299 results about "Performance enhancement" patented technology

The invention provides an arch dam shape efficient intelligent optimization method and system based on an agent model. The method comprises the steps that an arch dam physical-numerical model is constructed, a double optimization target with structural safety and economical efficiency as the core is determined, design parameters are selected, a constraint function is set, and an evaluation index system is established; samples are generated through Latin hypercube sampling, a data set is constructed in combination with finite element calculation, and a multi-task learning architecture is adopted to train a high-precision agent model. And then, coupling the proxy model with a multi-objective optimization algorithm, quickly searching a Pareto optimal solution set, and screening out a comprehensive optimal figure by using a multi-attribute decision-making method. And finally, through a finite element simulation verification result, prediction precision and performance improvement are ensured. The method has the advantages of lightweight modeling, efficient prediction and accurate search, and provides an effective tool for intelligent optimization and rapid decision making of hydraulic structures such as arch dams and the like.

The invention belongs to the technical field of artificial intelligence, and relates to a fine-grained instruction following evaluation method and system based on a large language model. The method comprises the steps that (1) an instruction is collected from an actual application scene, a corresponding response is generated for the instruction, constraints in the instruction are decomposed, a constraint check list is generated, a plurality of comments are generated for each constraint, and each comment comprises explanation and judgment; 2) selecting an optimal comment for each constraint through a comment filtering mechanism, and forming a training data set based on the instruction, the corresponding response, the constraint check list and the optimal comment; 3) supervising and fine-tuning the large language model to obtain a supervised and fine-tuned large language model; 4) performing preference optimization on the supervised and fine-tuned large language model to obtain a preference-optimized large language model, and 5) performing fine-grained instruction following evaluation by using the preference-optimized large language model, so that greater performance improvement can be brought under lower calculation overhead, and the reliability of instruction following evaluation is improved.

The invention discloses a formula optimization method and device of an insulating material, equipment and a storage medium, and relates to the technical field of chemical materials.The method comprises the steps that initial formula information corresponding to the current insulating material is obtained, and a physical constraint model is constructed according to the initial formula information; the physical constraint model is composed of a component compatibility constraint condition, a process feasibility constraint condition and a material performance threshold condition; generating an initial candidate formula according to a physical constraint model based on the initial formula information; performing iterative optimization on the initial candidate formula by adopting a gradient response optimization algorithm to obtain a current candidate formula; and performing physical property prediction on the current candidate formula, and determining a target optimization formula according to a physical property prediction result. The optimization process is based on a scientific model and algorithm, the limitation of traditional experience can be broken through, the relation between different performance indexes is balanced, the formula combination which may be ignored is explored, and a new possibility is provided for performance improvement of the insulating material.

The invention discloses a method and system for assisting weak supervision full slide image classification performance improvement in the technical field of computer vision and computational pathology, and the method comprises the steps: obtaining an image block instance set of WSI through a data preprocessing module, and obtaining an instance feature matrix through a pre-training feature extractor; learning features which are related to tasks but possibly contain deviation by using observation branches, and adaptively fusing observation representation and potential causal representation by intervention branches through a causal gate unit so as to correct false correlation; and combining the two-branch classification loss and cross-branch adversarial supervision loss joint optimization model, and finally outputting WSI classification prediction fusing double-branch results. According to the method and the system for assisting weak supervision full slide image classification performance improvement, shortcut learning can be effectively inhibited without explicitly annotating confounding factors, the generalization performance of a model on data outside distribution is remarkably improved, and the method and the system can be flexibly embedded into various mainstream MIL frames as a universal module.

The invention provides a GPU acceleration system and method based on medium sparseness large language model sparse matrix multiplication, and belongs to the technical field of computers. The system cuts a large model weight matrix through an unstructured cutting module to obtain a sparse weight matrix; a sparse matrix is converted into an MLSF format through a sparse tensor kernel layer, a slot filling layer and a residual element layer in the sparse matrix preprocessing module so as to minimize storage and decoding overhead; a sparse matrix multiplication GPU operation module is used, an MLSF format and a dense matrix are used as input, a fragmentation strategy and asynchronous loading are combined, a pipeline mechanism of data loading, decoding and calculation is constructed, and high overlapping of calculation and memory access operation is achieved. According to the method, dense matrix calculation with sparse matrix multiplication performance exceeding highly optimized performance is realized under medium sparseness, performance improvement of 5.29 times at most is obtained, and technical support is provided for efficient deployment of a large language model.

The embodiment of the invention discloses a thread priority adjustment method and related equipment which are applied to electronic equipment, and the electronic equipment runs a first application. A first process of the first application comprises a plurality of first threads; identifying an associated thread of the first thread from the threads of the second process; the second process is a process which is operated by the electronic equipment, except the first process and the resident process and provides service for the first application; when it is detected that the running state of the first application is changed, the priority of the first thread is adjusted, the priority of the associated thread is adjusted, and the priority adjustment trend of the associated thread is consistent with the priority adjustment trend of the first thread. By adopting the embodiment of the invention, the response speed and performance of the system can be improved, and the user experience is improved.

The invention discloses a blast furnace smelting safety reinforcement learning decision optimization method based on expert strategy constraint. According to the method, an optimization strategy can be learned from an off-line expert track on the premise that operation safety is ensured. Specifically, a conditional generative adversarial mechanism is introduced to realize the alignment of strategy distribution and expert decision, and the exploration ability in a security domain is retained while the expert experience is inherited. And an independent state-action safety evaluation network is designed and a discount factor is introduced, so that the long-term accumulation risk caused by the large hysteresis characteristic of the blast furnace is effectively dealt with. In addition, memory is adopted to enhance network coding historical information, incomplete state observation is supplemented, and information deviation in decision is reduced. According to the method, effective integration of triple guidance mechanisms is realized, knowledge inheritance is realized through distributed alignment, performance improvement is driven through reward optimization, and risk prevention and control are ensured through explicit security constraints.

The invention belongs to the technical field of software engineering, particularly relates to an automatic driving system performance enhancement method based on an expert hybrid architecture, and aims to solve the core problems that an end-to-end automatic driving system is confronted with semantic fuzziness to cause unreliable decision, multi-task interference hinders optimization planning, too long reasoning delay increases driving risks and the like. According to the method, an ExpertAD framework is provided, task key features are amplified through a perception adapter (PA), and the relevance of scene context understanding is guaranteed; related driving tasks are dynamically activated through a sparse expert mixture (MoSE), and task interference is minimized; and in combination with a customized training loss function, collaborative optimization of planning effectiveness and reasoning efficiency is realized. Experiments show that compared with an existing method, the method has the advantages that the average collision rate is reduced by 20%, the reasoning delay is reduced by 25%, higher multi-skill planning capacity is achieved in rare scenes (such as accident handling and first-aid vehicle avoiding), and good generalization is achieved for unseen urban environments.

The invention provides a remote sensing large model performance improvement method and device based on composite visual coding, and relates to the technical field of remote sensing image large language models. According to the method, a network structure is rewritten on the basis of an open-source visual universal language model, a composite visual coding ConvToMe layer used for extracting local details and capturing global semantics is added, the ConvToMe layer extracts the local details through a ConvNeXt layer, the global semantics are captured in combination with a ToMe ViT layer, a remote sensing large model with improved performance is obtained, and the performance of the remote sensing large model is improved. Through composite visual coding, the global context and local details are reserved at the same time, the robustness is improved, and the model precision is improved at the same time.

The invention discloses a method for improving the performance of double-independent quantum key distribution of actual reference system measurement equipment, belongs to the technical field of quantum key distribution (QKD), and aims to improve the performance of the double-independent quantum key distribution when non-ideal factors such as a post-pulse effect of a detector and finite sample statistical fluctuation are considered. The invention relates to a method and a system for improving the performance of a reference system measurement device by introducing an advantage extraction technology into a double independent quantum key distribution (RFI-MDI-QKD) protocol of the reference system measurement device. Simulation results show that when the post-pulse effect and the statistical fluctuation effect are considered, the advantage extraction method can effectively improve the security key rate and the security transmission distance of the RFI-MDI-QKD protocol on the premise of not changing optical hardware. According to the invention, a valuable reference technology can be provided for practical research of the RFI-MDI-QKD system.

The invention relates to the technical field of buck-boost switching control, and discloses a buck-boost seamless switching MPPT control method which has remarkable advantages. The single-side boost-buck function can stably control the output voltage in a full-voltage range, and the problem that the efficiency and stability of the system are affected due to the fact that the traditional topology cannot accurately regulate and control the power under the photovoltaic high-voltage working condition is effectively solved. A buck-boost mode is seamlessly switched, the response speed and reliability are improved, and faults caused by improper mode switching are avoided; when the method is applied to an energy storage product, the PV high voltage condition does not exist, the additional control logic can be reduced, the system complexity is reduced, the reliability is improved, meanwhile, the development cost and the maintenance difficulty are reduced, and powerful support is provided for photovoltaic system performance improvement and energy storage product optimization.

The invention provides a SiC controller electromagnetic-thermal-force full-coupling simulation design method and system, and belongs to the technical field of power electronic device simulation design, and the method comprises the steps: constructing a bidirectional real-time coupling model of an electromagnetic field, a thermal field and a mechanical stress field; the irradiation dose is determined based on the expected working environment of the SiC controller, irradiation damage factors are introduced to correct SiC material parameters, and a bidirectional real-time coupling model is input; different temperature conditions and stress conditions are input into the bidirectional real-time coupling model for multi-working-condition simulation, performance parameters of the SiC material are output, and a quantitative relation model of the performance parameters, the temperature and the stress is established to reveal the performance degradation law of the SiC material; and based on the bidirectional real-time coupling model and the performance degradation law of the SiC material, taking the switching loss and the temperature rise as optimization targets, and carrying out collaborative optimization on parameters of the SiC controller. The simulation precision is improved, the performance of the SiC controller is optimized, the reliability is improved, and the service life is prolonged.

The invention discloses a composite cutter bar design parameter optimization method. The method comprises the following steps: constructing a low-fidelity data set, a high-fidelity data set and a dual-stage deep learning model; in the first stage, low-fidelity features are learned based on an LSTM network, and a cutter bar performance prediction value is output; in the second stage, high-fidelity data and a low-fidelity prediction result are input for residual learning, and a comprehensive prediction model is obtained. Performing global optimization in the cutter bar design parameter space through a particle swarm optimization algorithm to obtain optimized design parameters; performing experimental verification to obtain actually-measured dynamic performance parameters and errors between the calculated predicted values and the actually-measured dynamic performance parameters; when the error does not meet the preset threshold value, according to the error distribution extension parameter value interval, new high-fidelity data is generated, then the comprehensive prediction model is updated, optimization verification is repeated, and final cutter bar design parameters are output. According to the method, efficient optimization and dynamic performance improvement of cutter bar design parameters are achieved.

The invention provides an ultra-thin wing anti-bending performance quantitative evaluation method which comprises the following steps: acquiring geometric parameters and material characteristic data of an ultra-thin wing, constructing a digital twin model by adopting a finite element analysis method, and simulating stress distribution under boundary conditions to obtain initial mechanical response characteristics; the obtained risk assessment matrix is adopted to calibrate the limitation of a static test, and boundary condition constraints are fused to obtain a calibrated anti-bending energy index set; constructing an evaluation framework of a quantitative system through the calibrated anti-bending energy index set, classifying performance levels in different load scenes by adopting a support vector machine algorithm, and determining an overall structure reliability score; and iteratively updating the parameter setting of the digital twin model according to the obtained anti-bending performance improvement scheme, simulating the optimized flight performance, and obtaining the final quantitative evaluation system output.

The invention discloses a transient performance enhancement-oriented multi-spacecraft attitude cooperative tracking control method, which provides a fixed time controller through an inversion control strategy, realizes quick response of a multi-spacecraft attitude system, and ensures that a consistency error is converged to a sufficiently small domain within fixed time. A fuzzy weight updating law is designed, dependence on system model information is reduced, and adaptability and high robustness are achieved. According to the asymmetric fractional obstacle Lyapunov function, adherence of multiple spacecrafts to time-varying constraints is guaranteed, meanwhile, the problem that the control amplitude related to a large constraint boundary in a traditional logarithmic obstacle Lyapunov function is too large is solved, and therefore the proposed method can process constrained and unconstrained scenes. A saturation threshold event triggering strategy is designed, a multi-spacecraft system is allowed to select three different triggering strategies by adjusting parameters, the transmission frequency of control signals is flexibly changed, and control bus resources are reasonably distributed.

The invention relates to the technical field of extremely-small curved surface structures, in particular to an optimal design method, system and equipment for an extremely-small curved surface structure and a storage medium. The minimum curved surface structure optimization design method comprises the steps that a multi-objective optimization problem is determined according to design variables and optimization objectives of a three-period minimum curved surface lattice structure; constructing an approximate function relationship between the optimization target and the design variable by utilizing a self-adaptive multi-target engineering optimization method of a Kriging agent model; latin hypercube sampling and a sequential quadratic programming optimization algorithm are combined to carry out optimization design on the three-period minimal curved surface lattice structure. According to the design method, the light / bearing / wave-absorbing minimum curved surface lattice sandwich structure with good structure-performance dual characteristics can be designed, the wave-absorbing structure is filled with the minimum curved surface lattice structure in a shape follow-up mode, design and manufacturing of a wave-absorbing and bearing integrated structure are achieved, and rapid iteration and performance improvement of design of the wave-absorbing and bearing structure are supported.

The invention relates to the technical field of Internet of Things communication, in particular to a LoRa network dynamic link self-adaptive control method and system based on environment awareness, and the method comprises the steps that the system firstly presets initial communication parameters for a terminal and collects the data and link information of the terminal; based on the building and traffic data on the transmission path, an environmental perception model is constructed to quantify environmental impact factors, and a dynamic evaluation matrix is established in combination with performance indexes. Calculating the comprehensive performance index of the current link based on the matrix, and predicting the change of the signal-to-noise ratio to obtain a relative difference value; the system generates optimization suggestions of communication parameters by analyzing a performance index change trend and applying an adaptive algorithm, and predicts brought performance improvement. By balancing energy consumption and time delay cost required by performance improvement and adjustment, parameter adjustment is executed only when the income is greater than the overhead. And finally, completing the optimization of all terminals within the fixed network distribution time, and evaluating the overall performance of the network. According to the invention, the accuracy of dynamic link adaptive control of the LoRa network is improved.

The invention relates to a dynamic power control method and system for a storage and charging all-in-one machine, and the method comprises the following steps: data collection and monitoring: collecting and analyzing key parameters in real time through multi-source data fusion; dynamic safety threshold calculation: dynamically calculating a safety operation threshold of each subsystem based on real-time data, and ensuring that the system operates in an optimal state; multi-objective optimization control is carried out, an intelligent algorithm is adopted, and an optimal balance point is found among battery safety, power grid stability and charging efficiency; according to a fault processing and redundancy mechanism, a multi-layer fault response system is established, and the safety and continuity of the system under the abnormal working condition are ensured; according to the energy efficiency management and strategy optimization, long-term performance improvement and economical efficiency optimization are achieved based on a data closed loop, and the problem that potential safety hazards are caused due to the fact that a traditional storage and charging all-in-one machine system neglects the common influence of multiple factors such as SOC, temperature and power grid loads is solved.

The invention belongs to the field of artificial intelligence and machine learning, aims to solve the problems of high manual generation cost, poor authenticity, low negative sample quality and insufficient scene coverage of multi-modal tool calling data, and comprises four core modules: a tool knowledge base module stores API metadata through PostgreSQL, and generates various user intentions and coherent calling sequences in combination with pre-training LLM; the multi-agent interaction module generates conversation and calling data containing multi-modal input through cooperation of a user, an assistant and an environment agent; the adversarial environment simulation module generates high-quality negative samples in batches by means of three adversarial strategies and automatically marks the high-quality negative samples; and the human feedback iteration module determines convergence by a performance improvement rate according to a model failure case optimization strategy. The system supports local / cloud deployment, adapts to LLM and Agent model training, and is suitable for training and robustness evaluation of dialogue type AI assistants and vertical field intelligent Agents.

This invention relates to the field of intelligent machine tool technology, specifically to an intelligent machine tool equipped with a transverse movement mechanism, including an up-and-down transverse movement mechanism, a left-and-right transverse movement mechanism, a left-and-right micro-movement mechanism, and a pressure mold. The left-and-right transverse movement mechanism is mounted on the outer side of the up-and-down transverse movement mechanism, and the left-and-right micro-movement mechanism is mounted on the outer side of the left-and-right transverse movement mechanism. The pressure mold is located on the top of the left-and-right micro-movement mechanism. The core of this invention lies in eliminating the independent dedicated mounting plate and instead directly connecting the mold-moving assembly to the machine head frame. By eliminating redundant mounting plates and auxiliary connectors, the overall mass and inertial load of the moving parts are effectively reduced, which is beneficial to improving the dynamic response performance and motion accuracy of the system. Secondly, the direct connection method significantly compresses the structural space, making the layout more compact and significantly improving space utilization. Based on the simplification of the overall structure, the miniaturization and lightweight design of the machine head module are realized, providing a structural foundation for the deployment and performance improvement of the equipment in a limited space.

The invention discloses a novel high-performance enhanced GaN power device and a preparation method thereof, the power device comprises a depletion mode HEMT and an enhanced HEMT which share a substrate structure, the depletion mode HEMT comprises a first composite channel layer, a first barrier layer, a first grid electrode, a first source electrode and a first drain electrode, the enhanced HEMT comprises a second composite channel layer, a second barrier layer, a p-GaN layer, a second grid electrode, a second source electrode and a second drain electrode, the first grid electrode and the second source electrode are connected through first interconnection metal, and the first source electrode and the second drain electrode are connected through second interconnection metal. According to the device, monolithic integration is adopted, parasitic parameters in the device can be reduced, the reliability of the grid electrode can be improved, and meanwhile the power device has the advantages of being high in breakdown voltage, large in current, low in on-resistance and the like through the multi-layer channel design.

The invention relates to a shock tube vibration and noise reduction performance improving device for engineering, and belongs to the technical field of engineering anti-explosion tests. A shock tube of the device is coaxially connected with a fragment filtering module, a to-be-tested piece is arranged between the shock tube and the fragment filtering module, and the fragment filtering module is coaxially connected with a vibration and noise reduction performance improving module; high-pressure shock waves are emitted from the shock tube to crush a to-be-tested piece, and fragments of the to-be-tested piece and the decelerated shock waves are attenuated and dispersed after passing through the fragment filtering module and the vibration and noise reduction performance improving module. According to the shock tube vibration and noise reduction performance improving device for the engineering, vibration and noise reduction of the shock tube for the engineering under compressed air explosion are achieved, the shock tube vibration and noise reduction performance improving device can be used for vibration and noise control of a large indoor shock tube test, and operability is high.

The invention relates to the technical field of snowboard accessories, and particularly discloses a snowboard performance enhancing gasket with both sliding and skipping functions and a stress dispersion function. The bolt fixing groove is formed in the central position of the angle adjusting chassis; the gasket base mechanism is arranged on the outer portion of the angle adjusting chassis in a sleeving mode, the gasket base mechanism comprises a base plate, and a plurality of hollow grooves are formed in the outer portion of the base plate; through the arrangement of the angle adjusting chassis, the bolt fixing groove, the gasket base mechanism and the cushioning sheet, the whole circle of arc-shaped curved surface is designed on the bottom surface of the gasket base, and jump-type take-off actions such as Nollie and Ollie fused in the carving and sliding process are combined; the dynamic adaptation characteristic of the arc-shaped curved surface base mechanism and the stress dispersion function of the cushioning sheet between the base and the snowboard are utilized to form double guarantees, and therefore the fracture risk of the snowboard under jumping impact is effectively reduced.

The application provides a reliability analysis method for positioning accuracy of a shield tool changing robot, realizes sensitivity analysis of error parameters by constructing a kinematic model and an error proxy model, and accurately evaluates the failure probability of positioning accuracy. The assembly and machining errors are comprehensively considered, the error parameters are treated as random variables, the accuracy and efficiency of the analysis are improved. The mean and standard deviation of the random variable are used for sensitivity calculation, and a failure calculation method of positioning accuracy sensitivity is obtained, which provides an important basis for optimization design. The application not only helps to reduce the tool changing time, improve the construction efficiency and reduce the safety risk, but also provides an accurate and reliable tool changing robot manufacturing scheme for designers. The method has universality and flexibility, and can be widely applied to performance improvement and fault prevention of the shield tool changing robot. The implementation of the application will promote the development of shield construction technology and provide strong support for industrial automation and efficient production.

This invention relates to the field of collision detection technology, and in particular to a general performance evaluation system for visual collision detection algorithms. The system includes a video data module for calibrating optical variables and storing motion modal videos of approaching collision objects to obtain a standard video dataset; a data transmission module for exporting data from the standard video dataset to a test algorithm model to obtain test results, and also for importing the test results into an analysis module; and an analysis module for analyzing the test algorithm model using performance analysis and algorithm theory analysis to obtain analysis results. This invention can comprehensively evaluate collision detection algorithms, helping to analyze the specific advantages and disadvantages of visual collision detection algorithms, promoting the overall development and performance improvement of such algorithms, and increasing the practical applicability of visual collision detection algorithms.

The invention discloses a self-adaptive anti-interference adjustment method and system for a Bluetooth headset, and the method comprises the steps: carrying out the normalization processing of each extracted feature, calculating a single-dimensional interference coefficient based on the normalized features, obtaining a comprehensive interference grade coefficient, and carrying out the comparison of the comprehensive interference grade coefficient and a preset threshold value, dividing interference levels; forming a state space by using the comprehensive interference level coefficient, the current signal-to-noise ratio, the current packet loss rate and the current data transmission delay, forming an action space by using adjustable parameters of the Bluetooth headset, and constructing a reward function by using the adjusted performance improvement effect to determine an optimal adjustment action set; issuing adjustment parameters corresponding to the optimal adjustment action set to a corresponding function module of the Bluetooth headset, continuously monitoring a preset period number, and then taking an average value as the adjusted performance; and calculating a reward value based on the difference between the performance after adjustment and the performance before adjustment, updating a state and action value table according to the reward value, realizing strategy iteration, and forming closed-loop feedback optimization.

The invention discloses a simulation optimization method for preparation parameters of a silicon carbide particle reinforced metal matrix composite, which comprises the following steps: S1, setting the temperatures of a preform and a sand box to be room temperature, and selecting circular ceramic particles for pouring to obtain a casting; s2, after three-dimensional graph drawing is carried out on a casting in Creo, a model is converted into an igs format and then imported into Procast, and then grid division is carried out in the Procast; s3, other variable parameters are kept constant, only the pore size of the ceramic particles is changed, and the optimal pore size is determined through simulation analysis. Accurate guidance is provided for performance improvement and practical application of the ceramic particle reinforced metal matrix composite, scientificity and preciseness of simulation research are ensured, and follow-up experimental verification and practical production and application of the material are powerfully supported.

The application discloses a kind of adjustable distributed optical fiber sensing performance promotion device, sensing system and method, device includes: narrow linewidth laser, acoustooptic modulator, first signal generator and frequency shift ring;Narrow linewidth laser is configured to output narrow linewidth laser to acoustooptic modulator;First signal generator is configured to generate pulse function signal, function signal controls acoustooptic modulator, and single-frequency laser is modulated into pulsed light;Acoustooptic modulator is configured to modulate single-frequency laser into corresponding pulsed light in acoustooptic modulator according to the pulse function signal generated by first signal generator;Frequency shift ring is configured to multiple cyclic frequency shifts of pulsed light into it, and generate broadband multi-frequency light output.The device generates MHz-level pulsed light by acoustooptic modulator modulation, and generates GHz-level broadband multi-frequency light by multiple cyclic frequency shifts of frequency shift ring, which is applied to sensing system, and the response bandwidth of sensing system can be improved, and the flexibility of sensing test is improved.

The invention discloses a green low-carbon production method of an HG785 steel plate by replacing hardening and tempering with direct quenching and tempering, the thickness of the steel plate is 30-50mm, and the production method comprises the following steps: smelting and continuously casting various components according to a reasonable proportion; heating and rolling; directly quenching; two-phase region coupling tempering; and after discharging, performing air cooling or water mist cooling to room temperature to obtain a finished steel plate. The production method provided by the invention is high in universality, low in modification requirement on an existing medium-thickness plate production line (with accelerated cooling and tempering capabilities), easy to realize industrial large-scale production and remarkable in economic benefit and social benefit; through the collaborative design of'component fine adjustment and gradient quenching + two-phase region coupling tempering ', the mechanical property of the steel plate comprehensively reaches the standard (the yield strength is larger than or equal to 785 MPa, and the impact energy stability at the temperature of-40 DEG C is larger than or equal to 80 J) while greenness and low carbon are achieved, it is proved that'green production and product performance improvement can be synchronously achieved', and the practicability and popularization value of the green low-carbon technology are further enhanced.

The invention belongs to the technical field of artificial intelligence, and relates to a dynamic controllability management method and system for long-term evolution of a large model, and the method comprises the steps: S1, obtaining the current state information of the large model; s2, calculating a controllability score of the large model based on the output distribution entropy, the distribution variable quantity and the logits drift of the large model, judging whether the current state of the large model deviates from a controllability critical zone or not based on the controllability score, if the current state deviates from the controllability critical zone, entering S3, and otherwise, directly entering S4; s3, actively correcting the offset behavior of the large model in the training process through a controllable calibration head, so that the current state of the large model does not deviate from the controllable critical zone; and S4, dynamically adjusting a training strategy of the large model based on the controllability score and the task category. According to the method, the large model has the capabilities of long-term plasticity maintenance, continuous performance improvement and rapid adaptation to new tasks, and the fundamental upgrade from a static optimization model to a self-evolution model system is realized.