Patents
Literature
Patsnap Eureka AI that helps you search prior art, draft patents, and assess FTO risks, powered by patent and scientific literature data.

17 results about "Fitting Problems" patented technology

CLIP-based single-stage zero-sample semantic segmentation method and apparatus

The invention provides a single-stage zero-sample semantic segmentation method and device based on CLIP, and belongs to the related technical field of computer vision and deep learning, a feature self-calibration module is introduced, and the attention ability of a model to local details of a foreground object is improved by recalibrating features of an image encoder, so that the problem of local detail loss is solved, and the accuracy of segmentation is improved. The fine task can be better processed by the model; an independent text-image decoder is designed to align visual features and text embedding of a middle layer, so that rich features of the middle layer are fully utilized, and the expression ability of the model is improved; by adopting a cascade feature fusion method, effective fusion of visual features of a middle layer and a last layer is realized, the problem that feature correlation is damaged is avoided, and the cross-modal alignment capability of the model is remarkably improved; the over-fitting problem is adjusted and relieved through deep cross-vision cues, and interaction of pre-training marks and cue marks is promoted by using a cross-attention mechanism to enhance the adaptability to unseen categories.
Owner:SHANDONG UNIV OF SCI & TECH

Adaptive feature weighted fusion fault diagnosis method for multi-source heterogeneous data of motor

The invention relates to an adaptive feature weighted fusion fault diagnosis method for multi-source heterogeneous data of a motor, and the method comprises the steps: constructing a multi-head deep convolutional neural network, designing a plurality of parallel convolutional channels in the neural network, deeply mining the abnormal features of original signals of current and vibration, and focusing on the signal features in different dimensions, thereby achieving the fault diagnosis of the multi-source heterogeneous data of the motor. And comprehensive capture of key feature information is ensured. And adaptive weighted fusion is performed on the extracted multi-dimensional features by adopting a correlation analysis method to highlight abnormal key features, so that the feature fusion effect of the multi-source heterogeneous data and the accuracy of anomaly detection are improved. An SELU activation function is adopted and a hyper-parameter optimization strategy of a dynamic attenuation learning rate is combined so that an overfitting problem which is easy to occur in a model training process can be effectively solved and convergence stability of the model and high precision of fault identification can be enhanced. Through adaptive fusion of multi-source heterogeneous data features, key fault features are effectively extracted, and the diagnosis accuracy in an electromechanical coupling fault scene is improved.
Owner:ZHENGZHOU UNIVERSITY OF LIGHT INDUSTRY

Automatic solving method for critical heat flux density of reactor based on principle of statistics

The invention discloses a method for automatically solving the critical heat flux density of a reactor based on a statistical principle, and belongs to the field of safety analysis of nuclear reactors. The problems that an over-fitting problem exists, a data set cannot be verified clearly through a fitting data set, and the development process is complicated when a traditional method is used for predicting independent variable combination items of a relational expression artificially given by CHF relational expression development are solved. The method comprises the following steps: collecting reactor CHF experimental data points; eliminating possible repeated experimental points and cold rod critical points to form a development database; based on a layered random sampling method, extracting data from the development database, and determining a fitting data set and a verification data set; based on the fitting data set, selecting a curve form of a single independent variable or a combination item; the independent variables and the combination items with the collinearity problem are removed; importance ranking analysis of the variables and the combination items is obtained through regression coefficient standardization; forming a final CHF relational expression; and based on the verification data set, verifying the CHF relational expression.
Owner:LANZHOU UNIVERSITY OF TECHNOLOGY

Switch cabinet temperature and humidity monitoring method based on multivariate neural network

The invention provides a switch cabinet temperature and humidity monitoring method based on a multivariate neural network, and the method specifically comprises the steps: collecting the temperature and humidity data of a switch cabinet, carrying out the preprocessing of the temperature and humidity data, and constructing a corresponding training data set; constructing a CNN-LSTM temperature and humidity prediction model, and training the CNN-LSTM temperature and humidity prediction model in combination with a sparrow optimization algorithm and the training data set; according to the current temperature and humidity data of the switch cabinet, temperature and humidity prediction data of the switch cabinet are obtained through prediction, condensation abnormity judgment of the switch cabinet is carried out, a switch cabinet adjusting strategy is set according to the condensation abnormity judgment result of the switch cabinet, and the temperature and humidity in the switch cabinet are adjusted. According to the invention, the multivariate neural network is used for processing a prediction scene with a large data volume, the over-fitting problem of the model is solved through the sparrow optimization algorithm, the prediction precision is improved, the accuracy of subsequent switch cabinet condensation abnormity judgment is improved, and the operation safety of the switch cabinet is guaranteed.
Owner:HUZHOU ELECTRIC POWER SUPPLY CO OF STATE GRID ZHEJIANG ELECTRIC POWER CO LTD +1

Deep neural network accelerator based on in-memory calculation and acceleration method thereof

The invention belongs to the technical field of artificial intelligence, and particularly discloses a deep neural network accelerator based on in-memory calculation and an acceleration method thereof. According to the application, the universal processor is used to determine the output data of the deep neural network in the forward propagation process for the to-be-solved problem, and the in-memory calculation module is used to determine the first gradient parameter of the first weight parameter of each network layer according to the first error information between the output data and the expected output; and the general processor updates the first weight parameter according to the first gradient parameter, so that the method does not depend on the network model type of the deep neural network, accelerates the pre-training process of the deep neural network, and improves the training efficiency of the deep neural network. Meanwhile, the deep neural network can be widely applied to solving of scientific calculation problems, such as solving of linear / nonlinear equation sets, solving of curve least square fitting problems and solving of steady-state partial differential equations, so that the deep neural network accelerator has certain universality.
Owner:HUAZHONG UNIV OF SCI & TECH

A method for fitting an error ellipse of any tightness to a planar coordinate point cloud

The present application relates to geodetic and engineering surveying technical field, specifically to a kind of plane coordinate point cloud fitting arbitrary close degree error ellipse in control network design stage.The method includes: the pre-processing of control point coordinate point cloud, removes abnormal value;Accurate spatial precision evaluation density field is constructed to calculate grid point density value;According to the specified confidence, the density threshold value is calculated, the contour is extracted and the error ellipse parameter is fitted.The present application solves the ellipse fitting problem of any specified confidence of non-normal distribution point cloud by adaptive bandwidth kernel density estimation, and significantly improves the reliability of precision evaluation under complex scene.
Owner:NO 3 ENG COMPANY LTD OF CCCC FIRST HARBOR ENG COMPANY +1

A callback-based deep neural network error localization method

The present invention discloses a callback-based deep neural network error location method, which is divided into three steps: model preparation, error detection, and statistical analysis. In the model preparation stage, a reasonable DNN model is selected to meet the problem requirements. In the error detection stage, key parameters are continuously tracked, and parameters such as training loss, verification loss, training accuracy, and verification accuracy are monitored through a callback mechanism. In the statistical analysis stage, methods such as sliding average, rate of change calculation, threshold determination, and visual analysis are used to accurately locate errors in the model and improve the diagnostic accuracy and comprehensiveness of the deep learning model. Error location is performed for underfitting and overfitting problems in the DNN model, which realizes real-time diagnosis and location of deep learning model training problems, improves the comprehensiveness and accuracy of location, and provides more comprehensive support for error diagnosis of deep learning models.
Owner:CHENGDU ZHENHUA WANXING TECHNOLOGY DEVELOPMENT CO LTD

Method for fitting ellipse with any adaptation error through plane coordinate point cloud

The invention relates to the technical field of geodetic surveying and engineering surveying, in particular to a method for fitting an ellipse with any adaptation error through plane coordinate point clouds in the design stage of a control network. Comprising the steps of preprocessing control point coordinate point cloud, and removing abnormal values; constructing an accurate spatial precision evaluation density field to calculate a grid point density value; and calculating a density threshold according to a specified confidence coefficient, extracting a contour line contour, and fitting an error ellipse parameter. According to the invention, through adaptive bandwidth kernel density estimation, the ellipse fitting problem of any specified confidence of the non-normal distribution point cloud is solved, and the precision evaluation reliability in a complex scene is significantly improved.
Owner:NO 3 ENG COMPANY LTD OF CCCC FIRST HARBOR ENG COMPANY +1

Blade tip clearance and arrival time synchronous measurement method and system based on symmetric reverse function fitting

The invention discloses a blade tip clearance and arrival time synchronous measurement method and device based on symmetric inverse function fitting, and the method comprises the steps: S1, a self-adaptive signal interception stage: carrying out the preprocessing of an original voltage signal collected by a capacitive sensor, including baseline correction, signal normalization and symmetric linear segment interception; s2, a symmetric inverse function fitting stage: constructing an inverse function data set and performing symmetric linear fitting; the symmetric linear fitting is that joint symmetric linear fitting is carried out on two inverse function data sets, and a peak moment, namely arrival time, and scale parameters are calculated through an objective function of a minimum time residual error; s3, at a least square linear fitting stage, the baseline offset, the peak moment and the scale parameter are substituted into a direct Gaussian fitting equation, a nonlinear fitting problem is converted into a linear fitting problem, the linear fitting problem is solved by adopting a least square method, and a closed-form solution of the signal amplitude is obtained; and inverting a blade tip clearance value based on a pre-calibrated relation curve between the signal amplitude and the blade tip clearance.
Owner:TIANJIN UNIV

Method and system for automatically checking fitting degree of welding lap joint surfaces

The invention provides a method and system for automatically checking the fitting degree of a welding lap joint surface. According to the method, part information is obtained by retrieving a welding spot file, normal lines of curved surfaces of all parts are made through welding spot sphere centers, and intersection points of the normal lines and the parts are calculated; comparing the coordinates of the intersection points in pairs, and counting the number of unique points with unequal coordinates; and when the number of the unique points is 2, performing coordinate exchange and fitting verification. In the lamination verification process, calculating vectors formed by adjacent points, unitizing the vectors, then calculating a dot product, and judging whether the vectors are in the same direction or not; and meanwhile, checking whether the coordinate values of adjacent points are equal in a tolerance range. The invention further provides a system for realizing the method. The system comprises a welding spot information acquisition module, a normal generation module, an intersection point calculation module and a fitting degree detection module. According to the method, a fitting problem in a three-dimensional space is converted into a vector analysis problem in a two-dimensional plane, the calculation complexity is simplified, and the detection efficiency and accuracy are improved.
Owner:DONGFENG MOTOR GRP

A multi-objective optimization method for the aerodynamic shape of the second-segment wing of a hypersonic winged reentry vehicle

The present invention discloses a multi-objective optimization method for the aerodynamic shape of a second-segment wing of a hypersonic winged reentry vehicle. Based on multi-fidelity data fusion using a deep learning neural network, the method fits the linear and nonlinear relationships between high-fidelity and low-fidelity aerodynamic data, thereby improving the accuracy and reliability of aerodynamic parameters at various flight state points during the design of the second-segment wing shape of the vehicle. On this basis, a proxy model constructed using a radial basis function neural network algorithm is combined with a simulated annealing optimization method, which has better applicability for complex nonlinear fitting problems in the field of aerodynamic shape design, while improving the globality and robustness of the optimization results of the geometric shape design variables, thereby providing a new, efficient and accurate design tool for hypersonic winged reentry vehicles.
Owner:NANJING UNIV OF AERONAUTICS & ASTRONAUTICS

Self-supervised real image denoising method based on dual random double-sampling mode

The invention discloses a self-supervision real image denoising method based on a dual random double sampling mode. Comprising the following steps: providing a random double-sampling method of dual denoising branches to solve a noise fitting problem and generate more sub-images for training; introducing complementary attributes of dual branches into the design of a loss function, and proposing a dual double-sampling cross loss function; putting forward a fixed'subgraph-sampling 'strategy according to a mode collapse phenomenon in a test process; in the pre-training and testing process, designing an artifact remover to further remove chessboard artifacts; and training the de-noising network of the dual branch structure through the training steps to obtain a de-noising model. And then, inputting an image with noise into the trained network model, and performing corresponding test steps to obtain a high-quality de-noised image. According to the method, clean images are not needed for training, the performance of the method is better than that of an existing self-supervision denoising method, and the method is an effective real image denoising method.
Owner:SICHUAN UNIV

Two-parameter optimized intelligent rating method for material pearlite spheroidization

ActiveCN119107502AEliminate underfitting problemsImprove the level of denoisingAcquiring/recognising microscopic objectsNeural learning methodsFeature extractionEngineering
The invention relates to the technical field of pearlite spheroidizing rating, in particular to a two-parameter optimized intelligent rating method for material pearlite spheroidizing, which is characterized in that a convolutional neural network is used for feature extraction, key parts of pearlite spheroidizing images, which have great influence on rating results, can be highlighted in a targeted manner through the setting of convolution kernels, and the accuracy of the rating results is improved. And the noise part of the pearlite spheroidized image generating negative interference on the rating result is weakened or eliminated, so that the de-noising degree of the image is improved. And then training the XGBoost algorithm through the feature image, and obtaining the number of decision trees which have great influence on the precision of the XGBoost algorithm and the optimal value of the depth of the decision trees, thereby obtaining an optimal classification model. According to the method, two weak classifiers, namely the convolutional neural network and the XGBoost algorithm, are combined to form a strong classifier, and the under-fitting problem of the convolutional neural network is eliminated through the XGBoost algorithm, so that the rating accuracy is effectively improved.
Owner:HEFEI GENERAL MACHINERY RES INST +2

A lightweight method for solar cell defect detection based on data enhancement

The invention discloses a lightweight method for defect detection of solar cell sheets based on data enhancement, comprising the following steps: S1: screening the acquired fault pictures of the solar cell sheets to obtain true fault samples; and manually marking the types of the pictures; S2: building an improved ACGAN network, using a generator to generate false fault samples and corresponding categories according to randomly generated noise data and type information; then inputting the false fault samples and the real pictures into a discriminant network together, and the discriminant network determines the true and false, and the following actions are repeated in a cycle, and the generating network and the discriminant network are trained in turn; the invention expands the data set based on the data enhancement algorithm of the ACGAN network, alleviates the overfitting problem caused by insufficient training data, and proposes optimizing the convolutional neural network structure, performing lightweight processing, reducing network training parameters, compressing the model, shortening the model training and defect detection time, and improving the defect detection efficiency.
Owner:XIANGYANG HUBEI UNIV OF TECH IND RES INST

Geological sample detection data dynamic correction method and system

The invention provides a geological sample detection data dynamic correction method and system, and belongs to the technical field of element measurement. The method comprises the following steps: acquiring initial energy spectrum detection data of a geological sample to be detected, initializing the detection data according to the signal-to-noise ratio of the initial energy spectrum detection data, dynamically correcting the variable window width, equally dividing a region corresponding to the variable window width into N sub-regions, calculating a characteristic value of the detection data of each sub-region, and fitting and denoising the characteristic values; and adjusting the width of the detection data dynamic correction variable window, and correcting the geological sample detection data based on the stored de-noised characteristic value. The variable window width is dynamically corrected only by adjusting the detection data, and the input variables of the correction method or algorithm are reduced; initializing the dynamic correction variable window width of the detection data through the signal-to-noise ratio of the initial energy spectrum detection data, and matching the data correction and the data pollution degree; the characteristic values are fitted by adopting a fitting formula, so that the over-fitting problem is not easy to occur.
Owner:BEIJING YIXINGYUAN PETROCHEMICAL TECHNOLOGY CO LTD

Financial product recommendation method and device, equipment and medium

The invention discloses a financial product recommendation method and device, equipment and a medium. The method comprises the steps of generating a virtual interaction record of an inactive customer according to a customer portrait of the inactive customer; generating a product recommendation knowledge graph according to the virtual interaction records and the actual interaction records of the inactive customers; and determining a target financial product recommended to the inactive customer according to the product recommendation knowledge graph. According to the embodiment of the invention, the accuracy of financial product recommendation can be improved, and an over-fitting problem is avoided.
Owner:INDUSTRIAL AND COMMERCIAL BANK OF CHINA

A method for modeling and optimizing compound noise in an NISQ device quantum circuit

This invention belongs to the field of quantum computing technology and relates to a method for modeling and optimizing composite noise in the quantum circuits of NISQ devices. Based on the Lindebrade master equation, a composite noise model integrating relaxation noise and phase noise is constructed, and the coupling effect of multi-source noise is simulated using Kraus operators. An improved zero-noise extrapolation technique is proposed, employing a zero-noise neural network extrapolation method. A multilayer perceptron is used to replace the analytical extrapolation function, transforming error correction from a fitting problem into a data-driven prediction problem. Using the one-dimensional transverse field Ising model as a benchmark, a ground state preparation and dynamic evolution circuit is constructed in a quantum cloud computing platform and Cqlib environment. Multi-noise level data is generated using a time scaling strategy to train the neural network. This invention reduces the average absolute error by 92.2% across the entire parameter region, including ferromagnetic phases, paramagnetic phases, and quantum critical points, demonstrating superior accuracy and stability compared to traditional extrapolation methods.
Owner:NANJING UNIV OF AERONAUTICS & ASTRONAUTICS +1