18 results about "Linear dynamical system" patented technology

The invention discloses a permanent magnet synchronous motor full-speed-domain position-sensorless hybrid control method and a permanent magnet synchronous motor full-speed-domain position-sensorless hybrid control system, and the permanent magnet synchronous motor full-speed-domain position-sensorless hybrid control method is constructed by taking a permanent magnet synchronous motor nonlinear dynamic system model as a starting point. A hybrid control method combining high-frequency pulsating square wave signal injection under a linear dynamic system model, a rotor polarity judgment and rotor initial position estimation algorithm and an unscented Kalman filtering extension algorithm in an additive noise form under a nonlinear discrete dynamic system model is adopted; and obtaining the optimal estimation observation value of the system state variable in the static, low-speed, medium-speed and high-speed full-speed domain range under the condition that the permanent magnet synchronous motor is not provided with a mechanical position sensor. The optimal estimation observation value is used for compensating and feeding back a permanent magnet synchronous motor drive control system adjusting strategy, disturbance of load torque is improved, the noise amount and disturbance amount introduced by three-phase current measurement are eliminated, and efficient and accurate control operation of the permanent magnet synchronous motor is achieved.

The present invention relates to methods and systems for improving the efficiency of artificial neural networks by configuring them to implement linear dynamical systems that compute state updates in linear time. More specifically, the present invention specifies methods and systems for setting the weights of least one linear artificial neural network layer by (a) selecting an set of basis vectors that define a desired impulse response for the layer, (b) deriving a matrix that produces this desired impulse response over a target temporal window, (c) deriving a matrix that dampens the impulse response to zero outside of the target window, and (d) combining these two matrices together through addition to obtain the layer's recurrent weights. Systems composed of at least one such linear layer are applied to a time series of input data elements to produce outputs that encode the results of pattern classification, signal processing, data representation, and data generation tasks.

Koopman operator theory is a widely used method to analyze, control, and predict the behavior of the states of a non-linear dynamical system using measurement functions in Hilbert space. Real time approximation of the Koopman operator is crucial in order to adapt and understand behavior of underlying non-linear dynamical system. Traditional approaches leverage matrix-based methods or artificial neural networks to approximate Koopman operator. However, such methods necessitate significant power and computational resources, therefore may not be suitable for applications that require real-time on-board processing. The problems of the conventional approaches are resolved based on a recent development of brain inspired spiking neural networks and neuromorphic computing platforms, as these offer extremely low-energy computation and real-time responses. Embodiments of the present disclosure provide implementation of a Spiking Neural Network (SNN) based architecture that efficiently approximate Koopman operator with minimal length of data and demonstrates significant computational savings.

The invention discloses a time sequence probability prediction method based on the Kumpmann theory and Kalman filtering, and the method comprises the following steps: 1, carrying out the patching of input time sequence data, and mapping the data to a high-dimensional potential space; 2, mapping the time sequence data after patch mapping to a high-dimensional measurement space, and modeling a linear dynamic system through a Kupman operator; step 3, optimizing the linear dynamic system by using a Kalman network, modeling the state of the linear dynamic system, updating state estimation, and outputting potential variables; and step 4, carrying out resampling from variational posteriori distribution, mapping the potential variables back to an original space, carrying out modeling to predict target distribution, and outputting a probabilistic prediction result of a time sequence. The invention further discloses a prediction system for realizing the prediction method, and the prediction system has a wide application scene.

The present application relates to a kind of 3D medical image mask self-encoding methods based on Mamba, to optimize the self-encoding task of processing 3D medical image, especially solve the problem of high memory requirement and computing resource consumption encountered when processing long sequence data.The Mamba network architecture introduced in the present application is used as the backbone network of the self-encoding model, to optimize the ability of the model to process long sequence data.Mamba is an advanced architecture based on state space model, which combines linear dynamic system theory and neural network concepts, effectively capturing the timing information and dynamic characteristics in the data.The core advantages of the present application include selective scanning algorithm, hardware-aware algorithm, and efficient data processing capability, especially when processing long sequence data.The present application not only reduces the burden of computing and storage, but also maintains or improves the accuracy and efficiency of image processing, especially suitable for processing 3D medical image data that requires a large amount of computation and high memory bandwidth.

The present application relates to the technical field of intelligent sensing and emotion computing, in particular to an end-to-end emotion recognition method and system based on a seat cushion pressure sensor array, which collects human sitting posture pressure distribution time series signals through a seat cushion pressure sensor array, obtains a standardized time series tensor through quality exploration, sliding window slicing and standardization processing; uses a pre-trained multi-scale spatiotemporal feature extraction network to extract emotion-related pressure spatiotemporal features, which are input into a multi-layer perceptron classifier after running normalization and linear dynamic system smoothing to output emotion categories. The system includes a pressure data acquisition module, a data preprocessing module, a multi-scale spatiotemporal feature extraction module, a feature post-processing module and an emotion classification output module. Through comparative learning pre-training and multi-scale spatiotemporal feature mining, the present application improves the cross-subject generalization ability and recognition stability, and realizes end-to-end direct recognition from original pressure signals to multi-category emotions.

The application discloses an industrial system mechanism modeling simulation platform, a unified system modeling simulation code system is constructed based on a Python language, a system identification model function library is constructed based on AI technology, and a dynamic model of a complex industrial system or equipment is identified through industrial big data; a general system modeling simulation environment is provided, a linear dynamic system and an artificial intelligence nonlinear dynamic model are incorporated into a unified modeling simulation system, visual combination modeling and industrial system simulation are carried out, and identification modeling simulation integration is supported; and the application is used for industrial system identification modeling simulation technology required by industrial internet and industrial intelligent construction, and provides a reliable system model for design of a high-end controller.

This invention discloses a scalable method and system for learning humanoid precision grasping skills in robots, involving the intersection of artificial intelligence and robotics. The method comprises: obtaining a robot's motion trajectory that imitates a human grasping; establishing a robot grasping skill model based on the motion trajectory that imitates the human grasping; learning and optimizing the parameters of the robot grasping skill model to obtain a grasping reference trajectory guided by the human's experience and knowledge; converting the grasping reference trajectory into a linear dynamic system and optimizing the parameters of the linear dynamic system to achieve global convergence of the robot's grasping skill; and reconstructing the robot grasping skill model using intermediate points to achieve rapid expansion of the robot's grasping skill. This invention can solve the problems of poor scalability and low accuracy in robot grasping path planning in unstructured environments.

The invention discloses a depth Koopman modeling method and system fused with differential quadratic programming, and the method comprises the steps: collecting and processing the historical operation data, state data and input data of a nonlinear dynamic system, and dividing the data into a training set and a verification set; a depth Koopman modeling framework integrated with a differentiable quadratic programming layer is constructed; and training the modeling framework by using a historical operation data set, and optimizing to-be-trained parameters through training to finally obtain a dimension raising mapping function of the modeled nonlinear dynamic system and an optimal high-dimensional global linear dynamic model corresponding to the function. According to the method, the dimension raising mapping function can be automatically learned and optimized, and the current optimal high-dimensional global linear dynamic model is calculated in real time by utilizing the differentiable programming layer in the learning process, so that the manual selection process of the dimension raising mapping function with subjectivity and blindness is avoided; and the optimality of the obtained high-dimensional global linear dynamic model can be effectively ensured. Therefore, the modeling precision of the method is effectively improved.

The invention discloses an optimization design method for a residual generator of a dynamic control system in a process industry, in order to solve the problems of a small fault detection time window and low fault detection sensitivity reflected by the residuals of an existing online residual signal generator applicable to a dynamic system. The invention targets two scenarios, namely, a known and an unknown mechanism model of a controlled object, and performs a unified optimization design based on the residuals of the two scenarios. The controlled system can be modeled as a linear dynamic system and its state space model is consistent. The residual signal is written as a parameterized form of a left coprime decomposition correlation matrix with the controlled system. The residual generator is designed offline through an optimization strategy, and is driven by input and output data collected online in real time, thereby realizing online optimal residual generation. The invention is suitable for residual generation of dynamic control systems in process industries.

The application discloses a kind of inertial navigation data dynamic fusion method and system, method includes: by inertial navigation system real-time acquisition of original inertial navigation data of different positions of traverse, first data is obtained after preprocessing;The inertial navigation system data of different positions in first data is carried out time synchronization processing, to obtain time synchronization inertial navigation data;Based on the accuracy and stability of inertial navigation system historical data, each inertial navigation system is allocated dynamic weight;The state of next time step is predicted by linear dynamic system equation, when new time synchronization inertial navigation data arrives, state estimation and error covariance matrix are updated using Kalman gain, dynamic weight and new time synchronization inertial navigation data, to output the fusion data set processed by Kalman filter.This application can effectively reduce the influence of environmental noise and equipment error on monitoring result;By dynamically adjusting weight and fusion strategy, it can better adapt to various complex and changeable environmental conditions, and maintain high stability.

The invention discloses a multi-time-delay processing system stability analysis method based on an updating precise integral method. The method comprises the following steps: S1, obtaining a state space according to a multi-time-delay n-dimensional linear dynamic system equation; s2, obtaining continuous output of the multi-time-delay processing system according to the ordinary differential equation solution; s3, obtaining a discrete output expression of the multi-time-delay processing system according to the second-order Newton polynomial; s4, further modeling the discrete output expression obtained in the step S3; s5, constructing a state transition matrix irrelevant to the initial value according to the Floquet theory; s6, judging the stability of the equation according to the characteristic value of the transfer matrix phi of the system; and S7, analyzing the stability of the multi-time-delay processing system based on the updated accurate integral. According to the method, high-order discretization is adopted, so that the calculation precision is improved; and only one matrix instead of m matrixes is constructed in one period, so that a large amount of calculation time can be saved, and the processing efficiency is improved.

The present application discloses a method and device for digital twin of the required amount of fused magnesia based on edge-cloud collaboration, which relates to the field of digital twin technology for complex furnace chambers of different batches. By correcting the non-linear part in the digital twin model of the required amount, it realizes the modeling of industrial process operation indicators whose dynamic characteristics change with the production process. The method includes: generating a dynamic model of the power of a multi-furnace batch group furnace; constructing a digital twin model of the required amount of a multi-furnace batch group furnace, including a linear model and a non-linear dynamic system; constructing an offline deep learning model of the non-linear dynamic system, constructing an online deep learning correction model and a cloud deep learning correction model; obtaining specified production process data to correct the online deep learning correction model and the cloud deep learning correction model, and completing the correction of the non-linear dynamic system; obtaining the sum value of the system output of the non-linear dynamic system and the model output of the linear model as the target required amount data, and sending the target required amount data to a display terminal for display.

The invention relates to a Mama-based 3D medical image mask self-encoding method, aims to optimize and process a self-encoding task of a 3D medical image, and particularly solves the problems of high video memory demand and computing resource consumption during processing of long-sequence data. According to the method, the introduced Mama network architecture is used as a backbone network of the self-encoding model, and the capability of the model for processing long-sequence data is optimized. Mamba is an advanced architecture based on a state space model, combines the concept of a linear dynamic system theory and a neural network, and effectively captures time sequence information and dynamic characteristics in data. The method has the core advantages of a selective scanning algorithm, a hardware sensing algorithm and efficient data processing capability, and particularly has more advantages during long-sequence data processing. Therefore, the method not only relieves the burden of calculation and storage, but also keeps or improves the precision and efficiency of image processing, and is particularly suitable for processing 3D medical image data needing a large amount of calculation and high memory bandwidth.

The invention discloses an FMCW laser ranging light source spectrum degradation suppression method based on strategy online deployment. Belongs to the laser radar field. The objective of the invention is to solve the technical problems of spectrum degradation and measurement precision reduction caused by frequency modulation nonlinearity of an FMCW laser ranging light source. Comprising the following steps: building an FMCW laser nonlinear dynamic system platform, and constructing a deep learning model to simulate a dynamic environment; designing a nonlinear correction process as a Markov decision process, and training a reinforcement learning agent by using a TD3 algorithm; carrying out lightweight processing on an Actor network in the policy network, wherein the lightweight processing comprises pruning and quantization operations so as to reduce the complexity of the model; a lightweight strategy network is deployed on a hardware platform, strategy network forward reasoning is achieved on an FPGA, the strategy network is cascaded with a state extraction module and an action output module to form a closed-loop correction system, and dynamic modulation current optimization is achieved. According to the invention, frequency spectrum degradation caused by frequency modulation nonlinearity can be inhibited on line, and the measurement precision is improved.

The present application discloses a state prediction method and device based on physical system state restoration. By applying the technical solution of the present application, in the sparse identification algorithm of the nonlinear dynamic system used in restoring the physical system in a certain data recognition scenario, a method of adaptively selecting a threshold for the nonlinear dynamic system can be achieved by adding a method of constructing a correlation function B with the threshold λ as an independent variable, so that the subsequent nonlinear dynamic system configured with a self-selected threshold is used to restore the state information of a physical system within a period of time. Thus, an object of correctly restoring the physical system state without manual threshold selection is achieved. It also solves the reduction error influence caused by missing function terms that may still occur in the process of using the nonlinear dynamic system to restore the physical system in the related art even when a relatively small threshold is selected.

The present invention proposes a method for dynamic prediction of injection and production of oil and gas fields based on a linear dynamic system. The method is aimed at low permeability oil and gas fields with water injection as the main driving method. The method includes: establishing a linear dynamic system (LDS) production prediction model based on the control material balance equation of the entire reservoir of the water-driven oil and gas field; using the EM method to estimate the model parameters in the LDS production prediction model; collating and dividing the dynamic data, dividing the data set into a training set and a test set according to a certain ratio; initializing the model parameters and calculating the parameters, integrating the prediction results and calculating the relative error. The introduction of hidden variables in the LDS production prediction model in this method better characterizes the injection-production relationship in actual production, and can make a more accurate prediction of the monthly production of low permeability oil and gas reservoirs.

The application relates to the technical field of nonlinear dynamic system fault diagnosis, in particular to a nonlinear dynamic system fault diagnosis method and device based on a mask attention subspace model. The method comprises the following steps: a subspace coding-decoding model is constructed, wherein the encoder serves as a state estimator, and the decoder serves as an output estimator; a mask attention layer is introduced into the encoder, and the attention range is constrained through an inter-neighbor mask module; a trained MAS-Net model is used to calculate residual signals and T2 statistics of test samples; a fault detection threshold is determined based on kernel density estimation, so that fault detection is realized; and fault variable identification and isolation are realized by analyzing the differences in attention weight under normal and fault states. The application solves the problems that traditional subspace identification methods are difficult to extract long-distance dependent features and have poor fault variable interpretability, can effectively process high-dimensional, nonlinear and strongly coupled industrial data, and simultaneously realizes high-precision fault detection and interpretable fault isolation.