40 results about "Line search" patented technology

The invention discloses a dynamic confrontation sample generation method and device based on non-convex optimization, and relates to the technical field of artificial intelligence safety and confrontation machine learning, and the method comprises the steps: training to obtain a pre-training model; applying noise to the original sample image to generate an initial confrontation sample, inputting the initial confrontation sample into the pre-training model to obtain model prediction output, and calculating a loss function value; applying symmetric differential disturbance in the neighborhood of the adversarial sample along a random orthogonal direction, estimating an interpolation gradient norm according to a loss function value, and optimizing disturbance and updating the adversarial sample along a gradient symbol direction to obtain a current adversarial sample; projecting the current adversarial sample to a legal disturbance space; judging whether the current disturbance update is valid or not based on a line search condition, and if yes, accepting the current disturbance update and entering the next round of iteration; if not, the dynamic projection range is dynamically adjusted until the iteration stop condition is met, the final adversarial sample is generated, and the efficiency and stability of adversarial sample generation are improved.

The invention relates to the technical field of computers, and discloses an information source coding matrix data processing method based on a modified conjugate gradient algorithm. According to the method, sparsity constraint and a fixed-point quantization mechanism are fused in iterative optimization, and a sparse fixed-point coding matrix meeting hardware deployment requirements is directly generated through gradient shielding, correction conjugate parameter calculation and Armijo criterion line search and synchronous execution of hard threshold sparseness and fixed-point quantization after each update. According to the method, the sparsity constraint and the fixed-point quantization mechanism are embedded in the iteration process of the modified conjugate gradient algorithm, so that end-to-end alignment of coding matrix optimization and hardware deployment requirements is realized.

The invention discloses a multi-converter parameter collaborative optimization method for improving the small interference stability of an active power distribution network, and belongs to the technical field of power systems, and the method comprises the steps: building a differential algebraic equation model, and carrying out the linearization of a steady-state operation point, and obtaining a state space model; and a control parameter optimization problem taking the minimization of the spectrum intercept construction as a target is defined, and control parameters are a GFL converter phase-locked loop parameter and a GFM converter virtual damping parameter. First-order sensitivity and second-order sensitivity of eigenvalues to control parameters are calculated based on a matrix perturbation theory and are converted into local quadratic constraint sub-problems. And performing convexity solution by adopting a positive semidefinite relaxation method, adjusting the step length in combination with a line search mechanism, iteratively updating control parameters, and outputting an optimization result after a convergence condition is met. According to the method, the spectral intercept is reduced through multi-converter parameter collaborative optimization, and the weak stability characteristic is improved; the convergence and solvability are improved by utilizing matrix perturbation and positive semidefinite relaxation; and the stability is enhanced and optimized by adopting line search step length adjustment.

The invention discloses a non-repeated scanning point cloud registration method based on energy consistency, and particularly relates to the technical field of three-dimensional point cloud processing. According to the method, a unified energy model fusing a scale normalization robust data item, an entropy regularization optimal transmission item, a normal consistency item and a spherical normal distribution alignment item is constructed aiming at the problems that the inherent geometrical relationship of a point cloud is damaged due to random sampling of a non-repeated scanning laser radar and a traditional registration method is easy to fail. Registration solution adopts a three-stage strategy: in a bootstrap stage, a high-quality initial pose is quickly obtained through spherical normal distribution alignment and robust translation estimation; in the main processing stage, the soft corresponding relation and the geometric increment are alternately optimized, and non-monotone line search is combined to ensure energy convergence; fine adjustment is conducted in the polishing stage to improve the final precision. According to the method, the matching problem in non-repeated scanning point clouds is effectively solved, and high-precision and high-robustness registration is realized in a scene with weak geometric features and repeated structures.

The invention discloses an unmanned aerial vehicle-mounted through-wall SAR motion error inversion method based on image sharpness and a GCBP algorithm. In order to solve the problem that SAR images are seriously defocused due to airborne platform motion errors, a traditional self-focusing method is popularized to the field of unmanned aerial vehicle-mounted through-wall SAR, and an accurate unmanned aerial vehicle-mounted through-wall SAR signal model is constructed based on the refraction effect of a wall; motion errors are analyzed from geometric characteristics, frequency dependence characteristics, time-varying characteristics and space-varying characteristics, and an accurate phase error model is constructed by adopting sub-aperture division and sub-band division strategies in combination with a GCBP algorithm; and finally, based on an image sharpness criterion and a motion continuity constraint, realizing accurate estimation of the motion error of the unmanned aerial vehicle-mounted through-wall SAR by adopting a gradient descent method combined with a line search strategy. The actual measurement experiment verifies that the method has good refocusing effect and parameter inversion performance in the field of unmanned aerial vehicle-mounted through-wall SAR (Synthetic Aperture Radar).

The application relates to the technical field of multi-spacecraft cooperative control, and discloses a multi-spacecraft grouping consistency restraint control method and device under optimized FLS, which comprises the following steps: acquiring the network topology of a heterogeneous multi-spacecraft system and constructing a multi-subgroup directed communication topology network, constructing a spacecraft attitude control system model based on modified Rodrigues parameters; approximating unknown nonlinear terms in the model through a fuzzy logic system, constructing a fuzzy logic system based on a gradient descent algorithm and Armijo line search criterion; constructing a restraint control term and using a backstepping method to construct a virtual control law, combining the output of the fuzzy logic system based on the gradient descent algorithm and the Armijo line search criterion to construct an actual control input law, so that grouping consistency restraint control of the heterogeneous multi-spacecraft system is realized. The application can improve real-time approximation precision and adaptive step length adjustment capability while guaranteeing grouping consistency, and can enhance topology applicability and task cooperation flexibility.

This invention relates to the field of computer technology and discloses a method for processing source coding matrix data based on a modified conjugate gradient algorithm. This method integrates sparsity constraints and fixed-point quantization mechanisms in iterative optimization. Through gradient masking, modified conjugate parameter calculation, and Armijo criterion line search, and by synchronously performing hard-threshold sparsification and fixed-point quantization after each update, it directly generates a sparse fixed-point coding matrix that meets hardware deployment requirements. This invention achieves end-to-end alignment between coding matrix optimization and hardware deployment requirements by embedding sparsity constraints and fixed-point quantization mechanisms within the iterative process of the modified conjugate gradient algorithm.

The application discloses an image processing method for tubular bubble level angle value measurement, which comprises the following steps: firstly, performing binaryzation processing on the collected bubble level image; secondly, performing first line search, and performing tilt correction on the image according to the direction angle of the line; thirdly, performing second line search, determining the left end point search area of the bubble, and obtaining the left end point of the bubble; then, calculating the physical distance from the left end point of the bubble to the left line main shaft; finally, calculating the absolute value of the physical distance difference from the left end point of the bubble in the two images before and after the bubble moves to the left line main shaft, and obtaining the axial moving distance of the bubble. The application preferably realizes the automatic measurement of the axial moving distance of the bubble which is necessary in the tubular bubble level angle value measurement process, completely eliminates the problems of time-consuming and laborious, large measurement error and the like caused by manual intervention, and effectively improves the accuracy and the measurement efficiency of the measurement result.

The invention provides a hair accurate collision detection method based on a pre-estimation correction method, and relates to the technical field of physical simulation of hair. The invention provides a collision detection algorithm based on estimation and correction, which optimizes the efficiency of SDF (signed distance field) collision detection while supporting accurate collision detection, and improves the simulation performance. A feedback type double-side sampling correction mechanism is adopted to effectively cope with predicted accumulated errors, a method based on SDF estimation correction is adopted to remarkably reduce sampling times required by SDF line search in hair collision detection, and in application of accurate hair collision detection, sampling times (more than 6 times during sharp object and low hair dispersion) in each sub-step are reduced to 3 times, so that the accuracy of the detection result is improved. Therefore, the overall efficiency of collision detection is improved. According to the method, the space and time change continuity of the minimum SDF point during hair movement is utilized, and a prediction method based on estimation and correction is used for replacing complete SDF line search, so that collision detection is carried out, and the accuracy and efficiency of collision detection can be considered at the same time.

The invention relates to a multi-line text searching and replacing method and system, and the method comprises the steps: carrying out the preprocessing of an original multi-line text, a multi-line searching content and a multi-line replacing content, removing the head and tail spaces of each line, and obtaining a corresponding logic text, a logic searching content and a logic replacing content, judging whether the logic search content contains a preset wildcard character or not, if the logic search content does not contain the preset wildcard character, carrying out simple character string matching on the logic text to obtain a logic index search result, converting the logic index search result into an original character index by utilizing a preset index mapping relation for positioning, and executing simple replacement operation, and if the logic search content contains the preset wildcard character, carrying out simple character string matching on the logic text to obtain a logic index search result; and if the logic text is matched with the fuzzy item list, regarding the logic search content as an integral matching block, performing sliding comparison according to rows in a row array corresponding to the logic text, if the matching succeeds, extracting the fuzzy item list and the starting and stopping row indexes of the matching block, performing positioning by utilizing an index mapping relation, and executing fuzzy replacement operation of inserting the fuzzy item list into a placeholder.

The invention discloses a route retrieval method and device, a readable storage medium and a computer program product, a target route is created according to an input operation of a user, and the target route is formed by correspondingly connecting a starting point, an ending point and a plurality of position points between the starting point and the ending point selected by the input operation; based on the target line and a time range input by a user, determining each mobile device accessing a base station covered by the target line within the time range; vectorizing the moving track of each mobile device and the target line to obtain a corresponding feature vector; and searching the mobile device matched with the target line by calculating the similarity between the feature vector corresponding to the target line and the feature vector corresponding to the moving track of each mobile device.

The application belongs to the field of non-stationary signal processing, health monitoring and fault diagnosis, and discloses an online instantaneous frequency ridge extraction method based on an adaptive cost function. The method innovatively proposes an online instantaneous frequency extraction algorithm framework, and through the construction of an adaptive cost function, introduces a search frequency band dynamic adjustment strategy, converts the ridge line search process into a constrained optimization problem, so that it can accurately identify and extract non-continuous or intermittent frequency components in a low signal-to-noise ratio environment, and effectively solve the feature extraction problem caused by local signal distortion or transients. The experimental results show that the method has good instantaneous frequency extraction accuracy and robustness, provides an efficient and reliable technical means for real-time operation state monitoring and early fault diagnosis of the compressor, and has important engineering application value.

The invention discloses an equation-oriented multiphase flow pipe network steady-state modeling method, and relates to the technical field of multiphase flow pipe network simulation. The method is executed by a computer system, an abstract pipe network is used as a node-branch mathematical model, node pressure and branch mass flow are defined as independent variables, and a nonlinear equation set comprising a branch pressure drop equation, a node mass conservation equation and a boundary condition equation is constructed; the method comprises the following steps: initializing based on a linearized Darcy equation, carrying out variable scaling, solving by adopting a Newton iteration method in combination with a line search technology, integrating a pipeline hydraulic model, a black oil model and a separator node model in the solving process, and processing pipeline pressure-temperature coupling calculation through a Runge-Kutta self-adaptive step length algorithm. The method solves the problems of poor convergence, difficulty in treating reverse flow and complex loop of the existing ordered through module method, has the advantages of high convergence efficiency, high robustness and wide adaptability, and provides technical support for pipe network design and optimization in the fields of oil and gas gathering and transportation, chemical process and the like.

The invention discloses a big language model optimization method and system based on inference tree bifurcation points and semantic gating, and the method comprises the steps: carrying out the system initialization, and defining a calculation model which comprises a teacher model, a student model and a semantic representation model; inputting a to-be-solved target question text, and constructing an inference tree based on Monte Carlo tree search by using a preset teacher model; performing node traversal and candidate sample pairing on the inference tree; carrying out preliminary screening on the sample pairs based on result differences; semantic similarity verification is introduced into the preliminarily screened sample pair, and a semantic gating mechanism is used for further filtering; training a student model based on the screened key bifurcation point data set, and constructing a contrast loss function; finally, multi-task joint optimization and parameter updating are carried out, and through the strategy of combining off-line search and on-line distillation, the problems that in the prior art, credit distribution is inaccurate and training convergence is slow in result-based reinforcement learning are solved.

This invention discloses a DC motor speed control identification method and system based on an improved quasi-Newton method, belonging to the technical field of DC motor speed control system modeling. First, a time-varying nonlinear DC motor speed control system is constructed, and an identification model of this system is obtained. Then, an identification method flow based on the quasi-Newton method under step-size line search is constructed. The optimal step size is obtained through step-size line search for parameter estimation, and a correction strategy is designed to ensure the positive definiteness of the approximation of the Hessian matrix inverse. This invention avoids the oscillation or slow convergence problems caused by a fixed step size by dynamically adjusting the step-size factor, and simultaneously introduces a correction strategy to forcibly guarantee the positive definiteness of the approximation of the Hessian matrix inverse, thereby obtaining more accurate motor model parameters.

The invention discloses a giant constellation distributed orbit determination solver and method based on a domestic super computer. The solver adopts a'base layer-algorithm layer-application layer-interface layer 'modular architecture, and a C-API is externally provided by an external' C 'handle, and can be directly called by Fortan / Python; a kernel realizes integration-correction integration through RKF7 (8) adaptive integration, Armijo non-precision line search and mixed precision PCG, and in combination with a CSR-D sparse format and topological perception domain decomposition, the memory complexity is reduced from O (N) to O (N) under the scale of ten thousand kernels, and the communication synchronization frequency is reduced by more than or equal to 70%. The calculation process comprises the steps of step length correction driven by local truncation errors, one-time completion of Armijo backtracking global stipulation, float-double residual error triggering type precision upgrading and Allan variance sliding window dynamic PPD early warning. The method supports check point continuous calculation and cross-language integration, and can be widely applied to operation control and space traffic management of giant constellations such as Beidou and Kiaji.

The application discloses a kind of wide-area backup protection potential fault element search method and device, method includes: obtaining the additional network structure of fault point where line fault point is and current fault component amplitude;The current fault component amplitude is associated with the branch connected with substation in the additional network structure of fault point;According to the size relationship of current fault component amplitude of the branch connected with substation, determine fault line;From the current fault component amplitude of each element in the fault line, determine potential fault element.Compared with prior art that can only search possible fault area, the application is based on the current fault component amplitude of line to search, can greatly reduce the suspected fault line searched, to improve the efficiency of search.

The application discloses a kind of multi-agent cooperative search and target exploration simulation system and its use method, it is related to marine environment detection technical field.The simulation system includes: regional division module, for dividing the task area into multiple sub-regions of different shape and size according to the ability of agent;Comb-like line search module, for global path planning to the task area;Local path re-planning module, for re-planning agent path to avoid obstacle;Task allocation module, for allocating target based on greedy algorithm;And target exploration module, for approaching exploration to the target allocated along the specified angle, and improved artificial potential field method is used to avoid obstacle in exploration stage to dangerous area.This application significantly reduces the failure rate of obstacle avoidance by dynamically adjusting the obstacle avoidance strategy.This application significantly improves the efficiency of target allocation by improving the greedy algorithm.

To efficiently collect and visualize information of a corporation affected by line disconnection by specifying a cause of the line disconnection and performing preventive maintenance.SOLUTION: The channel search unit performs channel search. The ping test unit executes a ping test on the line retrieved by the line retrieval unit. The data analysis unit predicts an abnormality of the line based on a result of the ping test executed by the ping test unit. The action execution unit automatically generates a preventive maintenance schedule based on the abnormality predicted by the data analysis unit. The log receiving unit receives a log from the service device. The second data analyzing unit analyzes the log received by the log receiving unit and specifies an error message. The second action execution unit executes an action based on the error message specified by the second data analysis unit.SELECTED DRAWING: Figure 1

To provide a route searching device that can suppress a guidance of routes including a blind alley or a dead end.SOLUTION: A route searching device comprises: a route information storage unit that stores road network data including information on a node, and information on a link linking between the nodes; and a route searching unit that searches for routes from a departure site to a destination site. The route searching unit is configured to, after excluding, from a search object, a terminal link serving as the link in which only one is connected to other link of the nodes located at either end of the link, search for routes from the road network data and cause the searched routes to output.SELECTED DRAWING: Figure 3

The utility model relates to the technical field of line searching and aligning, and discloses a line searching and aligning realization circuit applied to a cable maintenance device and the cable maintenance device, the line searching and aligning realization circuit comprises a line searching signal processing circuit and a line aligning signal transmission circuit, the line searching signal processing circuit comprises a line searching signal buffer circuit and a line searching signal transmission circuit, the first end of the alignment signal transmission circuit is electrically connected with the alignment control end of the first main control circuit and is used for grounding, and the second end of the alignment signal transmission circuit is electrically connected with the first cable interface; a first end of the hunting signal buffer circuit is electrically connected with a first hunting control end of the first main control circuit, a second end of the hunting signal buffer circuit is electrically connected with a second hunting control end of the first main control circuit, and a third end of the hunting signal buffer circuit is electrically connected with a first end of the hunting signal transmission circuit; the second end of the hunting signal transmission circuit is electrically connected with the first cable interface. Therefore, the line searching and aligning implementation circuit is simple in structure, and the power consumption of the line searching and aligning can be effectively reduced.

The application discloses a kind of mesh antenna electric performance weighting's form finding optimization method, comprising the following steps;Step (1): input the initial structure parameters of antenna and electric parameters in optimization model;Step (2): according to the initial force density input, solve mechanical model, obtain corresponding cable net node coordinates;Step (3): according to the aperture field distribution function of mesh reflector antenna, generate weighting function;Step (4): generate sensitivity matrix and vector;Step (5): find the optimal force density increment that meets the constraint condition;Step (6): when the solution accuracy of quadratic programming does not meet the requirement, one-dimensional line search is carried out, and the step length of next iteration is found;Step (7): update mechanical model, obtain next iteration value;Step (8): judge whether to meet convergence condition;Step (9): output force density design variable.The application improves calculation efficiency, effectively reduces antenna structure design and electric performance design iteration cycle time.

1. The name of the design product: multifunctional network line finder. 2. The use of the design product: for digital line search, network line search, also provided with NCV induction pen, LED lighting. 3. The design points of the design product: in shape. 4. The picture or photo that best indicates the design points: perspective view 1.

Systems, methods, and computer-readable media for solving large, sparse convex optimization problems on highly parallel hardware. A computing apparatus receives an initial point and iteratively updates an optimization point via a first subsystem that constructs and updates a Karush-Kuhn-Tucker (KKT) system and applies small steps in an improving direction. In parallel, a second subsystem incrementally solves the updated KKT system, while a third subsystem optimizes parameters such as preconditioners and matrix permutations. Outputs are exchanged among subsystems during each iteration to enable continuous refinement. A small-step criterion, enforced via line search or trust-region procedures, ensures convergence. Execution leverages GPUs or TPUs to partition and process KKT systems concurrently, dynamically adjusting step size based on convergence indicators.

The invention provides an off-line data search method and device, electronic equipment and a computer storage medium, and the method comprises the steps: training a machine learning model through taking maximization of a user click rate and / or a conversion rate in an off-line scene as a target, obtaining a scoring model, processing the multi-dimensional features of commodity information, and obtaining a comprehensive score of a commodity; generating a commodity recommendation set according to the comprehensive scores of all the commodities; therefore, the commodity recommendation set which is most representative, most comprehensive in coverage and most capable of meeting the requirement of the user for high relevance in first offline search is dynamically selected, the purposes of thoroughly eliminating network dependence of first starting of an application program and ensuring high relevance and richness of offline search results are achieved, and the use experience of the user is effectively improved.

The application discloses a high-precision solution method for steady-state magnetohydrodynamic equations based on an adaptive PINN model and belongs to the technical field of magnetohydrodynamic equation solution.The application constructs a complete closed-loop system from network construction, residual diagnosis to optimization adjustment. The sharing-branch network structure is used to realize differentiated learning of physical quantities, the adaptive residual mechanism is introduced in the Adam stage to dynamically focus on the high residual area, the L-BFGS stage is combined with the line search condition and the loss function scale transformation to perform fine optimization, the modules work collaboratively, and a systematic high-precision solution scheme is formed. Through the RAR or RBA mechanism, the model can autonomously diagnose the high residual area and dynamically adjust the training points during the training process, and the fitting capability for the boundary layer and the strong nonlinear region is significantly improved. Meanwhile, the Strong Wolfe line search and the loss function scale transformation are combined to adaptively balance the multi-objective loss scale, and the robustness of the step selection and the convergence precision are ensured to be improved.

The invention belongs to the technical field of large-span space structure design, and provides a non-linear form optimization method and system for a free-form surface single-layer latticed shell structure, which takes the minimum overall strain energy of the structure as an optimization target and coordinates of latticed shell nodes as optimization variables to construct a non-linear optimization equation of a single-layer latticed shell form and convert the non-linear optimization equation into an unconstrained optimization equation. Solving an optimization problem by adopting a conjugate gradient method; based on solution of a conjugate gradient method, the sensitivity of strain energy to optimization variables is determined; obtaining a nonlinear strain energy gradient based on the sensitivity so as to determine a conjugate gradient direction; based on the conjugate gradient direction, carrying out line search according to a preset load and a step length until the line search in the conjugate gradient direction meets a convergence condition; the overall strain energy minimization, obtained based on nonlinear calculation, of the structure serves as an optimization target, coordinates of latticed shell nodes serve as optimization variables, a conjugate gradient method serves as an optimization algorithm, and the influence of geometric nonlinearity of the structure under the load effect is fully considered.

1. The name of the design product: line search integrated detector (NF-288). 2. The use of the design product: for detecting network lines, line search, and line matching functions. 3. The design points of the design product: in shape. 4. The picture or photo that best indicates the design points: kit 1 perspective view 1. 5. Other circumstances that need to be explained: kit 1 is a detection signal transmitter, and kit 2 is a detection signal receiver.

The invention relates to the field of power grid reliability assessment, in particular to a power grid reliability assessment method fusing hyper-sphere space clipping and cross entropy important sampling, and the method comprises the steps: seeking an actual IS-PDF which is closest to a zero variance IS-PDF in a KL distance in a remaining subspace after hyper-sphere clipping; cross entropy iterative optimization of an important sampling density function and dynamic cutting of a hypersphere are fused together in a transformed high-dimensional standard normal space, that is, one-dimensional line search is carried out on all fault points generated by current iterative optimization of cross entropy parameters, so that a junction point set is obtained, and hypersphere cutting is carried out according to the junction point set; and next cross entropy parameter iterative optimization is carried out in the remaining subspace after the hyper-sphere is cut, so that the cutting accuracy of the hyper-sphere to the normal region and the cross entropy parameter optimization efficiency of the important sampling density function are remarkably improved, and the number of iterative optimization times is reduced.