40results about How to "Avoid local optima" patented technology

The invention relates to an Internet of Vehicles computing power scheduling method based on a graph neural network and deep reinforcement learning, and the method comprises the steps: S1, collecting the state information of network entities in the Internet of Vehicles and the association information between the entities in real time, and constructing a dynamic time-space attribute graph; s2, inputting the dynamic space-time attribute graph into a pre-trained graph neural network encoder, and outputting a node embedding feature set containing a high-order topological relation and global graph embedding features through multi-layer message passing and feature aggregation; s3, the node embedding feature set and / or the global graph embedding feature are / is used as the state of a deep reinforcement learning agent and input to a strategy network, the computing power scheduling action at the current moment is output, and the computing power scheduling action comprises the steps of assigning an unloading target node for a to-be-processed computing task and distributing corresponding computing and communication resources; s4, dispatching actions are distributed to the corresponding network entities to be executed, environment feedback is collected and used for model updating and next round of dispatching, and the method has the advantages of improving dispatching efficiency, adaptability and system energy efficiency and the like.

The invention relates to the technical field of control, in particular to an intelligent control method and system for a photovoltaic adhesive production line, and the method comprises the steps: obtaining multi-source technological parameters of the photovoltaic adhesive production line, and predicting the production quality of photovoltaic adhesive through a preset prediction model; calculating the instantaneous fluctuation degree of the multi-source process parameters in the production process and the prediction error of the prediction model in the historical time; calculating a prediction deviation degree of the prediction model according to the instantaneous fluctuation degree and the prediction error; and taking a sum value of the predicted deviation degree and a prediction result of the prediction model as a correction value of the production quality of the photovoltaic adhesive, and when the correction value exceeds a preset range, adjusting the multi-source process parameters. The stability of the production quality and the production efficiency of the photovoltaic adhesive production line are improved.

This invention provides a method and system for UAV path planning based on a fusion algorithm, relating to the field of UAV autonomous navigation and path planning technology. The method includes the following steps: acquiring the starting coordinates, ending coordinates, and obstacle information in the flight environment of a flight mission, and constructing a flight environment spatial model; defining path constraints and optimization objectives, and generating an initial set of feasible paths based on the flight environment spatial model, path constraints, and optimization objectives; performing weighted fusion and guided enhancement on the initial set of feasible paths to obtain a weighted fused path and a guided enhanced path; employing a particle swarm optimization algorithm to construct an initial particle swarm based on the initial feasible path set, the weighted fused path, and the guided enhanced path, and iteratively optimizing the initial particle swarm to obtain a globally optimal particle; and decoding the globally optimal particle to obtain the final flight path. This invention significantly improves the efficiency and quality of UAV path planning in complex static obstacle environments.

The application discloses a new energy station frequency control parameter collaborative setting method and application thereof, and the method comprises the following steps: constructing an optimized parameter vector x containing a network type and a network type station heterogeneous control parameter; establishing a comprehensive objective function f(x) which fuses the system frequency deviation minimization and the parameter adjustment cost minimization, and generating an initial training sample database corresponding to the comprehensive objective function based on sampling; constructing an anisotropic Kriging surrogate model based on the initial training sample database; constructing a sparse degree weighted adaptive expected improvement SW-AEI adding point criterion based on the anisotropic Kriging surrogate model, and screening an optimal candidate vector with the maximum sparse degree weighted adaptive expected improvement value; updating a new sample to the training sample database to recalibrate the anisotropic sensitivity hyperparameters in the anisotropic Kriging surrogate model until a convergence condition is met, and outputting the optimal collaborative control parameter.

The invention discloses an interactive intelligent teaching system based on multiple modes, and relates to the technical field of intelligent teaching management systems. The system comprises a multi-modal perception layer, a semantic and feature fusion engine, a cognition-emotion two-dimensional analysis module, a strategy decision center and a multi-dimensional interaction execution terminal. Multi-source data of vision, hearing, physiology, behaviors and the like are synchronously collected in real time, and an improved MLP regression network is utilized to construct a cognition-emotion two-dimensional state model of a learner; when the state of the student is monitored to be abnormal, the strategy decision center dynamically generates an optimal teaching path in combination with a knowledge graph and a historical successful case, and drives a virtual digital teacher to execute immersive intervention; according to the method, the problems of emotion perception deficiency, feedback lagging and insufficient individuation in traditional online teaching are solved, and intelligent closed loop and adaptive optimization of the teaching process are realized.

This invention discloses an intelligent path planning method for heavy-duty robotic arms used in the dismantling of decommissioned photovoltaic (PV) modules. The method involves modeling and training the dismantling task of decommissioned PV modules using an actor-critic algorithm, followed by intelligent path planning for the heavy-duty robotic arm based on the constructed model, generating locally optimal work trajectories. Heavy-duty robotic arms face scenarios with few obstacles in the dismantling of decommissioned PV modules, and their path planning problem can be reduced to an optimization task for a single robotic arm. This solution employs an actor-critic algorithm to achieve efficient and stable path generation, supporting automated dismantling. Furthermore, it introduces the concept of entropy to enhance exploratory capabilities, avoid local optima, and ensure the robotic arm can adapt to the irregular shapes of the modules and dynamic environmental changes.

The invention relates to a dynamic environment autonomous navigation method and system based on sub-target generation hierarchical reinforcement learning. The method comprises the following steps: constructing a double-layer framework in which a top layer is responsible for target scheduling and a bottom layer is responsible for high-frequency execution; the top layer constructs a local occupation grid map according to the laser radar, constructs strip-shaped sub-target probability distribution based on a global reference path, and regulates and controls the entropy of the strip-shaped sub-target probability distribution and the commitment time limit between the top layer and the bottom layer through the environment self-adaptive exploration temperature; and the bottom layer follows the current sub-target within the committed time limit, so that the robot drives to the real-time sub-target. According to the method, strip-shaped probability sub-target distribution is constructed around a global reference path, the distribution entropy and the underlying committed time limit are regulated and controlled by utilizing environment feature-driven exploration temperature single knob linkage, and sub-target screening and issuing are completed in combination with safety expansion and reachability constraint; and a bottom-layer high-frequency closed-loop actuator follows within a commitment period and triggers advanced re-mining when the risk is increased, so that safe and smooth real-time arrival of a final target is realized.

The application discloses a kind of factory inspection robot path planning methods based on improved big cane mouse algorithm and dynamic window method fusion, it is related to the field of path planning, this method first constructs factory workshop three-dimensional environment model;Second, the triple improvement is carried out to basic big cane mouse algorithm:design adaptive weight linearly decreasing with iteration number, dynamically balance global exploration and local development capability;Introduce sine-cosine strategy, through specific parameter control and sine-cosine direction optimization, search dimension is enriched to avoid local optimum;Adaptive factor is embedded to core calculation formula, introduce nonlinear function, improve global path planning precision and convergence speed, simultaneously, fusion dynamic window method constructs "global-local" double-layer path planning framework.Experiments show that, in complex dynamic environment of factory, global path planning error is reduced by 15%-20%, dynamic obstacle avoidance response time is shortened to within 0.3s, can satisfy robot all-weather inspection demand, guarantee inspection efficiency and operation safety.

The application discloses a kind of weakly supervised image semantic understanding methods based on multi-task learning, comprising the following steps: obtaining task missing image, constructing multi-level task sharing encoder, extracting high-level semantic information layer by layer, input corresponding decoder branch;Construct public space-task space feature mapping module, through the unaligned task fusion module and task interaction mapping module, update each subtask feature by mapping;Task adaptive feature update module is constructed, and multi-level iterative update unaligned task feature;Task adaptive weakly supervised image semantic understanding framework is constructed, model loss function is established, image data with task missing is input into model, and obtains multi-task prediction result such as semantic segmentation, depth estimation, surface normal estimation.The application is according to the data information of task label unaligned, through the mapping interaction of public space and task space, fully fuses unaligned task feature, iteratively generates high-quality multi-task prediction result, can effectively handle weakly supervised problem with task missing, and simultaneously improves each task prediction accuracy.

The application discloses an optimization method of a rolling wing blade connecting rod mechanism, and belongs to the field of rolling wing aircrafts.The blade swing angle is represented by structural parameters of the blade connecting rod mechanism, and the structural parameters of the blade connecting rod mechanism, including the lengths of the connecting rods, the coordinates of key hinge points and configuration bias, are defined as a variable vector; multiple target functions are constructed for multiple key working conditions such as hovering, forward flight and transition, the multiple target functions not only quantify the performance of the connecting rod mechanism itself, such as kinematic fitting error with a target motion law, dynamic stability index and force transmission efficiency index, but also cover aerodynamic efficiency index and aerodynamic acoustic index which are directly related to the performance of the whole machine; and the optimal structural parameters of the blade connecting rod mechanism are obtained by solving the multiple target functions under constraints, so that the global optimization of the structural parameters of the blade connecting rod mechanism under multiple working conditions, multiple targets and multiple constraints can be realized.

This invention discloses a method for assessing the health status of wind turbine blades based on LNN and an improved LightGBM model, relating to the field of wind turbine health status assessment technology, particularly for assessing the health status of wind turbine blades. The method involves acquiring acoustic signature signals, vibration signals, and SCADA data from the wind turbine, and preprocessing the data. An attention mechanism is used to fuse acoustic signature and vibration features, and a multi-dimensional feature matrix is ​​constructed using SCADA data. Dynamic feature depth extraction is performed using a liquid neural network (LNN) to obtain a high-dimensional feature vector characterizing the blade's health status. This vector is then input into the improved LightGBM model to assess the blade's health status level. This invention achieves accurate assessment of blade health status by correlating and reconstructing acoustic signature, vibration signals, and SCADA data, combining the advantages of LNN in capturing dynamic features with the efficient classification capabilities of the improved LightGBM model, thus providing a reliable decision-making basis for wind turbine blade maintenance.

This invention discloses a field-circuit collaborative analysis method and system based on reinforcement learning and Bayesian optimization. The invention utilizes a Bayesian optimization algorithm to optimize the parameter values ​​of each electronic component in a selected topology. Based on the optimized parameter values, a reinforcement learning-based parameter search environment for electronic components in the RF device circuit topology is constructed, modeled as a Markov decision process. An equivalent circuit model of the RF device is designed according to the optimal action. The equivalent circuit model of the RF device is connected to other circuit systems for circuit-level simulation analysis. By integrating reinforcement learning and Bayesian optimization, two intelligent algorithms, and fully leveraging their respective advantages in discrete structure search and continuous parameter optimization, a high degree of automation and intelligence is achieved throughout the entire field-circuit collaborative analysis process. While ensuring the physical rationality of the model, the efficiency and accuracy of the collaborative analysis are significantly improved.

This invention discloses a method for constructing a road cavity detection model based on 3D ground-penetrating radar (GPR). The method includes constructing a hybrid spatiotemporal detection model, which comprises a feature preprocessing module, a soft clipping module, a spatiotemporal feature extraction module, and an output module. The feature preprocessing module performs feature pre-extraction on preprocessed and labeled data, outputting a pre-extracted feature map. The soft clipping module performs spatiotemporal perception on the pre-extracted feature map, outputting a clipped feature map. The spatiotemporal feature extraction module analyzes the clipped feature map, outputting feature information with global spatiotemporal information. The output module integrates the feature information and predicts the existence of cavities and the range of their occurrence channels. By combining the correlations between data in the spatiotemporal dimension, it accurately detects underground cavities, solving the problems of high computational load and difficulty in adjusting model parameters in existing 3D GPR high-resolution data interpretation. This provides an efficient and reliable technical solution for intelligent road cavity detection and infrastructure safety maintenance.

The invention provides an energy interconnection framework parameter matching design method and system and a medium, and belongs to the technical field of power systems, and the method comprises the steps: determining energy interconnection framework parameters, including a framework type, a port type and a port number; determining constraint conditions of the energy interconnection framework parameters, constructing full life cycle use efficiency as a target function, calculating a fitness value according to the target function, optimizing the energy interconnection framework parameters through an improved litsea rotundifolia optimization algorithm, and obtaining the energy interconnection framework parameters with the optimal fitness value; and determining an energy interconnection framework parameter matching design scheme according to the optimal energy interconnection framework parameter. According to the method, the optimal energy interconnection framework parameter matching design scheme can be quickly generated, and the energy interconnection framework formed by the design scheme can realize the optimal use efficiency.

The invention provides a transportation line combination method, electronic equipment, a storage medium and a computer program product. The method comprises the following steps: acquiring a plurality of transportation lines in a logistics network; according to a line combination constraint, combining the plurality of transportation lines to obtain an initial line set; wherein the route combination constraint comprises a first deadhead constraint, and the first deadhead constraint is used for constraining the deadhead distance and / or deadhead time between different routes in the same initial route combination; and taking the initial line set as a line combination result and outputting the line combination result. In the transportation line combination process, the empty driving condition is considered, so that the initial line set obtained after combination has good performance in the empty driving dimension, and the vehicle empty driving condition is relieved.

The application provides a medical image segmentation method and system based on a CNN-Transformer parallel encoder, and relates to the technical field of medical image segmentation, and the specific steps comprise: inputting a standardized medical image into a parallel encoder, extracting local texture and global semantic feature maps, and aligning the resolution and spatial position through a spatial correlation matrix; screening lesion features based on an anatomical structure prior feature set, dynamically allocating weights to obtain fusion features; taking a lesion gold standard mask as a label, obtaining an initial lesion probability feature map through a decoder, and calculating a bias value; iteratively optimizing a parameter combination to output an optimal configuration parameter combination, and obtaining a final lesion probability feature map, which is classified and activated, thresholded, and outputted as a segmentation mask. The application effectively avoids the local optimal solution of parameter optimization, combines a bias value threshold-driven rapid screening mechanism, fine-tunes parameters for different modal images without full retraining, and effectively breaks through the bottleneck of poor adaptability of the prior art.

The invention provides a risk-adaptive robot hierarchical social navigation method and system, and relates to the technical field of strategy optimization, and the method comprises the steps: determining a speed fluctuation index and an acceleration fluctuation index based on the speed vector data of a pedestrian sample; performing weighted fusion on the speed fluctuation index and the acceleration fluctuation index to obtain an original uncertainty score, performing normalization on the original uncertainty score, and performing linear mapping to a preset unpredictability interval to obtain an unpredictability score; taking the robot state data sample, the pedestrian state data sample and the unpredictability score as input, taking a preset navigation point of the robot, an expected cruising speed of the robot and a prudent coefficient as output, and performing iterative training and optimization on the initial prediction network model in combination with near-end strategy optimization to obtain a target prediction network model; and inputting the current robot state data, the pedestrian state data and the unpredictability score into the target prediction network model, and determining a current navigation strategy.

The application belongs to the technical field of medical instrument navigation, and particularly relates to a needle knife navigation display system and method based on real-time ultrasound images. The system comprises a sequence graph construction module, a dynamic modeling module, a correlation tensor extraction module and an optimal navigation instruction generation module. Lesion images are collected by a high-frequency ultrasound probe and acoustic features are extracted. An infrared optical positioning camera acquires the probe and needle position information. A coupled sequence graph is generated through dynamic time warping and double attention fusion. The sequence graph is analyzed to determine the topological configuration and identify the cooperative morphological label. The configuration adaptive parameters are calculated to output the cooperative response prediction value. A cooperative phase space is constructed through space-time difference operation to extract the weighted space correlation feature tensor. The optimal navigation instruction is generated by embedding the simulated annealing algorithm, and is dynamically updated in real time and triggers an early warning when the cooperation or correlation intensity is low. The application improves the precision and stability of the needle knife navigation and ensures safe and efficient clinical operation.

The application discloses a dynamic AGV scheduling and conflict collaborative processing method and system based on DIWO-PVNS hybrid optimization, and belongs to the technical field of multi-AGV scheduling of intelligent production workshops. The method comprises the following steps: S1, establishing a mixed integer linear programming model; S2, establishing a real-time task list for event-triggered dynamic scheduling; S3, solving an optimal delay departure time of AGV; S4, adaptively adjusting algorithm parameters based on parameter level division of the number of tasks; S5, adopting a combination method for population initialization; S6, adopting a discrete invasive weed optimization algorithm to perform global search and optimization on the initial population to obtain an optimal solution; S7, adopting a population-based variable neighborhood search algorithm to perform local deep optimization on the optimal solution; S8, real-time response and rescheduling strategy of special working conditions; and S9, merging continuous faults into one-time rescheduling triggering. The application realizes dynamic scheduling and global conflict collaborative processing of multi-AGV in a workshop.

The application discloses a wireless power transmission system multi-parameter collaborative optimization method based on a VPSR-PSO algorithm, and belongs to the technical field of wireless power transmission, and comprises the following steps: step 1: according to the primary side input voltage, the primary side input current, the primary side output current, the primary side compensation inductance, the primary side compensation inductance internal resistance, the primary side coil self-induction internal resistance and the primary side coil self-induction of the system, the secondary side output current, the secondary side coil self-induction, the primary side coil series capacitor, the secondary side coil self-induction internal resistance and the secondary side coil series capacitor, and the mutual inductance between the primary side coil and the secondary side coil; step 2: constructing a target model according to the primary side compensation inductance, the mutual inductance between the primary side coil and the secondary side coil and the equivalent load resistance; and establishing the constraint condition of the target model; the application constructs a multi-objective model for simultaneously optimizing transmission efficiency and output power, and realizes efficient search and balanced regulation and control.

This invention discloses a reinforcement learning-based satellite pursuit and escape method and system based on orbital maneuver prediction. The method constructs a SAC policy network, designs a suitable loss function for backpropagation, and trains the initial policy of the model. A curiosity mechanism is introduced into the SAC algorithm, using the error between the prediction results and the actual values ​​from a fully connected network as a reward signal to encourage the agent to explore the action space. An input vector format and dataset for Transformer training are constructed, and the input vectors are positionally encoded upon input. A suitable loss function is designed for backpropagation of the Transformer. The Transformer's output is used as the input to the SAC algorithm, and the SAC observations are reconstructed, forming an integrated decision network. This invention solves the game problem of satellite pursuit and escape using pulse maneuvers.

The application discloses an Android application testing method based on Monte Carlo tree search and deep reinforcement learning, and belongs to the technical field of software testing.The method adopts deep reinforcement learning to perform adaptive learning on a testing strategy, a fine-grained state representation mode is designed for an application program interface state, a new reward function is used during exploration, state changes caused by interface jumps can be rewarded, and fine-grained control position and text changes in the interface can also be rewarded, so that the learned testing strategy is more comprehensive and detailed, the Monte Carlo tree search method is used to optimize the testing strategy, potential states are provided with opportunities for long-term exploration, and local optimization is avoided.The method can continuously reach some application program states that are difficult to traverse previously, high-code-coverage Android application program testing is realized, and the code coverage and fault detection rate performance of Android application testing are improved.

The application discloses a kind of intelligent decision system construction methods for traffic signal control.The method includes model training and deployment phase and application and evolution phase two parts, as follows: model training and deployment phase: S1, based on pre-stored road traffic conflict rules, generate traffic state structured data and the pairing training sample set of no conflict signal control instruction;Supervised fine-tuning is carried out to large language model using the pairing training sample set, and a basic model is obtained;S2, the basic model is accessed to traffic simulation environment and is reinforced learning training;In each training step, the basic model outputs signal control instruction according to current traffic state, and safety verification is carried out to the instruction according to the road traffic conflict rule, and generates safety reward signal, etc.The application constructs a safe and reliable, continuous evolution, and suitable for edge independent deployment traffic signal intelligent decision system.

The application is a crowd-sensing task allocation method based on proficiency cooperation. Firstly, the crowd-sensing scene is divided into multiple areas, and the users in each area are divided into a group, and one group leader is selected in each group. Secondly, the sensing platform sends a task set to the users. The users sort the tasks from early to late according to the deadlines, and obtain a task queue. When the users receive a new task, the new task is inserted into the task queue according to the deadline of the new task. Then, when the task quantity of the users does not increase suddenly, the users execute the tasks according to the order of the task queue. When the task quantity increases suddenly, the users send the task queue to the group leader. If the task quantity of the group leader increases suddenly, the group leader feeds back to the sensing platform and requests the nearby user group for assistance. The sudden increase of the task quantity refers to that the ratio of the to-be-completed task quantity of the user to the processing speed is greater than a set threshold. Finally, the group leader obtains the global optimal solution of the target function on the device of the group leader by using a swarm intelligence algorithm, and allocates the tasks in the task queue according to the global optimal solution. The method considers the balance of the task quantity and the balance of the processing speed of each user while striving to minimize the overall task completion time, thereby effectively reducing the operating cost.

The application relates to an energy storage power station intraday scheduling method based on a rule base hybrid model predictive control, wherein the method comprises the following steps: constructing a rule base considering actual operation of intraday optimization scheduling based on historical data of a target energy storage power station and scene application demand; constructing an improved particle swarm algorithm-model predictive control optimization model and generating an optimization prediction strategy fused with an adaptive mechanism; establishing an intraday scheduling hybrid optimization framework based on the rule base and the improved particle swarm algorithm-model predictive control optimization model; and generating an intraday scheduling optimization strategy based on the intraday scheduling hybrid optimization framework, the optimization prediction strategy, power deviation and operating conditions of the target energy storage power station. Thus, the problems in the prior art that the heuristic algorithm is highly dependent on initial parameter setting, is prone to falling into local optimization or divergence and non-convergence, and lacks an effective scheme for realizing rapid external feedback and internal rapid adjustment of an energy storage cluster without affecting a day-ahead planning curve are solved.

The application belongs to the technical field of unmanned aerial vehicle cooperative formation, and specifically discloses a layered intelligent-based unmanned aerial vehicle formation cooperative task planning method, which comprises the following steps: based on an improved deep Q network algorithm, an original formation is split into multiple sub-formations according to formation characteristics and target value information, and target distribution is performed; based on a position matching algorithm of bearing and distance, each unmanned aerial vehicle is smoothly transitioned from a current formation position to a target sub-formation formation position; based on an improved multi-population cooperative evolution algorithm, a cooperative track from a starting point to a target point is planned for each sub-formation, collision detection and track output are performed, and unmanned aerial vehicle formation cooperative task planning is completed. The application solves the problems that the prior art is difficult to adapt to a complex environment, high efficiency, stability and safety cannot be ensured when the original formation is split into sub-formations, and the cooperative nature of a track planned by a traditional genetic algorithm is insufficient.

The invention discloses a fuzzy PID (Proportion Integration Differentiation) control parameter online optimization method based on multi-population collaborative PSO (Particle Swarm Optimization), which comprises the following steps of: establishing a mathematical model of a control object, and solving a transfer function of the control object according to structural characteristics and a working principle of a system; establishing a fuzzy PID controller, setting a fuzzy rule and configuring related parameter values; operating an improved PSO particle swarm algorithm, wherein an algorithm optimization space is a value interval of a proportionality coefficient, an integral coefficient and a differential coefficient corresponding to the fuzzy rule; the speed and the position of each particle are updated, the fitness value of each particle is obtained in the learning process, and the extreme values of individual particles and population particles are updated in each iteration; and optimizing the output parameters of the fuzzy controller according to the result of the PSO particle swarm algorithm. According to the method, a gradient-oriented local optimization mechanism is introduced, the global search capability and convergence stability of the PSO algorithm are improved, parameters of the fuzzy controller are dynamically adjusted by improving the PSO algorithm, and the control performance of the system is improved.

This invention relates to the field of oil and gas reservoir exploration technology, specifically a method, apparatus, electronic device, and storage medium for predicting the content of multiple minerals. The method includes acquiring input features of the well section to be predicted; inputting these features into a multi-component mineral content prediction model; and outputting the corresponding prediction results for the content of each mineral component. The multi-component mineral content prediction model is trained using multiple samples on a pre-set deep learning model, which employs a strategy of bidirectional multi-scale feature extraction and cross-scale attention fusion. The model constructed by this invention can not only efficiently capture the global long-term trend and local abrupt fluctuations of logging curves, but also automatically learn the mutual constraints between minerals, achieving accurate mapping between multi-scale logging features and specific mineral categories. Therefore, the multi-component mineral content prediction model enables efficient and collaborative prediction of multiple minerals, providing an effective data foundation for oil and gas exploration and development, reservoir evaluation, and production capacity prediction.

The invention discloses a multi-agent system workflow optimization method based on a heterogeneous graph, and the method comprises the steps: modeling a multi-agent system into the heterogeneous graph, converting a MAS workflow generation problem into a heterogeneous graph adjacency matrix optimization problem, taking an agent and a tool as nodes, taking an interaction mode as an edge, and carrying out the optimization of an adjacent matrix of the heterogeneous graph; comprising a communication link between the intelligent agents, calling of tools and self-reflection of the intelligent agents; designing a sub-graph sampling strategy for sub-graph sampling by referring to an upper confidence bound algorithm; a two-stage matrix training process is introduced: in the first stage, through subgraph sampling and execution task evaluation, efficient nodes are quickly screened out by combining node sparsity and low-rank sparsity punishment; and the second stage is fine-grained edge optimization, interaction edge weights are optimized for the nodes selected in the first stage, effective edge weights are improved, redundant edge weights are reduced, and an optimal graph matrix is obtained. The method is suitable for the fields of scientific calculation, software development and the like, the adaptability and performance of the MAS system are improved, and the method has wide application prospects and practical value.

The application discloses a commodity information automatic question answering method based on an e-commerce knowledge graph, which is used for a knowledge graph question answering task in the e-commerce field and comprises the following steps: firstly, the number of entities and relations in the e-commerce knowledge graph is counted, and initial scores are respectively given to the entities and the relations; secondly, a natural language question proposed by a user is divided into words, and a theme entity is extracted; starting from a corresponding entity node in the e-commerce knowledge graph, the score of a tail entity is calculated by using the score of a head entity and the score of a relation of a triple, and a set of entity scores is updated, so that an interpretable knowledge graph question answering core path reasoning model is obtained. The application calculates a reasoning path in the e-commerce knowledge graph by extracting a natural language question feature vector, so that the accuracy and the interpretability of the commodity information automatic question answering are improved.