66 results about "Neural Network Simulation" patented technology

A neural network simulation represents components of neurons by finite state machines, called sectors, implemented using look-up tables. Each sector has an internal state represented by a compressed history of data input to the sector and is factorized into distinct historical time intervals of the data input. The compressed history of data input to the sector may be computed by compressing the data input to the sector during a time interval, storing the compressed history of data input to the sector in memory, and computing from the stored compressed history of data input to the sector the data output from the sector.

The invention relates to a human body measurement method, in particular to the non-contact human body measurement method for clothing design which belongs to the technical field of clothing design. The digital images of the front surface and the side surface of the human body are firstly obtained; the extraction of the image edges are carried out by adopting the color segmentation, the best threshold segmentation, the hole filling, the opening operation and the pixel communication image processing method, thereby obtaining the image edges of the front surface and the side surface of the human body; the human body measurement feature points and the lines are determined, the corresponding height and width measurement data of the human body for the clothing design are obtained by calculation; a mathematical model is established by applying the width and the thickness data of the human body to obtain the related circumference data, and the human body circumference measurement data for the clothing design is obtained by BP neural network simulation and regression prediction treatment. The method has simple device, convenient operation and low measurement cost, thereby being capable of meeting the requirements on the design and the production of the clothing industry and having broad popularization and application prospects.

A neural network simulation represents components of neurons by finite state machines, called sectors, implemented using look-up tables. Each sector has an internal state represented by a compressed history of data input to the sector and is factorized into distinct historical time intervals of the data input. The compressed history of data input to the sector may be computed by compressing the data input to the sector during a time interval, storing the compressed history of data input to the sector in memory, and computing from the stored compressed history of data input to the sector the data output from the sector.

The invention belongs to the field of the environment perception and autonomous obstacle avoidance of quadrotor unmanned aerial vehicles and relates to a depth Q learning-based UAV (unmanned aerial vehicle) environment perception and autonomous obstacle avoidance method. The invention aims to reduce resource loss and cost and satisfy the real-time performance, robustness and safety requirements ofthe autonomous obstacle avoidance of an unmanned aerial vehicle. According to the depth Q learning-based UAV (unmanned aerial vehicle) environment perception and autonomous obstacle avoidance methodprovided by the technical schemes of the invention, a radar is utilized to detect a path within a certain distance in front of an unmanned aerial vehicle, so that a distance between the radar and an obstacle and a distance between the radar and a target point are obtained and are adopted as the current states of the unmanned aerial vehicle; during a training process, a neural network is used to simulate a depth learning Q value corresponding to each state-action of the unmanned aerial vehicle; and when a training result gradually converges, a greedy algorithm is used to select an optimal action for the unmanned aerial vehicle under each specific state, and therefore, the autonomous obstacle avoidance of the unmanned aerial vehicle can be realized. The method of the invention is mainly applied to unmanned aerial vehicle environment perception and autonomous obstacle avoidance control conditions.

The invention discloses an embedded system data compression and decompression method based on BP neural network, which comprises the following steps: 1). Choice of the type of neural network; 2). The structure of the mapping relation; 3). Data compression of each standard string on a PC based on BP neural network; 4). Data decompression in the embedded system based on BP neural network; 5). To write the standard strings get from decompression in decompressed data file in turn; 6). To delete all special characters occurring at the end of the file. The invention has the advantages of simulating the mapping relation between line code and line data using neural network, meeting the purpose of data compression through using the information occupying less signal space to express the information occupying more signal space, breaking through the limit of traditionally only depending upon coding to lower data redundancy, realizing higher compression ratio, repeating multiple data compression to reach satisfactory compression ratio, effectively compressing compressed data with entropy coding and further improving compression effect.

A communications system comprising a plurality of terminals each having a neural network therein which has parameter values enabling the network to emulate a transmission processing stage in a transmission mode, and a transmission station for sending new parameter values to the terminals to change the operation of the neural networks to emulate a new transmission mode.

The invention provides a method for predicting energy consumption of buildings during holidays and festivals on the basis of time series and neural networks. The method essentially includes predicting energy consumption of the buildings by means of fitting by the aid of the time series; solving prediction errors of energy consumption of the buildings during holidays and festivals; simulating the neural networks by the aid of influence factors on the energy consumption of the buildings during holidays and festivals and the solved prediction errors; computing modification values of the energy consumption of the buildings during holidays and festivals; modifying prediction results of the energy consumption of the buildings during holidays and festivals. The method has the advantages that the energy consumption of the buildings during holidays and festivals can be predicted, and the prediction precision can be improved to a great extent.

The invention discloses a method for identifying water logging grades of an oil reservoir by using a neural network analogue cross plot. In the method, the conventional cross plot technology is improved by a neural network algorithm, so nonlinear identification and quantitative analysis functions of the cross plot are realized, and a back propagation (BP) neural network algorithm is used and the method comprises the following steps of screening object characteristic parameters, selecting network structure parameters, training a neural network model, testing the network model, and establishing a neural network analogue cross plot layout. The method specifically comprises the following steps of: according to various characteristics of oil, gas and water layers in reservoirs, accurately selecting parameter samples which can best reflect the characteristics of the oil, gas and water layers in the reservoirs from parameters calculated during well logging or the well logging curves relevant to oil and gas interpretation by a statistics method; selecting appropriate weight values and threshold values by the BP neural network algorithm to establish the network model, and training the model and checking errors; and judging the fluid type or water logging degree of the reservoir with the depth according to projective points of identification vectors which are obtained by network output on a plane.

Method intended for real-time modelling, by neural networks, of hydrodynamic characteristics of multiphase flows in transient phase in pipes. In order to specifically take account of the possible flow regimes of fluids in pipes, various neural or “expert” models are formed for several flow regimes (or subregimes) in the whole of the variation range of the hydrodynamic characteristics of the flows (preferably for each one of them), as well as a neural model estimating the probability of belonging of the flows to each flow regime or subregime, knowing some of the characteristics thereof. The probabilities obtained are used for weighting the estimations delivered by each neural model, the result of the weighted sum being then the estimation eventually retained. Applications to various industries and notably for modelling of hydrocarbon flows in pipelines.

Disclosed is an MCSKPCA based neural network fault diagnosis method for analog circuits, comprising acquiring the output voltage signal of an analog circuit to be diagnosed; performing wavelet transformation on the acquired output voltage signal; calculating the energy eigenvalues of the wavelet coefficients of the output voltage signal, obtained through the wavelet transformation; performing MCSKPCA feature extraction and dimensionality reduction on the energy eigenvalues, and obtaining an optical eigenvector; and sending the optical eigenvector to a BP neural network separator, and outputting a fault diagnosis result by the BP neural network separator. The method can be used for not only diagnosis of linear or nonlinear circuits and systems thereof, but also diagnosis of hard fault and soft fault in the linear or nonlinear circuits.

The invention relates to a prediction method for building energy consumption based on a neural network. The method mainly comprises the following steps of: step 1, collecting energy consumption data of a building and taking the energy consumption data as sample data, carrying out normalization on the sample data, and enabling range of the sample data to be [0,1]; step 2, carrying out neural network simulation, and establishing a first neural network model for predicting the building energy consumption; step 3, predicting the building energy consumption by the first neural network model, and counting prediction error of the building energy consumption under the conditions of festivals and holidays; step 4, carrying out neural network simulation again, and establishing a second neural network model for predicting the a building energy consumption modification value under the conditions of the festivals and the holidays; and step 5, respectively counting predicted values of the building energy consumption under the conditions of the festivals and the holidays as well as work days. The prediction method of the building energy consumption in the festivals and the holidays based on the neural network has the beneficial effects that due to the adoption of the technical scheme, the prediction precision of the building energy consumption can be greatly improved, particularly the prediction precision under the conditions of the festivals and the holidays can be greatly improved, and the prediction method has important significance in energy resource monitoring of buildings.

The invention discloses a rapid sub-grade settlement predicting method based on a static sounding and a BP (Back Propagation) neural network, comprising the following steps of: obtaining a predicted field data sample, collecting a similar field data sample, establishing a BP neural network model, training and test the BP neural network model, and predicting a sub-grade settlement. A similar field static sounding test result, a field subsidiary stress and sub-grade settlement observation data are obtained as BP neural network training and test data samples for training the BP neural network repeatedly, when a difference between a prediction value and actual measurement data is smaller than a prescriptive standard, the predicted field data sample is input into the BP neural network model subjected to training, so as to obtain a sub-grade settlement prediction value. According to the invention, sub-grade settlement and deformation can be predicted scientifically and rapidly with an on-site static sounding test and a BP neural network simulation experiment, so that the rapid sub-grade settlement predicting method disclosed by the invention can be used for predicting various sub-grade foundation settlement and deformation in the civil engineering field.

The invention discloses a fast vein imaging method for visible-light skin images, and belongs to the field of information perception and recognition. The method comprises steps as follows: N groups of skin images comprising visible-light images and near-infrared images which are completely synchronous are collected to build a skin image library; corresponding pixel blocks of each group of the visible-light images and near-infrared images are sequentially selected to form a training database; a three-layer feedforward neural network is adopted to simulate a mapping relation of visible-light pixels to near-infrared pixels in the training database, and training adjustment is performed; an RGB (red green blue) pixel value of a to-be-measured visible-light skin image is input to the three-layer feedforward neural network which is well trained so as to obtain vein imaging of the image. According to the pixel corresponding relation of visible-light and near-infrared synchronous images, the mapping between the visible-light and near-infrared synchronous images is realized by adopting the feedforward neural network, and then the vein imaging is realized through an energy result graph; the processing process is simple and easy to realize, and the method has high practical and popularizing value; extra special device and equipment are not required, and the imaging cost is greatly reduced.

The invention discloses an office building load prediction method based on a particle swarm neural network. The method includes the following steps of: determining the input feature variable and the output target vector of an office building load prediction neural network; initializing a particle swarm solution set; calculating the fitness value of each particle; updating the local optimal position and the global optimal position of each particle; updating speeds and positions of particles; judging ending conditions; is the ending conditions are met, outputting the current optimal position; assigning the neural network and simulating the neural network, and predicting the load of an office building. Through the office building load prediction method based on the neutral network, all internal disturbance and external disturbance factors influencing fluctuation of the official building load are comprehensively considered. Meanwhile, aiming at the special periodic electricity consumption characteristic of the office building, the periodic load change is also considered; the high-precision load prediction of the office building is achieved by using manually simulating the neutral network; the office building load prediction method based on the particle swarm neural network has the advantages of high load prediction precision and simple and easy to implement.

The invention discloses a method and a device for elastically transmitting telemetry data based on machine learning. The method comprises the following steps of: performing data acquisition and training learning, acquiring the telemetering data in real time by a sending end and sending the telemetering data to a receiving end, meanwhile performing training on a neural network of the sending end until the deviation of a training result and the telemetering data is lower than a preset threshold; performing parameters transmitting and data simulation, transmitting neural network parameters whichcomplete a training target to the receiving end by a transmitting end, constructing an isomorphism neural network according to the neural network parameters by the receiving end and generating telemetry simulation data by utilizing allowed prediction indication information sent by the sending end; and performing prediction comparison and outliers elimination, acquiring the telemetering data and comparing the telemetry data with an output result of the neural network by the sending end, and generating indication information in combination with flight control prior information for subsequent processing. According to the method for elastically transmitting the telemetry data based on machine learning, the time varying characteristics of the information entropy of the telemetry parameter can be dynamically adapted, the transmission capacity of the telemetry data is greatly reduced, and the reliability and the flexibility of end-to-end information transmission of the space wireless link areimproved.

The invention discloses a neural network method for precisely determining tropospheric delays in a region; the neural network method comprises the following steps of: A1. obtaining an approximate true value of a tropospheric wet delay in a control point observation station; A2. establishing a tropospheric wet delay computation module in the region through the analogy computation of a neutral network; A3. computing a tropospheric dry delay in the region; A4. computing a tropospheric total delay in the region; and for other points in the region, obtaining delta 0<w>, delta +<w> and delta<w> through respective computation according to formulas (5), (7) and (8) as long as four ground meteorological parameters (P0, T0, h0 and e0) are obtained through meteorological observation, then obtaining delta<d> through computation according to a formula (9), and finally obtaining delta according to the formula (10). The invention puts forward a method for precisely determining a tropospheric delay modification module in the region by using an aerological sounding balloon for observing the appropriate true value of the tropospheric delay in the information extraction region and adopting a neutral network technology.

The invention relates to an area quasi-geoid refining method based on an earth gravity model (EGM2008). The method comprises the following steps of: 1) determining an area range and distributing control points; 2) performing field measurement (data acquisition); 3) on the basis of the EGM2008, acquiring a gravity height anomaly of each control point; 4) fitting a quadratic polynomial; 5) measuring adjustment; 6) performing analog computation on a neural network; and 7) refining a model. By adoption of the method, the accuracy of an area height anomaly computation result is high, and the application range of a measurement result of a global positioning system (GPS) height is expanded. Compared with the conventional quadratic polynomial fitting method, the method provided by the invention has the advantages that by a great amount of engineering project application result analysis, the accuracy of a height anomaly computation result is improved by 20 to 50 percent; after the accuracy is improved, the GPS height can replace low-level leveling, so the workload of the conventional low-level leveling which is high in cost, high in difficulty and long in period is reduced to the greatest extent, and economic benefit is obvious; and the method is applicable to the technical field of geodesy.

A method for precisely confirming the abnormality of a regional elevation is a method for precisely confirming the abnormality of the regional elevation by utilizing an NN technology and a grid technology which includes: 1) confirming the regional range and stationing; 2) measuring in the field (collecting data); 3) fitting a quadratic polynomial; 4) measuring an adjustment; 5) simulating and calculating the NN; 6) refining a model; 7) carrying out regional gridding; 8) internally inserting the elevation abnormality. The result precise by using the method for calculating the elevation abnormality is high, thus enlarging the application range of a GPS elevation measuring result. Analyzed by the application results of a plurality of engineering examples, the precise of the calculating result of the elevation abnormality by using the method of the invention is higher by 20 to 60 percent than that of fitting the quadratic polynomial; after the precise is improved, the GPS elevation can replace the leveling of lower grades, thereby reducing the traditional working load of measuring with high cost, high difficulty, long period and low grades to the lowest extent; the economic benefit is remarkable.

The invention belongs to the technical field of spiking neural networks, and discloses a spiking neural network analog circuit based on reinforcement learning. The spiking neural network analog circuit comprises an input layer nerve cell, a hidden layer nerve cell, an output nerve cell and a synapse; the input layer neurons are connected with the hidden layer neurons through synapses, and the hidden layer neurons are connected with the output neurons through the synapses; and the synapse is used for adjusting the first pulse signal of the pre-stage neuron according to the weight value and thentransmitting the adjusted first pulse signal to the post-stage neuron, and is also used for receiving the second pulse signal output by the post-stage neuron and updating the weight value according to the time difference between the first pulse signal and the second pulse signal and the reward signal. Based on reinforcement learning, a pulse neural network circuit is built, and an XOR classification function is achieved. Compared with a traditional pulse neural network, the method has the advantages of higher training speed and higher accuracy.

According to the voiceprint recognition model training method, the storage medium and the computer equipment, the linguistic features containing the identity information of the speaker are extracted as the input features, multi-task training is performed by using the gender and other tags of the speaker, and the cross-channel problem is solved in combination with the adversarial training method. Finally, stable features reflecting the identity essence of the speaker are extracted. According to the method, the linguistic characteristics and the deep neural network are combined to simulate the learning mechanism of the human brain, so that the extraction capability, stability and interpretability of the identity essential characteristics of the speaker are improved, and finally, the accuracyand recall rate of automatic voiceprint recognition are improved.

According to the training device of the voiceprint recognition model, the phonetic features containing the identity information of the speaker are extracted as the input features, multi-task trainingis carried out by using labels such as the gender of the speaker, the cross-channel problem is solved in combination with an adversarial training method, and finally, the stable features reflecting the identity nature of the speaker are extracted. According to the invention, the linguistic characteristics and the deep neural network are combined to simulate the learning mechanism of the human brain, so that the extraction capability, stability and interpretability of the identity essential characteristics of the speaker are improved, and finally, the accuracy and recall rate of automatic voiceprint recognition are improved.

Optimizing the allocation of farmland between different crops is provided. First and second Deep Boltzmann machines (DBMs) are built, wherein the hidden layers of the DBMs are split into a plurality of neural networks, each neural network modeling a different timeframe of crop growth. A plurality of factors related to crop growth are fed into the first DBM, which is trained to produce a first multi-class output of predicted maximum crop yields within a specified overall timeframe. The first multi-class output is fed into the second DBM, which is trained to produce a second multi-class output of predicted crop yields. The second multi-class output is fed into a decision support system that generates a recommended allocation of the farmland among different crops during different timeframes to maximize total yield.

The invention relates to a mechanical arm movement rhythm control method based on a CPG neural network. The method comprises the following steps that (1) an upper computer is used for setting electricconductivity and reverse potential parameters corresponding to nerve cell continuous sodium current, potassium current and leakage current as well as an upper limit threshold value, a lower limit threshold value and a threshold value period during variable-threshold shaping, and transmitting the electric conductivity, the reverse potential parameters, the upper limit threshold value, the lower limit threshold value and the threshold value period to the FPGA through USB communication; (2) the FPGA is used for establishing a CPG neural network simulation model according to the set parameters, simulation is carried out, a discharge waveform is output in the simulation process, and the waveform is transmitted to the upper computer to be displayed; (3) the FPGA is used for achieving variable-threshold shaping, the discharge waveform output by the CPG neural network in the step 2 is set, moreover, a control signal is output to a mechanical arm so as to control and adjust the movement rhythmof the mechanical arm, and the angular displacement of a mechanical arm joint is transmitted to the upper computer to be displayed.

A method propagates a pulse of wave through the material to receive a set of echoes resulted from scattering the pulse by different portions of the material and simulates a propagation of the pulse in the material using a neural network to determine a simulated set of echoes. Each node in a layer of the neural network corresponds to a portion of the material and assigned a value the permittivity of the portion of the material, such that the values of the nodes at locations of the portions form the image of the distribution of the permittivity of the material. The connection between two layers in the neural network models a scattering event. The method updates the values of the nodes by reducing an error between the received set of echoes and the simulated set of echoes to produce an image of the distribution of the permittivity of the material.

The invention discloses a method for generating a fluidized bed formula based on a neural network system. The method comprises the following steps: 1. The neural network system collects training samples of the neural network and stores them in a data storage module; 2. Simulation training of the neural network system; 3. Application of neural network system to automatically generate recipes. The present invention uses neural network technology to set the production formula, automatically generates the production formula parameters to control the operation of the fluidized bed with the raw material parameters, and does not need experienced formula engineers to set the formula with experience, improves product performance and production efficiency, and improves formula setting scientific and accurate.

Method intended for real-time modelling, by neural networks, of hydrodynamic characteristics of multiphase flows in transient phase in pipes. In order to specifically take account of the possible flow regimes of fluids in pipes, various neural or “expert” models are formed for several flow regimes (or subregimes) in the whole of the variation range of the hydrodynamic characteristics of the flows (preferably for each one of them), as well as a neural model estimating the probability of belonging of the flows to each flow regime or subregime, knowing some of the characteristics thereof. The probabilities obtained are used for weighting the estimations delivered by each neural model, the result of the weighted sum being then the estimation eventually retained. Applications to various industries and notably for modelling of hydrocarbon flows in pipelines.

The method for optimizing and controlling pressure in gas-oil separation plants utilizes a genetic algorithm-based control method for controlling pressure in each stage of a multi-stage gas-oil separation plant to optimize oil production parameters. A neural network simulation model is used with an optimization procedure to provide on-line operation optimization of the multi-stage gas-oil separation plant. Pressure set points of each stage are automatically and continuously adjusted in the presence of fluctuating ambient temperatures and production rates to ensure optimal oil recovery and optimal quality of the produced oil.

The invention discloses a data flow-oriented test case generation method, which comprises the following steps of (1) performing data flow analysis on a program to be tested to obtain all definitions in the program; a usage pair; (2) designing and training a BP neural network to simulate a fitness function; and (3) generating a test case covering all definition-use pairs by using a genetic algorithm. The method has the beneficial effects that the test case is generated for the all-uses data flow criterion and is used for solving the test problem of the Java program; compared with a traditionalmethod, the neural network is used for simulating calculation of the fitness function, and the program instrumentation operation frequency in the genetic algorithm can be reduced; through the design of the neural network structure, the flow of the test case generation method based on the neural network is optimized, and the flow does not need to be executed for multiple times for multiple to-be-tested targets.

The invention discloses a method for simulating attention mobility by using a neural network, which adopts a visual image input layer, neuronal oscillator network oscillation layer and an attention mobility realizing layer, wherein the visual image input layer inputs a gray value of a gray level image into a neurodynamic network; the neuronal oscillator network oscillation layer couples dynamic systems, which are built by all oscillators in the neurodynamic network according to a FitzHugh-Nagumo model, to form the neurodynamic network; and the attention mobility realizing layer realizes the attention mobility between the synchronized different objects by changing parameters, so that the frequency output by nerves of a currently concerned object increases. On the basis of the neurodynamic system, the method forms simple simulation on human-eye attention mobility visual treatment by analyzing and modifying the FitzHugh-Nagumo model and has a great theoretical and practical significance for further research on the visual treatment mechanism of human.

The invention discloses a neural network simulation method and device, and aims to solve the problems that multi-layer continuous simulation verification cannot be completed and the efficiency is lowdue to the fact that a storage path of to-be-simulated layer data cannot be automatically acquired during simulation in an existing simulation verification system for a neural network model of an FPGA. According to the embodiment of the invention, generating the storage path of the current to-be-simulated layer data according to the pre-defined common path for storing the hidden layer data and thelayer identifier of the current to-be-simulated layer; performing simulation aiming at the current layer according to the data obtained from the file corresponding to the generated storage path; thehidden layer may comprise a plurality of layers; when simulation is carried out, storage paths of all hidden layer data do not need to be set independently, through the embodiment of the invention, the storage path of the current to-be-simulated layer data can be automatically generated, multi-layer continuous simulation is realized, the time for independently establishing simulation of each layeris saved, the simulation operation process is simplified, and the simulation efficiency is improved.