144results about How to "Slow convergence" patented technology

The invention discloses a self-adaptive composite service optimization method based on multi-agent reinforced learning. The method combines conceptions of the reinforced learning and agents, and defines the state set of reinforced learning to be the precondition and postcondition of the service, and the action set to be the Web service; parameters for Q learning including the learning rate, discount factors and Q value in reinforced learning are initialized; each agent is used for performing one composite optimizing task, and can perceive the current state, and select the optimal action under the current state as per the action selection strategy; the Q value is calculated and updated as per the Q learning algorithm; before the Q value is converged, the next round learning is performed after one learning round is finished, and finally the optimal strategy is obtained. According to the method, the corresponding self-adaptive action strategy is worked out on line as per the environment change at the time, so that higher flexibility, self-adaptability and practical value are realized.

A method and apparatus for reducing motion blur in a motion blurred image are provided. The method includes blurring a guess image based on the motion blurred image as a function of blur parameters of the motion blurred image. The blurred guess image is compared with the motion blurred image and an error image is generated. The error image is blurred and pixels in the blurred error image are weighted based on the steepness of edges proximal to corresponding pixels in the motion blurred image. The blurred and weighted error image and the guess image are combined thereby to update the guess image and correct for motion blur.

The invention discloses a short-term prediction method for occupancy of an effective parking space of a parking lot. The short-term prediction method comprises the steps as follows: 1) determining a time sequence of the occupancy of the effective parking space of the parking lot; 2) carrying out wavelet decomposition to the time sequence of the occupancy of the effective parking space through a wavelet function, thus obtaining a low-frequency coefficient vector and a high-frequency coefficient vector; implementing the wavelet reconstruction to the low-frequency coefficient vector and high-frequency coefficient vector, so as to obtain the time sequence of N+1 reconstructions; 3) establishing a wavelet neural network model to the time sequence of the N+1 reconstructions for predicting, thus obtaining N+1 prediction results; and 4) accumulating the N+1 prediction results, so as to obtain the prediction results corresponding to the time sequence of the occupancy of the effective parking space. According to the short-term prediction method disclosed by the invention, a wavelet analysis-wavelet neural network combinational prediction model is raised to perform short-term prediction to the occupancy of the effective parking space of the parking lot according to the short-term variation characteristic of the occupancy of the effective parking space of the parking lot, therefore, the prediction accuracy and the stability are improved.

A single-channel color image encryption method based on chaos and phase retrieval process comprises a color image chaos scrambling and decomposition step, a pure phase extraction and modulation step and a chaos diffusion step. Combining fractional Fourier transform and paired Logistic chaotic mapping, an encrypted image has nonlinearity and randomness, security and convergence speed of an encryption system are raised, and the problem of slow convergence speed existing in a present encryption method is solved. Meanwhile, complexity of the color image encryption system is reduced, the system is simple, and operation is convenient.

The invention belongs to the field of target attitude estimation, and discloses a single image indoor object attitude estimation method based on thermodynamic map. The object attitude estimation method based on thermodynamic map extracts candidate frames of a plurality of target objects through an RPN network. The thermodynamic images of eight vertices of each object bounding box on two-dimensional images are extracted by full convolution network (FCN), and then the 6D attitude estimation of the object is calculated by using PnP method. ShapeNet is used as CAD model library to synthesize a large amount of training data. The object posture estimation detection technology based on the thermodynamic map adopted by the invention has strong robustness, can estimate the posture of different indoor objects when the background is disorderly indoor scene and the object is partially occluded, has wide application range, is insensitive to light and does not require the object to have obvious texture appearance.

The invention relates to a determination method based on an ANSYS second development platform of the initial completion initial cable force of a cable-stayed bridge. The method uses the design cable force as the target value, considers the geometric non-linear effect and performs iteration for solving and correcting the cable initial strain under the action of dead load. The method comprises the following specific steps: any group of cable forces is assumed to be added on the stay cable in the mode of initial strain, dead load is added to calculate, the *do-loop command language is used to extract the calculated cable force of the stay cable and check whether the error between the cable force and the target completion cable force is within an available limit; if the error is large, the indifference method is used to correct the stay cable initial strain, then the error is calculated again until the error is within the available limit; and the initial completion initial cable force can be obtained by multiplying the last calculated group of stay cable initial strains by the elastic modulus of the stay cable which is corrected by considering the vertical effect. By using the method, the solving time of the completion initial cable force of the cable-stayed bridge can be reduced and the precision can be increased, thus the method has large practical engineering application value.

The invention relates to a self-adaptive beam-forming method based on related calculation and clutter covariance matrix reconstruction. The self-adaptive beam-forming method comprises following steps: solving an array sampling covariance matrix by limited sampling snapshot data; decomposing eigenvalue of the array sampling covariance matrix and estimating signal subspace; making related calculations between eigenvector corresponding to a signal and an assumed steering vector of a desired signal and solving eigenvector corresponding to the desired signal; resetting eigenvalue corresponding to the desired signal in order to reconstruct a new covariance matrix; solving self-adaptive weight vector based on the minimum variance orthoscopic principle by the reconstructed covariance matrix; and utilizing self-adaptive weight vector to performing weighting operation on echoed data.

Provided are a method and system for keyword expansion; the method performs a search by means of an initial keyword, the searched and obtained keyword serving as the basis for the next search, and the searches being performed by means of keyword iteration; when the error of two consecutively searched keywords is within a certain range, the searched keyword serves as an expansion keyword of the initial keyword; in this way, various expressions of the initial keyword and multi-faceted implied meanings of the word are obtained, and the initial keyword is effectively and reasonably expanded, solving the problem in the prior art requiring that a corpus be manually established; the method is a convenient and highly accurate method for keyword expansion. Also provided is a method and system for classifying corpora and automatic annotation; the method determines for each class one or more initial core keywords; expanded keywords of each class are obtained by means of initial core keyword expansion; a search is performed using the expanded keywords corresponding to the classes, and a class corpus is selected therefrom and annotated.

The invention relates to a mobile robot path planning method. The method comprises the following steps: finding the shortest path of an environment; taking grids safe from a dynamic barrier on the shortest path as local target points when a robot detects in progress the dynamic barrier and will bump into the dynamic barrier in the environment; determining the motion range of the dynamic barrier; the robot advancing along the grids with high pheromone concentration; and obtaining an optimal path that is away from the dynamic barrier and passes by a specified point. The problem of low convergence speed of an ant colony algorithm due to restrictions of the ant colony algorithm itself is solved. A group of optimal solutions is obtained after parameters of a conventional ant colony algorithm are optimized by a particle swarm algorithm. A genetic algorithm is adopted to prevent local optimum, other parameters of the above parameters keep unchanged, and the pheromone concentration is amplified by times. In this way, pheromone concentration differences on different paths are more obvious, and the convergence speed of the algorithm is increased.

The invention discloses a driving motor system performance evaluation method for an electric vehicle. The driving motor system performance evaluation method analyzes from different dimensions such as motor control performance, motor body design and enterprise qualification and ability of the driving motor system according to performance characteristics of a driving motor used for the electric vehicle, adopts an analytic hierarchy process to determine a driving motor performance evaluation index system and index weights thereof, establishes a BP neural network model for driving motor system performance evaluation, organically integrates a bat algorithm with a particle swarm algorithm to form a bat-particle particle swarm hybrid algorithm, and optimizes parameters of the neural network structural model by adopting the bat-particle particle swarm hybrid algorithm. Simulation examples show that, through training and testing data samples, the driving motor system performance evaluation method which optimizes the neural network based on the analytic hierarchy process and the bat-particle particle swarm hybrid algorithm has the advantages of fast evaluation speed and high accuracy rate, achieves satisfying evaluation results, and has certain promotion value in evaluation, selection and application of a driving motor system for the electric vehicle.

The invention relates to a super short-term wind power forecasting method based on a back propagation (BP) neural network, which is technically characterized by comprising the steps that: data is grouped through a cross grouping method as the input of a training process of the BP neural network; the training of the BP neural network is carried out through modified learning rate algorithm; and a forecasting result is obtained by calculating the wind power of a wind farm. The super short-term wind power forecasting method based on the BP neural network has a reasonable design, adopts the improved algorithm of the BP neural network, solves the problem that the convergence rate is slow during the training process by modifying the learning rate in the training process, can greatly improve the convergence rate in the training process, simultaneously selects data to train the neural network through the cross grouping method, can approach any continuous non-linear function by any precision; and with the increased scale of the wind farm and the non-linear increase of the wind speed changes, the adopted wind power forecasting method has very good robustness.

The invention discloses a false data injection attack design and defense method for a direct current microgrid. For a network communication message in the secondary control layer of the direct currentmicrogrid, malicious false data are injected. At the same time, considering the initial state estimation bias and measurement noise existing in the original detection process, a covert data injectionattack method is designed to cause disturbance to current / voltage without affecting the existing attack detection indicators. The covert data injection attack acquired through analysis affects the average voltage value of coupling points in the direct current microgrid. A proportional-integral dynamic average consistency algorithm is used to estimate the average voltage value of the coupling points in the direct current microgrid. The covert data injection attack is detected based on the average voltage deviation of the coupling points. Finally, a proportional-integral compensation mechanismis used to compensate the average voltage deviation of the coupling points. The secondary control input is adjusted. Voltage balance in the direct current microgrid is re-established.

The invention discloses field programmable gate array (FPGA) implementation equipment and an FPGA implementation method for self-adaptive clutter suppression of external radiation source radar. The method comprises the following steps of inputting four paths of channelized data to be processed into a finite impulse response (FIR) filter module through a first in first out (FIFO) memory inside a self-adaptive clutter suppression module of an FPGA chip; inputting two paths of data, channelized by an auxiliary antenna, into a step length calculation module through the FIFO memory, and calculating the step length through the step length calculation module; simultaneously inputting the output of the step length calculation module and two paths of output of the FIR filter module into a weight updating module; inputting an updated weight into the FIR filter module; and performing synchronous parallel output on clutter suppression results through in-phase / quadrature (I / Q) channels. The self-adaptive clutter suppression is controlled by five global clocks with the same rate and different phases. The self-adaptive clutter suppression of the external radiation source radar can be better realized. The problem of difficulty in meeting of a requirement on real-time performance caused by large conventional clutter suppression equipment amount is solved. The equipment and the method are high in processing efficiency, high in calculation speed and low in equipment complexity, and are used for implementing the self-adaptive clutter suppression of the external radiation source radar.

The invention provides an AUV (Autonomous Underwater Vehicle) track deviation estimation method based on multi-point terrain matching and positioning. The AUV track deviation estimation method comprises the following steps: (1) estimating a searching region of the terrain matching and positioning; (2) carrying out terrain matching and positioning under unknown tidal difference and measurement error conditions; (3) estimating measurement errors and estimating a positioning confidence interval; (4) estimating positioning errors; (5) estimating multi-point terrain matching and positioning initial track deviation; (6) carrying out track correlation and eliminating an incorrect positioning point; (7) carrying out secondary fitting of a track. The AUV track deviation estimation method provided by the invention has the advantages that characteristics that reckoning navigation is slowly changed in a time domain and an airspace and a terrain matching and positioning result has no diffusivity in the time domain and the airspace are combined, and a plurality of dense terrain matching and positioning points are obtained on a track line; track line of the reckoning navigation is fitted with multi-point positioning information so that the positioning precision and the positioning reliability are greatly increased.

A response of a channel may be estimated by correlating a received signal and a training sequence, by forming a matrix Γ based on a desired shape for the peaks of the correlation, by extracting a vector y from the received signal, and by estimating the channel response from a least-squares solution based on the matrix Γ, the vector y, and a matrix formed from the elements of the known training sequence.

Various configurations of improved wake-modifying devices for a power boat are provided herein. All such configurations provide a water channel and (optional) wedge for capturing and redirecting significantly more of the water flowing past the transom of the boat than conventional trim tabs consisting primarily of a flat plate. This increases the upward force applied to the deployed wake-modifying device and raises the stern of the boat on the non-surf side to list the boat to the surf side, producing a larger wake on the surf side. Some configurations provide a convergent water channel to produce longer, smoother wakes on the surf side of the boat. Other configurations provide a divergent water channel, which produces a comparatively shorter, taller wake on the surf side of the boat.

Disclosed is an active factor set membership proportional sub band self-adaption echo cancellation method. The method comprises the steps: A, dividing an adaptive echo cancellation sub band filter input vector X(n) formed by a discrete value of a far-end signal into a sub band signal X(n), B, performing N extraction on an input sub band signal X(n) and a near-end sub band signal d(n) to obtain a signal X(k) after the extraction and a d(k), C, obtaining a filtering value y(k) of the signal X(k) after the extraction via an adaptive echo cancellation sub band filter, D, subtracting a filtering value y(k) from the near-end sub band signal d(k), with an echo, after the extraction and then sending a different value to a far end, E, obtaining a sub band filter step length u(k) through calculation of an active factor f<l>(k) and a proportional matrix G(k) by utilizing a set membership filtering algorithm, and updating a weight coefficient vector W(k), and F, enabling k to meet an expression k=k+1, and repeating A-E steps until an end of a call. The method has the rapid convergence speed and the low steady-state error at one aspect, and has the rapid tracking capability at the other aspect; and the method has the good cancellation effect on an acoustic echo of a communication system.

The invention discloses a method for controlling the component and thermal treatment technological process of pre-hardening plastic die steel. The method includes the steps that firstly, an orthogonal test is designed according to structure characteristics and performance requirements of the pre-hardening plastic die steel, and thermal treatment technological parameters which have obvious influences on performance are found out through the test; then chemical components, the thermal treatment technological parameters and performance indexes are collected; an artificial neural network model is built, the model is trained, input of the model is the chemical components and the thermal treatment technological parameters, and output of the model is the performance indexes; the trained artificial neural network model is used for predicting steel performance and studying influence laws of input to output according to the chemical components and the thermal treatment technological parameters. According to the method, the complicated non-linear relationship between the chemical components, the thermal treatment technological parameters and mechanical properties is built, therefore, the component and thermal treatment technological process can be effectively controlled, the pre-hardening plastic die steel with excellent performance can be generated, and the service life of dies can be prolonged.

The invention discloses a method for optimizing a linear array antenna radiation pattern, which mainly solves the problems of low shaping and optimizing speed, low accuracy and low flexibility of the conventional antenna radiation pattern. The method comprises the following steps of: (1) setting parameters given by a system and a radiation pattern level value required by the system; (2) generating a level restraining fitness function, a level filling fitness function and a lower inclination angle fitness function according to the radiation pattern level value required by the system; (3) initializing excitation amplitude, an excitation phase and a unit distance parameter, and performing iterative optimization according to an optimizing formula until the requirement that the sum of values of the three fitness functions is equal to zero can be met; and (4) generating an antenna radiation pattern from the optimized excitation amplitude and / or the optimized excitation phase and / or the unit distance according to the theory of the array antenna radiation pattern so as to finish optimization on the radiation pattern. The method has advantages of high calculation speed, high accuracy and strong flexibility, and the optimized design, meeting the various system requirements, of the linear array antenna radiation pattern can be achieved.

The invention discloses a frequency domain blind separation method and a frequency domain blind separation system with low complexity. The method comprises the following steps: acquiring frequency domain separation signals collected by a microphone array; carrying out correlation analysis on the separation signals of the current frequency point and the pervious frequency point, thus completing thelocal sorting of the separation signal of each frequency point; carrying out correlation analysis on the separation signal of each frequency point after local sorting and the global central point, thus completing the global sorting of the separation signal of each frequency point; carrying out amplitude adjustment on the separation signals after global sorting; and carrying out Fourier transformon the adjusted separation signals, thus obtaining the separation signals of the time domain. With the method combining local sorting with global sorting provided by the invention, the accuracy and robustness of separation can be improved, meanwhile, the iteration times of the sorting process is reduced, and the computation complexity is relatively low.

An encryption method based on iteration FrFT and chaotic single-channel color images is characterized by comprising the steps that firstly the color images are placed in chaos and divided into three parts, then two parts are encrypted to obtain a middle result, and finally the amplitude of the middle result and the third part are encrypted to obtain a final cryptograph. The FrFT and the paired Logistic chaotic mapping are combined, so that the encrypted image has the nonlinearity and randomness, the safety and convergence rate of an encryption system are improved, the problem that an existing encryption method is low in convergence rate is solved, meanwhile, the complexity of the color image encryption system is reduced, the system is simple, and operation is convenient.

The invention discloses a satellite video dynamic target tracking method fusing a correlation filter and motion estimation, and relates to the field of spaceflight earth observation information processing and analysis. The satellite video dynamic target tracking method comprises the steps: obtaining a satellite video, and segmenting the satellite video into continuous single-frame images; selecting a to-be-tracked target from the single-frame image, constructing image features of the target, and inputting the image features into a related filter to obtain the position of the target; performingmotion estimation processing on the position of the target through a trajectory average algorithm and a Kalman filter, and when the Kalman filter is not stable, taking the processing result of the trajectory average algorithm as a motion estimation result; and when the Kalman filter is stable, taking a processing result of the Kalman filter as a motion estimation result. The satellite video dynamic target tracking method is suitable for satellite video ground target tracking, especially when the target is shielded, can still effectively track the target, and can complete multiple applicationssuch as target high-precision tracking, sensitive target real-time positioning and traffic flow monitoring.

The invention discloses a training method of a deep convolutional neural network for face recognition. The method comprises the following steps: 1) preparing a face image data set, dividing the face image data set into a training set and a verification set, and selecting the type, the structure, the hyper-parameter and the magnitude of a deep convolutional neural network model according to the scale and the complexity of the training set and the performance index of face recognition which should be achieved; 2) extracting features of face pictures input by the training set by using the model,and taking the features as input in the step 3); 3) constructing a loss layer, and iteratively calculating a loss value for the training; 4) comparing the loss value calculated in the step 3) with a preset threshold value, judging whether training is stopped or the gradient is calculated, and updating model parameters; and 5) verifying the model performance, and determining whether to stop training. According to the method, the human face features can be constrained by using a multivariate acting force from two aspects of an Euclidean space and an angle space during training, so that the deepconvolutional neural network model can learn the human face features with higher discrimination and robustness.

The invention discloses a VPN (Virtual Private Network) fast rerouting method and equipment and provider edge (PE) equipment. The VPN fast rerouting method comprises the following steps of: A, in a VPN of the PE equipment doubly attributive to CE (Customer Edge) equipment, receiving a label revoking message sent by network equipment on a main path by the PE equipment with a VPN rapid rerouting function; or, detecting that a BGP (Border Gateway Protocol) neighbor on the main path is broken off because the TCP (Transmission Control Protocol) fails in connection; B, determining whether an LSP (Label Switched Path) corresponding to the label revoking message or a BFD (Bidirectional Forwarding Detection) conversation associated to the BGP neighbor is in a communication state; and C, if so, carrying out main and spare path switching and canceling the BFD conversation. Due to adoption of the scheme, rapid convergence of VPN businesses can be ensured.

The compositional event based modeling of integrated applications (CINEMA) tool provides a way to extend a modeling environment using legacy event based applications, such as Graphical User Interface (GUI) APplications (GAPs). CINEMA allows modelers to extend the toolbox of the modeling environment by creating modeling elements that represent GUI objects of GAPs. CINEMA generates source code that allows an integrated system to control and manipulate the GUI objects of GAPs represented by the modeling elements used to create a model of the integrated system.

The invention relates to a high-precision numerical simulation calculation method for carrying out static characteristic analysis on concrete gravity dam based on h-p type finite element method, and belongs to the technical field of simulation. The high-precision numerical simulation calculation method for carrying out static characteristic analysis on concrete gravity dam based on h-p type finiteelement method comprises the following steps of building a calculation model of the hydraulic concrete gravity dam; calculating a displacement field and a stress field of the hydraulic concrete gravity dam model by a h-p type finite element method; judging whether the obtained displacement field and stress field meet the precision requirement or not, if the obtained displacement field and stressfield do not meet the precision requirement, encrypting the grid again, improving the order of the interpolation polynomial, and repeating the steps until a satisfactory result is obtained. The invention provides a novel high-precision finite element calculation and analysis method for a concrete gravity dam in order to solve the problems that an existing concrete gravity dam model based on traditional finite element static analysis is too long in calculation time, too large in error and low in precision.

Disclosed are a method and a system for multi-objective optimization of urban train operation. Firstly, speed limit information, slope information and curve radius information of a real train route are obtained, a section is segmented into non-equal sub-sections according to the above information about actual line characteristics, and then a longitudinal dynamics model of the train is constructed in combination with basic vehicle data of the train. Next, energy consumption of the train operation, section operation time, actual parking positions, and rates of acceleration change are calculated, so as to construct a multi-objective optimization model of train operation. Afterwards, a multi-objective differential evolution algorithm is used to solve the multi-objective optimization model, in order to obtain a Pareto optimal solution set of each operation district. Finally, an optimal solution is obtained which takes all objectives into comprehensive consideration, and an optimal train speed curve is generated.

The invention discloses an ultra-high density resistivity method parallel inversion method based on a weighted multi-strategy leapfrogging algorithm. The method comprises the steps: adding a weighting coefficient to the local search of the leapfrogging algorithm so as to improve the algorithm's capability of jumping out of a local extreme value; introducing the least square search strategy to the local search of the leapfrogging algorithm so as to achieve the multi-strategy local search; employing the directed search characteristics of the least square method to improve the local convergence speed of the algorithm; employing the CUDA technology for the parallel calculation of the adaptability of each fog and the local search process of each meme group, thereby improving the total calculation efficiency of the algorithm. The method can achieve the optimal inversion according to the features of an ultra-high density resistivity method, can remarkably improve the inversion quality and efficiency of the ultra-high density resistivity method, and is very high in practicality.

The invention belongs to the technical field of seismic and logging joint inversion, and discloses a CNN well-seismic joint inversion method and system, a storage medium, equipment and application. The method comprises the steps: searching an inversion mapping operator f1: y-> x from seismic data y to logging data x, i.e. X = f1 (y), with the seismic data y as the input and the logging data x as the output; reconstructing a logging curve in the forward direction; and reversely updating the weight and the bias. A four-layer network structure containing two hidden layers comprises an input layer, a first convolution layer, a second convolution layer and an output layer, and the two hidden layers are convolution layers. Some virtual logging curves are interpolated by using a Kriging interpolation technology, and virtual logging data and real logging data are used as training data for convolutional neural network learning. Under the condition that a real well is not additionally added, the number of learning samples can be increased through virtual well logging, an inversion mapping operator is searched for in a wider range, and over-fitting of local training data is prevented.

The invention discloses a multi-modulus blind equalization algorithm (MMA) optimized by a Memetic algorithm (MA). The concepts of individual evolution, social behaviors among individuals and the like are introduced into a blind equalization technology; the reciprocal of an MMA cost function is defined as a fitness function of the MA; individual optimal vectors are searched in a global scope by means of the population optimization mechanism and local area deep-searching capability of the MA and are taken as the initial optimization weight vectors of the MMA. Iteration is performed through the MMA to obtain an optimal weight vector of the MMA. Compared with a constant modulus blind equalization algorithm (CMA), the MMA and a genetic algorithm-based multi-modulus blind equalization algorithm (GA-MMA), the algorithm disclosed by the invention has the advantages of high convergence speed during equalization of high-order multi-modulus signals, smallest steady-state errors and clearest output signal constellation map.