219 results about "Weighing matrix" patented technology

An apparatus for achieving an accelerator of a sparse convolutional neural network is provided. The apparatus comprises a convolution and pooling unit, a full connection unit and a control unit. Convolution parameter information and input data and intermediate calculation data are read based on control information, and weight matrix position information of a full connection layer is also read. Then a convolution and pooling operation for a first iteration number of times is performed on the input data in accordance with the convolution parameter information, and then a full connection calculation for a second iteration number of times is performed in accordance with the weight matrix position information of the full connection layer. Each input data is divided into a plurality of sub-blocks, and the convolution and pooling unit and the full connection unit perform operations on the plurality of sub-blocks in parallel, respectively.

A memory-centric neural network system and operating method thereof includes: a processing unit; semiconductor memory devices coupled to the processing unit, the semiconductor memory devices contain instructions executed by the processing unit; weight matrixes including a positive weight matrix and a negative weight matrix constructed with rows and columns of memory cells, inputs of the memory cells of a same row are connected to one of Axons, outputs of the memory cells of a same column are connected to one of Neurons; timestamp registers registering timestamps of the Axons and the Neurons; and a lookup table containing adjusting values indexed in accordance with the timestamps, the processing unit updates the weight matrixes in accordance with the adjusting values.

The present invention relates to a method of generating two beams, having orthogonal polarizations, covering a selected area using an antenna (20) comprising multiple dual-polarized array elements (11). Each dual-polarized array element having a first phase centre (18) associated with a first polarization and a second phase centre (18) associated with a second polarization. The method comprises: designing a first weight matrix having a first non-zero weight vector for the first polarization and a second non-zero weight vector for the second polarization, calculating a second weight matrix based on the weight vectors of the first weight matrix, and applying the first and second weight matrix to the dual-polarized array elements to generate a second beam covering the selected area.

A method and system separates components in individual signals, such as time series data streams. A single sensor acquires concurrently multiple individual signals. Each individual signal is generated by a different source. An input non-negative matrix representing the individual signals is constructed. The columns of the input non-negative matrix represent features of the individual signals at different instances in time. The input non-negative matrix is factored into a set of non-negative bases matrices and a non-negative weight matrix. The set of bases matrices and the weight matrix represent the individual signals at the different instances of time.

The present invention smoothes a spherical graph signal composed of spherical signal points associated with graph vertices of a graph producing a smoothed spherical graph signal composed of smoothed spherical signal points. Each smoothed spherical signal point is computed by multiplying a vertex rotation matrix by the corresponding spherical signal point. The vertex rotation matrix is computed as a weighted average of neighbor rotation matrices using a local parameterization of the group of rotations. The present invention also filters anisotropically a graph signal composed signal points associated with graph vertices of a graph producing a filtered graph signal composed of filtered signal points. Each filtered signal point is computed as a weighted average of signal points corresponding to the corresponding graph vertices and neighbor graph vertices with neighbor weight matrices. The present invention also denoises the vertex positions of a polygon mesh without tangential drift. The face normals are smoothed on the dual graph of the polygon mesh. The smoothed face normals are used to construct neighbor weight matrices on the primal graph of the polygon mesh. The vertex positions are anisotropically filtered on the primal graph of the polygon mesh. The present invention also filters the vertex positions and face normals of a polygon mesh with interpolatory vertex positions and face normal constraints.

The invention belongs to the field of engineering risk evaluation, and particularly relates to an adjacent building safety quantitative evaluation method in tunnel construction. The adjacent building safety quantitative evaluation method in the tunnel construction mainly comprises the following steps of step 1, building an evaluation index system, step 2, building a standard extensible cloud model of evaluation indexes, step 3, computing and obtaining a weight matrix of the evaluation indexes through a subjective and objective weighting method, step 4, computing a relevancy matrix of a to-be-evaluated object through cloud relevancy computing, and obtaining a safety evaluation grade of the to-be-evaluated object through a weighted average method and combination of an evaluation index weighted value and the relevancy matrix, and step 5, measuring the reliability of safety evaluation grade. The adjacent building safety quantitative evaluation method in the tunnel construction combines advantages of a cloud model and an extensible theory, and provides effective decision support advices for protection and control of an adjacent building of different safety grades.

The invention discloses a method for detecting abnormal events of a wireless sensor network in a distributed way. The method comprises the following steps of: (1) acquiring the initial data of all the sensor nodes, and determining observation number of networks; (2) clustering all the nodes, acquiring the current observation number of each cluster, and constructing an initial vector for each cluster; (3) acquiring the current observation vector of each cluster; (4) acquiring the current reconstructed data vector of each cluster; (5) acquiring the current restored data vector of each cluster; (6) judging whether each node in each cluster has leaking detection or a false detection or not; (7) judging whether the times of weighing base tracing reconstruction of each cluster reach a preset value or not, if so, performing step (10), otherwise, updating weighing matrix parameters of the weighing base tracing reconstruction of each cluster; (8) updating the observation number of each cluster; (9) after updating the random Gauss matrix of each cluster, returning to the step (3); and (10) calculating the total number of elements of which the element value is 1 in the restored data vector of all the clusters.

A vehicle driving risk assessment method belongs to the technical of vehicles. The vehicle driving risk assessment method is technically characterized by comprising, S1, utilizing Internet of vehicle equipment to collect data of braking, steering, acceleration and travelled distance of vehicles to be tested; S2, establishing an assessment factor set; S3, structuring the membership functions of assessment factors; S4, providing object comments; S4, calculating the membership degree vector of every single factor to form a membership degree matrix; S6, establishing an assessment factor weight matrix; S7, performing operation of B=RoA on the membership degree matrix R and the assessment factor weight matrix A to obtain an assessment result. The vehicle driving risk assessment method can perform comprehensive assessment on driving risks as well as independent risk factors, thereby being beneficial to analyzing the influence of every dangerous driving factor on the driving risks.

A reconstruction process includes measuring structures formed on a substrate by a lithographic process, determining a reconstruction model for generating modeled patterns, computing and minimizing a multi- variable cost function including model errors. Errors induced by nuisance parameters are modeled based on statistical description of the nuisance parameters' behavior, described by probability density functions. From the statistical description model errors are calculated expressed in terms of average model errors and weighing matrices. These are used to modify the cost function so as to reduce the influence of the nuisance parameters in the reconstruction, without increasing the complexity of the reconstruction model. The nuisance parameters may be parameters of the modeled structure, and / or parameters of an inspection apparatus used in the reconstruction.

A method and system separates components in individual signals, such as time series data streams. A single sensor acquires concurrently multiple individual signals. Each individual signal is generated by a different source. An input non-negative matrix representing the individual signals is constructed. The columns of the input non-negative matrix represent features of the individual signals at different instances in time. The input non-negative matrix is factored into a set of non-negative bases matrices and a non-negative weight matrix. The set of bases matrices and the weight matrix represent the individual signals at the different instances of time.

The invention provides a sparse characteristics based video image de-blurring method, comprising the following steps: a modeling step of establishing a de-fuzzy model of weighted total variation regularization constraint described in the figure wherein / x represents a restored clear image, B a fuzzy kernel, y a generated blur image, [miu] an adjustable parameter, W a diagonal matrix of weights, and Vx includes the first-order differentials in the transverse, longitudinal and diagonal directions; and a solving step where an alternating iterative method is used to obtain the updated pixel weights of the / x and repetitive iterations are performed until the / x is converged. According to the invention, the de-blurring of a blurred image can be realized for a restored clear image. Moreover, the non-convex constraint model is improved through the method of weight assignment, and the convex optimization model is established to make the first-order differential constraint model of the image sparser and the possibility for rapid solution for the model.

The invention discloses a computer virtual sculpturing method for calligraphy Chinese characters. The method comprises the following steps of: binarizing the original image of a two-dimensional calligraphy work, extracting the font characteristic information of the calligraphy Chinese characters, and constructing a point light source boundary for all Chinese characters in the calligraphy work; calculating the initial depth map of the calligraphy Chinese characters based on an illumination attenuation model; obtaining the stroke vigor distribution information of the calligraphy Chinese characters according to the gray value of the image of the calligraphy Chinese characters, and calculating a depth weight matrix; optimizing the initial depth map of the calligraphy Chinese characters by use of the depth weight matrix so as to construct a triangular patch; and rendering to generate a three-dimensional virtual sculpture product of the calligraphy Chinese characters. Through the invention, a virtual sculpture product of calligraphy Chinese characters with natural depth change and smooth transition of the stoke cross part can be generated; and as the writing power factor is considered in the process of calculating the pixel point depth of the calligraphy Chinese characters, the vitality of the virtual sculpture product of the calligraphy Chinese characters is enhanced. The method disclosed by the invention can be applied to the fields of virtual reality, computer-aided manufacturing and the like.

The invention discloses a regulation and control method for improving diversity of nanocarbon to soil nematodes. The method comprises the steps of adopting plastic flowerpots of which the lengths are 25cm and the diameters are 2cm, cleaning and drying gauzes, fixing each gauze which is cut to a round shape with a proper size at the bottom of each flowerpot, and weighing matrix garden soil of 2kg in each container, wherein modified nanocarbon of which the weight is 1-5% that of the matrix garden soil is added into the garden soil respectively; respectively sowing festuca arundinacea seeds of 4.5g respectively; regularly quantitatively watering in the experiment period every day to meet the plant growth demands, culturing the plants in a laboratory, wherein during the cultivation period, the temperature is 25-34 DEG C, the relative humidity is 31-48%, the lighting condition is natural incident light, the experiment time is 120 days, the nematodes in the soil are separated to identify the type and number of the nematodes as the background data of the nematodes, and the type and number of the nematodes are processed after separation and identification after 4 months.

A system and method for extending partially labeled data graphs to unlabeled nodes in a single network classification by weighting the data with a weight matrix that uses a modified graph Laplacian based regularization framework and applying graph transduction methods to the weighted data. The technique may be applied to data graphs that are directed or undirected, that may or may not have attributes and that may be homogeneous or heterogeneous.

The present invention discloses a transformer substation energy efficiency assessment method based on an AHP (analytic hierarchy process). The method comprises: acquiring n index values for m to-be-assessed transformer substations; establishing an original index data matrix based on a gray relational analysis method, and carrying out standardized processing to generate a to-be-assessed index data matrix; establishing an index hierarchical structure, determining a relative weight,relative to a top level index, of each energy efficiency assessment index of a bottom layer of the index hierarchical structure to acquire a relative weight matrix; as for the to-be-assessed index data matrix, separately calculating an index value of each energy efficiency assessment index of each to-be-assessed transformer substation after the standardized processing, and a correlation coefficient between reference values of corresponding energy efficiency indexes in a reference line, thereby acquiring a correlation coefficient matrix; and calculating an integrated assessment result matrix of m to-be-assessed transformer substations. According to the method disclosed by the present invention, a reasonable degree of energy efficiency of the transformer substations can be analyzed, a power saving magnitude before and after reconstruction of the transformer substations can also be analyzed and assessed, so that the method has the advantages of high data assessment accuracy, and high-speed and efficient assessment.

The invention discloses a self-adaptive LQG / LTR (linear quadratic Gaussian with loop transfer recovery) controller design method for an aero-engine. The method comprises steps as follows: firstly, a linear mathematical model near a steady state point is established on the basis of a high-precision non-linear model of the aero-engine, and an LQG / LTR controller is primarily designed according to thelinear mathematical model; an error function of a BP (error back propagation) algorithm is defined and enabled to meet the requirement of system performance; finally, a Kalman filter gain matrix andan optimum state feedback gain matrix in the LQG / LTR controller are optimized online with the BP algorithm, and the error function is enabled to be the minimum. Aiming at solving the problem of weightmatrix selection in the LQG / LTR controller design process and avoiding a tedious manual trial and error process, all that is required is to supply any group of weight matrixes reasonably and solve the Kalman filter gain matrix and the optimum state feedback gain matrix, the LQG / LTR controller can be optimized online with the method, and the system is enabled to achieve good dynamic property and robustness.

The invention discloses a missile performance evaluation method based on an improved analytic hierarchy process. The method comprises the following steps: determining an association relationship and amembership relationship between missile performance evaluation factors, and establishing a hierarchical model; Taking each element in the hierarchical model as an evaluation index, arranging the relative quality order of each evaluation index according to a 7-scale method, and sequentially constructing a judgment matrix of the evaluation indexes; Calculating the weight of the judgment matrix andthe product of each element in each row of the judgment matrix, and carrying out normalization processing to obtain a weight vector matrix; By utilizing a calculation result of hierarchical single sorting, synthesizing a superior and inferior order of a higher hierarchy, and performing weighted calculation to realize hierarchical total sorting to obtain a scheme weight matrix; And calculating evaluation index values of the evaluated objects through an entropy weight method, and processing results of the analytic hierarchy process and the entropy weight method according to a Lagrangian multiplier method to obtain a new weight which is a final result. The method has the advantages of simple calculation and high efficiency, and the evaluation result is objective and reliable.

The invention relates to a local linear embedded algorithm based radio frequency map unsupervised classifying method and aims at solving the problem that existing radio frequency map classifying methods can only depend on space distribution of a to-be-positioned area for classification. The method includes setting an access reference point and a test point in a to-be-positioned indoor area; determining a matrix of a Radio map according to the position of the test point and RSS (radio-frequency signal strength) of the access reference point; processing data of the Radio map, namely, storing space coordinate information in the Radio map into storage equipment, and deleting the space coordinate information to acquire actual high-dimension data X=(x1, x2, ..., xt); and building a local covariance matrix Q according to adjacent points, calculating low-dimension embedding by utilizing a local reconstruction weigh matrix W, dividing low-dimension data into S classes and judging inter-class divergence and in-class divergence, merging the classes according to ratio conditions, and acquiring a final class information matrix. The local linear embedded algorithm based radio frequency map unsupervised classifying method can be widely applied to unsupervised classification of radio frequency maps.

The invention relates to a method for evaluating service quality, and aims to provide a method for evaluating quality of a whole process of an on-line education service. The method is to build a service quality evaluating standard system which comprises four first level dimensionalities and fifteen second level dimensionalities and uses the four first level dimensionalities and fifteen second level dimensionalities to evaluate quality during different life cycles of the whole process of on-line education. The method comprises the following steps: confirming relative weight matrixes of first level indexes and relative weight matrixes of second level indexes corresponding to the first level indexes by adopting index pairs of index number; performing consistency check to the relative weight matrixes; and computing the final weight of all the indexes by adopting the relative weight matrixes in case that the relative weight matrixes pass the consistency check. The invention puts forwards the dimensionality index for evaluating the on-line education and the method for evaluating the quality of the whole process of the on-line education service, provides the all-round evaluating index and evaluating method according to a real system, realizes the diversification of evaluation objectives, the polymorphism of evaluation contents and the variety of evaluation manners, and is closer to the requirements of users.

A method formats a weight matrix in a current layer included in a neural network. The method includes calculating a row length for each row of the weight matrix based on a number of elements each of which has non-zero value; storing rearrangement information including result of sorting rows in the order of row lengths; performing a row transformation or the row transformation and a column transformation on the weight matrix using the rearrangement information; distributing rows of a transformed weight matrix to a plurality of processing elements (PEs); and generating formatted data including one or more group data each including values and column information being processed in each of the PEs.

The invention relates to a design method for linear discrimination of a sparse representation classifier based on nuclear space. The method comprises the following steps of reading training samples, performing nonlinear transformation on the training samples to transform the training samples to the high-dimensional nuclear space, learning the training samples of each kind in the high-dimensional nuclear space, finding out the contribution (namely the weight) made by each individual in the training samples of the kind to constructing the subspace of the training sample of the kind, forming dictionaries through products of the training samples of the kind and a weight matrix, and sequentially arranging the dictionaries of all kinds to form a large dictionary matrix; obtaining linear discrimination sparse codes of the test samples inside the nuclear space on the basis of the dictionary matrix, and performing fitting on the test samples through the dictionaries of each kind and linear discrimination coding corresponding to the dictionaries; adopting the kind with the minimum fitting error as the category of the test samples. It can be ensured that sparse codes of the samples of the same kind are concentrated, sparse codes of the samples of different kinds are dispersed, the sample discrimination is effectively improved, and the performance of the classifier is improved.

The invention relates to a fuzzy variable-weight engineering geological environment quality evaluation method comprising the following steps: establishing an engineering geological environment quality evaluation hierarchical structure model; acquiring engineering geological data corresponding to all participating indexes in an area to be evaluated; dividing the area to be evaluated into m basic evaluation units; establishing the evaluation index matrix and the variable weight matrix of each basic evaluation unit; calculating the variable-weight comprehensive evaluation value of each basic evaluation unit according to the evaluation index matrix and the variable weight matrix corresponding to the basic evaluation unit; getting the engineering geological environment quality grade of each basic evaluation unit according to the variable-weight comprehensive evaluation value of the basic evaluation unit; and clustering the basic evaluation units of the same engineering geological environment quality grade through use of a clustering method to get an engineering geological environment quality comprehensive evaluation partition map. The method has the advantages of high evaluation precision, objective and reliable evaluation result, and the like.

The invention relates to a construction method for a special structure protograph QC-LDPC code based on a Pascal's triangle, and belongs to the technical field of channel coding research. The method utilizes the Pascal's triangle to construct a matrix, and specifically comprises the following steps of constructing a weight matrix of the protograph, namely a base matrix Hbase; performing anticlockwise rotation by using the Pascal's triangle structure to acquire an S matrix; filling elements in the S matrix into the weight matrix Hbase of the protograph according to a rule to acquire a base matrix H of the protograph; extending the protograph base matrix H via a zero matrix and a unit matrix to acquire a check matrix, and providing a fast iterative coding algorithm of the codon according tothe structure of H. The base matrix has a quasi bidiagonal structure, avoids four rings and can achieve fast coding, is reduced in coding complexity, and has better iterative decoding performance without significant error floors.

The invention discloses a method for safety precaution of a producing area of a farm product for a long term, which comprises the following steps: taking a standard data of producing area environment of a non-polluted farm product, a product standard data and a background data of actual environment of the producing area as basis to construct a weight matrix of indexes influencing composite index of the producing area environment; calculating composite index of respective year of the producing area through historical actual measurement data of the producing area environment; and finally outputting a line graph of the composite index and a predictor formula of the producing area by using a double exponential smoothing method of a time sequence so as to carry out prediction and warning of composite index and the deterioration speed of a producing area of a future farm product. The method discloses a law that the producing area safe condition changes along with the time through the historical data of the time sequence and predicts the future, has high reliability and accuracy, can carry out prediction and prewarning of an evolution trend, the direction and the speed of the producing area safety of different areas, and can make foreseeable judgment on a system and the like so that pertinence measures are taken, alarming source is eliminated, and agroecosystem safety can be ensured; and the method has wide applicability.

The invention discloses a wind power prediction method based on grey-combined weight. Fruit-fly least squares support vector machine model prediction and grey residual least-squares-support-vector machine model prediction are carried out on original wind power data respectively to obtain training data; the training data and practical data are compared and a grey relational degree analysis is carried out to obtain grey relational degree weight matrix; a target is predicted by using a fruit-fly least squares support vector machine and a grey least squares support vector machine, two obtained results are multiplied by the obtained grey relational degree weight matrix, and results are added, thereby obtaining a final prediction result. According to the weight combination algorithm of a grey relational degree model, a big error caused by an unknowable factor of a single model can be reduced based on the combined prediction; and the weight of the model output with high prediction precision in the result is improved.

The invention provides a maintenance method and device based on wind generation set part importance evaluation. The method comprises the following steps: dividing importance grades of parts and a factor set of evaluation, and calculating an interval number grey fuzzy matrix according to importance degree of each factor in the obtained factor set with respect to each grade in the importance grades, and sufficient degree of information included in the factor set; according to an interval analytic hierarchy process, calculating weight of each factor for wind generation set part significance evaluation, and determining a weight matrix of the factor set; and on the basis of an OWA operator, according to the interval number grey fuzzy matrix and the weight matrix of the factor set, determining importance degree of a part of a wind generation set, and determining a corresponding maintenance strategy according to the importance degree of the part of the wind generation set. The method and device fully consider the degree of membership of evaluation factors to the evaluation grades as well as influence of sufficient degree of the information on importance evaluation, thereby improving accuracy of wind generation set part importance evaluation.

The invention discloses a comprehensive intelligent method for predicting an impact danger level based on micro-seismic fractal, and is applicable to the field of mine pressure impact prediction in the mine safety field. The method comprises three parts of micro-seismic fractal index input, fuzzy intelligent processing analysis and intelligent prediction result output. The method comprises specific steps of calculating micro-seismic time, space and energy fractal indexes, establishing an impact danger level membership matrix, establishing an index abnormity index and a Gaussian membership function, establishing a single-index evaluation matrix R, and calculating and updating each index weight matrix in real time by adopting an F value scoring method in a confusion matrix; calculating the membership probability of each impact danger level, and determining the final impact danger level by combining a maximum membership degree principle (MMDP) and a variable fuzzy feature recognition model (VFPR). The method is clear in mathematical model, high in universality and operability, capable of achieving intelligent probability prediction of different impact danger levels and intelligent recognition of comprehensive results, good in application feasibility and high in intelligent prediction efficiency.

A network model block compression method for a neural network, comprising: a weight matrix obtaining step, obtaining a weight matrix of a trained network model of a neural network; a weight matrix blocking step, dividing the weight matrix into an array consisting of a plurality of initial sub-blocks according a predetermined array size; a step of concentrating the weighted elementto be cut, according to the absolute values and values of the weights of the matrix elements in the sub-blocks, concentrating the matrix elements with smaller weights to the sub-blocks to be cut through row and columnexchange, so that the absolute value and value of the weights of the matrix elements in the sub-blocks to be cut are made smaller relative to the absolute values and values of the weights of the matrix elements in other sub-blocks that are not in the sub-blocks to be cut; a sub-block cut step, cutting the weights of the matrix elements in the above-mentioned sub-blocks to be cut, and obtaining the final weight matrix to implement compression of the network model of the neural network. The resource overhead can be saved, and a large-scale neural network can be arranged under the condition of limited resources.