11 results about "State representation" patented technology

The invention relates to the technical field of reinforcement learning, and provides a decision support system and method based on a large language model and reinforcement learning, and the method comprises the steps: firstly obtaining a source state representation of reinforcement learning, carrying out the data enhancement of the source state representation through an external knowledge base, converting the enhanced state representation into a state representation of a preset large language model, and carrying out the recognition of the state representation. The method comprises the following steps: acquiring a state representation, inputting the state representation into a preset large language model to generate a reinforcement learning agent enhanced state representation function and an internal reward function, updating and maintaining a Lipexishz array based on the two functions to generate a state representation meeting a smooth condition, and finally inputting the state representation meeting the condition into a classifier decision model to generate a classifier decision model. And obtaining a decision result. Therefore, by introducing the external information, the source state characterization quality of reinforcement learning can be improved, so that the decision-making ability and adaptability of the intelligent agent are improved.

The invention discloses a control system and method of an intelligent servo press, relates to the technical field of press control, and constructs a uniform stamping state representation through deep excavation and fusion of force and displacement process data streams generated in the servo press. On the basis, a double-time-scale implicit state estimation algorithm is introduced, time sequence modeling and deep analysis are carried out on weak and gradual change characteristics contained in a full-stroke force displacement curve, and therefore accurate quantitative evaluation of the implicit state of mold abrasion is achieved. Then, the quantized wear value serves as a core decision factor, a self-adaptive control mechanism based on a compensation knowledge base is driven, and an optimal process compensation strategy and a sliding block target track are dynamically generated. Therefore, the product quality consistency in the precise stamping process can be improved, the service life of the die is effectively prolonged, the production cost is reduced, and the intelligent level of stamping production is comprehensively improved.

The invention provides a reinforcement learning state representation optimization method for intelligent control, and belongs to the technical field of reinforcement learning, and the method comprises the steps: firstly, carrying out the extraction of state features from the perspective of feature correlation and time continuity between a current state and a subsequent state, and then introducing a comparison cooperation module, robustness of state representation is enhanced by utilizing comparative learning, a query state and a key state are generated, parameters of a query encoder and a key encoder are optimized by adopting an exponential moving average (EMA) mechanism, and finally, a self-filtering module and a comparative collaboration module are combined, so that a self-filtering function is realized. The core problems in reinforcement learning tasks such as high-dimensional state representation, time-dependent modeling and state robustness improvement are successfully solved; the method is especially suitable for reinforcement learning tasks with complex environment dynamics, such as robot control and automatic driving.

The invention relates to the technical field of artificial intelligence and reinforcement learning, and discloses a cross-border trade intelligent decision optimization method based on reinforcement learning. The method comprises the following steps: constructing a composite state representation fusing tax, logistics, exchange rate, demand and competitive behavior; high-dimensional state compression is realized through an auto-encoder; designing a hierarchical action architecture to decouple a macroscopic strategy and a microscopic operation; a Nash equilibrium guided reward shaping function is introduced, and an equilibrium income deviation is estimated in combination with anti-factual reasoning; carrying out stable training by adopting a double-delay depth deterministic strategy gradient algorithm with state transition consistency constraint; an online fine tuning mechanism is deployed to adapt to a real business environment. According to the method, the strategy convergence speed, the annual profit rate and the responsiveness to policy mutation are remarkably improved.

This invention discloses a method, apparatus, equipment, and medium for coordinated operation of liquefied natural gas (LNG) receiving terminals based on multi-agent collaboration, relating to the field of artificial intelligence technology. The method includes: constructing a hierarchical collaborative multi-agent architecture comprising equipment agents, process agents, and safety agents; real-time collection of multi-source state data, which is then fused through a state representation layer to generate an operational profile containing a unified state ontology for the entire station; collaborative decision-making by process and equipment agents based on the unified state ontology and preset rules to generate a coordinated operation sequence; risk verification of operation instructions by the safety agent using a constraint shielding mechanism, followed by issuance to multiple execution devices for coordinated execution; real-time collection of execution feedback to update the operational profile and triggering a readjustment of the operation sequence within the architecture. This invention solves the problems of fragmented data representation, poor scheduling flexibility, and lack of dynamic safety verification in existing technologies.

The application discloses a high-dimensional data computing architecture, method, system and device based on recursive projection compression and closed audit. After the method obtains high-dimensional input data, at least one recursive projection compression is performed to form an intermediate audit state representation and / or a decision state representation; a state space deviation function and a state evolution trend prediction operator are calculated on the state representation; dimension hedging processing is performed between different recursive levels, different projection dimensions, different loop representations or different boundary representations, and combined with boundary legality audit, resource closed audit, capacity constraint audit, loop consistency audit, platform flatness audit and stability window audit, a control amount, a compensation amount, a reconstruction amount or a branch adjustment amount is generated; a deterministic early termination is triggered through consistency comparison of parallel audit loops, and in the preferred embodiment, the integrity of the compression result is verified through reverse checking. The application is suitable for industrial monitoring, omics data analysis and other high-dimensional complex system state audit scenarios.

Self-tuning database performance management system (100) comprising: a) a monitoring module configured to collect real-time performance data from cloud-based databases hosted on AWS and Snowflake; b) a feature extraction module configured to convert performance data into a state representation; c) a reinforcement learning module configured to receive the state representation and generate voting actions; and d) an action execution module configured to apply the reconciliation actions to the database environment.

The application discloses an intelligent optimization method and system for cross-regional power dispatching based on federal learning, and the application is implemented. Through heterogeneous data fusion, implicit association in unstructured data is converted into a calculable quantitative feature, which provides more comprehensive state representation information for subsequent model training, avoids the loss of key information caused by a single data type, and is beneficial to improving the accuracy of training data selected by the model. A model structure decoupled from public / private parameters is adopted to realize the compatibility of common sharing and individual isolation, and the practicality of the obtained model is improved. Through the execution of a differential privacy federal aggregation operation driven by differentiated weights, the dispatching model can iterate towards dispatching urgency, improve the adaptability of the dispatching model to the actual application scenario, and improve the security of the dispatching model. Finally, by setting an output mechanism of linkage between optimization and fault checking, the accuracy, reliability and security of the final output of the cross-regional dispatching scheme are improved.

The application discloses a kind of state representation generation and updating method and system of multi-source heterogeneous time series data, and relates to data processing technical field.The method comprises: collecting multi-source heterogeneous time series data and preprocessing into uniform format feature record;Perform alignment operation based on multiscale time window and semantic anchor point, obtain multi-source feature record;Calculate feature difference degree to identify conflict data and generate conflict score;Conflict data is weighted and dissolved based on data source reliability, to obtain fusion feature value and its confidence;The final fusion weight is calculated using the product model to integrate data sources, features, time effectiveness and confidence weight;According to final fusion weight and fusion feature value, generate the multi-dimensional state vector of target user;Multi-dimensional state vector is output to generate individualized planning scheme, collect review feedback events, and based on closed-loop state machine Trigger the collection and update of next round data.The application improves the accuracy of data fusion and the objectivity of state representation.

This invention discloses a method and system for analyzing the state of small industrial models using dual transfer learning, belonging to the field of industrial equipment state analysis technology. The method includes: constructing a set of state segments arranged in time window order by uniformly aligning and segmenting multi-channel time-series data generated during industrial equipment operation; mapping each state segment to a set of equipment operating state feature vectors; generating a target state representation vector set based on the set of equipment operating state feature vectors; performing dual transfer alignment calculations on the source and target state representation sets, and constraining the state representations with numerical alignment results; performing a confidence-based adaptive update calculation on the target state representation vector set, and outputting the final state representation result. This invention achieves robust analysis and cross-device adaptation of equipment operating states, significantly improving the stability, transferability, and engineering applicability of state representations.