42 results about "Allocation algorithm" patented technology

This invention discloses a large-scale visual-language model collaborative inference method for cloud-edge-device systems. Its features include: deploying a high-performance cloud-based LVLM model and a cloud-based archive in the cloud; deploying an edge-based LVLM model and an edge-based archive at the edge, wherein the edge LVLM model incorporates a retrieval-enhanced generation algorithm, and the edge archive and the cloud archive can dynamically interact in terms of knowledge; a task scheduler evaluates the complexity of received query tasks based on a task allocation algorithm, and allocates the query tasks to the edge for local processing or to the cloud for processing based on the complexity evaluation results. This invention can improve the inference efficiency within cloud-edge-device systems while maintaining high accuracy, effectively solving the technical problem of low inference efficiency caused by limited computing resources, intermittent communication windows, and bandwidth limitations of edge devices.

This invention relates to the fields of audio and video playback technology and network communication technology, specifically a dynamic control method and system for video playback caching. By introducing intelligent prediction models, dynamic weight allocation algorithms, and scene association mechanisms, it achieves precise adjustment of video player caching duration, download speed, and external parameters, thereby maximizing the utilization of the player's local cached content, improving bandwidth utilization, reducing CDN bandwidth costs, and minimizing playback stuttering. Furthermore, this invention provides dynamic parameter and intelligent algorithm configuration interfaces for external users, allowing them to adjust caching control strategies based on multiple different parameters and business scenarios to achieve the optimal caching strategy for their own business.

The application discloses an AI cluster perception-based dynamic power distribution system of a heterogeneous computing platform, which comprises a data acquisition module, a performance quantization module, a power consumption analysis module and a power distribution module.The data acquisition module acquires the real-time power consumption and utilization rate of each computing unit in the cluster, and forms a heterogeneous unit data set after preprocessing.The performance quantization module obtains the single-resource bottleneck coefficient of each resource type based on the heterogeneous unit data set, and calculates the comprehensive bottleneck coefficient of each unit by constructing an AI-based bottleneck perception space-time network.The power consumption analysis module identifies the critical path unit and the idle unit based on the comprehensive bottleneck coefficient by using a weighted power consumption redistribution algorithm based on critical path perception, and calculates the releasable power consumption factor of the critical path unit and the idle unit to perform optimized distribution in combination with the comprehensive bottleneck coefficient, so that the best target power consumption range of the computing platform is obtained.The power distribution module performs setting operation by calling the underlying hardware power consumption management interface based on the power consumption upper limit value set of different units, so that dynamic power distribution is realized.

According to one embodiment of the present invention, a system for detecting deep voice generated by generative artificial intelligence is provided, comprising: a memory for storing instructions; and one or more processors for executing said instructions, wherein the processor extracts a feature vector including frequency characteristics and temporal change patterns from input voice data, inputs said extracted feature vectors to an ensemble artificial intelligence model including a first AI model which is an acoustic model, a second AI model which is a language model, and a third AI model which is a pattern recognition model to calculate each individual deep voice probability, measures the signal quality of said input voice data to calculate a signal-to-noise ratio (SNR), calculates a final deep voice probability through an adaptive weight allocation algorithm that varies the weights to be applied to said AI model in real time based on said calculated SNR value, and generates a deep voice suspicion alarm when said final deep voice probability exceeds a threshold value.

This invention provides a method and apparatus for multi-task scheduling in federated learning. The method involves acquiring a target federated learning task and its current events, wherein the target federated learning task includes multiple events, which are divided into stages based on their execution time and required resources. An event priority ratio is determined based on the attribute information of the target federated learning task and the type of the current event. A cumulative priority is determined based on the event priority ratio and the time the current event entered the scheduling queue. A predefined adaptive resource allocation algorithm is used to allocate computing nodes to the current event, and the current event is executed based on the computing nodes and according to the cumulative priority. This invention ensures fairness in task execution, reduces the overall task execution time of the system, and improves resource utilization.

The application discloses a highway overweight management collaborative decision and disposal method based on multi-agent reinforcement learning. In view of technical problems in highway overweight management, such as difficulty in credit distribution of multiple departments, low strategy coordination efficiency under a partially observable environment, slow learning caused by sparse rewards and the like, a bidirectional attention credit distribution algorithm is designed, non-monotonic credit distribution is realized through a positive and negative credit flow separation mechanism, a graph attention collaborative decision algorithm is adopted, a department relationship graph is dynamically constructed to realize directional communication and reduce communication overhead, a double-layer adaptive internal reward algorithm is constructed, and internal reward symbols are bound with credit gradients to guide beneficial exploration, and the algorithm is interlocked and coupled through a joint loss function and is cooperatively optimized. The application realizes full-process automatic collaborative decision from state perception to disposal execution, and significantly improves response speed and strategy effect of overweight management.

The present application relates to the technical field of unmanned ship cluster cooperative control, and proposes a multi-unmanned ship cooperative task allocation algorithm based on deep learning optimization, comprising: step 1, constructing a multi-unmanned ship cooperative task allocation dataset according to an actual task scene to provide data support for network model training; step 2, using DQN as a core model and taking a multi-layer perception machine as an optimization network to complete training by mining the internal correlation of allocation strategies in different scenes; combining pre-training and transfer learning to enable the model to quickly adapt to a new scene and output time threshold, distance threshold and maximum depth key parameters; step 3, taking the parameters output by DQN as the input of the CBAA algorithm to guide it to complete task allocation and path planning and realize the cooperative allocation of DQN and the CBAA algorithm. The present application algorithm realizes intelligent parameter adjustment of CBAA, can automatically optimize a patrol path in a wide range of scenes, reduces manual intervention and improves the efficiency and reliability of multi-unmanned ship cooperative operation.

The application provides a multi-bit audio watermarking method based on phase distribution and high-efficiency bit mapping, and belongs to the field of audio information hiding. The application divides multiple independent phase subintervals based on the natural distribution characteristics of phases, and can realize one-to-many mapping from single feature to multi-bit watermarking by constructing unique phase features in the phase subintervals corresponding to watermarking information. In addition, the application introduces an optimization strategy, and can realize automatic phase redistribution. The application can significantly reduce the number of subspaces / feature modes required for multi-bit embedding, and reduce the design difficulty of the multi-bit watermarking algorithm. In addition, since an automatic phase redistribution algorithm based on optimization constraints is adopted, the algorithm can more effectively balance the inaudibility and robustness.

This invention provides a cross-domain intelligent target allocation method for heterogeneous aircraft swarms based on the COMA architecture, belonging to the field of multi-domain collaborative transportation technology. This method constructs a multi-dimensional feature vector containing aircraft launch costs, target assurance value of mission objectives, information reflecting the coupling relationship between resource usage and mission progress, transportation matching probabilities of aircraft and mission objective combinations, and joint transportation matching probabilities and transportation benefits before and after aircraft allocation to mission objectives. It breaks through the assumption of platform homogeneity in traditional methods, achieving precise collaboration among multi-domain heterogeneous aircraft and making the allocation scheme closer to real-world needs. Furthermore, this invention applies a counterfactual multi-agent policy gradient framework to the target allocation algorithm for multi-domain heterogeneous aircraft. In the environment of multi-domain heterogeneous aircraft collaborative transportation, the counterfactual learning mechanism can effectively solve the problem of ambiguity in credit allocation caused by capability differences between heterogeneous platforms, achieving efficient decision-making in cross-domain collaboration.

The application relates to the technical field of phosphorus content monitoring, and discloses an edge internet-of-things rural water body total phosphorus content monitoring method and system.The method comprises the following steps: collecting sensing data of rural water bodies at edge node positions; using a lightweight deep inference model to perform feature extraction and total phosphorus content evaluation on the sensing data of the rural water bodies, and constructing transmission data packets by the edge nodes; based on a cooperative protocol and a time slot allocation algorithm between the edge nodes, the transmission strategy of the transmission data packets is dynamically adjusted; the historical transmission data packets of the edge nodes are fused, the total phosphorus content in the transmission data packets is checked and adaptively corrected by using an edge monitoring correction mode, and the corrected total phosphorus content is output.The application combines adaptive correction, lightweight deep inference, transmission strategy optimization and space-time checking mechanisms, realizes real-time accurate monitoring and automatic correction of abnormal total phosphorus content, and effectively improves the reliability, stability and edge computing efficiency of the monitoring data.

The application provides a high-energy-efficiency point cloud neural network inference method and device. The method and device implement quantization of a point cloud neural network by relying on a quantization bit width allocation algorithm based on semantic importance prediction. Since the semantic importance prediction is introduced, a higher quantization bit width is reserved for semantic key points, and the quantization bit width of semantic non-key points is as low as possible. The average quantization bit width of the network can be greatly reduced under the premise that the final accuracy of the quantized network does not decrease or decreases very slightly. In addition, the accuracy and computational energy efficiency of the quantized point cloud neural network are evaluated by using a point cloud dataset. In the case that the accuracy and computational energy efficiency of the quantized point cloud neural network do not reach the expected target, the point cloud neural network is re-quantized, and iteration is repeated until the accuracy and computational energy efficiency reach the expected target. The quantized point cloud neural network has high inference energy efficiency.

This invention discloses a dynamic computing power allocation method for 5G smart chips supporting multimodal data, comprising the following steps: S1. Multimodal data acquisition and preprocessing; S2. Multimodal computing power demand prediction; S3. Initial allocation of computing power resources; S4. Dynamic computing power adjustment; S5. Energy consumption optimization and computing power constraints; S6. Computing power allocation feedback and model iteration; S7. Integrated computing and computing collaborative scheduling. This invention achieves deep collaboration between multimodal task priority, 5G communication rate, chip energy consumption, memory resources, and computing power allocation. It designs a dynamic weight allocation algorithm and cross-layer collaborative scheduling logic, considering not only the heterogeneity of multimodal data but also realizing the linkage between the 5G communication layer, data preprocessing layer, and chip computing layer.

The present application relates to the technical field of communication anti-interference, and particularly relates to a multi-system cooperative communication anti-interference method suitable for a strong electromagnetic environment, comprising the following steps: collecting multi-dimensional interference characteristic data of a communication link, obtaining communication anti-interference parameters, and synchronously obtaining a real-time running state of the communication link; intelligently allocating bandwidth resources according to the real-time running state of the communication link, combining a traffic prediction model and a dynamic weighted allocation algorithm, automatically adjusting the bandwidth proportion of each service, and synchronously generating a bandwidth allocation strategy; combining the real-time running state of the communication link and the bandwidth allocation strategy, constructing multiple redundant communication links, establishing a multi-dimensional quantitative evaluation model, judging whether the main link is interrupted by interference, determining the optimal redundant link switching after interruption, and synchronously adapting the communication anti-interference parameters and the real-time bandwidth allocation strategy. The present application improves the reliability of system communication through bandwidth allocation and link switching.

The application relates to the technical field of electric power communication, and discloses a digital converter station distributed fusion terminal time synchronization compensation method and device, which comprises the following steps: S1: according to the type of a service, priority of the service of a digital converter station is divided; S2: a preset routing and spectrum allocation algorithm is used to initialize a path and a time delay of the service after the priority is divided; S3: through adaptive setting of a to-be-forwarded waiting real-time delay in a pending node, differential compensation is carried out, time synchronization of a receiving end is realized, and a path and spectrum are reconstructed according to a minimized compensation routing spectrum; S4: whether there is a licensed spectrum slot is judged, if yes, step S5 is executed, and if not, step S2 is executed through re-routing; and S5: transmission delay of the service is recorded, and a blocking probability under a traffic load to which the service belongs is calculated. The method provided by the application guarantees time synchronization of a receiving end of a digital converter station, and has better performance in terms of transmission delay and blocking probability.

This invention discloses a dynamic task orchestration system and method for telephone lead lifecycle management, relating to the field of data processing technology. The system includes a task orchestration module, a concurrency control module, a lead management module, and a line management module: the task orchestration module dynamically obtains load weights based on load assessment parameters of each service node, and dynamically executes task migration processes based on the obtained load weights and service node loads; the concurrency control module uses an adaptive resource allocation algorithm combining proportional allocation and heap optimization to achieve distributed atomic concurrent control; the lead management module manages the entire lead lifecycle based on a lead status transition mechanism and a three-layer deduplication mechanism; the line management module performs availability checks on candidate lines before outbound calls, implements intelligent allocation of line resources based on a priority-based line scheduling algorithm, and uses a sliding window rate limiting mechanism to achieve periodic frequency control of lines.

The application discloses a kind of unmanned aerial vehicle cluster energy minimum distributed multi-target hunting interception method, in this method, the specific operation process of obtaining interception energy consumption is proposed, by the analysis and meaningful energy consumption obtained, dynamic MPT distribution can be designed by minimizing the total control cost of pursuer group. Compared with the series distribution and guidance method, the method well unifies the optimization index of two stages to avoid unnecessary energy consumption and improve the synergistic effect. In addition, a distributed allocation algorithm with allowable task set constraint is also proposed. Since the detection capability of each interceptor is usually limited in engineering applications, and only part of the target information is available for each interceptor, the method can well solve the practical problems in engineering to achieve target hunting interception at a lower cost.

The application discloses a data distribution method and device, equipment, medium and product, and relates to the field of resource distribution algorithm, and determines a user trust graph based on a communication request initiated by a request initiator to a request receiver, initial trust values of the request initiator and the request receiver, and a user library; determines a target trust value by updating the initial trust values in the user trust graph in real time based on a distribution task identifier, user registration information of the request initiator and the request receiver, a transfer request between the request initiator and the request receiver, and a third-party database; and distributes a to-be-distributed object to the request initiator and the request receiver based on the target trust value and a trust value threshold. By introducing the user trust graph on the block chain and the real-time updating mechanism, the technical scheme can ensure the reasonable distribution of a large amount of data resources and solve the problem of ensuring the effective distribution of data resources in a high-concurrency scenario.

This invention discloses a method for dynamic task allocation in a drone swarm based on a wolf-pack Nash equilibrium: Step 1: Initialize the drone swarm and task system; Step 2: Construct a Nash equilibrium model; Step 3: Design a wolf-pack-like task allocation algorithm; Step 4: Develop a game theory-based wolf-pack-like algorithm; Step 5: Develop performance metrics; Step 6: Output the drone swarm task allocation status. Advantages of this invention: 1) It utilizes the swarm intelligence optimization mechanism of the wolf-pack-like algorithm to perform a global search of the solution space, avoiding oscillations and local optima traps; 2) It innovatively introduces a Nash equilibrium model, ensuring the policy stability of the allocation scheme, meaning that any single drone unilaterally changing its strategy will not yield better returns, thus ensuring the stability and reliability of the system's decision-making; 3) It combines the stability guarantee of game theory with the global search capability of swarm intelligence, enabling the simultaneous balancing of multiple objectives such as maximizing task completion, minimizing cost, and load balancing.

The application discloses a design method of adaptive time slot scheduling based on software definition, and relates to the field of industrial internet of things. An adaptive time slot scheduling system based on software definition is designed; a superframe length is set; production data of the current time is transmitted to a resource control layer by a production device through the superframe; production data to be allocated at the next time and priorities thereof are acquired; the resource control layer predicts the number of time slots required by the production data to be allocated at the next time of the production device according to the production data of the current time by using a time slot prediction algorithm; the resource control layer generates a time slot allocation scheduling scheme according to the predicted number of time slots, and issues the time slot allocation scheduling scheme to the superframe; the superframe issues the time slot allocation scheduling scheme to the production device, and the production device performs the current round of communication according to the time slot scheduling scheme. The application solves the problems of allocating time slots depending on production experience, being unable to scientifically and reasonably allocate communication resources, and low communication efficiency in an industrial production site.

This invention discloses a method and apparatus for fine-tuning a large-scale smart health model based on gradient iterative pruning, relating to the field of model fine-tuning. The method includes: obtaining the updated weight matrix of the large-scale smart health model to be fine-tuned, performing singular value decomposition to obtain a first low-rank approximation matrix and a second low-rank approximation matrix, and iteratively updating the cumulative importance of each element. After multiple iterations, a first mask matrix and a second mask matrix are constructed, and iterative pruning is performed on the first and second low-rank approximation matrices respectively. An adaptive weight allocation algorithm with contribution calibration is used to calculate the first and second weight proportion coefficients, obtaining a calibration weight matrix. A conflict suppression regularization term, a key position constraint regularization term, and a sparse merging evaluation regularization term are introduced to construct a joint loss function. Fine-tuning is performed based on the joint loss function to obtain the fine-tuned large-scale smart health model. This invention solves the problems of element redundancy and weight distribution imbalance in low-rank matrices.

The application discloses a provable incentive mechanism and a reward distribution method for federated learning, and through the introduction of a trusted execution environment and remote attestation technology, the provable execution of the federated learning incentive mechanism is realized under the threat model that the server may not be honest, the whole process of contribution evaluation, reward distribution and model aggregation is ensured to be transparent and verifiable to the client, and the dishonest behavior of the server in tampering with the reward distribution is effectively restricted; meanwhile, through the design of a contribution-aware reward distribution algorithm based on a reverse auction game, the client reward is positively correlated with the real contribution degree, the high-quality client is effectively encouraged to continuously participate while the server budget constraint is considered; in addition, the core function is deployed in the trusted execution environment by using a modular architecture, the internal and external interaction and the code size are minimized, the security attack surface is reduced, and the stability, maintainability and scalability of the system are improved, so that a fair, trustworthy and efficient federated learning incentive ecosystem is constructed.

This invention relates to the technical field of ecological product value accounting, and in particular to an ecological product value accounting mapping system and method based on grid conservation allocation. The system includes a data source access module, a basic data management subsystem, a value accounting management subsystem, a spatialization and consistency engine, and a result output module. The system solves the problem of mismatch between multi-source heterogeneous data through online verification and consistency preprocessing. Utilizing the spatialization and consistency engine, it introduces vegetation net primary productivity weighting in material product allocation and suitability weighting in cultural service allocation, and employs a normalized conservation allocation algorithm. This invention can reveal spatial heterogeneity within a county and strictly ensures the conservation constraint between grid summary values ​​and statistical totals, realizing a closed-loop operation from raw data management to result mapping, improving the authenticity and consistency of accounting results, and solving the problems of spatial distortion, inconsistent definitions, and cumbersome operation in existing technologies.

The application discloses a kind of high-efficiency image video coding rate control methods based on deep learning, comprising: S1, current image frame and front state parameter are collected;S2, decoupling construction feature atlas is carried out to space-time characteristics and is fused, and output double-flow heterogeneous representation tensor;S3, by nonlinear mapping, lagrange multiplier dynamic adjustment coefficient is output;S4, the coefficient is used to correct reference multiplier to solve output target quantization parameter;S5, according to actual encoding error and buffer fullness, input improved NMPC algorithm, introduce hard-gate switching mechanism to deduce error trajectory, output closed-loop residual compensation bias;S6, bias is superimposed to generate modified bit budget, if buffer fullness is out of bounds, then trigger adaptive redistribution algorithm to carry out secondary correction to complete closed-loop control.The application effectively improves the perceptual rationality and system robustness of code rate allocation.

Disclosed are a blockchain sharding method and system based on a convolutional neural network-long short-term memory (CNN-LSTM) prediction model. The method includes: using a double-layer chain system architecture, namely a vehicle layer and an edge layer; and using a delegated byzantine fault tolerance (DBFT) as a consensus protocol of a network at the vehicle layer and using proof of authority (POA) as a consensus protocol at the edge layer. Load balancing of the blockchain sharding system is optimized through a CNN-LSTM prediction model. A transaction status of the entire sharding network in a next period is predicted. Hot accounts are effectively allocated through a designed account allocation algorithm. The blockchain sharding method can maintain load balancing of shards, reduce cross-shard transactions, and implement effective utilization of resources. Malicious behavior of nodes is restricted. Security of the network is ensured.

This invention discloses a method for logistics demand forecasting and flexible capacity allocation based on big data, relating to the field of big data processing. The method includes: establishing a high-quality data foundation through multi-source data collection and standardized processing; extracting core features using a hybrid model to achieve accurate demand forecasting; simultaneously integrating real-time capacity data to construct a dynamic evaluation system; generating an initial matching scheme based on the forecast results using a multi-objective optimization algorithm; and dynamically adjusting the scheme through real-time feedback; finally, evaluating the execution effect from four dimensions: timeliness, cost, efficiency, and service; and reverse-optimizing the forecasting model and allocation algorithm to form a closed-loop system of "data-driven - intelligent decision-making - dynamic control - feedback iteration". The advantages of this invention are: it uses full-link data-driven processing as its core, achieving accurate demand forecasting through multi-source data standardized processing and hybrid modeling; relying on IoT dynamic capacity management and multi-objective optimization algorithms to achieve flexible allocation; and efficiently balancing the multi-dimensional objectives of logistics operations.

The application relates to the field of low-altitude aircraft on-board power management, and provides a low-altitude aircraft task load plug-and-play power priority allocation method. The method acquires the current operation scene type of the low-altitude aircraft, collects relevant parameters of each task load connected to the low-altitude aircraft power output interface; based on an intelligent allocation algorithm model pre-trained based on historical power supply data of multiple operation scenes, outputs the dynamic power supply priority ranking and adaptive power supply control parameters corresponding to each task load; according to the output dynamic power supply priority ranking and power supply control parameters, performs dynamic power allocation, current limiting control and sequential enabling operation on all power output interfaces; collects the running state and power change data of each task load, and returns the collected running state and power change data to the intelligent allocation algorithm model, so as to realize plug-and-play stable power supply of all connected task loads. The switching efficiency and interface adaptation capability of multi-task operation are improved.

The application discloses a kind of ultra-low latency guarantee and resource scheduling method across data center network, first consider the resource allocation method of resource reservation fairness between multi-hop switches and the resource reservation mechanism that satisfies ultra-low queuing delay guarantee under the worst forwarding condition in single switch and corresponding switch scheduling mechanism are proposed.When a flow task request arrives, controller first calculates the minimum bandwidth resource required by each switch according to the traffic specification parameters of data flow, then judges whether the ultra-low latency demand of user for flow can be met under the resource reservation condition according to the switch scheduling mechanism proposed in the application, if it cannot be met, more resources are reserved in each switch according to the proportional fairness principle.This resource reservation allocation algorithm based on proportional fairness can meet the demand of ultra-low latency communication across data center network, and enable the network to bear more deterministic ultra-low latency flow.

The present application relates to the technical field of industrial automation control and real-time operating system, and particularly provides a virtual PLC user state deterministic memory management method and system, which comprises the following steps: a global large page memory pool is established by an operating system kernel layer through a swap-out prohibition mechanism to form metadata description required by a deterministic allocation algorithm; metadata of a user state deterministic memory management system is captured by a symbol interception dynamic library to form a private heap corresponding to each virtual programmable logic controller running instance, and the private heap is mapped from the global large page memory pool; the private heap is called by a memory calling interface to complete allocation and form an independent memory area under the control of a control logic of the virtual programmable logic controller running instance; and the private heap of each virtual programmable logic controller instance is continuously managed by a deterministic allocation algorithm and intercepted by a runtime interception of the symbol interception dynamic library to form a user state deterministic memory application and release closed loop. The present application forms a complete memory management closed loop under the user state.

The invention discloses a supply chain resource elastic reconstruction and self-execution intelligent contract system based on an industrial chain, and relates to the technical field of supply chain resource management. The system comprises a resource state sensing module which is used for collecting resource data of each node of a supply chain; the elastic resource allocation module is used for monitoring resource data in the resource dynamic data pool, starting a resource allocation algorithm when the resource data meets a preset triggering condition, and generating a resource allocation scheme based on the resource data; the self-execution smart contract module is used for converting the resource allocation scheme into a smart contract, deploying the smart contract and automatically executing the smart contract; and the feedback optimization module is used for collecting execution effect data of the smart contract and performing optimization adjustment on the resource allocation algorithm and the smart contract based on the execution effect data. According to the invention, resource utilization efficiency optimization and system adaptive iteration are realized, and the supply chain operation uncertainty is reduced.