1486results about How to "Easy to optimize" patented technology

The invention provides a unit test case generation system based on a large language model, and the system comprises a function dependency analysis engine module which is used for preprocessing a target project, extracting a function dependency relationship, and generating an abstract syntax tree and a source code file; the large model module is used for automatically generating a test case through the code context information and the optimized cue word; and the verification and optimization module is used for verifying whether the test cases can be successfully executed or not by compiling and running the test cases, marking part of the test cases needing to be further processed, automatically analyzing error reasons, optimizing cue words and realizing automatic repair of the part of the test cases. According to the unit test case generation system based on the large language model, the function dependency analysis and the large language model are combined, the high-quality test case can be automatically generated, dependence on manpower or existing test cases is not needed, manual intervention is reduced, and the test efficiency and comprehensiveness are improved.

The invention discloses a robot agent reinforcement learning training method in a complex scene, and the method comprises the steps: constructing a reinforcement learning algorithm environment and a state transition model which comprise a barrier, a dynamic target, environment constraints and other complex scenes and are suitable for a multi-agent system; an independent neural network is constructed for each robot agent by adopting a reinforcement learning algorithm; taking vectors of a given state, time sequence information and historical state information of the intelligent agent as input of the state transition model, and outputting a state transition probability and reward information of a prediction state; the action strategy and the action value output by the neural network are executed in the state transition model, and a new prediction state and reward information are generated; through adjusting hyper-parameters of the neural network, a large-scale distributed training architecture is used to carry out cooperative training on the decision network and the evaluation network, parallel learning and cooperative decision of multiple agents are realized, and performance and generalization ability of the agents in a complex scene are improved.

The invention discloses a lightweight industrial product surface defect detection method, and belongs to the field of computer vision and artificial intelligence application, and the method comprises the steps: obtaining an industrial product surface defect image data set, and carrying out the preprocessing of the data; the method comprises the following steps: constructing a YOLO-IDL industrial product surface defect detection model, designing a lightweight hierarchical multi-scale feature extraction network as a backbone network, designing a scale rhinoceros neck as a neck network, and adopting a YOLOv8 head network as a head network; and training a YOLO-IDL industrial product surface defect detection model by using the preprocessed industrial product surface defect image data set to obtain a well-trained YOLO-IDL industrial product surface defect detection model, and deploying the well-trained YOLO-IDL industrial product surface defect detection model so as to detect industrial product surface defects and output defect types and positions.

The invention provides an RFID-based power supply management method for a tilt angle sensor of a power tower, and the method comprises the steps: employing a support vector machine and a decision tree, analyzing the historical operation data of the tower, building a tower fault prediction model, predicting the state of the tower through the tower fault prediction model, and obtaining the fault probability and time of the tower; according to a tower fault prediction result, a preventive maintenance plan is made, priority maintenance is arranged for the tower with the fault probability larger than a preset threshold value, and a power supply strategy is adjusted in advance for the tower predicted to have a fault, and load pressure is reduced; a tower group intelligent power supply management system is adopted, state information and a power supply strategy of a single tower are integrated into a whole power grid, the power supply load of each tower is dynamically adjusted through a load balancing algorithm, and for a tower in a dangerous state, the power supply load of the tower is reduced, and the load is transferred to a tower in a good state.

The invention relates to the technical field of communication, and discloses a communication data transmission and temporary storage method and device and a storage medium, and the method comprises the following steps: S1, receiving and partitioning initial data, and storing the initial data in a cache region; s2, calculating weighting coefficients of the data blocks and performing priority ranking; s3, encrypting the data blocks according to the sequence; s4, transmitting the encrypted data to a target end through a security protocol; s5, the target end decrypts the data and verifies the integrity; s6, temporarily storing the verified data blocks in a temporary storage area until all the data blocks are transmitted; and S7, after the transmission is completed, recombining and storing the data block in the target equipment. According to the method, the data is subjected to blocking, priority ranking, encryption transmission and decryption recombination, so that the security, integrity and transmission efficiency of the data in the network transmission process are ensured, and the method is suitable for a communication system needing high security and high data integrity.

The invention discloses an energy-saving illumination control system and method for an urban illumination LED lamp group, and relates to the technical field of illumination control, and the system comprises an ambient light detection module, a traffic flow monitoring module, a time period division module, a weather information acquisition module, a weather information control module, an LED lamp group driving module, a wireless communication module, a high-precision photosensitive sensor for the ambient light detection module, a filtering and temperature compensation module, and an intensity measurement accurate data transmission module. The traffic flow monitoring module depends on geomagnetism or camera shooting, is high in precision and strong in recognition, and facilitates regulation and control through data; the time period division module accurately times and corrects time according to a clock chip, and sets illumination measures according to the time; the weather information acquisition module is networked to acquire meteorological data, and lamps are adjusted according to day conditions; and the control module multi-core microprocessor calculates illumination levels and brightness according to multi-source data to control lamp groups. According to multi-factor intelligent lamp control, energy saving is remarkable, lighting quality is good, comfort and safety are improved, remote monitoring and management can be achieved, troubleshooting is accurate and efficient, maintenance is convenient and fast, intelligent and green development of urban lighting is promoted, and the intelligent city construction requirement is met.

The invention discloses a power generation equipment fault diagnosis method based on a large language model. According to the method, firstly, an expert knowledge base is converted into a fault knowledge base composed of criterion nodes and edges, and fault retrieval is completed through the knowledge base; synthesis of natural language texts is completed by using input based on text function segmentation, cue word templates are constructed based on calculation types of different criteria, and a large model is promoted to complete calculation reasoning tasks of the criteria; multi-dimensional equipment state monitoring time sequence data are input, a cue word set at the current time is automatically generated in combination with the cue word template, the cue word set is input into a large language model for computational reasoning, and a current state set of equipment is generated; using a knowledge base to retrieve the state set to obtain a fault diagnosis result; and then a visualization system is designed to collect and visualize a large model output result and a fault diagnosis result, and input and interaction interfaces are provided for cue word template construction. The fault diagnosis completed by the method has good practicability and expansibility.

The invention provides a method for intelligently predicting the strength of a filling body by combining color and pore structure characteristics, and relates to the technical field of mine filling, the method comprises the following steps: firstly, selecting a filling material for mine production, making an experimental scheme on the basis of the actual proportion of mine production, and preparing and maintaining the filling body; then obtaining a color characteristic value, a pore structure characteristic value and a uniaxial compressive strength value of the filling body; and processing the color characteristic value, the pore structure characteristic value and the uniaxial compressive strength value, establishing a data set, constructing an initial strength prediction model, and performing evaluation and optimization to obtain an optimal strength prediction model.

The embodiment of the invention discloses a method for evaluating the ablation cracking degree of an environment-friendly gas high-voltage circuit breaker contact material. The method comprises the steps that thermal parameters and mechanical parameters of the high-voltage circuit breaker contact material are collected; establishing a three-dimensional geometric model of the high-voltage circuit breaker contact according to the thermal parameters and the mechanical parameters; analyzing the high-voltage circuit breaker contact according to an extended finite element method, and obtaining simulation results of temperature and stress distribution of the high-voltage circuit breaker contact in the working process; according to the simulation result, the ablation degree and the cracking degree of the high-voltage circuit breaker contact material are determined, and a corresponding evaluation result is obtained. Through accurate three-dimensional modeling, data processing and evaluation result obtaining, evaluation and optimization of the high-voltage circuit breaker contact material are facilitated, and a reference is provided for design and application of the high-voltage circuit breaker contact material under the large-current open-end condition.

The invention relates to the technical field of constructional engineering, and discloses a construction engineering crack detection device and method.The construction engineering crack detection device comprises a detection device body, the detection device body comprises an ultrasonic instrument, an ultrasonic connector is installed on one side of the ultrasonic instrument, and the ultrasonic connector is connected with a detection probe through a cable; the detection probe is installed on the wall surface through the installation device, the installation device comprises a positioning support and an installation support, the positioning support is installed on the wall surface through an adsorption assembly, the installation support is used for installing the detection probe, and the detection probe is fixed through insertion connection of the positioning support and the installation support. According to the scheme, the support is installed on the wall surface, after the stability of the probe is guaranteed, the instrument body is operated, one worker is liberated, and the problem that the efficiency is low due to the fact that two workers need to be consumed in an existing detection mode is solved.

The invention relates to the technical field of power grid investment management and decision support systems, in particular to a dynamic power grid investment benefit evaluation and consultation platform which comprises a cross-domain data fusion module, a real-time intelligent analysis module, a comprehensive evaluation index generation module, an investment decision support module, a strategy dynamic updating module and a user interaction module. Wherein the cross-domain data fusion module is used for collecting heterogeneous data in multiple aspects; the real-time intelligent analysis module is used for carrying out dynamic analysis and generating intermediate data containing power grid investment benefit evaluation; the comprehensive evaluation index generation module is used for generating a comprehensive power grid investment benefit index; the investment decision support module is used for generating optimized investment strategy suggestions; and the strategy dynamic updating module dynamically adjusts the power grid investment strategy. According to the invention, by integrating multi-source data in real time and providing an interaction function, the adaptability, the accuracy and the participation degree of power grid investment decision making are remarkably improved, so that the return on investment is optimized and the risk is reduced.

The invention provides an image processing method, device and equipment for scene reconstruction and rendering. The method comprises the following steps: initializing sparse point clouds corresponding to each scene image to be reconstructed to obtain a plurality of initial Gaussian surface elements; optimizing each initial Gaussian surface element based on an inverse depth map, a rendering scene image and a normal vector map obtained by rendering the initial Gaussian surface elements to obtain a plurality of first target Gaussian surface elements; based on the importance of the first target Gaussian surface elements, determining a second target Gaussian surface element, the importance being determined based on a plurality of sub-models and a target scene image to be reconstructed matched with each sub-model, the plurality of sub-models being obtained by performing mesh generation on the plurality of first target Gaussian surface elements, and the plurality of sub-models being obtained by performing mesh generation on the plurality of first target Gaussian surface elements; the target scene image to be reconstructed matched with each sub-model is determined based on the pose of the scene image to be reconstructed and the rendering result of the sub-model; and determining a third target Gaussian surface element for rendering the reconstructed scene image based on the spatial range of the grid.

The invention relates to the field of medical model training, and particularly discloses a medical treatment guide model training method and system, a terminal and a medium, and the method comprises the steps: carrying out the incremental pre-training of a selected large language model through priori knowledge, and obtaining a basic large language model; acquiring a public real doctor-patient dialogue data set; constructing a doctor-patient dialogue template, and generating and constructing a doctor-patient dialogue data set by using a third-party large language model based on the doctor-patient dialogue template; combining the real doctor-patient dialogue data set and the constructed doctor-patient dialogue data set into a data set, and recording the data set as an initial doctor-patient dialogue data set; performing duplicate removal preprocessing on the initial doctor-patient dialogue data set to obtain a doctor-patient dialogue training data set; and using the doctor-patient dialogue training data set to carry out supervision fine tuning training on the basic large language model to obtain a doctor-seeing and hospital-guiding model. According to the invention, the joint training data set is formed based on the public data set and the automatic construction data set, so that the accuracy of the medical treatment guide model is improved.

The invention belongs to the technical field of sound equipment control, and particularly relates to an intelligent sound equipment control method and system and a storage medium, and the method enables the interaction control of the intelligent sound equipment on home equipment to be based on the control effect, and can screen out the interaction control record with the abnormal control effect based on the historical interaction control record. According to the method and the device, the voice control instruction is subjected to abnormal traceability and voice control instruction complexity correlation analysis, so that the targeted adjustment direction of the sound interaction control is obtained, feedback closed-loop adjustment of the sound interaction control is realized, the blank of the existing sound interaction control on feedback control is effectively filled, and the user experience is improved. Continuous optimization and improvement of sound interaction control are accelerated and promoted to a certain extent, so that the user experience is improved continuously, and sustainable development of the intelligent sound is facilitated.

The invention discloses a hotspot mining system and method based on Internet finance and economics information and a medium, and relates to the field of Internet finance and economics hotspot mining systems, and the system comprises an information collection module which collects finance and economics information in a multi-source manner, determines a collection period according to source characteristics, and carries out the screening and storage; the text preprocessing module is used for cleaning and analyzing texts, standardizing terms and carrying out classification labeling; the hotspot evaluation index calculation module is used for calculating indexes such as classification weighting, TF-IDF and the like; the hotspot mining and analyzing module is used for setting threshold mining hotspots, analyzing emotion and the like; and the visual display module is used for visually displaying the hot spots and supporting interaction and pushing. According to the invention, financial information is integrated, and hotspots are mined accurately. The information acquisition efficiency is improved, and decision-making investment and research are assisted. Information spreading is promoted, and financial field development is promoted.

The invention discloses an intelligent self-adaptive real-time task scheduling method and system based on multi-dimensional optimization. The method comprises the following steps: S1, inputting a task; s2, a DAG task is decomposed into independent sequential sub-tasks; s3, dynamically calculating the priorities of the tasks by adopting a multi-factor dynamic priority scheduling module, and updating the priorities of the tasks; s4, dynamically adjusting an acceleration factor by adopting a self-adaptive acceleration factor management module; s5, vertex correlation analysis is carried out, and a variance of processor load imbalance is calculated by adopting a load balanced distribution algorithm and is minimized; s6, in combination with static analysis and dynamic inspection, a hybrid schedulability judgment module is adopted to judge the tasks, and judgment results are fused; and S7, fusing decisions by adopting a decision fusion algorithm, obtaining a scheduling strategy through a scheduling strategy coordinator, executing the task, and monitoring the task in real time. Real-time monitoring and real-time optimization scheduling are realized, and the utilization rate of resources is improved.

The invention discloses a GNN and GRU-based bridge structure time-varying fire response prediction method and system, and the method comprises the steps: building a to-be-detected bridge structure FEM model, randomly generating a fire curve, and generating a data set through the combination of the FEM model and the fire curve; establishing a graph structure according to the FEM model, and embedding static and dynamic characteristics for temperature, stress and displacement prediction in nodes of the graph structure; a message transmission layer is constructed, the GNN is responsible for space information transmission, and the GRU is responsible for time information transmission; simulating a heat conduction formula by using a message passing formula combining the two; receiving node input vectors including temperature, stress and displacement prediction characteristics from the graph structure; determining a loss function and separately training GNN + GRU networks for temperature, stress and displacement prediction; and when the to-be-detected bridge structure encounters a fire, acquiring a fire response condition of the to-be-detected bridge structure through the trained model. The problem that a traditional GNN model can only predict non-time-varying boundary fire response is solved.

The invention provides a PWI modulation method for an inverter. Belongs to the technical field of power electronics, and comprises the following steps: setting basic working parameters of an inverter, initializing an initial value and a change range of a phase weighting factor, and presetting a PWF regulation strategy according to load characteristics and waveform quality requirements; loading and initializing a control algorithm of the inverter, and generating a standard sine wave as a reference wave according to a preset output voltage and frequency; calculating a PWF value of each phase point according to the current load state and a preset PWF adjustment strategy, and multiplying the PWF value by a standard sine wave to generate a target modulation waveform; and selecting a carrier signal, comparing the target modulation waveform with the carrier signal, and generating a PWM signal according to a comparison result. By presetting the PWF adjusting strategy and dynamically adjusting the PWF value, the waveform quality of the output voltage of the inverter can be effectively improved, and different load characteristics and waveform quality requirements are met.

The invention discloses a hospital internal network security isolation method based on an all-optical network architecture. The method comprises the following steps: S1, dividing a hospital internal network into a plurality of virtual optical slices by using an all-optical network slicing technology; s2, performing time domain isolation by using an optical time domain multiplexing technology, and transmitting data on different optical wavelengths by using a wavelength selective switching technology; s3, monitoring and managing bandwidth allocation through a software defined network control platform; s4, traffic prediction and resource optimization are carried out based on a decision tree integration algorithm; s5, detecting potential anomalies and security threats by using the graph neural network; s6, abnormal behaviors are detected through an isolation forest algorithm, and abnormal traffic is marked and isolated; s7, recording a network traffic log through a block chain technology; and S8, when a network fault occurs, using an intelligent self-healing mechanism to automatically switch redundant optical path resources and recover the optical path. According to the invention, all-optical network slicing and artificial intelligence technologies are fully utilized, and security isolation and resource optimization of a hospital internal network are realized.

The invention provides a distributed fault detection and dynamic task scheduling system for an unmanned ship cluster, and the system achieves the efficient cooperation of the unmanned ship cluster in a complex marine environment through the cooperation of a sensor data collection module, a data processing and feature extraction module, a fault state classification module, an information sharing module, a weighted voting decision module, a task scheduling module and the like. The system firstly collects state information of each unmanned ship and adjacent ships, performs feature extraction and classification on data, and judges whether a fault exists or not. Fault information is shared to each boat in a cluster through a multi-path redundancy communication mechanism, and weighted decision is carried out by a voting mechanism, so that the accuracy of fault detection is improved. After a fault is confirmed, the system dynamically adjusts task allocation according to task priorities, key tasks are transferred to the unmanned surface vehicle in a good state, and continuity and stability of cluster tasks are guaranteed. Different from traditional centralized control, the distributed control method disclosed by the invention enhances the damage resistance and the self-adaptability of the system.

The invention discloses an image forgery detecting and positioning method based on noise auxiliary prompt learning, which introduces a contrast language image pre-training model (CLIP) and prompt learning into the field of image forgery detecting and positioning, and comprises the following steps of: 1, respectively processing an input image and noise thereof by using an image encoder of the CLIP and a visual transformer to obtain double-branch features, the forgery perception capability of the network is enhanced through multi-domain fusion and a memory mechanism; 2, creating a pair of learnable prompts as negative and positive samples to replace discrete prompts, and finely adjusting the prompts according to the features and categories of each image; and 3, completing forgery detection by calculating similarity between prompt and image features, and integrating image and text features through feature linear modulation to obtain a forgery positioning result. According to the invention, the accuracy and generalization of image forgery detection and positioning can be improved, and the false alarm problem can be reduced, so that the performance of image forgery detection and positioning can be improved.

The invention discloses a machining control method, device and system based on a five-axis machining machine tool and a medium. The method comprises the steps that geometrical characteristic parameters of a to-be-machined workpiece and tool characteristic information are obtained, a cutting force model is established according to the obtained geometrical characteristic parameters, and cutting force distribution conditions under different paths are calculated; in the cutting force model, based on the cutting force distribution condition and the geometric feature parameters, generating an optimal cutter path corresponding to the cutter feature information by using a path optimization algorithm; according to the generated tool path and tool feature information, establishing a simulation environment of the five-axis machining tool, and simulating motion states of the tool in different paths; in the simulation environment, setting a reward function including path smoothness and cutting force stability; using a reinforcement learning algorithm to optimize the fairness of the tool path, and generating a smooth tool path; verifying the generated smooth tool path in a simulation environment; and triggering a processing execution instruction. The machining precision and the surface quality are improved.

The invention discloses an intelligent assistance method and system for software development, particularly relates to the field of software development assistance, is used for solving the problem of code snippet screening and recommendation, and aims to analyze an abstract syntax tree of code snippets through a systematic method, construct a dependency graph and a time sequence dependency graph, and calculate early warning isolation and a functional functional rate. Screening high-value code snippets by combining the indexes; converting the code snippets into high-dimensional vectors by using a deep learning model, capturing grammar and semantic features, understanding the intention of a developer in real time through a situation analysis model, and adjusting a code recommendation strategy; clustering and classifying the code snippets by using an unsupervised learning algorithm to form a code snippet library with similar semantics and related functions; the accuracy of classification, clustering and similarity calculation of the code snippets is improved, code search and recommendation are effectively supported, the development process is optimized, the reuse rate and the system stability are improved, and the maintenance cost is reduced.

The invention provides a method and system for optimizing a commodity title and computing equipment, and the method comprises the steps: obtaining commodity information which comprises an original title of a commodity and a commodity attribute; obtaining related flow words associated with the commodity by using an algorithm recommendation system; based on the commodity information, first candidate traffic words used for being added into the original title are selected from the related traffic words through a large language model; based on the commodity information and the first recommended traffic word, utilizing a large language model to generate a second candidate traffic word used for being added into the original title; based on the commodity information, the first candidate traffic word and the second candidate traffic word, determining excluded risk traffic words by using a large language model to obtain a third candidate traffic word; and selecting an optimized traffic word from the third candidate traffic words by utilizing a large language model based on the commodity information, and adding the optimized traffic word into the original title, thereby generating a new title. According to the technical scheme, merchants can be better helped to screen drainage keywords, commodity titles can be more conveniently optimized, and commodity competitiveness is improved.

The invention discloses a port multi-agent task allocation method and system based on reinforcement learning, and belongs to the field of logistics management, and the method comprises the steps: constructing a task set for various operation tasks in a port multi-task environment; aiming at different types of port intelligent equipment, constructing a dynamic model of the heterogeneous multi-agent; constructing a global environment model based on the task set and the dynamical model; constructing a collaborative multi-agent task allocation mechanism, and performing task allocation and scheduling by using the task allocation mechanism; and constructing a collaborative reinforcement learning model based on the task set, the dynamic model, the global environment model and the task allocation mechanism, and training a task execution strategy of the intelligent agent by using the collaborative reinforcement learning model. Through the hierarchical design of the high-layer strategy network and the low-layer strategy network, a complex task can be decomposed into a plurality of sub-tasks, management and optimization are facilitated, a multi-task environment can be effectively processed, and an intelligent agent can flexibly cope with different task sets.

The invention discloses a 2D-CNN network wake flow electromagnetic field detection method based on wavelet transform time-frequency spectrum characteristics. The method comprises the steps of (1) generating an underwater vehicle wake flow electromagnetic field calculation data set through a multi-physics field coupling method of hydrodynamics and electromagnetism, (2) extracting time-frequency spectrum features of a wake flow electromagnetic field through wavelet transform, and (3) developing a wake flow electromagnetic field detection model based on a 2D-CNN deep neural network. And training the model by using the data after the wavelet transform for wake flow electromagnetic field detection. According to the 2D-CNN network wake flow electromagnetic field detection method based on the wavelet transform time-frequency spectrum characteristics, the detection capability of the wake flow electromagnetic field can be effectively improved, a new technical means is provided for detection of the wake flow electromagnetic field of an underwater vehicle, and the method has a good application prospect in the field of marine target electromagnetic field monitoring and early warning.

The invention discloses a multi-degree-of-freedom ultrasonic-assisted laser cladding remanufacturing device and a working method thereof, and belongs to the field of laser cladding. Comprising a workbench, a to-be-machined workpiece arranged at the top end of the workbench through a positioning mechanism, a laser cladding mechanism erected above the to-be-machined workpiece and focusing type ultrasonic vibration mechanisms symmetrically arranged on the two sides of the laser cladding mechanism through a multi-degree-of-freedom variable-angle clamping mechanism. And focused ultrasound emitted by the focused ultrasonic vibration mechanism is aligned with a cladding area of the laser cladding mechanism. By adopting the multi-degree-of-freedom ultrasonic-assisted laser cladding remanufacturing device and the working method thereof, high-strength focusing of ultrasonic waves is realized through a multi-stage focusing method, and ultrasonic vibration can be uniformly applied during laser irradiation, so that the defects of cracks, air holes and the like are effectively reduced, and the processing quality and efficiency of laser cladding remanufacturing are improved.

The invention discloses a power equipment communication transmission system based on a power wireless private network. The system comprises a path planning module, a communication parameter adjusting module and an execution module. According to the power equipment communication transmission system based on the power wireless private network, efficient, stable and safe data transmission of power equipment in a complex network environment is realized through a low-delay and high-reliability wireless private network communication technology; the real-time performance and the stability of communication are improved through the path planning module, the communication parameter adjusting module and the execution module, the self-adaptive adjusting function is achieved, the anti-interference capability is enhanced, maintenance and optimization are facilitated, and the operation efficiency, the reliability and the safety of the power equipment are remarkably improved.

The invention discloses a traceability tracking method and system for cosmetic electronic two-dimensional codes, and relates to the technical field of cosmetic traceability tracking. According to the traceability tracking method of the cosmetic electronic two-dimensional code, after a production link is finished, data such as production original ingredient quality, component stability and production environment are stored in the two-dimensional code, scanning traceability analysis is carried out in the next link, production quality related indexes are calculated, whether the product is qualified or not is comprehensively judged, qualified products continue to be produced, and unqualified products are screened and processed. According to the invention, after each production link is finished, the production data including the original ingredient quality, the component stability, the production environment and other information are dynamically stored, a complete link recording chain is formed, the data are updated in real time through the electronic two-dimensional code, the synchronization of the traceability information and the actual production is ensured, the real-time performance and the traceability efficiency of the data are improved, and the production efficiency is improved. According to the link-by-link tracking mechanism, the transparency of the cosmetic production process is remarkably improved, so that the specific production link can be quickly positioned when the quality problem occurs.

The invention belongs to the technical field of artificial intelligence, and particularly relates to a dynamic reasoning method and system based on a language model and a knowledge graph. The method comprises the following steps: step 1, encoding an input QA context text by using a language model encoder; reasoning the knowledge graph by using a graph encoder; 2, iteration is carried out in a hierarchical aggregation module, a joint reasoning module and a dynamic pruning module, so that multi-modal fusion is carried out on the features of the QA context text and the knowledge graph; and step 3, calculating a final output answer. The dynamic reasoning method and the dynamic reasoning system have better answer accuracy, have the advantages of convenience and wide application range, can provide new thoughts and choices for the multi-modal fusion technology in the field of natural languages, and have good application prospects.