1000results about How to "Reduce latency" patented technology

The invention discloses a cloud edge cooperative control method, which comprises the following steps: establishing a cloud edge cooperative architecture, collecting data by edge nodes, and selecting a corresponding data compression strategy to compress the data; generating an edge-cloud collaborative uploading data stream; the cooperative control layer executes adaptive load balancing based on a cloud-side node state, task characteristics and a network condition which are monitored in real time so as to realize task allocation and scheduling between the cloud and the side; the cloud receives the data stream transmitted by the edge node, decompresses and processes the compressed data according to the metadata information, completes the task and feeds back an execution result; a node performance prediction model is constructed by using system state monitoring information, a task allocation strategy and a resource allocation and data compression strategy are adjusted in real time according to a prediction result, and cloud-side collaborative optimization is realized. Efficient data transmission, flexible task scheduling and self-adaptive resource management are realized, and the performance of the system in a complex and changeable environment is comprehensively improved.

The invention relates to a traffic flow prediction optimization method based on an intelligent optimization algorithm, which is applied to the technical field of traffic flow prediction. The system comprises a multi-modal data acquisition and fusion module, a hybrid deep learning prediction module, an intelligent parameter optimization module, an integrated learning module, a multi-scale error correction module, a prediction result analysis and visualization module, a distributed learning and privacy protection module and the like. Wherein the hybrid deep learning prediction module adopts an attention mechanism to fuse models such as CNN, Transform and LSTM, and introduces an adaptive error correction mechanism, so that the prediction precision is remarkably improved; the intelligent parameter optimization module accelerates the model search efficiency by using technologies such as quantum computing and swarm intelligence optimization; the integrated learning module and the adversarial training strategy enhance the robustness and generalization of the model; and the multi-scale error correction module adopts a time sequence decomposition and self-adaptive online learning method to improve the accuracy of long-term and short-term prediction.

The invention discloses a tensor core-based method for accelerating arbitrary precision sparse matrix multiplication and addition operation, which comprises the following steps of: aiming at an architecture of a tensor core on A100 and a programming model thereof, combining program characteristics of deep learning parallel processing sparse matrix, combining an arbitrary precision sparse matrix multiplication and addition operation operator with parallel computing acceleration equipment, and performing parallel computing acceleration on the sparse matrix. The method comprises the following steps: decomposing a general matrix into a sparse matrix in a set format, partitioning the sparse matrix and assigning a corresponding parallel computing hardware unit for the sparse matrix, reducing memory conflicts and access delay by using a shared memory, calling a high-level language API function library to execute loading, computing and storing processes of a computing data volume, and finally performing specification on output data of a function. Therefore, the final result of the random-precision sparse matrix multiplication and addition operation on the parallel computing acceleration equipment is obtained. According to the method, parallel computing resources such as the tensor core on the A100 can be fully utilized, the program calculation amount and the communication overhead are effectively reduced, and the program operation performance is greatly improved.

The invention relates to a task planning and scheduling method which is quicker, lighter and finer in task splitting based on a large model, and aims to improve the capability of processing complex tasks by calling a tool through the large model. According to the method, on the basis of a FastAPI framework, a Text-to-SQL technology, a local knowledge base, Function Calling and other core technologies are integrated, user intention understanding is completed through a large model, dynamic calling and fusion of an API interface, the knowledge base and a database are achieved on the basis of intention understanding, and the task planning, tool calling and decision support capacity of the large model in a complex scene is remarkably enhanced. By introducing a thinking chain technology of multi-level deep task decomposition, the system can decompose a complex task into a plurality of sub-tasks, an execution path is optimized, and the accuracy and execution efficiency of a multi-step reasoning task are improved. A user query tool calling module is researched and developed, user requirements can be accurately recognized and decomposed, proper tools are automatically selected and called for task processing, and efficient solving of complex problems and optimal configuration of resources are achieved.

The invention relates to the technical field of cold-chain logistics information, and discloses a multi-mode AIGC cold-chain logistics information processing method and system, and the method comprises the steps: carrying out the feature extraction and feature enhancement of video data, temperature data and position data in a cold-chain logistics system based on a logistics information AIGC model, and obtaining an enhanced feature matrix; carrying out digital twinborn modeling to obtain a digital twinborn model of the logistics system, and constructing a state parameter space; performing deep reinforcement learning and information delay index calculation to generate a real-time data synchronization scheme; constructing a distribution network graph structure, and performing depth map network coding on the distribution network graph structure to obtain a distribution network representation vector; and performing multi-agent collaborative decision-making on the distribution network representation vector to obtain a cold chain distribution scheduling scheme, and updating the cold chain distribution scheduling scheme to the logistics system digital twinborn model. Therefore, intelligent scheduling and accurate control of the cold-chain logistics system in a complex scene are realized.

The invention relates to a construction method of a Ti-SA model for pilot situational awareness assessment, belongs to the technical field of aviation safety and situational awareness, and solves the technical problems that an existing perception model cannot assess the situational awareness ability of a pilot in real time, and selected data cannot adapt to a complex environment, so that the effectiveness is poor. Comprising the steps that S1, eye movement trace time sequence data and AOI fixation time sequence data in the simulated flight process of multiple pilots are acquired to serve as sample data, and the sample data and generated corresponding category labels form a training data set; wherein the category label is a level label of the scene awareness of the pilot; step S2, constructing a Ti-SA model used for pilot scene awareness assessment; and training the Ti-SA model by using the training data set until the trained Ti-SA model is obtained when the loss function converges. The method achieves the real-time evaluation of the situational awareness capability of the pilot through the Ti-SA model.

The invention discloses a mine tunnel operation period deformation monitoring method and system, and relates to the technical field of mine tunnel monitoring, and the method comprises the following steps: carrying out the real-time monitoring of the deformation condition in a mine tunnel through a sensor device, and transmitting the collected monitoring data to a data processing center through a wireless transmission module. Real-time monitoring of mine tunnel deformation is achieved through the multiple sensors and the wireless transmission technology, the problem of data lag is solved, and monitoring accuracy and timeliness are improved. And in combination with a multi-objective optimization algorithm, the system automatically adjusts key parameters such as a supporting structure and a mining sequence according to real-time data, so that an optimal safety control scheme is ensured. According to the system, the operation efficiency and safety are improved, manual misjudgment is reduced, and the control strategy can be continuously optimized according to the operation stage.

The invention discloses a large model parallel task scheduling method based on space-time two-dimensional segmentation and intelligent sharing, and particularly provides a model segmentation strategy based on space-time two dimensions, and based on significant differences of training and reasoning tasks in calculation properties, targeted optimization strategies are adopted respectively, so that the task execution efficiency is improved; the invention further provides a model placement strategy based on the heterogeneous resources and the interference model, the defects that a traditional method lacks intelligent sharing judgment and cannot be dynamically adjusted are overcome, the flexibility and adaptability of scheduling are improved, and therefore the maximum benefit of the shared resources is guaranteed. The method solves the problems that when an existing heterogeneous intelligent computing cluster scheduling method is used for processing large model training and reasoning tasks, the model placement optimization technology of model segmentation and heterogeneous resource adaptation is not fully utilized, intelligent sharing scheduling of computing resources in a cluster is lacked, system task scheduling is unreasonable, and the scheduling efficiency is high. And the utilization rate of computing resources is low.

The invention relates to the technical field of operating systems, in particular to a process scheduling method, system and equipment based on a heterogeneous multi-core processor, and the method comprises the following steps: monitoring and recording an operation behavior of a user in real time, and generating a log file of the user behavior; analyzing the log file of the user behavior by using a machine learning algorithm, and predicting the next operation behavior of the user; dynamically loading the application predicted to be used into a memory based on the prediction result; determining the optimal processor core of the task by comprehensively considering the execution time, energy consumption, dependency relationship and processor resource utilization rate of the task according to the prediction result in combination with the state of each processor core in the system and the current process scheduling strategy; and allocating different tasks to the corresponding optimal processor cores for execution. The overall performance of the system is improved, energy consumption is reduced, and optimization of energy efficiency is achieved.

The invention discloses a DBA management method and device in an FTTR scene and an optical line terminal. The method is applied to a PON MAC chip, wherein the PON MAC chip comprises a CPU unit and a logic unit. The method comprises the following steps that: a CPU (Central Processing Unit) determines the uplink service bandwidth of each AllocID in each sub-period in the current DBA period; the CPU unit takes at least part of the sub-periods as sub-distribution periods, and determines BWMap data of the first sub-distribution period and the second sub-distribution period; the CPU unit sends the length of the silent window, the BWMap data of the first sub-allocation period, the BWMap data of the second sub-allocation period and the total length of the current DBA period to the logic unit; the logic unit determines the number of the remaining sub-allocation periods in the current DBA, and takes the BWMap data of the second sub-allocation period as the BWMap data of each remaining sub-allocation period to obtain complete BWMap data in the current DBA period; and the logic unit sends the BWMap data of the current DBA period, so that the problem that the DBA cannot adapt to the use requirement of the FTTR scene is solved, and the low-delay requirement of the DBA is met.

The invention belongs to the technical field of distributed data access, discloses an RDMA (Remote Direct Memory Access) distributed data transmission method supporting high-concurrency message transmission, and aims to improve system performance in a high-concurrency message transmission environment. The method comprises the following core steps: 1) peer-to-peer server design: each device is an independent server, and connection is established by a connection channel monitored at any time; 2) establishing a connection resource pool in advance; 3) creating and registering an RPC message; 4) transmitting the RDMA to a memory mailbox pool of a target server; 5) the client and the server perform high-concurrency processing; 6) dynamically polling the server; (7) zero-copy transmission of data is achieved through the RDMA technology. According to the invention, the dynamic expansion capability of the architecture is improved by adopting a connection resource decoupling mode, QP jitter can be reduced and the stability of the system can be enhanced by registering a memory of a continuous large page at one time, and the message transmission efficiency under the distributed network architecture is effectively improved by adopting a combined sending strategy.

The invention relates to an Internet of Vehicles task unloading method based on federated learning, and belongs to the technical field of information. Aiming at the problem of how to make an optimal unloading decision in a dynamic environment of the Internet of Vehicles, a framework based on an edge intelligent system is constructed, a communication and calculation model between the Internet of Vehicles and an edge server is established, a calculation unloading strategy based on federated learning is designed, the Internet of Vehicles and the federated learning are fused, and the optimal unloading decision is made. The processing time delay of the vehicle calculation task is minimized, the calculation task can be unloaded to an edge server to be processed based on the Internet of Vehicles, simulation experiments prove that the method reduces the time delay to a certain extent, and the vehicle task unloading efficiency is improved.

The invention belongs to the technical field of artificial intelligence information security, particularly discloses a neural network privacy reasoning method and device based on winograd sparse convolution and differentiable activation function approximation, and aims at solving the problems that existing neural network privacy reasoning is high in calculation time delay and large in communication cost. And in combination with a multi-granularity weight pruning algorithm, the sparsity of a linear convolutional layer is maximized, and the communication cost required by the convolutional layer with dense multiplication is reduced. Meanwhile, collaborative optimization is carried out on a nonlinear activation function, a new activation function sensitivity and importance quantitative evaluation algorithm is designed, and an adaptive learnable polynomial function is introduced, so that the fitting precision of the activation function is improved, and the polynomial replacement loss is reduced; and the powerful prediction performance of the neural network model is reserved by using a multi-stage knowledge distillation method, and the calculation and communication efficiency of privacy protection neural network reasoning is effectively improved on the premise of not influencing the prediction precision.

The invention discloses a heterogeneous computing power network high-reliability resource scheduling optimization method based on coding distributed computing and a hypergraph neural network, which comprises the following steps of: firstly, constructing a multi-level resource representation model, and capturing high-order correlation among heterogeneous nodes by using the hypergraph neural network so as to optimize the precision and efficiency of dynamic resource scheduling; secondly, a coding distributed computing strategy is introduced, edge node failure is dealt with through redundant load distribution, the stability of the system is enhanced, in order to further improve scheduling optimization, a hybrid algorithm framework combining a hypergraph neural network and particle swarm optimization is provided, the framework captures high-order relevance of heterogeneous nodes in the network through hypergraph modeling, and the scheduling optimization is improved. And meanwhile, the particle swarm optimization algorithm is combined, so that the discreteness and discontinuity problems can be effectively solved. Finally, the hybrid algorithm improves the operation efficiency while guaranteeing the solving quality, and achieves efficient resource scheduling optimization. The wide application potential in the future 6G application scene is shown.

The invention discloses a multi-UAV (unmanned aerial vehicle) task unloading and caching optimization method and system based on edge calculation. The method comprises the following steps: establishing a system model composed of a T-UAV layer, a B-UAV layer and a ground user layer; and obtaining an optimization problem by taking the minimum total cost of the system model as a target under the condition that a preset constraint condition is met, modeling as a multi-agent Markov decision process, and solving to obtain an optimal task unloading and caching joint strategy. In the invention, in order to reduce the calculation load of an edge server and improve the system response capability, an unmanned aerial vehicle for storing a calculation result and an unmanned aerial vehicle for assisting edge calculation are divided; when the unmanned aerial vehicle encounters repetitive tasks, task results are selectively stored through a caching mechanism, similar calculation tasks can share calculation results submitted by a previous task, and a joint decision of unloading and caching is proposed to effectively reduce time delay and energy consumption of task unloading and processing. And the efficiency and accuracy of task unloading and caching of the multiple unmanned aerial vehicles are further improved.

The invention discloses a road traffic accident monitoring and alarming method and system, and relates to the technical field of data processing. The method comprises the following steps: acquiring vehicle speed values, vehicle distance values and driving tracks of a plurality of vehicles on a road to be monitored; calculating a speed change rate according to the vehicle speed value of each vehicle, calculating a distance change rate according to the vehicle distance value of each vehicle, and calculating a vehicle transverse offset according to each driving track; when the speed change rate, the distance change rate or the vehicle lateral offset exceeds the corresponding standard threshold value, determining that a traffic accident occurs on the road to be monitored; receiving the lane occupancy rate, the traffic flow density and the average passing speed of the road to be monitored; and determining an accident grade corresponding to the traffic accident based on the lane occupancy rate, the traffic flow density and the average passing speed, and generating alarm information of the accident grade corresponding to the traffic accident. By implementing the technical scheme provided by the invention, the effect of improving the road traffic accident monitoring and alarming efficiency is achieved.

The invention relates to an edge computing migration system for an intelligent factory digital twin system, and the system comprises a scene obtaining module which is used for obtaining the real-time data of automation equipment in a factory; wherein the real-time data comprises a running state and task information; the digital twin module is used for virtually mapping the real-time data of the automation equipment and synchronizing the state between a factory physical system and a virtual system; the model construction module is used for constructing an edge computing network time delay and energy consumption model according to the real-time data of the automation equipment; and the optimization algorithm module is used for carrying out real-time optimization on allocation of calculation tasks and resources by adopting a multi-agent deep reinforcement learning algorithm based on the edge calculation network time delay and energy consumption model. The system provided by the invention has the characteristics of low time delay and high energy efficiency, is suitable for highly dynamic industrial scenes, can optimize the scheduling of computing resources, and improves the overall operation efficiency of factories.

The invention provides an inverse knowledge distillation method and system based on a federal large model, and relates to the field of federal learning, and the method comprises the steps: a client uses a private data set to update the weight parameters of a small parameter model, and obtains a client model; outputting client model soft logic by using the public data set and sending the client model soft logic to a server; the large-parameter model of the server uses the public data set to output server model soft logic; calculating the weight of each client according to the correlation between the soft logic of the server model and the soft logic of the plurality of client models and the accuracy of each client model to the public data set; calculating weighted client soft logic according to the client weight and the client model soft logic; knowledge distillation is performed using server model soft logic and weighted client soft logic. According to the method, knowledge distillation is carried out on a single large-parameter model by using a plurality of small-parameter models, and a brand-new method is provided for updating and iteration of the large-parameter models in a server.

The invention discloses a dual-mode barrier gate control system and method based on LoRa and Wi-Fi. The method comprises the following steps: S1, system initialization; s2, dynamically selecting a communication mode by adopting a Markov decision process algorithm; s3, analyzing the historical behavior data of the visitor by combining a hypergeometric learning algorithm and a manifold learning algorithm, and generating a dynamic two-dimensional code; s4, encrypting the dynamic two-dimensional code, and transmitting the dynamic two-dimensional code to the visitor terminal through LoRa or Wi-Fi; s5, the visitor scans the two-dimensional code to request for passing, and the edge computing device locally verifies the validity of the two-dimensional code and device binding information; s6, when an abnormal behavior is detected, the system triggers a multi-factor authentication mechanism; s7, after the authentication is passed, the system controls a barrier gate execution mechanism to open a gate, and traffic information is recorded and uploaded to the cloud; and S8, the system collects visitor data for a long time, generates a behavior model by using big data, and optimizes an approval process and two-dimensional code management. According to the invention, the dynamic two-dimensional code is generated by combining LoRa and Wi-Fi dual-mode communication and an intelligent algorithm, so that the dynamic management of visitors is realized.

The invention is suitable for the technical field of redundancy coding, and particularly relates to a data redundancy coding method and system based on deterministic network cooperative transmission, and the method comprises the steps: delimiting an industrial production area, and positioning a relay node which at least comprises edge equipment and a network switch, key indexes of the relay nodes are collected, and the relay nodes are divided into main nodes and common nodes; the industrial production tasks are collected, the master control server and the execution end are extracted, the control command sent by the master control server is received, and the industrial production tasks are coded according to a preset coding sequence. By constructing the hash chain, the control command can be decoded, the traceability of the control command is improved, meanwhile, the fault tolerance of the industrial automation system is further enhanced, it is guaranteed that the control command can be accurately and timely transmitted to the execution end, the response efficiency of the execution end is greatly improved, and normal operation of industrial production is guaranteed.

The invention relates to the field of artificial intelligence, and particularly discloses a time delay and energy consumption optimization method based on federated learning, and the method comprises the steps: enabling a terminal layer, an edge layer and a cloud layer to work cooperatively: the terminal layer is responsible for collecting data and extracting features through a convolutional neural network; the edge layer receives the features and trains an edge model, and then uploads parameters to the cloud layer; and the cloud layer aggregates the parameters to form a global model and sends the global model back to the edge layer for the next round of training. In addition, a global time delay and energy consumption model is also constructed, and optimization is carried out through a particle swarm optimization algorithm, so that the time delay and energy consumption in the training process are further reduced, and the efficiency and performance of model training are improved. According to the method, resource utilization is optimized, and a more efficient model training process is realized by optimizing time delay and energy consumption of the terminal layer, the edge layer and the cloud layer.

The invention provides an efficient AI model management and safe loading system and method oriented to a Serverless framework, and aims to solve the high concurrency performance bottleneck and potential safety hazards of AI model loading in a Serverless environment. According to the system, an AI model management module is introduced, efficient semantic cutting is carried out on an uploaded AI model, and fine-grained encryption model blocks are generated; a high-performance P2P network is constructed by utilizing a Serverless cluster, and distributed storage and fast routing of model blocks are realized; a function instance loads a model block in a high-concurrency mode through a P2P network, a multi-stage intelligent cache is constructed locally, and dynamic caching and prefetching are achieved. According to the method, the model loading parallelism degree is improved by adopting a heuristic cutting algorithm, lightweight model updating is realized by utilizing incremental updating transactions, and the loading overhead is minimized. According to the method, the AI model loading performance and efficiency in the Serverless scene are effectively improved, meanwhile, the flexibility and safety of model management are considered, and the method has remarkable technical innovation and practical value.

The invention provides a cloud edge collaborative anomaly detection method, system and equipment for a virtual power plant, and is applied to the technical field of data mining. The method comprises the following steps: performing anomaly detection on real-time operation data of each distributed resource in a virtual power plant based on a global anomaly detection model obtained from a cloud server; sending the detection result to monitoring equipment associated with each distributed resource corresponding to the edge end equipment; the global anomaly detection model is obtained by the cloud server through fusion processing according to local anomaly detection models trained by the edge end devices respectively. The problem that abnormal data generated in the data transmission process of the virtual power plant system affects the accuracy and reliability of energy management and decision making is solved.

The invention relates to the technical field of serial deserializer design, and discloses an FPGA-based direct-current balanced digital audio and video signal serial deserializer, which comprises a serializer, a deserializer, a control configuration module, a power management module, a configuration storage module and a time sequence control module, the serializer is used for converting parallel digital audio and video data into high-speed serial data streams; the deserializer is used for detecting and repairing abnormal data in the high-speed serial data stream by using a machine learning technology, and converting the repaired high-speed serial data stream back to parallel digital audio and video data; the control configuration module is used for configuring working parameters of the serializer and the deserializer through an external interface; the time sequence control module is used for managing the time sequence relation in the system and ensuring synchronization of data flow between different modules. According to the method, the defects of a traditional deserializer in the aspects of anomaly detection and repair, signal quality evaluation and parameter optimization are overcome, and the stability and quality of high-resolution digital audio and video data transmission are improved.

The invention belongs to the technical field of wireless communication, and discloses an optimization method based on active reasoning in an MEC-enabled low-altitude element cosmic system, which realizes low energy consumption and low delay of Internet of Things equipment on the premise of ensuring user service quality through joint optimization of a flight path and a resource allocation strategy of an unmanned aerial vehicle, and improves user experience. And meanwhile, an auxiliary means is provided for improving the economic benefits of operators. Aiming at the dynamic nature of the network, the uncertainty of information acquisition and the incompleteness of data use, an optimization problem is converted into a Markov decision process, an improved deep learning algorithm based on active reasoning is adopted, factors except the reward dimension are additionally considered on the guidance of dynamic learning, and the dynamic learning algorithm is optimized. And a personalized strategy is explored according to the preference of the intelligent agent.

The invention provides an edge processing method and device, equipment and a storage medium, relates to the technical field of communication, and is used for realizing edge credible verification of an artificial intelligence model so as to safely realize a service information processing process in a low-delay manner, avoid a service data leakage risk and effectively protect an artificial intelligence model application program. The method is applied to business side equipment, and comprises the following steps: sending a service request to edge computing equipment; the service request is used for requesting an information processing service of an artificial intelligence model on the edge computing device; receiving remote attestation information from the edge computing device; the remote certification information is certification information of a trusted execution environment instance which is registered in a remote certification server in advance and is deployed with an artificial intelligence model; sending a first attestation request to a remote attestation server; and sending an information processing request carrying the to-be-processed information to the edge computing equipment under the condition that a first passing response from the remote attestation server is received.

The invention discloses a large-depth pressure-resistant communication antenna of an unmanned underwater vehicle and the underwater vehicle, and belongs to the technical field of underwater vehicles, the antenna comprises an antenna shell, a pressure-resistant cabin body and a fin-shaped component, the pressure-resistant cabin body is a component forming a cabin body segment of the underwater vehicle, and the fin-shaped component is a fin-shaped component. The antenna shell comprises a radial supporting part fixed to the pressure-resistant cabin and a shell part fixed to the top end of the radial supporting part, the fin-shaped component is fixed to the radial supporting part, an antenna unit is installed in the shell part, and a wireless communication function unit connected with the antenna unit is arranged in the pressure-resistant cabin. The large-depth pressure-resistant communication antenna of the unmanned underwater vehicle has the advantages of being excellent in communication quality, high in pressure resistance, optimized in size, high in space utilization rate, stable and reliable for a long time and the like, and can meet operation requirements in a deep sea environment.

The invention relates to the technical field of tunnel engineering monitoring, in particular to a tunnel health monitoring method and system based on edge computing, and the method comprises the steps: a data collection module obtains the original state data of a tunnel through a sensor deployment network, carries out the primary processing of the original state data based on edge computing, and obtains the actual state data; during the period, the data acquisition module performs data filtering on the original state data based on a state data threshold value, rejects invalid data in the original state data and marks abnormal data in the valid data according to an anomaly detection algorithm, and the remaining unmarked original state data is the actual state data of the tunnel; and the monitoring cloud platform obtains the actual state data every a preset period, and carries out data summarization, data storage and data analysis processing on all the actual state data. The system has the beneficial effects that part of calculation tasks are issued to the data acquisition module, so that the processing burden of the monitoring cloud platform is reduced, and the real-time performance and effectiveness of monitoring and early warning are improved.

The invention discloses a body intelligent execution and training method and device based on an edge-end cooperation large model, and the method comprises the steps: transmitting a voice instruction to a robot body, converting the voice instruction into a text instruction through the robot body, and transmitting the text instruction, machine vision feedback and a visual image to an edge side unit; a multi-modal large model is arranged in the edge side unit; reasoning the text instruction, the machine vision feedback and the vision image by the multi-modal large model, obtaining the intention of the text instruction after reasoning, generating a task decomposition instruction according to the intention, and sending the task decomposition instruction to the robot body; the robot body calls a chassis moving API and an atomic skill library API according to the task decomposition instruction to complete a target task; and after the target task is completed, the robot body feeds back process algorithm data and an execution effect to the edge side unit, and the edge side unit carries out post-training upgrading according to the feedback. According to the method, the large model reasoning time delay and cost can be effectively reduced, the load of the cloud server is reduced, and the data security and privacy are improved.

The invention discloses a dependent task computing unloading method meeting random preference of a user, and belongs to the technical field of computing unloading of a mobile communication technology. In order to solve the problem that in the prior art, when a satellite depends on task calculation unloading, the requirement for random change of a user optimization target is often ignored, the application of a ground mobile terminal is divided into a workflow composed of a plurality of tasks, and the workflow is represented by a DAG; determining a scheduling sequence of the dependent tasks according to the DAG; constructing a random preference dependent task calculation unloading optimization model and function; converting a calculation unloading problem into an MDP decision process according to the created model, and solving an unloading decision by adopting multi-target reinforcement learning; and comprehensively considering the dynamic preference demand of the optimization target, and designing a composite reward function to obtain an optimal unloading decision. According to the method, the optimal unloading decision model meeting different preference requirements is provided, the solving efficiency and flexibility are improved, and the completion time and energy consumption of the dependent task are reduced.