33results about How to "Improve convergence efficiency" patented technology

The invention discloses a multi-modal remote sensing small target recognition method based on common-differential mode collaborative interactive fusion, and relates to the technical field of image recognition. Performing feature extraction and interactive fusion based on a double-flow backbone feature extraction network in the multi-modal remote sensing small target recognition model to obtain an infrared enhanced fusion feature and a visible light enhanced fusion feature respectively; the infrared enhancement fusion features and the visible light enhancement fusion features output by the corresponding convolution down-sampling module and the output layer in the two backbone feature extraction channels are spliced and then are subjected to neck feature fusion output to obtain a prediction image with a target recognition result. According to the method, infrared and visible light modal features are fully utilized, and common features and difference features of the infrared and visible light modal features are subjected to collaborative modeling and interactive fusion, so that the characterization level of the small target is effectively improved, and recognition of the remote sensing small target is more robust.

The invention discloses a satellite-ground enhanced signal conversion method and a satellite-ground enhanced signal conversion model based on PPP-RTK, which are used for solving the problems that conventional RTK in a ground-free network area fails, and satellite-based PPP-RTK depends on a special terminal and is slow in convergence. Generating an RTCM-OSR correction number by analyzing the SSR correction number, fusing the orbital clock correction, ionosphere and troposphere correction and performing weighted interpolation; the ambiguity is solved in combination with the observation value of the base station, dual-path verification and weighted fusion are carried out in combination with M-W and LAMBDA algorithms, and the wide lane ambiguity fixing reliability is improved; and when the ground-based signal is interrupted for more than 1 second, inheriting and maintaining Kalman filtering continuity by using the state vector and generating a switching identifier. The model integrates a signal processing module, a correction generation module, an ambiguity verification module and a switching control module, so that a conventional RTK terminal obtains centimeter-level positioning (level 2.3 cm and elevation 4.1 cm) without a public network, the interruption recovery is less than or equal to 1.5 seconds, and the scene upgrading cost of power inspection, power transmission monitoring, wind power plant monitoring and the like is reduced.

The invention relates to the technical field of optical fiber wireless communication and mobile forward transmission, in particular to a digital analog optical fiber wireless forward transmission transmitting end parameter self-adaptive optimization method and system based on reinforcement learning. Decomposing the normalized wireless signal into digital parts of all levels and a final analog residual error part through multi-level modulation, and transmitting the digital parts and the final analog residual error part in a time division multiplexing manner; a receiving end represents inter-symbol interference caused by link nonlinear distortion and bandwidth limitation, and the signal-to-noise ratio of a recovery signal is calculated based on the error vector amplitude. Parameters of a transmitting end are optimized and modeled into a Markov decision process, states comprise rounding factors, scaling factors, constellation forming factors and pre-equalization filter tap coefficients of all levels, and actions are discrete increments of the parameters; a reinforcement learning agent is constructed by using a dual deep action value network, and training is updated in combination with experience playback and a target network. The reward function comprises a threshold satisfaction item and a signal-to-noise ratio change item; parameters are interactively updated in the training stage, and online self-adaptive adjustment is performed in the reasoning stage.

The application provides a path planning method and device based on a multi-factor genetic algorithm, node coordinate information in a database is used to construct a planning topology graph, and on this basis, a plurality of task paths, i.e., a plurality of individuals, are generated for each task, all individuals corresponding to each task and corresponding optimization factors are corresponded, through a plurality of rounds of genetic iteration, in the process of each round of genetic iteration, offspring populations are obtained by randomly crossing and mutating each individual, and offspring individuals in the offspring populations vertically inherit optimization factors of parent individuals to maintain population stability, and the offspring individuals are screened according to the optimization factors to obtain a next-generation input population with higher superiority, through a plurality of rounds of iteration, a relatively optimal individual corresponding to each task is obtained as an optimal path of the task, and the convergence efficiency of the algorithm in a multi-task concurrent scenario of path planning and the optimization effect of resource consumption are improved.

The invention discloses a face gear on-machine measurement comprehensive error compensation method based on sequential decoupling. Specifically, a sequential decoupling compensation strategy is provided for solving the problem that the measurement precision is limited due to mutual coupling of a measuring head pre-stroke error and a workpiece installation coordinate system error in the prior art. The method comprises the following steps: firstly, establishing a measuring head pre-stroke error model through standard ball calibration, and performing anisotropic compensation on a tooth surface measuring point by utilizing Delou inner triangulation and a weighted interpolation method; secondly, based on the compensated data, the outer cylindrical surface and the tooth top plane of the face gear are accurately fitted through an iterative optimization algorithm, an accurate measurement coordinate system is established, and the initial phase of the C axis is optimized; and finally, reconstructing an actual tooth surface by adopting a two-stage NURBS curved surface fitting method, and calculating a normal deviation. According to the method, an error propagation chain is effectively cut off through a strict execution sequence, the overall precision and reliability of on-machine measurement are remarkably improved, and a key technical support is provided for face gear closed-loop manufacturing.

The invention discloses a multi-label smell description prediction method, and relates to the field of compound smell prediction, and the method comprises the steps: obtaining compound identification information, molecular structure descriptors and smell label data, and constructing a multi-label smell data set; generating a molecular structure feature vector through a molecular fingerprint coding technology, and extracting a multi-dimensional descriptor reflecting the physicochemical properties of molecules; compressing the molecular fingerprint features to a low-dimensional space through a dimension reduction algorithm; performing unbalanced data processing on the training set, fusing the dimension-reduced molecular fingerprints with the molecular descriptors to form a joint feature matrix, and configuring a class weight balance mechanism and overfitting suppression parameters by adopting a multi-label classification architecture; independently optimizing a probability threshold for each odor label based on the verification set; and outputting a multi-odor label combination prediction result according to the target molecule identification information. According to the scheme, the multi-odor characteristics of the compound can be accurately depicted, and the combined recognition accuracy of the compound odor is remarkably improved.

The invention relates to the technical field of parameter control, in particular to a rotary furnace parameter optimization recommendation method based on data analysis, and the method comprises the steps: obtaining the maximum allowable variable quantity of each control parameter of a rotary furnace; generating an initial population containing a plurality of parameter vectors; and executing an iteration process of the differential evolution algorithm, carrying out iteration for multiple times until a preset termination condition is met, and outputting an optimal recommendation parameter. According to the technical scheme, key parameters which have obvious influence on the quality of finished products can be focused in the iteration process, the recommendation accuracy of the optimal process parameters in a complex industrial baking scene is guaranteed, and intelligent control over the rotary furnace is achieved.

The present application relates to the technical field of underground structure anti-floating engineering, and aims to provide a prestressed anchor rod arrangement method, device and medium based on virtual stiffness iteration. The present application can simplify the coupling calculation between prestress application and real stiffness of the anchor rod, and through iterative optimization on the virtual stiffness of the prestressed anchor rod and arrangement nodes, the reasonable arrangement of the prestressed anchor rod is realized on the basis of meeting the deformation coordination and vertical stress requirements of the upper structure, and the safety and economy of the anti-floating design of the underground structure are improved.

The invention belongs to the technical field of power distribution network leakage current detection, and particularly relates to a CNN and FiLM-based leakage current type intelligent identification method. Comprising the following steps: S1, collecting real-time operation data of a typical power supply area, obtaining leakage current waveforms and related environment characteristic parameters under different working conditions, and constructing an original leakage current sample data set; s2, a CNN-based leakage current classification model is constructed, a FiLM condition modulation module is introduced, a multi-task learning framework is used at the tail of the network, a main task is leakage current type identification, and an auxiliary task is grounding system discrimination; s3, using a weighted cross entropy loss function to alleviate a class sample imbalance problem; an OneCycleLR dynamic learning rate scheduling strategy is introduced; s4, evaluating the performance of the model on the test set, and using the accuracy and the confusion matrix as evaluation indexes; according to the method, high-precision identification of multiple types of faults such as single-phase grounding, arc type electric leakage and direct current system electric leakage is realized, different grounding systems can be adapted, and the accuracy and robustness of system diagnosis are improved.

The application discloses a kind of low-carbon scheduling method, system, equipment and medium of power system, belong to power grid low-carbon scheduling technical field, method includes initialization parameter and constructs collaborative space;By particle swarm module optimization reinforcement learning hyperparameter;Reinforcement learning module interactive experience data generation;Extract priority sample and elite particle realize two-way experience sharing;Converge after output scheduling strategy.System includes initialization module, collaborative space construction module, hyperparameter optimization module, interactive storage module, two-way experience sharing module and strategy output module.The application is coupled by constructing dynamic collaborative optimization space and two-way experience sharing mechanism, particle swarm and deep reinforcement learning.Feedback guides particle swarm to realize adaptive disturbance, avoid local optimum trap.At the same time, spontaneous optimization network parameter, significantly enhance the adaptability and generalization ability of algorithm.The scheme breaks through single algorithm decision bottleneck, improves low-carbon scheduling efficiency and power system reliability.

The invention provides an active power distribution network low-carbon economic dispatching optimization method based on carbon-energy coupling, and belongs to the technical field of power distribution network dispatching. Comprising the following steps: constructing a double-chromosome model consisting of a carbon chromosome and an energetic chromosome; binding carbon-energy coupling constraint for the carbon chromosome and the energy chromosome, and initializing the double chromosome model; constructing a total cost function including an energy cost item and a carbon constraint penalty item; constructing a selection operator based on the total cost function, a crossover operator based on carbon violation feedback and a mutation operator of the genetic algorithm; constructing a local optimization function of the genetic algorithm; and performing iterative optimization on the double-chromosome model by using the optimized genetic algorithm, and obtaining an optimal scheduling scheme after iteration is finished. According to the method, through the double-chromosome model, the unified cost function, the adaptive genetic operator and the enhanced local optimization function, collaborative optimization of economical efficiency and low-carbon targets of the power distribution network can be realized, and a high-quality optimization result is provided for actual dispatching operation.

This disclosure relates to a supply chain network configuration method, apparatus, electronic device, and storage medium, comprising: acquiring a mixed-integer linear programming model; decomposing the model into a main problem and multiple sub-problems, wherein the main problem is used to determine facility location and capacity allocation, and the sub-problems are used to determine logistics flow allocation under various scenarios; iteratively solving the main problem and multiple sub-problems to obtain the upper and lower bounds of the model's optimal value; wherein, during the iterative solution process, effective inequalities are added to the main problem to narrow its feasible region, dual variables are obtained by solving each sub-problem in parallel, and the optimal aspect of each sub-problem is generated based on the dual variables to modify the main problem and optimize the supply chain network planning decision; and outputting the optimal supply chain network configuration scheme when the relative error between the lower and upper bounds is no greater than a preset threshold. By using the method of this disclosure, a supply chain network configuration scheme that satisfies the constraints can be output quickly.

The invention discloses an FPSO mooring system optimization design method, and particularly belongs to the technical field of ocean engineering mooring design. The method comprises the following steps: constructing an initial mooring model; optimization variables are defined, constraint conditions are determined, multiple sets of values of each set of optimization variables are assigned to the initial mooring model, and a multi-objective function is obtained; dynamically analyzing the optimization variables to generate an initial data set; based on the initial data set, establishing a multi-target agent model through an LSTM network; generating a new sample based on the initial data set by adopting an adaptive Latin hypercube sampling method; optimizing by using a genetic algorithm to obtain an optimized solution set; and carrying out real calculation on an optimal solution in the optimal solution set to obtain a simulation result, and judging whether constraint and performance requirements are met or not. According to the method, the problems of high cost and low efficiency of mooring design under complex sea conditions are solved, experience driving of the mooring scheme is changed into intelligent algorithm driving, and reliable technical guarantee is provided for large-scale deployment of the floating platform.

The invention discloses a three-dimensional human body posture estimation method and device and a medium, and belongs to the technical field of computer vision. The method comprises the following steps: extracting two-dimensional human body posture key points based on an acquired video picture to obtain a two-dimensional human body posture key point sequence; projecting the two-dimensional key point sequence to a feature space through nonlinear high-dimensional mapping to obtain a high-dimensional feature space matrix; inputting a three-dimensional human body posture estimation model based on the high-dimensional feature matrix to obtain a three-dimensional human body posture key point sequence; based on the three-dimensional human body posture key point sequence, a three-dimensional human body posture estimation result is obtained through the three-dimensional coordinate point positions. According to the method, through the three-dimensional human body posture estimation model, the anti-interference capability of feature extraction is enhanced, and the robustness of three-dimensional posture estimation in a complex dynamic scene is remarkably improved.

The invention provides a method, a system, equipment and a medium for toughness partitioning and fault recovery of a power distribution cooperative network, and the method comprises the steps: constructing a dynamic weighted undirected graph based on a node admittance matrix, and introducing the physical connection state and operation characteristics of the power distribution cooperative network into a partitioning modeling process; and the partitioning result can truly reflect the line coupling strength and the change of the equipment operation state. Initial population construction is carried out by adopting a position-based node coding strategy and combining historical operation data, so that an invalid search space is reduced. The internal stability of the power supply island and the partition adjustment cost are taken as optimization objectives, and a partition scheme is iteratively optimized by using a multi-objective non-dominated sorting genetic algorithm NSGA-II, so that the operation and maintenance cost is reduced. According to the method, the dynamic toughness partition map is generated from the Pareto optimal partition scheme, and the scheduling scheme is generated, so that emergency power supply scheduling and line maintenance decisions can be cooperatively generated with partition results, and the pertinence and overall efficiency of fault recovery are improved.

The application discloses a method for generating a malicious traffic detection model, a storage medium and an electronic device. The method comprises: obtaining a plurality of Tor traffic data, and constructing the plurality of Tor traffic data into a plurality of standardized traffic sequences respectively. A plurality of different scale feature data of the plurality of traffic sequences are determined, and the plurality of different scale feature data comprises a first behavior feature of a data block scale, a transmission direction feature of a traffic sequence scale and a second behavior feature of a data flow scale. Training samples are constructed based on the first behavior feature, the transmission direction feature and the second behavior feature of the plurality of traffic sequences respectively. The classification model is trained by using the training samples to obtain a detection model, and the detection model can determine whether the to-be-detected Tor traffic data is malicious traffic. By using the three scale features, the traffic is described from three dimensions of microscopic interaction, direction mode and macroscopic evolution, the one-sidedness of a single granularity feature is overcome, and accurate identification of the malicious traffic is realized.

The invention relates to a DDCPN-based time delay precision balance optimization method in vehicle infrastructure cooperative infrastructure perception, which belongs to the field of Internet of Vehicles and comprises the following steps: constructing an Internet of Vehicles system consisting of a plurality of vehicles and a single RSU; establishing an optimization problem to minimize the total delay of all task vehicles completing the perception task within the duration; modeling the optimization problem as a Markov decision process MDP; and task unloading and resource allocation optimization are carried out based on a discrete drive continuous policy network DDCPN algorithm. The algorithm has good convergence efficiency. Simulation results show that the algorithm has a faster convergence rate under the constraint conditions of meeting fast reasoning precision, energy consumption, computing resource utilization rate and the like, effectively reduces task execution delay, and better adapts to a dynamic and complex vehicle-mounted network environment.

The application provides a hierarchical collaborative optimization and settlement method for large-scale industrial demand response, relates to the technical field of power systems, and comprises the following steps: dividing large-scale industrial demand response resources into multiple levels and constructing an optimization model; decomposing a global optimization problem into local sub-problems, defining consistent variable constraints for global consistency, and generating scheduling instructions through iterative calculation; obtaining actual response data in response to the scheduling instructions, performing hash operation and digital signature processing on the actual response data to generate a signed digest; verifying the signed digest by using a smart contract, executing real-time settlement after verification to obtain real-time settlement data, and summarizing to obtain periodic settlement data; calculating performance indicators according to the scheduling instructions and the actual response data, and performing numerical correction on the real-time settlement data and the periodic settlement data by using the performance indicators to obtain adjusted settlement data. The application realizes fast scheduling and convergence of large-scale resources, and effectively improves the enthusiasm and fairness of users participating in scheduling.

This invention discloses a digital human portrait modeling method based on reinforcement learning, belonging to the field of computer vision technology, including the following steps: S1, registering multi-view identity images with 3D model parameters to form a portrait object; S2, dividing category intervals and writing identity visual evidence to form a detection object; S3, connecting to an improved DreamerV3 model, changing the RSSM classification random state to a pairwise cumulative category boundary state; S4, advancing the same-direction boundary, updating the posterior random state with the intersection of the identity visual evidence interval and the advancing range; S5, aggregating the category embeddings between boundaries, generating candidate actions, and calculating the boundary shrinkage amount; S6, using the egret flock optimization algorithm to determine the optimized action, rendering the image and updating the posterior random state; S7, writing the stable boundary interval into the 3D model to form the portrait result. This invention achieves adaptive shrinkage of the local parameter range of digital humans and multi-view identity consistency portrait modeling.

The present application relates to the field of optical microscopic imaging technology, and provides a three-dimensional differential phase contrast refractive index tomographic reconstruction method, system, device and program, the method comprising: collecting multi-mode intensity data; the multi-mode intensity comprises a three-dimensional intensity image stack and a focus and defocus angle change intensity set; determining a transfer function according to the three-dimensional intensity image stack, and obtaining an initial scattering potential spectrum by using a deconvolution algorithm; based on the focus and defocus angle change intensity set, performing a loop single LED intensity constraint update on the initial scattering potential spectrum to obtain a scattering potential estimate; using the three-dimensional intensity image stack, performing a linear gradient constraint update on the scattering potential estimate, and executing the linear gradient constraint update cyclically until a convergence condition is met, performing a three-dimensional inverse Fourier transform according to the final obtained scattering potential spectrum, and outputting a three-dimensional refractive index distribution result. The present application substantially improves the accuracy and quality of three-dimensional refractive index reconstruction while keeping the advantages of non-interference and label-free imaging and the simplicity of system hardware.

The invention provides a rapid ICP matching method in combination with semantic priori constraint of a semantic image, and belongs to the technical field of machine vision and machine perception. The method comprises the following steps: synchronously inputting an RGB image frame and a point cloud frame, obtaining a rotation prior initialization attitude by utilizing PTZ equipment, extracting image semantic features by adopting a deep semantic segmentation network, and realizing semantic correspondence between a point cloud and an image and constructing a reinforcement learning adaptive strategy through a calibration matrix. According to the method, a semantic-geometry joint error function is innovatively introduced, Lie algebraic increment optimization and a self-adaptive step length mechanism are combined, and the registration contribution of a significant structure is enhanced through a semantic region weighting strategy. Compared with a traditional ICP algorithm, the method has the advantages that registration stability in a structured environment is improved through semantic constraint, learning is reinforced, dynamic optimization parameters are optimized, convergence instability is avoided, PTZ prior reduction and optimization of a search space are achieved, high-precision and robust real-time point cloud registration is achieved through a comprehensive strategy, and matching efficiency and precision in a complex scene are remarkably improved.

This invention discloses an intelligent optimization method for the coordinated scheduling of cascade reservoir groups oriented towards multi-objective equilibrium, belonging to the field of swarm intelligence technology. The method includes: real-time acquisition of time-series monitoring data, simultaneously acquiring hydrological ensemble forecast data; generating a set of feasible scheduling trajectories satisfying rigid constraints using a fast non-dominated sorting algorithm, and generating an initial scheduling strategy population by combining a typical scenario library and a transfer learning mechanism; using a robust optimization algorithm to perform rolling solutions on the feasible scheduling trajectories and the initial strategy population, generating water level control intervals and unit output command sets; and using a multi-objective decision visualization platform, displaying the water level change intervals and risk probabilities of each scheme with a 3D interactive interface, generating Pareto front schemes for dispatchers to select and execute based on real-time preferences. This invention can effectively improve the scheduling robustness and multi-objective equilibrium capability of cascade reservoir groups under hydrological uncertainty, providing intelligent decision support for the safe and efficient operation of reservoir groups.

The invention discloses a ship common rail oil injector health assessment method and system, and relates to the field of ship equipment health monitoring, and the method comprises the steps: obtaining a normal pressure signal of a common rail pipe of an oil injector in normal work; preprocessing the normal pressure signal, extracting normal time domain features, and normalizing the normal time domain features; constructing and training a self-organizing mapping neural network by using the normalized normal time domain features to obtain an optimal matching unit and obtain a reference distance threshold value; wherein the optimal matching unit is a node with the minimum Euclidean distance from the input normal time domain feature in an output layer of the self-organizing mapping neural network; s2, processing the real-time pressure signals collected in real time according to the step S2, obtaining real-time time domain features, calculating the Euclidean distance between the real-time time domain features and the optimal matching unit, and comparing the Euclidean distance with a reference distance threshold value; and when the reference distance threshold value is larger, it is judged that the oil injector is in a healthy state. The method has the effect of high real-time performance.

The invention provides an equilibrium strategy generation method, device and equipment for a large-scale unmanned cluster game, relates to the technical field of large-scale autonomous system decision control, and aims to solve the defects of poor stability and low calculation efficiency when an equilibrium strategy is generated for a large-scale unmanned cluster in the prior art. And efficient collaboration and autonomous decision-making of a large-scale cluster in a complex environment are realized. Comprising the following steps: dividing devices included in a device cluster into a plurality of populations based on characteristic parameters of each device included in the device cluster; constructing an average field game model of the plurality of populations, wherein the average field game model is used for reflecting an interaction effect among the plurality of populations; iteration is carried out based on a target mode and the average field game model, a balance control strategy of the equipment cluster is obtained, and the target mode comprises at least one of exponential weighted aggregation, near-regularization, bootstrap regularization and heuristic acceleration; and controlling the equipment cluster to execute the target task based on the balance control strategy.

The invention provides a dynamic compilation method, system and equipment for multiple tasks of an unmanned aerial vehicle and a medium. The method comprises the following steps: constructing a killing chain element path model through task logic of the unmanned aerial vehicle to obtain killing chain element path expressions, and grouping high-correlation killing chain element path expressions through feature vector construction and cosine similarity calculation to obtain task groups; based on the degree of adaptation of the unmanned aerial vehicle and the task group, generating an initial population, constructing a fitness function, and evaluating the individual fitness of the initial population, thereby improving the proportion of an initial effective solution, reducing the elimination cost of an invalid solution in the earlier stage of iteration, and accelerating the convergence efficiency; iterative optimization is carried out by adopting a genetic operator method and reinforcement learning based on individual fitness, the problem of poor adaptability in different scenes caused by manual parameter adjustment of a traditional genetic algorithm is solved, the applicability of an optimization result is improved, and multi-task compilation of the unmanned aerial vehicle can be dynamically provided in real time.

The invention discloses a multi-unmanned aerial vehicle urban inspection task allocation method considering a time-sensitive target, and relates to the technical field of task allocation. According to the method, photoelectric effective load capacity and target characteristics in actual city inspection are considered, heterogeneous targets are divided into point targets, line targets, surface targets and body targets, and a three-dimensional Dubins dynamic multi-vehicle path model considering a time window is established, so that the difference between an estimated path and an actual path is reduced, and the accuracy of the estimated path is improved. Therefore, the feasibility and accuracy of task allocation in the urban environment are improved; a self-adaptive large neighborhood search algorithm of a dynamic priority strategy is adopted, destruction and repair operators with time window constraints are designed to improve convergence efficiency, and local optimum is avoided in combination with a simulated annealing mechanism. Besides, a priority criterion strategy of a dynamic target is customized and integrated into an adaptive large neighborhood search framework, so that the target with a dynamic time window is processed in real time, and efficient planning of a reconnaissance task allocation scheme in a dynamic environment is realized.

The invention relates to an intelligent routing system and method based on a digital twin network and a neural network, and the system comprises a digital twin scene engine module which is used for constructing topologies corresponding to real networks in a one-to-one manner, and supporting the path planning of an initial path, cutover drilling, business migration, and a new business scene; the output end of the digital twin scene engine module is connected with the data acquisition module, and the data acquisition module is used for acquiring link information and forming graph structure features during the scene operation period; the output end of the data acquisition module is connected with a neural network strategy module, and the neural network strategy module is used for realizing path planning by adopting a graph neural network based on the graph structure characteristics and by taking a Qos service index as an optimization target; the output end of the neural network strategy module is connected with the service scene implementation module, and the service scene implementation module is used for implementing a predetermined service scene. According to the method, the optimal path in different operation and maintenance scenes such as cutover drilling, business migration and new business can be quickly planned.

The invention discloses a hybrid electric vehicle energy management method considering working condition recognition, and the method comprises the steps: carrying out the real-time classification of vehicle driving working conditions based on a working condition recognition network of a long and short term memory attention mechanism, and designing a self-adaptive adversarial dual depth Q network AD-D3QN energy management algorithm based on the classification. According to the algorithm, dynamic planning expert demonstration experience and a reward weight self-adaptive adjustment mechanism are combined, and optimal distribution of engine and motor torque is achieved. The method has the advantages that working condition adaptability is improved, the training convergence speed is increased, and economical efficiency and stability are improved.

This invention provides a method, apparatus, computer device, and storage medium for robot navigation, belonging to the field of robot navigation. The method includes: real-time acquisition of the target distance between the robot and its destination position, and the obstacle distance to the nearest obstacle; constructing two algorithms to control the robot's discrete and continuous actions respectively; determining the attractive and repulsive potential fields respectively, and constructing an artificial potential field based on the attractive and repulsive potential fields; combining the two algorithms to construct a unified navigation model based on the artificial potential field; determining the robot's potential energy change based on the artificial potential field, using the potential energy change of the robot at each step as a reward function to guide the robot to move in the direction of decreasing potential energy; and determining that the robot has completed navigation when the target distance is less than a preset threshold. Thus, this discrete-continuous action hybrid control mode can adapt to complex environments and provide the robot with a precise navigation path.