234 results about "Neural network modeling" patented technology

The invention provides a method and a system for deep nerve translation based on character encoding. A combined nerve network model is established by using an RNN to cover the whole translation process, and translation tasks are directly completed from the perspective of an encoder-decoder framework. The method comprises the following steps: A, word vector generation: performing word segmentation on character-level input data through neural network modeling and generating a word vector; B, language model generation: establishing grammar rules by utilizing the characteristic of memory of the recurrent neural network in time; C, word alignment model generation: obtaining the probability of translating multiple words in a source language statement into target language words; D, output: translating an inputted source language into a target language; E, translation model combination: establishing a deep nerve translation model (RNN-embed) based on character encoding in combination with neural network models in the four steps and accelerating model training by using CPU parallel computation.

The invention discloses a reading understanding model method based on paragraph internal reasoning and joint question answer matching, and the method comprises the following steps: S1, constructing avector for each candidate answer, the vector representing the interaction of a paragraph with a question and an answer, and then enabling the vectors of all candidate answers to be used for selectinganswers; S2, carrying out experiment. According to the model provided by the invention, paragraphs are firstly segmented into blocks under multiple granularities; an encoder is used for summing the intra-block word embedding vectors by utilizing neural word bag expression; then, a relationship between blocks with different granularities where each word is located through a two-layer forward neuralnetwork is modeled to construct a gating function, so that the model has greater context information and captures paragraph internal reasoning at the same time. Compared with a baseline neural network model such as a Stanford AR model and a GA Reader, the accuracy of the model is improved by 9-10%. Compared with a recent model SurfaceLR, the accurcay is at least improved by 3% and is about 1% higher than that of a single model of the TriAN, and in addition, the model effect can also be improved through pre-training on an RACE data set.

The invention discloses a neural network modeling method and a neural network modeling system. The method comprises: a data preprocessing step of acquiring historical data in actual processes and screening sample data out; a neural network initializing step of initializing parameters of a neural network model; an output calculation step of calculating the output of the neural network model according to the model and the sample input; a gain calculation step of calculating the gain of the output of the model relative to the input according to the parameters of the model and the sample data; and a model training step of instructing the neural network model to perform parameter iterative learning according to an error between the output of the model and the sample data and the input output gain of the model until the model meeting accuracy requirement and gain restriction is obtained. Through the method and the system, gain relationship between the input and the output of the neural network model can be definitely represented and the established model can be more consistent with actual conditions through the gain restriction.

A method for modeling error-driven adaptive control of an aircraft. Normal aircraft plant dynamics is modeled, using an original plant description in which a controller responds to a tracking error e(k) to drive the component to a normal reference value according to an asymptote curve. Where the system senses that (1) at least one aircraft plant component is experiencing an excursion and (2) the return of this component value toward its reference value is not proceeding according to the expected controller characteristics, neural network (NN) modeling of aircraft plant operation may be changed. However, if (1) is satisfied but the error component is returning toward its reference value according to expected controller characteristics, the NN will continue to model operation of the aircraft plant according to an original description.

A propylene polymerization production process optimal soft-measurement meter based on genetic algorithm optimized BP neural network comprises a propylene polymerization production process, a site intelligent meter, a control station, a DCS databank used for storing data, an optimal soft measurement model based on genetic algorithm optimized BP neural network, and a melting index soft-measurement value indicator. The site intelligent meter and the control station are connected with the propylene polymerization production process and the DCS databank; the optimal soft-measurement model is connected with the DCS databank and the soft-measurement value indicator. The optimal soft measurement model based on genetic algorithm optimized BP neural network comprises a data pre-processing module, an ICA dependent-component analysis module, a BP neural network modeling module and a genetic algorithm optimized BP neural network module. The invention also provides a soft measurement method adopting the soft measurement meter. The invention can realize on-line measurement and on-line automatic parameter optimization, with quick calculation, automatic model updating, strong anti-interference capability and high accuracy.

The invention discloses a soft-sensing modeling method and a soft meter of a multi-model neural network in a biological fermentation process. The method comprises the following steps: a data preprocessing module preprocesses input variable data by a normalization and principle component analysis method; and then the data preprocessing module carries out cluster division on a preprocessed principle component variable set; through and then a BP neural network model module respectively establishes sub neural networks according to different clusters and finally establishes a soft-sensing model of the multi-model neural network. The soft-sensing model of the multi-model neural network is used for measuring biomass concentration in a fermentation process on line, and a measurement value is displayed through a biomass concentration soft-sensing value displayer. The invention introduces a core fuzzy C mean clustering algorithm based on a particle swarm algorithm and combines the mean clustering algorithm with the modeling method of the multi-model neural network, and the established model is simple, realizes the on-line measurement of the biomass concentration and has timely control, high measurement accuracy and strong capacity of resisting disturbance.

Embodiments of the invention can include methods and systems for controlling clearances in a turbine. In one embodiment, a method can include applying at least one operating parameter as an input to at least one neural network model, modeling via the neural network model a thermal expansion of at least one turbine component, and taking a control action based at least in part on the modeled thermal expansion of the one or more turbine components. An example system can include a controller operable to determine and apply the operating parameters as inputs to the neural network model, model thermal expansion via the neural network model, and generate a control action based at least in part on the modeled thermal expansion.

The invention discloses a performance prediction method applicable to dynamic scheduling for a semiconductor production line. An extreme learning machine (ELM) is applied to prediction and modeling in the performance prediction method. Feeding control and scheduling rules are considered in a unified manner in the method, short-term scheduling key performance indexes such as an equipment utilization rate and a movement step number are predicted on the basis of a real-time state of a system, and a foundation is provided for dynamic real-time scheduling. A novel feed-forward neural network of the ELM is introduced into the semiconductor manufacturing system, and a prediction model is built by the aid of available data of the production line. As shown by test results, ideal prediction results can be quickly acquired by the method implemented by the aid of the ELM, the method has obvious advantages and an obvious application prospect in the aspects of parameter selection and learning speed as compared with the traditional neural network modeling method, and a new idea is provided for online optimal control.

The invention discloses an abnormal gait identification method capable of facilitating screening Parkinsonism. The method is characterized in that the gait plantar pressure characteristics can be extracted, and the modeling and the identification of the neural network of the gait system of the normal healthy people and the patients suffering from the Parkinsonism can be dynamically carried out; the constant neutral network can be established; a dynamic estimator can be built by using the constant neutral network, and based on the difference between the gait modes of the normal healthy people and the patients suffering from the Parkinsonism in the gait system dynamics, the abnormal gait caused by the Parkinsonism and the normal gait of the normal healthy people can be distinguished according to the minimum error principle, and the screening detection of the Parkinsonism can be facilitated. By arranging a pressure sensing floor system or wearing the special shoes provided with the pressure sensor insole, the plantar pressure characteristics can be acquired, and the abnormal gait caused by the Parkinsonism and the normal gait of the normal healthy people can be distinguished conveniently, simply, and non-invasively, and therefore the daily gait monitoring of the family members can be realized, and the screening detection of the Parkinsonism can be facilitated.

The invention relates to a construction method of a fuzzy neural network expert system for the water quality assessment in turbot culture, which comprises neural network modeling and network model testing. The method is characterized in that the neural network modeling comprises the following steps of: setting network training parameters, wherein sensitive indexes of the growth of turbots after three-tiered classification, which include four expert data of temperature, salinity, pH and dissolved oxygen, are used as input parameters; determining the network topology: constructing a three-layer neural network by using the input layers of four input nodes, the hidden layers of two hidden nodes and the output layer of one output node; and determining training samples. The network model testing comprises the steps of: leading in tested samples and assessing the network model. The invention combines the neural network, the fuzzy system and the on-line monitoring system for the water quality assessment in industrial turbot culture for the first time, avoids the operations such as manual setting and the like in the traditional assessment method, and overcomes the influence of human factors and the like on the assessment in the traditional turbot water quality assessment.

The invention discloses an ultra-short-period photovoltaic prediction method. The method comprises the following steps: selecting training data x; performing normalization processing on the training data; performing data exception handling on the training data; performing data functional transformation; performing significance analysis; training a generalized regression neural network model; and predicating the generalized regression neural network model. According to the ultra-short-period photovoltaic prediction method, a generalized regression neural network modeling theory and method is adopted; partial approximation is further accurate by adding a primary function in a hidden layer, and global optimum is achieved; significance extraction and improvement is carried out specific to the model input information; the correlation of historical data is enhanced through the functional transformation, and the historical data, used as the input signal, enters the generalized regression neural network prediction model, so that the prediction efficiency is effectively improved; in addition, after a training sample is chosen, the generalized regression neural network structure and the weight are determined automatically by only requiring to adjust smoothing parameters, so that the computational process for circuit training is avoided, and the global approximation study and prediction capability is realized more rapidly.

A method for a computer-aided control of a technical system is provided. The method involves use of a cooperative learning method and artificial neural networks. In this context, feed-forward networks are linked to one another such that the architecture as a whole meets an optimality criterion. The network approximates the rewards observed to the expected rewards as an appraiser. In this way, exclusively observations which have actually been made are used in optimum fashion to determine a quality function. In the network, the optimum action in respect of the quality function is modeled by a neural network, the neural network supplying the optimum action selection rule for the given control problem. The method is specifically used to control a gas turbine.

The invention discloses a multi-view angle gait recognition method based on Kinect. The multi-view angle gait recognition method based on Kinect comprises the steps that three-dimensional space position information of a framework joint point is collected and normalized to side view angles through view angle normalization; gait features existing after view angle normalization are extracted, and neural network modeling and recognition are conducted on gait system dynamics at different view angles in a training set; a constant value neural network is established, a dynamic estimator is established, and through the differences between gait modes at different view angles on the gait system dynamics, a test mode is recognized according to the minimum error rule. According to the multi-view angle gait recognition method based on Kinect, the three-dimensional space position information of the framework joint point is obtained, the help of other sensing devices is not needed, image processing is not needed, the complexity of a system is lowered, and the extraction precision of feature data is improved.

The invention relates to a wind driven generator operation state analyzing method, in particular to a wind driven generator operation stat analyzing method based on a neural network model. The method comprises the steps of reading a database, preprocessing data, modeling a neural network, storing a wind energy converting power characteristic curve model, reading the wind energy converting power characteristic curve model, acquiring data, calculating expected output power, carrying out statistics on the tolerance between the expected output power and actual output power, and judging the operation state of a wind driven generator. The wind driven generator operation stat analyzing method based on the neural network model can effectively and accurately monitor and forecast the operation state of the wind driven generator and namely provides effective and accurate decision support for maintenance of the wind driven generator.

The invention discloses an abnormal gait detection method based on a determined learning theory. The abnormal gait detection method based on the determined learning theory includes the steps that features are extracted, neural network modeling and identification are dynamically carried out on a gait system of healthy and normal people and patients with motor neurodegenerative diseases of different types on the basis of the extracted gait features, a constant neural network is built, a dynamic estimator is built by the utilization of the constant neural network, and abnormal gaits caused by the motor neurodegenerative diseases are distinguished from normal gaits of the general healthy people according to the minimum error principle on the basis of differences between the gait mode of the healthy and normal people and the gait mode of the patients with the motor neurodegenerative diseases of different types on the gait system dynamics, so that the abnormal gaits are detected accurately and a detection result is evaluated. The abnormal gait detection method based on the determined learning theory has the advantages that the method is convenient and easy to implement and is in a non-invasion mode, and under the intelligent home environment, daily gait monitoring on family members can be achieved by mounting a pressure sensing floor system or wearing special shoes with sensor insoles.

The present invention relates to a rockburst dynamic prediction method based on BP neural network modeling. The method comprises the steps of: determining and acquiring rockburst influence factors; performing quantification processing on qualitative description parts in influence factor indexes, and obtaining an initial population; performing BP neural network training on the eight acquired influence factors separately; optimizing a number of neurons, an algorithm learning rate and momentum factors by using a genetic algorithm, and obtaining an optimal hidden layer node number; and performing prediction on rockburst of a mine by using a BP neural network algorithm model obtained through training, and obtaining a risk level of the rockburst of the mine. The method provided by the present invention has relatively high reliability, overcomes the defect of no association between the rockburst and the influence factors of the rockburst in the current rockburst prediction process, implements middle and short term dynamic prediction on the rockburst, and can be widely applied to the field of mine rockburst prediction.

The invention discloses a real-time yield predicting method for a catalytic cracking device. According to the real-time yield predicting method for the catalytic cracking device, kinetic parameters and device parameters of a catalytic cracking reaction are corrected in real time by processing field real-time data by adopting a data reconciliation technology, and combining an improved differential evolution algorithm, so that the actual operating situations of the device can be described accurately by using a catalytic cracking device mechanism model. The method comprises the following steps: on the basis of a corrected model, analyzing the influence on the yield of a catalytic cracking product caused by key operation / process conditions, such as an operating temperature, a feeding load, a raw material preheating temperature, a reaction pressure, a residue adding ratio, a regenerator temperature, a catalyst-to-oil ratio and the like; performing piecewise linearization according to an influence trend, solving a linear equation to obtain corresponding Delta-Base yield data, associating the operating conditions and the Delta-Base yield data by combining a neural network modeling technology, and establishing a yield agent model, so that the yield data calculating speed is improved; the real-time yield predicting of a continuous catalytic cracking device is realized; a theoretical support is provided for establishing an accurate plan optimization PIMS model.

The invention discloses an analog circuit test node selecting method based on dynamic feedback neural network modeling. The method comprises the steps of selecting the frequency of a test signal, inputting the test signal into a circuit to be tested, simulating various typical fault conditions, collecting the voltage values of a normal sample and a fault sample of the circuit on a test node to be selected of the circuit to be tested so as to construct a fault dictionary table; according to a fault fuzzy voltage interval, analyzing a fuzzy fault set and obtaining a fault integer encoding table; constructing an initial training sample set, training an initial dynamic feedback neural network, and utilizing the dynamic feedback neural network for fitting the nonlinear mapping relation between the test node and the fault; and according to the target function calculated by the genetic algorithm output by the network, obtaining the optimal test node set by utilizing the genetic optimization algorithm. In the method, a fault dictionary is analyzed by the intelligent algorithm, so that the global optimum test node set can be found, and the subsequent diagnostic accuracy can be further improved.

The invention belongs to a neural network modeling method which is applicable to the electron-beam welding techniques of various metal materials, and relates to the neural network modeling method of shape factors in a fusing region of the electron-beam welding. The neural network modeling method adopts neural network methods and systems to set up a mathematical model of the shape factors in the fusing region of the electron-beam welding, takes multiple non-related factors into consideration in all aspects to be used as an input layer for model solving, and belongs to the modeling methods with non-related multi-input and multi-output processing.

The invention relates to an RBF neural network modeling method based on feature clustering, which belongs to the field of automatic control, information technology and advanced manufacture. The invention particularly relates to an RBF neural network modeling method based on feature extraction function clustering, which can solve the modeling problem that data can be scattered. The method is characterized by comprising the following steps: defining a feature extraction function based on existing mechanism knowledge, determining an RBF network center in a clustering algorithm based on the feature extraction function, and determining a weight value from the hidden layer to the output layer of the RBF network in a least square method. The invention also provides a clustering algorithm based on the feature extraction function, which is not used for directly clustering data, but is used for clustering data with scattering features through introduction of the feature extraction function based on the mechanism knowledge. The obtained clustering center is used as the RBF network center, and the weight value from the hidden layer to the output layer of the RBF network can be obtained with a linear interpolation method. The invention can effectively solve the modeling problem that the data has scattering features, and can achieve high modeling accuracy.

The invention discloses a micro-gyroscope control system based on a neural network. The micro-gyroscope control system comprises a control system which is in connection communication with a micro-gyroscope, wherein the control system comprises a nominal controller and a neural network compensator. The control scheme of a feedback controller based on a micro-gyroscope nominal model and the neural network compensator is applicable to control of the micro-gyroscope, so that the tracking effect and robustness of the micro-gyroscope control system are improved. The adaptive law of a weight value of the neural network is designed based on a Lyapunov stability theory, so that the stability of the whole control system can be guaranteed under the condition of presence of a modeling error of the neural network.

The invention discloses a tricholoma matsutake fast nondestructive testing system and method based on a convolutional neural network. The system comprises a deep learning convolutional neural networkmodel, a control end and a consumer terminal; the deep learning convolutional neural network model comprises sample collection, data acquisition, and deep learning convolutional neural network modeling and optimization; in the sample collection, a sample set is established by finishing sample screening of a detection object, and the sample set is divided into a training set, a verification set anda test set; the data acquisition comprises sample chemical content measurement and spectral data acquisition; in the deep learning convolutional neural network modeling and optimization, preprocessedspectral data and corresponding chemical content are modeled through the deep learning convolutional neural network model and pooling processing; a detection result of the tricholoma matsutake by thedeep learning convolutional neural network model is stored in the control end; and the consumer terminal can obtain detection data of the tricholoma matsutake by accessing the control end. Accordingto the tricholoma matsutake fast nondestructive testing system and method, the detection cost can be effectively reduced, and the market supervision of the supervision department can be facilitated.

The invention belongs to the field of equipment maintenance time prediction, and relates to a selective wear-based equipment optimal maintenance time prediction method. The method mainly comprises two steps of: solving a selective wear possibility value of each part of equipment in the current state by utilizing an association rule algorithm; and taking the solved possibility value as input, and solving the optimal maintenance time by neural network modeling. The method comprises the following steps of: constructing an association rule library; acquiring state monitoring data, extracting characteristic values from the data, and establishing an equipment monitoring data set; matching the equipment monitoring data set with the association rule library, and calculating the wear possibility value of each part under the condition of successful matching; and training a self-organizing competitive neural network model, and predicting the optical maintenance time by utilizing the model.

Based on principle of minimization risk of configuration, combined with cooperative collaboration evolution algorithm, and learning network structure of neural network and connection weight value, the invention obtains optimal compromise between network structure and connection weight value finally. The method includes three basic steps: data processing, network learning and network estimated forecast. Configuring network and learning connection weight value are carried out at same time in the invention so as to better solve practical problems existed in traditional neural network learning: correlation between result and initial value, slow convergence rate, easy to run to local minimum value as well as derivable error function needed and over learning. The invention raises learning capability and generalization capability of network, applicable to intelligent diagnosing heart disease, fault diagnosis in industries, stock and goods price forecasting etc.

The invention relates to an optimal soft measurement instrument based on EGA optimization of a propylene polymerization production process, which comprises the propylene polymerization production process, an on-site intelligent instrument, a control station, a DCS database for storing data, an optimal soft measurement model based on EGA optimization and a melt index soft measurement value indicator; the on-site intelligent instrument and the control station are connected with the propylene polymerization production process and the DCS database; and the optimal soft measurement model is connected with the DCS database and the soft measurement value indicator. The optimal soft measurement model based on EGA optimization comprises a data preprocessing module, an ICA independent component analyzing module, a BP neural network modeling module and an EGA optimizing module. The invention also provides a soft measurement method realized by the soft measurement instrument. The soft measurement instrument and the soft measurement method realize online measurement, automatic online parameter optimization, high calculation speed, automatic updating of the model, strong anti-interference capability and high precision.

The present invention relates to the field of sewage treatment, and is realized based on BP neural network modeling. By establishing the output layer variable of the neural network, the effluent COD concentration, and selecting the input layer variable, the transfer function of the hidden layer and the transfer function of the output layer of the neural network, A BP neural network model is established; sample data of the input layer variables and output layer variables of the neural network are selected, abnormal data processing is performed on the sample data, and the BP neural network model is used for training and prediction to obtain the COD concentration of the effluent. This method can quickly and accurately predict the effluent COD concentration during A2O wastewater treatment, providing reliable control conditions for A2O wastewater treatment.

The invention relates to an image data extraction and neural network modeling-based platinum flotation grade estimation method. The method comprises the following steps: acquiring the correlation degrees between six variables of aeration rate, pulp density, a collecting agent, an activating agent, a foaming agent and an inhibitor and flotation grade and recovery rate by an variable experiment; carrying out collection and pretreatment on a platinum foam image; extracting five image data of energy, entropy, inertia, homogeneity and gray correlation from four tolerant feature images such as gray level images, histogram equalization, contrast enhancement of images and image binarization obtained by pretreatment; and building a multi-layer perceptron neural network model comprising a three-node input layer, a hidden layer and a dual-node output layer. According to the image data extraction and neural network modeling-based platinum flotation grade estimation method provided by the invention, flotation grade and recovery rate are effectively estimated through the foam image, and the target of monitoring grade and recovery rate in the flotation process in real time is achieved.

A method for a computer-aided control of a technical system is provided. The method involves use of a cooperative learning method and artificial neural networks. In this context, feed-forward networks are linked to one another such that the architecture as a whole meets an optimality criterion. The network approximates the rewards observed to the expected rewards as an appraiser. In this way, exclusively observations which have actually been made are used in optimum fashion to determine a quality function. In the network, the optimum action in respect of the quality function is modeled by a neural network, the neural network supplying the optimum action selection rule for the given control problem. The method is specifically used to control a gas turbine.

The invention discloses a full-process modeling method for an oil refining process. Based on the mechanisms and running characteristics of all production devices in the oil refining process and a corrected model, the influence of key operation / process conditions of all the devices on the product yield is analyzed. Piecewise linearization is carried out according to the influence trend, a linear equation is solved, corresponding Delta-Base yield data are obtained, a neural network modeling technology is combined, operation conditions and the Delta-Base data are related, a yield surrogate model is built, the yield data calculating speed is improved, real-time prediction on the product yield in the oil refining process is achieved, and theoretical supports are provided for building a precise plan optimized PIMS model.

The invention discloses a deterministic learning theory based gait recognition method irrelevant to a visual angle, and belongs to the technical field of pattern recognition. The method comprises the following steps: preprocessing; extracting characteristics; on the basis of extracted gait characteristics, dynamically carrying out neural network modeling and identification to a gait system at different visual angles in a training set; establishing a literal neural network; and constructing a dynamic estimator, and realizing the accurate classification recognition of test modes according to a least error principle by utilizing differences among gait modes on gait system dynamics at different visual angles. The dynamic local accurate modeling and identification of the human body gait system at different visual angles can be realized, meanwhile, gait modes at different visual angles are formed into a uniform training gait mode library, a problem of a traditional method that corresponding training sets need to be independently constructed for gait modes at different visual angles can be overcome, the gait recognition irrelevant to the visual angle is realized, and the invention exhibits higher robustness and practicality.