129 results about "Neural network design" patented technology

Fractal computers are neural network architectures that exploit the characteristics of fractal attractors to perform general computation. This disclosure explains neural network implementations for each of the critical components of computation: composition, minimalization, and recursion. It then describes the creation of fractal attractors within these implementations by means of selective amplification or inhibition of input signals, and it describes how to estimate critical parameters for each implementation by using results from studies of fractal percolation. These implementation provide standardizable implicit alternatives to traditional neural network designs. Consequently, fractal computers permit the exploitation of alternative technologies for computation based on dynamic systems with underlying fractal attractors.

The invention, which belongs to the robot system control field, relates to a distributed adaptive-neural-network continuous tracking control method of a multi-robot system. According to the existing coordinated tracking and controlling method of the multi-robot system, problems of parameter uncertainty and external interference existence in the multi-robot system exist. The provided method comprises: under the circumstances that only parts of followers can obtain dynamic navigator state information, a distributed observer design is implemented with limitation of communication tine delay existence, so that all followers can obtain the dynamic navigator state information; and with consideration of the parameter uncertainty and external interference existence in the system, controlling is carried out by using a distributed adaptive tracking control expression designed based on two neural networks, so that the approximate error is close to zero. In addition, the control algorithm of the distributed adaptive tracking control expression is in a continuous control mode, no buffet is caused at the system and the great practical application value is created. Besides, validity of the control algorithm is verified by the simulation experiment.

The invention discloses an RBF neural network based mobile manipulator self-adaptive control method. The method comprises the following steps of S1, establishing a standard mobile manipulator dynamical model; S2, constructing an RBF neural network of the robot dynamic model; S3, designing a mobile manipulator trajectory tracking method with the adaptive capability through the constructed neural network; S4, automatically identifying unknown mobile platform and manipulator dynamic parameters through online learning, and conducting closed-loop identification and compensation on the unknown dynamic parameters, wherein the parameters of the RBF neural network can be updated on line, and finally, the feasibility and effectiveness of the simulation verification control method are verified. Through the method, output errors caused by the unknown dynamic parameters and external disturbance can be eliminated completely without an accurate robot dynamic model; the deficiency that a model-based robot control scheme cannot be implemented without the accurate dynamic model is made up; and the dynamic performance of a mobile manipulator and the trajectory tracking precision of a joint space areimproved.

The invention discloses a BP neural network construction and identification method for identifying different brown rice grains, adopting the following technical scheme: including the following steps: 1) acquiring images: 2) image preprocessing: 3) extracting image feature information of different brown rice grains: 4) main Component analysis reduces dimensionality of image feature information; 5) Design BP neural network structure: 6) Train neural network, and use the BP network neural network constructed by any of the above-mentioned construction methods to identify different brown rice grains. The method proposed by the present invention obtains images of different types of brown rice grains through black as the background, uses image processing technology to obtain its characteristic information, and uses principal component analysis to reduce the dimensionality of the characteristic information, and finally uses BP neural network to perform different types of brown rice grains. identify. The method can identify different types of grains objectively, accurately and quickly, and overcomes the disadvantages of traditional manual detection.

The invention discloses a short text automatic abstracting method and system based on double encoders, belongs to the technical field of information processing, and is characterized by comprising thefollowing steps: 1, preprocessing data, 2, designing a double encoder with a bidirectional recurrent neural network, and 3, arranging an attention mechanism fusing global and local semantics 4, arranging a decoder with empirical probability distribution and using a decoder designed by adopting a double-layer unidirectional neural network; 5, adding word embedding characteristics, 6, optimizing word embedding dimensions, and 7, carrying out preprocessing and testing on the news corpus data from the Sogou laboratory, substituting the news corpus data into a Seq2Seq model with double encoders andaccompanying empirical probability distribution to carry out calculation, and carrying out experimental evaluation through a text abstract quality evaluation system Rouge. According to the invention,traditional weaving is carried out; and the decoding framework is subjected to optimization research, so that the model can fully understand text semantics, and the fluency and precision of text abstracts are improved.

The invention discloses a general image classification and identification system and method based on neural networks. The system comprises a parameter acquisition module, an expert system database, an algorithm generation module and a parameter adjustment module. Based on a neural network design template embedded in the expert system database, a whole set of corresponding algorithm suitable for the types of the neural networks are automatically generated according to specific application of a user, and a parameter space is automatically explored through learning and testing, and parameters are adjusted to get the best results. The design difficulty of neural networks is greatly reduced, and the user does not need very complicated professional background to directly use the neural networks.

A data analytics platform is provided for forecasting future states of commodities and other assets, based on processing of both textual and numerical data sources. The platform includes a multi-layer machine learning-based model that extracts sentiment from textual data in a natural language processing engine, evaluates numerical data in a time-series analysis, and generates an initial forecast for the commodity or asset being analyzed. The platform includes multiple applications of neural networks to develop augmented forecasts from further analysis of relevant information as it is collected. These include commodity-specific neural networks designed to continually develop taxonomies used to process commodity sentiment, and applications of reinforcement learning, symbolic networks, and unsupervised meta learning to improve overall performance and accuracy of the forecasts generated.

The invention relates to a water ship trajectory tracking control method for actuator asymmetric saturation. The method comprises steps: 1, an expected tracking trajectory is given, an expected plane position (x<d>, y<d>) is given, an expected yaw angle psi<d> is given, and the expected trajectory is presented to be eta<d>=[ x<d>, y<d>, psi<d>]<T>; 2, a trajectory tracking error is calculated, and the error, z1=eta - eta<d>, between the actual trajectory and the expected trajectory is calculated; 3, an expected speed is calculated, and the expected speed alpha needed for eliminating the error between the expected trajectory and the actual trajectory is calculated; 4, a neural network is designed, an uncertainty delta in a kinetic equation uses the neural network delta=W<*T> phi(theta) + epsilon to be approximately expressed; 5, an auxiliary control system is designed, a Gauss error function is used, and a smooth asymmetric saturation actuator model is designed; and 6, a model control law is calculated, a control quantity tau needed for eliminating the error between the expected trajectory and the actual trajectory is calculated. The method can approximate a model uncertainty, suppress external disturbance influences, resist asymmetric saturation of an actuating mechanism, track any expected trajectory and ensure gradual stability of the system.

A method for stimulation of hydrocarbon production via imbibition by utilization of surfactants. The method includes use of fuzzy logic and neural network architecture constructs to determine surfactant use.

Fractal computers are neural network architectures that exploit the characteristics of fractal attractors to perform general computation. This disclosure explains neural network implementations for each of the critical components of computation: composition, minimalization, and recursion. It then describes the creation of fractal attractors within these implementations by means of selective amplification or inhibition of input signals, and it describes how to estimate critical parameters for each implementation by using results from studies of fractal percolation. These implementation provide standardizable implicit alternatives to traditional neural network designs. Consequently, fractal computers permit the exploitation of alternative technologies for computation based on dynamic systems with underlying fractal attractors.

The invention discloses a flexible manipulator composite control method based on output redefinition, which is a flexible manipulator control strategy based on output redefinition combined with compound learning and disturbance observation, belongs to the field of robot control, and is used for solving the existing problem that the joint angle cannot be controlled with high precision by the flexible manipulator under the influence of external disturbance and internal unknown information. The method adopts the output redefinition processing system dynamics equation to make the system achieve zero dynamic stability; a disturbance observer is designed to process disturbance terms due to external disturbance and joint friction; and a composite learning control algorithm is designed to construct system modeling errors, and the adaptive updating law including the modeling errors of the RBF neural network design is further combined. The interaction between disturbance observation and compoundlearning realizes a more comprehensive estimation of system uncertain information, which is suitable for engineering high-precision control.

The invention relates to a voice visualization method based on an integration characteristic and a neural network. The method is characterized by comprising the following eight steps of: preprocessing a voice signal, extracting characteristics, optimizing the characteristics, designing the neural network, mapping position information, encoding a main color, encoding pattern information and synthesizing an image. Different voice characteristics are integrated in an image to create a voice signal readable mode for a deaf mute, and images at different positions have different colors, so that the advantage that the deaf mute has higher color stimulated visual memory capacity is better utilized; moreover, tone characteristics are adopted to encode the pattern information in order to reduce a screen accommodating load and an observer memory burden; therefore, voices consisting of the same final and different tones are displayed at the same position. Compared with the conventional method, the voice visualization method has high robustness and classification positioning capacity, and has a good effect of assisting the learning of the deaf mute.

The invention discloses a PMSM chaotic system rapid terminal sliding mode control method based on a nerve network, comprising steps of establishing a PMSM chaotic system model, initializing a system state and a control parameter, designing a nerve network, approximating the non-linear uncertain item in the system, and designing a rapid terminal sliding mode controller on the basis of the nerve network. The PMSM chaotic system rapid terminal sliding mode control method based on the nerve network can effectively solve the calmness problem of the PMSM chaotic system, improves the rapid convergence performance of the system and realizes the accuracy control of the PMSM chaotic system.

A method for stimulation of gas hydrocarbon production via imbibition by utilization of surfactants. The method includes use of fuzzy logic and neural network architecture constructs to determine surfactant use.

A method for constructing a neural network, a device, a computer device and a storage medium, wherein the method comprises: according to the user 's operation in the preset neural network design interface, determining each target network element in the neural network to be constructed, according to the name of each target network element, the parameter information of each target network element being obtained, the hierarchical relationship of each target network element being determined according to the directional connecting line between each target network element, and the neural network tobe constructed being generated according to the hierarchical relationship of each target network element and the parameter information of each target network element. The method realizes that the neural network can be constructed only according to the directional connecting line between the target network element and the target network element selected by the user on the visualized neural networkdesign interface, and the code is not required by the user, thus reducing the development difficulty and improving the construction efficiency.

The invention provides an impact localization method based on phase-sensitive optical reflection and deep learning of a convolutional neural network. The impact localization method includes the following steps: 1) two optical fiber monitoring network topological structures for impact load localization, based on a phase-sensitive optical time domain reflection principle; 2) structure design of a phase-sensitive optical time domain reflection sensing probe for impact load monitoring; 3) construction of a distributed phase-sensitive optical time domain reflection sensing impact monitoring systemand meshing of a thin plate monitoring area; 4) phase-difference-based phase-sensitive optical time domain reflection technique for impact load localization; 5) generation of an impact response samplelibrary based on a Phi-OTDR sensor; 6) data pre-processing and deep learning convolutional neural network design; and 7) using the trained deep learning convolutional neural network to identify the impact response data of the Phi-OTDR sensor.

The invention discloses a micro-gyroscope double-feedback regression neural network sliding-mode control method. On the basis of a conventional single-layer regression neural network, the method adds an output regression item, and carries out simulation on an MEMS micro-gyroscope model. A designed sliding-mode neural network controller can guarantee that a system tracking error is converged to be zero. Moreover, the method employs a double-feedback regression neural network and a common regression neural network, and the results of a neural sliding mode controller of the common regression neural network indicate that the double-feedback regression neural network is better in approaching effect and is more stable in internal state. The neural network designed in the invention can set a central vector and the initial value of a base width randomly, and the central vector and the base width can be stabilized to be optimal values along with a designed adaptive algorithm according to different inputs.

A mechanical-arm servo-system neural-network full-order sliding mode control method with dead-zone compensation is disclosed. Aiming at a mechanical arm servo system which contains a dynamic execution mechanism and is with unknown dead-zone input, a full-order sliding mode control method is used and a neural network is combined so as to design the mechanical-arm servo-system neural-network full-order sliding mode control method with the dead-zone compensation. A dead zone is converted into a linear time-varying system, and then the neural network is used to approach an unknown function so as to compensate an additional influence of a traditional unknown dead zone and an unknown parameter of the system. In addition, a full-order sliding mode surface is designed so as to guarantee rapid and stable convergence of the system; generation of a differential term is avoided in an actual control system so that buffeting is improved and a singular problem is solved. The invention provides the control method which can improve a buffeting problem of the sliding mode surface, solve the singular problem and can effectively compensate a system unknown dynamic parameter and unknown dead zone input so that rapid and stable control of the system is realized.

The invention discloses a novel voice signal feature fusion method. The method comprises the following steps: S1, constructing a model framework; S2, designing a neural network; and S3, carrying out audio feature representation and extraction. According to the method, a CNN variant and an LSTM variant are parallelly fused to form the novel hybrid neural network, meanwhile, the signal features having the maximal influences on the final emotion are extracted by combining with an Attention mechanism, finally, the audio emotion vector capable of more accurately and more rapidly classifying emotionis obtained, the generalization ability is strong, the structure is clear, the integration with or separation from other modules is easy, and in addition, the problem that the traditional voice signal features can not realize effective fusion is solved.

The present invention relates to an emotion visualization method facing voice and a facial expression signal. The method mainly comprises emotional signal acquisition, emotional signal preprocessing, emotional feature parameter extraction, neural network design and image generation. Firstly, the obtained emotional signal is subjected to feature extraction, then the obtained emotion feature is used as the input of a neural network, the output of the neural network is corresponding pattern information, then a visualization image is generated through an image generation module, and the visualization of six types of human basic emotions which are neutral, joy, anger, surprise, sadness and fear is realized. Through combining different modes of emotional features into an image, the reading mode of emotion is created for people, the display is visual, and the robustness is high.

The invention discloses a semantic text similarity calculation method based on attention. The method comprises the steps: 1, carrying out preprocessing on each text pair in a data set so as to obtaincorresponding text data samples; 2, dividing all the samples into a training set and a verification set, establishing a neural network, and training the neural network by using the samples in the training set to obtain a network model for semantic text similarity calculation; and 3, preprocessing a text pair to be calculated, and inputting the preprocessed text pair into the network model to obtain a semantic similarity calculation result of the text pair. The neural network designed by the invention can more effectively extract semantic information of texts and extract finer-grained interaction information between two texts; and the neural network uses a plurality of attention mechanisms to enhance important information in interaction information between two texts and improve semantic text similarity calculation accuracy.

The invention provides a risk situation analysis processing method and processing device based on a BP neural network. The processing method comprises the steps of constructing a case library according to historical risk factor data of a workshop; preprocessing the historical risk factor data in the case library to obtain preprocessed case data; establishing a neural network model according to risk situation prediction requirements; training the neural network model according to the case data to obtain a trained target neural network; and obtaining current risk factor data of the workshop, andinputting the current risk factor data into the target neural network to obtain a risk situation prediction result. According to the embodiment of the invention, the risk situation analysis model isdesigned by adopting the BP neural network, so that the nonlinear relationship between non-deterministic risk factors and the risk situation is effectively solved, and the adaptability and reliabilityof an analysis system are improved; and meanwhile, workshop risk points in a budding state can be checked and solved in time, and the accident rate is reduced.

The invention discloses a multi-scale image segmentation method based on weight learning. According to the neural network designed by the invention, the target features are fully extracted on different scales, the edge and morphological features of the target are fully ensured, and the learnable weight is used to reserve useful features to eliminate noisy features. The network model mainly comprises an encoding part and a decoding part, in the encoding stage, features under different scales are extracted through a network, the features of a multi-layer feature space are fused in the decoding stage, a category probability distribution diagram can be obtained through each enhanced feature, and the obtained enhanced features are weighted through learnable self-adaptive weights to obtain finalfeatures. Experiments show that the method provided by the invention has higher accuracy and recall rate in road extraction, and is closer to the contour of a real road in the aspect of road appearance.

The invention provides an in-memory pulse neural network based on current integration. Calculation based on a charge domain is naturally compatible with a working mechanism of neurons. In one aspect, in order to avoid non-ideality of employing NVM materials, memory cells of a synaptic array in an architecture employ silicon-based SRAM cells. In addition, the modified NVM unit can benefit from the architecture of the built-in pulse neural network designed by the invention. When the synaptic array adopts an SRAM (Static Random Access Memory) unit as a storage unit, the design of a post-neuron circuit corresponds to the SRAM unit, so that the SNN architecture in the storage can be used for calculating a multi-bit synaptic weight, and the combined column number is programmable. Further, in order to improve the use efficiency of the area and save energy efficiency, in the calculation of the multi-bit synaptic weight, the circuit is designed to be in a time multiplexing form of resource sharing. Finally, an automatic calibration circuit is provided to counteract the change of the conduction current caused by factors such as process, voltage, temperature (PVT) and the like, so that the calculation result is more accurate.

A new method for stock selection, portfolio optimization and asset allocation using specially constructed neural networks and ranking algorithms. Each stock in a portfolio is modeled by a neural network designed to learn from the statistical differences between the stock and the target benchmark. The neural network outputs, representing each model's forward looking rate of return (FLRoR) relative to the benchmark, are then ranked in descending order. Portfolio optimization is performed by rebalancing the positions based on their FLRoR rankings subject to the portfolio constraints including regulatory requirements and risk level. The method has been implemented in computer programs.

The invention discloses an intelligent motion control method for a motor servo system. The intelligent motion control method belongs to the field of electromechanical servo control. The intelligent motion control method for the motor servo system is characterized by comprising the specific steps of: step 1, establishing a mathematic model of the motor servo system, and simplifying an electrical dynamic state of a motor as a proportional link according to Newton's second law; step 2, designing a multi-layer neural network to estimate unknown function disturbance suffered by the motor servo system; step 3, designing an extended state observer to estimate time-varying external interference of the motor servo system in combination with the three-layer neural network; step 4, and designing a motor servo system position tracking controller based on the multi-layer neural network and disturbance feedforward compensation. The intelligent motion control method can ensure that the position output of the motor servo system can precisely track an expected position instruction, thereby being more beneficial to application under complex working conditions in engineering practice.

The invention discloses an intelligent autonomous operation system of a GEO satellite. The system comprises the steps of: designing a controlling law of orbit keeping and controlling by a neural network based on relative motion equation; estimating the satellite state during orbit keeping and controlling period in real-time by a neural network state observer and taking the satellite state as an autonomous orbit determination result of the orbit keeping and controlling period; and transmitting the estimated state to a controller and realizing the autonomous orbit keeping and controlling of theGEO satellite by the controller, thereby realizing the autonomous operation of the GEO navigation satellite.

The invention relates to the technical field of computational intelligence, in particular to an interval type II intuitionistic fuzzy random vector function connection neural network design method. The interval type II intuitionistic fuzzy random vector function connection neural network design method comprises the following steps: constructing an interval type II intuitionistic fuzzy random vector function connection neural network; adopting a hybrid learning method combining quantum clustering algorithm, swarm intelligence algorithm and least square method to adjust the structure and parameters of the neural network. The neural network adopts the membership function and the non-membership function based on the beta function as the forepart of the fuzzy rule, and adopts the random vectorfunction to connect the neural network as the afterpart of the interval fuzzy rule. The IT2IF-RVFLNN designed by the invention can achieve global approximation.