216 results about "Learning dynamics" patented technology

The invention belongs to the technical field of power equipment, and discloses a power equipment fault early warning method based on multi-source data fusion, and the method comprises the steps: constructing multi-dimensional feature association through multi-modal data time-space association collection and hierarchical fusion driven by a knowledge graph; a space-time weight matrix is used for correcting sampling deviation, fault mechanism knowledge is combined to strengthen key feature contribution degree, false alarm and missing alarm caused by data isolation are effectively avoided, early recognition of hidden defects of equipment is realized, and global perception capability of early warning is improved. A meta-learning enhanced cross-equipment early warning model and reinforcement learning dynamic threshold decision are adopted, cross-equipment rapid adaptation under a small number of samples is realized through a ''meta-micro'' double-circulation mechanism, and a nonlinear law of fault evolution can be accurately described by combining a three-dimensional dynamic threshold matrix to balance an equipment state, an environment and an operation and maintenance strategy. The model generalization problem of different types of equipment in a complex environment is solved, and the adaptability to scenes such as load fluctuation and environment sudden change is improved.

The invention relates to the technical field of organic display data processing, in particular to an organic display interface multichannel signal modulation method, which comprises the following steps of: generating a standardized data set through gamma correction and gamut mapping, and inputting the standardized data set into time sequence, amplitude and frequency sub-models deployed in a federal learning framework; and respectively analyzing the pulse width-refresh rate relevance, the driving voltage-brightness nonlinear relationship and the conduction period-mobility dynamic response, generating an optimized weight coefficient, and fusing the optimized weight coefficient into a boundary constraint condition of a cross-sub-model. A migration rate parameter index, amplitude gradient calculation and duty ratio correction service of chain calling is constructed based on a service grid, a reinforcement learning dynamic routing strategy is combined to avoid resource competition nodes, and brightness error characteristics are periodically fed back to reversely optimize gradient updating. According to the method, the conflict between time sequence deviation accumulation and local optimum is effectively suppressed, the color gradation continuity, the dynamic range and the edge sharpness are improved, and efficient cooperative modulation of multi-channel signals is realized.

The invention discloses an intelligent engineering vehicle scheduling method based on artificial intelligence, and the method comprises the steps: constructing a scheduling objective function which at least comprises the engineering vehicle transportation cost, the task completion time cost, the path congestion cost and the energy consumption cost; establishing a multi-dimensional task demand prediction model based on historical task data and real-time traffic data, and outputting a task distribution prediction value and a traffic state prediction value in a future time period; inputting the model into an engineering vehicle dynamic scheduling model established based on a deep reinforcement learning algorithm for iterative optimization; and generating a real-time scheduling instruction based on the optimized dynamic scheduling model, dynamically allocating task paths and resources of the engineering vehicle, and monitoring an execution state in real time to adjust a scheduling strategy. According to the method, efficient, economical and environment-friendly intelligent scheduling of the engineering vehicle is realized through construction of a scheduling objective function, multi-dimensional demand prediction, deep reinforcement learning dynamic optimization and real-time scheduling and monitoring.

Systems and methods for generating simulation data based on real-world environments are provided. A method includes obtaining multi-modal sensor data indicative of a dynamic object within an environment of a robotic platform. The multi-modal sensor data is associated with a plurality of timesteps including a first timestep and a second timestep. The method includes providing the multi-modal sensor data indicative of the dynamic object within the environment as an input to a machine-learned dynamic object removal model. And, the method includes receiving as an output of the machine-learned dynamic object removal model, in response to receipt of the multi-modal sensor data, a scene representation indicative of at least a portion of the environment including a reconstructed region based at least in part on removal of the dynamic object and multiple levels of granularity. The scene representation is used as a template for generating different simulations within the depicted environment.

The invention discloses an intelligent optimization method for a multi-type well seam joint control fine injection-production mode, and relates to the technical field of oil-gas field development. The method comprises the following steps: setting a well seam joint control fine injection-production mode, establishing an oil reservoir numerical simulation model in oil reservoir numerical simulation software, obtaining multiple groups of oil reservoir injection-production schemes based on a Latin hypercube sampling method, performing simulation according to each group of oil reservoir injection-production schemes by utilizing the oil reservoir numerical simulation model, generating multiple pieces of sample data, and establishing a sample database; a deep learning agent model is established, after the sample database is utilized to train and train the deep learning agent model, a particle swarm optimization algorithm is adopted to carry out single-target pre-search global optimization to obtain a preferred reference strategy, a reinforcement learning dynamic decision model is established, and a reinforcement learning agent is obtained through training based on a PPO near-end strategy optimization algorithm; and the optimal injection-production development scheme of the oil reservoir is obtained by utilizing the reinforcement learning agent, so that rapid optimization and decision support of the oil reservoir injection-production scheme in a new multi-type well seam joint control mode are realized.

The invention relates to the technical field of artificial intelligence, can be applied to business scenes such as personal intelligence, financial science and technology, medical health and the like, and discloses a task execution strategy generation and adjustment method, device, equipment and medium. Encoding the visual information, the audio information and the language instruction information to obtain a visual feature, an audio feature and a language feature, fusing the visual feature, the audio feature and the language feature to generate a comprehensive feature, generating an initial task execution strategy based on the comprehensive feature and executing a corresponding action, and in the execution process, according to real-time feedback information of the environment, executing the corresponding action according to the initial task execution strategy. And dynamically adjusting the initial task execution strategy by adopting a reinforcement learning model to obtain an updated task execution strategy. Through multi-modal information fusion and reinforcement learning dynamic adjustment, optimization and flexible updating of a task execution strategy in a complex environment are realized, and the autonomous decision-making capability of the intelligent equipment is improved.

The invention discloses a deep geothermal target area intelligent positioning method based on a machine learning algorithm, and relates to the technical field of geothermal resource exploration. Comprising the following steps: S1, multi-source heterogeneous data fusion acquisition and preprocessing; s2, geological feature entropy quantification and spatial autocorrelation analysis are carried out; s3, deep feature extraction and multi-modal feature fusion are carried out; s4, performing multi-model adaptive integration and dynamic weight optimization; s5, reinforcement learning dynamic adjustment participation abnormal threshold determination; and S6, performing intelligent positioning and risk assessment on the three-dimensional geothermal target area. According to the positioning technology, geological structure, geophysical field, geochemistry and remote sensing data are integrated through a machine learning algorithm, high-dimensional feature vectors are constructed, the limitation that the prior art depends on physical detection singly is solved, reinforcement learning is introduced to dynamically optimize model parameters, real-time geothermal well data updating is combined, and the positioning accuracy is improved. The model can adapt to geological condition changes, and the prediction precision is greatly improved compared with a traditional method.

The invention provides a DL-MVCNN-based wind power subsynchronous oscillation traceability method and system, and relates to the technical field of wind power grid connection and oscillation traceability, and the method comprises the steps: constructing a multi-wind power plant grid connection model, and obtaining instantaneous current and power high-dimension subsynchronous oscillation data; modeling is carried out on instantaneous current and power subsynchronous oscillation intrinsic dynamics based on sampling determination learning, subsynchronous oscillation dynamic characteristics are extracted in a given state space, and a subsynchronous oscillation dynamic diagram is generated; inputting the subsynchronous oscillation kinetic graph into a multi-modal oscillation traceability model, extracting respective kinetic mode characteristics of each wind field, and fusing the kinetic mode characteristics of each wind field through a channel attention mechanism in the multi-modal oscillation traceability model to obtain fused characteristics; different oscillation sources correspond to different determined learning dynamics characteristics, based on fusion characteristics, adjusting a weight giving channel, calculating a corresponding channel weight, and outputting a classification result of wind field traceability.

The invention discloses a dynamic time sequence rapid analysis method based on a graph neural network. The method comprises the following steps: modeling a topological structure and working load characteristics of a target circuit; constructing a pre-training graph representation learning model based on GAE, capturing static topological information of the circuit in an unsupervised mode, and learning a dynamic input state and pin overturning information in a supervised mode; extracting circuit structure characteristics by utilizing the trained GNN, and performing representation learning on an input node state in combination with dynamic working load information to generate a uniform circuit embedding vector; constructing a dynamic time sequence prediction model based on GAT, and predicting dynamic signal arrival time under different loads through regression learning in combination with pre-trained graph embedding information; and S5, verifying the trained model on a test data set, and carrying out dynamic time sequence prediction by utilizing an optimized model framework, so that compared with a traditional method, the prediction precision is improved, and calculation acceleration is realized. The method is suitable for the field of chip design and optimization, and can be used for dynamic time sequence analysis of an advanced process.

The invention discloses a knowledge distillation method for target detection. The knowledge distillation method comprises two major designs including a dynamic teacher-student mutual learning mechanism and interpretable feature decoupling. The RGB detector is used as a teacher guidance event detector to learn static semantic knowledge in a conventional illumination scene, and the event detector is used as a teacher guidance RGB detector to learn dynamic robust features in an unconventional illumination scene, so that dual-mode collaborative optimization is realized; and the original teacher and student feature space is decoupled into three parts of mode universality, mode specificity and mode irrelevance, and effective knowledge migration based on the mode universality feature is carried out between the two modes by constructing bidirectional distillation loss. The objective of the invention is to enable an RGB detector and an event detector to have a more accurate target recognition effect.

The invention relates to an intelligent drainage basin management method, device and equipment based on digital twinning and a medium of the intelligent drainage basin management method and device based on digital twinning. The method comprises the steps that core terrain attributes are extracted through terrain feature decoupling and coding, and a disaster response function of an optimal source drainage basin is migrated based on terrain and rainfall similarity; a hydrodynamic model and a migration function are combined to generate a physically constrained synthetic disaster situation data set, a two-channel neural network model fusing topographic features and rainfall dynamics is constructed, and model parameters are dynamically calibrated through Bayesian continuous learning by using real-time monitoring data of a target drainage basin. Finally, a digital twinborn system with flood routing prediction capability is formed, the problem of rapid construction of a flood prediction model under the condition of no historical data is solved, and the timeliness and accuracy of early warning of sudden flood in small and medium-sized watersheds are remarkably improved.

The invention discloses a 3D point cloud large model dialogue security protection system and method based on reinforcement learning and a protection layer, a reinforcement learning security alignment module takes a GPT4Point framework as a base, constructs a point cloud-language joint embedding space, extracts global semantic features by using a multi-layer Transform architecture, realizes efficient alignment of 3D point cloud and text instructions, and improves the security of the 3D point cloud large model dialogue. According to the multi-modal protection layer architecture, lightweight Lama Guard serves as a core model, and cross-modal risk interception is achieved by analyzing geometric features of text entities and associated point clouds generated by a target model in real time. According to the invention, through fusion of reinforcement learning security alignment, a large model protection layer and a GPT4Point multi-mode framework, a multi-level security protection system oriented to a 3D point cloud large model dialogue system is constructed. According to the collaborative design of a reinforcement learning dynamic optimization strategy and a Lama Guard filtering mechanism, the robustness of the model under attack resistance and the adaptability of the model in a natural distortion scene are remarkably improved, and the unified architecture of GPT4Point provides efficient support for point cloud-language understanding and generation.

The invention relates to the technical field of dynamic optimization, and discloses an intelligent learning dynamic optimization system introducing a time sequence. The system comprises a multi-source data preprocessing module, an intelligent learning knowledge graph establishing module, a learning path optimizing module and a learning resource matching module. Firstly, a data detection model is constructed based on a neural network to perform invalid data filtering and abnormal data detection; secondly, knowledge node association strength is calculated based on a time sequence, and an intelligent learning knowledge graph is established; marking a learning path in the intelligent learning knowledge graph, and performing dynamic path planning by using an improved Harris eagle optimization algorithm to find an optimal learning path; and finally, generating learning resource configuration according to the optimal learning path, and establishing a learning resource feature library for resource matching to obtain an intelligent learning optimization strategy. According to the method, the learning behavior data are analyzed and processed by introducing the time sequence, and the purpose of intelligent learning dynamic optimization is achieved.

The invention belongs to the technical field of artificial intelligence, embedded systems and action confrontation command, and discloses an action model learning system and method based on embedded AI. The system comprises an information collection module used for collecting historical action information and carrying out data preprocessing, and the historical action information comprises action environment data, confrontation situation information and decision data; the feature extraction and model construction module is used for extracting features from the preprocessed historical action information by using a deep learning algorithm to obtain action dynamic change features and enemy action mode features, and matching the action dynamic change features and the enemy action mode features with corresponding decision data results to form a training set; training the embedded AI model by using the training set to generate an action model; according to the invention, dynamic data from actions can be processed and learned in real time, so that the action model is continuously optimized, and the command decision-making capability is improved.

The invention provides a federal learning dynamic cutting method and device based on gradient correlation, and the method comprises the steps: receiving a global model issued by a server; training the global model according to the target local data to obtain a local model; determining an update of each level of the local model and an update of each level of the global model; determining a correlation value for each level between the local model and the global model; determining hierarchies of which the correlation values are smaller than a preset correlation threshold value in the local model as cutting hierarchies of the local model; determining hierarchies of which the correlation values are greater than a preset correlation threshold value in the local model as reserved hierarchies of the local model; cutting the invalid update corresponding to the cutting hierarchy of the local model, and returning the valid update corresponding to the reserved hierarchy of the local model to the server; according to the invention, federal learning efficiency and model performance can be improved.

The invention discloses a hybrid sequential network power consumption prediction method for reinforcement learning dynamic calibration, and the method comprises the steps: collecting multivariable power consumption time sequence data, completing the data preprocessing through resampling, feature engineering, normalization and sliding window technologies, generating a supervised learning sample set, and dividing the supervised learning sample set into a training set, a verification set and a test set; constructing a hybrid prediction model comprising a dynamic capture module, a long-term dependence modeling module, a regression prediction module and a reinforcement learning dynamic fine tuning module; training and optimizing by adopting a staged training strategy to obtain a hybrid prediction model; multivariable power consumption time sequence data are collected in real time and preprocessed, the preprocessed data serve as input, real-time prediction of future total consumption is conducted through the mixed prediction model, and a final prediction result after dynamic fine adjustment is output. According to the method, accurate and efficient prediction of the power grid load can be realized, and a reliable technical solution can be provided for power system scheduling optimization, demand side management, market transaction and other scenes.

The invention discloses a global scanning image splicing and defect positioning system for an ultra-large-diameter quartz crucible, and relates to the technical field of quartz crucibles, and the system comprises a scanning control module which constructs a fusion model through multiple sensors, evaluates a risk coefficient, obtains the fusion model, dynamically plans a multi-arm path of a robot based on reinforcement learning, and sends the multi-arm path of the robot; adaptively adjusting the acquisition frequency of the corresponding area based on the risk coefficient, and outputting multi-modal data through a multi-round feedback iteration optimization acquisition process; the image splicing module is used for acquiring multi-modal data for local splicing, and generating a panoramic image in combination with a multi-arm path of the robot; the defect positioning module is used for constructing a quartz crucible model through the panoramic image, detecting defects in a layered manner by utilizing a YOLOv5 model, tracking the defects through extended state Kalman filtering, quantifying the defect volume and the hazard level by combining point cloud semantic segmentation, and performing 3D positioning and evaluation on the defects; the output data is ensured to always meet the subsequent image splicing and defect positioning requirements, and the system stability is guaranteed.

The invention discloses a continuous monitoring method for a composite roof collaborative filling surface, and belongs to the field of filling surface monitoring, and the method comprises the steps: arranging a multi-source sensor based on stope and roadway geometry and filling technology, obtaining and carrying out the time-space alignment of a multi-source signal, extracting the space and time features through a graph neural network and a time sequence neural network, and carrying out the time-space alignment of the multi-source signal; a continuous state field is generated through multi-modal attention fusion; the strategy network outputs filling parameter instructions according to a state field and physical constraints to achieve closed-loop regulation and control, and meanwhile, continuous, high-precision and dynamic monitoring and self-adaptive control of the composite roof collaborative filling surface are achieved in combination with an online incremental learning dynamic optimization model.

The invention provides a power grid standard intelligent recommendation method based on reinforcement learning and post system management. The power grid standard intelligent recommendation method comprises the following steps of 1, collecting and cleaning power grid standard data; step 2, post-standard system knowledge graph construction; 3, designing a reward function for the power grid standard recommendation system, and realizing a personalized recommendation strategy by taking post responsibilities and skill levels as dynamic weight parameters; and step 4, post-driven online incremental training is carried out. According to the method, a reinforcement learning dynamic optimization engine is adopted to learn user behaviors (click rate and task completion score) in real time, and a standard recommendation strategy is adjusted; and when the standard is updated, the post capability weight is automatically adjusted, closed-loop feedback optimization is realized, and the problem of updating delay of a traditional system is solved.

The invention discloses an intelligent virtual power plant cross-domain collaborative optimization system. The system comprises a data acquisition unit which comprises an energy controller, an intelligent electric meter, a sensor and an edge gateway; the edge calculation unit comprises a federated learning optimization module, and the federated learning optimization module is used for carrying out federated learning optimization; based on a lightweight federated learning node local training model, adopting a differential privacy technology to desensitize data; the cloud collaboration unit comprises a multi-modal large model, and the multi-modal large model is used for aggregating edge model parameters and generating a global optimization strategy; and the execution unit comprises an intelligent contract. According to the method, global optimization of industrial, commercial and resident resources is realized through cross-domain collaborative scheduling, and compared with existing single-group scheduling, the efficiency of the system is improved; the communication overhead is reduced by balancing privacy and efficiency, locally processing sensitive data by the edge computing layer and only transmitting model parameters by the cloud, and the system response time is greatly shortened by executing real-time dynamic response and combining reinforcement learning dynamic adjustment strategies.

The invention discloses a power grid energy storage demand response multi-agent reinforcement learning dynamic optimization method, which belongs to the technical field of power system scheduling and optimization control, and comprises the following steps of: constructing a joint optimization model based on a Markov decision process, representing an energy storage charge state, demand response regulation potential and renewable energy output fluctuation, defining a state space, and establishing a dynamic optimization model; establishing a joint action space; constructing a dual-network structure; an improved hierarchical time memory network is introduced, spatial-temporal features are extracted through a multi-level memory unit and a spatial-temporal attention mechanism, and the spatial-temporal features are used for strategy optimization and value evaluation; optimization learning is carried out by adopting multi-agent interaction and centralized training decentralized execution, and strategy updating is carried out by combining an experience playback pool and target network soft updating; and outputting a power grid dynamic scheduling strategy, realizing energy storage and demand response joint optimization, and improving system stability and adaptive capacity under a high-proportion renewable energy access condition.

The invention relates to the technical field of education, and discloses an AI-based learning weakness promotion training test question generation system, which comprises a data acquisition module used for acquiring emotional data and cognitive load data of students in real time, the emotional data comprises emotional fluctuation information of the students, and the cognitive load data comprises emotional fluctuation information of the students; the cognitive load data comprises question solving time and question answering error deviation information of the students; the emotion and cognition analysis module is used for analyzing the current emotion fluctuation and cognition load of the student based on the collected emotion data and cognition load data, and generating a corresponding emotion fluctuation value and a cognition load value; and the test question generation module is used for dynamically generating test questions suitable for the current learning state of the student according to the emotion fluctuation value and the cognitive load value of the student, and adjusting the difficulty and the type of the questions. According to the method, the difficulty and the type of the test questions are dynamically adjusted by combining emotion fluctuation and cognitive load analysis, so that cognitive overload and emotion pressure are effectively avoided, and the learning efficiency and the learning motivation are improved.

The invention provides a computer network data information identification system, which relates to the technical field of computer network security and data processing and comprises an edge cloud collaborative data acquisition module, a multi-modal feature fusion module, a federated learning dynamic model training module and an intelligent decision and response module. The method has the advantages that the edge cloud collaborative architecture is adopted, data preprocessing and feature extraction are conducted on the network edge, the transmission quantity and delay are reduced, and the real-time performance and the processing efficiency are improved; a self-attention mechanism is used for fusing multi-modal features, so that the recognition accuracy is improved; the data privacy security is protected based on a federated learning framework training model; a reinforcement learning algorithm is introduced to optimize federated learning and decision strategies, and the adaptability and intelligence of the system are enhanced; and response abnormal data can be intelligently dispatched according to an identification result to ensure safe and stable operation of the network.

The invention provides a language learning dynamic resource configuration and interaction system based on an internet platform, relates to the technical field of data processing systems, and provides a double-layer cascade diagnosis mechanism. Through a first calculation module, a state anomaly score is calculated based on the stability of user interaction behaviors instead of simple correctness, so that beneficial struggling and harmful fatigue are accurately distinguished; when the score exceeds a threshold value, a second calculation module is activated, a specific learning fragment is analyzed in combination with eye movement trajectory data, and a cognitive deviation value for quantifying specific cognitive impairment is calculated; on one hand, through accurate state recognition, wrong intervention during deep thinking of the user is avoided, and the learning heart stream is effectively protected; and on the other hand, through accurate cognitive attribution, the system can provide targeted accurate assistance, the tutoring efficiency and the learning effect are improved, and intelligent teaching upgrading is realized.

The invention discloses an efficient acquisition and processing system for safety production monitoring data of an electric power steel structure, and the system comprises a space-time sparse sampling module which analyzes the spatial distribution density of monitoring points of the electric power steel structure and the time domain characteristics of vibration signals, generates a dynamic sampling strategy, and carries out the acquisition operation through distributed sensing nodes; and the data transmission scheduling module encapsulates data according to a Predict Steel Core pre-research platform protocol, and performs directional transmission by adopting a time division multiplexing mechanism. The physical information prediction module imports an explicit time-domain physical information dynamic response prediction model and generates state prediction data in combination with the structure parameters; and the threshold early-warning analysis module loads an online incremental learning dynamic threshold early-warning algorithm to carry out real-time comparison. The platform interaction storage module stores data in a classified mode and provides a calling interface, the control coordination module dynamically adjusts operation parameters and data circulation time sequences of all the modules, efficient operation of the whole link is guaranteed, and the processing efficiency and accuracy of the safety production monitoring data of the electric power steel structure are improved.

The invention discloses an aerodynamic noise time sequence prediction method based on a VMD-ESN, and relates to the technical field of hydromechanics noise prediction.The method comprises the steps that firstly, aerodynamic noise signals are collected in real time through a sound pressure sensor; secondly, performing adaptive frequency domain decomposition on the original noise signal by adopting VMD, obtaining a plurality of orthogonal narrowband intrinsic mode functions through a constraint variation optimization framework, effectively separating vortex shedding harmonic and turbulence pulsation characteristics, and inhibiting spectrum aliasing; then, inputting each IMF into an ESN, performing high-dimensional mapping on a modal time sequence evolution rule by using a dynamic reserve pool of sparse connection of a neural network, and training an output layer weight matrix through a ridge regression algorithm; and finally, linearly superposing prediction results of all modes to generate a complete aerodynamic noise time sequence. The method does not need to depend on high-resolution grid iterative calculation, and achieves the accurate prediction of the aerodynamic noise time sequence in a long time interval through a cooperation mechanism of VMD signal adaptive decomposition and ESN machine learning dynamic modeling, and greatly improves the calculation efficiency.

The invention discloses an intelligent management and control system and method for the whole process of deep vertical shaft construction based on deep learning, relates to the crossing field of mine engineering and artificial intelligence, and aims to solve the problems of difficulty in complex spatial-temporal feature extraction, scarcity of labeled samples and weak dynamic working condition adaptation in deep vertical shaft construction monitoring. The system comprises three core modules of spatio-temporal feature extraction, collaborative enhancement semi-supervised optimization and reinforcement learning dynamic decision, firstly, multi-source heterogeneous data is collected through a sensor, deep spatio-temporal fusion features are extracted through an improved spatio-temporal convolutional network, and then a model is optimized by combining a small amount of annotated data with a large amount of unannotated data through a semi-supervised framework. According to the system, the monitoring accuracy is improved, the sample dependence and cost are reduced, the dynamic adaptability is enhanced, the intelligent management and control of the whole construction process are realized, the safety is guaranteed, and the efficiency is improved.

The invention discloses an edge reasoning optimization method and system based on segmented knowledge distillation. Firstly, a segmented knowledge distillation framework is constructed, a teacher-student model is divided into corresponding sub-modules with balanced parameters, parallel distillation training is adopted, middle feature dimension reduction and space alignment are achieved in combination with a principal component analysis method, the knowledge transmission efficiency is improved, and convergence is accelerated. Secondly, proposing an equipment perception self-adaptive pruning strategy, dynamically distributing a differential pruning proportion according to the real-time calculation capability and resource state of heterogeneous edge equipment, and balancing the load of low-performance equipment and the precision of high-performance equipment; and finally, establishing a deep reinforcement learning dynamic scheduling mechanism, generating a module delay-energy consumption file through offline analysis, adaptively selecting a device combination by an intelligent agent in an online stage, determining an optimal partition and deployment strategy through a threshold value distribution algorithm, and realizing joint optimization of energy consumption and reasoning time while meeting delay constraint.

The invention discloses a target detection online learning dynamic sample selection method and system, computer equipment and a storage medium. The method comprises the steps that classification uncertainty and positioning uncertainty of samples to be screened are obtained through Monte Carlo Dropout sampling; constructing multi-dimensional feature vectors including classification uncertainty, positioning uncertainty, knowledge gap matching degree and the like; constructing a dynamic weight learning network based on an attention mechanism, and calculating the weight of each feature in combination with a model verification set performance index; and performing multi-dimensional value scoring on the samples according to the feature weights, and screening out an optimal sample subset in combination with calculation power limitation so as to complete online updating of the model. According to the method, the sample value is comprehensively evaluated through multi-dimensional feature fusion, different scene requirements are adapted by utilizing dynamic weights, and resource consumption and updating effects are balanced in combination with computing power perception sampling, so that the adaptability and detection precision of the model in a dynamic scene are effectively improved, and meanwhile, the dependence on manual annotation is reduced.

The invention relates to the field of data processing, and provides a multi-dimensional feature driven B2B2C collaborative recommendation method and system. The method comprises the steps of performing multi-dimensional collection on B-end merchant features, C-end user features and commodity features through a heterogeneous data source interface to obtain standardized multi-dimensional features; performing dynamic weight learning on the standardized multi-dimensional feature data set through a multi-head self-attention mechanism to obtain a fusion feature vector; performing three-layer cooperative matrix construction on the fusion feature vector based on tensor decomposition to obtain a multi-dimensional factor matrix; performing causal relationship modeling on the multi-dimensional factor matrix through a causal graph structure-based collaborative filtering algorithm to obtain a deep collaborative network model; and performing real-time recommendation of to-be-recommended items through the deep collaborative network model to obtain a personalized B2B2C recommendation list. According to the method, the complex mode in the business scene can be captured, and the accuracy of the personalized recommendation result is improved.