228 results about "Newton's method" patented technology

A computerized system and method for large scale semi-supervised learning is provided. The training set comprises a mix of labeled and unlabeled examples. Linear classifiers based on support vector machine principles are built using these examples. One embodiment uses a fast design of a linear transductive support vector machine using multiple switching. In another embodiment, mean field annealing is used to form a very effective semi-supervised support vector machine. For both these embodiments, the finite Newton method is used as the base method for achieving fast training

A computerized system and method for large scale semi-supervised learning is provided. The training set comprises a mix of labeled and unlabeled examples. Linear classifiers based on support vector machine principles are built using these examples. One embodiment uses a fast design of a linear transductive support vector machine using multiple switching. In another embodiment, mean field annealing is used to form a very effective semi-supervised support vector machine. For both these embodiments the finite Newton method is used as the base method for achieving fast training.

The invention discloses a power system short-term load forecasting method based on improving extreme learning machine (IELM) methods, which adopts an extreme learning machine (ELM) as the basic structure of a forecasting model and is an iteration-analysis learning algorithm which puts forward BFGS (Broyden, Fletcher, Goldfarb, Shanno) simulated Newton's method to give optimization, training and adjustment to network left metrics and analyze to get right metric parameters. The invention establishes the short-term load forecasting model based on improving extreme learning machine (IELM) method, puts forward extreme learning machine network reference hidden node concepts, trains the equidimensional extreme learning machine networks which have the same numbers of hidden nodes and samples, orderly clusters module values of equidimensional network right metrics vector, finds out a plurality of corresponding module value break points, and regards the break points as the reference hidden nodes of predicting networks. The method of the invention is also greatly improved in the aspect of prediction precision and speed.

The invention discloses a rectangular coordinate Newton method load flow calculation method with a changeable Jacobian matrix. The method includes the following steps that original data input and voltage initialization are conducted; a node admittance matrix is formed; power and voltage deviations are calculated, and the maximum amount of unbalance delta Wmax is obtained; the Jacobian matrix J is formed; a correction equation is solved and a real part e and an imaginary part f of voltage are corrected; node and circuit data are output. According to the rectangular coordinate Newton method load flow calculation method, a Jacobian matrix calculation method different from that used in the following iteration processes is adopted in the initial iteration process, and the convergence problem of rectangular coordinate Newton method load flow calculation in analyzing a system with a small-impedance branch circuit is solved. When misconvergence happens with conventional rectangular coordinate Newton method load flow calculation, the rectangular coordinate Newton method load flow calculation method with the changeable Jacobian matrix can achieve reliable convergence, and the number of iterations is fewer compared with the prior art. The rectangular coordinate Newton method load flow calculation method with the changeable Jacobian matrix can effectively solve the convergence problem of the conventional rectangular coordinate Newton method load flow calculation in analyzing the system with the small-impedance circuit branch, and meanwhile load flow calculation can be performed on normal systems, and adverse effects are avoided.

A method for analyzing limit pretension force of an angular contact ball bearing under fixed-position pretension of a high-speed electric spindle belongs to the technical field of bearing pretension. According to the method, firstly, parameters including a bearing type, a working condition, maximum permissible stress of a material and the like are determined; secondly, based on a Hertz contact theory, the influences of pretension force, centrifugal force and gyroscopic torque of the bearing are considered, and a kinetic model of the angular contact ball bearing under the fixed-position pretension is established, wherein the kinetic model includes a contact region model of the angular contact ball bearing, a pretension model of the angular contact ball bearing in a static state of the spindle, and a pretension model of the angular contact ball bearing in a rotating state of the spindle; thirdly, a relation between the pretension force and maximum contact stress of a bearing rolling body and bearing inner and outer rings under the condition of rated speed is solved with a Newton method; and finally, the pretension force, i.e., the limit pretension force, of the bearing under the permissible stress of the material is analyzed out.

The present invention provides a five-axis machining track contour error double closed-loop compensation method, belongs to the multi-axis numerical control process technology field, and relates to a contour error double closed-loop compensation method configured to improve the five-axis machining precision based on inner ring prediction compensation and outer ring feedback compensation of contour errors. Based on model prediction and feedback correction, the method estimates motion position of each physical axis at the next moment, employs tangential inversing and the Newton's method to calculate tip point contour errors and cutter axis direction contour errors of the next moment and the current moment, and the contour errors are subjected to prediction compensation at the inner ring and the contour errors are subjected to prediction compensation at the outer ring according to inversed Jacobian matrix so as to improve the five-axis machining track contour precision. The five-axis machining track contour error double closed-loop compensation method inhibits the residual errors caused by outside disturbance and model errors on the basis of realizing control of the five-axis machining contour errors in advance so as to have important meanings for improving the five-axis numerical control machining precision.

A system and method for implementing a maximum likelihood estimator for making a joint estimation of range, range rate, and acceleration of a target utilizing a pulse doppler radar. The MLE of target motion parameters are determined by keystone processing a baseband signal, and generating a first estimate of the motion parameters based on the processed signal. The first estimate is utilized to set up sampling intervals for the performance of a coarse search. Then a fine search is performed using Newton's method to determine the MLE.

The invention discloses a frequency selectivity MIMO (multiple input multiple output) system space-time blind equalizer method based on a MNM (modified Newton method). The method comprises the following steps that firstly, a multimode method and a soft decision guide method are combined for building a cost function; then, a batch processing mode is adopted, the training sequence length is selected, and the initial value of the blind equalizer is estimated from the known emitted data and the corresponding observing data by a least square method; a newly provided MNM is utilized for calculating the optimal equalizer from observing samples; and finally, the equalizer is utilized for evenly receiving signals, the interference among signals and the interference among codes are eliminated, and resources signals are recovered. The method overcomes the defects that when the exiting MNM is applied to a MIMO system, the calculation quantity is great, and the equalization precision of the existing blind equalization algorithm is low. Compared with the exiting method, the method has the advantages that MNM, MMA and SDD only need little training sample data, and the equalizer can be enabled to be correctly converged to the optimum MMSE (minimum mean square error) equalizer.

The present invention discloses a Matlab sparse matrix-based Cartesian coordinate Newton power flow calculation method. Matrix operation and complex operation are adopted, and a sparse matrix technique provided by the Matlab is used. The method of the invention is realized on the Matlab platform, so that scientific researchers can be facilitated to use various tools and functions provided by the Matlab to test and analyze calculation results; and the main part of the method of the invention adopts the matrix operation and complex operation, and therefore, program codes can be reduced, programming is simplified, a program is more clear, the scientific researchers can be facilitated to modify, debug and improve the program and add new features. The sparse matrix technique of the Matlab and the matrix operation are adopted, and therefore, calculation speed can be greatly improved; and the sparse matrix of the Matlab is very easy to use, and line and column numbers can be adopted to directly use the elements of the sparse matrix just as a full matrix, and an excellent analytical tool is provided for the scientific research work of the scientific researchers.

The invention discloses a deep learning target detection method. The method comprises the following steps: inputting a picture, and extracting features of a target in the picture through a convolutional neural network backbone network; firstly, generating a plurality of sliding windows with different scales and length-width ratios through a regional candidate network, and then obtaining candidateboxes possibly containing targets through classification and frame regression operation; mapping to a feature layer of a target to be detected through the candidate box, and performing pooling operation on an interested target area to obtain a pooled feature map with the same dimension of the target area; respectively performing classification and frame regression operations on the feature layer of the target to be detected; and obtaining a final target detection box after non-maximum suppression. The method has the advantages that the problem that the target function of the detection networkcannot be optimized when the prediction frame and the real frame are not overlapped is avoided; a Newton method can be used to update network weight, and detection network training is faster. High target detection result accuracy is realized.

The invention discloses an improving method of energy flow calculation of a natural gas system. The method comprises the following steps: firstly, inputting network parameters of the natural gas system to establish a natural gas system energy flow calculation model; optimizing the initial value of the natural gas system by means of a genetic algorithm; and finally, calculating the energy flow of the natural gas system by means of a Newton method. By combining global search of the genetic algorithm and local searching abilities of the Newton method, the problem of selecting the initial value of natural system energy flow calculation is effectively solved.