248 results about "Random search" patented technology

A method is described for improving the prediction accuracy and generalization performance of artificial neural network models in presence of input-output example data containing instrumental noise and/or measurement errors, the presence of noise and/or errors in the input-output example data used for training the network models create difficulties in learning accurately the nonlinear relationships existing between the inputs and the outputs, to effectively learn the noisy relationships, the methodology envisages creation of a large-sized noise-superimposed sample input-output dataset using computer simulations, here, a specific amount of Gaussian noise is added to each input/output variable in the example set and the enlarged sample data set created thereby is used as the training set for constructing the artificial neural network model, the amount of noise to be added is specific to an input/output variable and its optimal value is determined using a stochastic search and optimization technique, namely, genetic algorithms, the network trained on the noise-superimposed enlarged training set shows significant improvements in its prediction accuracy and generalization performance, the invented methodology is illustrated by its successful application to the example data comprising instrumental errors and/or measurement noise from an industrial polymerization reactor and a continuous stirred tank reactor (CSTR).

The invention discloses an improved RRT<*> obstacle avoidance motion planning method based on a multi-degree-of-freedom mechanical arm, and belongs to the field of mechanical arm motion planning. A six-degree-of-freedom mechanical arm model with seven connecting rods and six rotary joints is built; parameters in a to-be-searched space are determined; if the distance is shorter than the distance of a path with lowest cost, the distances between a near node in a set to an initial point and the distance between the node to a random point are temporarily determined as the minimum path; a newly generated sigma is subjected to collision detection, and the node and the path are added if the newly generated path does not collide an obstacle interval; the steps are repeated until the optimal path is found; and the generated path is added into a path planning device. Compared with the prior art, the method has the following advantages that the random search characteristic is changed in a mode of adding normal distribution, the algorithm convergence rate can be increased through the heuristic search, the RRT<*> algorithm has the evolutionary optimization path, and a large number of calculations is not needed; and after Gaussian distribution of an inspiration point near a target point is added, the convergence rate is increased, and the search time is shortened.

The invention relates to the mobile communication technology field, in particular to a network frequency optimization method based on measurement reports. The method includes the following steps: firstly extracting a measurement report and establishing an interference matrix; calculating the adaptability of each frequency distribution proposal in a frequency distribution proposal group according to the interference matrix; establishing the proportional distribution according to the size of the adaptability and implementing random search selection; generating a new group of frequency distribution proposals through random frequency point modification or frequency point interconversion; recalculating the adaptability of each frequency distribution proposal in the new frequency distribution proposal group; continuously repeating the above steps until the adaptability of a new frequency distribution proposal meets the requirement; finally the frequency distribution proposal with largest adaptability in the frequency distribution proposal group becoming the network frequency distribution proposal after the optimization of the present network. The method provided by the invention takes the measurement report of the present network as the basis of frequency optimization and fully considers the real district interference condition of the present network, thus achieving the minimum interference of the network after frequency optimization.

A computer-implemented method for solving a constraint satisfaction problem (CSP), which is defined by variables and constraints applicable to the variables, and which has states corresponding to respective sets of values of the variables. The method includes assigning cost functions to the constraints so that the states have respective costs determined by application of the cost functions to the respective sets of values of the variables, the respective costs defining a problem topography of the CSP having global extrema corresponding to solutions of the CSP. The constraints of the CSP are reformulated so as to perform at least one of increasing a density of the solutions in the problem topography and smoothing a gradient of the problem topography. One or more of the solutions of the CSP are found by applying a stochastic CSP solver to the reformulated constraints.

Measurements at multiple distinct polarization measurement states are taken to define the polarization state of an input, for example to calculate a Stokes vector. High accuracy and/or capability of frequent recalibration are needed, due to the sensitivity of measurement to retardation of the input signal. A multiple measurement technique takes a set of spatially and/or temporally distinct intensity measurements through distinct waveplates and polarizers. These can be optimized as to orientation and retardation using initial choices and also using tunable elements, especially controllable birefringence elements. A device matrix defines the response of the device at each of the measurement states. The matrix can be corrected using an iterative technique to revise the device matrix, potentially by automated recalibration. Two input signals (or preferably the same signal before and after a polarization transform) that are known to have a common polarization attribute or other attribute relationship are measured and the common attribute and/or attribute relationship is derived for each and compared. The device matrix is revised, for example by iterative correction or by random search of candidates to improve the accuracy of the device matrix. Optional tunable spectral and temporal discrimination provide additional functions.

The invention relates to the technical field of hydroelectric generating set fault diagnosis, in particular to a hydroelectric generating set fault diagnosis method and system based on a DdAE (Difference Differential Algebraic Equations) deep learning model. The method and the system are established on the basis of the analysis of the original vibration data of the hydroelectric generating set, adeep learning characteristic extraction method based on a multilayer neural network model is adopted, a complex manual processing and feature extraction process is not required, an ASFA (Aquatic Sciences and Fisheries Abstracts) method based on random search is adopted to carry out the structural parameter adjustment and optimization of the DdAE to achieve a purpose of strategy optimization. A deep denoising automatic encoder model is used for realizing the distributed expression of original data, and reconstruction data subjected to feature extraction is input into a Softmax regression modelto judge the work state and the fault type of the hydroelectric generating set. The analysis of a network experiment result indicates that the method can be effectively applied to the hydroelectric generating set fault diagnosis.

The invention discloses a sensor target assignment method for a multi-objective optimization differential evolution algorithm. The method includes the steps that objective importance degree calculation is carried out according to objective information, a sensor target assignment constraint multi-objective optimization function is built, distribution scheme codes and initial population chromosomes are generated, offspring scheme populations are generated through the differential evolution algorithm, population combination and screening are carried out, and a distribution scheme Pareto front-end solution set is obtained. The method is combined with the differential evolution algorithm, is easy to use in terms of population difference heuristic random search, is good in robustness and has the advantages of being high in global search ability and the like. A Pareto set multi-objective optimization assignment strategy is provided. A sensor utilization rate function is added on the basis of a sensor target monitoring efficiency function, an assignment problem is converted into a multi-objective optimization problem, sensor resources can be saved as much as possible on the condition that monitoring precision requirements are met, and reasonable and effective assignment of the sensor resources is achieved.

The invention provides a small area-level ultra-short-term load prediction and visualization method based on a Long Short-Term Memory (LSTM) network, the method comprising: step 1: an input and outputvariable of the model is determined; step 2: The input and output data sets are separately preprocessed; step 3: the depth LSTM load prediction model is constructed and the random search method is used to find the appropriate hyperparameters until the test set prediction error is minimized. step 4: The t-SNE visualization technology is used to visually characterize the network hidden layer vector, and the correlation coefficient heat map is formed according to the hidden layer vector to perform correlation quantitative analysis, thereby reflecting the network's ability to extract feature datafrom the input data. The method aims to utilize the feature extraction capabilities of the deep learning model and the LSTM temporal correlation learning capabilities to achieve higher prediction accuracy than the machine learning model.

It is described a method for determining values of geometry parameters of a cone beam computer tomography apparatus, the method comprising: (a) obtaining x-ray projection data captured by the detector from at least three calibration objects arranged at mutually different positions, the x-ray projection data comprising for each calibration object plural projections at different rotation angles; (b) determining for each calibration object a respective ellipse representation from the respective plural projections; (c) performing a random search across candidate values of the geometry parameters for determining the values of the geometry parameters, wherein a cost function depending on the ellipse representations and the geometry parameters is optimized.

The invention discloses a method for cloud manufacturing resource optimization configuration based on an improved whale algorithm, and the method comprises the steps: building a problem model, and defining a fitness function; setting improved whale algorithm parameters, and generating an initial population; Calculating fitness values of all individuals in the population, obtaining a current optimal resource allocation scheme and converting the current optimal resource allocation scheme into whale individual position vectors; Introducing a parameter p, and judging whether p is less than or equal to 0.5; If not, performing spiral motion iteration updating to complete population updating; If yes, whether the value A (1) of the coefficient vector of the improved whale algorithm is met or not is judged; If yes, performing shrinkage encircling iteration updating; If not, performing random search predation iteration updating; Obtaining a current optimal resource configuration scheme; Adding 1to the number of iterations, and judging whether the current number of iterations is smaller than the maximum number of iterations; If yes, repeating the operation; And if not, outputting the currentoptimal resource configuration scheme. The whale algorithm is improved, so that the algorithm convergence speed is higher, the optimal solution is easier to achieve, and a new method is provided forsolving the problem of resource allocation.

A chromosome is decoded by a decoding unit, and converted into parameters of a problem model calculation unit. In the problem model calculation unit, a controller executes a local search method unit, a GA search unit, or a stochastic search unit while suitably selecting any of them, so that a solution is generated. If a constraint violation is detected by a constraint violation determination unit during a solution generation process, an added part (a part which causes a constraint violation) is removed from a current solution by the constraint violation processing unit, and the solution generation process is continued.

The invention discloses a high-frequency high-voltage transformer design optimization method based on a genetic algorithm. On the basis of a minimum loss formula of a transformer and the insulation dimension and the iron core shape of the transformer, a mathematical model is established, a genetic algorithm is adopted for optimizing the transformer by taking the number of turns of primary sides and the layer number of secondary sides as optimization variables and taking efficiency as an optimization target, so that the efficiency of the transformer is maximized, the loss of the transformer is minimized, temperature rise is minimized under an equal condition, the leakage inductance and the distribution capacitance of the transformer are fully utilized to participate in the work of a power system to form an LCC (Life Cycle Costs) resonance circuit, and the loss of the transformer is reduced so as to lower the temperature rise. Compared with the prior art, the invention introduces the genetic algorithm, the genetic algorithm exhibits a problem non-relevant quick and random search capability, searches by starting from a group and has the characteristic of concurrency, and a plurality of individuals can be simultaneously compared to greatly quicken optimal solution search speed.

The invention discloses an ant colony algorithm and an intelligent communication service deployment method for an electric power communication network. The method comprises the following steps: step 1, importing a network topology model of the electric power communication network; step 2, inputting an initial node and a target node; step 3, performing network topology analysis by using the ant colony algorithm according to any one of claim 1 or 2, and outputting the optimal solution path; step 4, carrying out feasibility analysis, and if the requirement is satisfied, performing intelligent deploying of communication services of the electric power communication network according to the path obtained through the step 3; and otherwise, returning to the step 3 to carry out network topology analysis again. Defects of random search and easily resulting local optimum in existing ant colony algorithms can be effectively avoided; the routing deploying capabilities and operational support capabilities of the electric power communication network are improved; and the cost of routing analysis and deployment are saved.

The invention discloses a project constraint parameter optimizing method based on an improved artificial bee colony algorithm. According to the method, the problem of the project constrained parameter optimization is described by the adoption of an objective function and an equality/non-equality constraint; an artificial bee colony is initialized according to the value range of parameters; partial parameters in a parameter vector is selected according to the probability M to serve as the adjusted object, and step size in search is adjusted in a self adaptive mode, so that a guide bee can search nectar sources randomly in an intra area; according to the corresponding cost function value fi of the nectar sources, the fitness function value fiti is acquired through fi, the probability Pi of follow bees being transferred to the nectar sources is further acquired, and whether position updating is conducted or not is judged; the current optimal solution is recorded in every iterative search process, and the optimized estimated value of the parameters is acquired through the finite iterative search. The step size in search changes in a self adaptive mode with the times of search, on the premise that search accuracy is not affected, search time is reduced effectively, and search efficiency is improved.

The invention provides a fitness random search behavior-based multi-threshold image segmentation method, which comprises the following steps of: establishing a multi-threshold segmentation fitness function and calculating the optimal segmentation threshold; according to the optimal segmentation threshold, establishing and initializing the first generation of particle swarms; according to the multi-threshold segmentation fitness function, calculating a fitness value of each particle, and calculating the individual optimal position of each particle and the global optimal position of all particles; updating a speed and a position vector of each particle, the individual optimal position of each particle and the global optimal position of all the particles by utilizing a particle swarm iterative formula; and repeatedly executing the steps until the condition that the number of iterations of the particle swarm iterative formula u=Umax is met. The method has the advantages of good segmentation stability, high speed and high segmentation accuracy, greatly improves the segmentation speed and accuracy, and makes the subsequent work of image processing possible.

The invention discloses an image automatic annotation method based on digraph unequal probability random search, which comprises the following steps: inputting an image to be annotated and an annotated image set; extracting a plurality of feature vectors of the image to be annotated; selecting an adjacent image set; constructing a digraph model of the image to be annotated; calculating a word similarity matrix Se between tags and a symbiotic relationship matrix Co between tags; fusing the word similarity matrix Se between tags and the symbiotic relationship matrix Co between tags, so as to obtain a tag similarity matrix TT; and carrying out unequal probability random search on each candidate tag in a candidate tag set in the digraph model, so as to calculate the score, and obtaining a plurality of high-score candidate tags to be used as the label results. The invention also discloses an image automatic annotation device based on digraph unequal probability random search. In the invention, the dependency relation between images and similarity relation between tags are utilized fully and reasonably, thus the image automatic annotation can be effectively carried out, and the annotation effect is better.

A system for optimizing shelf space placement for a product receives decision variables and constraints, and executes a Randomized Search (“RS”) using the decision variables and constraints until an RS solution is below a pre-determined improvement threshold. The system then solves a Mixed-Integer Linear Program (“MILP”) problem using the decision variables and constraints, and using the RS solution as a starting point, to generate a MILP solution. The system repeats the RS executing and MILP solving as long as the MILP solution is not within a predetermined accuracy or does not exceed a predetermined time duration. The system then, based on the final MILP solution, outputs a shelf position and a number of facings for the product.

The invention provides a bearing fault classification method and system based on a deep learning network, and the method comprises the steps: setting a sampling frequency, and collecting the vibrationsignal data of a bearing under different working conditions; segmenting the obtained vibration signal data to construct a plurality of samples; decomposing the vibration signal data of each sample toobtain a plurality of modal components so as to realize effective component separation; constructing a deep network with a residual error unit, and determining an appropriate network depth by using arandom search method; inputting the training set into a deep residual network for iterative training and obtaining a classification model; and inputting the test set into the classification model toobtain a fault classification result. According to the classification method provided by the invention, variational mode decomposition and a deep residual network are combined; the problems that noiseinterference exists in input data, cross aliasing exists in effective components, network deepening causes identification gradient disappearance, and performance degradation causes poor classification effect are solved, fault feature extraction not affected by rotating speed changes is achieved, and the fault classification accuracy is improved.

The invention relates to a low-quality finger vein image enhancement method, which comprises four processing stages of: image preprocessing, image segmentation, parameter correction and skeleton line tracking, wherein the image preprocessing stage comprises four steps of: target positioning and cutting, target size normalization, image filtering and image rotation, so that influence of difference of an acquisition device on adaptive capacity of the image enhancement method can be weakened; the image segmentation stage comprises three steps of: adaptive threshold image segmentation, mathematical morphologic filtering and image thinning, so that a finger vein binary image and a skeleton image are acquired to serve subsequent processing; and in the parameter correction stage, average width of finger vein is calculated according to the finger vein binary image and the skeleton image, and a width parameter and a distance parameter in the method are corrected according to the width. The skeleton line tracking stage comprises steps of acquiring a tracking starting point set, initializing a track space, selecting a tracking starting point, generating a random search direction, performing dark line tracking and the like. By adopting the method, a low-quality finger vein image can be quickly and effectively enhanced, and high adaptive capacity is guaranteed.

The invention relates to the rehabilitation training field and aims to optimize the quantifying factor and scale factor of a fuzzy controller and the fuzzy control rules, then control the current mode of an FES system accurately, stably and instantly and effectively improve the accuracy and stability of the FES system. The technical scheme adopted by the invention is as follows: the walking aid electrostimulation fine control method based on genetic-ant colony fusion fuzzy controller comprises the following steps: firstly, converting the selection of fuzzy control decision variable to the combinational optimization problem adapting to the genetic-ant colony algorithm, coding the decision variable, randomly generating a chromosome composed of n-numbered individuals; secondly, using the genetic algorithm to generate the initial pheromone distribution of the ant algorithm, utilizing the ant colony algorithm to randomly search and optimize the membership function, quantifying factor and scale factor of the fuzzy controller; and performing repeated self-learning and self-regulating according to the system output, and finally using the processes in the FES system. The invention is mainly used for rehabilitation training.

The invention discloses a performance reliability simulation method based on improved self-adaption selective sampling, successively comprising four steps of: 1. extracting an initial failed sample: taking the original mean value of parameters as a sampling center, taking 1-3 times of the original variance as a sampling covariance to carry out random search, and finding the initial failed sample by iteration; 2. extracting a batch of failed samples: randomly sampling by taking the failed sample searched in the step 1 as the initial sampling center of the step 2, adjusting the sampling center in the next circulation, and circulating until the failed sample of appointed number is extracted; 3. carrying out self-adaption selective sampling: calculating the initial sampling center and the initial sampling covariance in the step 3 according to the failed sample extracted in the step 2 , and carrying out cycling sampling stimulation; and 4. calculating: after one-time sampling, calculating the sampling center and the sampling covariance again, stopping stimulation when the stimulation result of failure probability is almost stable, and finishing the reliability stimulation of self-adaption selective sampling.

The invention relates to a multi-target point path planning method based on a fast random search tree, and belongs to the field of robot path planning. The method of the invention uses a two-layer tree structure. The bottom layer tree is composed of multiple trees extending from multiple target points, wherein each tree has a weight determined by the surrounding environment, and each tree searchesthe free space by using a fast random search tree algorithm. An effective collision-free path is generated when two trees are close enough, and the effective path and nodes constituting the effectivepath will be transferred to the top layer tree. The top layer tree carries out re-planning on the paths and the nodes by using an improved minimum spanning tree algorithm and finally acquires the shortest path capable of traversing multiple target points. The multi-target point path planning method provided by the invention can be effectively implemented in various obstacle environments. In addition, the calculation speed is high, and the path can be directly navigated by a mobile robot.

A storage medium for recording subtitle information based on text corresponding to moving picture data having a plurality of playback routes and an apparatus for and a method of reproducing an image including subtitles corresponding to the data. The storage medium includes moving picture data having a plurality of playback routes; a plurality of subtitle data items corresponding to the playback routes and supporting random search for a subtitle; and mapping information linking the moving picture data and the subtitle data. Accordingly, compatibility with a bitmap image method that is a subtitle processing method of a DVD or a bluray disc can be maintained, and multiple story playback is enabled. When random search or playback is performed, the search time for subtitle data is reduced such that search efficiency is enhanced.