202 results about "Function approximation" patented technology

A function approximation processing method comprises inputting and binarizing image data, extracting contour from the binarized image data, estimating tangent points in horizontal and vertical directions from the contour, and approximating contour between adjacent tangent points among the estimated tangent points with a predetermined function, thereby an input object such as a character or a figure can be processed at high speed, and outline data with high image quality can be generated with a reduced number of points.

A method and system for creating hypercomplex representations of data includes, in one exemplary embodiment, at least one set of training data with associated labels or desired response values, transforming the data and labels into hypercomplex values, methods for defining hypercomplex graphs of functions, training algorithms to minimize the cost of an error function over the parameters in the graph, and methods for reading hierarchical data representations from the resulting graph. Another exemplary embodiment learns hierarchical representations from unlabeled data. The method and system, in another exemplary embodiment, may be employed for biometric identity verification by combining multimodal data collected using many sensors, including, data, for example, such as anatomical characteristics, behavioral characteristics, demographic indicators, artificial characteristics. In other exemplary embodiments, the system and method may learn hypercomplex function approximations in one environment and transfer the learning to other target environments. Other exemplary applications of the hypercomplex deep learning framework include: image segmentation; image quality evaluation; image steganalysis; face recognition; event embedding in natural language processing; machine translation between languages; object recognition; medical applications such as breast cancer mass classification; multispectral imaging; audio processing; color image filtering; and clothing identification.

Method of incrementally forming and adaptively updating a neural net model are provided. A function approximation node is incrementally added to the neural net model. Function parameters for the function approximation node are determined and function parameters of other nodes in the neural network model are updated, by using the function parameters of the other nodes prior to addition of the function approximation node to the neural network model.

The present invention is a method and an apparatus for reward-based learning of policies for managing or controlling a system or plant. In one embodiment, a method for reward-based learning includes receiving a set of one or more exemplars, where at least two of the exemplars comprise a (state, action) pair for a system, and at least one of the exemplars includes an immediate reward responsive to a (state, action) pair. A distance metric and a distance-based function approximator estimating long-range expected value are then initialized, where the distance metric computes a distance between two (state, action) pairs, and the distance metric and function approximator are adjusted such that a Bellman error measure of the function approximator on the set of exemplars is minimized. A management policy is then derived based on the trained distance metric and function approximator.

A method is described for predistorting an input signal to compensate for non-linearities caused to the input signal in producing an output signal. The method comprises: providing an input for receiving a first input signal as a plurality of signal samples, x[n], to be transmitted over a non-linear element; providing at least one digital predistortion block comprising, a plurality of IQ predistorter cells coupled to the input, each comprising a lookup table (LUT) for generating an LUT output The at least one digital predistortion block block is configured to apply interpolation between LUT entries for the, plurality of LUTs; and generate an output signal, y[n], by each of the plurality of IQ predistorter cells by adaptively modifying the first input signal using interpolated LUT entries to compensate for distortion effects in the non-linear element. A combiner may be provided configured to combine the output signal samples, yQ, from the plurality of IQ predistorter cells into a combined signal to generate the output signal, y[n], for transmission to the non-linear element. An error calculation block may be coupled to a digital predistortion adaptation block to determine and modify a predistortion performance.

A method for diagnosing analog circuit fault based on vector-valued regularized kernel function approximation, includes steps of: step (1) acquiring a fault response voltage signal of an analog circuit; step (2) carrying out wavelet packet transform on the collected signal, and calculating a wavelet packet coefficient energy value as a characteristic parameter; step (3) utilizing a quantum particle swarm optimization algorithm to optimize a regularization parameter and kernel parameter of vector-valued regularized kernel function approximation, and training a fault diagnosis model; and step (4) utilizing the trained diagnosis model to recognize circuit faults.

The invention discloses a method for predicating the output power of a photovoltaic power generation system based on a BP (Back Propagation) neural network model. The method comprises the following steps of selecting influencing factors of the output power of the photovoltaic power generation system; generating input vectors according to the historical data of the selected influencing factors, and utilizing the historical data of the output power of the corresponding photovoltaic power generation system as output, thereby obtaining training samples; training a BP neural network by utilizing the training samples, thereby obtaining the trained BP neural network; generating test input vectors according to real data at the to-be-predicated moment of the selected influencing factors, inputting the test input vectors into the trained BP neural network, so that the output is the predicted value of the output power of the photovoltaic power generation system at the to-be predicated moment. According to the method for predicating the output power of the photovoltaic power generation system, modeling prediction is performed; the invention provides a predication method based on the BP neural network; the favorable nonlinear function approximation capability of the BP neural network is utilized, so that the precision and generalization capability of a prediction model are improved.

A method and apparatus are used to predistort input signal samples according to Volterra Series Approximation Model using one or more digital predistortion blocks (300) having a plurality of predistorter cells (301-303), each including an input multiplication stage (366-367) for combining absolute sample values received from an absolute sample delay line (362) into a first stage output, a lookup table (368) connected to be addressed by the first stage output for generating an LUT output, and a plurality of output multiplication stages (371-372, 373-374) for combining the LUT output with samples received from the amplitude sample delay line (362) and signal sample delay line (363) to generate an output signal sample yQ from said predistorter cell, where the output signal samples yQ from the predistorter cells are combined at an output adder circuit (375) to generate one or more Volterra terms of a combined signal (yOUT[n]).

A method for the problem of reconstructing a watertight surface defined by an implicit equation from a finite set of oriented points. As in other surface reconstruction approaches disctretizations of this continuous formulation reduce to the solution of sparse least-squares problems. Rather than forcing the implicit function to approximate the indicator function of the volume bounded by the surface, in the present formulation the implicit function is a smooth approximation of the signed distance function to the surface. Then solution thus introduced is a very simple hybrid FE / FD discretization, which together with an octree partitioning of space, and the Dual Marching Cubes algorithm produces accurate and adaptive meshes.

The invention discloses a photovoltaic power generation power prediction method and system. A correlation analysis method is adopted to analyze historical data and determine a radiation intensity predication correlation time and a power generation power predication correlation time. A BP neural network is adopted to train a solar radiation intensity prediction sample and a photovoltaic power generation power prediction sample so as to obtain a solar radiation intensity prediction model and a photovoltaic power generation power prediction model. The solar radiation intensity prediction model is utilized to compute sun radiation intensity at prediction time of a prediction day; the photovoltaic power generation power prediction model is utilized to compute photovoltaic power generation power at prediction time of the prediction day. A grey relational analysis method is adopted to remove solar radiation intensity at the radiation intensity correlation time with a low relational degree in the historical data, and the predication accuracy of the solar radiation intensity is improved. By the adoption of the good nonlinear function approximation capability of the BP neural network, the solar radiation intensity prediction sample and the photovoltaic power generation power prediction sample are trained, the prediction models are built, and predication accuracy of the prediction models is improved.

The invention discloses the optimal consistency control method and system of a nonlinear multi-agent system. The method is characterized by establishing a reference behavior model according to the individual dynamic characteristic of a heterogeneous multi-agent system, and using a leader-follower control model to form a multi-agent system formed by reference behavior models; then, constructing a dynamic graph game global error dynamical model according to the network topology structure of multiple agents, defining a multi-agent local performance index function, and according to the global Nashequilibrium, acquiring a Bellman optimal equation; and then, under the condition of only using local agent information, using an execution-evaluation execution network framework mode based on value function approximation to carry out online iterative learning, and acquiring an optimal consistency protocol to achieve the consistency of each reference model behavior. Compared with the prior art, byusing the method and the system of the invention, under the condition of guaranteeing optimal control performance, the consistency problem of the complex multi-agent system can be high-efficiently solved, and an actual application value and high scalability are achieved.

The invention provides a dynamic vehicle weighing method and system; all shafts of the vehicle to be weighed subsequently drive pass a weighing platform, wherein, the method includes the following steps: the shaft load signal and the shaft type signal of all the shafts of the vehicle to be weighed are collected; the dynamic weighed data and shaft type data of the shafts are generated according to the shaft load signal and the shaft type signal of the shafts; the dynamic weighed data is processed to obtain and weigh effective data; trigonometric function approximation processing is conducted on the shaft weight effective data to obtain the shaft weight benchmark data and the revised data; the shaft weight is obtained according to the shaft type data and the shaft weight benchmark data and the revised data. The invention is used to solve the problem of dynamic weighing of vehicles passing the weighing platform at medium and low speed, and provide weighing schemes meeting the weighing precision requirement for different vehicle types.

The invention discloses a method for forecasting a short term load under demand response. The method comprises the following steps of: analyzing and selecting influence factors for forecasting the load, and acquiring historical data of the load to obtain a training sample set; generating an input variable according to the historical data of the influence factors and taking the historical data of a corresponding load valve as an output to obtain a training sample; training a support vector machine model by using the training sample to obtain the trained support vector machine model; and generating a test input vector according to actual data of the influence factors at the moment to be forecasted, inputting the test input vector to the trained support vector machine model and taking output of the trained support vector machine model as a load forecasting value of the moment to be predicted. The method for forecasting the short term load under the demand response, which is disclosed by the invention, is a short term load forecasting method based on the support vector machine; and the precision and the generalization capability of a forecasting model are improved by using the favorable nonlinear function approximation capability of the support vector machine.

The invention provides a color calibration optimization algorithm based on a projector-camera system, which is used for eliminating influence on immersion and reality sense of an observer caused by color difference of a display system of a large multi-projection screen. In the invention, for the problem of neglecting influence of a camera in the previous algorithm, a high dynamic range image (HDRI) synthesis algorithm based on Pedro image segmentation is proposed so as to establish a mapping relation between an input pixel and an output pixel of the camera; brightness and chrominance of a projector are calibrated at the same time by virtue of a combined function approximation method for a YUV (luminance and chrominance) color space transformation model so as to generate a color lookup table (CLUT); graphic processing unit (GPU) rendering for color calibration is realized based on shading language; and finally the three-channel annular multi-projection screen is taken as an experimental platform to verify effectiveness of the algorithm.

The invention discloses a two-wheeled self-balance robot self-adaptive sliding mode changing structure control method and system. According to a classical mechanics analytic method and based on a Lagrange algorithm based on energy analysis, molding of a kinetic equation of a two-wheeled self-balance robot of is realized, a sliding mode changing structure controller is designed according to the kinetic equation. The sliding mode changing structure controller comprises a speed sliding mold changing structure controller and an angle sliding mold changing structure controller; the speed sliding mold changing structure controller and the angle sliding mold changing structure controller give each other feedbacks; a feedback equation is that theta r=beta V; and self-adaptive controlling is carried out on the system on the basis of a function approximation mode. Through adoption of the technical scheme of the present invention, the modeling process is enabled to be more simplified and comprehensive; the robustness and the respond speed of the system are raised; simultaneously, since a mutual feedback relation exists between the speed and the angle of the system, when an inclination angle of the system is overlarge, the system will automatically decelerate; while the speed is reduced, the system will return to the balance position; in a condition of facing different road surface conditions, the system can adapt to external environment and large scope load changes, thereby guaranteeing the safety and stability of the system.

The invention discloses a neutral network theory-based non-linear system adaptive proportional integral control method. The method comprises the following steps: in a first step, a mathematical model for a non-linear system is built; in a second step, a smooth function is used for approximating a non-smooth performer saturation function; in a third step, a neutral network adaptive proportional plus integral controller is designed for control. According to the method disclosed in the invention, as for a non-linear system with input saturation, the smooth function is used for approximating the performer saturation function, a BLF is referenced, that neutral network input is maintained in a bounded compact set range can be ensured, and normal operation of a neutral network can be ensured; compared with a conventional PI gain adjustment, an adjustment method put forward in the invention is advantageous in that 1) proportion integral gain of the PI controller is not a fixed constant but a time variant; 2) proportional gain and integral gain are not designed individually but associated with each other via a certain coefficient, and therefore system analysis can be facilitated; 3) the method has certain robustness for nondeterminacy and input saturation of the system.

A method and apparatus are used to predistort input signal samples according to Volterra Series Approximation Model using one or more digital predistortion blocks (300) having a plurality of predistorter cells (301-303), each including an input multiplication stage (366-367) for combining absolute sample values received from an absolute sample delay line (362) into a first stage output, a lookup table (368) connected to be addressed by the first stage output for generating an LUT output, and a plurality of output multiplication stages (371-372, 373-374) for combining the LUT output with samples received from the amplitude sample delay line (362) and signal sample delay line (363) to generate an output signal sample yQ from said predistorter cell, where the output signal samples yQ from the predistorter cells are combined at an output adder circuit (375) to generate one or more Volterra terms of a combined signal (yOUT[n]).

An input image is divided into a plurality of regions, and it is determined whether each of the divided regions is suitable for thinning processing. In accordance with a result of the determination, an outlining processing is selected to generate outline data (vector data) for each of the regions. The generated outline data is output. For example, a character region and a drawing region are discriminated from each other and outline data having a format suitable for a discriminated type of region is generated. In addition, generation of outline data (vector data) which pass through a center line of the line drawing or generation of outline data (vector data) indicating a contour of a drawing is automatically selected. Furthermore, in accordance with the size of a character, function-approximation processing may be selected.

The invention provides a rapid evaluation method of the disturbing gravity along a flight trajectory for the first time. The rapid approximation method with ballistic missiles as research objects aims at solving the problem of rapid evaluation on the disturbing gravity in the missile flight process. The rapid evaluation method comprises generating a standard trajectory according to a launch task; generating a flight pipeline with the standard trajectory as a benchmark and performing airspace subdivision; performing evaluation on the node disturbing gravity by a point quality method or a high-order spherical harmonic function method; calculating a disturbing gravity value of any point on an actual trajectory based on a network function approximation theory inside a calculation unit and achieving the whole rapid evaluation calculation on the disturbing gravity along the flight trajectory. Compared with the existing method, the rapid evaluation on the disturbing gravity along any flight trajectory can be implemented, the trajectory calculation requirements for the evaluation precision of the rapid evaluation method are met, the evaluation speed, the data storage and other indexes are excellent, and the real-time evaluation on the disturbing gravity missiles is implemented.

A discriminating method of parallel series patter and its application in the machinery smell, the characteristics is that regarding the big scale smell sorts and intension confirm problem as assortment problem, and then regarding as approximation problem, giving settling respectively for the mode which is made up of classifier layer and function approximation mode layer. It includes the four implementation steps: (1) n pieces one output neural net of classifier layer takes charge of ensuring the types of n kinds of smell, one to one correspondence. The training subset of a network is made up of all representative smell samples and the part apposition smell samples. (2) n pieces one output neural net of function approximation mode layer takes charge of ensuring the intensity of n kinds of smell, one to one correspondence. The training subset of a network is made up of all representative smell samples. (3) The suppositional balance of imbalance training subset. (4) n+1 pieces one output neural net puts up decision. By using the invention, the machinery olfactometer can solve the simultaneous estimation problem of thousands upon thousands smell types and intensity.

The invention discloses a network flow estimating method which concretely comprises the steps of initializing, obtaining a sigma point set, predicting the state, estimating the state and updating the process. In the method, the flow matrix estimating problem is modeled into a nonlinear system which more meets the OD (Origin-Destination) flow real characteristic; the sigma point set obtained by UT (Unscented Transformation) is subjected to nonlinear transformation; the coefficient of the system sate equation needed by nonlinear transformation is obtained concretely by Chebyshev polynomial fitting instead of approximately obtaining the system state equation by the traditional local linearization; accordingly, the state equation of the system does not need to meet the available linear function approximating condition; a symmetrical sampling policy is adopted for the UT; the particle point set approaches the probability density function distribution of a nonlinear function to obtain the higher-order approximation of state estimation, the result of OD flow estimation has higher precision, and the calculating complexity of the system is reduced.

The invention relates to a fractional order electromagnetic anomalous diffusion three-dimensional simulation method of rational function approximation, and aims at calculating three-dimensional time-domain induction-polarization double field response of a fractional order Cole-Cole model. The method mainly comprises the steps of establishing a Cole-Cole model fractional order transfer function andan n-odrer rational approximation function based on a frequency domain rational function approximation method, and making the sum of the real part and the imaginary part of an error function as a target function; achieving linearization of the target function through an instrumental variable technique, and obtaining an optimal approximation rational function by adopting a linear programming approach; obtaining a time-domain mode of a conductivity by adopting a partial fraction expansion method and inverse laplace transformation; putting the time-domain mode in a Maxwell equation, deducing aniterative equation of an electromagnetic field based on a finite difference method, and achieving fractional order Cole-Cole model three-dimensional electromagnetic response value calculation. The fractional order electromagnetic anomalous diffusion three-dimensional simulation method of the rational function approximation has the advantages of fast and accurately simulating the fractional order Cole-Cole model three-dimensional time-domain electromagnetic response and providing theoretical foundation for research on abnormal electromagnetic diffusion in a polarization medium.

A method and a system for image editing are provided. The method includes the following steps: firstly, search for characteristic curves of an image; next, approximate the characteristic curves with spline functions; then, when editing the image, manipulate the characteristic curves through control points of the corresponding spline functions; and then, reconstruct the image according to the manipulated characteristic curves.

Function approximation performed when a raster image is converted into a vector image is performed in a simple manner with high accuracy, without using feedback. When anchor points are extracted from a coordinate point sequence obtained from the raster image, and function approximation is performed on the coordinate point sequence between anchor points, an appropriate point among coordinate points defined in a unit approximation section that is partitioned by anchor points is selected, and after setting the direction of the corresponding coordinate point as a tangential direction, correction is performed such that the position of a control point obtained from a tangent line does not intersect another control point.

The invention discloses a design method for designing a two-path orthogonal graph filter group, models a design problem of the two-path orthogonal graph filter group into a band constraint optimization problem, takes the reconstruction error of a filter as a target function and takes stopband attenuation as a constraint condition. Therefore, an iterative method is adopted to solve the problem. In single-step iteration, through a Taylor formula and functional approximation, a highly nonlinear non-convex target function is converted into a convex quadratic function, and a non-convex optimization problem is approximate to the subproblem of convex optimization. By use of the method, the two-path orthogonal graph filter group with better integral performance can be obtained.

An image identification apparatus includes an acquisition unit configured to acquire an input image, a classification unit configured to calculate a classification score of the input image based on a neural network having a weight coefficient in each layer determined to calculate the classification score indicating a degree of similarity between a training image and an image to be processed, an anomaly determination unit configured to calculate an anomaly score of the input image based on a function approximator constructed by machine learning based on the training image of correct answer, and an identification unit configured to classify the input image into a good image having a high degree of similarity to the training image of the correct answer or a bad image having low degree of similarity to the training image of correct answer based on the classification score and the anomaly score.

The invention discloses an SVDD radar target one-dimensional distance image identification method based on density weight and a hybrid kernel function and belongs to the radar target identification field. The K-type kernel function has strong generalization capability, extraction of global characteristics of training data is facilitated, complex index operation of a radial basic kernel function isavoided, and properties of small calculation complexity of a polynomial kernel function and high approximation precision of the radial basic kernel function are realized. The radial basic kernel function has quite good local characteristics, the K-type kernel function and the radial basic kernel function are combined to replace a kernel function of a traditional SVDD algorithm, moreover, a localdensity algorithm based on interception distance is employed to calculate local density between a support vector and training sample data in the high-dimensional kernel characteristic space, accordingto density distribution, the shape of a super-closed ball is adjusted, and identification performance of radar one-dimensional distance image single type targets is effectively improved.

Systems and methods are provided for controlling coherence of a magnetic resonance signal of spin species. The small difference between hard π pulses and their delta-function approximation is exploited to provide new classes of spin echoes which have applications in nuclear magnetic resonance (NMR) spectroscopy, magnetic resonance imaging (MRI) and magnetic resonance microscopy (MRM), and related spectroscopies of solids, and mixtures of solids and liquids. Systems and methods are also provided for controlling coherence of the resonance signal from pseudospin species.