44 results about "Nonlinear neural networks" patented technology

A neural network classifier provides the ability to separate and categorize multiple arbitrary and previously unknown audio sources down-mixed to a single monophonic audio signal. This is accomplished by breaking the monophonic audio signal into baseline frames (possibly overlapping), windowing the frames, extracting a number of descriptive features in each frame, and employing a pre-trained nonlinear neural network as a classifier. Each neural network output manifests the presence of a pre-determined type of audio source in each baseline frame of the monophonic audio signal. The neural network classifier is well suited to address widely changing parameters of the signal and sources, time and frequency domain overlapping of the sources, and reverberation and occlusions in real-life signals. The classifier outputs can be used as a front-end to create multiple audio channels for a source separation algorithm (e.g., ICA) or as parameters in a post-processing algorithm (e.g. categorize music, track sources, generate audio indexes for the purposes of navigation, re-mixing, security and surveillance, telephone and wireless communications, and teleconferencing).

A method and system for objectively evaluating the quality of speech in a voice communication system. A plurality of speech reference vectors is first obtained based on a plurality of clean speech samples. A corrupted speech signal is received and processed to determine a plurality of distortions derived from a plurality of distortion measures based on the plurality of speech reference vectors. The plurality of distortions are processed by a non-linear neural network model to generate a subjective score representing user acceptance of the corrupted speech signal. The non-linear neural network model is first trained on clean speech samples as well as corrupted speech samples through the use of backpropagation to obtain the weights and bias terms necessary to predict subjective scores from several objective measures.

The invention discloses an information recommendation method based on graph convolution and neural collaborative filtering. In combination with the advantages of a graph convolution neural network model, fusion processing can be carried out on various information in an intuitive manner, so that not only feature information of a user but also attribute information of the user can be received, and relatively good recommendation performance can be achieved for sparse score data; and input and parameters of the model are subjected to optimization modeling by using multiple skills, so that the detail problems encountered possibly are solved. In addition, a nonlinear neural network-based collaborative filtering method is introduced as a decoder part of the model, so that user and article codes output by a graph convolution encoder can be well utilized, and through an end-to-end model, all processes run in the same framework without being trained separately. Through the processing of input data and the training and prediction of the model, a complete score prediction matrix can be obtained.

A neural network classifier provides the ability to separate and categorize multiple arbitrary and previously unknown audio sources down-mixed to a single monophonic audio signal. This is accomplished by breaking the monophonic audio signal into baseline frames (possibly overlapping), windowing the frames, extracting a number of descriptive features in each frame, and employing a pre-trained nonlinear neural network as a classifier. Each neural network output manifests the presence of a pre-determined type of audio source in each baseline frame of the monophonic audio signal. The neural network classifier is well suited to address widely changing parameters of the signal and sources, time and frequency domain overlapping of the sources, and reverberation and occlusions in real-life signals. The classifier outputs can be used as a front-end to create multiple audio channels for a source separation algorithm (e.g., ICA) or as parameters in a post-processing algorithm (e.g. categorize music, track sources, generate audio indexes for the purposes of navigation, re-mixing, security and surveillance, telephone and wireless communications, and teleconferencing).

The invention relates to a method for the optimization of operation and control of energy conservation for a reserve ball-grinding powder system. Firstly, a non-linear neural network model about the consumption of powder of a powder-making system and operating parameters is established through collecting the service data of the powder-making system, then the negative pressure and the temperature at the entrance of a coal mill and the temperature at the outlet of the coal mill are obtained through the non-linear optimization with restricted conditions as a set value for controlling the system, wherein, the negative pressure, the temperature at the entrance and the outlet lead to lower consumption of powder; the entire optimized controlling of the powder-making system is carried out through the intelligent control of rotation of the coal mill and multivariable decoupling control of parameters at the entrance of the coal mill based on the operating experience as well as an intelligent down draft control technology, so as to ensure that the powder-making system strictly runs at an optimal condition and effectively reduces the powder consumption. The method provided by the invention can effectively inhibit the fluctuations of a primary air pressure in the on / off process of the powder-making system, put an end to serious accidents such as flameout in a boiler caused by down draft and effectively reduce the operating intensity for the staff.

A wafer temperature estimator calibrates contact-type temperature sensor measurements that are used by a temperature controller to control substrate temperature in a high temperature processing chamber. Wafer temperature estimator parameters provide an estimated wafer temperature from contact-type temperature sensor measurements. The estimator parameters are refined using non-contact-type temperature sensor measurements during periods when the substrate temperature is decreasing or the heaters are off. A corresponding temperature control system includes a heater, a contact-type temperature sensor in close proximity to the substrate, and an optical pyrometer placed to read temperature directly from the substrate. A wafer temperature estimator uses the estimator parameters and measurements from the contact-type sensor to determine an estimated wafer temperature. A temperature controller reads the estimated wafer temperature and makes changes to the heater power accordingly. The wafer temperature estimator has a nonlinear neural network system that is trained using inputs from the various sensors.

The invention discloses a multi-model self-adaptive controller and control method designed for a zero-order closely-bounded nonlinear multivariable system. The multi-model controller is provided with a nonlinear robust self-adaptive controller and a nonlinear neural network self-adaptive controller. Due to the fact that a nonlinear complementation item is introduced, the identification model of the nonlinear robust self-adaptive controller and the output errors of a real system are guaranteed to be bounded. The BIBO stability of the system can be guaranteed through a one-step advanced control law designed by utilizing model errors. System control signals are generated through the switching of the two controllers, and the performance of the system is improved through the neural network self-adaptive controller.

Methods and apparatus are provided pertaining to a design of experiments. The method comprises generating a data set from historical data; identifying and removing any fault data points in the data set so as to create a revised data set; supplying the data points from the revised data set into a nonlinear neural network model; and deriving a simulator model characterizing a relationship between the input variables and the output variables. The apparatus comprises means for generating a data set from historical data; means for identifying and removing any fault data points in the data set so as to create a revised data set; means for supplying the data points from the revised data set into a nonlinear neural network model; and means for deriving a simulator model characterizing a relationship between the input variables and the output variables.

The invention discloses a nonlinear neural network optimizing PID control method for the temperature of an electric heating furnace. An RBF neural network is trained offline according to historical input and output information of an electric heating furnace control system, neural network related parameters are obtained, and a trained neural network serves as a prediction model of the system; and the RBF neural network is combined with the PID controller, and the RBF neural network is used to self-set the parameters of the PID controller online. Thus, the difficulty in establishing a nonlinear electric heating furnace model is overcome, the RBF neural network is used to self-set the parameters of the PID controller, and the problem that the electric heating furnace control system is hard to set the parameters of the PID controller in the practical control process is solved.

The invention discloses a speed control method and a speed control device and relates to the technical field of a computer. A nonlinear neural network model can simulate non-linear relationship between velocity instructions under different environments and speed values; a current velocity instruction and historical velocity instructions within a preset time as well as historical velocity values are input to the nonlinear neural network model to obtain a corrected velocity instruction, wherein the corrected velocity instruction is obtained by carrying out correction on the current velocity instruction according to the non-linear relationship between the velocity instructions and the speed values; and the corrected velocity instruction is input to a system based on PID control to obtain an expected speed value, thereby improving speed control accuracy and precision.

Methods and apparatus are provided pertaining to a design of experiments. The method comprises generating a data set from historical data; identifying and removing any fault data points in the data set so as to create a revised data set; supplying the data points from the revised data set into a nonlinear neural network model; and deriving a simulator model characterizing a relationship between the input variables and the output variables. The apparatus comprises means for generating a data set from historical data; means for identifying and removing any fault data points in the data set so as to create a revised data set; means for supplying the data points from the revised data set into a nonlinear neural network model; and means for deriving a simulator model characterizing a relationship between the input variables and the output variables.

The invention discloses a piezoelectric actuator hysteresis neural network compensation method for helicopter body vibration active control, and belongs to the field of helicopter vibration control. Aiming at the problems that the helicopter body vibration has the vibration characteristic of multi-order harmonic response, and the control effect becomes poor due to higher harmonic response caused by hysteresis nonlinearity of a piezoelectric actuator in the active control process of helicopter body vibration driven by the piezoelectric actuator, based on a neural network and a nonlinear autoregressive exogenous (NARX) input model, a piezoelectric actuator hysteresis nonlinear neural network and a nonlinear compensation neural network under the driving of two-order harmonic signals are provided, and the nonlinear compensation neural network is used in a helicopter body vibration active control system driven by a piezoelectric actuator. The piezoelectric actuator hysteresis neural networkcompensation method provided by the invention can obviously improve the control effect of the helicopter body vibration active control system.

The invention discloses a polynomial auxiliary neural network behavior modeling system and method for a power amplifier. The modeling system is characterized in that the modeling system comprises a polynomial auxiliary module and a neural network module, the polynomial auxiliary module utilizes prior information of the power amplifier to fit main nonlinearity of the power amplifier, and the neuralnetwork module compensates for characteristics which cannot be represented by the polynomial auxiliary module and carries out fine fitting on nonlinear behaviors of the power amplifier. The polynomial auxiliary module and the neural network module are integrated in the same neural network, and coefficients of the two modules are updated at the same time by adopting a back propagation algorithm. The invention further discloses a polynomial auxiliary neural network behavior modeling method for the power amplifier. The prior information of the power amplifier is embedded into the neural networkmodel, so that the complexity of the model is greatly reduced under the condition that the modeling precision is not lost.

Methods and apparatus are provided for determining refrigerant charge in a vapor compressor system (VCS) of an aircraft. The methods and apparatus comprise the following steps of, and / or means for, generating a data set from historical data representative of a plurality of VCS operating conditions over time, identifying one or more steady-state data points in the generated data set, forming a revised data set that includes at least the steady-state data points, using principal components analysis (PCA) to derive values for a plurality of minimally correlated input variables, supplying the derived values for the plurality of minimally correlated input variables and the corresponding values for the VCS refrigerant charge in the revised data set to a nonlinear neural network model, and deriving a simulator model characterizing a relationship between the plurality of minimally correlated input variables and the VCS refrigerant charge.

The invention discloses a solenoid valve type vibration damper control method based on a non-linear neural fuzzy logic controller. The method includes the steps of deeply optimizing input fuzzy variables, namely vertical vibration acceleration and vertical vibration speed of a vibration damping platform, and output fuzzy variables, namely membership function of a damping force coefficient, through a non-linear neural network method in smart control theory, and finally conducting real-time control on the solenoid valve type vibration damper through a control core. The solenoid valve type vibration damper control method solves the complex non-linear characteristic problem of a solenoid valve type vibration damper, and ensures riding comfortableness and operation stability. The vibration damping effect and the driving performance of a motor vehicle can be improved, and the service lifetime of the motor vehicle can be prolonged. The solenoid valve type vibration damper control method can be conveniently applied on semi-active suspension systems of various motor vehicles.

The present invention relates to a smart city system artificial intelligence evaluation method based on a nonlinear neural network which solves the technical problem that the intelligence quantitative evaluation cannot be carried out on a smart city system. By adopting the technical scheme of establishing an initial database; pre-processing the initial database to obtain a prediction database; according to the prediction database and a non-linear autoregression network with the external input, establishing a non-linear autoregression smart city system performance prediction model p(t) about time series; according to the non-linear autoregression smart city system performance prediction model p(t), exciting a function and training an algorithm, carrying out the real-time non-linear autoregression smart city system performance prediction, and inputting the performance prediction results in the initial database, the method solves the problem better, and can be used for the smart city system performance evaluation.

The invention discloses a ship automatic berthing nonlinear neural network control method. The method comprises the following steps: performing a large amount of successful berthing by utilizing ship,acquiring training data, designing a controller by utilizing three-layers of nonlinear neural network, and training network weight and deviation by utilizing a BP algorithm; and realizing the outputof the ship automatic berthing control under different initial input conditions by utilizing a generalization performance of the ANN. The remote monitoring and control of the unmanned ship can be realized through the technical scheme provided by the invention; the provided automatic berthing control method is independent of a virtual navigation route with uncertain element; in the actual control process, the computation load is reduced, an ideal control effect is reached when the control instantaneity is improved, and the voyage practice requirement is satisfied.

The invention discloses a nonlinear neural network model for a modeling wide band RF (Radio Frequency) power amplifier. The model comprises an input layer, a hidden layer and an output layer, wherein the input data of the input layer comprises advance items x (n+1), |x (n+1)|3, ..., |x (n+1)|<2Q+1>, aligning items x(n), |x(n)|, |x(n)|[3], ..., |x (n)|<2Q+1>, and delay items x (n-1), ..., x (n-M[1]), |x (n-1)|, |x (n-1)|, ..., |x (n-M[2])|, ..., |x (n-1)|<2Q+1>, ..., |x (n-M[Q+2]|<2Q+1>, wherein the x (n+1) is base band complex data of an input end of RF power amplifier at current time, and the output of the output layer is y(n). The nonlinear neutral network has the advantages that a generalized memory effect (memory effects at the delay time and the advance time shall be considered) is considered based on a super-strong memory effect and a strong static nonlinearity of the modeling RF power amplifier; meanwhile, an input signal of an input layer does not only comprises a base band signal, but also comprises a model of a base band complex signal and a high power of the model, and the output signal of the output layer is a plural signal, therefore the modeling precision is higher and can be improved by 5dB in comparison with a real time delay neural network model.

The invention discloses a joint design method of IRS reflection pattern and channel estimation based on deep learning, and the method aims at an intelligent reflection surface assisted wireless communication system, and comprises the following steps: (1) generating a training data set needed for training a nonlinear neural network; (2) building a nonlinear neural network, and jointly training the nonlinear neural network by using the training data set generated in the step (1) to obtain an intelligent reflection surface reflection pattern and channel estimation; (3) enabling the base station to send the reflection pattern obtained through training in the step (3) to the intelligent reflection surface and configure the reflection pattern; and (4) enabling the base station to carry out online channel estimation by adopting the channel estimation nonlinear neural network obtained by training in the step (3). Compared with a traditional channel estimation method, the method has the advantages that the overhead of the pilot frequency can be remarkably reduced on the premise of the same channel estimation precision, the online calculation complexity is low, and engineering implementation is facilitated.

The invention discloses a supercritical fluid heat transfer correlation type proxy model construction method based on machine learning, and aims to solve the problems that a traditional or developed heat transfer experience correlation type is difficult to predict and poor in precision due to nonlinear physical property change of supercritical fluid. The method comprises the following steps: firstly, widely collecting experimental data, and evaluating and selecting thermal boundary, geometric and physical property dimensionless parameter factors which potentially influence a heat transfer grade; then, on the basis of a singular value decomposition technology, carrying out data order reduction processing, and achieving main flowing heat exchange feature recognition and extraction of samples; establishing a mathematical expression of the supercritical heat transfer model and a nonlinear RBF-MLP neural network structure, and training, verifying and optimally selecting the number of neurons of an input layer, a hidden layer and an output layer; and finally, enabling a prediction result to show that the heat transfer correlation type agent model has the characteristics of high prediction precision and small network error. The scheme is simple and reliable, and the purposes of accurately predicting the wall surface temperature and the heat transfer coefficient and reducing the test cost can be quickly achieved.

The invention provides a fault diagnosis method for an aircraft airspeed tube based on neural network analysis and redundancy output. Sufficient historical data of input and output of the aircraft airspeed tube under normal working conditions is obtained, a non-linear neural net model for describing the input and output characteristics of the airspeed tube is established, and the neural network model is trained by using historical training data of input and output under the normal conditions to construct a neural network analytical model of the aircraft airspeed tube. After common fault modesof the airspeed tube are determined, the residual data of the actual output of the airspeed tube and a signal output by the neural network model are collected, when the residual data is greater than atolerance value, it is determined that a fault occurs in the airspeed tube, and the actual output of the airspeed tube and the analytical output signal of the neural network model are subjected to linear regression to identify fault characteristic parameters, thereby identifying fault categories occurring in the airspeed tube and achieving the fault diagnosis of the aircraft airspeed tube.

The invention discloses an MRAC control method of a hydraulic servo system based on a nonlinear neural network. Aiming at matched and unmatched interference and parameter uncertainty in the hydraulicservo system, a nonlinear neural network is adopted to approach state-related interference so as to carry out feed-forward compensation, and meanwhile, parameters related to input are updated on linein order to further improve the accuracy of feed-forward compensation. In the aspect of theoretical proof, the symbolic function robust integral control strategy (RISE) is combined with the MRAC, theapproximation error of the neural network is suppressed through the RISE, asymptotic tracking is realized without utilizing an acceleration signal, and finally the effect of the method is verified through experiments.

The invention discloses an image reconstruction method and device based on a peak potential distribution time interval. The method comprises the following steps: extracting distribution time intervalcharacteristics of a peak potential; constructing a reconstruction model by using a neural network learning algorithm according to the distribution time interval characteristics of the peak potential;and reconstructing the image based on the constructed reconstruction model. According to the method, the issuing time interval characteristics in the peak potential signals are extracted, and the nonlinear neural network reconstruction model is combined, so that a complex gray level image is effectively reconstructed.

The invention relates to a method for the optimization of operation and control of energy conservation for a reserve ball-grinding powder system. Firstly, a non-linear neural network model about the consumption of powder of a powder-making system and operating parameters is established through collecting the service data of the powder-making system, then the negative pressure and the temperature at the entrance of a coal mill and the temperature at the outlet of the coal mill are obtained through the non-linear optimization with restricted conditions as a set value for controlling the system,wherein, the negative pressure, the temperature at the entrance and the outlet lead to lower consumption of powder; the entire optimized controlling of the powder-making system is carried out throughthe intelligent control of rotation of the coal mill and multivariable decoupling control of parameters at the entrance of the coal mill based on the operating experience as well as an intelligent down draft control technology, so as to ensure that the powder-making system strictly runs at an optimal condition and effectively reduces the powder consumption. The method provided by the invention caneffectively inhibit the fluctuations of a primary air pressure in the on / off process of the powder-making system, put an end to serious accidents such as flameout in a boiler caused by down draft andeffectively reduce the operating intensity for the staff.