114 results about "Linear approximation" patented technology

The invention relates to a vehicle and unmanned aerial vehicle combined scheduling method for wide-range low-cost inspection. The method comprises the following steps: acquiring prior information; modeling the unmanned aerial vehicle inspection problem of each target area according to the prior information to obtain a mixed integer non-convex optimization problem with the goal of minimizing the weighted sum of the total execution time and the energy consumption of all the inspection unmanned aerial vehicles; performing linearization on a non-convex bilinear term in the mixed integer non-convex optimization problem, and performing discretization processing on a nonlinear function by adopting piecewise linear approximation; an approximate mixed integer linear programming problem is obtained and solved, and an unmanned aerial vehicle scheduling strategy is obtained; modeling according to the unmanned aerial vehicle scheduling strategy and the prior information to obtain an inspection vehicle path planning problem taking the comprehensive driving cost as a target; the routing inspection vehicle path planning problem is converted and modeled into a Markov decision process, a routing inspection vehicle is used as an intelligent agent, a state, an action and a reward function are defined, and a routing inspection vehicle scheduling strategy is obtained. Therefore, combined inspection of the inspection vehicle and the unmanned aerial vehicle is realized, and the inspection range is expanded.

The invention discloses a rope-driven mechanical arm motion control method based on model predictive control, which comprises the following steps: taking a joint angle error as a state variable, taking a rope-driven rotation angle as an input variable, and establishing a nonlinear state-space equation of a rope-driven mechanical arm according to the state variable and the input variable; the nonlinear state-space equation is subjected to linearization processing, a linear approximation model is generated, and the linear approximation model comprises a linearization matrix; defining a prediction interval, expanding the linear approximation model into a multi-step prediction form, and generating a prediction model; constructing a cost function, and combining the prediction model into the cost function to solve and obtain an optimal control sequence; and extracting the optimal control increment at the current moment from the optimal control sequence, and updating the state variable, the input variable and the linearization matrix according to the optimal control increment to form closed-loop control. Accurate path tracking control can be carried out on the rope-driven mechanical arm under complex dynamic constraints, and it is ensured that the mechanical arm moves according to the planned trajectory.

The invention discloses a squeeze film damper active control method based on RBF neural network PID optimization. The method comprises the following steps: acquiring state parameters of a rotor system through a data acquisition module; inputting the state parameters into an RBF neural network, and performing online optimization on PID control parameters by using the RBF neural network; based on the optimized PID parameters, a piezoelectric ceramic actuator is driven through a piezoelectric ceramic controller, the damping force of the extrusion oil film damper is adjusted, and active control over vibration of the rotor system is achieved. According to the method, RBF neural network nonlinear approximation and PID control stability are fused, the hysteresis and locking problems of traditional passive control are solved, active suppression of rotor system vibration is achieved, and the control precision and reliability are greatly improved.

The invention discloses a real-time solution reconstruction planning method for a man-machine interaction rope traction parallel robot, which belongs to the field of robot configuration planning, and comprises the following steps of: 1, constructing kinematics and dynamics models according to the position relationship between a rope leading-out point and a movable platform, and establishing an admittance model; 2, according to the dynamics and admittance model, a hyperplane movement method is used for representing a force feasible condition, and an optimization problem objective function is set according to the force feasible condition; 3, expressing a rope leading-out point solving problem as a nonlinear optimization problem through constraint, approximating the nonlinear optimization problem as a linear optimization problem through linear approximation, and solving an approximate optimal solution of the linear optimization problem; and 4, an artificial potential field is set according to dynamic characteristics, the approximate optimal solution is corrected, the solved rope leading-out point position is not located at the boundary of a solution space while the solving speed is guaranteed, and reconstruction planning of real-time solving of the robot rope leading-out point is completed. According to the method, the solving instantaneity and the man-machine interaction stability can be ensured.

This invention proposes a GQA-LUT method, utilizing a genetic algorithm and LUT-based circuit to efficiently approximate non-linear operators in Transformers. It adaptively finds optimal solutions for various non-linear functions, outperforming conventional neural network methods. A novel rounding mutation (RM) algorithm enhances approximation accuracy during quantization, improving low-bit integer precision. The invention also introduces a LayerNorm folding strategy as a near-memory computing principle, reducing IO and energy overheads with a two-stage memory hierarchy. Additionally, an additive partial sum quantization method is proposed to reduce energy consumption by quantizing accumulated PSUMs in matrix multiplication, alongside a PSQ-APSQ grouping strategy and floating-point regularization.

The invention provides a GQA-LUT method. According to the GQA-LUT method, a non-linear operator in a converter is effectively approximated by utilizing a genetic algorithm and an LUT-based circuit. The GQA-LUT method can adaptively find optimal solutions of various nonlinear functions, and is superior to a conventional neural network method. A novel rounding variation (RM) algorithm enhances approximation accuracy during quantization, thereby improving low order integer precision. In the invention, a LayerNorm folding strategy is also introduced as a near memory calculation principle, so that IO and energy overhead of the hierarchical structure of the two-stage memory is reduced. Furthermore, an additive partial and quantization method is proposed to reduce energy consumption by quantizing accumulated PSUM in matrix multiplication as well as a PSQ-APSQ grouping policy and floating point regularization.

This specification pertains to the field of refining and chemical production processing technology, and particularly relates to a method and apparatus for optimizing refining and chemical production plans for a target secondary processing unit. The method includes: obtaining parameters of the secondary processing unit under a refining and chemical production optimization scenario; constructing a nonlinear constraint equation for product output based on the parameters of the secondary processing unit; performing an equivalent transformation on the nonlinear constraint equation to obtain an equivalent transformation equation; extracting the nonlinear constraint terms and performing a Taylor expansion; transforming the Taylor expansion formula into a linear approximation equation based on the proportion of the feed material to the output material of the blending tank, the feed material quantity to the blending tank, and the feed property values ​​of the blending tank, according to the property balance constraints of the blending tank; and then substituting these into the nonlinear constraint equation for product output, thereby transforming the nonlinear constraint equation for product output into a linear approximation equation containing only material information variables. This method effectively ensures model convergence and shortens the solution time.

A computer-implemented method, computer program product, and computer system for automated model predictive control. The computer system trains multiple step look-ahead regression models, using historical states and historical actions for a to-be-optimized system, for each timestep of a past time horizon. Regression models may be either linear or nonlinear in order to capture process dynamics and nonlinearity. The computer system generates optimization constraints for each timestep of a future time horizon. The computer system generates optimization variables, based on the multiple step look-ahead regression models, for each timestep of the future time horizon. The computer system constructs a mixed integer linear programming based optimization model that includes an objective function, the optimization constraints, and the optimization variables. Nonlinear regression models are converted into piecewise linear approximation functions. The computer system solves the optimization model to produce actions for the to-be-optimized system, over the future time horizon, and recommend commitment-look-ahead actions.

The invention discloses a reservoir group medium-and-long-term power generation optimization scheduling method based on gallery iteration static optimization, and the method comprises the steps: firstly calibrating convex hull parameters based on a planar convex hull linear approximation method, constructing a scheduling model, and solving an initial solution (including the storage capacity and the output flow of each time period); dividing the whole output feasible region into small-scale regions in a high-density manner, and respectively calibrating convex hull parameters of each region; comparing the initial solution with the small-scale region boundary period by period, determining the region to which the initial solution belongs, and updating the convex hull parameter of the corresponding period; constructing a time period corridor by taking the initial solution as a reference, and constructing a new model in combination with the updated parameters to solve a current iterative solution; and if the target function value change is smaller than the threshold value, converging and outputting the optimal solution, otherwise, updating the scale parameter and repeating iteration. According to the method, the optimal solution is approximated through static calibration-parameter selection, linear errors are corrected, efficient solution can be achieved, operation of the cascade hydropower stations can be guided, and coordinated utilization of water resources is optimized.

The present application discloses a method, device, terminal and medium for constructing a digital twin simulation model of a power device. The solution provided by the present application first constructs a geometric model of the target power device, then establishes multiple physical field models related to the target power device, determines the type of coupling relationship between different physical field models according to the field quantity change relationship between different physical field models, and then constructs a physical field coupling relational expression according to different coupling relationship types to achieve the purpose of coupling multiple physical fields. Then, through the simulation test method, the material change characteristic data of the target power device is obtained to construct the material characteristic function of the target power device, and the material characteristic function is subjected to Taylor expansion and linear approximation processing to complete the material characteristic mapping, so as to obtain the digital twin simulation model of the target power device, realize the lightweight reconstruction of the power device in the multi-physical field coupling scenario, and solve the technical problem of low modeling efficiency of the existing digital twin model of the power device.

The application discloses a wireless resource allocation method for a LEO satellite-ground fusion uplink communication system, which comprises user base station selection, power and carrier allocation and power and carrier allocation methods for base station to satellite link, first establishes a system energy efficiency maximization model for subcarrier and power allocation; then quotes a slack variable to convert the model into a target function lower bound model; uses a symbolic function to couple power allocation variables and carrier allocation variables; and further decomposes the target function lower bound model into a base station access and carrier allocation optimization model and a power allocation optimization model; linear approximation method and successive convex approximation method are respectively used for solving, so that the wireless resource allocation of the uplink communication system is realized. By using the method, 0-1 integer optimization variables are removed, the calculation complexity is low, and higher uplink system energy efficiency is obtained.

The invention discloses a linear approximation attention operator acceleration method based on NPU, and the method combines linear approximation attention with an Ascend C programming language, and achieves an Ascend C operator for calculating the linear approximation attention. According to hardware characteristics of the mercuration NPU, optimization of vector calculation is carried out, the number of instructions used for line-by-line summation and division is reduced, and time consumption is reduced; according to hardware characteristics of the mercuric chloride NPU, pipeline parallel optimization is carried out, large matrix multiplication is decomposed into multiple small matrix multiplication and asynchronous matrix multiplication, execution of matrix multiplication and vector calculation tasks can be coordinated more efficiently, and the utilization rate of hardware resources is maximized; cache occupation is optimized according to hardware characteristics of the mercuric chloride NPU, and the number of blocks of matrixes such as Q, K, P, Z and the like can be reduced under some input scales, so that the number of cycles of calculation and carrying is reduced, and the operation time of operators is shortened.

The present application relates to the technical field of power system, in order to solve the problem that the current reactive power flow calculation result cannot be applied to the actual operation system, provide a kind of power system optimal reactive power flow calculation method and system.Power system optimal reactive power flow calculation method includes with the active power network loss in power system as minimum to construct objective function, then combine linearized power flow equation, the reactive power output limit value corresponding to synchronous generator in power system and the linearization approximation constraint converted by the constraint of the internal potential and output current of inverter, construct optimal reactive power flow model and solve to obtain the new operation mode of power system;After obtaining the new operation mode of power system, recalculate the power flow distribution of power system, obtain the active power network loss of power system under new operation mode, in the process of successive linear approximation, the active power network loss is continuously reduced.It can obtain the optimal reactive power flow calculation result, guarantee the stability of power system.

The application discloses a corona discharge load state detection method based on an oscillation circuit voltage ratio relationship and relates to the technical field of corona discharge detection. An ideal Royer circuit is constructed; a peak value protection circuit is used for collecting a peak voltage Vm; a direct current input voltage of a sine wave self-excitation oscillation boost circuit is Vin; a ratio relationship of Vm and Vin in the ideal Royer circuit under no-load is calculated to obtain a first relationship formula; a ratio relationship of Vm and Vin in the ideal Royer circuit under load is calculated to obtain a second relationship formula; first-order linear approximation fitting is performed on the first relationship formula and the second relationship formula to obtain a target relationship formula; and the corona discharge load state is determined according to the target relationship formula. Through establishing a quantitative relationship between the ratio of the peak voltage of the primary side of the Royer circuit which is easy to measure and the input voltage and the secondary direct current high voltage and the load current reflecting the needle tip state, safe, quantitative and adaptive contamination detection is realized.

The application discloses a kind of based on ST-GAN's multi-source data dynamic risk early warning method and system, the method includes the following steps: S1. constructing the spatiotemporal generation confrontation network suitable for spatiotemporal data characteristics, utilize the network to generate extreme precipitation data, realize spatiotemporal unbalanced data reduction, obtain the equalization data of final output;S2. the equalization data of final output with high-dimensional space variable is carried out multi-source data fusion, and nonlinear spatiotemporal information conversion equation, and by local linearization obtains linear approximation model, to predict future time series;S3. under the present situation that extreme rainfall data amount is relatively insufficient, neural network based on dual learning theory accurately learns the parameter of nonlinear spatiotemporal conversion, estimates extreme weather event.The application is through spatiotemporal generation confrontation network (ST-GAN) and multi-source data fusion engine, significantly improves the precision and efficiency of natural disaster warning.

The invention belongs to the technical field of micro-grid optimization scheduling, provides a micro-grid model prediction control method and system based on bilateral opportunity constraint, and aims at solving the problem of micro-grid model prediction control optimization operation control by combining historical data of renewable energy prediction errors and updated data in real-time operation. A non-parametric Bayesian model and an incremental Gaussian mixture model are respectively adopted to establish and update a Gaussian mixture model of a renewable energy prediction error, and bilateral chance constraints of a storage battery energy storage level are introduced in a micro-grid model prediction control optimization operation process, so that safe operation of a micro-grid is ensured. A piecewise linear approximation method and an algorithm for accelerating the solving process are provided, so that the opportunity constraint can be efficiently solved by using a commercial solver; compared with the prior art, the method has the advantages that the calculation efficiency and the operation economy and conservative property are well balanced.

This invention claims protection for a CNN-LSTM algorithm neural network accelerator based on FPGA. The CNN hardware implementation includes a data input line buffer module, a convolution calculation module, a ReLU activation function module, an intermediate result buffer module, and a pooling calculation module. The LSTM hardware implementation includes an LSTM control module, a gate function calculation module, and a sigmoid activation function linear approximation module. The FC hardware implementation includes an FC control module, a fully connected layer calculation module, a ReLU activation function module, and a data output buffer. The purpose of this invention is to design a high-performance, low-power, and highly flexible CNN-LSTM neural network accelerator tailored to specific application scenarios. The innovation lies in the fact that, compared to traditional neural network accelerators, this invention uses a parallel pipelined design method to implement a CNN-LSTM algorithm neural network accelerator, which significantly improves the low power consumption and data throughput of the neural network accelerator. Furthermore, the parallel processing capabilities of the FPGA enable the algorithm to run at a faster speed.

The invention relates to a computer-implemented method for creating a model for detecting an anomaly or an operating state or a change in an operating state in a technical system (1), comprising the following steps: - Providing (S1) time series of operational variables, each indicating a temporal progression of an operational variable, as training data; - Performing (S2) a piecewise linear approximation on the time series of the operating variables to obtain segments of concatenated linear functions for each of the operating variables, obtaining change points (T) as time points between the segments of all operating variables, and summarizing the change points (T) of all operating variables to define state segments between two temporally successive state segments that define time intervals of the time series of the operating variables; - Determining (S3) characteristics of the time-dependent behavior of the operating variables for each of the state segments in order to form a state vector for each (S4); - Training (S6) a classification model based on the state vectors; - Implement (S7) the classification model in a control unit of the technical system (1).

The invention discloses a multi-stage pipe network leakage positioning method and system based on a hydraulic model sensitivity matrix and sparse inversion, and relates to the field of monitoring and leakage detection of an urban water supply pipe network, and the method comprises the steps: S1, linearizing a pipe network continuity equation and an energy equation at a normal operation point of a baseline working condition, obtaining a sensitivity matrix of the leakage rate and the demand disturbance to the pressure residual error, and establishing a linear approximation relation; s2, constructing an optimization problem with L1 regularization based on the relationship; eliminating a demand disturbance item, converting into a non-negative LASSO problem, and completing suspicious node positioning; s3, convex combination approximation of sensitivity of nodes at the two ends is introduced for the suspicious pipe section, an optimization target is constructed, and the internal leakage continuous position and the leakage amount of the pipe section are estimated; and S4, a joint optimization problem is constructed based on the pressure residual data of the multiple time periods, consistent selection of leakage positions of the multiple time periods is kept, and final leakage positioning is realized. According to the method, pipe network leakage positioning is accurately and efficiently completed through multi-stage progressive positioning.

To efficiently evaluate an error of a quantum gate.SOLUTION: Acquiring measurement data indicating an execution result of a quantum computer for a quantum circuit in which a quantum gate array including a first quantum gate is repeated N times (N is an integer of 1 or more); determining, from a first matrix indicating an ideal value of the quantum gate array, a period k (k is an integer of 1 or more) at which the first matrix raised to the power of k becomes a unit matrix; Processing for generating a linear approximation function for linearly approximating the influence of an error on measurement data by approximating the repetition of a quantum gate array and processing for estimating the error by using the linear approximation function and the measurement data are executed.SELECTED DRAWING: Figure 1

The invention discloses an elevator energy consumption control system based on multi-objective optimization, and the system comprises a parameterization module which generates a self-adaptive non-uniform B-spline control point set and a corresponding non-uniform node vector; the elevator discrete nonlinear dynamics module is used for constructing an elevator discrete nonlinear dynamics model; the index function construction module is used for constructing an elevator finite time domain nonlinear optimal control performance index function; the iteration module constructs a linear approximation system; the Newton-GMRES module is improved, and the optimal correction amount of the adaptive non-uniform B-spline control point set is obtained; the updating module is used for generating an updated self-adaptive non-uniform B-spline track; the motor drives the signal generation module to drive the elevator to run along the updated self-adaptive non-uniform B spline track; and the circulation module is used for repeatedly executing the steps from the index function construction module to the motor driving signal generation module. According to the method, the speed and precision of track generation and adaptive adjustment are remarkably improved.

The application provides a secant approximation method for nonlinear constraints of a redundant drive system, and belongs to the technical field of dynamic control distribution of the redundant drive system. The method comprises the following steps: firstly, according to the control input model of the redundant drive system with any pair of constraint components being nonlinear constraints, a closed region formed by the intersection of a rectangle and an ellipse in a geometric plane is obtained; then, after the closed region is divided into the union of the rectangle and the ellipse triangle, the approximation result of the closed region is obtained by performing the approximation of the rectangle and the triangle combination on the ellipse triangle, so that the linear approximation of the pair of nonlinear constraint components is realized. The application jointly uses the triangle and the rectangle to perform the approximation on the region surrounded by the nonlinear constraints, converts the nonlinear constraints into multiple linear constraints, converts the nonlinear constraint set into a linear constraint set, effectively solves the problem that the control reachable set cannot be determined due to the nonlinear constraint relationship between the actuators in the redundant drive system, and is helpful to realize the real-time control of the redundant drive system.

A battery diagnosis apparatus includes a voltage sensor to generate a voltage signal indicating a battery voltage of a battery, a current sensor to generate a current signal indicating a battery current of the battery and a control circuit. The control circuit determines a capacity curve indicating a relationship between the battery voltage and a charge capacity in a set voltage range based on the voltage signal and the current signal collected at each unit time for a constant current charging period. The control circuit determines a differential curve indicating a relationship between the battery voltage and a differential capacity in the set voltage range based on the capacity curve. The control circuit determines an approximate straight line of the differential curve using a linear approximation algorithm, and determines whether lithium deposition is present in the battery based on the approximate straight line.

The invention discloses a pluggable power supply module regulation and control and network port power supply optimization method based on deep learning, and the method comprises the following steps: S1, collecting power parameters of a power supply module and a port, building a bipartite graph topology, and forming historical data; s2, bus voltage ripples are injected and decoded, a sideband signaling channel is established, and a feature sequence is generated through alignment; s3, performing time sequence prediction on the feature sequence to form a prediction result; s4, Koopman dimension raising is executed, a linear approximation model is generated, and a control strategy is calculated; s5, executing a control strategy to carry out output adjustment and port allocation, and issuing a signaling to implement current limiting; and S6, voltage, current and temperature are collected, deviation comparison is carried out, a condition maintenance strategy is met, and otherwise, protection is triggered and write-back adjustment is carried out. According to the invention, through deep learning prediction, Koopman dimension raising control and port energy mutual assistance, intelligent regulation and control of the pluggable power supply module and network port power supply optimization are realized.

The application provides a trusted federated learning system and method based on a blockchain and zero-knowledge proof, comprising a disturbance module.In the disturbance module, a non-interactive zero-knowledge protocol is applied, so that the noise added to the local model can be verified. The trainer and the verifier independently and respectively use linear approximation to construct the same circuit containing specific Gaussian noise, and the trainer uses the circuit and a public generator and model parameters to construct a proof. The verifier uses their circuit to verify the received proof; a verification module. The verification module constructs a sharded blockchain consensus process. Each verifier first verifies the accuracy of the local model gradient update using a public data set. The verification result is used as a voting summary to form a consensus in the network. For block data containing local model selection information, a Byzantine fault tolerance protocol is introduced to ensure data credibility. The application can be proved to achieve differential privacy and can analyze the upper bound of the influence of noise on the model performance.

The present invention addresses the problem of calculating the root of a power function. An arithmetic device according to the present invention calculates a root Y of a power function (Y = AX, wherein A is a constant, and X is a prescribed numerical range). The arithmetic device comprises: an input unit into which input data indicating an exponent X is input; a first arithmetic circuit that executes a first arithmetic process on the basis of a linear approximation equation of a graph of a power function in a first numerical region of a numerical range of the exponent X; a second arithmetic circuit that executes a second arithmetic process on the basis of a linear approximation equation of a graph of a power function in a second numerical region of the numerical range, the second numerical region differing from the first numerical region; and a selector that outputs output data indicating the root Y calculated by the first arithmetic circuit when if exponent X indicated by the input data inputted into the input unit belongs to the first numerical region, and outputs output data indicating the root Y calculated by the second arithmetic circuit if the exponent X indicated by the input data inputted into the input unit belongs to the second numerical region.

A control model generation device includes: an observation value acquisition unit to acquire a plurality of observation values that are outputs of a control target having non-linear characteristics; a state space model estimation unit to estimate a state space model that expresses a linear approximate curve related to the plurality of observation values acquired; and an upper bound model estimation unit to calculate an estimation error, and estimate an upper bound model that expresses an upper bound of the estimation error, the estimation error being an error between each of the observation values acquired and the linear approximate curve expressed by the state space model estimated. Furthermore, the control model generation device includes a control model generation unit to generate a control model that expresses an equation of motion of the control target using the state space model estimated and the upper bound model estimated.

The invention discloses a convolutional neural network adaptive structured pruning method for end-side equipment image classification, a storage medium and equipment, and belongs to the technical field of image processing. The method mainly comprises the following steps: based on a sparse linear approximation theory, constructing an index-oriented sparse index; iteratively executing multiple rounds of unstructured pruning and recovery training on the pre-training model; and after iteration is finished, redundant parameters and invalid convolution kernels in the current image classification model are scanned, the invalid convolution kernels are physically deleted from the memory, a computational graph of the image classification model is reconstructed, and structured pruning and network reconstruction are completed. According to the method, the importance of each convolutional layer in image feature extraction is automatically evaluated by constructing an index-oriented sparse index, the number of reserved weight parameters of each layer is dynamically calculated, unstructured rarefaction is gradually realized by adopting an iterative'pruning-recovery training 'mechanism, the image classification precision can be kept under a high compression ratio, and the image classification efficiency is improved. Full-adaptive pruning is realized, and the method is suitable for resource-limited scenes.

This disclosure describes systems, methods, and apparatus for an adaptive engine with a bifurcated nonlinear model. The adaptive controller uses a nonlinear model having a control portion and an estimation portion, wherein the estimation portion uses a time-varying linear system to approximate nonlinear behavior of the system. Further, the time-varying linear system receives a structure of the underlying matrices for every frame of control samples allowing the time-varying linear system to model large nonlinearities and to pre-process this linear approximation for each frame. At the same time, the time-varying linear system also uses estimated model parameter tensors in the underlying matrices that are updated or adapted every control cycle, in real-time, throughout a frame, such that the linear approximation is also able to approximate small nonlinearities in the system. This bifurcation of a linearized model provides a faster and more robust adaptive controller.

A state-observed and updated model predictive control strategy is proposed for image-based visual servoing (IBVS) of micro air vehicles (MAVs). The proposed strategy can accurately adjust the pose of MAVs without global positioning system (GPS). Specifically, image features are defined on a virtual image plane to decouple the translational motion of MAVs. Then, a linear velocity observer is developed to provide high-quality linear velocity information for MAVs in real time. The image dynamics on the virtual image plane are linearized by first-order Taylor expansion, and a controller is constructed based on model predictive control to efficiently solve the optimal control input. Furthermore, the state input of the controller is updated at each control period to eliminate the cumulative error of the rolling optimization on the linear approximation dynamics, ensuring the accuracy of IBVS. Experimental results demonstrate the performance of the proposed observer and control strategy.