highlighted Diffusion Model as Plug-in Priors for Polyp Segmentation.

Automated polyp segmentation from colonoscopy images is crucial for colorectal cancer diagnosis. The accuracy of such segmentation, however, is challenged by two main factors. First, the variability in polyps' size, shape, and color, coupled with the scarcity of well-annotated data due to the need for specialized manual annotation, hampers the efficacy of existing deep learning methods. Second, concealed polyps often blend with adjacent intestinal tissues, leading to poor contrast that challenges segmentation models. Recently, diffusion models have been explored and adapted for polyp segmentation tasks. However, the significant domain gap between RGB-colonoscopy images and grayscale segmentation masks, along with the low efficiency of the diffusion generation process, hinders the practical implementation of these models. To mitigate these challenges, we introduce the Highlighted Diffusion Model Plus (HDM+), a two-stage polyp segmentation framework. This framework incorporates the Highlighted Diffusion Model (HDM) to provide explicit semantic guidance, thereby enhancing segmentation accuracy. In the initial stage, the HDM is trained using highlighted ground-truth data, which emphasizes polyp regions while suppressing the background in the images. This approach reduces the domain gap by focusing on the image itself rather than on the segmentation mask. In the subsequent second stage, we employ the highlighted features from the trained HDM's U-Net model as plug-in priors for polyp segmentation, rather than generating highlighted images, thereby increasing efficiency. Extensive experiments conducted on six polyp segmentation benchmarks demonstrate the effectiveness of our approach.

High-order brain network feature extraction and classification method of first-episode schizophrenia: an EEG study

IntroductionA multimodal persistent topological feature extraction and classification method is proposed to enhance the recognition accuracy of first-episode schizophrenia patients. This approach addresses the limitations of traditional higher-order brain network analyses that rely on single persistent features (e.g., persistent images).MethodsThe study utilized resting-state EEG data from 198 subjects recruited at Huilongguan Hospital in Beijing, comprising 102 males and 96 females, with a mean age of 30 years and mean education of 14 years. Persistent topological features were extracted using adaptive thresholding during persistent homology (PH) filtrations. The distribution of these features was visualized through heatmaps and persistence entropies, while the generation process was elucidated using Betti curves and persistence landscapes.ResultsThe classification performance of the multimodal persistent topological features was assessed using various machine learning classifiers. The classifier yielding the highest performance was selected for comparison with traditional brain network features derived from graph theory and single persistent topological features. The results revealed significant topological changes in first-episode schizophrenia patients throughout the persistent homology filtering compared to healthy subjects. The univariate feature selection algorithm achieved a classification accuracy of 94.6% with a combination of attributes meeting the criterion of AC ≥ 0.6.DiscussionThe proposed method demonstrates clinical significance for the early identification and diagnosis of first-episode schizophrenia patients, offering a new research perspective for constructing higher-order functional connectivity networks and extracting topological structure features.

The effect of cinnamic acid derivative on the activity of mitochondrial enzymes in brain tissue under conditions of experimental Parkinson's disease

The aim: to evaluate the effect of cinnamic acid derivative on changes in mitochondrial enzymes activity in rats brain tissue under conditions of experimental Parkinson's disease. Materials and methods. Parkinson's disease was modeled in male Wistar rats by direct injection of rotenone solution (5 mg/ml) into the striatum. The analyzed compound 4-hydroxy-3,5-di-tretbutyl cinnamic acid and the reference ethylmethylhydroxypyridine succinate were administered orally in equivalent doses (100 mg/kg) for 30 days from the moment of pathology modeling. Next, a brain supernatant was obtained by differential centrifugation, in which changes in the activity of enzymes: citrate synthase, succinate dehydrogenase, cytochrome c oxidase and aconitase were evaluated. The obtained results were processed statistically. During the analysis, the StatPlus 7.0 application software package was used. Results. During the study, it was found that the administration of 4-hydroxy-3,5-di-tretbutyl cinnamic acid to rats increased the activity of citrate synthase, succinate dehydrogenase, cytochrome c oxidase and aconitase relative to untreated animals by 109.7% (p0.05); 53.6% (p0.05); 65.0% (p0.05) and 63.1% (p0.05), respectively, whereas against the background of the use of the reference, the activity of these enzymes increased by 56.3% (p0.05); 57.7% (p0.05); 71.7% (p0.05) and 49.1% (p0.05), respectively. At the same time, the activity of citrate synthase in animals treated by 4-hydroxy-3,5-di-tretbutyl cinnamic acid was higher than that in rats treated by the reference by 34.2% (p0.05). Conclusions. The study showed that the course administration of 4-hydroxy-3,5-di-tretbutyl cinnamic acid to animals with experimental Parkinson's disease is accompanied by an increase in the activity of mitochondrial enzymes, which may reflect the significant effect of this compound on the processes of mitochondrial biogenesis, mitophagy and generation of mitochondrial reactive oxygen species.

Enhanced Chaotic Pseudorandom Number Generation Using Multiple Bernoulli Maps with Field Programmable Gate Array Optimizations

Certain methods for implementing chaotic maps can lead to dynamic degradation of the generated number sequences. To solve such a problem, we develop a method for generating pseudorandom number sequences based on multiple one-dimensional chaotic maps. In particular, we introduce a Bernoulli chaotic map that utilizes function transformations and constraints on its control parameter, covering complementary regions of the phase space. This approach allows the generation of chaotic number sequences with a wide coverage of phase space, thereby increasing the uncertainty in the number sequence generation process. Moreover, by incorporating a scaling factor and a sine function, we develop a robust chaotic map, called the Sine-Multiple Modified Bernoulli Chaotic Map (SM-MBCM), which ensures a high degree of randomness, validated through statistical mechanics analysis tools. Using the SM-MBCM, we propose a chaotic PRNG (CPRNG) and evaluate its quality through correlation coefficient analysis, key sensitivity tests, statistical and entropy analysis, key space evaluation, linear complexity analysis, and performance tests. Furthermore, we present an FPGA-based implementation scheme that leverages equivalent MBCM variants to optimize the electronic implementation process. Finally, we compare the proposed system with existing designs in terms of throughput and key space.

Proof-of-principle experiment on quasi-phase matching of high-order harmonic generation via the selected-zoning method with a transverse disruptive pulse

We demonstrate the proof-of-principle experiment on all-optical quasi-phase matching (QPM) of high-order harmonic generation (HHG) via the transverse selective-zoning method. The dephasing length and the out-of-phase region in the interacting medium are identified in advance by applying a tomographic diagnosis of the generation process. Then a transverse disruptive pulse is employed to suppress the harmonic emission in the out-of-phase region to achieve QPM. The final output is enhanced by 35% with just one QPM zone. This proof-of-principle result provides the feasibility of ion-based HHG for higher cut-off energies, where phase matching is difficult to achieve.

Field study on proactive pipe condition assessment using hydroacoustic noise

Condition assessment of water transmission pipelines is of important necessity for prioritizing rehabilitation and preventing catastrophic pipe failure. Hydraulic transient analysis has been investigated for three decades for this purpose and applied to many field studies. However, the significant pipe vibrations caused by the transient generation process and the limited energy and bandwidth of conventional discrete transient pressure waves (e.g., step and pulse waves) limit the accuracy and resolution of the analysis. This paper reports a pilot field study of noninvasive and nondestructive pipe condition assessment using small amplitude and persistent hydroacoustic noise instead of conventional large and discrete hydraulic transient waves. By opening a discharge valve installed at an existing access point, such as an air valve, on the field pipe, hydroacoustic noise was generated at the valve due to the turbulence of the discharge. The hydroacoustic noise was measured by pressure transducers at this access point as well as some other access points along the field pipe. A signal deconvolution process was then applied to the measured hydraulic pressures to transfer them to a compositional impulse response function (IRF), which is composed of IRFs of different pipe sections. A mathematical model was derived to interpret the anomaly-induced spikes on the deconvolution trace. A time-shifting process on the compositional IRFs was proposed to find the locations with pipe condition changes from a specific direction of the pipe. The field study has shown that small-amplitude persistent hydroacoustic noise can replace conventional large hydraulic transient waves for pipe condition assessment.

Geological and geochemical aspects of modeling of hydrocarbon systems

Forecasting the phase-genetic types of hydrocarbons is the most important stage in modelling geological hydrocarbon systems. This is because accurate qualitative and quantitative assessment of potential reserves is necessary for making informed investment decisions in exploration activities. Improving the precision of qualitative and quantitative oil and gas forecast can help reduce geological risks in asset development and increase investment efficiency. Therefore, technologies that allow modelling of hydrocarbon systems are becoming ever more important. This study aims to explore the key geological and geochemical aspects of hydrocarbon system modelling. Hydrocarbon systems refer to interconnected elements and geological processes influencing hydrocarbon formation in sedimentary basins. The study emphasizes that the "oil window" concept, or the theory of hydrocarbon formation phases, provides the theoretical basis for hydrocarbon systems. This concept consistently encompasses the entire history of sedimentary basin development, the evolution of thermal and thermobaric regimes in sedimentary deposits, the transformation of organic matter, hydrocarbon generation, migration, accumulation, and preservation processes. Using the results of modelling of hydrocarbon systems to evaluate geological risks and the economic impact of exploration activities can be a powerful instrument for companies that make investment decisions in geological exploration.

A Method for Generating Toolpaths in the Manufacturing of Orthosis Molds with a Five-Axis Computer Numerical Control Machine

This paper proposes a new algorithm for the automatic generation of toolpaths for machining complex geometric positions, such as molds used in orthosis production. The production of individualized orthoses often requires the use of multi-axis machining systems, such as five-axis machines or industrial robots. Typically, complex and expensive CAD/CAM systems are used to generate toolpaths for these machines, requiring the definition of a machining strategy for each surface. While this approach can achieve a reliable and high-quality machining process, it is very time-consuming and makes it challenging to meet the criteria for rapid production of orthopedic aids. Given that their production is a custom-made process using individual shapes as inputs, the toolpath generation process becomes even more demanding. To address these challenges, this paper proposes an algorithm suitable for the automatic generation of toolpaths for such complex positions. The proposed algorithm has been tested and has proven to be robust and applicable.

Automatic animal monitoring systems based on AutoML technologies

In contemporary agricultural contexts, the sector is experiencing active processes of automation, underscoring the need for effective tools to develop such automated systems. The utilization of automation tools enhances the efficiency of numerous processes, including those in the domain of animal monitoring. This article examines the application of the AutoML approach as a means for automating the process of generating deep learning models employed in automatic monitoring systems. The VGG19 architecture has been chosen as a testbed for demonstrating the capabilities of the developed technologies. This well-established architecture for deep learning models is designed for object recognition in images.The present study implements a technology for automated structural-parametric synthesis of VGG19 models and the optimization of their hyperparameters. Such an approach allows for the automated creation of models tailored to specific applied problems, even for users lacking specialized knowledge in deep learning.The system delineated in this work is developed on the AutoGenNet software platform, which embodies the No-Code development concept. This concept conceals complex aspects of model creation and training processes from users, significantly lowering the entry barrier for newcomers. Additionally, the AutoGenNet platform incorporates a mechanism for the automatic generation of software wrappers, facilitating efficient interaction with trained models.All aforementioned aspects have contributed to the effective implementation of the AutoML approach for automating the generation and training processes of the VGG19 model. Consequently, the processes associated with solving automatic monitoring tasks reliant on deep learning models have been significantly simplified and expedited.The developed system has been tested on the task of recognizing individual cows. Test results indicated that the system possesses a high degree of scalability and can be adapted for the automated generation of other object recognition models, thereby opening avenues for addressing a diverse array of applied challenges related to the monitoring of various animal species.

Mechanisms of Reservoir Space Preservation in Ultra-Deep Shales: Insights from the Ordovician–Silurian Wufeng–Longmaxi Formation, Eastern Sichuan Basin

This study aims to explore the reservoir characteristics and formation mechanisms of ultra-deep shale gas in the Ordovician–Silurian Wufeng–Longmaxi Formation in the Sichuan Basin in order to provide theoretical support and practical guidance for the exploration and development of ultra-deep shale gas. With recent breakthroughs in ultra-deep shale gas exploration, understanding its organic matter development, mineral composition, and reservoir space characteristics has become particularly important. The background of this research lies in the significant potential of ultra-deep shale gas, which remains inadequately understood, necessitating an in-depth analysis of its pore structure and reservoir quality. Through a systematic study of the ultra-deep shale in well FS1 of Sichuan Basin, that the following was found: (i) The ultra-deep shale in the Wufeng–Longmaxi Formation is mainly composed of quartz and clay minerals, exhibiting high total organic carbon (TOC) and high porosity characteristics, indicating it is in an overmature thermal evolution stage. (ii) Organic pores and microcracks in the ultra-deep shale are more developed compared to middle-shallow and deep shale, forming a complex pore structure that is conducive to gas storage. (iii) In the diagenesis process, the dissolution and recrystallization of the biogenic skeleton promote the cementation between autogenetic quartz particles, forming a rigid skeleton that effectively inhibits the impact of mechanical compaction. (iv) The overpressure environment created by the hydrocarbon generation process, along with gas production from hydrocarbon cracking, can effectively offset the mechanical compaction of overburden pressure on micropores, and this overpressure environment also promotes the further development of microfractures, which is beneficial for the development and preservation of ultra-deep shale pores. In summary, this study not only reveals the reservoir characteristics and formation mechanisms of ultra-deep shale but also provides essential references for the exploration and development of ultra-deep shale gas in the Sichuan Basin and similar regions, emphasizing the ongoing significance of research in this field.

Statistical analysis of repertoire data demonstrates the influence of microhomology in V(D)J recombination.

Humans rely on diverse adaptive immune receptor repertoires to effectively defend against infections. The receptor generation process, known as V(D)J recombination, is designed to create this diversity by stochastically joining V(D)J gene segments and modifying their junctions through nucleotide deletions and insertions. Previous studies, conducted in vitro, have suggested that short stretches of homologous nucleotides between gene segments can bias these recombination steps. In this study, we explore the extent to which these homologous nucleotides influence V(D)J recombination in humans using statistical inference on large-scale receptor repertoire sequencing data. Our findings reveal that microhomology significantly biases several recombination steps, with important practical implications for the analysis, processing, and interpretation of receptor sequences.

Dynamic Retrieval Augmented Generation of Ontologies using Artificial Intelligence (DRAGON-AI)

BackgroundOntologies are fundamental components of informatics infrastructure in domains such as biomedical, environmental, and food sciences, representing consensus knowledge in an accurate and computable form. However, their construction and maintenance demand substantial resources and necessitate substantial collaboration between domain experts, curators, and ontology experts.We present Dynamic Retrieval Augmented Generation of Ontologies using AI (DRAGON-AI), an ontology generation method employing Large Language Models (LLMs) and Retrieval Augmented Generation (RAG). DRAGON-AI can generate textual and logical ontology components, drawing from existing knowledge in multiple ontologies and unstructured text sources.ResultsWe assessed performance of DRAGON-AI on de novo term construction across ten diverse ontologies, making use of extensive manual evaluation of results. Our method has high precision for relationship generation, but has slightly lower precision than from logic-based reasoning. Our method is also able to generate definitions deemed acceptable by expert evaluators, but these scored worse than human-authored definitions. Notably, evaluators with the highest level of confidence in a domain were better able to discern flaws in AI-generated definitions. We also demonstrated the ability of DRAGON-AI to incorporate natural language instructions in the form of GitHub issues.ConclusionsThese findings suggest DRAGON-AI's potential to substantially aid the manual ontology construction process. However, our results also underscore the importance of having expert curators and ontology editors drive the ontology generation process.

Experimental and performance analysis of organic fraction of municipal solid waste in downdraft gasifier

It is anticipated that the production of municipal solid waste will increase from 2.10 billion tonnes in 2023 to 3.80 billion tonnes by 2050. About 170 billion metric tonnes of biomass is produced worldwide each year. On a wet basis, the composition of municipal solid waste (MSW) includes approximately 20% of wood, 70% of food waste, 60% of yard trimmings, and 50% of other organic materials. Gasification of waste biomass is one of the most efficient processes for creating gaseous fuel and subsequently power with reduced environmental impact. The current work presents a novel strategy to reduce MSW by converting waste into briquettes and feeding it in gasifiers to produce syngas. The downdraft gasifier is employed in the syngas generation process. With a calorific value of 22.6 MJ/kg and a higher lignin concentration of approximately 31%, MSW proves to be suitable for thermochemical conversion processes. This article presents the results during gasification of briquette made from organic fraction of municipal solid waste (OFMSW) with significant heating value of 22.6 MJ/kg. The fuel produced by the gasifier has a heating value ranging from 920 to 1150 kcal/m3, making it clean-burning. Its primary components consist of 19% to 23% carbon monoxide and 17% to 21% hydrogen. This study offers a unique investigation into the gasification behavior of the organic fraction of OFMSW in a downdraft gasifier, with an emphasis on optimizing operational parameters and evaluating the yield and quality of syngas. Although gasification has been widely studied as a thermochemical conversion method for biomass, the specific performance of OFMSW in this process has received limited attention. By focusing on this underexplored area, the study enhances the understanding of OFMSW's potential in gasification and its contribution to sustainable energy production, thereby addressing key issues in both waste management and renewable energy generation.

Exploiting Contextual Embeddings in Hierarchical Topic Modeling and Investigating the Limits of the Current Evaluation Metrics

Abstract We investigate two essential challenges in the context of Hierarchical Topic Modeling (HTM) – (i) the impact of data representation and (ii) topic evaluation. The Data representation directly influences the performance of the topic generation, and the impact of new representations such as contextual embeddings in this task has been under-investigated. Topic evaluation, responsible for driving the advances in the field, assesses the overall quality of the topic generation process. HTM studies exploit the exact topic modeling (TM) evaluation metrics as traditional TM to measure the quality of topics. One significant result of our work is demonstrating that the HTM’s hierarchical nature demands novel ways of evaluating the quality of topics. As our main contribution, we propose two new topic quality metrics to assess the topical quality of the hierarchical structures. Uniqueness considers topic topological consistency, while the Semantic Hierarchical Structure (SHS) captures the semantic relatedness of the hierarchies. We also present an additional advance to the state-of-the-art by proposing the c-CluHTM. To the best of our knowledge, c-CluHTM is the first method that exploits contextual embeddings into NMF in HTM tasks. c-CluHTM enhances the topics’ semantics while preserving the hierarchical structure. We perform an experimental evaluation, and our results demonstrate the superiority of our proposal with gains between 12% and 21%, regarding NPMI and Coherence over the best baselines. Regarding the newly proposed metrics, our results reveal that Uniqueness and SHS can capture relevant information about the structure of the hierarchical topics that traditional metrics cannot.

U-Net Semantic Segmentation-Based Calorific Value Estimation of Straw Multifuels for Combined Heat and Power Generation Processes

This paper proposes a system for real-time estimation of the calorific value of mixed straw fuels based on an improved U-Net semantic segmentation model. This system aims to address the uncertainty in heat and power generation per unit time in combined heat and power generation (CHPG) systems caused by fluctuations in the calorific value of straw fuels. The system integrates an industrial camera, moisture detector, and quality sensors to capture images of the multi-fuel straw. It applies the improved U-Net segmentation network for semantic segmentation of the images, accurately calculating the proportion of each type of straw. The improved U-Net network introduces a self-attention mechanism in the skip connections of the final layer of the encoder, replacing traditional convolutions by depthwise separable convolutions, as well as replacing the traditional convolutional bottleneck layers with Transformer encoder. These changes ensure that the model achieves high segmentation accuracy and strong generalization capability while maintaining good real-time performance. The semantic segmentation results of the straw images are used to calculate the proportions of different types of straw and, combined with moisture content and quality data, the calorific value of the mixed fuel is estimated in real time based on the elemental composition of each straw type. Validation using images captured from an actual thermal power plant shows that, under the same conditions, the proposed model has only a 0.2% decrease in accuracy compared to the traditional U-Net segmentation network, while the number of parameters is significantly reduced by 74%, and inference speed is improved 23%.