Risun (China)

Parkinson’s disease (PD) is an incurable neurological disorder that often begins insidiously with sleep disturbances and somatic symptoms, progressing to whole-body motor and cognitive symptoms1–5. Dysfunction of the somato-cognitive action network (SCAN)—which is thought to control action execution6,7 by coordinating arousal, organ physiology and whole-body motor plans with behavioural motivation—is a potential contributor to the diverse clinical manifestations of PD. To investigate the role of the SCAN in PD pathophysiology and treatments (medications, deep-brain stimulation (DBS), transcranial magnetic stimulation (TMS) and MRI-guided focused ultrasound stimulation (MRgFUS)), we built a large (n = 863), multimodal, multi-intervention clinical imaging dataset. Resting-state functional connectivity revealed that the substantia nigra and all PD DBS targets (subthalamic nucleus, globus pallidus and ventral intermediate thalamus) are selectively connected to the SCAN rather than to effector-specific motor regions. Importantly, PD was characterized by specific hyperconnectivity between the SCAN and the subcortex. We therefore followed six PD cohorts undergoing DBS, TMS, MRgFUS and levodopa therapy using precision resting-state functional connectivity and electrocorticography recording. Efficacious treatments reduced SCAN-to-subcortex hyperconnectivity. Targeting the SCAN instead of effector regions doubled the efficacy of TMS treatments. Focused ultrasound treatment benefits increased when the target was closer to the thalamic SCAN sweet spot. Thus, SCAN hyperconnectivity is central to PD pathophysiology and its alleviation is a hallmark of successful neuromodulation. Targeting functionally defined subcortical SCAN nodes may improve existing therapies (DBS, MRgFUS), whereas cortical SCAN targets offer effective non-invasive or minimally invasive neuromodulation for PD.

The performance of photovoltaic (PV) modules can be assessed using several models, with the most common being the Osterwald model, which states PV power is proportional to solar irradiance. However, the models based on electrical equivalent circuits, which have a variable number of parameters affected by environmental conditions like irradiance and temperature, can be used to increase the accuracy of PV power estimation. Actually, understanding these parameters in any weather condition can help predict PV generators' performance with higher accuracy than commonly used analytical models. This study proposes a method for evaluating PV power and evaluates its application to 10 PV modules tested at two different installation sites (Turin, Italy, and Jaén, Spain). The devices under test have different technologies, rated power (up to 450 W) and efficiency (up to <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\approx$</tex-math></inline-formula>23%). Such a technique is based on the determination of the single diode model parameters, obtaining normalized root mean square errors in the current-voltage curves tracing lower than 1% for any condition. Moreover, its performance is validated by comparing the energy estimation with that of most common models used in the literature. In particular, the performance of the model is excellent for the PV modules with the newest technology, achieving accuracy improvements in energy estimation higher than 70% with respect to the other models.

ABSTRACT This study investigated the low‐velocity impact behavior of carbon fiber‐reinforced composite laminates through both experimental testing and finite element simulation. One‐, two‐, and three‐time low‐velocity impact tests were conducted using a drop hammer testing machine, and corresponding finite element simulations were performed in ABAQUS. Digital image correlation was employed to capture the damage evolution and deformation behavior during impact. The effects of impact number and impact energy on the mechanical response and failure mechanisms of the laminates were analyzed. The experimental and simulation results demonstrate that both parameters significantly influence the impact response of the composites. At impact energies of 16.17 and 18.865 J, the peak impact force increased with the number of impacts. However, at 21.56 J, the peak force during the third impact decreased due to severe damage accumulation from the second impact. These findings provide valuable data and insights for the analysis and structural design of laminates subjected to multiple low‐velocity impacts.

Machine Learning Operations (MLOps) extends Development and Operations (DevOps) practices to manage the end-to-end lifecycle of machine learning models. However, the alignment between MLOps and the traditional software development lifecycle (SDLC) remains insufficiently defined. This study presents a systematic literature review of peer-reviewed works published between 2018 and 2025 in IEEE Xplore, Scopus, and the ACM Digital Library. The review was conducted following a protocol for evidence-based software engineering. A total of 71 primary studies were included after deduplication and screening. A phase-based taxonomy of MLOps tools was synthesized and an SDLC-MLOps compatibility framework was derived to classify alignment levels as high, medium, or low across development phases. Inter-rater agreement, measured using Cohen's Kappa, yielded a value of 0.91, while the average study quality score was 0.73. The results show convergence around pipelines, model registries, and containerized deployment, alongside persistent fragmentation in process models and organizational practices. The findings provide implications for practice, including maturity-based adoption strategies, role definition, governance mechanisms, and continuous monitoring. Additionally, opportunities for future research are identified in the areas of ML-specific testing and harmonized metrics across software engineering and data science disciplines.

Key-value (KV) cache is a crucial optimization for large language model (LLM) serving, particularly in long-context inference scenarios. While existing KV stores suffer from a fundamental mismatch between the CPU-centric KV cache storage and the GPU-accelerated computation: (1) the low-parallelism CPUs cannot satisfy the fine-grained, highly parallel I/O demand, and (2) the synchronization between CPUs and GPUs introduced by layerwise asynchronous KV operations prevents the adoption of low-level optimizations (such as CUDA Graph).

Abstract The proportion of new energy power generation installed capacity in the new power system is getting higher and higher, which leads to the gradual reduction of the support capacity of the grid voltage and the increase of the risk of system instability. Aiming at this problem, this paper establishes an optimal configuration model of energy storage considering the improvement of the voltage support capacity of the new energy base. Firstly, by analyzing the influencing factors of the voltage support strength of the power system, the generalized operating short circuit ratio is used to characterize the voltage support strength of the new energy base, and the mechanism model of the energy storage equipment on the voltage support strength of the new energy base is established. Then, to maximize the generalized operating short-circuit ratio(GOSCR), an optimal configuration model of energy storage considering the improvement of voltage support capacity of the new energy base is established. Finally, the effectiveness of the optimal configuration model of energy storage proposed in this paper is verified by an example simulation. Compared with the new energy base without optimal configuration of energy storage, the model proposed in this paper effectively improves the voltage and power stability of the new energy base.

In response to the limited accuracy of the original UNET in image segmentation, this paper presents innovative optimizations to the loss function, resulting in a significant enhancement of UNET's segmentation accuracy. Through experiments conducted on a retinal vascular dataset, UNET demonstrates outstanding segmentation performance. To address the pronounced imbalance between foreground and background pixels in remote sensing road images, we introduce Dice loss to the UNET model, thereby proposing DC-UNET. Evaluation on the Road_segmentation_ideal dataset reveals that DC-UNET achieves an average improvement of 1.06% in the test set crossover ratio compared to the original model. This research not only enhances the accuracy of road segmentation but also amplifies the practicality and utility of deep learning models in real-world applications such as road detection.

Besides the voltage drop issue by the droop control, dynamic current unbalance due to inertia inconsistency would also happen in multi-parallel energy storage system (ESS) supplied DC charging station. For this, a new multi-phase interleaved-parallel DC converter combined with capacitor-integrated electric spring (C-ES) is developed. And dynamic inertia unified loop and energy-balance-combined dynamic voltage drop compensation control are proposed for this C-ES-combined interleaved-parallel ESS architecture. Dynamic current balance and voltage drop compensation can be simultaneously achieved. In detail, the multi-phase interleaved-parallel structure is taken as the front stage and the C-ES is adopted as the back stage. The working principle is illustrated by analyzing the switching state of the devices one by one. In terms of the control, the dynamic unified inertia loop is designed for the multi-parallel interleaved structure, and the inertia characteristics of each energy unit is reshaped to be consistent and the dynamic current balance can be guaranteed. Thus, the dynamic current balance is achieved by the consistency of physical structure and control performance. Then, an energy-balance-combined dynamic voltage drop compensation control is developed for the C-ES to compensate for both the dynamic and steady-state voltage drop by charging the C-ES. Moreover, the small-signal model is built to reveal the essential mechanism of dynamic current balance and voltage drop compensation. Finally, the effectiveness of C-ES-combined interleaved-parallel ESS architecture on dynamic current balance and voltage drop compensation is validated by simulation and experiment results.

This article explores the application of Large Language Models (LLMs), including proprietary models such as OpenAI’s ChatGPT 4o and ChatGPT 4o-mini, Anthropic’s Claude 3.5 Sonnet and Claude 3.7 Sonnet, and Google’s Gemini 1.5 Pro, Gemini 2.0 Flash, and Gemini 2.0 Flash-Lite, as well as open-source alternatives including Qwen2.5-14B-Instruct-1M, and commercially accessed models such as DeepSeek R1 and DeepSeek V3, which were tested via APIs despite having open-source variants, to automate validation and verification in Application Programming Interface (API) testing within a Service-Oriented Architecture (SOA). Our system compares internal responses from the Enuygun Web Server against third-party API outputs in both JSON and XML formats, validating critical parameters such as flight prices, baggage allowances, and seat availability. We generated 100 diverse test scenarios across varying complexities (1–4 flight results) by randomly altering request and response parameters. Experimental results show that Google Gemini 2.0 Flash achieved high accuracy (up to 99.98%) with the lowest completion time (85.34 seconds), while Qwen2.5-14B-Instruct-1M exhibited limited capability in processing complex formats. Models such as OpenAI’s ChatGPT and Anthropic’s Claude Sonnet models also demonstrated strong performance in single-flight validation scenarios, making them suitable for low-latency, high-precision tasks. Our findings indicate that some open-source models can offer promising cost-effective alternatives, though performance significantly varies. This integration of LLMs reduced manual workload, improved test scalability, and enabled real-time validation across large-scale datasets. As LLM technologies mature, we anticipate further advances in automation, accuracy, and efficiency in software validation systems.