55 results about "Algorithmic skeleton" patented technology

The invention provides a multi-modal feature perception and meta reinforcement learning adaptive algorithm scheduling system and method, and the system comprises a multi-modal feature coding module which is used for carrying out the cross-modal feature alignment and fusion of input image data, physical signals and structured parameters; the deep reinforcement learning decision module is used for constructing a multi-objective optimization strategy by adopting a TD3 algorithm framework; the self-adaptive scheduling execution module is used for realizing strategy distillation and dynamic priority scheduling; and the multi-modal feature coding module comprises a dynamic graph convolutional network and a hierarchical cross attention mechanism. Through technical innovation, comprehensive breakthrough is achieved in the aspects of multi-modal data processing precision and efficiency, reinforcement learning scheduling stability and adaptability, engineering deployment lightweight and robustness and the like, the defects in a dynamic scene in the prior art are effectively overcome, and the intelligent level and the practical application value of an algorithm scheduling system are remarkably improved.

The invention discloses a body intelligent controller architecture based on an end side intelligent model and an operation method, and belongs to the technical field of industrial body intelligence. According to the architecture, a CPU, a GPU and an FPGA are integrated by adopting a heterogeneous multi-core processor, and high-speed interconnection is realized through a star-type + NoC hybrid topology; a multi-level task scheduling mechanism is constructed based on a real-time operating system, and it is ensured that control tasks are executed preferentially. End-side visual model lightweight (YOLOv5s model parameters are reduced to 3.5 M) and FPGA hardware acceleration (control instruction time delay is 10ns) are innovatively realized on an algorithm framework layer, and a transmission bottleneck is eliminated through a zero-copy data path. The system supports model hot switching less than or equal to 10ms and FPGA bit stream online updating less than or equal to 50ms, and has MD5 check and an automatic rollback mechanism. Compared with a traditional separation type scheme, the processing time delay is reduced by 25 times, the hardware cost is reduced, the size is reduced, and the real-time performance, reliability and deployment flexibility of the industrial robot and other intelligent agents with the body are remarkably improved.

The invention discloses an unmanned ship control method based on deep reinforcement learning in a multi-task scene. The method comprises the following steps: constructing dynamics and kinematics models of an unmanned ship; the method comprises the following steps: constructing a high-fidelity simulation environment based on Isaac Sim and a parallel training framework thereof, and respectively designing a state space and an action space for various unmanned ship tasks; reward functions are respectively designed; constructing an unmanned ship control strategy, and designing an algorithm framework of a centralized importance sampling and shearing strategy optimization mechanism based on an end-to-end deep reinforcement learning algorithm; and for different task scenes, multiple times of training-verification are performed on the strategy network through a PPO algorithm, and the trained strategy network is used for realizing multiple control tasks of the unmanned ship. According to the method, a high-fidelity simulation environment is constructed through Isaac Sim and a parallel training framework thereof, parallelization support which is crucial to deep reinforcement learning training is achieved, and various actual control tasks of the unmanned ship can be achieved.

The invention discloses a task allocation method and system based on an improved whale optimization algorithm framework, relates to the technical field, and is used for optimizing order type adaptive cross-domain traffic control network task allocation and improving the key task response capability of a time-sensitive traffic system. ICWOA initializes a population through Chebyshev mapping, and introduces Levy flight disturbance to enhance the optimization ability; dynamically balancing local and global search by means of adaptive parameters; relieving population diversity attenuation through randomness retention, diversity maintenance and boundary constraint; dimension type pinhole imaging reverse learning is fused to reduce high-dimensional optimization dimension interference. According to the algorithm, the convergence speed and the solving precision are better, sub-second calculation time is kept under different task scales, the task distribution efficiency is improved, and the time-sensitive scene task success rate is remarkably improved. The method solves the problems that an existing algorithm is insufficient in adaptive order type'order application-order sending 'structure and prone to falling into local optimum, population diversity attenuation and calculation speed.

The invention discloses a civil aviation element configuration optimization method based on mixed integer dynamic programming. The method comprises the following steps: S1, constructing a Chinese civil aviation core element and carbon emission spatial-temporal characteristic database; s2, establishing a mixed integer dynamic programming model with the purpose of minimizing the total cost of the system in the programming period; and S3, establishing decision variables, constraint conditions and a fusion solution algorithm framework required by the mixed integer dynamic programming model. And S4, based on the fusion solution algorithm framework, solving the mixed integer dynamic programming model by using a mathematical programming solver integrated with a robust optimization module so as to obtain a civil aviation core element dynamic configuration scheme which is optimal in total system cost in multiple periods in the future and meets the uncertainty scene. According to the method provided by the invention, a multi-source data fusion algorithm is optimized, a carbon emission measuring and calculating method is deepened, and solver parameters are adjusted, so that the collaborative requirements of improving the operation efficiency and reducing the carbon emission from the perspective of carbon cost are met.

The invention relates to a multi-target logistics distribution path optimization method based on AI, and the method comprises the following steps: obtaining order data, road network data and vehicle data of logistics distribution of a customer point, extracting basic information from the three types of data, and constructing the multi-dimensional features of the customer point; inputting the multi-dimensional features of the customer points into a pre-trained graph attention network for feature embedding, generating a distribution network feature graph fusing customer demands and a road network relationship, inputting a pre-trained reinforcement learning agent, and generating an initial path population with quality and diversity balance; performing iterative optimization on the initial path population based on a Memetic algorithm framework; and calculating the fitness of each individual on a plurality of preset optimization objectives, maintaining a Pareto solution set by adopting a multi-objective evolutionary algorithm with reference points, and recommending a final path scheme from the Pareto solution set based on user preferences. The method has the effect of remarkably improving the quality, efficiency and dynamic adaptability of path optimization.

The invention discloses an AUV trajectory tracking deep reinforcement learning method based on an improved curiosity mechanism. On the basis of an internal curiosity mechanism, a self-adaptive internal reward coefficient mechanism is provided, and an improved internal curiosity module IICM is constructed, so that the AUV can dynamically adjust the exploration capability of the AUV according to the actual tracking effect. Meanwhile, the IICM is combined on the SAC algorithm framework, an SAC + IICM algorithm is provided, the exploration behavior of the AUV is stimulated through a self-adaptive internal reward mechanism, and the understanding depth of the AUV on the environment is improved. Besides, in order to improve the tracking effect and the training efficiency, a composite reward function fusing factors such as path errors, speed changes and yaw angle errors is designed, and a state and action space highly matched with a tracking task is constructed. The method has the advantages of being high in autonomy, good in adaptability, high in convergence speed, high in control precision, high in robustness and the like, and is suitable for an AUV autonomous operation scene in a complex marine environment.

Aiming at the construction of an automatic warehouse logistics system of a digital workshop production line, the invention provides a single-load automatic guided vehicle workshop path planning method based on directed acyclic graph modeling, and the method comprises the functions of task selection, task execution sequence planning and the like. The method comprises a fusion algorithm framework based on a binary discrete particle swarm optimization algorithm and a genetic algorithm component, and can solve and fully consider the diversity of task selection strategies in a short time according to information such as known task arrival time and transportation starting point, so that normal production is ensured; and the task execution sequence with the shortest total driving distance is the optimal (or close to the optimal) task execution sequence.

The invention provides an interpretable safe longitudinal federal representation learning method and device for multi-modal big data. According to the method, for the problem of target domain modal data scarcity, in the target domain modal representation learning process, an attention mechanism is used for supplementing information of a source domain modal into a target domain modal, and an algorithm framework is established under a longitudinal federated learning framework, so that on one hand, it is guaranteed that data of a source domain and data of a target domain are not locally output, and the algorithm framework is established under a longitudinal federated learning framework; therefore, the data security is improved; on the other hand, the problem of insufficient modal data of the target domain is solved, and the performance of downstream tasks is improved; besides, in the process of constructing the loss function, an information bottleneck theory is introduced, redundant information between input and intermediate representation is minimized, and related information between representation and a target task is maximized, so that efficient information compression and feature extraction are realized, and the interpretability of extracted representation for downstream tasks is improved.

The present invention belongs to the field of video analysis technology, and discloses a method for software-hardware collaborative optimization of a video structured analysis system based on parameter value selection, including step 1: construction of multi-module algorithm control parameters and implementation parameter sets; based on the industry's mainstream YOLO+byteTrack+ResNet+SIFT CDVS feature video structured analysis system framework, combined with the encoding algorithm characteristics and pipeline constraints of heterogeneous multi-computing platforms, a hardware-friendly algorithm framework is selected for each module; step 2: multi-module coupling parameter optimization value selection; according to the pipeline throughput constraint, the control parameters of a certain concurrency granularity and concurrency intensity are configured to achieve a balance between system throughput and hardware consumption. The present invention supports the joint optimization of six algorithm modules, supports the joint optimization of target performance such as bit rate, computational complexity, algorithm accuracy, and system processing retrieval delay; and realizes the joint optimization of multi-target performance of multiple algorithm modules.

Traditional algorithms for solving constrained optimization problems are complicated to implement, difficult to interpret, and require significant computational resources. Disclosed embodiments convert constrained optimization problems into parametric optimization problems, in which at least a subset of the constraints are converted into parametric quadratic penalty (PQP) terms that each depends on a translational parameter. The parametric optimization problem may be used for optimization in a power system (e.g., for optimal power flow, economic dispatch, etc.). When solving the parametric optimization problem, the translational parameters are updated to ensure convergence. The parametric optimization problem can be solved with reduced computational expense, using only a linear equation solver to solve a sequence of primal variables only, thereby reducing computational complexity and expense. In addition, the disclosed embodiments provide a means to incorporate constraints into machine-learning algorithms. The disclosed algorithmic framework also provides interpretability and insights for analysis.

The invention discloses a multi-agent-based micro-grid energy management method and device containing flexible resources, and a storage medium. The method comprises the following steps: step 1, determining each main body of a micro-grid; 2, constructing a state space, an action space and a reward function of each main body in the three main bodies, thereby constructing a sequential decision model of the micro-grid intelligent body under a GPRO algorithm framework; 3, determining an objective function and constraint conditions; step 4, generating a training sample, and performing multiple rounds of training on the sequential decision model of the micro-grid intelligent agent under a GPRO algorithm framework through the training sample to obtain a micro-grid energy management model; and 5, inputting the current state of each main body in the micro-grid to be managed into the micro-grid energy management model to realize micro-grid energy management. The device and the storage medium are used for implementing the method. According to the method, flexible resources on the load side are fully considered, memory occupation is reduced, and training resources are remarkably reduced.

The application provides a mechanical state monitoring method based on an interpretable sparse optimization unfolding network, comprising the following steps: S100, performing mechanical state data monitoring test to obtain mechanical state signals; S200, constructing a sparse optimization model, and using sparse feature representation for the extracted mechanical state features; S300, using an alternating multiplier method to derive an iterative optimization solving algorithm of the sparse optimization model; S400, introducing learnable parameters to replace the parameters in the iterative optimization solving algorithm; S500, using an unfolding algorithm framework to construct an interpretable sparse optimization unfolding network; S600, using the interpretable sparse optimization unfolding network to identify the health state of a test machine; the application embeds a Q-adjusted wavelet transform into a sparse optimization model, and uses the multi-scale characteristics and wavelet dictionary structure for sparse feature representation, so that accurate capture of key state information can be realized in the feature extraction stage.

The invention discloses a robot path planning method based on an adaptive fuzzy and hybrid strategy, and belongs to the technical field of robot motion planning, and the method comprises the steps: carrying out environment perception, and obtaining environment information; the method comprises the following steps: on the basis of an RRT-Connect algorithm framework, fusing a fuzzy control theory, a hierarchical hybrid expansion mechanism and a dynamic path optimization thought to obtain an improved RRT-Connect algorithm; and an improved RRT-Connect algorithm is utilized, and robot path planning is realized based on environment information. According to the method, the fuzzy control theory is utilized to endow the path planning process with a macroscopic intelligent decision-making capability, and a microscopic accurate execution and fault-tolerant capability is provided through a hierarchical mixed strategy, so that the method is an efficient and robust universal solution; the success rate, efficiency and path quality of path planning of the robot in a complex high-dimensional space are remarkably improved, and more efficient and better collision-free path generation can be achieved.

The application provides an IRS transmission power optimization method based on a quasi-affine transformation evolution algorithm, and comprises the following steps: step 1, system model construction and parameter initialization; step 2, channel modeling; step 3, constructing a target for minimizing transmission power while meeting the constraint condition of the signal-to-noise ratio of all users; step 4, searching by using a quasi-affine evolution-based meta-heuristic algorithm; and step 5, outputting an IRS phase shift vector and an AP beamforming vector matrix. The application has the beneficial effect of improving the quasi-affine transformation evolution algorithm framework, aiming to overcome the problems of the traditional optimization method, such as the dramatic increase in the calculation complexity under the condition of a large number of IRS unit numbers N and user numbers K, and the defects of the existing heuristic algorithm (such as the particle swarm optimization) such as being prone to local optimization and low search efficiency.

The invention relates to the field of human resource management, in particular to a human resource management method based on multi-modal data fusion and adaptive learning, and adopts the technical scheme that the method comprises the following steps: constructing a time sequence data portrait of a target object through static attribute data, dynamic behavior data and external environment data; and inputting the multi-modal data into a preset feature extraction model for feature extraction to obtain a multi-dimensional feature vector, inputting the multi-dimensional feature vector into a pre-trained human resource analysis model to obtain an analysis result, and finally executing a corresponding human resource management operation according to the analysis result. The personnel and post matching degree score, the demission risk probability and the recommended development path are analyzed, intelligent resume screening and personnel and post matching optimization can be carried out, employee demission risk prediction and intervention suggestion can be carried out, personalized employee development path planning can be carried out, and the employee development path planning efficiency is improved. And deep learning and dynamic optimization capabilities are improved through an algorithm framework of multi-modal data fusion and dynamic adaptive learning.

The invention provides a mechanical state monitoring method based on an interpretable sparse optimization expansion network, and the method comprises the following steps: S100, carrying out the monitoring and testing of mechanical state data, so as to obtain a mechanical state signal; s200, constructing a sparse optimization model, and representing the extracted mechanical state features by adopting sparse features; s300, deducing an iterative optimization solution algorithm of the sparse optimization model by using an alternating multiplier method; s400, introducing learnable parameters to replace parameters in an iterative optimization solution algorithm; s500, constructing an interpretable sparse optimization expansion network by using an expansion algorithm framework; s600, using the interpretable sparse optimization expansion network to identify the health state of the test machine; according to the method, the Q-switched wavelet transform is embedded into the sparse optimization model, and sparse feature representation is carried out by using the multi-scale characteristic and the wavelet dictionary structure, so that the key state information can be accurately captured in the feature extraction stage.

The invention discloses a risk electric appliance identification online continuous updating method based on an improved aTLAS algorithm, and the method comprises the steps: 1) carrying out risk electric appliance characteristic analysis and task vector initialization, constructing a risk electric appliance characteristic library, and generating an initial task vector library; 2) constructing a general learning algorithm framework, generating a combined model by the initial task vector library through electric power fingerprint-oriented anisotropic scaling, and outputting a final prediction result through a decoupling head prediction structure; 3) realizing online continuous updating of the identification model, and dynamically updating the task vector library through a new sample detection and anomaly detection module; and 4) edge deployment and real-time reasoning are carried out, and when a new risk electric appliance is detected, the task vector library is updated, and dynamic iteration of the model is realized. According to the method, high-precision identification and dynamic self-adaptive evolution of risk electric appliances are realized, memory occupation and evolution time are remarkably reduced, and edge equipment deployment requirements are met.

The application belongs to the technical field of neural network simulation and analog, and particularly relates to a large-scale cortex pulse neural network asynchronous parallel simulation algorithm based on GPU. The application utilizes the advantages of multi-thread and texture memory of a computing graphics card, combines the general form of a biological brain receiving external stimulation and the general connection mode between neurons in the cortex, designs an asynchronous parallel algorithm framework of GPU and CPU, GPU is responsible for parallel evolution of neuron dynamics equations and block parallel calculation of isotropic connection in a local network, CPU is responsible for processing anisotropic long-range connection, and different neuron dynamics equations and plasticity learning rules can be compatible. Compared with the prior art, the application can effectively improve simulation speed, provides a tool for simulating a biological brain cortex in a single computing node, and is suitable for a single node multi-graphics card and a multi-node multi-graphics card distributed operation model.

The invention discloses a target detection algorithm based on stable learning, and relates to the technical field of computer vision, and the algorithm comprises the following steps: S1, constructing an algorithm framework; s2, data preprocessing; s3, feature extraction; s4, stable learning; and S5, target prediction. According to the target detection algorithm based on stable learning, the weight of the training sample is dynamically adjusted through the stable learning module, the dependence of the model on irrelevant features is reduced, and the influence of data noise and abnormal values on model training is reduced, so that the model is more stable in the training process, and the convergence speed is higher; the algorithm of the invention can effectively identify and utilize key features really related to labels, improves the adaptability of the model to different data distributions, and can still maintain high detection precision and significantly enhance generalization ability under the condition that the distribution of training data and the distribution of test data are different.

The present invention discloses a coordinated control method for a slow-dynamic unknown coal-fired power generation system based on T-S fuzzy and TD3, comprising the following steps: first, the coal-fired power generation system is decomposed into a fast subsystem and a slow subsystem using singular perturbation theory, and the original control task is decomposed into a stabilization task of the fast subsystem and a tracking task of the slow subsystem. For the fast subsystem, the slow-varying characteristics of steam pressure are utilized to select fuzzy sets on its definition domain to construct a T-S fuzzy model. The fast subsystem controller is obtained by solving the algebraic Riccati equations corresponding to multiple linear systems. For the slow subsystem, a dual-Q network is used to reduce the overestimation of the Q value, and a dynamic learning rate and batch size adjustment mechanism are introduced to accelerate training convergence. The control input of the slow subsystem is obtained by learning under the TD3 algorithm framework. The control inputs of the fast and slow subsystems are combined and acted on the original system to obtain the state information at the next moment. The intelligent agent interacts with the original system to complete collaborative optimization.

The invention discloses a multi-unmanned aerial vehicle cooperative task unloading and trajectory optimization method, and relates to the field of vehicle networking and unmanned aerial vehicle cooperative computing. According to the invention, the space-time attention mechanism of Transform is combined with a multi-agent depth deterministic strategy algorithm framework, and intelligent collaboration and dynamic optimization of an unmanned aerial vehicle group are realized from three aspects of system architecture, feature modeling and decision strategy. The method is composed of four main stages: system architecture construction, feature modeling and state characterization, strategy generation and decision execution, and model training and parameter updating. A closed loop is formed among the four stages, and the whole-process collaboration from environment perception to intelligent decision-making to strategy optimization is realized step by step.

The invention discloses a nonlinear acceleration method suitable for real-time speed control of a stepping motor, which takes a high-efficiency multiply-add recursive form without division and root division as a core, aims to overcome the limitation of a basic algorithm and provides a hybrid algorithm framework which comprises the following steps of: 1, aiming at the problem of insufficient precision of a low-speed region, selecting a high-efficiency multiply-add recursive form with a high-efficiency multiply-add recursive form with a high-efficiency multiply-add recursive form; compensating through a high-order formula or a lookup table (LUT); 2, when the running speed passes through a resonance area of the stepping motor, the acceleration is smoothly increased and recovered by dynamically adjusting a parameter m related to the acceleration in a smooth transition interval, so that the stepping motor rapidly passes through the resonance area while the impact is reduced; and 3, when the speed of the motor approaches to the final target speed, switching to a substitution recursive control method which carries out high-order nonlinear correction according to the virtual step number, and slowly reducing the acceleration to zero in a mathematical optimization manner, thereby generating a perfect S-shaped speed curve turn. The method has real-time performance, flexibility and resonance resistance.

This invention provides an automatic solution method and system for job scheduling problems based on a multi-agent framework. The method includes: acquiring a natural language description of the job scheduling problem and loading functions from a preset heuristic algorithm framework; calling a manager agent to parse the natural language description to obtain specific constraints and objectives, and identifying the target function to be modified in the heuristic algorithm framework based on the specific constraints and objectives; calling a code generation agent to parse the functional description of the target function and the specific constraints and objectives to generate reconstructed function code; calling a code detection agent to verify the reconstructed function code; if the verification fails, calling a code modification agent to modify the reconstructed function code, and calling the code detection agent again to verify the modified reconstructed function code, until the reconstructed function code passes verification; if the verification passes, reconstructing the heuristic algorithm framework based on the reconstructed function code, and executing each function in the reconstructed heuristic algorithm framework to output a job scheduling scheme.

This invention discloses a method for processing error data of smart energy meters based on edge computing, belonging to the technical field of error data processing methods. The invention includes: S1, calculating the mean and standard deviation of the total energy increment sequence within a sliding window, defining a first-order autoregressive hysteresis correction term, setting an adaptive threshold, defining the state machine system state, and finding the computer increment sequence during inactive periods by constructing an inactive period criterion; S2, storing the filtered effective data increments, calculating the energy increment of each sub-meter, defining a continuous zero increment detection function, collecting effective rearranged data points by setting the meter failure detection window and parameters, and constructing an observation matrix X and an observation vector Y; S3, using an improved genetic optimization algorithm framework, calculating eigenvalues ​​and condition numbers, and iteratively obtaining optimal and distinct individuals; S4, determining the optimal regularization parameters using the L-curve method, and calculating the error coefficients of each energy meter using improved Tikhonov regularization.

The invention discloses a multi-stage mixed flow assembly production scheduling optimization method based on a heuristic algorithm. Firstly, a multi-stage mixed flow assembly scheduling model is constructed, production is divided into L stages, each stage contains Jl stations, and a multi-objective function is established. Secondly, initializing a heuristic algorithm framework, and constructing a set of a high-level heuristic (HLH) strategy and a low-level heuristic (LLH) method; through multi-stage iterative optimization, an initial scheduling solution is generated randomly or according to a priority rule, then a new solution is generated by selecting an LLH method based on an HLH strategy, a current solution is updated by using a dominant solution, and inter-stage collaborative parameters are optimized by using a particle swarm algorithm. Data are collected in real time, and a trigger threshold value is dynamically adjusted. And when the scheduling scheme converges or reaches the maximum number of iterations, outputting the optimal scheduling scheme. According to the method, multi-objective optimization and strategy combination are realized through a hyper-heuristic framework, dynamic adjustment and reinforcement learning are performed in combination with a particle swarm algorithm and a triangular fuzzy number, the system coordination is improved, and the method adapts to personalized manufacturing complexity.

The invention discloses an MO-KTO law model enhancement method, device and equipment and a storage medium, and the method comprises the steps: obtaining a plurality of law-related optimization targets, and generating a multi-bit binary signal tag according to the performance of an original law text sample in each optimization target; inputting the law big language model, the multi-bit binary signal data labeled according to the multi-bit binary signal label and each optimization target into an MO-KTO algorithm framework to obtain training process data and an optimized target law big language model; performing multi-dimensional capability evaluation on the target law big language model by using a preset law test set to obtain performance indexes under each optimization target, and judging whether the target law big language model is successfully enhanced or not; performance reduction caused by target conflicts in traditional multi-target reinforcement learning can be effectively avoided; it is ensured that key law dimensions are remarkably improved in a balanced mode, and therefore it is accurately verified that enhancement success of the target law large language model is achieved.

The invention discloses an offline reinforcement learning data enhancement method and device based on uncertainty guide diffusion, and belongs to the technical field of reinforcement learning. The method comprises the following steps: acquiring an offline data set; training a diffusion model to learn off-line data distribution; constructing a guide buffer area in a strategy training process, and storing a high-uncertainty state-action pair; the training classifier distinguishes the line data and the guide buffer area data; a classifier gradient is introduced in the diffusion generation process to serve as a guide item, and synthetic data is generated; and mixing the synthetic data with the original data for strategy training. The device comprises an offline data storage module, a diffusion model generation module, a strategy training module, an uncertainty estimation and guide buffer module, a classifier training module and a data mixing module. According to the method, through an uncertainty guide diffusion mechanism, on the premise that an original algorithm framework is not changed, the problem of Q value over-estimation caused by distribution offset is remarkably relieved, and the generalization performance and the training stability of the strategy are improved.