19results about How to "Good calculation efficiency" patented technology

The invention discloses a lightweight space-time traffic flow prediction method based on GAT and Mama, and relates to the technical field of intelligent traffic systems. According to the method, the feature expression ability of the model is improved through multi-dimensional feature learning (time, date and space-time) of an embedded layer, rich features of input data can be deeply captured, and the learning ability of the model is enhanced; a graph attention module is used to adaptively pay attention to the relative importance between nodes, so that the graph structure relationship in the data can be more effectively captured and utilized, and the analysis accuracy of the model on the spatial data is improved; the GAT-Mama module optimizes the time sequence analysis and spatial data processing capability, and the model can more accurately capture the dynamic dependency relationship between time points and between spatial positions through a multi-layer graph attention mechanism and residual connection; the GAT-Mama layer structure supports flexible adjustment, a user can adjust the depth of the model according to specific task requirements so as to adapt to spatio-temporal data prediction tasks with different complexities, and good adaptability is achieved.

The invention discloses a medium metasurface reverse optimization method based on a space coupling mode theory, and relates to the field of optics, and the method comprises the following steps: building a cut-off waveguide array based on a metasurface, calculating a guided mode and a propagation constant of each waveguide according to the cut-off waveguide array, and coupling adjacent waveguides through a coupling matrix, a waveguide array is obtained; an input light field is coupled into a waveguide array, propagation of light in waveguides is simulated based on a spatial coupling mode theory model, and near-field intensity distribution is converted into far-field intensity distribution by using a Rayleigh-Sommerfeld diffraction theory; and calculating the difference between the target intensity and the far-field intensity through a loss function, evaluating the effect of the current metasurface design, and gradually adjusting the geometric parameters of the waveguide unit through an optimization algorithm. According to the method, the problems of efficient modeling and optimization in metasurface design are solved through the space coupling mode theoretical model.

The invention discloses a power distribution network unknown measurement information completion method based on data flow mapping, and relates to the technical field of power distribution network monitoring and control. The invention discloses a method for complementing unknown measurement information of a power distribution network based on data power flow mapping. The method comprises the following steps: step 1, constructing a data power flow mapping model; step 2, speculating and complementing missing data; 3, real-time updating and intelligent optimization are carried out; step 4, system integration and application; wherein the step 1 comprises the following sub-steps; step 1.1, data acquisition and preprocessing; step 1.2, feature extraction and analysis; 1.3, constructing a mapping model; wherein the step 2 comprises the following sub-steps; step 2.1, identification and speculation of missing data; step 2.2, constructing a speculative model; and step 2.3, data completion.

The invention relates to the technical field of mechanical cutting machining, and discloses a method for constructing a milling force prediction model for milling a variable-curvature part based on RIME-SVM, and the method comprises the following steps: S1, obtaining prediction model data; s2, building an RIME-SVM milling force prediction model based on a support vector machine; s3, constructing a cutting force prediction model for machining a conventional straight surface by the end mill; s4, verifying the precision of the cutting force prediction model of the end mill for processing the conventional straight surface; s5, constructing a cutting force prediction model for machining different geometrical characteristics of the end mill; and S6, verifying the precision of the cutting force prediction model for processing different geometrical characteristics by the end mill. The RIME-SVM model combines the advantages of SVM processing nonlinear regression and the high efficiency of the RIME optimization process. According to the RIME-SVM model, the prediction precision is improved, the experiment cost is reduced, the generalization ability is high, unseen data can be accurately predicted, and the method is suitable for practical engineering application.

The present invention belongs to the field of computer application technology, and in particular to a method for matrix accelerated calculation based on GPU. The present invention provides a method for matrix accelerated calculation based on GPU, which uses a CPU / GPU heterogeneous computing system as a computing platform to optimize the load of the CPU and GPU, thereby ensuring that the calculations of both are completed at the same time, so that the computing efficiency of the collaborative processing of the entire CPU / GPU heterogeneous computing system is optimized.

The invention discloses a social network link prediction-oriented time sequence diagram network parallel training acceleration method. The method comprises the following steps: firstly, calculating a redundancy score and an old score for each user interaction edge in a training set, and calculating a comprehensive information loss score according to the redundancy score and the old score; secondly, according to a preset core set retention proportion alpha, selecting all user interaction side comprehensive information loss scores and upper alpha quantiles of repetitiveness as threshold values, and discarding user interaction with the comprehensive information loss scores lower than the threshold values, so that a simplified core set is obtained through single-time preprocessing; then carrying out adaptive batch division on the obtained core set, and dynamically determining an acceptable maximum user interaction number in each batch according to a redundancy and old comprehensive information loss score; and finally, training the time sequence diagram neural network based on the divided batches to realize social network link prediction. The method not only improves the training efficiency, but also greatly improves the precision of the model.

The invention discloses a training method and system for a medical image recognition model, and particularly relates to the technical field of image processing, and the method comprises the following steps: feature extractor training: obtaining a training set sample image, and inputting the training sample image into a network model; calculating an error between the output of the output layer and the target label through forward propagation of the network model; judging whether the network model is converged or not, and if the network model is converged, obtaining a network model trained by the feature extractor; and classifier training: inputting the training image into the feature extractor after feature extraction training is completed. The classifier adopted by the invention starts from the probability density function, the regression value of the dependent variable to the independent variable is directly calculated by using the training sample, iterative operation is not needed, the consumed time is short, the calculation efficiency is ideal, the classification processing capability of the classification model is ensured, the convergence speed of the model training process is high, and the method is suitable for large-scale popularization and application. And the recognition precision, the real-time performance and the accuracy of the recognition model are improved.

This invention discloses a method for evaluating the anti-sliding stability of transmission tower pile foundations on slopes under rainfall conditions. The method decomposes the stability evaluation problem of transmission tower foundations under rainfall-induced landslide conditions into several sub-problems: First, the basic parameters of the slope and future rainfall information released by the meteorological station are obtained to determine the shear strength parameters of the slope soil after rainfall time t. Based on these parameters, a numerical method is used to determine the stability state of the slope. If the slope is unstable, the relative position of the foundation and the potential sliding surface is first determined. If the foundation is above the potential sliding surface, the transmission tower will fail. If the foundation is partially below the potential sliding surface, the rotation angle of the foundation is calculated based on the soil conditions at the bottom of the foundation. The stability of the transmission tower is then determined according to the maximum allowable tilt value of the transmission tower specified in the "Operation Regulations for Overhead Transmission Lines".

The invention relates to the communication field and the AI field, in particular to a general calculation co-fusion-based baseband processor and a control method, and the general calculation co-fusion-based baseband processor comprises a signal input module, a signal preprocessing module, a dynamic precision adjustable PE module, a signal decoding module, a neural network preprocessing module and an AI result output module. MIMO equalization and AI calculation share a shared PE unit, an optimal scheduling algorithm is provided to realize dynamic allocation of the PE unit, and the problems that a traditional baseband processor cannot realize resource sharing of communication optimization and neural network calculation and cannot dynamically allocate resources according to communication requirements are solved.

The invention discloses a method for alternately calculating a large-scale new energy transmission line fault based on a time domain and a phasor domain. The method comprises the following steps: 1, establishing a controlled current source mathematical model of a new energy inverter; 2, establishing a composite sequence network model of the new energy grid-connected system; 3, solving the new energy grid-connected point voltage in the fault period in the composite sequence network; 4, converting the new energy grid-connected point voltage from a phasor domain to a time domain; 5, solving the mathematical model of the new energy grid-connected inverter by using a numerical integration method to obtain the output current of the new energy grid-connected inverter; 6, converting the output current of the new energy from the time domain to a phasor domain; 7, selecting a corresponding step length according to the dynamic change of the current, and updating the calculation time; and 8, repeating the steps 3-7 until the calculation is finished. The method can be suitable for nonlinear scenes involving control switching and controller saturation, the calculation process is simple, the applicability is high, and when system-level fault calculation with large processing scale and high complexity is carried out, the calculation efficiency is remarkable.

The application discloses an image classification method, system, device, medium and product based on Wasserstein discriminative dictionary learning, relates to the field of image classification, and comprises the following steps: acquiring a training data set; constructing a target function of Wasserstein discriminative dictionary learning based on the training data set; the target function comprises a reconstruction error term based on a Wasserstein distance, an improved local constraint term and an intra-class sharing term; an alternating iterative optimization algorithm is used to solve the target function, so that an optimal dictionary and an optimal coefficient matrix are obtained; a test coefficient vector is determined based on the optimal dictionary and an image to be classified; a category membership matrix of a dictionary atom in the optimal dictionary is constructed based on the optimal coefficient matrix; category scores of the image to be classified are calculated based on the test coefficient vector and the category membership matrix, and the category of the image to be classified is determined according to the maximum score. The application can improve the precision and efficiency of image classification.

The invention belongs to the technical field of blockchain security, and discloses an industrial Internet of Things anonymous responsibility-traceable data editing method and system based on a blockchain. The method comprises the following steps: generating a chameleon Hash trap door by a trusted mechanism, splitting the chameleon Hash trap door into shares by using threshold secret sharing, distributing the shares to verifier nodes, and initializing a dynamic accumulator; the owner encrypts the production data and formulates a modification strategy, and generates a hash value and an initial voucher by using chameleon hash and uploads the hash value and the initial voucher to a chain; the modifier initiates a request containing zero-knowledge proof, after verification is passed, the verifier node generates partial vouchers by using the trap door share, and the modifier aggregates the partial vouchers into a one-time voucher to complete editing; and the trusted mechanism decrypts the identity of the malicious modifier, moves the malicious modifier out of the dynamic accumulator and updates the revocation list. According to the method, trap door leakage is prevented through a threshold sharing mechanism, zero knowledge proof and a dynamic accumulator are combined, and effective tracking and permission revocation of malicious behaviors are realized while identity privacy of a modifier is guaranteed.

The application is based on a dual-flow gate violence detection method and system of an expanded 3D convolutional network and a Transformer, and realizes classification through the following steps; S1: obtaining a video stream, segmenting the video stream into frames, and calculating the dense optical flow between the continuous frames of the video stream; S2: segmenting the video stream segment through a sliding window, and suppressing non-motion elements in the RGB mode by using the optical flow mode; S3: performing spatiotemporal modeling on the dense optical flow and the SRGB mode by using a 3D convolutional neural network, and performing feature fusion to create a feature map; S4: performing time modeling on the spatiotemporal feature sequence of the video stream segment by using a Transformer encoder, and generating a time feature map; S5: performing average pooling by using a 1D convolutional neural network, and performing classification by using a multilayer perceptron, and classifying the video as violence or non-violence. The application solves the deficiency of the violence detection model in processing spatiotemporal features, improves the accuracy and efficiency of violence detection, and especially has an application effect in a dynamic video monitoring scene.

The invention relates to an edge end load dynamic balancing method and an edge computing system.The edge end load dynamic balancing method comprises the steps that when the load state of at least one edge computing node is high load, the load state of an edge computing cluster is judged to be high load, and otherwise, the load state of the edge computing cluster is judged to be low load; when the load state of the edge computing cluster is low load, allocating edge computing tasks to each edge computing node in turn; and when the load state of the edge computing cluster is a high load, preferentially allocating an edge computing task to the edge computing node with the highest resource score, calculating the resource data of each edge computing node through a non-integral rank sum ratio method to obtain the resource score of each edge computing node, and negatively correlating the resource score of each edge computing node with the load degree. The invention solves the problem that the current edge end load balancing method cannot dynamically allocate the calculation task for the edge node.

The application provides a motor target parameter estimation and long-time phase correlation accumulation detection method and system, which first carries out radio frequency conditioning and sampling on a radar original echo, carries out pulse compression after obtaining a discrete complex baseband signal, and outputs a range focusing but still containing range migration frequency domain signal. The core step utilizes a second order Keystone transformation to correct the second order range migration of the target, outputs a second order correction signal eliminating the range-azimuth coupling, and then searches the optimal speed through a Bayesian optimization and CPU parallel acceleration algorithm. On this basis, the IAR algorithm is utilized in combination with the optimal speed estimation value to correct the first order range migration, and phase compensation is carried out according to the speed matching phase factor to obtain a phase matching signal. Finally, the matching signal is subjected to a non-uniform fast Fourier transformation in a slow time domain to realize energy focusing, and the complete motion parameters of the target are solved through analyzing the peak position and amplitude characteristics in the Doppler domain.

This invention discloses a method and system for reinforcement learning-based backbone modules for remote sensing images in the fields of remote sensing image processing and deep learning. The method includes the following steps: input preprocessing and construction of a heterogeneous primitive library; heterogeneous feature alignment and normalization representation; construction of a dynamic routing decision space and generation of routing actions; updating the reward network and routing strategy based on bi-layer primitive optimization; stabilizing policy updates based on reference policy regularization; and performing dynamic feature aggregation, task prediction, and controllable inference. This invention can adaptively allocate computational resources according to the complexity of remote sensing image samples, significantly reducing redundant computation while ensuring performance in detecting complex scenes, small targets, and dense targets. Furthermore, a single model can adapt to deployment scenarios with different computational constraints, making it widely applicable in engineering.

The invention discloses a fractional time-delay multivariable grey model for carbon emission prediction, and belongs to the field of grey prediction models. The model comprises the following steps: S1, collecting dependent variable data and related influence factor data; s2, preprocessing the data; s3, constructing a parameter matrix, and calculating a parameter column of the model; s4, using a particle swarm optimization algorithm to optimize and identify to obtain a value of an optimal time-delay parameter, and constructing a fractional time-delay multivariable grey model; and S5, outputting a prediction result of the carbon emission system. The method has high interpretability and prediction accuracy, is applied to prediction of the total carbon dioxide emission in China, and is an efficient carbon emission prediction method.

The invention belongs to the technical field of spatial big data processing, and discloses a grid-based spatial object topological relation calculation method and system. According to the grid-based spatial object topological relation calculation method provided by the invention, firstly, adaptive subdivision is performed on a spatial object which is a point object or a non-point object, and each point object is adaptively subdivided into a fixed-level grid; and adaptively dividing each non-point object into a set of multi-level grid units according to a set division threshold, and then filtering and solving based on an adaptive grid. The computing system provided by the invention comprises a self-adaptive subdivision module, a filtering module and a refinement module. According to the method, the calculation efficiency of the spatial object topological relation can be effectively improved.

The invention belongs to but is not limited to the technical field of reactor core power distribution, and discloses a small power reactor core power distribution rapid calculation method and system, and the method comprises the steps: selecting a proper physical calculation program, carrying out the modeling of a fuel assembly of a small power reactor core, designing proper input and output parameters, and constructing a power distribution data spectrum library; a full-connection neural network is selected to construct a data-driven power distribution calculation model; the training sample data set is substituted into the data-driven calculation model based on the full-connection neural network, multiple rounds of training are carried out, and a convergent data-driven calculation model is obtained; and performing test verification on the test sample set by using the convergence model to obtain a test result. The method provided by the invention has extremely high calculation speed, can achieve calculation precision similar to that of a physical calculation program of the reactor core, and can meet basic requirements of online supervision of the reactor core of the small reactor for the high-maneuverability ship.