20results about How to "Increase the compression ratio" patented technology

This application relates to the field of natural language processing, and more particularly to a method for ultra-long context tensor parallel incremental attention and differential compression transmission. It employs a multi-level precision sliding window attention mechanism, dynamically allocating precision levels such as fp16 / int8 / int4 based on the distance between the window and the current computation position, and the information entropy of the window content. A tensor parallel incremental attention engine enables cross-GPU / node incremental attention computation and asynchronous All-Reduce synchronization, improving computational efficiency. Semantic differential compression transmission technology is used, transmitting only the differential data based on similarity detection, significantly reducing cross-node transmission overhead. Combined with an extreme performance optimization and security control strategy, a balance between performance and security is achieved through hardware monitoring and dynamic power consumption / frequency adjustment. This invention can significantly reduce memory usage and transmission overhead in ultra-long context processing, improve computational efficiency and hardware utilization, enhance the understanding ability of large models for long texts, adapt to heterogeneous hardware environments, and is suitable for large model inference and training scenarios.

The application discloses a low-temperature oil-free centrifugal compressor placed in an adiabatic environment, comprising an adiabatic cold box, a centrifugal compressor unit, connecting pipelines and a vacuum pump. The centrifugal compressor is placed in the adiabatic cold box and comprises a motor, an impeller, a volute and a bearing. The impeller and the motor at two ends of the centrifugal compressor are placed in different chambers of the adiabatic cold box. The centrifugal compressor can adopt the same rotor structure at two ends. The motor does not need to be cooled. The centrifugal compressor is provided with a motor heat preservation loop. The bearing adopts a gas bearing or a gas-magnetic hybrid bearing, and the gas bearing type is full dynamic pressure, dynamic-static pressure or full static pressure. The low-temperature oil-free centrifugal compressor placed in the adiabatic environment can greatly reduce the power consumption and cold leakage loss of the compressor. The compressor with the same structure at two sides does not need to consider the axial force. Through the measures of setting different adiabatic chambers and heat preservation loops, the compressor can operate at an import temperature of -268 to -20 DEG C. The low-temperature centrifugal compressor is purely oil-free, and the system volume and weight are greatly reduced.

The application provides a database monitoring operation and maintenance method and system based on time series data compression, which realizes real-time monitoring of various index data of a database platform through efficient data acquisition and compression technology. The system can efficiently collect various performance data generated during the operation of the database, and optimize the time series data through a compression algorithm to reduce the redundant storage of data. In the data storage and management process, the first-order difference sequence and the second-order difference sequence are calculated to reduce the amount of stored data. The use of difference removes the trend in the original data sequence, making the subsequent difference sequence easier to compress. At the same time, for non-constant difference, by setting a dynamic threshold, only the difference values greater than the threshold are recorded, and the small changes are ignored on the premise of not losing important information of the data, thereby improving the compression rate. On the other hand, by using S relation The regularity of the difference sequence is determined, and the compression strategy can be selectively applied, so that suitable compression methods can be applied to time series data with different characteristics, thereby improving the processing efficiency. Through the above design, the storage and processing efficiency of the compressed data are improved, the system can quickly decompress and analyze the data, and the real-time monitoring of the running state of the database is realized.

This invention provides a lossless frame compression method that is easy to implement in HEVC hardware, comprising: S1. Data alignment of the image, expanding the width to a multiple of 64 and the height to a multiple of 16; S2. Block partitioning; S3. Compression of 4*64 size luma PLCU and chroma PLCU, each PLCU being an independent compression unit, with compression performed in units of sbk; obtaining the data_val of all pixels in the luma and chroma PLCU; setting different data_len values ​​according to the characteristics of the data_val of pixels in sbk; S4. Generating the Body, compressing pixels according to the information obtained above; S5. Generating the Head. This method breaks away from the original encoding framework, is more adaptable, and achieves a better compression ratio and faster compression speed through appropriate block partitioning, better luma and chroma component processing, and fewer syntax elements.

The utility model discloses a kind of axial flow centrifugal compressor for compressed air energy storage, it is related to energy storage technical field, comprising: axial flow section, centrifugal section and main shaft;The axial flow section has casing and the axial flow rotor in the casing, the centrifugal section has volute and the centrifugal rotor in the volute, the casing is connected with the volute;The axial flow rotor is connected with the one end of the main shaft by fastener, the centrifugal rotor is sleeved on the main shaft, the upstream end of the centrifugal rotor is supported on the boss of the main shaft the downstream end of the centrifugal rotor is detachably connected with the main shaft by connecting structure.The utility model technical scheme, axial flow rotor and centrifugal rotor are respectively set as detachably connected with main shaft, so that axial flow rotor, centrifugal rotor and main shaft can be separately produced, for large flow, high pressure ratio compressor, independent production facilitates each part manufacturing.

This invention discloses a high-efficiency video analysis method based on cloud-edge collaboration and adaptive MoE. The method includes: when an edge device performs a video analysis task, it first constructs a meta-request containing current task data and device resource constraints, and sends it to the cloud; the cloud uses a pre-trained basic model to extract features from the video data and clusters the data based on the feature space, dividing complex video scenes into multiple semantic subdomains; a lightweight expert model is trained in the cloud for each semantic subdomain; a routing decision model is further trained in the cloud; after deploying an expert model pool and routers on the edge device, for input data, the router first calculates the matching degree of each expert model and performs a comprehensive evaluation based on its computational cost. Through knowledge distillation, the large-scale basic model is decomposed into lightweight expert models, achieving a significant compression of the model size, enabling efficient deployment on resource-constrained edge devices.

This invention provides a method and system for compressing power messages, relating to the field of power system control technology. The method includes: acquiring power messages via a first FPGA; timestamping the acquired power messages; caching the timestamped messages in a first shared memory; extracting messages from the first shared memory; parsing the messages according to their type and classifying and storing them in a second shared memory; extracting messages from the second shared memory by control block using a second FPGA; compressing the compressed packages corresponding to the control blocks; storing the compressed packages in a third shared memory; packaging and secondary compression according to a set time period; generating a total compressed package for that time period; and storing it on an SSD for dual-disk storage. This invention effectively reduces the storage space requirements for historical messages, improves the storage time of historical messages, and solves the problem of frequent failures in storing historical messages for post-event analysis of power systems.

This invention discloses a method for producing large-size H13 round steel, comprising the following steps: (1) Electric furnace smelting: the steel charge to the furnace includes 5-10t of DRI direct reduced iron, and the steel output C≥0.05%; (2) Refining: using the LF+RH process; the LF furnace performs a slag return operation, with a slag return amount of 500-1000kg; in the LF furnace, after slag formation and deoxidation, 20-30t of H13 continuous casting head billet and / or tail billet are added for partial alloying, with a white slag time of 110-160min; the vacuum time in the RH furnace is 30-35min; (3) Vertical continuous casting: casting is performed using a vertical continuous casting machine with a casting speed of 0.36-0.40m / min. This invention improves the steelmaking and rolling processes, thereby increasing the internal flaw detection pass rate of large-size H13 round steel, significantly improving the core quality of the round steel, and obtaining green, high-quality large-size H13 round steel that meets market demands.

The present application relates to a high-fidelity key cache compression method and system based on spectral quantization, comprising: step S1: spectral transformation and feature analysis process; receiving original key cache data from the model, mapping the original key cache data from the time domain to the frequency domain, and deeply analyzing the signal structure by using the energy concentration characteristics unique to the discrete cosine transform (DCT); step S2: frequency domain hybrid quantization encoding process; by using the main frequency extraction mechanism, the high frequency pre-emphasis technology and the hybrid bit width allocation strategy, the frequency domain coefficient is compressed into a sparse representation with extremely low bit; step S3: fusion decoding and sparse accumulation process; by using the hardware-aware fusion operator and the delay inverse transform technology, the attention score is calculated directly in the compression domain. The present application realizes the limit compression and efficient inference under the lossless precision.

This invention relates to a method for manufacturing high-surface-quality steel wire for chain pins, comprising the following steps: steel smelting → continuous casting → heating of continuously cast billets → billet preparation → flaw detection and grinding of intermediate billets → wire rod heating → high-pressure water descaling → annealing → pickling and phosphating → wire drawing → finished product. The method employs a two-stage forming process, with a high compression ratio in billet rolling, wire rod rolling, and wire drawing, which reduces the impact of defects in the raw material on the finished product. During billet rolling, flame cleaning is used to remove surface defects from the billet. Before wire rod rolling, the intermediate billet undergoes surface flaw detection and full grinding to ensure a defect-free surface. Eddy current testing and high-reduction-ratio wire drawing ensure the depth of surface defects in the steel wire, and the high-reduction-ratio drawing results in higher wire hardness, improving the cutting performance for pin processing. Finally, the steel wire used in pin production can control the single-sided grinding residue to within 0.03mm, significantly improving material yield and surface defect compliance. Simultaneously, the high wire hardness provides excellent cutting performance for pin processing.

This invention provides a method for accelerating the compression of intelligent models on edge terminals, relating to the field of edge intelligence. Its key feature is that it includes: acquiring real-time operating status data of the edge terminal and feature information of the current input data, wherein the real-time operating status data includes computing resource load, memory occupancy, battery level, and device temperature. The advantages of this invention are: by sensing the device status and input complexity in real time, and utilizing reinforcement learning to dynamically generate an optimal strategy combination including pruning, hybrid quantization, and early termination mechanisms, this method overcomes the limitations of traditional static compression, achieving a dynamic balance between computing power, accuracy, and energy consumption. It not only significantly reduces inference latency and memory usage but also adaptively adjusts based on battery level and temperature, effectively solving the problems of difficult deployment, high heat generation, and short battery life of large models on resource-constrained edge devices, and greatly improving the real-time performance and stability of intelligent applications.

This invention discloses a solid waste treatment device for chemical safety, relating to the field of waste treatment technology. It includes a base plate and front, rear, left, and right plates forming a compression chamber. By incorporating inclined slide rails and a drive cylinder, the left plate works in conjunction with the front and rear plates to perform multi-directional compression, eliminating compression dead zones. Automatic reset is achieved with the help of connecting springs, solving the problems of poor unidirectional compression and easy interference in multi-directional drive. Its core lies in the integrated conversion mechanism between the drive cover plate and the mechanical claw. This mechanism, through the descent of a single longitudinal axis moving platform, sequentially closes the compression chamber with the cover plate and triggers a connecting rod to open the flap, rotating the drive plate to a horizontal position, allowing the mechanical claw to extend and grasp the waste. This invention achieves the switching between top sealing during compression and grasping during unloading with a single drive source, simplifying the equipment structure, reducing manufacturing costs, and achieving a high degree of integration of multi-directional, efficient compression and grasping functions.

The application provides a data read-write method, a data reading method, a three-dimensional memory and an electronic device. The data read-write method comprises the following steps: a logic chip determines first write data; the logic chip compresses the first write data according to sample data to obtain a redundancy code; the bit difference between the first write data and the sample data is within a preset range; the redundancy code and the storage address of the sample data are stored in a redundancy storage area of a storage chip; and the logic chip and the storage chip are three-dimensionally integrated. The method of the application can realize high compression rate, reduce the delay of memory access data, and reduce the power consumption of a CPU. In addition, when the capacity of the cache area is insufficient, the method of the application can transfer at least part of the data, expand the storage capacity, and reduce the CPU overhead.

The present application relates to the technical field of data processing, and especially relates to a point cloud data compression method, device, equipment and storage medium. The method comprises the following steps: obtaining an octree sequence corresponding to target object point cloud data, and obtaining an image of the target object; inputting the octree sequence into a preset first neural network model to obtain a first probability distribution output by the first neural network model; inputting the image into a preset second neural network model to obtain a second probability distribution output by the second neural network model; obtaining a merged probability distribution based on the first probability distribution and the second probability distribution; and compressing the octree sequence based on the merged probability distribution to obtain point cloud compression data. The present application is used to solve the defect of low point cloud data compression rate in the prior art.

The utility model discloses a folding tent, including roof and support frame, support frame detachable installation in roof bottom, support frame includes two groups of rotatable support components, when two groups of support components relative rotation, can be between the conversion of unfolded state and folding state, wherein, support component all includes two movable rods and a crosspiece, two movable rods are connected respectively in crosspiece both ends below, and the movable rod between two groups of support components can be rotatably connected, makes two groups of support frames between the conversion of unfolded state and folding state, and the movable rod and crosspiece between every group of support components can rotate, makes movable rod and crosspiece between can be between the conversion of unfolded state and folding state. The utility model movable rod between, movable and crosspiece between all can rotate each other, makes support frame can multiple folding, reduces the storage space, and sets up stabilizer and support rod, can stabilize movable rod between, movable rod and crosspiece between, effectively suppresses support frame deformation, and wind load capacity promotion.

The application discloses a hardware compression and decompression method based on Huffman decoding table, wherein original data is compressed on a PC end first, a Huffman decoding table and compressed data are dynamically generated in the compression process, and the two parts are burned into a hardware storage device; a decompression process is carried out in an embedded system, the embedded system sends corresponding instructions and addresses to a hardware decompression module, the decompression module addresses and searches the Huffman decoding table according to the corresponding instructions to decompress the compressed data, and the decompressed data is sent into the embedded system. The open source Gzip compression code is improved, a dynamic Huffman decoding table suitable for the hardware decompression module of the application is generated in the compression process, and the dynamic Huffman decoding table is stored in a corresponding format, so that the decoding table is directly used in the decompression process, and the decompression efficiency and accuracy are greatly improved.

This invention relates to the field of data processing technology, providing a log processing method and system. The method includes: recording the location information of sub-variables that need to be compressed in a log block as key positions and corresponding metadata of the encoded stream into a preset data structure based on a preset template; compressing the log block into different storage units based on the preset data structure; locating the storage unit corresponding to the search keyword based on the distribution of the location vectors corresponding to each log block on the location matrix, and completing the retrieval of log entries based on the matching result of the search keyword and the storage unit, where the location vector is the matching result of the search keyword and the decompressed preset template. This invention efficiently extracts dynamic patterns from logs through preset templates and stores log data as fine-grained units, thereby improving log compression bandwidth; it uses vectorization technology to achieve low-latency retrieval of fine-grained units, thus providing effective support for log analysis and system modeling.

This invention relates to the field of industrial lighting equipment technology, specifically disclosing a dual-purpose emergency lighting fixture for garages and gas stations with adjustable parameters and an expansion interface. It includes a lamp body, a lamp holder, a central first LED light source, and second LED light sources arranged around the first LED light source. The lamp body and lamp holder form an IP66 sealed cavity via three-sided bolts and sealing rings, and the lamp body is equipped with a standardized expansion interface. An integrated emergency circuit drive power supply is installed within the cavity, and the control device independently adjusts the power, color temperature, and beam angle via a DIP switch group. The innovations of this invention are: 1) Hardware-level three-parameter adjustment enhances scene adaptability; 2) The standardized expansion interface supports plug-and-play Bluetooth / sensor modules; 3) The emergency circuit is deeply compatible with intelligent mode, switching within 0.3 seconds of mains power interruption; 4) The three-bolt sealing structure achieves a balance between protection and convenient maintenance. This invention solves the problems of fixed functions, poor expandability, and damage to the sealing ring during disassembly and assembly in traditional lighting fixtures.

This invention discloses a method and system for time-series data persistence based on a data service platform, belonging to the field of time-series database technology. The technical problem this invention aims to solve is how to improve the query performance of total time-series data on a data service platform while simultaneously increasing file utilization. The technical solution adopted is as follows: The method involves writing the collected time-series data into memory, dividing it into multiple blocks of fixed size, and filling them with the data. When a block is full, the data is filed into the file system. Specifically, the following steps are taken: defining the file structure, including a data file (.data) and an index file (.idx); converting row-based data to column-based data; compressing by column, configuring the `compress` parameter to enable or disable compression; using different compression algorithms for each column based on its data type; and persisting the compressed data to the file system.