58 results about "Global iteration" patented technology

The invention discloses a federated learning model training method and system based on differential privacy image noise addition. The method comprises the following steps: S1, initializing an evaluation network, a target network and hyper-parameters; s2, initializing a global model and parameters; s3, acquiring a selected client and a privacy budget; s4, the selected client performs local training and uploads the disturbance gradient to the central server; s5, the central server aggregates the disturbance gradient of the client to obtain an updated global model and a reward corresponding to the current state; s6, the next state of federated learning is obtained, and the evaluation network and the target network are updated; s7, repeatedly executing the steps S3 to S6 until the set number of global iterations is reached; and S8, updating the exploration rate and repeatedly executing the steps S2 to S7 until the set federal learning times are reached. The invention provides a dynamic differential privacy method in combination with a federal learning framework, which can improve the image recognition rate while ensuring the privacy of the user.

The invention belongs to the related technical field of task scheduling optimization, and discloses a transport vehicle operation path and scheduling optimization method and system, and the optimization method comprises the steps: generating a path library, carrying out the global search of a particle swarm search algorithm on the basis of the path library, and during the execution of the particle swarm search algorithm, on the basis of a current global optimal solution, carrying out the global search of the particle swarm search algorithm; circularly executing local search, namely generating a new solution by using a damage and repair strategy and performing conflict adjustment, then updating a global optimal solution based on the new solution, then updating particle positions and speeds, and performing particle swarm iterative search again, namely executing global iteration; and after the number of iterations is reached and an optimal solution is obtained, decoding is carried out to generate a scheduling scheme, that is, tasks in the task set are allocated to each transport vehicle, a vehicle scheduling scheme and path selection are generated, and transport vehicle operation path and scheduling optimization for solving space-time conflicts by multiple vehicles and multiple task points is efficiently realized.

A distributed model training method includes that: the training of an internal iteration with a preset number of internal iterations is performed on a preset model through a computation process, to obtain a first node model parameter value of a current global iteration; a second node model parameter value of the current global iteration of a second computing node in a distributed system is acquired through a communication process running in parallel with the computation process, and a first ALLReduce model parameter value of the current global iteration is determined according to the first node model parameter value and the second node model parameter value; and external iteration is performed through the computation process by using a second ALLReduce model parameter value of a last global iteration, to obtain a target model parameter value of the current global iteration.

The present invention belongs to the field of distributed machine learning, and relates to a decentralized learning method based on differential privacy protection for heterogeneous data. It includes: building a decentralized scenario, setting global parameters, and completing user settings; users send model parameters to neighbor users, calculate self-gradients and cross-gradients, clip self-gradients and cross-gradients respectively, add noise to self-gradients and cross-gradients, and pass cross-gradients to neighbors; each user obtains a corrected gradient based on self-gradients and received cross-gradients after similarity correction; update the model, broadcast the updated momentum and model to neighbors, and perform weighted averaging based on the received momentum and model; repeat the above steps until the model of each user converges or reaches the upper limit of the global iteration number. The method disclosed in the present invention obtains a model that effectively processes heterogeneous data within a limited number of learning rounds in a decentralized environment, and protects user data privacy during the learning process to meet differential privacy requirements.

The invention provides a hybrid intelligent path planning method and system for enteromorpha tracking and monitoring, and relates to the technical field of marine ecological environment monitoring, and the method comprises the steps: obtaining the data of a monitoring region, generating an initial path meeting basic constraints based on a rasterization cost model through employing a Dijkstra algorithm, and then taking the path as an initial solution, and carrying out the optimal path planning through employing a Dijkstra algorithm; carrying out global iterative optimization with monitoring coverage rate, equipment energy consumption and obstacle avoidance safety as targets by adopting a snakelike optimization algorithm, and carrying out smoothing processing on an optimized path; and meanwhile, path re-planning is dynamically triggered based on real-time feedback data. According to the method, the technical problem that the initial feasibility, the global optimality and the dynamic adaptability are difficult to consider in unmanned monitoring equipment path planning in a complex marine environment is effectively solved, and the efficiency, the accuracy and the reliability of enteromorpha tracking monitoring are remarkably improved.

The invention discloses an adaptive optical anti-shake control method for a camera module. The adaptive optical anti-shake control method comprises the following steps: S1, collecting real-time vibration data of the camera module in a working process; s2, preprocessing the real-time vibration data to obtain pose change information; s3, constructing a global search model, and defining an anti-shake compensation parameter search space; s4, performing multiple rounds of global iteration by using an improved seagull optimization algorithm, and determining a candidate parameter solution and a candidate region; s5, performing local sampling and iterative tuning by adopting a Bayesian optimization algorithm to obtain an optimal anti-shake compensation parameter; s6, the optimal anti-shake compensation parameters are issued to a camera module control unit, and an optical anti-shake assembly is driven to adjust the position or angle of a lens group; and S7, carrying out quality evaluation on the compensated image, and triggering whether to carry out global search or local optimization again or not. According to the invention, by constructing an adaptive optimization control mechanism, high-precision and low-delay real-time optical anti-shake compensation of the camera module in a dynamic environment is realized.

The invention relates to the technical field of industrial adaptive control, in particular to an automatic processing parameter optimization method and system for a five-axis numerical control machine tool. The specific implementation process comprises the following steps: carrying out discrete analysis on machine tool parameters containing a cutter shaft vector and a translation shaft position coordinate, and generating time-varying geometric features; constructing a numerical control machine tool simulation model according to the material constitutive equation and the frequency domain response function, calculating a regenerative flutter prediction sequence, and dividing a real-time parameter feasible region by combining global rigidity and local flexibility data; and performing global iteration optimization on the processing parameter function group by using a multi-target optimization algorithm in the feasible region, and locking a Pareto optimal solution set to perform automatic optimization on the processing process. According to the method, the simulation model containing the physical dynamic characteristics and the real-time parameter feasible region are constructed, flutter and errors caused by stiffness matching imbalance are avoided, collaborative optimization of multiple target variables is achieved, and the productivity efficiency of the numerical control machine tool is remarkably improved on the premise that the machining precision is guaranteed.

The invention discloses a pulmonary nodule detection system and method based on federated learning and a target detection algorithm, and relates to the technical field of medical image detection and auxiliary diagnosis, the system comprises a plurality of medical institution clients and a cloud central server, the central server initializes to generate a global model, and deploys the global model to each client; each client uses a local CT data set to train a global model, model parameters are updated through training until loss is minimized so as to complete training, a local model is generated, and the updated model parameters are uploaded to the central server; wherein the total loss of the local model is the dynamic weighted sum of various losses; the central server adopts a dynamic weighted average method to aggregate the updated model parameters of all the clients, generates an updated global model, then deploys the updated global model, and gradually optimizes the global model through multiple times of global iteration updating; and each client identifies the to-be-detected CT image by using the locally deployed optimized global model, and outputs a more accurate pulmonary nodule identification result.

The present invention discloses a method and device for optimizing the security capacity of a drone-assisted backscatter communication system, which relates to the field of wireless communication technology. The system includes one drone, K backscatter devices and M eavesdroppers, and includes the following steps: obtaining collected energy, inputting the collected energy into the backscatter device for backscatter communication, optimizing time allocation, drone trajectory and backscatter device scheduling based on preset energy constraints and security performance constraints of backscatter communication; solving time allocation, drone trajectory and backscatter device scheduling respectively to obtain three optimal solutions, performing global iteration on the three optimal solutions to obtain a global optimal solution as the security capacity optimization result, thereby maximizing the security capacity of the minimum link.

The invention provides a communication network deployment method, device, equipment and medium, and the method aims at a target relay node randomly determined from an initial network deployment scheme, if the migration probability of the target relay node is not smaller than a migration probability threshold value, the target relay node is migrated to obtain a first intermediate network deployment scheme, and the first intermediate network deployment scheme is a second intermediate network deployment scheme. And on the basis of the first intermediate network deployment scheme, local position optimization is performed on at least part of relay nodes in the area to which the relay nodes belong, so that a candidate network deployment scheme is obtained, and if the candidate network deployment scheme meets a preset condition, the candidate network deployment scheme is used as an initial network deployment scheme to perform global iteration, so that the initial network deployment scheme is obtained. And finally determining a target network deployment scheme to perform entity network deployment. According to the embodiment of the invention, the entity network deployment precision can be improved.

The application provides a mixed intelligent path planning method and system for Enteromorpha tracking monitoring, and relates to the technical field of marine ecological environment monitoring. The method comprises the following steps: obtaining monitoring area data, generating an initial path meeting basic constraints by using a Dijkstra algorithm based on a gridding cost model, then taking the path as an initial solution, performing global iterative optimization with a snake optimization algorithm for the purpose of monitoring coverage, device energy consumption and obstacle avoidance safety, and performing smoothing processing on the optimized path; meanwhile, path re-planning is dynamically triggered based on real-time feedback data. The application effectively solves the technical problem that path planning of unmanned monitoring equipment in complex marine environment is difficult to balance initial feasibility, global optimality and dynamic adaptability, and significantly improves the efficiency, accuracy and reliability of Enteromorpha tracking monitoring.

The application belongs to the technical field of state monitoring, and specifically relates to a robot operation state real-time monitoring method and system based on end-cloud cooperation, which comprises the following steps: collecting original data, completing noise reduction, alignment, repair and coding according to a kinematic model to form a standardized time series data set. The local model and early warning rules are used to carry out risk preliminary judgment, divide abnormal types, and schedule transmission to output encrypted data packets. The cloud analyzes data and local judgment results, combines historical working conditions and environmental information to carry out space-time fusion analysis, revises and corrects with the aid of a risk model, and outputs a comprehensive judgment report. The platform formulates an optimization scheme according to task priority and field layout, and pushes control instructions in a chain link. The terminal executes instructions and returns state data, and the end-cloud synchronously updates the model and rule library relying on incremental learning, and the cloud completes global iteration to realize system autonomous evolution. In the application, data research and risk judgment are managed through multiple links to build a robot self-evolution monitoring system.

The application relates to the technical field of tunnel engineering disaster prediction, and discloses a soft rock tunnel large deformation prediction method based on a particle swarm optimization neural network, which comprises the following steps: selecting geological construction characteristic parameters such as water content of surrounding rock and construction methods as input factors, and establishing a qualitative index and quantitative value mapping; based on orthogonal test design and numerical simulation, engineering sample data sets are constructed and normalized; subsequently, the initial connection weight and bias of an error back propagation neural network are globally iteratively optimized by using a particle swarm optimization algorithm, so that optimal initial values are obtained by taking network errors as fitness values; finally, data training is carried out to generate a mature model, and a predicted deformation value is output and combined with a double check mechanism to carry out engineering early warning. The application solves the problems that traditional neural networks are slow in convergence and prone to falling into local optimization due to random parameter initialization, and effectively improves the prediction accuracy of soft rock tunnel large deformation and the reliability of engineering early warning.

The invention discloses an Internet of Vehicles target identification method based on personalized federated learning, which assumes that an edge server and N moving vehicles exist in an Internet of Vehicles scene, and data sets owned by each vehicle are distributed in a non-independent identical distribution mode. Each vehicle uses local data to train a local target detection model and uploads model parameters to an edge server, the edge server collects the model parameters, reasones data similarity and hierarchical clustering based on a public data set, updates personalized layer and non-personalized layer parameters of the next round of global iteration of each vehicle, and sends the updated personalized layer and non-personalized layer parameters to the edge server; therefore, the next round of global iterative training is carried out until convergence. According to the method, by optimizing the cooperative relationship between the moving vehicles, the personalized model is customized for the moving vehicles with different data distributions, and the problem of training model divergence caused by data isomerism is solved, so that the convergence speed is increased, and the model precision is improved.

This invention provides a method for reconstructing continuous wheel-rail forces based on discrete fastener reaction forces, comprising the following steps: Step 1, establishing a wheel load distribution model; Step 2, establishing an algorithmic framework for the inversion method, by introducing a responsibility region and a local deconvolution strategy, mapping the discrete fastener reaction force FRF to the continuous wheel-rail force WRF. This invention effectively solves the problems of traditional onboard sensors' difficulty in long-term fixed-point monitoring of specific road sections, and the inability of existing ground discrete monitoring points to fully characterize the continuous wheel-rail interaction behavior. By constructing an inversion framework based on Gaussian load distribution using a "responsibility region + local deconvolution" approach, it fully utilizes the load transfer characteristics of the vehicle-track coupling system, achieving high-precision reconstruction of continuous wheel-rail forces using ground discrete fastener reaction force data. This eliminates the need for complex global iterative optimization and can accurately restore the dynamic characteristics of wheel-rail contact even with only a limited number of discrete measurement points.

The invention discloses an unmanned aerial vehicle assisted symbiotic radio system capacity optimization method and a related device, and the method comprises the steps: carrying out the channel allocation between backscattering equipment and a main user according to the number of backscattering equipment in a system, the number of main users, and the distance between the backscattering equipment and the main user; a backscatter communication capacity optimization model of the system is decomposed into a base station transmitting power optimization model, a backscatter coefficient optimization model and an unmanned aerial vehicle position optimization model, global iteration is carried out, a global optimal solution is used as a capacity optimization result of the system, and the system can be configured based on the optimization result so that the system can operate under the optimal capacity.

The application discloses a federated learning method, device, system, storage medium and electronic equipment. The method comprises the following steps: in each global iteration process, receiving a modified gradient item in the current global iteration process; in each local iteration process in the current global iteration process, determining a local gradient item of the machine learning model in the current local iteration; correcting the local gradient item of the current local iteration based on the modified gradient item, updating the model parameter of the machine learning model based on the corrected target gradient item, and executing the next local iteration process based on the updated model parameter; and in the case that the local iteration process is completed, sending the model parameter change in the current global iteration process to the center server node. The embodiment of the application avoids the case that each computing node falls into a local optimal value in the local training process, and improves the generalization performance of the machine learning model obtained by training.

The invention belongs to the technical field of automatic driving, and particularly relates to a local-to-global iterative optimization lane line detection method and application, and the method comprises the steps: obtaining an input road image, and extracting a multi-scale feature map of the road image; global feature coding is carried out on the multi-scale feature map to output confidence, the first K query vectors are dynamically selected from a group of learnable query vectors based on the confidence, and the query vectors correspond to potential lane line instances; initializing the selected query vector into a learnable lane line anchor box; inputting the initialized lane line anchor frame and the coded multi-scale feature map into a multi-stage decoder for iterative optimization; and calculating to obtain a lane line detection result by using the lane line parameters output by the last-stage decoder. The method is high in adaptability to complex scenes, and the lane line expression normal form can reduce the training complexity while having the fine-grained fitting capability.

The present invention discloses a method and system for training a federated learning model based on differential privacy image noise addition. The method includes: S1: Initialize the evaluation network, the target network, and hyperparameters; S2: Initialize the global model and its parameters; S3: Obtain the selected clients and the privacy budget; S4: The selected clients perform local training and upload the perturbed gradients to the central server; S5: The central server aggregates the perturbed gradients of the clients to obtain the updated global model and the reward corresponding to the current state; S6: Obtain the next state of the federated learning and update the evaluation network and the target network; S7: Repeat S3 to S6 until the set global iteration number is reached; S8: Update the exploration rate and repeat S2 to S7 until the set federated learning number is reached. The present invention proposes a dynamic differential privacy method in combination with the federated learning framework, which can improve the image recognition rate while ensuring user privacy.

The present application relates to the technical field of industrial adaptive control, in particular to a machining parameter automatic optimization method and system for five-axis CNC machine tools. The specific implementation process includes: discretely analyzing machine tool parameters containing tool axis vector and translational axis position coordinates to generate time-varying geometric characteristics; constructing a CNC machine tool simulation model according to a material constitutive equation and a frequency domain response function, calculating a regenerative chatter prediction sequence, and dividing a real-time parameter feasible region in combination with global stiffness and local flexibility data; performing global iterative optimization on a machining parameter function group in the feasible region by using a multi-objective optimization algorithm to lock a Pareto optimal solution set for automatic optimization of the machining process. The present application avoids chatter and errors caused by unbalanced stiffness matching by constructing a simulation model containing physical dynamic characteristics and a real-time parameter feasible region, and simultaneously realizes the collaborative optimization of multi-objective variables, thereby significantly improving the productivity efficiency of the CNC machine tool under the premise of ensuring machining accuracy.

The application belongs to the technical field of model training, and provides a differential privacy federated learning method based on noise scale allocation and related equipment. Embodiments of the application obtain a total noise scale sequence of a target client under a total training step number, and allocate a noise scale sub-sequence to a plurality of local training steps of the target client in a current global iteration based on the total noise scale sequence; the total noise scale sequence is an increasing sequence determined by minimizing the convergence upper bound of a global model corresponding to a server under the constraint of a total privacy budget of the target client; the target client performs the current global iteration based on a decreasing learning rate sequence and a mini-batch stochastic gradient descent algorithm, and performs gradient noise processing according to the noise scale sub-sequence. Finally, the current global model is updated based on the aggregated cumulative noise gradient. Embodiments of the application can be applied to differential privacy federated learning of mini-batch local training, and can improve model performance while protecting privacy.

The invention discloses a beam forming and trajectory joint optimization method for an unmanned aerial vehicle communication and inductance integrated system, and belongs to the technical field of wireless communication. The method comprises the following steps: clustering ground users, and selecting a user cluster for service in each time slot; establishing a joint optimization problem for maximizing the average downlink communication rate of the UAV-ISAC system under the conditions of transmitting power constraint, user minimum rate constraint, unmanned aerial vehicle kinematics constraint and sensing threshold constraint; decomposing a joint optimization problem into three optimization sub-problems of user cluster scheduling, common rate allocation and beam forming and unmanned aerial vehicle trajectory; a linear programming method, a positive semi-definite relaxation and successive convex approximation method and a continuous convex approximation method are respectively adopted to solve the sub-problems of user cluster scheduling, non-common rate allocation and beam forming and unmanned aerial vehicle trajectory optimization, and finally a joint optimal solution is obtained through global iteration. The communication rate of the system can be effectively improved on the premise that the sensing threshold is met, and cooperative gain of communication and sensing is achieved.

The invention relates to the technical field of federated learning, and provides a collaborative security and incentive federated learning method, which is characterized by comprising the following steps that: a server selects an optimal client subset; each client in the optimal client subset performs performance evaluation on the global model and the local model and generates an alarm decision; calculating an optimal local training batch by each client in the optimal client subset; the clients participating in this round of training generate model updating based on the alarm decision; the server judges a benign client set based on the alarm information; aggregating the model update corresponding to the benign client set to generate a new generation of global model; broadcasting a new generation of global model to the optimal client subset for the next round of training; and repeatedly executing until the total number of rounds of global iteration is reached. According to the invention, the problem of inherent conflict and tradeoff between security and incentive in federated learning for a long time is successfully solved.

This invention discloses a method and system model for detecting user electricity theft based on consortium blockchain and federated learning. The detection method includes the following sequential steps: 1) Node registration; 2) Training initialization; 3) Key distribution; 4) Local training: Selected participants download the global model and iterate through it using a local optimization strategy, then encrypt the model and send the encrypted data to the proxy nodes; 5) Model aggregation: A cluster of proxy nodes that meet the threshold jointly recovers the decryption key and calls a smart contract to execute a two-stage aggregation, with the result uploaded to the blockchain after consensus; 6) Global model update: After global iteration meets the conditions, global training ends, and the electricity theft detection model is updated; 7) Electricity theft detection. This invention can obtain a model with performance exceeding that of local individual training while also ensuring the privacy of local data, thus improving security, and enabling continuous iteration and updating of the user electricity theft detection model.

This invention discloses a data-driven adaptive parametric modeling method for automotive rooflines, comprising the following steps: acquiring multiple sets of side-view roofline coordinate data of real vehicles, dividing the side-view roofline into a front guide segment and a rear main body segment; establishing a normalized coordinate system for each segment, and then discretizing it into several sampling points to form a bottom-level sample database; defining control variables, including the spatial position of control points in the normalized coordinate system, the number of control points, and the type of reconstructed model; employing a two-layer optimization strategy to perform global iterative optimization of the control variables for each segment to obtain the optimal control variables; constructing a comprehensive evaluation index, and selecting the control variable with the optimal comprehensive evaluation index from all parametric reconstructed models that have undergone inner-layer optimization as the optimal control variable, thus obtaining the optimal parametric reconstructed model. This invention aims to solve the problems of existing designs relying on manual experience and the difficulty in achieving an optimal balance between fitting accuracy and model complexity.

The invention relates to the field of intelligent crossing of material mechanical property testing and calculation, in particular to an intelligent global iteration recognition method for indentation impact factors based on composite indentation data. The method comprises the steps of establishing a mapping database based on a material constitutive model; updating the mapping database according to the residual indentation morphology and the maximum load in the mapping database; constructing a forward artificial neural network, and training the forward artificial neural network by using the updated mapping database; acquiring a plurality of groups of residual indentation morphology and maximum load in the indentation experiment of the to-be-tested material, and converting the residual indentation morphology and maximum load into actual standardized characteristic quantity irrelevant to the size of the indenter; and constructing a composite error function related to the residual indentation morphology and the maximum load, carrying out global optimization on the indentation influence factor, and outputting the indentation influence factor of the to-be-tested material. In this way, multi-source indentation information can be fused, friction and depth changes are fully considered, and the method has high calculation efficiency and strong generalization ability.

This application relates to the field of data processing technology and discloses a method for processing heterogeneous multi-scale data in Gaussian process regression. The method first acquires the raw data and performs feature transformation, standardization, and dimensionality reduction preprocessing to construct a multi-level validation dataset. Second, it reconstructs the Gaussian process regression kernel function, introducing a learnable Minkowski distance parameter to replace the traditional fixed Euclidean distance metric. Next, it uses a tree-structured Parzen estimator strategy to perform global iterative optimization on the validation set with the goal of minimizing prediction error, determining the optimal Minkowski distance parameter. Subsequently, it trains the model based on the optimal parameter and an automatic correlation determination mechanism, jointly optimizing hyperparameters such as signal variance and length scale through maximum likelihood estimation. Finally, it performs prediction and outputs the mean and variance. This invention, through adaptive parameterization of the kernel function metric, effectively overcomes the shortcomings of traditional models in adapting to non-spherical distributions and multi-dimensional heterogeneous features, significantly improving prediction accuracy and generalization performance.

The invention discloses an intelligent production scheduling method and system for cable flexible manufacturing, and relates to the field of production management scheduling, and the method comprises the steps: firstly, through data solidification, comprehensively capturing dynamic information such as an emergency order, a current plan, a factory state and model changing time, and forming a context basis of a decision; then, a rapid heuristic algorithm is used to carry out cost quantitative evaluation on all possible insertion schemes of the emergency order, and a high-quality initial scheduling solution is rapidly generated; and finally, introducing a simulated annealing element heuristic algorithm, and carrying out global iterative search by taking the initial solution as a starting point so as to find an optimal solution with minimum disturbance to the whole production plan. According to the method, a complex experience decision is converted into a data-driven automatic optimization process, rapid and accurate adjustment of the production plan is realized, and the overall stability and efficiency of the production line are improved.

The invention relates to an optimization method for large-scale cloud service processes, including: randomly allocating a plurality of execution plans to a plurality of parallel nodes; performing, by each parallel node, a local iteration in parallel to process the execution plans inside the parallel node until the local iteration stops, where during each local iteration, the parallel node needs to process the execution plans by using DMBO algorithm and a dependency-conflict repair algorithm; then gathering together execution plans obtained after the local iterations in all the parallel nodes are completed; and determining whether a global iteration stop condition is satisfied, where if yes, optimal execution plans are directly outputted, or otherwise, the gathered execution plans are reallocated to the plurality of parallel nodes, and then the foregoing steps are repeated. The present invention improves the efficiency and accuracy of solution seeking of the optimization method for cloud service processes.

Embodiments of the present application provide a method and device for constitutive model construction and parameter optimization of honeycomb materials, computer equipment and medium, relating to the technical field of electronic digital data processing, wherein the method comprises the following steps: obtaining original experimental data sets generated by mechanical experiments of the honeycomb material under multiple different working conditions; fitting to obtain a shape function representing the stress-strain relationship under a reference working condition, constructing a density function representing the influence of density on mechanical properties, using an exponential nonlinear function, constructing a strain rate function representing the influence of strain rate on mechanical properties, and constructing a temperature function representing the influence of temperature on mechanical properties; constructing a constitutive model of the honeycomb material based on multi-factor correction factors; taking the minimum mean square error between the stress-strain curve predicted by the constitutive model and the target curve as the optimization target, performing global iterative optimization and outputting. Through the constructed constitutive model, the problem of strong dependence of traditional model parameter fitting and poor universality is solved.