27 results about "Double loop" patented technology

The invention discloses a digital twinning multi-agent reinforcement learning intelligent decision-making system and method with a secure memory playback mechanism, and the system comprises a digital twinning module which is used for constructing a virtual model and synchronizing the virtual model with a physical entity in real time; the multi-agent reinforcement learning module is used for carrying out strategy learning based on a constrained Markov decision process and balancing performance and safety through a Lagrange multiplier; the safe memory playback module is used for weighting and playing back the experience samples according to the risk and the timeliness so as to improve the learning safety; the reversible grey influence network module is used for causal modeling and reasoning and enhancing decision interpretability; the double-loop self-constraint control module ensures that a control action is always in a physical safety boundary through a barrier function and safety projection; and the convergence and stability criterion module is used for verifying strategy security convergence and system asymptotic stability. According to the method, the problems of strategy border crossing, virtual-real mismatching and the like in the high-risk manufacturing process are solved, and multi-target optimal control under the safety constraint is realized.

The invention discloses a permanent magnet synchronous motor PI double-loop control method based on deep reinforcement learning. The method comprises the following steps: establishing a permanent magnet synchronous motor double-loop coupling mathematical model; a traditional FOC composite controller is constructed; designing a state feedback and reward mechanism of the double-ring TD3; constructing a speed loop and current loop TD3 + PI composite controller; and carrying out system stability analysis. According to the method, a system is divided into an inner ring and an outer ring based on hybrid cooperative control fusing deep reinforcement learning and classical control, the problem that robustness is insufficient under system disturbance through PI control and the problem that tracking precision is reduced due to load sudden change and noise interference in PMSM operation are combined, nonlinear disturbance is dynamically compensated through the strategy learning ability of a TD3 algorithm, and the tracking precision is improved. Meanwhile, soft update is introduced to improve convergence stability, and a compound control strategy based on double-loop TD3 + PI is provided. According to the invention, complex non-linear processing is simplified, and the stability of the system and the accuracy of rotating speed tracking are improved at the same time.

The invention discloses a robot motion control method and system based on cerebellum reinforcement learning, and the method comprises the steps: obtaining a current environment state vector, and inputting the current environment state vector to a main strategy channel and a cerebellum compensation channel in parallel; the main strategy channel outputs a basic action based on a long-term task target, and the cerebellum compensation channel outputs a compensation action responding to real-time dynamic through an efficient query mechanism; synthesizing the basic action vector and the compensation action vector into a synthesized action vector driving robot; feeding back latest data after the robot drives the motion action vector, and determining a sensory prediction error based on the latest data; and updating the original parameters of the cerebellum compensation channel based on the sensory prediction error, and optimizing the original strategy parameters of the main strategy channel based on the latest data. According to the invention, by constructing a parallel double-channel architecture, functional decoupling of advanced decision and rapid adaptation is realized, and unification of rapid adaptation and continuous optimization is realized through a double-loop hierarchical learning system.

The invention relates to the field of artificial intelligence and optimization algorithms, in particular to an intelligent evolution optimization method based on large language model driving. According to the method, semantic comprehension and reasoning capabilities of large language models such as DeepSeek and the like are deeply embedded into a core process of a traditional evolutionary optimization algorithm (PSO, GA), a'semantic cognition-dynamic decision 'double-loop mechanism is constructed, and full-process optimization from problem modeling, intelligent search to explainable output is realized. According to the method, the effectiveness of the method is verified on two typical optimization scenes of a 0-1 knapsack problem (discrete type) and Rosenbrock and Sphere function (continuous type), and through semantic-driven population initialization, dynamic parameter regulation and control and a closed loop feedback mechanism, the core problems that a traditional algorithm is low in initial solution quality, prone to local optimum, insufficient in calculation efficiency and the like are effectively solved.

The invention discloses a real-time signaling flow intelligent arrangement system based on queue optimization, and relates to the technical field of communication networks, and the system comprises a multi-dimensional feature perception and fusion module which is used for collecting and outputting various dynamic attribute parameters of signaling in real time; and the self-adaptive scheduling decision module is connected with the multi-dimensional feature perception and fusion module and is used for receiving the dynamic attribute parameters and generating a scheduling instruction. According to the real-time signaling flow intelligent arrangement system based on queue optimization, by introducing a double-loop intelligent architecture of prediction and decision verification, the perspectiveness and reliability of signaling flow scheduling are effectively improved. According to the technical scheme, the system can make decisions by integrating the multi-dimensional dynamic factors, and through double inspection of the rule base and micro-simulation, the reasonability of the scheduling strategy and the system stability are still kept in the face of complex burst traffic, and the end-to-end service quality is guaranteed.

The invention relates to a federated learning method and system based on dynamic clustering and double-ring cooperative training, and relates to the field of federated learning. The problems that a clustering mechanism is insufficient in adaptability and low in knowledge sharing efficiency are solved. The method comprises the following steps: step 1, receiving non-sensitive features from each industrial control device participating in federal learning; 2, clustering the industrial control equipment according to the similarity of each non-sensitive feature to form a plurality of federal sub-clusters; 3, independently executing inner ring knowledge distillation training by each federal sub-cluster, executing outer ring model sharing training among the federal sub-clusters, and judging whether a training ending condition is met or not; and step 4, updating the non-sensitive features, and repeating the step 2, the step 3 and the step 4 until a training ending condition is met.

The application discloses a kind of robot motion control method and system based on cerebellum reinforcement learning, it includes: obtaining current environment state vector, and parallel input to main strategy channel and cerebellum compensation channel;Main strategy channel outputs the basic action based on long-term task target, and cerebellum compensation channel then outputs compensation action by efficient query mechanism to cope with real-time dynamics;The basic action vector and compensation action vector are synthesized into synthesized action vector to drive robot;After robot drive movement action vector, feedback latest data, and determine sensory prediction error based on latest data;Based on sensory prediction error, the original parameters of cerebellum compensation channel are updated, and the original strategy parameters of main strategy channel are optimized based on latest data.The application realizes the functional decoupling of high-level decision and rapid adaptation by constructing parallel double-channel architecture, and the hierarchical learning system of double loop realizes the unity of rapid adaptation and continuous optimization.

The present application relates to the technical field of industrial process control, in particular to an intelligent control method for multi-process collaborative management of injection stretch blow molding machines, comprising: in the mold cooling main process for medical consumable production, when the monitored main process variable reaches the preset threshold value, based on the material property model associated with the rheological melt index, the operation instruction of the preparatory process executor is adaptively generated, and the preparatory process parallel to the main process is started; using the model prediction double-loop control loop, through real-time control of the inner loop, according to the dynamic prediction of the synchronization deviation of the two process procedures, the preparatory process executor is continuously adjusted to realize synchronization, and at the same time, through online learning of the outer loop, according to historical error feedback, the prediction model inside the loop is continuously self-corrected. Through the establishment of a model prediction double-loop self-correcting control loop, the present application realizes predictive collaborative control of multiple asynchronous executed process procedures.

The application relates to the technical field of laser system control, and discloses a multi-stage PID parameter regulation and control method for a laser frequency locking system based on deep learning, which comprises the following steps: acquiring an error signal time sequence of a PDH laser frequency locking system; outputting a frequency locking state score and fast-slow double-loop PID parameters through a deep learning model; judging whether the frequency locking state score and the fast-slow double-loop PID parameters both satisfy corresponding preset conditions; if the frequency locking state score and the fast-slow double-loop PID parameters do not both satisfy the corresponding preset conditions, iteratively optimizing the deep learning model until the frequency locking state score and the fast-slow double-loop PID parameters both satisfy the corresponding preset conditions; and outputting the fast-slow double-loop PID parameters of the optimized deep learning model. The application can realize online self-adaptive reasoning and closed-loop optimization of fast-slow double-loop multi-stage PID control parameters, and improve the intelligent level and the ability to adapt to complex working conditions of the PDH laser frequency locking system.

The application discloses a multi-target distribution and path planning method based on a simulated annealing algorithm, and first, a multi-target distribution model is established, relevant variables are defined, and a target function is constructed to minimize total flight cost while considering task execution time and fuel limit; secondly, a double-loop model is constructed, the problem is abstracted into a Hamilton loop problem, including an outer Hamilton loop and an inner Hamilton loop, which are responsible for path planning between regional nodes and path planning of targets in a region respectively; finally, a simulated annealing algorithm is used to solve the TSP model of the double Hamilton loop, and through parameter setting, initial solution generation, solution transformation, Metropolis criterion application and cooling strategy, an optimal multi-target distribution strategy and corresponding path planning are found. The method can effectively handle large-scale, multi-target and multi-constraint complex problems, improve calculation efficiency, and find a solution close to the global optimum, and has important practical application value and market prospect.

The application relates to a PID parameter optimization method for reducing motion control position deviation uncertainty, which comprises the following steps: firstly, determining the PID parameter optimization form and constraint conditions, and constructing an optimization objective function for the motion control position deviation uncertainty; secondly, establishing a Gaussian process regression proxy model between the PID parameters and the optimization objective function; thirdly, determining a next group of control parameters according to an expected enhancement acquisition function and measuring the corresponding optimization objective function value; finally, optimizing the PID parameters by using a double loop, wherein the inner loop uses randomly set initial control parameters to iteratively execute and obtain inner loop optimization control parameters, the outer loop is used for starting the inner loop optimization process multiple times, and finally, the final tuning parameter value is calculated by using non-convex scenario optimization; the selection of the optimization objective function of the application considers the average position deviation and the position deviation uncertainty, the optimization method adopts a double loop Bayesian optimization mode, and the method has the advantages of not being prone to local optimization and good consistency of optimization results under the action of uncertainty disturbance.

The application relates to an air-ground cooperative navigation control method and system based on global priori of an unmanned aerial vehicle and elastic preset performance tracking of an unmanned vehicle, and belongs to the technical field of autonomous navigation and cooperative control of unmanned systems. In view of the problems of a single system, such as a large blind area of sensing, a prominent calculation bottleneck, easy falling into a local minimum value in local obstacle avoidance, weak anti-interference performance in trajectory tracking and waste of communication resources in a complex environment, a layered cooperative architecture is adopted, the unmanned aerial vehicle is used to complete mapping, two-dimensional projection and global path planning and issue a discrete topological path, the unmanned vehicle generates a local trajectory through spatial domain robust differential game, dynamic optimization is realized by combining elastic preset performance control and internal and external double-loop reinforcement learning, high-precision tracking and global optimization of energy consumption are realized, and an event triggering mechanism is used to reduce communication load. The application improves navigation reliability, trajectory stability and resource utilization efficiency in a complex dynamic environment.

The invention relates to the technical field of machine learning and intelligent optimization, and discloses a multi-objective collaborative optimization and decision-making method for laser cladding process parameters, and the method comprises the steps: an initialization stage: constructing a physical correlation model and a Gaussian process agent model; in the fast loop optimization stage, the estimation performance is obtained by utilizing the physical correlation model so as to quickly update the Gaussian process agent model; in a slow loop calibration stage, concurrently obtaining foundation truth value data of the key points; and a slow loop reconstruction stage: reversely updating the physical correlation model by using the ground-based truth value, and globally reconstructing the Gaussian process proxy model. According to the method, a double-ring optimization architecture with cooperative work of fast-ring fast exploration and slow-ring concurrent calibration is constructed, so that dynamic correction of system deviation of the low-cost physical correlation model by using a high-cost foundation truth value is realized, and parameter exploration efficiency is greatly improved while optimization precision is ensured; the problem that an existing optimization method is difficult to balance among speed, cost and precision is solved.

This invention relates to a multi-agent autonomous learning method for hydraulic supports, belonging to the field of intelligent coal mine technology. It includes: using the scene features input by a learning-type real-loop agent and the output displacement / pressure-time action sequence as input to a learning-type dual-loop agent; the learning-type dual-loop agent corrects the displacement / pressure-time action sequence to obtain a corrected displacement / pressure-time action sequence; the corrected displacement / pressure-time action sequence is then returned to the learning-type simulation loop agent to verify for conflicts; if conflicts are found after correction, the correction continues, iterating until the corrected displacement / pressure-time action sequence is conflict-free; finally, the final actual action strategy is expanded to a preset strategy set for training a higher-quality learning-type real-loop agent. This invention achieves closed-loop data fusion analysis that combines real data and simulated data.

The invention relates to the technical field of distributed collaborative learning and elastic response, in particular to an eBPF-based distributed collaborative learning and elastic response method, system, equipment and medium, which comprises the following steps of: performing global analysis in a user mode based on system behavior data acquired by a kernel mode eBPF program to construct a normal behavior baseline model; the baseline model is synchronized to a kernel mode after being lightened; in the kernel mode, deviation detection is carried out on system behaviors generated in real time based on the synchronous baseline model, and a risk score is output; determining a response level in combination with the risk score and a preset asset criticality; and in a kernel mode, executing a response action corresponding to the response level. The method has the beneficial effects that the inherent contradiction between the response speed and the detection precision in the prior art is successfully solved by constructing the fast and slow double-loop collaborative security system.

The present invention discloses a network traffic anomaly detection and calculation method based on ESN, which relates to the field of traffic detection technology. The method includes the following steps: constructing a causal reasoning type reserve pool: embedding a causal neuron cluster in the ESN double-loop reserve pool to obtain a causal reasoning type reserve pool; pre-detection based on causal reasoning: analyzing the reserve pool state in real time based on the causal neuron cluster; adaptive resonance threshold adjustment: calculating the characteristic resonance degree between the reserve pool state and the historical normal mode in real time; multi-scale time gating dynamic adjustment: using the multi-scale time gating mechanism to dynamically adjust the detection window width; anomaly judgment and feedback learning: judging whether the current traffic is abnormal. The present invention realizes causal reasoning modeling by embedding a causal neuron cluster in the ESN double-loop reserve pool, thereby achieving high-precision detection of network traffic anomalies, solving the problem in the prior art that a single detection window lacks causal reasoning capabilities and is difficult to take into account multi-dimensional anomaly patterns.

This invention provides an auxiliary decision-making method and system for the renovation of cultural relic-related water conservancy facilities, belonging to the field of water conservancy engineering management technology. The decision-making method achieves quantitative collaborative decision-making for the dual objectives of engineering safety and cultural relic protection, transforming the qualitative requirements of cultural relic protection into structured data that can be calculated alongside engineering indicators, thus overcoming the previous decision-making dilemma of separating these two dimensions. It establishes a complete technical closed loop from assessment to design, decomposing complex decision-making problems into three logically clear stages: functional zoning, strategy optimization, and structural matching, ensuring the rigor and traceability of the decision-making logic. The introduction of a dual-loop feedback mechanism significantly improves the rationality throughout the entire lifecycle. The inner loop feedback ensures the mechanical consistency between the design scheme and the decision-making objectives, avoiding performance deviations during the design phase; the outer loop feedback uses actual operation and maintenance data to back-optimize the initial decision-making model, enabling the system to have self-learning and self-evolution capabilities.

The invention relates to the technical field of automatic testing, in particular to a multi-time-scale energy optimization scheduling method. The invention relates to a multi-time scale energy optimization scheduling method, which comprises an energy system and is characterized in that the energy system comprises a data and constraint convergence layer; performing feature engineering and state estimation; generating a scene library; a timing predictor; gNN flow proxy is carried out; performing day-ahead distribution robust optimization; performing intra-day rolling consistency correction; a real-time controller; modeling health and life; consistency coordination and double-loop verification are carried out; training and distilling; and monitoring and auditing. Compared with the prior art, the multi-time scale energy optimization scheduling method provided by the invention comprises the following steps: generating a tail scene through a diffusion model and incorporating the tail scene into robust optimization; a DC-TSP is used to extract multi-scale time sequence features; the Auto-FRL is combined with the safety barrier projection to realize millisecond-level control; gNN is fast approximate to the power flow, and NeuralODE accurately describes energy storage degradation; and the global consistency of day-ahead, day-intra and real-time solutions is ensured through double-ring verification.

The invention relates to a multi-agent autonomous learning method for a hydraulic support, and belongs to the technical field of coal mine intellectualization. Comprising the steps that scene features input by a learning type reality ring Agent and a displacement / pressure-time action sequence output by the learning type reality ring Agent serve as input of a learning type double-ring Agent, the learning type double-ring Agent corrects the displacement / pressure-time action sequence, and the corrected displacement / pressure-time action sequence is obtained; the corrected displacement / pressure-time action sequence is returned to the learning type simulation ring Agent again to verify whether conflicts exist or not, if conflicts exist after correction, correction is continued, loop iteration is conducted till the corrected displacement / pressure-time action sequence does not have conflicts, and a final actual action strategy is expanded to a preset strategy set. The method is used for training the learning type reality ring Agent with higher quality. According to the invention, closed-loop data fusion analysis combining real data and simulation data is realized.

The application discloses a beyond-visual-range air combat double-loop coupling autonomous maneuver decision-making method and device based on situation driving, a medium and a product, and relates to the technical field of aerospace. The method comprises the following steps: using a trained LSTM model, respectively according to state control information of an enemy target and a local machine, performing recursive prediction to obtain multi-step track prediction information of the enemy target and the local machine; calculating the situation change gradient of both sides; inputting the track information of the local machine and the state control information of the enemy target into a trained first reinforcement learning model to generate a main action instruction; inputting the track information of the local machine, the state control information of the enemy target and the situation change gradient of both sides into a trained second reinforcement learning model to generate a preloaded action instruction; and using a null space behavior method to fuse the main action instruction and the preloaded action instruction to obtain an autonomous maneuver decision-making instruction. The application can reduce the decision-making risk under incomplete information and realize smooth tactical conversion of a combat aircraft.

The application discloses a kind of real-time signaling stream intelligent arrangement systems based on queue optimization, and the application relates to communication network technical field, comprising: multi-dimensional feature perception and fusion module, for real-time acquisition and output the multiple dynamic attribute parameters of signaling;Adaptive scheduling decision module is connected multi-dimensional feature perception and fusion module, for receiving the dynamic attribute parameter and generating scheduling instruction.The real-time signaling stream intelligent arrangement systems based on queue optimization, by introducing the double-loop intelligent architecture of prediction and decision verification, effectively improves the foresight and reliability of signaling stream scheduling.System can make decision by comprehensively multi-dimensional dynamic factor, and after the double inspection of rule base and micro simulation, thereby still maintain the rationality of scheduling strategy and system stability when facing complex burst traffic, guarantee the quality of service of end to end.

The invention belongs to the technical field of generative artificial intelligence, and particularly relates to a complex task-oriented double-loop language model agent reasoning method which comprises the following specific steps: S1, task whiteboard establishment: establishing a task whiteboard of a'state-action-result 'graph structure, and recording task targets, resources, available tools and constraint conditions; s2, loop-first planning: the planning Agent generates a plurality of candidate action chains, scores based on a target distance, resource consumption and a historical success rate, and selects an optimal scheme; and S3, executing loop attempt: calling a tool by the execution Agent according to a scheme, collecting feedback in real time and updating a task whiteboard. According to the method, 3-5 candidate action chains instead of a single linear path are generated before execution of the first-view ring, so that the jamming risk of walking to the bottom by one path is avoided from the source.

The application provides a man-machine co-driving closed-loop optimization method and system based on end-to-end and world model fusion, and belongs to the technical field of intelligent driving. In view of the technical problems that the existing man-machine cooperation signal cannot be fully utilized, the closed-loop optimization mechanism is imperfect, and the world model scene generation lacks risk assessment and strong man-machine cooperation constraint, the end-to-end automatic driving model is deeply fused with the multi-level dynamic world model, and a data and cognition double-driven man-machine co-driving cooperative interaction framework is constructed. Through the man-machine bidirectional mutual feedback closed-loop structure, the driver takeover, intervention and preference signal is fed into the system in real time to drive the decision optimization and model iteration, and based on the risk state, scene prediction and cause inversion result, the active safety bottom protection and dynamic interaction guidance are provided to the driver. Through the prediction and planning iteration mechanism coupled with the world model and the double-loop closed-loop evolution mechanism, the system is continuously optimized.

The invention discloses an automobile door hinge production line dynamic scheduling method and system based on digital twinning, and the method comprises the steps: S1, collecting the original production state data of an automobile door hinge production line, and carrying out the data cleaning preprocessing of the original production state data, and obtaining the production state data; s2, a production line digital twinborn model is obtained through combination of a preset mechanism model and a real-time data driving function, and the production state data is input into the production line digital twinborn model for simulation initialization to obtain a simulation prediction result; s3, constructing a double-loop reward agent combining a reinforcement learning agent and a double-loop reward function, inputting a simulation prediction result into the double-loop reward agent for online learning optimization, and outputting an optimal scheduling strategy parameter; and S4, the optimal scheduling strategy parameters are issued to an execution mechanism at the bottom layer of the production line, and real-time dynamic scheduling of the automobile door hinge production line is achieved.

The application discloses a dynamic scheduling method and system for an automobile door hinge production line based on digital twinning, and comprises the following steps: S1, collecting original production state data of the automobile door hinge production line, performing data cleaning and preprocessing on the original production state data to obtain production state data; S2, obtaining a production line digital twinning model by combining a preset mechanism model and a real-time data driving function, inputting the production state data into the production line digital twinning model for simulation initialization to obtain simulation prediction results; S3, constructing a double-loop reward intelligent agent combined with a reinforcement learning intelligent agent and a double-loop reward function, inputting the simulation prediction results into the double-loop reward intelligent agent for online learning optimization to output optimal scheduling strategy parameters; and S4, issuing the optimal scheduling strategy parameters to an execution mechanism at a bottom layer of the production line to realize real-time dynamic scheduling of the automobile door hinge production line.

This invention discloses an adaptive control method and system for denim pre-shrinking process based on machine learning, belonging to the field of fabric pre-shrinking technology. The method includes: acquiring denim production process data and real-time images, and inputting them into a pre-established lightweight physical simulation network to obtain a virtual state of the fabric surface; inputting the virtual state of the fabric surface into an adaptive control network for optimization processing to obtain pre-shrinking process parameters; based on the pre-shrinking process parameters, executing a double-loop learning process to obtain sorting control commands for controlling the pre-shrinking machine; and driving the pre-shrinking machine to perform actions according to the sorting control commands, completing the adaptive control of the denim pre-shrinking process. This invention uses a lightweight physical simulation network as a differentiable forward simulator, which can accurately deduce the virtual moisture content, shrinkage rate, and smoothness distribution of the fabric surface, providing accurate virtual quality feedback for parameter optimization. Combined with the double-loop learning mechanism, it ensures the uniformity and high standards of pre-shrinking quality from both the source and the process.

The invention discloses an beyond-visual-range air combat double-loop coupling autonomous maneuver decision-making method and device based on situation driving, a medium and a product, and relates to the technical field of aerospace, and the method comprises the steps: carrying out the recursive prediction through employing a trained LSTM model according to the state control information of an enemy target and a local machine, and obtaining a state control result; obtaining multi-step track prediction information of the enemy target and the local aircraft; calculating situation change gradients of the two parties; inputting the flight path information of the aircraft and the state control information of the enemy target into a trained first reinforcement learning model to generate a main action instruction; inputting the flight path information of the local aircraft, the state control information of the enemy target and the situation change gradient of the two parties into a trained second reinforcement learning model to generate a preloading action instruction; and fusing the active action instruction and the preloading action instruction by using a null-space action method to obtain an autonomous maneuvering decision instruction. According to the method, the decision risk under incomplete information can be reduced, and smooth tactical conversion of a fighter can be realized.