249 results about "Graph match" patented technology

The invention relates to the technical field of Internet of Things management, in particular to a lightweight Internet of Things management system based on a star flash protocol stack, which comprises a plurality of modules such as a star flash protocol communication module, a dynamic topology management module and a resource virtualization module. The satellite flash protocol communication module realizes low-power-consumption efficient connection of equipment; the dynamic topology management module optimizes the network topology by using an improved multi-agent Q learning algorithm; the resource virtualization module realizes accurate resource allocation through a weighted bipartite graph matching model; the safety protection module adopts an attention mechanism to detect abnormities and guarantee data safety; the edge co-processing module realizes intelligent task unloading by means of deep reinforcement learning; all the modules cooperatively work under the overall planning of the cross-module coordination controller, and the information barrier is broken. According to the invention, the problems of high communication energy consumption, poor resource allocation, weak security protection, low task processing efficiency, insufficient module collaboration and the like of a traditional Internet of Things system are effectively solved, the system communication efficiency, the resource utilization rate and the security are remarkably improved, and the task processing capability is enhanced.

The invention relates to a self-localization and motion perception method based on deep learning, and the method comprises the following steps: multi-modal perception: collecting RGB frames and event streams through employing a DAVIS346 event camera, obtaining texture and depth information through employing an RGB-D camera, and supplementing 3D structure data through a laser radar; hybrid optical flow driven motion perception: adopting a double-branch architecture of an improved RAFT network and an event optical flow private network; multi-modal 3D detection and trajectory management: fusing multi-modal data based on VoxelNeXt-Lite to generate a 3D detection frame, filtering false detection by combining point cloud density clustering and an event density threshold, then constructing a space-time diagram associated trajectory through a Spatio-Temporal Graph Transformer, and optimizing pedestrian trajectory prediction continuity by using an LSTM (Long Short Term Memory) model perceived by a gait cycle; multi-view semantic graph matching: constructing a 3D semantic voxel map containing vertical features, calculating scene similarity in combination with top view NetVLAD features and deformable graph matching, and dynamically adjusting semantic weight by using a situation encoder and knowledge graph reasoning; and carrying out multi-sensor fusion and robust positioning.

The invention discloses a large model reasoning enhancement method based on knowledge graph sub-graph matching. The method comprises the following steps: firstly, establishing an index based on attribute information of nodes and edges in a multi-hop knowledge graph; and then, according to the established index, performing retrieval based on a nearest neighbor algorithm. And then, according to a retrieval result, constructing a minimum correlation subgraph, and performing reasoning enhancement. And finally, based on the minimum correlation subgraph, extracting and processing information, and generating a natural language answer understood by the user. According to the method, the multi-hop reasoning problem under a complex graph structure is effectively solved through the multi-stage optimization process, and the response quality of user query and the practicability of the system are improved.

The invention discloses a real-time dynamic trajectory tracking method and system for millimeter wave radar gesture recognition, and relates to the technical field of gesture recognition tracking, and the method comprises the steps: receiving an echo signal reflected by a gesture, extracting a potential target point cloud, and carrying out the clustering generation of a gesture point cloud sequence; establishing a multi-modal motion model library, dynamically selecting an optimal motion model by adopting graph matching, and generating a prediction state in combination with a gesture point cloud sequence; on the basis of a Poisson multi-Bernoulli hybrid filtering framework, according to the signal-to-noise ratio and spatial distribution of the current gesture point cloud sequence, dynamically adjusting the observation weight, optimizing the observation point cloud, and carrying out optimal association by combining Mahalanobis distance with dynamic time warping; a multi-hypothesis tracking strategy is adopted to maintain trajectory hypothesis, an optimal trajectory is selected through a trajectory scoring mechanism, and Kalman filtering smoothing processing is performed on the optimal trajectory. According to the method, high-precision and low-delay tracking of gesture motion is realized, gesture habits of different users and complex environment interference can be adapted, and meanwhile, relatively high track precision is kept.

The invention relates to the technical field of electric power topology identification, and discloses an electric power topology rapid identification method based on depth map matching, and the method comprises the following steps: S1, carrying out the multi-source heterogeneous data fusion and adaptive noise reduction, and employing an adaptive threshold noise reduction algorithm; s2, modeling a dynamic space-time diagram; s3, attention-driven graph matching is carried out, and a two-channel graph neural network is used; and S4, updating the incremental topology library. For the condition that the real-time data of the low-voltage power distribution network generates significant noise due to equipment aging and acquisition errors, a self-adaptive threshold noise reduction algorithm is adopted, high-frequency noise components are accurately filtered out according to a threshold value, missing data are intelligently repaired in combination with a space-time Kriging interpolation method, the purity and integrity of the data are ensured, and a data basis is provided for subsequent topology recognition.

The invention discloses a data governance method and system based on artificial intelligence, and relates to the technical field of data governance, and the method comprises the steps: effectively improving the recognition capability of structural abnormal behaviors in a device file through introducing an abnormal track extraction mechanism combining image enhancement and cuckoo search, and improving the recognition efficiency. The method combines an isolated forest model optimized by a genetic algorithm, achieves the high-confidence screening of candidate abnormal data, remarkably improves the accuracy and robustness of anomaly detection, and achieves the recognition of a redundant entity through the construction of a semantic conflict graph and the introduction of a graph matching network. The problem that in a traditional method, the field similarity is insufficient to support complex conflict judgment is effectively solved, finally, a standard equipment file is generated through a field-level fusion strategy, a governance evolution graph is synchronously updated, a complete closed loop from discovery, evaluation and fusion to optimization is formed, automation, intelligentization and self-adaptive evolution of equipment master data governance are achieved, and the efficiency is improved. And the consistency, availability and governance efficiency of the equipment data are remarkably improved.

The invention discloses a task dynamic decomposition management system and method based on an AI framework, and relates to the technical field of task management, and the method comprises the steps: dividing task parameters into a plurality of dimension features, carrying out the fusion of the multi-dimension features through a space-time attention network, and constructing a three-dimensional feature tensor; recursively decomposing the three-dimensional feature tensor through an adaptive secondary screening algorithm, and outputting a decoupled modal set; predicting a resource load and quantifying a modal demand level based on a long short-term memory network, constructing a bipartite graph matching model, and adaptively allocating resource instances through dynamic matching; time performance and resource consumption data of task execution are collected in real time, a modal deviation value is calculated, if the modal deviation value exceeds a deviation threshold value, dynamic re-decomposition is triggered, resources are reallocated based on an updated modal set, and dynamic capture, efficient decomposition and accurate resource allocation of task features are achieved; therefore, the task execution efficiency and the resource utilization rate are improved.

The invention provides a multi-target tracking method and device based on secondary graph matching, and belongs to the technical field of computer vision, and the method comprises the steps: employing an improved YOLOX target detection network, combining with a multi-task learning framework, and carrying out the target detection and feature extraction of an input video frame image; carrying out target classification, target association and target tracking processing by adopting a secondary graph matching strategy of target machine matching and region-level matching; after the second-level graph matching processing is completed, track prediction is carried out by adopting Kalman filtering, a multi-target tracking result is output after track updating and fusion processing are carried out in combination with background modeling based on a Gaussian mixture model, matching is carried out at a target level by adopting a GNN graph neural network, and the target identity retention capability is optimized; and then, in each target area, fine-grained matching is carried out by utilizing foreground point matching and adopting GCN graph convolution, so that the target association stability under the shielding condition is improved, and the stability of multi-target tracking is improved by adopting secondary graph matching and combining Kalman filtering and background modeling.

The invention discloses a single tree trunk structure extraction method and system based on deep learning, and the method comprises the steps: obtaining two-dimensional image data of a single tree, marking the two-dimensional image data, and constructing a single tree trunk segmentation data set; constructing a spatial domain and frequency domain double-branch network based on the segmented data set, respectively extracting spatial domain features and frequency domain features, and fusing the double-branch features to generate a coding feature map; the coding feature map is decoded, in the decoding process, coordinate convolution CoordConv is adopted to enhance position perception, and mask segmentation and semantic label extraction are respectively carried out through a dynamic mask reconstruction branch DRMask Branch and an instance branch Inst Branch; based on a bipartite graph matching strategy, associating results of the mask segmentation and instance branches, and realizing segmentation of a single tree trunk and matching of instance-level labels; according to the method, the boundary precision and the detail reconstruction capability of trunk segmentation are remarkably improved, and the problem of feature loss of a traditional method in a complex under-forest environment is solved.

The invention discloses a cross-language code semantic alignment method, which constructs a shared semantic space through a unified abstract syntax tree (AST) and a graph matching network (GMN) so as to reduce the difference of different programming languages in syntax structure and node representation. The method comprises the following steps: (1) mapping a multi-language AST node to a unified general label set and performing structure enhancement; (2) performing node feature coding on the unified AST, and realizing cross-language interaction in combination with a cross-graph attention mechanism; (3) node representation is generated through intra-graph loop updating, and an overall semantic vector is obtained through global attention pooling; and (4) through comparative learning training in the shared space, the distance between semantically equivalent positive sample vectors is shortened, and the distance between non-equivalent negative sample vectors is shortened, so that the discrimination capability of cross-language semantic representation is enhanced. According to the method, the semantic consistency of the functional level can be effectively captured, and the accuracy and efficiency of cross-language code understanding, multiplexing and retrieval are remarkably improved.

The invention discloses an APT attack detection system for graph matching threat intelligence. The APT attack detection system comprises a cloud data layer, a management decision layer and a terminal equipment layer, the detection method comprises the following steps: step 1, static graph matching similarity calculation based on a graph neural network; step 2, dynamic abnormal flow detection; step 3, trust evaluation; the method has the characteristics that the APT attack is detected by fusing graph matching and threat intelligence, and the cross-domain abnormal traffic detection efficiency of the industrial internet is improved.

The invention provides a query graph sub-graph matching method based on a graph neural network, relates to the technical field of data processing, and aims to avoid direct global aggregation of a whole graph by adopting a query graph division strategy based on betweenness centrality so as to reduce the calculation complexity. By designing a multi-round bidirectional cross attention mechanism, the matching results of different sub-graphs are effectively fused, and the problems of node representation smoothing and global structure information loss in the existing method are solved, so that the consistency of the overall matching results is improved. By introducing the reinforcement learning pruning strategy, the decision in the matching process is adaptively optimized, so that the efficiency and precision of the algorithm in processing complex graph data are improved.

The invention discloses a new energy power station inspection method based on a multi-modal large model small sample open set, and the method constructs a multivariate text learnable prompt, and guides a detection model to form a finer-grained category decision boundary through the text information related to an aggregation task. Due to the lack of real unknown class samples in the training process, the mining of unknown class pseudo samples is regarded as a bipartite graph matching task for the first time, and the model is optimized to form a compact unknown class decision boundary by adding unknown class virtual nodes, constructing a cost matrix and mining the unknown class pseudo samples. In order to solve the problem that known classes and unknown classes of small samples are prone to confusion, the method proposes unknown class optimization loss based on cost perception, considers the classification and positioning quality of the unknown classes, and improves the open set detection performance of the model. According to the method, transformation from a single-mode vision small model to a multi-mode vision large model is realized, and good generalized small sample open set target detection performance can be obtained only by a small amount of training data.

The invention discloses an end-to-end category level object pose estimation method based on space-time implicit anchor point query, which comprises the following steps of: inputting an RGB (Red, Green and Blue) image sequence into a visual encoder, extracting multi-scale semantic features and fusing depth and normal vector geometric features to generate 2.5-dimensional multi-scale features; a 3D implicit anchor queue is established based on the features, and queries in the queue are processed through confidence screening and camera pose offset transformation. Performing modeling on space-time correlation by using multi-head attention, outputting time sequence enhanced query features, and performing geometric perception feature sampling; and finally, mapping the implicit 3D query into 9-degree-of-freedom attitude parameters through bipartite graph matching to realize end-to-end training.

The invention discloses a unified representation and dynamic knowledge reasoning method and device for an intelligent cluster system, and the method comprises the steps: determining a decyclization task sub-graph based on a composite task state space model of the intelligent cluster system through the priori information; constructing a real-time data sub-graph by using a real-time data feedback mechanism, and constructing a traceability mode sub-graph based on the decyclization task sub-graph; candidate traceability sub-graphs with the highest matching similarity with the traceability mode sub-graphs are determined from the real-time data sub-graphs to serve as traceability sub-graphs; constructing a reasoning path set based on the decyclization task subgraph, and calculating a probability value of each reasoning path in the reasoning path set by using a path probability model constructed based on a Bayesian conditional probability chain rule; and screening out the reasoning paths of which the probability values are smaller than a preset probability value in the reasoning path set by utilizing a dual path pruning strategy, and adding the reasoning paths containing abnormal nodes in the reasoning path set into an alarm path set so as to provide support for state prediction and risk management of the intelligent cluster system.

The invention relates to the field of relation triple extraction, in particular to a relation triple joint extraction method based on information enhancement and bidirectional modeling. According to the method, through entity-to-relation and relation-to-entity double-branch collaborative modeling, semantic information is enhanced through double-branch potential information complementation: a bipartite graph matching method is combined with subjects and objects extracted by an entity extraction module, and relation classification is optimized through an entity boundary mask attention enhancement method; a potential relation extraction module is used for guiding subject and object entity extraction, and entity boundary information is utilized to crosswise mask attention of a large related area; the two branches filter redundant triads through a relation triad cutting module; combining bidirectional results and outputting a complete triple set; according to the method, a bidirectional interaction method is introduced to realize mutual enhancement of entities and relationships, and attention distribution is optimized in combination with entity boundary masks, so that the problem of high dependence on an initial extraction result caused by unidirectional modeling in traditional joint extraction is effectively solved.

A graph representing the source code is computed, the graph comprising a plurality of nodes connected by edges, each node representing a software component of the source code and each edge representing a relationship between software components. A domain model of a domain related to the legacy code base is computed, the domain model comprising a plurality of concepts and relationships between the concepts. A node of the graph matching one of the concepts is identified. A functional role of model is assigned to the identified node. The knowledge of the assigned functional role is used to trigger any of: migration, deployment as a microservice, deployment as a containerised service, deployment as a cloud-native application, maintenance, refactoring.

The invention relates to the technical field of data analysis, in particular to a power grid project cost rationality analysis system, which comprises a data fusion module for collecting heterogeneous data sources of a power grid project, fusing multi-source data in real time, and converting the data into standard data through a unified data cleaning engine; the dynamic analysis module is used for setting a dynamic threshold value according to the standard data, tracking a single index of a project progress and the dynamic threshold value based on a machine learning model, and positioning abnormal data; according to entities and relations in the standard data, when the system is used, synonyms or different-name entities in multiple sources are efficiently recognized and unified through the graph matching and embedding technology, manual account checking is reduced, a dynamic threshold value is combined with a machine learning model, the execution condition of each task in a project is continuously tracked, and the construction efficiency is improved. Different from a traditional fixed threshold value, the system supports personalized monitoring according to historical behaviors, target curves and time dynamic changes, and is beneficial to being closer to actual construction fluctuation.

The invention relates to the technical field of computer vision and pattern recognition, and discloses a multi-component object matching method and system based on graph matching. The method comprises the following steps: identifying each discrete component and component type label in a to-be-detected object image; forming a hierarchical set structure corresponding to the hierarchical relationship of the preset template; establishing undirected full-connection edges between adjacent levels, and filtering invalid edges to construct a dynamic graph; searching candidate sub-graphs matched with the topological structure of the preset template in the dynamic graph; calculating an affine transformation matrix between each candidate subgraph and a preset template; calculating a matching degree between the remaining candidate sub-images and a preset template, taking a matching result of which the matching degree exceeds a preset threshold value as effective matching, and outputting an object position coordinate and a geometric transformation parameter corresponding to each effective matching; according to the method, deep learning detection and graph matching technologies are fused, and hierarchical dynamic graph construction and backtracking affine two-stage verification are combined, so that the detection rate and the processing efficiency are greatly improved, and the mismatching rate is effectively reduced.

The invention provides a character interaction detection method based on open vocabularies in an unmanned aerial vehicle scene, and the method comprises the steps: extracting global features of an image through employing a pre-trained CLIP visual encoder, segmenting the image into a plurality of image blocks, and carrying out the coding of the image blocks; an MLR module is introduced to extract global context information, character interaction is decoded from a multi-level feature map, and when bipartite graph matching is carried out between a prediction result and a real result, a loss function is designed to guide a low-level feature map to correspond to a character pair with a small distance and guide a high-level feature map to correspond to a character pair with a large distance; a large language model is used for generating human body part state description related to character interaction in an image, character interaction category names and the state description of the related human body parts are coded into text embedding, and the embedding is combined with the output of a character interaction decoder and global context information extracted by an MLR module to obtain an instance-level detection score. According to the invention, the model can better adapt to interaction detection requirements under different distances.

The invention discloses an online vectorization high-precision map generation method based on point features, which is characterized in that real-time high-quality generation of an online vectorization high-precision map is realized by utilizing a Point MapNet network, and the Point MapNet network is obtained by improving a BEV feature coding module and a map element decoding module of an original MapTR network; the BEV feature coding module is improved in the following steps: replacing BEV features sampled at fixed positions with position learnable point features, naming the improved BEV coding module as a point feature coding module, enabling the point features to learn a spatial position containing a detection object in a sensing area through bipartite graph matching and distance loss calculation, and updating the position of the point features; the map element decoding module is improved as follows: attention masks based on distance are used, so that map element features are only interacted with point features of a near space. Through the improvement, the reasoning speed is increased and the training cost is reduced while higher map generation quality is achieved.

The invention provides a deep learning-based dynamic early warning analysis method for disseminated intravascular coagulation, which comprises the following steps of: acquiring and structuring a core physiological mechanism knowledge graph of the disseminated intravascular coagulation, preprocessing standardized and clean multi-modal clinical time sequence data, generating a physiological consistency constraint vector by utilizing a dynamic graph matching and graph embedding technology, and performing dynamic early warning analysis on the disseminated intravascular coagulation core physiological mechanism knowledge graph. Inputting the medical logic correlation variable into a time sequence encoder fused with a knowledge graph attention mechanism to realize efficient modeling of the medical logic correlation variable; anti-factual reasoning is carried out based on model output, the risk of dispersive intravascular coagulation of a patient is dynamically scored, a grading early warning signal is triggered, and clinical decision making is supported; the model can update the knowledge graph and optimize parameters according to new cases and clinical feedback increments, the adaptability and generalization are enhanced, and the accuracy, early warning ability and medical interpretation of risk assessment of disseminated intravascular coagulation are effectively improved.

The invention discloses an intelligent AI interactive question-answering method and system applied to science popularization and a storage medium, and the method comprises the steps: constructing a science popularization knowledge spectrogram based on multi-source science popularization knowledge; extracting key feature data based on a voice or text question input by a user, and generating a question feature vector; similarity matching is carried out on the basis of the feature vectors and a node embedding space of the science popularization knowledge spectrogram, and accurate knowledge sub-graphs related to questions are extracted; after completing the response of the current question, recording a storage positioning index of the knowledge sub-graph corresponding to the question in the popular science knowledge spectrogram, and generating knowledge sub-graph index sequence data; and based on the next question of the user, carrying out correlation analysis, including questions of the same and different knowledge types, and carrying out sub-graph matching or full-graph matching, and if a plurality of knowledge fields are involved, carrying out sub-graph matching and full-graph matching. The method has the advantages that efficient knowledge matching is realized by fusing multi-source popular science knowledge and feature extraction, user questions are quickly responded, and the matching precision is continuously optimized.

The invention discloses a timing sequence statement positioning model training method and device based on proposal selection and anchor point distribution, and relates to the technical field of timing sequence statement positioning. The method comprises the following steps: initializing a fixed number of learnable queries according to a first static anchor point set; on the basis of the first learnable query, according to the unpruned long video and the natural language description text, performing proposal generation through a time sequence statement positioning model; based on the proposal selection module, redundant proposal filtering is carried out by using a non-maximum suppression algorithm; based on an anchor point distribution module, according to the untrimmed long video, the first static anchor point set and the candidate proposal set, performing bipartite graph matching by using a Hungary algorithm; and carrying out loss function weighting calculation according to the unpruned long video and the optimal proposal set, and carrying out parameter optimization on the time sequence statement positioning model to obtain an optimized time sequence statement positioning model. The method is an efficient and accurate time sequence statement positioning model training method combining proposal selection and anchor point distribution.

The embodiment of the invention provides a network surveying and mapping behavior anomaly detection method and system based on machine learning, and belongs to the technical field of network surveying and mapping behavior anomaly detection. The method comprises the steps of collecting double-source flow data, and generating a structured log data set through a double-source log fusion engine; performing sub-graph matching calculation to obtain a surveying and mapping behavior deviation degree; generating communication data containing the watermark identifier in a communication path corresponding to the session; and verifying whether an attack event carries the watermark identifier, and generating a network surveying and mapping behavior anomaly detection report. According to the method, the self-adaptive attack behavior model is constructed through the multi-modal feature vector based on the structured log and the graph protocol mapping rule base, the cognitive robustness of protocol camouflage and path drifting is fundamentally enhanced, a detection result real-time verification chain is established, and through cross verification of a watermark carrying state and a behavior track, the detection accuracy is improved. And traditional passive detection is converted into self-proof active defense.

Systems and methods for entity linking using a graph neural network are disclosed. In one aspect, a method for entity linking can include extracting a first attribute set of an unknown entity from an information source and retrieving second attribute sets of known entities from a database, wherein each of the second attribute sets corresponds to one of the known entities. The method can further include generating an unknown entity graph based on the first attribute set, generating known entity graphs based on the second attribute sets, generating an unknown entity graph embedding by applying the unknown entity graph to a graph neural network, and generating known entity graph embeddings by applying the known entity graphs to the graph neural network. The method can further include assigning the information source to one of the known entities based on the unknown entity graph embedding and the known entity graph embeddings.

The invention discloses a sheet metal part manufacturability reasoning method based on space-semantic map alignment, and relates to the field of manufacturing-oriented design evaluation and industrial knowledge reasoning, and the method comprises the following steps: carrying out geometric analysis on a CAD geometric model of a sheet metal part to be evaluated; abstracting the geometric features and the topological / metric spatial relationship thereof into a computable spatial semantic graph; performing semantic analysis on the process specification described by a natural language, converting the process rule into formalized logic check expression by using a large language model through context learning, and generating an executable domain-specific language check script; executing the script on a spatial semantic graph, realizing deterministic reasoning through graph matching and attribute verification, and completing accurate mapping and violation detection of text rules and geometric features; and outputting an interpretable diagnosis result containing violation feature positioning, triggering rules and numerical evidence. In order to solve the problems that a process rule'natural language-geometric model 'has a semantic gap, a traditional rule system is poor in adaptability, and an end-to-end learning method is high in data dependence and cannot be explained, a new rule can be quickly adapted under the condition that a large amount of data does not need to be labeled and a model does not need to be retrained; the method can accurately identify the violation of the micro-size and spatial relationship, and has reasoning preciseness, interpretability and engineering availability.

The invention discloses a traditional Chinese medicinal material defect detection method based on machine vision, relates to the technical field of intelligent detection, and solves the problems that in the prior art, a threshold cannot be migrated, so that an inter-batch judgment standard is unstable, and high-resolution block reasoning causes boundary pseudo defects and instance wrong affiliation. According to the method, a reference area is constructed in a batch domain, a statistical descriptor is extracted, and monotone mapping calibration is performed on defect scores so as to realize cross-batch unified judgment; meanwhile, a seam continuity field is introduced in the image block splicing process, defect connected domains are fused through a cross-window adjacency relation, and accurate attribution of defects and medicinal material pieces is achieved based on bipartite graph matching; according to the method, the judgment stability and the statistical reliability in the traditional Chinese medicinal material defect detection process are remarkably improved.

The invention provides a graph matching medical text scoring method and device based on a fact structure, and the method comprises the steps: obtaining a standard medical text and a test medical text; inputting the standard medical text and the test medical text into a pre-trained language model to obtain a plurality of standard keywords of the standard medical text and a plurality of test keywords of the test medical text; respectively constructing a standard triple of the standard medical text and a test triple of the test medical text based on the plurality of standard keywords and the plurality of test keywords; respectively constructing a standard knowledge graph of the standard medical text and a test knowledge graph of the test medical text based on the standard triple and the test triple; based on a preset graph editing distance algorithm, determining a graph editing distance between the standard knowledge graph and the test knowledge graph; performing normalization based on the graph editing distance to obtain an evaluation score of the test medical text; according to the invention, more accurate and comprehensive quality evaluation can be provided for the medical text.

The invention discloses a vehicle path problem solving method, system and equipment and a medium, and particularly relates to a vehicle path problem solving method based on clustering decomposition and graph matching, which comprises the following steps: S1, receiving a large-scale vehicle path problem instance to be solved, modeling into a weighted undirected graph, and combining a series of constraint conditions and optimization targets; s2, constructing an offline knowledge base with a diversified structure, inputting a group of preset parameters in the step, and outputting a knowledge base stored in the memory; s3, aiming at the large-scale vehicle path problem instance to be solved, executing initialization operation to generate a global initial solution; and S4, by taking the global initial solution as a starting point, executing an iterative local search framework to carry out deep optimization on the solution until a preset termination bar is met. The technical problems that in the prior art, when a large-scale complex logistics network is processed, the calculation time consumption is long, the planning cost is high, and the result is unstable are solved.