Abstract：Intelligent work order systems are core platforms for enterprise digital transformation. However, current systems face challenges in cross-modal redundancy and semantic conflicts caused by multi-source heterogeneous data. Traditional large language models (LLMs) relying on single-modal architectures exhibit high rates of critical information omission and poor data cleaning performance, severely hindering system intelligence. To address this, the Mixture of Experts (MoE) model adaptively distributes multimodal data to specialized expert networks through dynamic routing mechanisms, enhancing cross-modal feature fusion accuracy while optimizing computational efficiency. This paper proposes an MoE-based intelligent processing method for multimodal work order data. Firstly, we design a semantic analysis model using the DeepSeekMoE architecture to achieve feature decoupling and key content extraction from cross-modal data. Secondly, we introduce a Thinker-Talker multimodal feature fusion framework to improve redundancy utilization and semantic consistency. Finally, we develope an unstructured data cleaning and structured form generation algorithm to denoise raw data, enhance semantics, and output standardized work orders. Ablation experiments demonstrate that our method achieves 92.7% information extraction accuracy on a private dataset, with a 36.2% improvement in compliance with processing standards compared to traditional methods, providing an extensible technical paradigm for intelligent work order systems.

Abstract：To improve the passing efficiency of the inspected vehicles, an automatic vehicles inspection method in the fixed region based on differential GPS+multi-line lidar is proposed. Firstly, according to the parking lot environment, the topology diagram is planned, and the electronic map based on fixed weight mark is proposed. Secondly, based on the data of multi-line lidar, the proposed adaptive clustering method is used to detect the targets. Finally, with the GPS robot positioning, the electronic map and target information, the global + local path planning algorithm is proposed to dynamically complete the path planning. Theoretical analysis and test results show that the method can effectively improve the target passing efficiency.

Abstract：Electricity demand forecasting, as a crucial component of energy planning and grid scheduling, plays a significant role in achieving energy system stability and sustainable development. This study proposes an electricity demand forecasting method based on data transformation fusion and model integration. By combining Copula and Yeo-Johnson transformations, the method achieves unified data distribution while preserving feature correlations. Feature importance is analyzed using the XGBoost algorithm, which serves as the basis for constructing an LSTM network model optimized with feature weights. Additionally, Bayesian optimization and thread pool techniques are employed to efficiently optimize hyperparameters, further improving model performance. Experimental results show that the proposed fusion model reduces the root mean squared error (RMSE) by approximately 11.85% compared to traditional LSTM models, significantly enhancing prediction accuracy and generalization ability. This study provides a reliable solution for complex time series forecasting tasks and offers valuable insights for both theoretical research and practical applications in electricity demand forecasting.

Abstract：To improve the measurement accuracy of the rotor position signal for resolver motors, a dual-channel resolver decoding system based on CS32F407 was designed and developed. The system consists of two resolver encoding information acquisition input circuits, two sinusoidal excitation signal output circuits, an MCU control unit, a communication unit, an interface unit, and a power supply circuit. With the CS32F407 MCU as the processing core, data from the coarse and fine resolver decoding chips CAD2S1210 were collected via dual SPI interfaces, and the mechanical shaft angle was calculated using an improved multi-pole resolver decoding algorithm, with position information uploaded to the display and control unit through the USART interface. The system supports a maximum position resolution of 21 bits and an excitation signal frequency of 2 kHz. Prototype verification shows that the position error is less than 0.01°, and operational stability meets environmental requirements ranging from -40℃ to 60℃. The results demonstrate high measurement accuracy, strong anti-interference capability, and applicability to complex scenarios such as aerospace and industrial control.

Abstract：The effects of design parameter on power handling capability of radio-frequency silicon-on-insulator (RF SOI) switch were comprehensively investigated in this work, including the width of Metal-Oxide-Semiconductor Field-Effect Transistor (MOSFET) under different stack, and the Gate/Body bias resistance. The impacts of larger 1st/2nd stage MOSFET width, larger N stage MOSFET width, gradual change MOSFET width and different Gate/Body bias resistance on RF switch insertion loss, isolation and power handling capability were discussed in detail. The result can be reference as RF switch circuit design.

Abstract：The traditional spreading physical design can no longer meet the needs of VLSI physical design, and hierarchical physical design has become the mainstream method of VLSI design. In the process of VLSI hierarchical physical design, the clock tree has a great impact on the overall chip timing convergence. Regarding the issue where the billion-gate hierarchical physical design clock tree cannot read the clock tree delay in the submodule, causing significant actual clock deviation, a solution has been proposed to set the actual clock delay of the submodules during the top-level clock tree ccopt phase. This approach effectively reduces clock deviations, thereby providing a solid foundation for subsequent timing optimizations.

Abstract：As a crucial passive device in the RF front-end, the power divider continues to evolve towards miniaturization and higher power capabilities. A novel Gysel power divider based on SiC IPD technology is proposed. The distributed circuit is converted into a lumped topology using π-shaped and Bridged T-coil equivalent circuits, significantly achieving miniaturization of the device. Theoretical derivation of the lumped equivalent circuit is provided using the odd-even mode method. The circuit topology is modeled and analyzed based on HFSS full-wave simulation. The lumped inductors and capacitors are fabricated using semiconductor processes, featuring high fabrication precision and accurate values, which ensures stable device performance. The SiC substrate exhibits excellent heat dissipation capability, effectively enhancing power capacity. The measured results demonstrate that the return loss is greater than 10 dB, insertion loss is less than 0.9 dB in 6.3~7.7 GHz operating frequency. The isolation is more than 15 dB across 5~9 GHz, and the ultra-compact electrical dimensions is 0.10×0.05 (λg×λg@f0). These results validate its high engineering application value.

Abstract：To address the uncertainty of Werner deconvolution results in geological structure depth estimation, the K-means clustering algorithm in unsupervised machine learning was applied for optimized analysis of Werner solutions. A synthetic magnetic field model containing two dike-like bodies was constructed and tested, with random noise added to increase complexity. Werner solutions from both synthetic and actual data were analyzed using the clustering algorithm. The results showed that the algorithm accurately identified two geological bodies with depths of 5 m and 8 m in the synthetic model and three bodies with depths of 536 m, 635 m, and 530 m in actual data, consistent with previous exploration findings in the region. The study demonstrated that this algorithm effectively determined the number of geological bodies and accurately estimated their depths, maintaining stability even with noise.

Abstract：To address the challenge of Advanced Persistent Threats (APTs) leveraging Component Object Model (COM) interfaces to obfuscate malicious activities, rendering traditional tracing methods ineffective, this study proposes COMLink, a real-time provenance analysis system based on data matching. COMLink leverages the data association characteristics inherent in data exchange between client and server processes during COM calls, employing a prefix similarity algorithm to achieve precise, thread-level provenance for sensitive behaviors. COMLink is capable of cross-process tracing of COM call chains, enabling effective provenance analysis even when malware exploits trusted processes to execute malicious actions. Experimental results demonstrate that COMLink can trace COM-based attack behaviors with an accuracy of 82% in a test environment comprising 6 known exploitable COM interfaces. COMLink's impact on system performance is negligible, with a performance loss of less than 2%, significantly enhancing APT detection and attribution capabilities.

Abstract：Integrated Sensing and Communication (ISAC),as a core technology for 5.5G/6G systems,breaks the functional boundaries of traditional systems by deeply integrating communication and sensing capabilities,enabling shared spectrum and hardware resources,and driving communication networks toward multi-dimensional sensing.This paper systematically explores the core principles,technical challenges,and future directions of ISAC.Technically,ISAC leverages the transmission and reflection characteristics of wireless signals to achieve efficient fusion of environmental sensing and data transmission through joint waveform design,multi-station collaborative sensing architectures,and self-interference cancellation.However, its development faces critical challenges such as joint optimization of communication-sensing waveforms,dynamic scheduling of network resources,multi-station synchronization,and standardization.Typical application scenarios include intelligent transportation,industrial Internet,and space-air-sea-ground collaborative networks,requiring sub-meter-level sensing accuracy,millisecond-level latency,and high reliability.Future advancements focus on four key directions: optimization of terahertz-band devices and intelligent metasurfaces,AI-driven joint communication-sensing frameworks,intent-driven cloud-edge-terminal dynamic architectures,and lightweight security mechanisms.By addressing foundational theories like ISAC joint information theory and energy efficiency models,and promoting industry-academia-research collaborative validation,ISAC will accelerate the deep integration of the physical and digital worlds,emerging as a cornerstone enabling technology for global intelligent networks.

Abstract：The command display system of computer system faces increasing challenges in terms of stability, reliability, and flexibility. This paper proposes a command display framework based on the Qt framework, integrating data processing, interface management, and multi-task support to meet the diverse needs of complex space missions. Additionally, a multi-process page interaction display technology based on continuous data protection is adopted, which combines the main process with independent page display processes, significantly enhancing the system's stability and security. Furthermore, this paper applies continuity situation display technology and improves the adaptive periodic time bucket algorithm, optimizing event synchronization and rollback mechanisms to further enhance the accuracy and efficiency of task replay. These technological improvements address the requirements of the next-generation command display system of the software system, providing strong support for the successful execution of complex space missions.

Abstract：To solve the difficulties of complex operation in traditional distortion correction, a geometric feature-based distortion correction method is proposed. This method analyzes the camera model of radial distortion and uses a single image to calculate the radial geometric distortion center and distortion coefficient of the image. By robustly evaluating the curves in the image and solving for the degree of curvature of the curves in two different directions, the optimal geometric distortion center of the optical image is found. Then, the recursive least squares method is used to calculate the polynomial coefficients of the radial geometric distortion, and the optimal distortion coefficient is obtained to correct the distorted image. Through simulation and actual image testing, the results show that this method does not require a calibration board and can effectively correct radial geometric distortions in imaging measurement systems by only capturing a single real scene image, and can accurately correct image distortions.

Abstract：In order to solve the problem that the airline's website is easy to cause system anomalies and difficult to manage quickly when the number of visitors fluctuates greatly, this paper proposes an intelligent service governance Scheme based on Long Short Term Memory (LSTM) model， which is dased on SpringCloud microservice governance framework, incorporates LSTM model to analyze and predict the system metrics, and combines the Cloud-Native architecture. Using LSTM to train and output the prediction model for historical index data, it realizes the trend prediction of indexes; combining with Cloud-Native infrastructure, it generates active horizontal container group auto scaler based on the prediction results, realizing the pre-scaling of basic resources; according to the index prediction results and the change trend, it automatically sends out and recycles the management policy, realizing intelligent traffic management. Practice has shown that this scheme can effectively mitigate system risks by preparing for resource scaling and service management in advance based on model prediction results.

Abstract：A hybrid positioning method combining Ultra Wide Band (UWB) signal confidence, a location neural network model, Time-of-Flight (TOF) ranging, and Kalman filtering is proposed to address the challenges of severe signal reflection in narrow and deep coal mine galleries. The method first trains a location neural network model offline using UWB communication data between tags and base stations, along with known true distances of the tags. During real-time positioning, a UWB signal confidence discriminant method is introduced to refine TOF measurements. Finally, Kalman filtering is employed to further enhance positioning accuracy. Experimental results demonstrate that this approach effectively reduces positioning errors caused by signal reflections in underground coal mines, achieving hybrid positioning accuracy below 10 cm with advantages of high speed and excellent stability.

Abstract：This paper summarizes the problems and challenges faced in the heat dissipation design of multi-channel high-power T/R module, and analyzes the heat dissipation capabilities of various heat dissipation solutions. Microchannel and manifold microchannel heat dissipation technology has received widespread attention in high heat flux density heat dissipation scenarios, and has become an effective heat dissipation method for high-power, high heat flux density integrated circuits. At the same time, multi-channel flow distribution is an important factor affecting temperature uniformity. Applying the above technology, this paper proposes a design method for heat dissipation solutions for multi-channel T/R component modules and applies it to the design of specific heat dissipation solutions. The Computational Fluid Dynamics(CFD) method is used to analyze, compare and evaluate various solutions. The manifold microchannel and parallel shunt solution embedded in the carrier can effectively solve the cooling problem of multi-channel T/R components with local power density greater than 500 W/cm2, and the thermal resistance and temperature uniformity between channels are less than 2 degrees Celsius.

Abstract：In order to solve the problems of cumbersome and low efficiency in the traditional vision detection process, a vision detection system design based on gesture recognition was proposed. The system uses STM32 microcontroller as the core control, and realizes intelligent and self-service vision detection through laser ranging module, gesture recognition module and eye chart display module. The system displays the visual target through the TFT LCD display screen, and after the ranging reaches the standard range, the gesture recognition sensor recognizes the gesture action of the measured person, and collects, compares and judges through the microprocessor, and completes the visual acuity grade display with the single visual target judgment criterion and dichotomy retrieval strategy. The feasibility of the system is verified by experiments, and the system can achieve the correct recognition of the four gestures, and the average accurate recognition rate of vision detection reaches 82%, which has good accuracy and high efficiency.

Abstract：In order to effectively test the actual anti-interference performance of communication radio stations, a design scheme for Gaussian white noise comb spectrum interference signals based on FPGA is proposed. Compared to traditional random phase frequency modulation comb spectrum interference, it is more effective in combating broadband communication systems. This scheme is based on FPGA and improveds Ziggurat algorithm and DDS technology to generate comb spectrum interference signals. It has strong flexibility while meeting high-speed and high-performance requirements. And through the DA chip output, 3~11 comb spectrum signals were generated. The experimental results show that the frequency control error of the generated comb spectrum interference signal is less than 0.001%, and the interference effect is quantified as an indicator.

Abstract：To improve the production and testing efficiency of meteorological satellite occultation detection payloads, liberate productivity, and release equipment resources, an intelligent automated testing system is proposed. The principle and basic testing process of the satellite simulation system carried by the occultation detection payload were specifically introduced, and the intelligent automation testing system scheme and implementation method were elaborated in detail. The core architecture of the system software has achieved the migration from the traditional TestStand and LabVIEW architectures to the Python technology stack, which can realize intelligent interface adaptation, intelligent test planning, intelligent fault detection, and automatic generation of test reports, supporting the full process testing of occultation detection payload research and development, testing, and delivery. Studying the intelligent automated testing technology for meteorological satellite occultation detection payloads can open up the standard process for automated testing of meteorological satellite occultation detection payloads, promote the development of occultation detection payload technology, and assist in the improvement and application of meteorological observation technology.

Abstract：A design method of low inclination satellite orbit which aims at special latitude regions is proposed. Firstly, the character of low inclination satellite orbit is analyzed. Secondly，by taking the optimal coverage effect of a hot region as the performance index，a optimized model is built to design the satellite orbit．Finally，based on the interconnection of STK and MATLAB, this paper gets the result which fulfills the need of the model with the use of STK′s coverage analytical ability and MATLAB′s GA toolkit. The method is simple and easy to implement, which combines the advantages of two kinds software and presents good engineering application value．

Abstract：With the increasing complexity of aerospace missions, traditional software designs for command and control display systems can no longer meet the demands of high-density, multi-task scenarios. This paper proposes a modeling-based method for the rapid construction of aerospace launch mission scenarios. By standardizing task decomposition and designing scenario templates, it enables the extraction and reuse of system functional components. The research is centered on a meta-model, combined with hierarchical modeling concepts, to establish a standardized modeling framework. Complex mission requirements are decomposed into standardized scene modules, and task dependencies are identified. Following modular design principles, a scene template library is developed to support efficient reuse and dynamic composition of scenarios. Experimental results demonstrate that this method significantly improves the efficiency of scenario construction, achieving a scene template reuse rate of 85%~90% and an average time optimization rate of 83%, effectively enhancing the overall performance of task modeling and scenario configuration.

Abstract：A large amount of data will be generated during the aerospace-grade chip loading test. In order to realize the storage, management and transmission of data and adapt to the characteristics of embedded devices with limited memory resources, this paper designs a data storage and transmission system using Secure Digital(SD) card and User Datagram Protocol(UDP). First, the data storage and transmission system architecture is designed. The main control chip communicates with the SD card through the SPI bus, and communicates with the loaded chip and the host computer through UDP. Then, the selection and software design of the SD card and the file system are introduced. Next, the software design of the data transmission part is carried out based on Lightweight IP(LwIP). Finally, the effectiveness of the data storage and transmission system is verified by experiments, and the transmission rate of the system is as high as 4.5 Mb/s and the memory occupancy is only 57 KB. It meets the needs of data storage and transmission during the chip loading test, and compared with the traditional storage and transmission methods, this method greatly improves the transmission rate and reduces the memory occupancy.