89results about How to "Avoid local convergence" patented technology

The invention relates to a BP neural network image segmentation method and device based on an adaptive genetic algorithm, and the method comprises the following steps: 1), analyzing a to-be-segmented image, and generating a training sample of a neural network; 2), setting the parameters of the neural network and population parameters, and carrying out the chromosome coding; 3), inputting the training sample for the training of the network, optimizing the weight value and threshold value of the network through employing a new adaptive genetic algorithm, adapting to the crossing and mutation operations, and introducing an adjustment coefficient; 4), inputting the to-be-segmented image, carrying out classifying of the trained neural network, and achieving the image segmentation. The device comprises a training sample generation module, a neural network structure determining module, a network training module, and an image segmentation module. The method introduces the adjustment coefficient which is related with the evolution generations, solves a problem that the individual evolution stagnates at the initial stage of population evolution, and also solves a problem of local convergence caused when the individual adaption degrees are close, thereby obtaining the neural network which can maximize representation of the image features, and achieving the more precise image segmentation.

The invention discloses a method for regional water-source optimized configuration based on an MAEPSO algorithm. The method comprises the steps that basis information data of a water resource system is acquired at first; secondly, a water-resource multi-target configuration model is established; then the algorithm based on MAEPSO is executed, and a Pareto optimal solution set of multi-target configuration of the regional water resource system is solved; and finally, a final solution is selected from the optimal solution set according to certain rules. According to the invention, global optimization is achieved; computation efficiency is increased; and a requirement for selecting a multi-target optimal configuration scheme of the water resource system can be satisfied.

The invention belongs to the real time positioning technical field of a distributed multi-point positioning monitoring system, and especially relates to a positive semi-definite relaxation time difference positioning method for the distributed multi-point positioning monitoring system; the method comprises the following steps: firstly building a time difference positioning equation; carrying out maximum likelihood estimation for a signal source position, i.e., a to-be-positioned object position; importing an auxiliary vector; converting a distance difference positioning equation into a constraint least square problem; carrying out weight least square solving for the auxiliary vector; using the weight least square solved auxiliary vector to primarily estimate relaxation equality constraint, and building a novel cost function; using a protruding positive semi-definite program to optimize and solve the auxiliary vector value and a variable value transpositioned by the auxiliary vector; using a characteristic constant to decompose the obtained auxiliary vector value; finally obtaining the signal source position information according to a relation between the solved auxiliary vector value and the signal source position. The method can prevent the local convergence and solving divergence problems caused by a conventional iteration algorithm, thus improving the positioning precision.

The invention relates to a method for movie recommendation on the basis of an orthogonal and cluster pruning based improved multi-objective genetic algorithm. An improved algorithm OTNSGA-II is provided aiming at defects in distributivity and convergence of NSGA-II (non-dominated sorting genetic algorithm-II) and can be used for solving various multi-objective function optimization problems. By design of fault multi-objective orthogonal experiment initialization population, distributive deficiency caused by individual nonuniformity is avoided; by application of self-adaptive cluster pruning strategies, a population evolution process is maintained, and inferior individuals in an appropriate quantity are removed to keep convergence and distributivity of the population. By combination with information mining of user behaviors and movie properties, the algorithm is applied to solving of a practical problem of personalized movie recommendation, universality and effectiveness of the algorithm are explained by test comparison with existing algorithms, better recommendation results are obtained, recommendation accuracy rate, recall rate and coverage rate are increased, rich recommendation scheme combinations are provided, and interest points of users can be mined beneficially to provide more reliable recommendation services.

The invention discloses a vehicle Inerter hanger-bracket parameter optimization design method. The vehicle Inerter hanger-bracket parameter optimization design method includes the following steps that 1, hanger-bracket models are set; 2, basic parameters and road surface input of vehicle types are determined; 3, design variables and the ranges of the design variables are determined; 4, the selected variables are subjected to normalization processing; 5, evaluation index systems of hanger-bracket systems are set; 6, hanger-bracket models of different parameters are simulated; 7, optimal solutions are recorded; 8, domains of the optimal solutions are determined, whether the solution domains reach the designed targets or not is judged, and if the solution domains reach the designed targets, optimization is completed; 9, the ranges of the variables of parameters are changed, or the evaluation index systems of hanger-bracket systems are modified, and optimization design is newly carried out. By means of the vehicle Inerter hanger-bracket parameter optimization design method, the parameter optimal solution domains of hanger brackets can be rapidly and accurately obtained, the problem that multiple existing optimization algorithms are subjected to local convergence can be solved, and the optimal parameters can be selected through the global optimum solution domains. The vehicle Inerter hanger-bracket parameter optimization design method provides a novel method and idea for design and performance researching of the hanger brackets.

The invention discloses a path planning method for multiple unmanned aerial vehicles to arrive at a designated place simultaneously in a three-dimensional environment, and the method comprises the steps of building a population, wherein a chromosomes represents a comprehensive flight path, and the comprehensive flight path is formed by the series connection of the flight paths of all unmanned aerial vehicles; setting the fitness of the genetic algorithm, and comprehensively evaluating the set violation degree W of the maneuverability constraint of the unmanned aerial vehicle by the total tracklength L of all unmanned aerial vehicles, the difference degree T of arrival time of different unmanned aerial vehicles and the track; and optimizing the chromosome by adopting the genetic algorithm,and obtaining an optimal chromosome after the iteration of the genetic algorithm, and decomposing the optimal chromosome to obtain the track of each unmanned aerial vehicle. By using the method, thetechnical purposes of gathering multiple unmanned aerial vehicles and reaching the task points at the same time can be achieved, and meanwhile time errors are reduced as much as possible.

The invention provides a 3D gesture recognition method and system based on a deep convolutional neural network. The method comprises the steps that firstly, using a first deep convolutional neural network or conducting pre-segmentation on a large number of color images containing hands, and extracting hand action parts; secondly, performing hand joint node detection on the extracted hand by usinga second deep convolutional neural network; Performing gesture 3D reconstruction on the detected joint nodes by using a double-flow deep convolutional network; And finally, constructing a softmax network comprising three full connection layers to identify the 3D reconstructed gesture. The technical scheme provided by the invention has the beneficial effects that the gesture recognition precision can be effectively improved. From the perspective of application range, an object of the method is an RGB image collected by a monocular camera, required equipment is simple and cheap, and the application scene is wider.

The invention discloses a method for calibrating the installation position error of a laser distance-measuring sensor based on the Levenberg-Marquardt algorithm and belongs to the technical field of robots. According to the method, a device is composed of an industrial robot, a plane test plate, a laser distance-measuring sensor, a sensor installation bracket and a laser tracking instrument. The plane test plate is arranged at the tail end of the industrial robot. The laser distance-measuring sensor is arranged on the sensor installation bracket. The sensor installation bracket is held stationary. According to the method, a mathematical model is modeled for the measurement distance of the laser distance-measuring sensor. Based on the above model, an error model is obtained. After that, theLevenberg-Marquardt algorithm is adopted for the parameter identification of the installation position error of the laser distance-measuring sensor, so that the measurement accuracy of the distance-measuring sensor is improved. By adopting the method, the installation position error of the distance-measuring sensor can be well calibrated. The local convergence problem of the ordinary least squaremethod can be avoided.

The invention provides a blind equalization method and a blind equalization system. The method comprises the following steps of performing input transformation on input signals and generating input vectors after transformation; acquiring output signals through a feedforward network by means of the input vectors after transformation and recursive updating parameters; when desired signals are uncertain, acquiring input decision information through input signals and acquiring output decision information through output signals; designing blind equalization algorithm by using both the input decision information and the output decision information to acquire a feedback error; and updating the recursive updating parameters through the feedback process through the feedback error and the input vectors after transformation. By additionally using the input decision information to assist the feedback process, the method can effectively prevent common local convergence problems generated during blind equalization due to the influence of the feedforward network structure and the recursive updating parameters. Robust convergence of blind equalization results is guaranteed. Meanwhile, the blind equalization performance is also ensured.

The invention proposes a positioning method of optimal joint time synchronization and positioning under a TDOA condition for solving the technical problem of low positioning accuracy in the existing mobile terminal positioning technology. The method comprises the following steps: setting system parameters; establishing an observation model of joint time synchronization and positioning arrival timedifference TDOA; establishing a double-constraint mobile terminal position estimation target function based on the observation model of joint time synchronization and positioning arrival time difference TDOA; and solving the double-constraint mobile terminal position estimation target function to obtain a globally optimal solution of a to-be-estimated vector y, and reading the globally optimal solution of a mobile terminal position vector p. By adoption of the positioning method, the transmission of positioning errors is avoided, the globally optimal solution of the positioning non-convex problem is obtained, the positioning accuracy is improved, and thus the positioning method can be applied to mobile terminal positioning estimation depending on a plurality of cellular base stations in asight distance scene.

The invention provides a user preference-based dynamic computing migration method and device for a smart city, and the method comprises the steps of initializing a set of input tasks; stipulating an algorithm stop standard, a population maximum iteration frequency, the number of neighborhood vector sets of each particle and a population initial migration strategy, and defining a group of weight vector sets required to be used in the algorithm; then, on the basis of an MOEA/D algorithm, continuously updating a migration strategy of the task by taking optimization of total energy consumption andtotal time delay of the mobile equipment task of the user side from generation to completion as a target; meanwhile, in order to meet the requirements of the user, adding an elitist strategy which can be changed in a directed mode according to the requirements and preferences of the user; according to the invention, an elitist strategy is adopted, energy consumption and time delay generated by task processing are comprehensively considered while user preferences are met, an appropriate calculation migration strategy is formulated for user tasks in an MEC environment, and the purpose of multi-objective optimization is achieved.

The invention provides an image super-resolution reconstruction method based on maximum posteriori probability and non-local low-rank prior. The method comprises the steps that a continuous image sequence is adopted as data input, the similarity between a single image and a continuous image is used as priori knowledge, block matching is conducted on similar blocks through an image block local grouping mode, and the spatial structure relation of the image pixel level is mined; modeling is performed with a maximum posterior probability framework, Gaussian distribution and Gibbs distribution areused for fitting model parameters, and the generalization ability of the model is improved; noise interference is suppressed in a low-rank truncation mode; a non-local low-rank constraint regularization image reconstruction process is adopted, and local information in a single image and local information between continuous images are utilized to improve the quality of a target image. Parameters inthe model are alternately optimized in each iteration, the robustness of the model is improved, and local convergence is avoided. Finally, weighted averaging is performed on the reconstructed image blocks to obtain a target high-resolution image.

The invention relates to a robust multiuser detector design method, which solves the technical problem of high error rate of the traditional robust multiuser detector in an impact noise channel environment. The robust multiuser detector design method comprises the steps of: initializing algorithm parameters; using an opposition-based learning method to initialize a parent population, and determining three wolves in the parent population; updating the parent population by adopting an improved gray wolf algorithm position updating equation, and sorting population individuals according to fitnessvalues from large to small; generating offspring crossover mutants by utilizing the parent population, performing position information differential operation on an evolutionary direction of the offspring crossover mutants and successful crossover mutation probabilistic information when the fitness values of the offspring crossover mutants are superior to that of the parent population, acquiring new evolutionary direction information and saving the new evolutionary direction information, and updating positions of the three wolves. With the adoption of the robust multiuser detector design method adopting a Huber theory and utilizing a residual non-rapid-increasing function to design a multi-user detector in an impact noise channel, the mentioned problem is effectively solved, and the robustmultiuser detector design method can be used in multi-user detector design.

The invention provides a positioning method with optimal joint of time synchronization and positioning under a TOA condition, used for solving the technical problem of low positioning precision and efficiency of the mobile terminal positioning technology in the prior art. The positioning method with the optimal joint of the time synchronization and the positioning under the TOA condition comprisesthe following steps: setting system parameters; establishing an observation model jointing the time synchronization and the positioning time of advent TOA; establishing a constrained target functionof the positioning estimation of a mobile terminal according to the observation model jointing the time synchronization and the positioning time of advent TOA; and solving the constrained target function of the positioning estimation of the mobile terminal by means of the generalized trust region method to obtain the global optimal solution of a combined vector y and read out the global optimal solution of a position vector p of the mobile terminal. The invention avoids the transfer of the positioning error, obtains the global optimal solution of positioning non-convex problem, and improves the positioning accuracy; and the calculation efficiency of the positioning estimation is improved by a simple iterative operation. The positioning method with the optimal joint of the time synchronization and the positioning under the TOA condition can be used for the positioning estimation of the mobile terminal depending on a plurality of cellular base stations in a stadia scene.

The invention discloses a particle swarm exchange long-time accumulation implementation method based on a multi-core DSP. During long time accumulation, a target cross-distance unit and a Doppler unitcan cause reduction of accumulation detection performance. The long-time accumulation module firstly segments target echo data to enable a target in each segment not to cross a distance unit and a Doppler unit, then uses a particle swarm exchange PSO algorithm to search parameters such as distance, speed and acceleration of particles, and iteratively searches global optimal particles to serve asthe distance, speed and acceleration of the target. The method is particularly suitable for increasing the radar action distance of stealth and other weak echo targets, good target detection performance can be obtained, and the radar action distance of stealth and other weak echo targets is equivalently increased.

The present invention discloses an automatic guide rail carrying device and robot cooperative carrying method and system. The method comprises the following steps of: the step 1: arranging a ground guide rail arranged between workbenches from a transportation starting point and a transportation end point and guide rails arranged on the workbenches, and arranging a landmark on the ground guide rail; the step 2: grabbing an object from an object-fetching assigned position by a desktop robot located on an object-fetching workbench; the step 3: allowing a mobile robot to move through the ground guide rail and grab an object to the other workbench; the step 4: after a desktop robot located on an object-put workbench grabs the object, moving the desktop robot to an object-put assigned position of the object-put workbench; and the step 5: combining an extreme learning machine and a wavelet neural network to establish an electric quantity prediction model to perform decision of the next motionof the mobile robot. The automatic guide rail carrying device and robot cooperative carrying method and system complete timing and fixed-point transportation of industrial laboratory objects throughcooperation of the desktop robot and the mobile robot so as to achieve round-the-clock transportation of the industrial laboratory.

The invention relates to a flood prevention material rescue distribution method based on a non-dominated artificial bee colony. The method comprises: establishing a flood prevention material distribution optimization model based on warehouse material storage capacity constraints, material demand constraints and flood prevention material arrival time constraints; initializing a food source by adopting three methods of large-scale storage warehouse priority, disaster-affected demand priority and random allocation, and proposing a demand fitness value, a material satisfaction rate variance fitness value and a transmission time fitness value of the food source; in the employed bee stage, using a high-dimensional matrix local search method, and the space complexity being reduced through population classification and exclusion elimination; in the bee following stage, updating food sources through cross operation, and expanding the search range of the population; and in the bee reconnaissancestage, selecting a tail food source, a repeated food source and a food source reaching the evolution upper limit frequency to reconstruct. The distribution of flood prevention materials can be optimized, the distribution time of rescue materials is saved, the material distribution of each affected area is balanced, and the overall distribution efficiency of the flood prevention materials is improved.

The invention discloses a discretized region scanning subarray-level sparse optimization method and system, and belongs to the technical field of array antenna design, and the method comprises the following steps: S1, discretizing an initial population; s2, carrying out region scanning processing; and s3, optimization processing. When the method and the system are applied to sub-array-level sparseoptimization of a large-scale circular aperture antenna, under the condition that the sub-array spacing meets the spacing requirement, the optimization time is greatly reduced, the sidelobe level isreduced, a new way is provided for rapidly solving effective position arrangement of array elements under the specific array element spacing limitation condition, the diversity of a solution set in the optimization process is improved, the probability of finding the optimal solution is improved, and local convergence of the optimization algorithm is effectively avoided; and when the designed arrayantenna is applied to a large-scale radar system, the radar has the advantages of long operating distance, high resolution, low cost, light weight, high engineering realizability and the like, and isworthy of popularization and application.

The invention discloses an engineering parameter optimization method, and the method comprises the steps: an objective function corresponding to a preset engineering problem is constructed at first; N populations are generated randomly; S iterative updates are performed on populations separately; a global optimal particle is screened out from updated N populations, and then optimal engineering parameters of the preset engineering problem are determined; wherein an iterative update process can be performed on any current population, the process comprises the steps: when a first probability condition is met, a dimension value of a particle is generated in a population range corresponding to the current population, when a second probability condition is met, a dimension value of the particle is generated outside the population range corresponding to the current population, and an iteratively updated population of this time can be obtained correspondingly. A problem of occurring local convergence in the optimal process is avoided, so that a globally optimal solution can be obtained finally. In addition, the invention also discloses an engineering parameter optimization system.