43 results about "Data efficiency" patented technology

The invention discloses a drug combination patent intelligent identification method and device based on a large language model, and the method comprises the steps: obtaining a standard data set built based on drug combination labeling information of patent data, carrying out the data expansion of the standard data set, and generating an initial training set; a parameter efficient fine tuning technology is adopted, the initial training set is utilized to train the base large language model, and an initial recognition model is generated; screening and identifying difficult samples in the unlabeled patent data by using the initial identification model, and adding the difficult samples into the initial training set to form an enhanced training set; performing iterative training on the initial recognition model by using the enhanced training set and adopting a multi-task learning framework in combination with a comparative learning mechanism to obtain a drug combination patent recognition model; the to-be-recognized patent data is input to the drug combination patent recognition model, and a drug combination information recognition result is obtained.The accuracy, data efficiency and model interpretability of drug combination patent recognition can be remarkably improved, and the computing resource requirement and the manual review cost are reduced.

The invention provides an autonomous operation obstacle avoidance method and system for a low-altitude aircraft, and the method is executed by the low-altitude aircraft, and comprises the following steps: S1, data acquisition: collecting data through a multi-source heterogeneous sensor group carried by the low-altitude aircraft, airspace control information, real-time meteorological data and collaborative sensing data of other aircrafts sent by the ground sensing base station are received through the air-ground communication interface; preprocessing the collected original data, wherein the preprocessing comprises data format standardization, abnormal value elimination and time synchronization; and S2, data fusion processing: carrying out credibility evaluation on the preprocessed multi-source heterogeneous sensing data and collaborative sensing data. According to the invention, through three-dimensional credibility evaluation and an air-ground data compensation mechanism, the perception precision is greatly optimized, the perception data efficiency is improved in complex environments such as low illumination, strong electromagnetic interference, rainstorm and the like, and obstacle avoidance decision errors caused by data distortion of a single sensor are avoided.

The invention discloses a system and method for improving data configuration efficiency of a test machine, and relates to the technical field of board card testing, the system comprises an FPGA PCIE module, a DMA address data analysis module and a data transmission module, a general data frame is divided into an independent 32b address field and an independent 32b data field through the arranged DMA address data analysis module, then bit [31] and 0 of an 32b address are analyzed to represent write operation, and the data transmission module is connected with the FPGA PCIE module. 1 represents read operation, an identifier is independently extracted and embedded into a subsequent instruction packet, then a bit [27: 0] of a 32b address is extracted, a target daughter card number is determined according to a preset address mapping table, and data is packaged after determination, so that data distribution of different address fields is realized, the data transmission module receives the data and then sends the data to each post-stage daughter card module in parallel, and data read-write is realized. Therefore, the read-write speed of the test machine is greatly increased, and the read-write efficiency is further improved.

The application provides a data query method and device, equipment and a storage medium. It can be used in the field of big data technology. The method comprises the following steps: receiving a data query command sent by a control device; the data query command comprises target query information; the target query information reflects target query data; determining a corresponding index type according to the target query information; the index type is a Hash index or a B+ tree index; in response to the index type being a Hash index, directly querying position information corresponding to the target query data from preset mapping information according to the target query information to obtain target position information; the preset mapping information comprises preset data information and position information corresponding thereto; in response to the index type being a B+ tree index, querying position information corresponding to the target query data according to the target query information and a B+ tree index strategy to obtain target position information. The application can reduce the number of scans, speed up the response and improve the efficiency of querying target query data.

The invention discloses a low-power-consumption quick start awakening method for an acquisition terminal. According to the scheme, the acquisition terminal comprises a main control module, an auxiliary control module and a power supply switching circuit, and a rechargeable low-leakage-current battery capacitor is adopted as a standby power supply; during normal power-on, the terminal is powered by the main power module, the main control module works in a high-performance mode, the clock module and the liquid crystal screen module are switched to the main control module through the multi-channel analog switch, and the auxiliary control module performs voltage sampling of the main power supply and the standby power supply and charging control of the standby power supply. The main control module and the auxiliary control module are simultaneously communicated with the metering chip to read data; when the power-off key is awakened, the multi-channel analog switch is switched to the auxiliary control module, and only a few modules such as the auxiliary control module work. Through cooperation of the main power module, the standby power module, the power supply switching circuit and the like, the problems of long key wakeup starting time, high power consumption and low data reading efficiency are solved, and rapid key wakeup starting, low power consumption and improvement of the data reading efficiency of the acquisition terminal are realized.

The invention discloses an information management net value mail cleaning and analyzing method and system based on a large model, and the method comprises the steps: obtaining a mail of a user mailbox, and extracting and storing the metadata of the mail; performing multiple identification on the mails based on a large model, confirming net value mails, and excluding non-net value mails; aiming at net value mail contents in different formats, adopting a corresponding extraction mode and carrying out data verification; standardizing the extracted net value data according to a preset term mapping table and a standardized data dictionary; performing automatic analysis based on the net value data after standardization processing; and carrying out visual display on the analysis result. According to the method, automatic identification and acquisition of net value mails are realized, net value data in mails in different formats can be extracted and automatically analyzed, and analysis results such as income trend and risk assessment are generated in real time. The method solves the problems that in the prior art, the efficiency of obtaining and analyzing the net value data in the asset management mail is low, errors are prone to occurring, format adaptation is difficult, and a dynamic analysis result is difficult to generate in time.

The invention provides a zero sample anomaly detection method based on bimodal prompt enhancement, and the method comprises the following steps: S1, carrying out the input coding and multi-modal feature extraction: carrying out the prompt coding of an input image and a text into feature vectors, and achieving the preliminary alignment of the features of the image and the text; s2, dynamic vision-text prompt interaction: adopting a PIM module to realize layer-by-layer dynamic guidance of text prompt on visual features through a learnable projection matrix; s3, constructing a hierarchical abnormal prototype memory library: constructing a multi-granularity abnormal prototype library by adopting a PMB module, and enhancing fine-granularity representation through micro-texture and macro-component prototypes; s4, an uncertainty perception anomaly scoring mechanism; calculating an abnormal score by combining multi-granularity prototype matching and the uncertainty weight; and S5, anomaly detection and output: integrating the text-visual similarity and the PMB score to complete anomaly detection. According to the method, the modal semantic deviation is solved, the fine granularity detection precision is improved, and the data efficiency is high.

This invention relates to the field of automotive communication technology, providing a method and system for converting and managing automotive communication database formats. It aims to solve the problems of low efficiency, error-proneness, and difficulty in collaboration associated with traditional manual processing of multi-source heterogeneous communication data. This method automatically parses source data in multiple formats such as DBC and Excel, converting them into a pre-defined intermediate data model. It then automatically generates ARXML files conforming to the AUTOSAR standard and finally uploads the files to a cloud database, associating them with project information to achieve centralized storage, version management, and collaborative work. Through collaboration between local and cloud platforms, multiple verifications and version management can also be achieved. This invention efficiently and accurately converts scattered and heterogeneous communication data into industry-standard files through an automated process, and leverages the cloud to achieve centralized file management and project-based collaboration, effectively improving the efficiency, consistency, and traceability of in-vehicle communication data processing.

The invention provides a construction method of a multi-source heterogeneous equipment collaborative energy scheduling model, a medium and a system, and belongs to the technical field of energy scheduling. A fault propagation prediction model based on a heterogeneous graph structure and an attention mechanism is constructed, and the uncertainty is quantitatively predicted by using a Bayesian neural network; solving the collaborative scheduling mathematical model by adopting a self-adaptive robust optimization method to obtain an initial scheduling scheme, dynamically adjusting charging and discharging power according to a comprehensive aging index, guiding active learning through a sampling priority value to carry out dense sampling on high-uncertainty nodes, and carrying out incremental training on the model; and adjusting the power limitation of the energy storage unit and the control strategy of the photovoltaic inverter according to the node fault probability distribution, thereby solving the technical problem of insufficient reliability of the scheduling scheme caused by the lack of an artificial intelligence model training mechanism with high data efficiency in the fault propagation scene of the multi-source heterogeneous equipment of the micro-grid system.

The invention discloses a stateless robot motion control method and system based on vision, and belongs to the field of robot intelligent control, and the method comprises the steps: collecting multi-view visual observation data through a top main camera and double-wide-angle wrist cameras installed above and below an end effector; inputting the visual observation data into a pre-trained strategy network, wherein the strategy network outputs a relative pose increment of an end effector; and the current pose of the end effector is updated according to the relative pose increment, and the robot is controlled to execute corresponding actions. The strategy network is trained based on a stateless imitation learning framework, only visual observation data and corresponding expert action labels are used in the training process, and state information is sensed without depending on a robot body; according to the method, pure-vision-driven end-to-end control is realized, and the method has strong space generalization ability, high data efficiency and excellent cross-platform adaptability, and is suitable for complex operation tasks such as industrial manufacturing and service robots.

This invention relates to the field of coal mine sequestration technology and discloses a method for sequestering carbon dioxide underground in closed coal mines. The method includes: constructing a causal structure of a knowledge base for the evolution of multiphase geological processes in coal mines to form a multi-scale knowledge logic framework; interpreting historical sequestration projects in coal mines using multiple modalities and encapsulating the interpretation results as medium-term prediction capabilities; extrapolating the future state of the medium-term prediction capabilities output under multiple scenarios based on the multi-scale knowledge logic framework to obtain a long-term sequestration efficiency trajectory; performing multi-objective Pareto front analysis on the long-term sequestration efficiency trajectory and compiling the analysis results into an optimized sequestration scheme; applying the optimized sequestration scheme and using dual-track data efficiency to verify the real-time monitoring data and the long-term sequestration efficiency trajectory to form a decision-making source deviation representation; analyzing the decision-making source deviation representation through deviation attribution and encoding the analysis results into a multi-scale knowledge logic framework and medium-term prediction capability collaborative reconstruction instruction. This invention can improve the efficiency of carbon dioxide sequestration in closed coal mines.

The invention discloses a lithium ion battery health state estimation method based on physical information Transform. The method comprises the following steps: acquiring voltage and current time sequence data of a lithium ion battery under a pulse charging condition; feature parameters are extracted from the time series data, and a training data set containing the feature parameters and corresponding health state labels is constructed; and constructing a physical information neural network model based on a Transform architecture, inputting the training data set into the physical information neural network model for training, processing pulse charging data of the lithium ion battery to be tested by using the trained physical information neural network model, and outputting a battery health state estimation result. According to the method, a framework which is high in principle and can be popularized is established, and an important contribution is made for constructing a battery health prediction model which is higher in interpretability, better in data efficiency and higher in physical credibility.

The semi-supervised microseismic first arrival intelligent picking method combined with SimMatch and improved TransUGA, Step1, obtaining the real labeled signal in microseismic data, generating forward simulation signals under different main frequencies and different speeds; Step2, using the real signal and the generated simulation signal to make the original data set, inputting the made original data set into the constructed UGATIT generative adversarial network, generating pseudo signal samples through the mutual game of the generator and the discriminator, and screening out the generated pseudo real signal samples; Step3, taking the original labeled data and the unlabeled samples generated by the UGATIT generative adversarial network as a new data set, and semi-supervised training the improved TransUNet network; Step4, obtaining the trained TransUNet network and high confidence pseudo label; Step5, inputting the microseismic data to be tested into the trained TransUNet network to pick up the first arrival signal. Effectively solve the problems of low efficiency of manual labeling data, poor first arrival signal picking accuracy under low signal-to-noise ratio, and difficulty in learning signal features from small sample data.

The application provides a multi-controller write data method and device, electronic equipment and a storage medium, wherein the method comprises: receiving a write data request, determining to-be-written data corresponding to the write data request and a target data stripe in a RAID to which the to-be-written data is to be written; accessing metadata in an associated memory to determine a target value of a bit corresponding to the target data stripe in the metadata; determining whether the target data stripe is occupied based on the target value, and determining whether to execute a write data operation corresponding to the to-be-written data. Through memory consistency interconnection between multiple storage controllers, memory resource sharing between the storage controllers is realized, the multiple storage controllers are directly associated together through the interconnection memory, and series connection at a physical layer is realized. Through a mode in which the multiple storage controllers jointly operate the same piece of metadata in the interconnection memory, a stripe lock mechanism is abolished, write IO processing delay is reduced, and write data efficiency is improved.

The invention discloses a low-light image enhancement and adaptive dimming method based on semi-supervision, and the method comprises the following steps: constructing a dimming module based on a mixed density network, and the dimming module is used for simulating the color distortion and darkening process under a low-light condition; a dimming module is used for converting the high-quality image into a low-illumination image, and a synthesized low-light-normal-light image pair is generated to serve as training data; training an image enhancement model of a UNet architecture on the synthetic training data in an unsupervised manner, and performing fine tuning by using a small number of real images; in the image enhancement process, the low illumination effect is controlled by setting a dimming factor, and the brightness level of an output image is controlled through condition input. The method has the beneficial effects that the technical scheme realizes multi-dimensional breakthrough in the aspects of data efficiency, reality sense simulation and enhancement quality and flexibility through a semi-supervised learning strategy, a physically-driven dimming module and adaptive brightness control, and effectively solves the core pain point of a traditional low-light enhancement method.

In order to further improve the quality of texture mapping and solve the problem of large dynamic random memory bandwidth consumption and data efficiency bottleneck, the application provides a method and device for texture mapping hardware acceleration, which improves the method of output pixel texture mapping to texture pixel points, replaces the bilinear interpolation with bicubic interpolation, and designs a static memory as a multi-path group associative cache, which can set different cache layers according to different application scenarios to reduce the consumption of memory resources, and effectively reduce the redundant bandwidth. The application can flexibly configure the texture cache hierarchy size according to the required texture mapping scene, realize the hardware acceleration of texture mapping through bicubic interpolation, and is suitable for various types of three-dimensional graphics systems.

The invention discloses a data set management method and a related product. In the scheme, the files meeting the preset conditions are marked as the data set files according to the meta-information of the files; performing multi-dimensional judgment on the data set file to obtain a multi-dimensional judgment result; the multi-dimensional judgment comprises file activeness judgment, storage consumption judgment, creator personnel state judgment and replica redundancy judgment; and performing weighted summation on the multi-dimensional judgment result, and determining a to-be-cleaned data set file. Compared with a manual regular cleaning mode in the prior art, the method has the obvious advantages that the process is time-consuming and labor-consuming, massive, multi-modal and frequently-updated data generated in artificial intelligence development are difficult to deal with, and the efficiency is low.

The invention discloses an electric power inspection credible detection method and system based on a visual language model thinking chain and rule perception reinforcement learning, and relates to the technical field of computer vision, large model application and artificial intelligence safety monitoring. A large visual language model is driven to generate an explicit thinking chain text before outputting a detection box, and the problems that existing detectors such as YOLO / DETR are lack of semantic reasoning ability, black box decision cannot be explained and generalization ability is poor under the small sample condition are solved. According to the method, an end-to-end large visual language model architecture is adopted, a rule perception reinforcement learning mechanism is combined, an electric power safety regulation is converted into a computable logic reward function, and physical constraint and compliance verification are carried out on a reasoning process. According to the method, high-precision detection is guaranteed, semantic-level interpretation of violation behaviors is achieved, the data efficiency and the system credibility in a few-sample scene are remarkably improved, and the method is suitable for power operation safety monitoring.

This invention proposes a method and apparatus for identifying ground motion pulses based on a hybrid convolutional and self-attention neural network, belonging to the interdisciplinary field of earthquake engineering and artificial intelligence. It addresses the technical problems of low accuracy and low data efficiency in the automatic identification of pulse-type ground motions in existing technologies. The method includes: acquiring raw ground motion velocity time-history records; constructing a CVT hybrid model and processing the two-dimensional waveform image to obtain a trainable model; training the model: optimizing the CVT hybrid model using a labeled smoothed cross-entropy loss function combined with L2 weight regularization; employing the AdamW optimizer with a cosine annealing learning rate scheduling strategy; performing iterative training; evaluating the trained model to obtain a validated CVT hybrid model that can distinguish ground motion pulse types; performing pulse type discrimination and outputting classification results representing pulse-type and non-pulse-type ground motions. This invention is applied to the automatic identification of near-fault pulse-type ground motions.

The invention discloses a density meter voltage withstanding test device and method, and solves the problems of low safety coefficient, large labor consumption, incapability of quantitatively measuring insulation parameters, low data recording efficiency and the like in density meter voltage withstanding test, and the density meter voltage withstanding test device comprises a base which is provided with a test area and a placement area; the plurality of conveying rails are rotationally connected to the top surface of the base, and the plurality of conveying rails penetrate through the placing area and the testing area; a photoelectric sensor is fixed on the top surface of the test area; the clamp is fixed on the top surface of the conveying track and can clamp the density meter; the connecting assembly is fixed on the top surface of the base so as to be connected with a test interface of the density meter; the insulating cover is fixed on the top surface, corresponding to the test area, of the base; the protection cover is hinged to one side of the insulation cover and covers the placement area, an inner cavity of the protection cover is communicated with an inner cavity of the insulation cover, and the width of the protection cover is smaller than that of the protection cover; and the code scanning gun is fixed on one side wall, facing the protective cover, of the insulating cover.

The application discloses a data migration method and system of a data acquisition and monitoring system of an oil and gas pipeline, and belongs to the technical field of oil and gas pipeline data acquisition and monitoring control system engineering. The method comprises the following steps: first, uniformly converting data into data in a standard format; and then, based on a preset conversion database, converting the data in the standard format into data in a target format, so that the conversion and migration processes of the standardized data are realized, the data in different formats of various software is conveniently converted into the data in the target format, the efficiency is relatively high, the problem of low data conversion and migration efficiency in the related art is solved, and the engineering data migration efficiency of the data acquisition and monitoring system of the oil and gas pipeline is improved.

The application discloses a kind of linear dynamic modeling method and system based on generalized momentum and Koopman theory, for the interaction and operation law of multiple energy in wisdom transformer area system, construct the mathematical model of wisdom transformer area system including a variety of different energy, power electronic equipment control logic and power flow equation;Comprehensively consider multiple constraints, construct the constraint condition set of the mathematical model of wisdom transformer area system;By introducing momentum base state space representation, linear decoupling is carried out from state-dependent dynamics to control input channel, a linear Koopman dynamic model only needs to learn unexcited dynamics, which significantly reduces model complexity and improves data efficiency;Further integrate linear generalized extended state observer, realize real-time estimation and compensation to external disturbance and unmodeled dynamic characteristics of system;The proposed method is applied to the actual data of wisdom transformer area system, while improving the accuracy of dynamic characteristic extraction, significantly enhances the stability and robustness of the system under uncertain conditions.

The invention discloses a model training method based on a high-dimensional efficiency adaptive sparse training normal form, and belongs to the technical field of computer science, and the method comprises the following steps: 1, data and infrastructure preparation; 2, mixed sparse pre-training is carried out; and step 3, aligning and post-training. The invention provides a set of complete, end-to-end and efficient LLM training and reasoning control system. According to the system, through underlying innovation of a framework, a training target and a post-training alignment mechanism, three bottlenecks of calculation efficiency, data efficiency and alignment accuracy faced by traditional LLM training are thoroughly solved.

The invention relates to the technical field of data processing, particularly provides a peak signal fitting method, device and equipment and a medium, and aims to solve the technical problem of low peak data fitting efficiency caused by the fact that parameters need to be manually adjusted and a large amount of time needs to be spent on calculating fitting data in the process of fitting peak signals. The peak signal fitting method comprises the following steps: obtaining a fitting model corresponding to a peak signal according to the peak signal; for one parameter in the peak parameters, in any round, according to the range of the peak parameters, the nth adjustment step length of the nth peak parameter is obtained; according to the nth adjustment step length, adjusting the (n-1) th peak parameter to obtain an nth peak parameter; substituting the nth peak parameter and other peak parameters into the fitting model to obtain an nth fitting signal; obtaining an nth residual sum of squares according to the nth fitting signal and the peak signal; and obtaining a residual sum of squares change value according to the nth residual sum of squares and the (n-1) th residual sum of squares.

This invention relates to the field of blockchain technology, specifically to a blockchain-based method and system for protecting the privacy of tourism data. The method includes: constructing a distributed caching layer deployed between blockchain nodes and user terminals to dynamically capture newly added tourism data from users, and aggregating this data into data block groups based on preset time windows or data volume thresholds. This invention avoids performing a complete privacy protection process on each piece of new data separately by constructing a distributed caching layer between blockchain nodes and user terminals, thus reducing redundant computation and verification operations. The data block groups are encrypted using a dual-layer approach: lightweight symmetric encryption ensures basic data efficiency, while asymmetric encryption strengthens group-level metadata security. Furthermore, it initiates sharded parallel verification and utilizes message queues to achieve lock-free operation, shortening verification time.

This invention discloses a system for analyzing and optimizing the usage efficiency of components within a medical consumable kit using AI, relating to the field of intelligent management of medical consumables. The system includes a component usage status perception module, a component usage efficiency analysis module, and a component configuration AI optimization module. The status perception module acquires usage status data for each component within the kit under a target surgical procedure. The efficiency analysis module calculates component-level usage efficiency indicators based on the usage status data. The AI ​​optimization module is built based on a reinforcement learning algorithm, using the component-level usage efficiency indicator as the status input, the component configuration scheme as the action output, and a comprehensive score based on usage efficiency feedback results or usage efficiency and low-carbon feedback results as the reward signal. Through continuous iterative learning, it generates and dynamically optimizes the optimal component configuration scheme. Thus, through the collaborative work of the aforementioned modules, a refined analysis of the usage efficiency of medical consumable kit components and an AI-driven continuous optimization closed loop can be achieved.

The invention discloses a data query method and device, electronic equipment and a program product, relates to the field of artificial intelligence, and is applied to the field of financial science and technology, the method comprises the following steps: receiving a query request of a target user, and inputting the query request into an intelligent agent, the query request being a data query request for various data storage systems; query instruction information is generated through the intelligent agent according to the query request, and the query instruction information at least comprises data source information, a read-only operation instruction and a desensitization rule; creating a data read-only connection through the data access interface according to the data source information; and performing data query based on the data read-only connection, the read-only operation instruction and the desensitization rule to obtain a target query result corresponding to the query request. By means of the data query method and device, the problem that in the related technology, when data are queried, a user needs to know query instructions of different data storage systems, and consequently the data query efficiency of the user is low is solved.

Embodiments of the present application disclose a task processing method and device, electronic equipment and a storage medium. The method comprises: performing a set number of test tasks based on test data to obtain a task processing time, wherein the test task comprises data production and data consumption; calculating an optimal data processing efficiency based on the task processing time; determining an optimal speed configuration of a formal task based on the optimal data processing efficiency, and executing the formal task. The method can calculate the optimal data processing efficiency based on the multiple task processing times obtained by multiple test tasks, and on this basis, the optimal speed configuration of the formal task can be determined based on the optimal data efficiency to effectively control the data processing speed of the task, thereby improving the efficiency and reliability of task processing.

The embodiment of the invention discloses a 4D HOI modeling method and device, a storage medium and a program product. The method comprises the following steps: receiving a 3D model of a target object, generating time sequence human body motion based on the guidance of a multi-view structure of the 3D model, cooperatively reconstructing the dynamic state of the object, and obtaining a 4D HOI sample; training a dynamic character interaction generation model by using the 4D HOI sample; calling the trained dynamic character interaction generation model for reasoning, and using a time sequence to guide and sample a human body motion sequence and a target object motion sequence in a reasoning stage; and the human body motion sequence and the target object motion sequence are aligned according to a time sequence and then are combined to generate a 4D HOI sequence. According to the method, various 4D HOI samples can be generated by using the target object 3D model, and the model generalization ability is enhanced, so that the data efficiency is improved, and new objects and new tasks can be quickly adapted.

A machine-learning based, rapid, physics-aware thermal simulator, drawing inspiration from the Fourier's law and the Fourier-Biot equation, the first and second derivatives of the temperature map, is provided. The learning objective evolves from merely translating images to approximating natural phenomena such as the thermal gradient and thermal Laplacian. By adding an additional encoder during training and substituting the image-based loss with the thermal-aware loss, the proposed model achieves lower prediction error, higher data efficiency, and more physically accurate behavior.