52 results about "Rapid convergence" patented technology

This invention proposes a fast convergence path planning method suitable for robotic arms in complex environments. First, in collision detection, an obstacle expansion method is used to treat expanded obstacles as collision detection obstacles, preventing unnecessary potential collisions between the algorithm's execution unit and obstacles during physical experiments. Second, a better constraint sampling method is adopted to further narrow the sampling interval and sample within the restricted interval. Finally, a sampling point optimization strategy is introduced, designing a cost function and performing weight adaptation. The cost of multiple sampled candidate points is calculated, and the optimal point is selected as the final sampling point. This invention solves the problem of low path generation quality in complex environments for robotic arms.

This paper presents a graph neural network node classification method that integrates meta-learning and mini-batch training, belonging to the field of information technology. First, the method utilizes the METIS algorithm to divide the original large-scale graph data into multiple non-overlapping connected subgraphs. Then, by constructing a hybrid selection mechanism based on label coverage and label entropy, subgraphs with high information content and strong representativeness are selected from the subgraph pool as the meta-learning task. Subsequently, iterative training is performed on the selected subgraphs using the meta-learning framework to capture the general prior features of the graph structure, thereby obtaining a set of initial parameters for the model with rapid adaptability. Finally, these optimized initial parameters are transferred to the mini-batch training stage on the full dataset, guiding the model to achieve rapid convergence through high-quality initialization. On large-scale benchmark datasets, this method significantly reduces the number of iterations required by the model while maintaining the same classification accuracy as current mainstream graph neural network models, thus greatly shortening the overall training time.

This invention relates to the field of unmanned surface vessel (USV) path tracking control technology, specifically disclosing a method, computer equipment, and storage medium for path tracking of USVs based on prioritizing experience replay optimization of near-end policy. The method includes setting up a prioritizing experience replay memory pool, an action space, a state space, and a reward function; constructing a network model based on a near-end policy optimization algorithm, and training the network model using samples from the prioritizing experience replay memory pool; generating current state information based on images acquired by the USV, the current position information of the USV, and the position information of matching points on the reference trajectory; and prioritizing the training of samples with poor action performance based on priority sampling data. This invention employs a prioritizing experience replay memory pool to ensure rapid model convergence, and the network model is constructed based on near-end policy optimization. The use of a state space and reward function tailored to environmental conditions improves the algorithm's adaptability and stability to the environment.

The present disclosure discloses a method for reconstructing the working face of coal mine based on search space pruning and multi-population dynamic adjustment, which involves the technical field of working face reconstruction of coal mine under incomplete projection conditions, comprising: based on prior knowledge of the working face reconstruction model structure, an initial grid partitioning of the exploration area is performed. Subsequently, multi-scale grid partitioning is achieved through search space pruning, guided by the sum of ray intercepts within each grid unit; to meet the requirements of population diversity and rapid convergence in the multi-population genetic algorithm, a multi-scale reconstruction objective function is constructed based on the partitioned multi-scale grids, this objective function is then solved using a dynamic multi-population genetic algorithm. Therefore, the above-mentioned method for reconstructing the working face of coal mine based on search space pruning and multi-population dynamic adjustment may dynamically adjust and optimize the number of populations, to ensure that the diversity of populations may be maintained, and the search efficiency may be improved, thereby improving the stability and convergence speed of the working face reconstruction operation.

This invention discloses a hybrid power allocation method for an IRS-assisted VLC downlink NOMA system, belonging to the technical field of power allocation methods. Addressing the problems of easy occlusion of line-of-sight links, low efficiency of fixed power allocation, and high complexity of pure metaphysical heuristic algorithms in existing technologies, this invention jointly optimizes the reflection angle of intelligent reflective surfaces and the three-dimensional position of the UAV to construct a composite line-of-sight and non-line-of-sight communication link, enhancing anti-occlusion capabilities. Simultaneously, it proposes a modified fixed power allocation method, introducing a dynamic allocation factor based on user channel gain, achieving adaptive power allocation with the weakest user as the benchmark, improving fairness and efficiency. Furthermore, it integrates the modified fixed power allocation with the Ocean Predator algorithm to form a hybrid power allocation framework. This framework reduces the search space with low-complexity structured allocation, and then uses the Ocean Predator algorithm to globally and jointly optimize the UAV position, reflection angle, and dynamic allocation factor, balancing high system summation rate and fast convergence performance.

The application discloses a kind of multi-micronet robust game optimization scheduling method and system based on dynamic auction official algorithm, and relates to energy system scheduling and optimization technical field.The method constructs system physical model and double-layer robust optimization model based on Stackelberg master-slave game, upper layer maximizes system operator's profit to formulate price signal, lower layer maximizes the worst scenario income of multi-micronet alliance to optimize resource scheduling, combined with multiple constraint conditions, Stackelberg equilibrium is solved iteratively using dynamic auction official algorithm.The application converges to global optimal solution quickly through three-dimensional hybrid driving price updating mechanism, realizes system economic benefit maximization, operation robust and reliable, and has good explainability and practical application value.

The application relates to the technical field of flight pod control, and discloses a flight pod pulse power source adaptive charging method based on a heterogeneous architecture, which comprises the following steps: synchronously collecting 6-way sensor data and performing parallel filtering; performing main-backup redundancy weighted fusion and fault determination on key sensors; generating LLC PWM driving and performing feedback voltage DMA sampling and 3-frame anti-shake robust estimation; performing MLP calibration on pretreated sensor data to quantize a 16-bit fixed-point model, realize batch production error self-compensation, and replace original data with the calibrated result; selecting parameter fine-tuning actions according to environmental states and energy states, updating a state-action value table according to a reward function and an epsilon-greedy strategy, and realizing 5-10 rounds of rapid convergence to near-optimal parameters. The application provides a charging control method which has high-speed processing capability, strong robustness and self-adaptive capability, so as to solve the bottleneck problem of existing MCU and DSP schemes in the application of high-power microwave flight pods.

This invention discloses a method for glioma identification with flexible access control and privacy protection, belonging to the field of medical artificial intelligence and privacy computing. The method includes: secure initialization and fine-grained access control configuration; local training and dynamic parameter updates through bidirectional comparative learning on each client, and selection of the optimal model as a reference source based on a dynamic validation set to achieve cross-site parameter alignment; encryption and signing of uploaded parameters, and secure aggregation on the server side to complete global model iterative updates. The fine-grained access control employs attribute-based encryption based on complex attribute predicates to achieve precise authorized access to model parameters, combines symmetric and homomorphic encryption to protect privacy during transmission and aggregation, and provides verifiability through digital signatures. This invention can achieve stable and rapid convergence in both IID and non-IID scenarios, improving classification accuracy and efficiency, while effectively reducing the risk of leakage in parameter interaction and storage.

Disclosed in the present invention is a feedback-free transmission and dynamic resource allocation method for a fully-decoupled space-air-ground integrated network, which provides a solution for resource management of space-air-ground multi-node cooperation. The method comprises: on the basis of a deep learning-based channel state information prediction model, replacing a traditional feedback signal by means of user geographic location information, so as to implement multiple-input multiple-output transmission; and designing a many-to-one matching model for enabling adaptive allocation of spectrum resource blocks between space-air-ground heterogeneous nodes, so as to ensure matching stability and have the characteristics of low complexity and fast convergence. For cooperative transmission of multiple nodes and resource allocation between the heterogeneous nodes, the spectrum efficiency of a system is improved by means of feedback-free channel state information prediction and a resource allocation mechanism. Compared with a single-connection mode and polling-based resource scheduling in a traditional network, the flexible resource allocation algorithm of the present invention significantly increases the network capacity and improves user communication quality.

This invention relates to a time-coordinated distributed control execution method and system for vehicular Ethernet. The method includes: establishing a unified global time base, initializing the theoretical transmission time of control messages and the historical delay parameters of each control group, and determining the calibration transmission time of control messages accordingly; scheduling control messages based on the calibration transmission time, calculating the actual execution time of control commands, filling it into the control information field of the control message, and sending it to the controlled node; the controlled node executes the control commands according to the actual execution time, records the actual reception timestamp of the control message, calculates the corresponding end-to-end delay, fills it into the feedback information field of the feedback message, and sends it to the control node; the control node updates the corresponding parameters according to the end-to-end delay and stores them. Compared with the prior art, this invention has the advantages of achieving unified coordination of the control execution timing of multiple controlled nodes, controllable constraints, and fast convergence.

This application provides a method, apparatus, device, and storage medium for optimizing UAV path planning parameters, belonging to the field of UAV applications. By defining UAV operational constraints and basic algorithm parameters, an optimization framework for path planning is established. Then, the core parameters of the ant colony are encoded to construct chromosomes, achieving a standardized expression of the optimization object. In the genetic iteration, the weights of energy consumption and convergence speed are dynamically adjusted in stages: initially focusing on rapid convergence to explore the solution space, and later focusing on low energy consumption for precise development, thus balancing the contradiction between global search and local optimization. Offspring solutions are generated by combining elite retention, crossover mutation, and other operations, and fitness is calculated in real time through path verification to ensure that the optimization results closely match the actual scenario. Finally, the optimal parameter combination is output and directly applied to UAV path planning, significantly improving optimization efficiency and path quality, and solving the problems of local optima and low efficiency caused by fixed weights in traditional methods.

The application discloses a power distribution communication network intelligent computing scheduling method and system based on a perception-computation integration, through constructing a perception-computation integration system, unified modeling and intelligent collaboration of three types of resources of perception, communication and computation are realized, cloud, edge and terminal computing resources can be integrated, and overall utilization is improved; based on the perception-computation integration system, a unit allocation cost of task execution on a node is defined according to communication delay, computation delay, energy consumption and task migration cost, a first objective function of minimizing a global weighted cost is established, accurate resource matching can be realized, and the allocation inefficiency problem is solved; a teacher-student joint learning combined with a reinforcement learning method is used, network dynamic changes can be adaptively adapted, fast convergence can be realized on a resource-limited edge computing platform, communication fluctuations, task bursts and other scenes can be coped with, real-time adjustment of the computing scheduling strategy is ensured, system response speed and robustness are improved, computing power is efficiently and timely allocated, and computing power utilization is improved.

This invention discloses a method and apparatus for optimizing the feeding frequency of particulate friction stir additive manufacturing. First, feeding experiments are conducted under different operating conditions to collect pressure and displacement data. A mechanism-empirical hybrid model of the dynamic coupling relationship between pressure, displacement, and feeding frequency is established. An uncertainty set is defined. A robust optimization model for the feeding frequency is established. The dual objective is transformed into a single objective through weighted summation. The inner worst-case scenario is solved. The outer optimization employs adsorption kinetics-enhanced gradient descent. This invention uses an improved gradient descent algorithm as the core optimizer, requiring only one calculation of the inner worst-case scenario and gradient per generation. The computational load is significantly less than that of population-based algorithms, making it suitable for embedding in real-time control systems for online optimization. Combined with adaptive learning rate, dynamic diffusion intensity, and desorption probability adjustment mechanisms, the algorithm achieves rapid convergence while maintaining global search capability, providing an efficient and feasible technical solution for online dynamic tuning of the feeding frequency.

The application relates to a clock calibration method, which is applied to a client, periodically sends time requests to a standard time server for several times according to a preset period, acquires several groups of time samples in the preset period, comprehensively analyzes the time samples to obtain an effective clock offset in the current round, and adopts different correction operations with different correction amplitudes according to the absolute value of the effective clock offset in the current round, quickly corrects when the offset is large, and carefully corrects when the offset is small, so that frequent correction caused by network fluctuation is avoided, and the fast convergence and operation stability of clock calibration are realized.

This invention discloses a modeling method for grid-type multi-converter based on sub-microsecond simulation, belonging to the field of converter modeling and simulation technology. The method involves discretizing the switching elements to construct a generalized constant admittance switching model based on the LC equivalent circuit. Then, by discretizing the system state matrix, an extended modeling from a single converter to a multi-converter system is achieved. Furthermore, by combining the inertia characteristics of a virtual synchronous generator, a complete grid-type multi-converter model including control elements is established. Finally, a simulation model of the grid-type multi-converter system is built in PSCAD / EMTDC, verifying the correctness and effectiveness of the method. This invention demonstrates good applicability and stability in integrating the generalized constant admittance switching model with the inertia characteristics of a virtual synchronous generator at a sub-microsecond simulation step size. It significantly improves the simulation accuracy and efficiency while achieving rapid convergence of transient processes in the grid-type multi-converter and active support for grid voltage and frequency.

The application discloses an elevator group control scheduling method based on an improved NSGA-II algorithm and belongs to the field of elevator group control scheduling methods. The method comprises the following steps: obtaining elevator call data and elevator state data, establishing an elevator operation mathematical model, taking total passenger waiting time and system total energy consumption as optimization targets, and constructing a fitness function of the targets; introducing heuristic population initialization, self-adaptive crossover mutation rate and other methods to improve the NSGA-II algorithm, solving multiple targets through the improved NSGA-II algorithm, and obtaining a pareto solution set; automatically judging a current traffic mode according to the distribution of the pareto solution set; and making a decision on the pareto solution set according to the traffic mode, the fitness value of the solution and a decision scheme to obtain an optimal solution, namely, an elevator dispatching scheme. The method can automatically judge the traffic mode in the solving process, can solve the dependence of an optimization algorithm on the traffic mode before calculation, and can quickly converge and jump out of a local optimum.

The present application relates to a kind of particle swarm-based clean energy station multi-unmanned aerial vehicle task allocation method, comprising: obtaining clean energy station area data information, and the division of patrolling area is carried out;Clean energy station multi-unmanned aerial vehicle task allocation model is constructed;According to individual optimal position and global optimal position, the speed and position of particle are adjusted, and objective function is optimized;When the iteration number reaches upper limit, the task allocation solution corresponding to global optimal particle is returned, otherwise, continue to update particle state, optimization solution.Nonlinear dynamic collaborative improvement is carried out to inertia weight and learning factor, inertia weight adopts nonlinear self-adaptive decreasing strategy, iteration initial period is kept larger value, enhances the global traversal ability of unmanned aerial vehicle formation, widely searches various sub-regions and task combination;Rapidly attenuate in iteration later period, strengthen local precision search, ensure fast convergence to optimal cost combination.Golden sinusoidal algorithm is fused to reconstruct position update mechanism, realize the dynamic balance of global and local search.

The application discloses a kind of electromagnetic super surface unit structure optimization method based on constraint space mapping, belong to electromagnetic super material design technical field.The application is to solve the problem of low optimization efficiency and difficult to handle constraint condition in the process of super surface unit structure optimization in prior art, mainly using mapping multidimensional structure variable to the convex constraint space formed by inequality constraint condition, and combining multi-objective optimization algorithm for efficient optimization.By using direction vector, scale factor to represent structure variable vector, the initial solution generation process that satisfies complex constraints is simplified, and the global search ability and convergence speed in the constraint space are improved.The method can quickly converge to local optimal solution under complex constraint conditions, and effectively find the Pareto optimal solution, thereby providing an efficient and accurate solution for multi-objective optimization of electromagnetic super surface unit structure.

This invention discloses an online self-calibration method and system for wake model parameters based on the DE-SQP joint framework. The method includes constructing a wake calculation system comprising a wind turbine model, a wind speed model, and a wake model; acquiring measured power data of the wind turbine generator under a target operating condition, and calculating predicted power data for the corresponding operating condition using the wind turbine model and wind speed model in the wake calculation system; setting an optimization objective function based on the measured and predicted power data; using a differential evolution algorithm to globally search and optimize the multidimensional empirical parameters of the wake model to obtain a globally optimal preliminary parameter vector; using the preliminary parameter vector as the initial point, employing a sequential quadratic programming algorithm for local fine-tuning, utilizing its fast convergence characteristic to obtain the final calibration parameters; and applying the final calibration parameters to the wake model in real time to complete the online self-calibration of the model parameters. This invention can significantly improve the prediction accuracy, robustness, and generalization ability of the model across the entire wake region.

Embodiments of the present application provide a memory management method, a computer device and a storage medium, and belong to the computer field. The method comprises the following steps: receiving a memory allocation request, and recording behavior information of a memory operating system after the memory allocation request fails; determining an adjustment parameter of a to-be-adjusted memory configuration parameter based on the behavior information, wherein the to-be-adjusted memory configuration parameter and the behavior information have a corresponding relationship; and adjusting the memory configuration parameter according to the adjustment parameter. The technical solution of the embodiments of the present application can improve the adjustment efficiency of the memory configuration parameter and realize the rapid convergence of the memory configuration parameter.

The application relates to the technical field of semantic processing, and discloses a semantic understanding system for user intention in a complex environment, which comprises an input text serialization segmentation unit, a semantic graph topological mapping unit, a dynamic uncertainty convergence state control unit and an intention matching arbitration decision unit; the input text serialization segmentation unit reconstructs a word unit flow sequence and filters low-weight edges, the semantic graph topological mapping unit establishes a semantic topological node, the dynamic uncertainty convergence state control unit activates a search switching module when the uncertainty value is out of limit, and the intention matching arbitration decision unit outputs matching data; the application regulates graph connection weights through the entropy variation characteristics of a symbol stream, in-situ cuts off a scattered manifold and cuts off a low-probability scattered branch to make an intention feature vector quickly converge, effectively alleviates search bifurcation and index tree recalculation oscillation caused by input defects, and reduces the flow text processing delay.

An AI chip device with a fast-converging network that manipulates a large number of variables and finds inter variable relations and hidden patterns, which it learns and converges rapidly. It utilizes a self-learning device (SALS Device) for finding the divergence to provide a parameter for feedback, and converges rapidly. It employs methods such as the “SALS Model” and the “AIW Method” (approximation by iteration of weight) to converge the network fast and to determine the relation between two variables to combine the feature variables into an output variable using a weight and the combined variables are combined again and again until a single output is obtained. It is capable of supervised and unsupervised learning by using its property of convergence.

The application discloses a kind of self-adapting global fractional order terminal sliding mode serpentine robot control methods.In order to solve the problem of high complexity in modeling process, it is difficult to solve, an error dynamics model is constructed, and the design difficulty of control system is further simplified. In response to the uncertain terms and unknown disturbances that may exist in the model, a radial basis function neural network observer is introduced, which effectively approximates the estimation of these uncertain factors. In order to ensure that the joint serpentine robot system can converge quickly, an adaptive constant rate control strategy is designed, and combined with the global fractional order terminal sliding surface, not only the high-precision control of the system is realized, but also the robustness of the system is significantly enhanced.

The application discloses a fast convergence networked anti-interference decision method and system of a wireless self-organizing network, and an implementation scheme thereof comprises the following steps: constructing a key networking state information set; setting multiple task scenarios, collecting interaction trajectories and dividing the interaction trajectories into a support set and a query set; constructing a meta-learning network model comprising a context inference network, a strategy network and a value network, taking a weighted sum of an expected cumulative return and a KL divergence penalty term as an optimization target, and obtaining a scene potential variable by decoupling a public strategy and scene-specific characteristics from the support set trajectory; inputting the potential variable and the key networking state information into the trained meta-learning network model in combination, and outputting a cross-layer joint anti-interference decision action comprising physical layer waveform selection, working frequency point selection and network layer routing selection. The application can effectively reduce state interaction and calculation overhead, improve the anti-interference capability of the wireless self-organizing network, and be used for combined strategy selection of physical layer waveforms and network layer routing protocols in the wireless self-organizing network.

This invention discloses a timing adjustment device and method for a satellite communication system, including a timing adjuster, a timing offset estimator, a timing offset statistician, a timing offset coefficient calculator, and an SNR estimator. The timing adjuster adjusts the reception timing of the N subframe based on the timing offset value provided by the timing offset statistician for the N-1 subframe and the timing adjustment coefficient, thereby adjusting the timing offset. The timing offset estimator outputs the sum of a fractional multiple of the timing offset and an integer multiple of the timing offset based on the timing offset of the current N subframe. The SNR estimator calculates the SNR value of the N subframe signal and outputs the SNR value. The timing offset coefficient calculator selects timing offset filter coefficients and timing adjustment coefficients with different weights based on the input SNR value. The timing offset statistician updates the timing offset filter value based on the timing offset filter coefficients provided by the timing offset coefficient calculator and the timing adjustment amount issued by the timing adjuster for the N subframe, and feeds back the N subframe timing offset value to the timing adjuster. This process is repeated to achieve rapid timing convergence.

This invention belongs to the field of intelligent manufacturing and production scheduling technology, and discloses a dynamic scheduling method and system for flexible workshops that considers machine aging. The invention designs a hierarchical environmental response strategy, which first assesses the severity of environmental changes. When the changes are not drastic, only a lightweight local optimization strategy is used for fine-tuning; when the changes are more drastic, a global reconstruction strategy combining knowledge transfer, re-initialization, and targeted repair is activated. This allows the algorithm to intelligently allocate computing resources according to the severity of environmental changes, avoiding blind global searches and greatly improving the algorithm's response speed and operating efficiency. This design effectively balances the algorithm's exploration and utilization capabilities, ensuring rapid convergence while maintaining population diversity, thereby obtaining a set of Pareto optimal solutions with good convergence and wider distribution.

This application discloses a thermal effect-driven chip layout planning method, apparatus, device, and medium, relating to the field of integrated circuit automated planning technology based on reinforcement learning. The method includes: generating an initial chip layout based on a chip configuration file; constructing a state vector by combining the layout information, line length, and simulation operating temperature of the initial chip layout; inputting the state vector into a pre-constructed deep Q-network to adjust and obtain a new chip layout; calculating a reward function based on the new chip layout; determining the target chip layout and corresponding type label for the current iteration based on the reward function; if the iteration threshold is not reached, continuing iteration until the threshold is reached; and determining the optimal chip layout based on the type label of the target chip layout for each iteration. This application effectively overcomes the shortcomings of traditional methods, such as low optimization efficiency, slow convergence, and susceptibility to local optima. It enables self-learning and rapid convergence in the optimization process, significantly reducing runtime overhead.

The application discloses a robot non-calibration visual servo control method with fast convergence, which comprises the following steps: constructing a perspective projection model containing unknown camera internal and external parameters; linear parameterization; designing an auxiliary matrix, an auxiliary vector and a sliding variable; designing an adaptive law for online updating unknown parameter information; designing a torque controller of a robot visual servo system; and feeding back the calculated torque value to the mathematical model of the degree of freedom robot system. The application realizes the visual servo control of the robot under the condition that the camera internal and external parameters are unknown, controls the robot to move to the desired position quickly, and guarantees the safety and reliability of the robot visual servo system in the complex and changeable working environment.

The application discloses a task-aware-based large model fine-tuning method and a legal information analysis method. The fine-tuning method comprises the following steps: constructing a target field knowledge graph and injecting an initial large model to obtain a first large model, establishing a mapping relationship between a task type and a knowledge subgraph to obtain a second large model, and configuring a task self-adaptive optimization mechanism to obtain a third large model; selecting and expanding initial data samples to obtain target training samples; finally, the third large model is fine-tuned by using a progressive unfreezing strategy and mixed precision training. The method deeply couples the two, uses a task-knowledge mapping table to drive the unfreezing sequence and dynamically update the precision bit width, converges quickly with the least video memory and parameters in the early stage, gradually releases the capacity and precision bit width in the later stage, realizes the triple balance of gradient stability, video memory saving and performance optimization, and the fine-tuned large model realizes the deep fusion of professional field knowledge and parameters, can dynamically adjust the internal representation and reasoning mode according to different task characteristics, and meets the diversified task analysis demand.

The application belongs to the technical field of block chain, and particularly relates to a lightweight consensus method for large-scale distributed intelligent computing scenarios. The method is as follows: firstly, a threshold key is initialized among replica nodes and a signature share is generated. In the chain consensus stage, the leader node broadcasts a proposal, and the replica verifies and generates a threshold signature share and returns after verification; the leader node aggregates enough shares to form a legal certificate to promote the pipeline. If a certificate is not formed within a timeout, the view synchronization and recovery stage is entered. When adjacent view chain certificates are reached, the final submission can be accelerated; otherwise, the block needs to be submitted after meeting the continuous three-view condition. In abnormal conditions, the view synchronization mechanism is enhanced to aggregate the view switching threshold share, so that the whole network quickly converges to the new view and recovers the consensus. The technical scheme of the application can be widely applied to large-scale distributed sharded block chain networks through the collaborative optimization of the cryptographic primitives and the sharding structure level.