95 results about "Model switching" patented technology

The invention discloses a method for intelligently regulating and controlling production parameters in a fruit concentrated juice production process, which comprises the following steps of: acquiring multi-dimensional process parameters such as temperature, pressure, flow, concentration, equipment state and the like in real time through a multi-channel sensor network, and forming a standardized data sequence after filtering, normalization and drift correction; extracting stage features by using technologies such as a sliding window and Fourier transform, and inputting the stage features into the lightweight classification model to realize production stage identification; in combination with an identification result, dynamically calling a corresponding multi-target optimization sub-model, and realizing nonlinear prediction and optimal solution selection of process parameter setting by adopting an LSTM and a multi-target genetic algorithm; on the basis of real-time feedback, the performance of the model is automatically evaluated, self-adaptive adjustment and optimization of the optimization algorithm are achieved through reinforcement learning and an incremental updating mechanism, multi-target collaborative optimization, self-adaptive adjustment and model switching in the production process can be achieved, and the consistency of production efficiency and product quality is improved.

The invention relates to the technical field of cross of storage controllers and artificial intelligence acceleration, and discloses a large model reasoning system and method based on combination of a flash memory controller and an NPU (Network Processing Unit), and the large model reasoning system comprises the flash memory controller, the NPU and a flash memory array, the flash memory controller integrates a host interface module, a flash memory interface module, an independent AI acceleration interface module and an AI management engine, and the AI management engine autonomously completes NPU initialization, model weight direct loading, KV Cache hierarchical management, RAG knowledge base retrieval and model switching; the flash memory array is divided into a firmware partition, an AI special partition and a user storage partition, and different data storage requirements are met. According to the method, large model reasoning with low delay and low CPU dependence can be realized, and the model loading delay is reduced from 5-30 seconds to lt; after 500 milliseconds, the CPU occupancy rate of the host is reduced from 15-25% to lt; 2%, and concurrent operation of 4-8 models is supported. According to the invention, integration of storage and calculation is realized, and edge end, data center and mobile equipment scenes are adapted.

The invention relates to the technical field of FSRU regasification control systems, and particularly discloses an FSRU regasification heat source intelligent switching control system based on multi-source coupling optimization, and the system comprises a multi-source grading collection and edge preprocessing module which is used for collecting heat source-load-environment-heat storage-abnormity five-dimensional basic data and dynamic operation constraint data, and sending the data to a data processing module; carrying out hierarchical transmission and normalization processing on the data; and the dual-mode adaptive multi-source-heat storage coupling model module is used for pre-constructing a lightweight coupling model and a high-precision coupling model, and dynamically switching between the lightweight coupling model and the high-precision coupling model according to a working condition trigger signal. By introducing a transient asynchronous compensation mechanism, the system effectively solves the problem of control window period caused by time consumption of high-precision model calculation, ensures that the system can still maintain smooth and continuous instruction output at the moment of extreme load fluctuation, and eliminates the risk of pressure oscillation caused by model switching.

The application discloses a low-altitude intelligent networking dynamic collaborative reasoning method based on multi-agent reinforcement learning, relates to the technical field of low-altitude intelligent networks and edge artificial intelligence, and comprises the following steps: preloading a light model and a complex model on each unmanned aerial vehicle (UAV), deploying a complete complex model by means of a ground station, and constructing an air-ground integrated intelligent reasoning system. By introducing an enhanced multi-agent deep reinforcement learning algorithm, each UAV can dynamically select a model type, determine a model segmentation point, and reasonably allocate bandwidth and ground computing resources based on the state of the UAV, network conditions and task characteristics during task execution, so that multi-DNN flow heterogeneous resource-aware collaborative reasoning is realized. The application can effectively improve reasoning accuracy and reduce average delay under different device performance, bandwidth conditions and task density, has good system scalability and adaptability, and solves the problems of limited single-machine processing capacity, restricted communication resources and inefficient model switching.

The embodiment of the invention provides a management operation execution method and device, a storage medium and an electronic device, and the method comprises the steps: determining a priority set for at least one model in a first node, and determining the occupation number of model processing units (MPU) of each model in the at least one model and the processing time of each model, the MPU occupation number and the processing time have a first association relationship; and executing a management operation on the at least one model through the priority and / or the first association relationship, wherein the management operation comprises at least one of the following: deactivating the model, adjusting the number of occupied MPUs, activating the model, selecting the model, switching the model, updating the model and adjusting the processing time. By adopting the technical scheme, the technical problem of how to manage a plurality of artificial intelligence models or computing units at the same time under the condition of limited computing, storage and other resources at the terminal side in the related technology is solved.

The application relates to the technical field of low-orbit satellite communication, and discloses a satellite communication-oriented traffic prediction method, device, equipment and medium, which are applied to a satellite-borne system, wherein the method comprises the following steps: receiving a self-description model package sent by a ground terminal in a standby area, and keeping an old model currently running in an active area to execute an inference task; obtaining original traffic data in a preset sliding window, determining a data correction factor of an adaptive ground model based on the difference between the original traffic data and statistical information; loading the data correction factor to the ground model to obtain a corrected new model; and in response to a model switching instruction, updating a global active model pointer from a memory address pointing to the old model to a memory address pointing to the new model by atomic operation. The technical scheme provided by the application can realize seamless hot updating of a model from the ground to the satellite without interrupting real-time inference services of the satellite-borne system.

The invention discloses an asymmetric fault equivalence method and system for short-circuit current calculation of a flexible direct network construction system, and the method comprises the steps: building a fault equivalence model of network-following and network-construction power electronic equipment through obtaining the information of a power grid, building a system-level composite sequence network based on the fault type, and carrying out the fault equivalence calculation of the short-circuit current of the flexible direct network construction system. And constructing a system-level asymmetric fault equivalent model of the new energy base sending end power grid under the flexible direct construction network, and constructing a fault composite sequence network of the new energy base sending end power grid under the flexible direct construction network. According to the method, the problems of poor model switching index applicability and insufficient asymmetric fault modeling in the prior art are solved, and the reliability of fault analysis of the flexible direct structure network system is improved.

The invention relates to the field of model deployment, and discloses an edge AI model control method and device, edge equipment and a storage medium, and the method comprises the steps: obtaining network state parameters and hardware performance parameters of the edge equipment in real time, judging whether the network state parameters and the hardware performance parameters are matched with a currently deployed AI model or not, and triggering a model switching condition when the network state parameters and the hardware performance parameters are not matched with the currently deployed AI model. And selecting a target model matched with the current environment parameters from a model library comprising the reference model and a plurality of lightweight AI models with different compression ratios, and executing seamless switching. The problem that an existing edge deployment scheme cannot adapt to diversified device performance and dynamic network environments due to the fact that a single fixed model is adopted is solved, dynamic balance between the model performance and resource occupation is achieved, and the self-adaptive capacity and user experience of the edge device in a cross-border scene are remarkably improved.

This invention discloses a collaborative scheduling method, device, and medium for decoupling parameters and computation, applied to an AIGC inference system containing at least one computing node. The method includes: receiving a user inference request, the inference request indicating a target model; in response to determining that computing resources need to be allocated to the target model, performing a resource allocation operation on the target computing node based on a preset collaborative tidal scheduling strategy, the resource allocation operation including an atomically executed ebb and flow operation; wherein: the ebb operation includes: releasing computing resources on the target computing node that have been allocated to one or more low-priority tasks; the flow operation includes: allocating the released computing resources to the target model to process the user inference request. This invention enables high resource utilization, low-latency model switching, and intelligent resource scheduling for AIGC inference services.

The application relates to a multi-model optimal load prediction method and device based on reinforcement learning. The method can automatically select an optimal model or weight combination according to real-time characteristics and historical performance of the model based on a Markov decision mode. The method uses a reinforcement learning decision mechanism to realize dynamic selection and adaptive fusion of the prediction model, thereby overcoming the shortcomings of the prior art in terms of working condition adaptability, prediction accuracy and stability. Meanwhile, through a reward function and an error feedback mechanism, the method can realize continuous optimization of prediction performance. By using the reward function to consider risk-sensitive rewards and switching penalty terms, the method can avoid oscillation caused by frequent model switching, improve model prediction stability, and ensure the accuracy of the results.

A quality monitoring apparatus includes a processor configured to execute monitoring processing of monitoring a quality of a manufacturing process of a biopharmaceutical by using a plurality of state prediction models that predict a state of a liquid produced in the manufacturing process, which is related to the quality of the manufacturing process, using spectroscopic spectrum acquired inline from the liquid as input data, and model switching processing of switching the state prediction models during the monitoring processing based on step management information for managing at least one step included in the manufacturing process, in which the plurality of state prediction models have been trained before the monitoring processing is started, and the step management information includes at least one of step identification information, usage device information, or step quality information.

The application discloses a multi-mechanism cold supply load prediction method and system, which establishes a machine learning model based on load physical composition and containing a sub-model cooperation prediction system, and a historical data matching model that can construct a hybrid model system with self-defined parameters and adaptive matching algorithms, and automatically selects and runs an adaptive adjustment model for actual prediction. The model self-improvement mechanism is also deployed for the machine learning model and the historical data matching model, respectively, to ensure that the model has self-improvement characteristics during operation, which will ensure that the two models are continuously optimized and always use the better model, significantly improving the scene coverage rate and the ability of the model after data accumulation.

The application discloses a valve cavity air pressure control method and device and related media, the method comprising collecting a multi-dimensional time series data set, and performing data preprocessing on the multi-dimensional time series data set to obtain a feature engineering data set; the feature engineering data set is divided according to working conditions and pressure segments to obtain a two-dimensional model matrix; a model mapping table is established based on the two-dimensional model matrix to obtain a model switching configuration object; the current cavity air pressure is obtained according to the model switching configuration object, and data comparison is performed to obtain a pressure segment identification result; the pressure segment identification result is used to determine a pressure increase / decrease identifier and / or a pressure adjustment identifier, so that a sub-model of the two-dimensional model matrix is selected for reasoning to generate a control instruction. The pressure increase / decrease identifier and / or the pressure adjustment identifier are determined according to the pressure segment identification result obtained by calculation, so that the sub-model of the two-dimensional model matrix is selected for reasoning to generate the control instruction. In this way, the valve cavity air pressure can be stably and accurately controlled in the full pressure range.

The invention relates to a multi-model optimal load prediction method and device based on reinforcement learning. According to the method, an optimal model or a weight combination can be automatically selected based on a Markov decision mode according to real-time features and historical expressions of models; according to the method, a reinforcement learning decision mechanism is utilized to realize dynamic selection and adaptive fusion of a prediction model, so that the defects in the aspects of working condition self-adaption, prediction precision and stability in the prior art are overcome. Meanwhile, through a reward function and an error feedback mechanism, continuous optimization of prediction performance can be realized; the risk-sensitive reward and the switching penalty term are considered by using the reward function, so that oscillation caused by frequent model switching can be avoided, the model prediction stability is improved, and the result accuracy is ensured.

A system and method for automatically detecting the type of product being processed in an industrial machine and switching to the corresponding product model without manual operator input. The system includes a detection module configured to determine the product type based on real-time data such as visual classification, lot number association, or processing schedule. Upon identifying the product type, a model selection engine automatically activates the appropriate product model from among multiple models stored on the machine. This automation eliminates the risk of human error associated with manual switching, improves operational reliability, and ensures accurate processing for environments where multiple product types are handled on the same equipment.