66results about How to "Balance accuracy" patented technology

The invention relates to the technical field of numerical simulation, in particular to a grid model generation method for the fluid displacement process of well cementation engineering. The method comprises the following steps: acquiring geological data of a well cementation project area, and constructing an initial geological model; well cementation engineering design data are obtained, well cementation engineering topological structure modeling is carried out, and a well cementation engineering topological structure model is obtained; according to the initial geologic model and the well cementation engineering topological structure model, fluid dynamic simulation is carried out, and a fluid characteristic matrix is obtained; performing fluid displacement process simulation according to the fluid characteristic matrix to obtain fluid displacement process data; performing fluid displacement process optimization on the well cementation engineering design data based on the fluid displacement process data to obtain optimized fluid displacement process data, and performing resource allocation redistribution on the well cementation engineering design data according to the optimized fluid displacement process data to obtain optimized well cementation engineering design data. According to the invention, the accuracy of fluid displacement simulation can be improved.

The invention provides a dynamic sliding window-based shadowboxing coherence scoring method and related equipment. The method comprises the following steps of: obtaining a shadowboxing video; 3D skeleton points and space coordinates are extracted from the shadowboxing video; according to the shadowboxing style to which the frame picture belongs, distributing a weight level for each 3D skeleton point; according to the weight levels of the 3D skeleton points, assigning a sliding window range for each 3D skeleton point; judging whether the space coordinates of the 3D skeleton points are changed or not in a specified sliding window range, and if not, judging that the action is lagging; respectively counting the number of lagging times of each 3D skeleton point, and calculating a greeting style coherence score of each shadowboxing greeting style according to the number of lagging times; according to the invention, different sliding window ranges are allocated according to the weight levels of the 3D skeleton points, so that the computing resources of the system and the accuracy of detection are balanced; the jamming times of each 3D skeleton point are counted, and the coherence score of each Tai Chi style is calculated according to the jamming times, so that the fluency degree of the action can be evaluated more comprehensively.

The invention provides a grey wolf optimization feature selection method, device and equipment based on an adaptive mechanism, and relates to the technical field of data processing. The method comprises the following steps: acquiring a data set to be subjected to feature selection, and initializing a grey wolf optimization algorithm; the fitness value of each grey wolf is calculated with the purpose of obtaining the current optimal solution, and three current optimal wolves are obtained; then taking the current optimal three wolves as search guidance, combining a nonlinear parameter control strategy, adopting an improved grey wolf step length updating calculation mode, adaptively adjusting variation intensity in the search process, updating and calculating related parameters, positions and fitness values of each grey wolf, and continuously converging the distance between the prey and the grey wolf in the search process; and calculating the score of each grey wolf, and selecting the solution candidate with the highest score as the optimal solution in the current population until the maximum number of iterations is reached, thereby obtaining a global optimal solution. The method is used for processing feature selection in high-dimensional and complex classification, and is high in convergence speed and high in accuracy.

The invention discloses a high-performance and low-delay deep pulse neural network conversion method. A corresponding SNN neural network is converted with the minimum time step length. Inspired by quantitative perceptual training, a QReLU activation function is introduced in the training process, coding errors are eliminated, in addition, a multi-bit pulse sequence is integrated in the QReLU activation function for training, conversion performance can be improved, and finally, a non-leakage IF (Integrating-and-Fire) neuron with a cache is designed to reduce reasoning errors. Experiments of classification and detection tasks prove that the method can achieve the performance of the artificial neural network level by only using 16 time steps.

The embodiment of the invention provides a model evaluation method and device, a storage medium and a program product. In the model evaluation method, after a reply result output by a target model for a target question is obtained, a question type corresponding to the target question can be obtained, and a target evaluation model matched with the question type in a plurality of evaluation models is called to evaluate the reply result. In the embodiment, in the multiple evaluation models, different evaluation models correspond to different question types, and different evaluation models have different parameter quantities. Based on this, a scheme for cooperatively evaluating the reply results output by the target model for different types of questions based on the evaluation models with different parameter quantities is realized, so that the evaluation model with low parameter magnitude can be conveniently adopted to evaluate the reply results of some questions with low complexity, the resource consumption is effectively reduced, and the evaluation efficiency is improved. And an evaluation model with relatively high parameter quantity can be adopted to evaluate reply results of some questions with relatively high complexity, so that the evaluation accuracy is improved.

The invention discloses a new energy power station operation and maintenance management system, relates to the technical field of new energy operation and maintenance, and solves the technical field that collected data cannot be fully utilized to mine operation and maintenance related problems in the prior art. The system comprises a data acquisition subsystem which is used for acquiring operation data of a power generation assembly and storing the operation data in a database; the data analysis subsystem is used for processing the operation data to obtain a first-class feature value and a second-class feature value; the fault diagnosis subsystem is used for acquiring a first-class prediction result and a second-class prediction result; the strategy generation subsystem is used for determining a processing strategy according to the working state and issuing the working state and the corresponding processing strategy to the user side; the user side is used for receiving operation data and processing strategies of the power generation assembly in real time; according to the invention, operation and maintenance personnel are timely reminded of maintenance when the power generation assembly breaks down, and replacement is timely reminded when the data acquisition subsystem breaks down, so that the situation that the corresponding situation cannot be obtained when the power generation assembly breaks down due to untimely data acquisition is avoided.

The invention discloses a scene perception information fusion decision-making method for industrial field multi-source heterogeneous data, which comprises an edge layer, a cloud layer and an interaction layer which form a three-level collaborative architecture, and belongs to the technical field of industrial field data processing. The method comprises the specific steps that S1, an edge layer collects industrial field multi-source heterogeneous original data and carries out self-adaptive preprocessing to obtain cleaned multi-source feature data; s2, the edge layer processes the multi-source feature data through a lightweight scene perception model, and outputs a preliminary scene judgment result; s3, the cloud layer constructs and updates a causal relationship graph based on industrial business knowledge and incremental data of the edge layer, processes data uploaded by the edge layer through a causal reasoning fusion model, outputs a decision result and issues the decision result through the interaction layer; and S4, the cloud layer optimizes parameters of the causal reasoning fusion model and the lightweight scene perception model based on a decision execution result fed back by the edge layer, and issues the optimized parameters to the edge layer through the interaction layer to realize dynamic iteration.

The invention discloses a global neural drawing method and system oriented to hybrid representation optimization, and belongs to the technical field of computer graphics, and the method comprises the steps: dividing a screen into a plurality of tiles, and obtaining a first workload index on each tile in a first drawing stage and a second workload index on each tile in a second drawing stage; generating a cross-tile and cross-stage unified scheduling queue based on the first workload index and the second workload index, and executing the drawing tasks on different tiles in parallel according to the unified scheduling queue; and triggering a subsequent feature fusion and neural global illumination drawing stage when the inter-stage dependency condition is met so as to generate an output image with a global illumination effect. According to the method, the end-to-end frame rate oriented to the modern AI accelerator / AI-GPU can be remarkably improved and the tail delay can be reduced on the premise of ensuring the visual quality and the time sequence stability, and the method is suitable for application scenes such as real-time global illumination, large-scale complex scene drawing and interactive neural rendering.

The present invention relates to the technical field of human posture estimation, and solves the technical problem that it is difficult for traditional methods to improve the efficiency of key points on single-scale features, and in particular to a multi-level efficient human posture estimation system and method, the system comprising: extracting features from an input image in sequence, and generating a local feature extraction module, a spatial feature extraction module, and a global feature extraction module for low-level features, intermediate features, and high-level features, respectively. The method comprises: obtaining low-level features, intermediate features, and high-level features of the input image in sequence, and converting the spatial scale of the high-level features into two-dimensional spatial features, and using deconvolution to restore the scale of the two-dimensional spatial features to obtain the key point positions corresponding to the high-level features in the input image. The present invention overcomes the defect that it is difficult for traditional methods to improve the efficiency of key points on single-scale features, can effectively improve the accuracy of key points, and maintain a good running speed.

The application discloses a kind of federal positioning method based on individualized patch and dynamic fusion, belong to indoor positioning and federal learning technical field.When applying federal learning to indoor Wi-Fi fingerprint positioning of Internet of Things, global model is often difficult to capture local data feature preference, local training cost is high and poor adaptability alone, the present application is by setting individualized patch and dynamic parameter fusion mechanism, reduce the influence of above-mentioned problem on positioning performance.The algorithm configures private position decoder as individualized patch for each local client, extracts local data features in training and uploads to server, server calculates the similarity of data distribution between clients to generate combination weight, for local clients with similar data distribution, individualized patch can selectively linearly combine with the decoder of the local model with higher relevance, avoid local model overfitting, improve the positioning performance of global model in local scene.

The present application relates to the field of artificial intelligence technology and early warning analysis, in particular to an engineering machinery identification and safety risk early warning method based on continuous learning, comprising: collecting optical cable line video, environmental parameters and positioning data for frame synchronization alignment to build an initial identification model; the edge side preliminarily screens suspected targets and uploads to the cloud; the cloud accurately identifies the mechanical category, working state and confidence, combines the positioning data to build multi-modal risk judgment features; based on compliance verification and distance determination to determine the risk level, generate graded early warning and work order; based on operation and maintenance feedback to screen difficult samples, update the model through incremental training and synchronize to the edge side, realize the continuous iteration of the identification model, balance real-time and accuracy, reduce the false positive rate, and improve the automation level of optical cable inspection.

The application relates to the field of 5G communication technology, and discloses a 5G network slice security isolation method and system.The 5G physical network is divided into multiple independent slices, each slice contains an access node, a core node and an output node, the access node is used for security authentication, the terminal matching index is dynamically calculated based on terminal state parameters and authentication matching times; the core node is used for a security management module, the flow security index is obtained based on the weighted comparison of the current flow and a reference value and historical flow; the output node is used for security evaluation, the output security index is obtained by the weighted comparison of the current communication times and a reference value and historical times; the access, flow and output indexes are compared with corresponding threshold values; if any one is abnormal, the current slice is directly isolated; the deviation of each security index and the threshold value is weighted calculated and compared with the threshold value; if greater than or equal to the corresponding threshold value, the current slice is isolated, multi-dimensional security isolation of the 5G network slice is realized, and the balance between precision and calculation efficiency is achieved.

The invention discloses a lithium ion battery health state evaluation method and equipment. The method comprises the following steps: presetting a cell attenuation rate matrix in a battery management system, wherein the matrix defines single-circle capacity attenuation rates under different temperature and discharge depth combinations; acquiring the duration time and the corresponding discharge depth of the target battery in different temperature intervals in the complete time period; calculating cumulative capacity attenuation according to actual operation data and the matrix; and determining the current SOH according to the initial capacity sum. According to the method, simulation and actual measurement are combined to construct a high-precision matrix, a dynamic look-up table accumulation algorithm is utilized, high-precision and low-cost online evaluation of the SOH of the battery under complex variable working conditions is achieved, and the problems that a traditional method is insufficient in precision and high in calculation complexity in a dynamic scene are effectively solved.

The invention relates to the technical field of cultural relic identification and visual large model fusion, in particular to a ceramic cultural relic full-dimensional identification system and method based on multi-modal visual large model and microscopic feature fusion. After high-quality applicable data are screened, other data are calibrated with the optimal mode as the reference, cross-mode deviation is eliminated, and a data foundation is built firmly; the core features are converted into topological nodes, association strength is calculated, feature association edges with guidance are established, a traceable and explainable identification guidance link is generated, and the strong ceramic identification logic requirement is met; the optimal link is screened to be combined with the pre-training model, and the identification precision and the careful property are balanced by depending on a confidence hierarchical judgment mechanism, so that the black box pain point of the traditional model is effectively solved, and the credibility of the identification result is improved.

The present invention discloses a network traffic anomaly detection method based on online continuous learning, which uses an MLP neural network based on wavelet transform to detect network traffic anomalies in an online continuous learning manner, has continuous adaptive intelligent detection capabilities, efficient system performance, excellent detection effects, strong practicality, solves the model forgetting problem in traditional methods, balances real-time and accuracy, reduces resource consumption, improves system scalability, and sets a network traffic reconstruction model based on time series analysis theory and wavelet transform, which significantly improves computational efficiency while maintaining a high sensitivity to sudden anomalies. After using Haar wavelet to perform discrete wavelet transform on network traffic, only a single-layer MLP network is used to learn the network traffic pattern. This lightweight design greatly reduces resource occupation, enables the system to be flexibly deployed on various network devices, and effectively supports anomaly detection tasks in large-scale network environments.

This application discloses a method, apparatus, storage medium, and electronic device for updating the mesh scale, relating to the field of smart home technology. The method includes: upon obtaining the contour cross-sectional data of the object to be modeled, processing the curve path corresponding to the contour cross-sectional data using preset modeling control parameters and a first formula to obtain multiple sampling points; calculating the tangent direction of each sampling point on the path, and determining all two-dimensional vertices constituting the preset contour cross-sectional shape based on the contour cross-sectional data; projecting each two-dimensional vertex along the tangent direction of the corresponding sampling point to obtain a three-dimensional contour surface consistent with the position and direction of the sampling point; adjusting the mesh subdivision density of the three-dimensional contour surface, and updating the mesh scale of the object to be modeled during three-dimensional modeling based on the adjusted mesh subdivision density. This method solves the problems of insufficient accuracy and low efficiency in traditional modeling methods when processing complex curve paths.

This invention discloses an AI-based physical education exam scoring method, belonging to the field of exam scoring methods, including the following steps: (1) activating the cameras deployed in the exam room to collect motion images and facial expression images of the examinee during pull-ups; (2) analyzing the collected facial expression images using AI image recognition technology to divide the exam process into three stages; (3) for each stage, analyzing the examinee's motion images using AI technology to determine whether each pull-up motion is a qualified motion; (4) counting the number of qualified motions in each stage and scoring them separately, adding the scores of the three stages to obtain the examinee's final exam score. This invention can divide the exam into three stages based on facial expression recognition, perform differentiated analysis on the motion characteristics of each stage, better fit the examinee's physical fitness change patterns, and make the evaluation results more targeted.

The application discloses a kind of based on image data's track area abnormal detection method and device of machine tool, it is related to image recognition technical field.The method includes: using image recognition algorithm, introduce multi-objective optimization algorithm, build machine tool target detection model, and connect to candidate region segmentation engine and space-time correlation filter engine;Video stream of track area is collected, and pretreatment is carried out, and the image data of continuous frame is obtained;The image data of continuous frame is input to candidate region segmentation engine, and the ROI region of each frame of image data is obtained;ROI region is input to machine tool target detection model, and the detection frame of machine tool target is obtained;The detection frame of machine tool target is input to space-time correlation filter engine, identifies stationary effective target, and carries out abnormal detection to stationary effective target.The application solves the problems of low detection accuracy, difficult model deployment and poor anti-interference ability in the prior art.

The application provides a semiconductor production business process construction method and system based on analysis node feedback self-optimization, which comprises the following steps: constructing an execution link of a business process based on basic information of a business scenario and preset basic function nodes, wherein the basic function nodes at least include business nodes and analysis nodes; executing the business process according to the execution link; when the analysis node is executed, collecting current associated data and obtaining corresponding historical associated data, wherein the current associated data and the historical associated data both include control parameters and corresponding output parameters; fitting the historical associated data by using a preset regression algorithm model, obtaining a model update weight coefficient, and calculating a fluctuation index; determining a preset threshold interval to which the fluctuation index belongs, and if the fluctuation index belongs to a preset allowed update interval, storing the current associated data into a historical database and saving the model update weight coefficient. The application can improve the adaptability and continuous optimization capability of the business process to a dynamic environment.

The application discloses a Chinese human phenotype ontology recognition method and related equipment, which can be applied to the technical field of text information processing. The application can effectively improve the recall rate by expanding the human phenotype ontology concept synonym set through a large language model, generating an expanded dictionary with the initial dictionary, significantly enhancing the discrimination ability of the recognition model for semantically ambiguous concepts through two-stage comparative learning, balancing the precision and recall rate by constructing an initial dictionary based on the human phenotype ontology official database and the Chinese human phenotype ontology official data, merging and disambiguating the current input text after accurate matching and semantic similarity matching to obtain a fused candidate result list, and finally obtaining the structured recognition result corresponding to each current input text by post-processing each candidate fragment in the fused candidate result list, which can provide effective data support for disease diagnosis.

The present invention provides a model optimization method and system for federated learning with fairness guarantee, the method comprising: receiving first optimization parameters and images to be tested of a classification model to be optimized sent by multiple clients; calculating the accuracy of the classification model to be optimized of each client on different images to be tested, and obtaining a contribution matrix formed by the accuracy corresponding to all clients; according to the contribution matrix, obtaining the total accuracy corresponding to each client, and constructing an optimization model according to the total accuracy corresponding to each client, the first number and the second number, and the variance between the total accuracy corresponding to all clients; according to the optimal solution of the optimization model, allocating a second optimization parameter to each client, so that each client optimizes the classification model to be optimized according to the second optimization parameter. The present invention realizes comprehensive improvement of the performance of the classification model to be optimized after all clients are optimized, and ensures that the distributed system composed of servers and clients can be sustained for a long time and stably.

The invention provides a battery health state estimation method based on battery local fragment data extraction, and belongs to the technical field of lithium ion batteries, and the method comprises the steps: S1, constructing an estimation model; s2, in m charge-discharge cycles, determining an interval section for feature extraction on a battery charging curve corresponding to each charge-discharge cycle, selecting n voltage sections at the same interval V, and obtaining an input feature matrix; s3, obtaining an optimal interval Vp, and obtaining an optimal input feature matrix; s4, repeating the step S2 and the step S3 for a set number of times, selecting voltage segments at different intervals V in each time of repetition, and obtaining an optimal input characteristic matrix corresponding to the current time of repetition; and S5, inputting each optimal input feature matrix into the estimation model to determine a final optimal interval, and repeating the step S2 at the final optimal interval to obtain a final input feature matrix. The complexity and accuracy of the estimation model can be balanced.

This application relates to the field of data verification technology, specifically to a data verification method and apparatus for a chip programming system. The method includes: for a multi-channel parallel chip programming system, statistically analyzing the clock cycles of each channel within each time window; determining the clock phase offset of each channel; evaluating the deviation between each channel and a second reference channel in the current clock cycle and the deviation of the clock phase offset, obtaining the timing deviation of each channel; and dynamically determining the verification delay time of each channel using the timing deviation and clock cycle, generating the verification trigger time for each channel for data verification. This balances verification accuracy and overall efficiency.

A mixed precision inference method, device, equipment, computer program product and computer readable storage medium for a deep neural network are provided. The method divides the deep neural network into functional modules, determines the precision level of each module by switching the precision of each module and comparing the changes in the output indicators, and based on the precision level, conditionally inserts a data type conversion operation on the data flow path to avoid unnecessary overhead. Finally, the calculation is performed according to the level of each module, and the precision conversion is completed according to the conversion operation. The method balances the precision and speed at the module granularity, conditionally inserts the conversion to greatly reduce the type conversion overhead, and guarantees efficient execution and framework compatibility through static planning, thereby maximizing the performance benefits of low-precision calculation while ensuring the output precision.

The invention discloses an infant paper diaper defect detection method. The method comprises the following steps: firstly, collecting baby diaper pictures and constructing a training data set; thirdly, constructing an infant paper diaper surface defect detection model, and training the infant paper diaper surface defect detection model by utilizing the training data set to obtain a trained infant paper diaper surface defect detection model; the infant paper diaper surface defect detection model is obtained by improving a yo11n model. And finally, inputting a to-be-detected baby paper diaper picture into the trained baby paper diaper surface defect detection model, and outputting a defect detection result by the model. According to the infant paper diaper surface defect detection model provided by the invention, on the premise that the model parameter quantity and the calculation quantity are remarkably reduced, the average precision can still be kept at the same level as that of an existing reference model, and good balance of precision and efficiency is realized.

The invention discloses a wireless federated learning energy consumption optimization method and a medium, and belongs to the technical field of wireless communication, and the method comprises the steps: building a bidirectional model compression frame based on a wireless federated learning system model, and building a joint optimization problem with the minimization of the total energy consumption of the system as a target and a model convergence condition as a constraint; solving the joint optimization problem by adopting an alternating iterative optimization algorithm, and solving a sub-problem of optimal bandwidth allocation and a sub-problem of optimal bidirectional sparseness parameters; and executing compressed global model issuing and local model uploading between the central server and the client according to the solved optimal bandwidth allocation and the optimal bidirectional sparseness parameter. According to the method, the minimization of the total energy consumption of the system is realized on the premise of ensuring the model precision by constructing the bidirectional model compression framework and the bandwidth joint optimization, the method is particularly suitable for large-scale deployment, the communication overhead can be reduced, the model performance of federal learning can be maintained, and the collaborative optimization of the energy efficiency and the precision can be realized.

The application provides a method and system for generating forward simulation data, the method comprising: constructing a three-dimensional voxel model corresponding to a simulation scene, dividing a binning region according to the numerical characteristics of electromagnetic parameters corresponding to each voxel contained in the three-dimensional voxel model, and the binning region describing the value range of the numerical characteristics of the electromagnetic parameters; mapping and converting the electromagnetic parameters with different numerical characteristics corresponding to multiple voxels in one binning region into electromagnetic parameters with the same numerical characteristics; determining a target connected domain in the three-dimensional voxel model, wherein the voxels contained in the target connected domain correspond to electromagnetic parameters with the same numerical characteristics and the voxels are connected in space in the three-dimensional voxel model; selecting a local region of interest from the three-dimensional voxel model, generating geometric commands corresponding to the target connected domain included in the local region of interest; and executing the geometric commands to generate forward simulation data. The technical problem of excessive number of geometric commands and low efficiency of simulation data generation is solved.

This invention discloses a heuristic dynamic memory weighting and recall reordering method, comprising six steps: initial impression value calculation, Ebbinghaus forgetting curve modeling, recall depth parameterization, multi-dimensional independent sorting, multi-dimensional result fusion, and recall-triggered impression value inversely proportional increment. This invention simulates the natural decay of memory through an 8-stage Ebbinghaus forgetting curve, achieves a recall-as-reinforcement effect through an inverse impression value increment mechanism, comprehensively considers semantic similarity, temporal freshness, and memory importance through multi-dimensional reordering, and flexibly adjusts the breadth and depth of recall through recall depth parameterization. This invention provides two forgetting curve implementation methods: discrete-stage lookup table method and continuous curve calculation method (including exponential decay type and power-law decay type), suitable for different application scenarios. This invention significantly improves the intelligence level and anthropomorphism of AI long-term memory systems, enabling AI to perform natural memory recall like humans.

The present invention provides a multi-time-scale joint optimization scheduling method and system for an electric-gas interconnected system, which obtains operating parameter data of an electric power network and a natural gas network; based on the obtained operating parameter data, a typical scenario with probability is generated according to a long-term wind power forecast result of the day before, and the day-ahead optimization is performed with the minimum total operating cost as the goal, so as to obtain the combination strategy of the units under each typical scenario; based on the combination strategy of the units, intraday optimization is performed with the minimum expected sum of the total operating cost under each typical scenario as the goal, so as to obtain the optimal intraday unit scheduling strategy; based on the obtained optimal economic scheduling strategy, real-time optimization is performed with the minimum deviation between the real-time stage output and the intraday stage reference as the goal, so as to obtain the optimal real-time scheduling strategy of the units; the present invention applies the random model predictive control method to the renewable uncertainty problem in IEGS, so as to balance the economy and accuracy of the operation strategy.

The invention relates to the technical field of airfoil profile optimization design, and particularly discloses an airfoil profile geometric parameterization method and program with self-adaptive orders, equipment and a storage medium. The method aims to solve the problem that precision and efficiency are difficult to balance due to the fact that the order of a Bernstein polynomial depends on manual presetting in an existing CST parameterization technology. According to the core scheme, the method comprises the steps of obtaining airfoil profile coordinate data and performing normalization processing; determining a class function according to the airfoil profile type; automatically searching and determining the optimal order of the Bernstein polynomial by combining a genetic optimization algorithm and taking the relative change rate of the fitting error as a judgment criterion; and constructing a parameterized model based on the optimal order and outputting geometric parameters. The method has the beneficial effects that the self-adaptive determination of the order is realized, overfitting and calculation redundancy are effectively avoided, the design efficiency is improved while the precision is ensured, and the method has good universality and is suitable for parameterization design of various airfoils.