9 results about "Gradient method" patented technology

This invention aims to optimize the input sequence of actuators with significant time delays in marine vehicles (such as variable buoyancy hydraulic pumps and valves, and slow-response actuators like rudders). Under complex sea conditions and model uncertainties, this method achieves rapid and stable attainment of the target operational state with minimal action cost, while simultaneously satisfying safety and lifespan constraints. The method disclosed in this invention uses a parameter identification model as the prediction kernel, incorporating terminal task deviation, the number of execution actions and non-zero duration, and the safety volume boundary into a multi-objective cost function. It employs an improved LM algorithm with adaptive damping for sequence-level optimization and rolling updates, thereby overcoming the shortcomings of conventional closed-loop compensation and pure gradient methods in terms of slow convergence, oscillation, and difficulty in balancing safety and wear on time-delayed objects. This significantly improves convergence speed, control accuracy, and reliability, while reducing energy consumption and mechanical wear.

This application relates to the field of wireless communication technology, providing an antenna robustness design method and apparatus based on hybrid deep learning. This invention simulates manufacturing process errors by combining sampling methods within the range of process errors, obtaining design variables and... S Sensitivity analysis was performed on the Gaussian distribution curves of the statistical mapping relationship between parameter responses to identify variables more sensitive to manufacturing errors, thereby reducing the dimensionality of variables and decreasing the complexity of subsequent antenna robustness optimization. A hybrid deep learning model was used to construct an antenna response substitution model, replacing traditional electromagnetic simulation, significantly shortening the optimization cycle while ensuring the accuracy of response prediction and avoiding distortion of optimization results due to model errors. By constructing a robustness objective function and employing a genetic algorithm to optimize antenna parameter robustness, a highly robust optimal solution was found, avoiding premature convergence of the gradient method in multi-peaked environments caused by random errors.

The application provides a global sparse texture filtering method based on edge structure preservation, including introducing a texture inhibition function in a penalty term, and constraining the gradient of an output image, the texture inhibition function inhibits texture, noise and unnecessary detail information in the image by setting two threshold values, then using the inhibited gradient as the input of the denominator of the penalty term, so that the penalty term can sufficiently distinguish texture and structure; sparse regular L1 norm is used to constrain the penalty term, non-convex optimization is converted into a convex optimization problem by introducing a sub-gradient, and an alternating direction multiplier method is used for iterative solution, so that better edge preservation is achieved; sparse L p Norm is used to constrain the penalty term and a preconditioned conjugate gradient method is used to accelerate and improve the calculation efficiency, so that more robust and sparse image smoothing effect is achieved. The application can improve the robustness of the algorithm in distinguishing texture and structure, retain better semantic information, and achieve better edge structure preservation and smoothing performance.

The application relates to a time-optimal three-axis reorientation method and device for an inertia-symmetrical rigid spacecraft, which determines an angular velocity analytical expression according to inertia symmetry of the spacecraft and a time-optimal control switch structure set in advance; through introduction of a new inertia axis coordinate system, the symmetry characteristic of a reorientation track is explicitly derived; a Radau pseudo-spectrum discrete method is used to discretize a quaternion dynamic equation, so that the time-optimal reorientation problem is converted into a nonlinear programming problem; through given two parameter values, the discretized quaternion dynamic equation is converted into a linear equation set and is solved, a discrete quaternion value sequence obtained through decomposition is used, a gradient method or a Newton method is used to optimize the two parameters, and a time-optimal reorientation track is determined according to an optimization result. The method does not need to perform numerical integration or solve a large-scale nonlinear programming problem, and the calculation efficiency is significantly improved, so that fast and accurate track generation of an inertial rigid spacecraft is realized.

This application relates to a method and system for solving conjugate gradient finite element models based on sparse convolution preprocessing. The method includes establishing a structural finite element model, generating a structural stiffness matrix A and a load vector b, and constructing a linear equation system Ax=b, where A is a sparse symmetric positive definite matrix. A preprocessing sub-generator is constructed by training different structures using a sparse convolutional neural network. The stiffness matrix A is input into the preprocessing sub-generator to obtain a preprocessing factor. A symmetric positive definite preprocessor is constructed based on the preprocessing factor. The linear equation system Ax=b is solved using the preprocessed conjugate gradient method to obtain the displacement response vector x. This application optimizes the condition number of the preprocessed matrix to improve convergence speed and reduce solution time. It adapts the sparse convolutional U-net structure to large-scale sparse matrices, improving training efficiency and forming a unified and scalable preprocessing framework for structural engineering, providing a general and efficient preprocessing strategy for large-scale finite element model analysis.

This invention discloses a method and apparatus for collaborative optimization of microgrids based on event triggering and packet loss compensation. The method includes: constructing a collaborative operation model of the microgrid system using the alternating direction multiplier method; constructing an augmented Lagrangian function based on the collaborative operation model; iteratively solving the augmented Lagrangian function to obtain the scheduling scheme of each microgrid; each iteration includes: each microgrid solving its local coupling variables and comparing them with the coupling variables solved in the previous round; if the deviation exceeds a preset threshold, noise is added to the coupling variables before sending them to the scheduling center; otherwise, multiplexing symbols are sent; the scheduling center receives the data sent by each microgrid; if data packets are lost, the coupling variables are linearly extrapolated and predicted based on historical momentum; if the data is a multiplexing symbol, the coupling variables from the previous round are reused; the consistency variables and dual variables are updated based on the coupling variables of each microgrid using the Nesterov accelerated gradient method, and fed back to the microgrids that have not lost packets.

This invention discloses a variable-length set decision-making multi-objective optimization method for selecting key nodes in communication networks, relating to the field of communication technology. The method includes: acquiring data from the original communication network and abstracting it into a graph structure; for each node, constructing a feature vector and defining node attributes and a node state mask; setting multiple constraint thresholds related to node attributes; constructing a multi-objective evaluation function and a constraint violation function; constructing a policy model based on an RL-GNN according to the graph structure, training the policy model using a policy gradient method to obtain the final policy; selecting candidate nodes based on the current state and the final policy to generate variable-length candidate solutions; generating multiple candidate solutions, constructing an initial population, and combining this with the establishment of an external elite database, using the multi-objective evaluation function and constraint violation function for evolutionary iteration; obtaining and outputting the non-dominated solution set from the external elite database, providing multiple options for engineering decision-making.

This invention discloses a B-spline curve fitting method for image edge-finding, comprising: extracting a set of points to be fitted from the image and initializing the set; constructing a basis function matrix of the B-spline curve based on the initialization result; calculating a normal matrix based on the basis function matrix, preprocessing the normal matrix to extract corresponding node feature vectors and edge feature vectors; inputting the node feature vectors and edge feature vectors into a graph neural network to construct a precondition submatrix; iteratively updating the control points using the preprocessed conjugate gradient method based on the precondition submatrix; and generating the final required B-spline curve based on the basis functions and the control points obtained when the iteration stops. This invention uses a precondition submatrix constructed by a graph neural network to adjust the update vectors of the B-spline curve control points, reducing the number of iterations and improving the fitting speed of the B-spline curve in image edge-finding.

The application relates to a train precise parking control method based on deep reinforcement learning. ATO parking data is collected and preprocessed to obtain an expert data set X. Behavior cloning imitation learning is performed based on the expert data set X to initialize a policy network of a train. A deep determination policy gradient method is used to train the policy network online to obtain a deep optimization network. The deep optimization network is output and saved, and is used for train precise parking control. The following technical effects can be achieved: the ATO historical data of an existing line is fully utilized to perform offline imitation learning; the coupling relationship between PI control parameters and vehicle characteristics is decoupled by relying on the generalization ability of a neural network; the technology can reduce the online debugging time of ATO software and improve parking precision; the policy network and the value network of reinforcement learning are adjusted according to the change of train characteristics, and the effect of lifelong learning is realized.