74 results about "Learning agent" patented technology

This invention discloses an online reinforcement learning method and system based on a virtual sandbox. The online reinforcement learning method includes: constructing a virtual sandbox that is functionally identical to the real environment but physically isolated; simulating real-world operational challenges within the sandbox environment; a reinforcement learning agent to be trained; environmental state monitoring and reward evaluation; receiving experience trajectories generated by the agent's interaction with the sandbox; and using an online reinforcement learning algorithm to update the agent's policy network in real time. This enables the agent to learn autonomously through direct interaction with the environment in a safe and high-fidelity environment, with its reward signal based on objective and authentic feedback obtained from the interaction with the environment, eliminating the need for manually labeled data.

A federated learning framework including household agents configured to continuously learn model parameters for managing charging periods and discharging periods of household batteries, and microgrid agents to maximize use of local energy based on a pricing policy, including accessing power from other microgrids when there is insufficient local energy to cover local demand. and selling surplus energy to the other microgrids when power generation by the microgrid surpasses the local demand. Each household machine learning agent is configured to control household energy demand from and supply to a microgrid which they are connected in order to minimize household energy cost while adapting to changes in the energy price that is determined based on the pricing policy of the microgrid agent that encourages reduction of carbon emission. A federated learning engine combines the model parameters from the household machine learning agents to update a global household machine learning agent.

The application relates to a government affair cross-domain collaborative response method and system based on a large model and RPA dynamic arrangement. The method converts user appeal into structured instructions by adopting a domain fine-tuning large language model through an intention decoupling module; then, a reinforcement learning agent hub in a dynamic assembly module is generated according to the instruction and the real-time state of an RPA cluster, and a dynamic scheduling strategy is output; further, a cross-domain execution module drives an RPA robot to interact with a heterogeneous government affair system in a non-intrusive manner to execute a task according to the strategy; finally, a work order derivation module generates a standard work order by using a conditional generative adversarial network (cGAN) to aggregate returned data. The application has the beneficial effect that the constructed cognitive-to-execution automation closed loop can effectively improve the efficiency, accuracy and flexibility of government affair services while ensuring the safety of existing systems.

The application provides a method for generative heterogeneous logic optimization based on critical path partitioning. The method includes: converting a digital logic circuit into an And-Inverter Graph (AIG) format; partitioning the AIG using reinforcement learning based on critical path information of the circuit to obtain critical-path-aware partitions so as to reduce influence of critical path changes brought by the partitioning on timing performance; respectively and automatically exploring an optimized structure and synthesis flow suitable for each partition using the reinforcement learning method; selecting optimization operators for various structures using a reinforcement learning agent guided by area-delay metrics and optimizing each partition; and merging the optimized partitions in the AIG format. In the application, the digital logic circuit is partitioned based on critical paths, and reinforcement learning is used to explore and optimize structures and strategies based on area and delay.

The application discloses a kind of based on crystal plastic finite element and sequence condition generation formula deep learning agent model's oriented electrical steel cold rolling texture fast prediction method, system and medium, with cold rolling process parameter set as input, with ODF slice image as output;Through the crystal plastic finite element model driven by physical mechanism, "process path-ODF" virtual labeling data pair is generated in batch in pre-set cold rolling process design space, and data set is constructed;Then using sequence condition generation formula deep learning agent model realizes the end-to-end nonlinear mapping from cold rolling process path to high-dimensional ODF image, so that the premise of not needing to call crystal plastic finite element model successively, realize the fast prediction, visualization and path regulation of cold rolling texture evolution.This application solves the problem of existing fusion crystal plastic simulation precision advantage and deep learning model high efficient generalization ability, realizes the fast and accurate prediction of cold rolling texture evolution.

The application provides a composite gas cylinder mechanical property prediction method and system based on micro damage evolution, and relates to the technical field of composite material mechanical property analysis. The method establishes a micro representative volume element model and performs multi-axial progressive loading analysis, extracts damage initialization conditions and evolution rules, and obtains homogenization output parameters through homogenization processing. The mapping relationship between the microstructure parameters and the macro damage parameters is established by using a machine learning agent model, and the macro finite element model is implanted, and finally the predicted failure pressure and the predicted fatigue life are output. The application realizes high-precision mechanical property prediction of the composite gas cylinder from the micro damage mechanism, and has the advantages of high prediction accuracy and self-calibration.

A computing system (10) including one or more processing devices (12) configured to receive a semantic entitlement (30) that semantically specifies an access permission scope (34) of a machine learning (ML) agent (22) included in an ML system (20). The semantic entitlement has a natural language format. At least in part by processing the semantic entitlement at a generative language model (32) included in the ML system, the one or more processing devices identify one or more resources (52) that are included in the access permission scope indicated in the semantic entitlement. The one or more processing devices grant an ML agent of the plurality of ML agents access to the one or more identified resources. At the ML agent, the one or more processing devices compute an agent output (60) based at least in part on the one or more identified resources. The one or more processing devices output the agent output to an additional computing process (62).

Method and system are provided for performing physical-aware logic synthesis. A synthesis environment receives an RTL design, associated design constraints, and one or more design flow and physical design parameter settings. The RTL design is synthesized into a gate-level netlist while incorporating physical design considerations. A scoring result is computed from a logic synthesis result including the synthesized gate-level netlist to provide a quantitative measure of design quality. In certain aspects, the scoring result includes a scattering metric that evaluates placement density of standard cell instances, rankings across multiple modules, PPA results, and runtime prediction estimates. In further aspects, the design parameters are iteratively adjusted using artificial intelligence, reinforcement learning, or machine learning agents to optimize design quality. The invention enables improved integration of logical and physical design stages at advanced semiconductor process nodes.

The application discloses an engine electronic control method based on an AI agent, and relates to the field of engine electronic control.The method comprises the following steps: S1, based on bottom support, collecting engine state signals, receiving vehicle instructions, collecting deviation data, and executing control instructions; the bottom support comprises a heterogeneous multi-core SoC hardware architecture; S2, based on the heterogeneous multi-core SoC hardware architecture, at least three physically isolated computing domains are divided, and the following is executed respectively: reaction agent logic, planning agent logic and learning agent logic; S6, further comprising deploying a shared digital twin engine: a digital twin model based on an integrated physical mechanism model and a neural network model, which simulates the running state and performance parameters of the engine in real time, and provides a prediction basis for the planning agent logic and an update object for the learning agent logic.

This invention provides a method for determining the adjustable power range of an aluminum electrolytic cell. The method includes establishing a multiphysics CFD simulation model for the electrolytic cell, and using this model to solve for the distribution characteristics of temperature, voltage, and concentration under different input power P, as a training dataset. A dual-agent model is constructed, comprising a cell safety function model and a production economy function model, and supervised learning is used to train the dual-agent model using the training dataset. A multi-agent reinforcement learning system is constructed to solve for the minimum descent power and maximum escalation power of the aluminum electrolytic cell. This invention accurately characterizes the thermal field, electric field, and concentration distribution within the aluminum electrolytic cell through CFD simulation, and combines this with a safety agent model to evaluate in real time whether key indicators such as temperature, voltage, and concentration exceed limits, ensuring that the adjustment range is always within the safety boundary. The reinforcement learning agent of this invention autonomously seeks the power point with maximum output / lowest energy consumption while ensuring safety.

This invention provides a vehicle dynamic stability control method based on reinforcement learning and variable pole configuration, relating to the field of vehicle dynamics control. This method constructs multi-dimensional criteria for the centroid sideslip angle phase plane and energy phase plane to identify the vehicle's stability state and energy evolution trend in real time. It utilizes a reinforcement learning agent to make online decisions about the expected closed-loop pole positions of the system based on the current state, enabling the configuration of aggressive poles to improve flexibility under stable conditions and conservative poles to ensure stability under unstable conditions. Furthermore, this invention designs a safety monitoring mechanism that includes pole physical domain constraints, enabling the reshaping of the vehicle's dynamic characteristics through feedback control without altering the driver's steering intentions, effectively solving the challenge of real-time trade-off between flexibility and safety in intelligent chassis control.

This invention provides a method, device, and medium for optimizing the crystal structure of high-entropy alloys based on a reinforcement learning agent, relating to the fields of materials computation and artificial intelligence. The method includes: acquiring the X-ray diffraction pattern of the target material as the optimization target; constructing a reinforcement learning agent environment, wherein the environment uses element sequences as state representations and modifies lattice site elements as the action space; generating a crystal structure based on the current element sequence and calculating its X-ray diffraction pattern; calculating the similarity between the predicted pattern and the target pattern; calculating a reward value based on the similarity improvement; training the agent using a proximal policy optimization algorithm or a Monte Carlo tree search algorithm; iteratively optimizing the element sequence until the similarity reaches a preset threshold; and outputting the optimized element sequence and the corresponding crystal structure. This addresses the problems of existing high-entropy alloy design relying on human experience, high trial-and-error costs, and low optimization efficiency.

The present application relates to a kind of heuristic Q-learning-based multi-robot full coverage path planning method, including constructing global gridding map and dividing mutually non-overlapping connected sub-region by DARP algorithm;Each robot independent sub-environment is constructed and Q-learning agent is initialized;Agent uses the enhanced state definition of "absolute coordinate+local environment topological feature", and the optimal action is selected by heuristic strategy of unvisited priority and backtrack distance guidance;The Q value table is updated based on the comprehensive function of coverage reward, dynamic repeated access penalty and completion reward;Path is generated after iteration training to full coverage or reaching preset number.The present application reduces 23.8% of total moving steps under the premise of ensuring 100% coverage, reduces path repetition rate from 26.35% to 3.36%, reduces 44.1% of turning number, significantly improves work efficiency, and can be widely applied to collaborative work scenarios such as cleaning robot cluster and agricultural automatic harvester group.

This invention discloses an intelligent water quality prediction method for watershed floods based on dynamic weight optimization and deep reinforcement learning. The method comprises the following steps: collecting historical flood water quality observation data and historical flood hydrological data; standardizing the historical flood hydrological data to obtain comparison sequences and reference sequences, and then performing dimensionless processing; calculating dynamic grey relational coefficients and weighted grey relational degrees using the dimensionless comparison sequences and reference sequences, and sorting them to form a weighted grey relational degree sequence; using the weighted grey relational degree sequence to filter out effective historical flood fields, and performing dynamic weight optimization based on a long short-term memory network and a reinforcement learning agent based on the TD3 algorithm; and calculating the final predicted flood water quality data using the historical flood water quality observation data and the historical flood weights. This invention's prediction method enables intelligent water quality detection during flood processes, providing accuracy, real-time performance, and adaptability.

The application relates to a building initial fire spread situation prediction method, system, equipment and medium based on space-time evolution modeling. The method comprises the following steps: based on the building information model data of a target building, extracting space units and connection relationships and constructing a building topology graph; initializing a graph neural network fire spread prediction model and an evacuation behavior simulation model containing a reinforcement learning agent group based on the topology graph; coupling the two models and conducting adversarial training on the evacuation agent group; obtaining building fire data to input the model for fire state evolution calculation, calling the trained model to simulate personnel evacuation behavior, and outputting a joint prediction situation containing a future space-time fire spread range and a dynamic evacuation path of personnel. The method can overcome the defects of traditional separate simulation methods that ignore the two-way dynamic interaction of fire and personnel behavior, realize coupled space-time evolution prediction of the fire spread and evacuation process, and improve the accuracy and timeliness of building fire emergency situation judgment.

The application provides a multi-dimensional cognitive state vector simulation agent digital textbook dynamic adaptability evaluation method, and relates to the technical field of artificial intelligence assisted education. The method comprises the following steps: extracting teaching entities and semantic relationships between the entities in a source file of a digital textbook to be evaluated through a multi-modal analysis model, and constructing a knowledge graph; initializing multiple groups of virtual learning agents based on a pre-trained large language model, and configuring differentiated cognitive state vectors for each group of virtual learning agents; driving each group of virtual learning agents to traverse the teaching knowledge graph according to a topological sorting by using a knowledge boundary mask mechanism, obtaining understanding feedback of each node by each group of virtual learning agents, and generating confusion degrees and thought chains in the understanding feedback process; calculating a cognitive block index and an adaptability distribution curve of the source file of the digital textbook by using a weighted summation algorithm, and generating an evaluation report. The application simulates human cognitive processes by using a large language model agent, and evaluates the quality and adaptability of a digital textbook.

In one example aspect, the present disclosure provides an example computer-implemented method for generating feedback signals for training a machine-learned agent model. The example method can include obtaining an output of a machine-learned agent model, the output including a next state feature generated by the machine-learned agent model based on a sequence of preceding states. The example method can include processing, using a machine-learned reward model, the output and the sequence of preceding states to generate a quality indicator indicating a quality of the next state feature in view of the preceding states. The machine-learned reward model could be trained by retrieving reference data from a reference data source and computing one or more quality indicators in view of a respective training input and output(s), and the reference data. The example method can include outputting the quality indicator to a model trainer for updating the machine-learned agent model.

The present disclosure provides systems and methods for determining an anomaly score for defect detection in connection with a kinematic implementation, such as a vehicle implementation. The method includes receiving one or more design parameters relating to the vehicle implementation (104). Furthermore, one or more shape-function parameters relating to the implementation are received. A machine learning engine (ML engine) (316) is also employed, based on the one or more design parameters and the one or more shape-function parameters, to create an environment such that a learning agent associated with the environment is configured to analyze the design parameters and the shape-function parameters to identify anomalous patterns associated with the vehicle implementation (104).The anomaly score for error detection related to vehicle operation is then calculated based on the identified anomalous patterns.

The application relates to a risk assessment method and device for an urban lifeline pipe network, and relates to the field of urban pipe network system safety monitoring; the method can accurately quantify pipe network risks and provide data support for urban risk decision-making. The method comprises the following steps: acquiring attribute parameters of each pipe section in a pipe network system and soil body attributes, and constructing a finite element model of the pipe network system; acquiring seismic motion records of a region where the pipe network system is located, performing dynamic analysis based on amplitude-modulated acceleration time histories, and obtaining stress data of the pipe sections under different seismic motion peak accelerations; determining the damage levels of the pipe sections through the stress data, and counting the number of times when the damage levels reach preset levels to obtain failure probabilities of the pipe sections; training a machine learning agent model through the seismic motion records and the failure probabilities, and determining the failure probabilities of to-be-measured pipe sections through the model; determining importance indexes of the to-be-measured pipe sections through a weighted graph theory model; and determining risk assessment results of the to-be-measured pipe sections based on the failure probabilities and the importance indexes.

The application provides a kind of adaptive modal injection type current source converter energy efficiency optimization control method based on PPO algorithm, it is related to power supply system field, it is applied to injection type current source converter, injection type current source converter includes twelve pulse inverter and multiple injection branches cascaded in the DC side of twelve pulse inverter, and the method comprises: collecting real-time meteorological data and real-time power grid state data;Through the reinforcement learning agent based on proximal policy optimization algorithm, based on real-time meteorological data and real-time power grid state data, determine the optimal action, wherein the optimal action at least includes target level number, the number of injection branches that need to be started and the number of injection branches that need to be disabled;According to optimal action, control multiple injection branches run, with the advantage of realizing the optimal operation of injection type current source converter under all operating conditions.

The application discloses a kind of based on cellular automaton and the rapid prediction and process reverse design method, system and medium of high temperature annealing organization evolution of oriented electrical steel of bidirectional consistency deep learning agent model, with initial state feature s and annealing process vector u as input, generate the virtual data set of "(s,u)→ODF" in batch through cellular automaton model in preset process space, and train bidirectional consistency deep learning agent model, realize the rapid prediction of high temperature annealing texture and the process of target texture guide reverse push.This application can realize rapid forward prediction on the basis of absorbing mechanism model credibility, and further realize executable annealing process path from target texture reverse calculation, so as to support stable, replicable high temperature annealing process design and quality consistency control.

The application discloses a service function chain arrangement and scheduling method and device and medium. The method adopts a security reinforcement learning method framework, allocates a series of security reinforcement learning agents to each service function chain, and the agents are simultaneously guided by a reward function and a constraint function for optimization. Each agent can control the service function deployment position, transmission path, processing bandwidth and transmission bandwidth of each data packet according to the specific service quality requirement of the service function chain corresponding to the business and the local network state observed at the moment. The constraint-oriented optimization mechanism of the method can improve the network bandwidth resource utilization rate, meet the deterministic end-to-end delay and jitter performance requirements, and provide a certain and efficient network service for the delay-sensitive business in the industrial scene.

The application discloses a simulation warm-up method based on historical simulation data migration learning, and relates to the technical field of simulation warm-up methods.The simulation warm-up method comprises the following steps: obtaining thermal physical parameters of a historical scene spacecraft and corresponding temperature field distribution to construct a training data set; training a deep learning agent model based on the training data set; migrating model parameters of the deep learning agent model to initial parameters of a deep learning agent model in a new scene; obtaining thermal physical parameters of the spacecraft in the new scene and corresponding temperature field distribution to construct a fine-tuning data set; fine-tuning model parameters of the deep learning agent model in the new scene based on the fine-tuning data set; and predicting the temperature field distribution based on a simulation task in the new scene and by using the fine-tuned model parameters of the deep learning agent model in the new scene.The application can effectively reduce the iteration calculation period and accelerate the solution process of fine grid simulation by quickly inferring the temperature field in a new thermal simulation scene.

Deep reinforcement learning agents for demand response in home energy management systems are provided via training an agent via power availability data, electricity use data for an electrical load in a household, and an effect on a length of service of a power supply device to optimize a reward function that rewards: reduced electricity usage at peak demand times for the power grid according to the power availability data, increased user satisfaction with activation of the electrical load, and increased length of service for electrical devices used to delivery electricity to the electrical load; deploying the agent to the household; and activating electrical devices that are part of the electrical load for the household according to a schedule generated to optimize the reward function for delivery of power from at least one of the power grid and the power supply device to the household.

Processing system (100) for executing an artificial intelligence or Al application comprising: a wearable device (110) comprising a first electronic unit (111) and a first battery (112); an electronic mobile terminal (120) connected to the wearable device (110) through a wireless connection (140), said electronic mobile terminal (120) comprising a second electronic unit 121 and a second battery (122); a cloud or edge server (130) connected to the electronic mobile terminal (120) through a telecommunications network (150), said server (130) comprising a third electronic unit 121; wherein the processing system is configured so that, at the first request of execution of the Al application, the Al application is executed by the first electronic unit (111), the second electronic unit (121) and the third electronic unit (131) according to a first predefined computational load distribution, and wherein the second electronic unit (121) is programmed for executing, after the first execution of the Al application, a Reinforcement Learning (RL) agent configured for determining at least one second distribution of the computational load on the basis of wearable device state data, electronic mobile terminal state data, server state data, wireless connection state and telecommunication network state data so as to minimize the energy consumption of the first battery maintaining an end-to-end latency smaller than the application time.

This invention discloses an artificial intelligence-based intelligent dismantling sequence planning method for power batteries, belonging to the field of power battery dismantling technology. The method includes initializing a dismantling sequence simulation environment, loading a dismantling action constraint relationship network, a dismantling tool action model, and a dismantling safety rule base; employing a reinforcement learning agent to drive Monte Carlo tree search, simulating component dismantling order to generate multiple candidate sequences, and simultaneously collecting simulation data such as tool selection, time consumption, safety risk score, and resource consumption assessment for each step; constructing a multi-objective evaluation model with four dimensions: dismantling efficiency, safety, resource cost, and component integrity recovery rate, quantifying and fusing data, and calculating a comprehensive score; selecting the highest-scoring sequence as the recommended sequence, and outputting an executable process guidance document with detailed steps, a tool list, and risk warnings. This invention overcomes the limitations of human experience, covers multiple paths under complex constraints, quantifies and balances multi-objective conflicts, and improves the scientific nature of power battery dismantling.

The application provides an interaction control method and device based on an agent telephone customer service, equipment, a storage medium and a program product, relates to the technical field of artificial intelligence, and the method comprises the following steps: acquiring a speech recognition result, semantic integrity information and user speaking state information output by an audio recognition module; inputting the speech recognition result, the semantic integrity information and the user speaking state information into a reinforcement learning model in a reinforcement learning agent to obtain a decision action output by the reinforcement learning model; and performing a calling operation on a business agent based on the decision action. By deploying the reinforcement learning agent in the voice customer service system, the application decides the calling time of the business agent according to the semantic integrity and the speaking state of the user voice, effectively avoids the mis-triggering problem caused by user pauses or voice punctuation, and improves the accuracy and fluency of voice interaction without modifying the existing business agent.

This application provides a method, apparatus, device, and medium for autonomous driving data acquisition based on multimodal uncertainty fusion. The autonomous driving data acquisition method includes: acquiring multimodal data of the vehicle's surrounding environment in real time; inputting the multimodal data into an uncertainty estimation model, performing uncertainty fusion estimation processing on the multimodal data to determine a fusion uncertainty value; inputting the fusion uncertainty value into a reinforcement learning agent, generating a data acquisition optimization strategy based on the current environment state vector and the fusion uncertainty value; acquiring new data according to the data acquisition optimization strategy, and using the new data to adjust the autonomous driving model online. Based on the fusion uncertainty value predicted by the model, a data acquisition optimization strategy is dynamically generated to optimize autonomous driving data acquisition, improve data quality, and enhance system robustness.