93 results about "Task network" patented technology

The invention belongs to the technical field of computing power distribution, and particularly relates to a dynamic computing power distribution method and system based on reinforcement learning, and the method comprises the following specific steps: S1, covering cloud, edge and end full-node scenes, and collecting computing power resource states, task demand features and cross-domain network condition data in real time; s2, on the basis of standardized data output by a cross-domain computing power sensing module, by constructing a state space fusing computing power, tasks and a network, defining an action space of computing power scheduling direction and proportion, and designing a multi-target reward function for balancing the resource utilization rate, the task satisfaction rate and long-term conflict avoidance; and realizing self-learning and self-iteration scheduling strategy generation based on a reinforcement learning algorithm. According to the invention, the reinforcement learning agent autonomously learns the computing power demand of the emergency scene and the new type of task, the rule does not need to be manually preset and modified, and the method has the advantage of realizing dynamic adaptation of computing power distribution to complex and changeable scenes.

The invention discloses a rock slag multi-dimensional intelligent identification and real-time early warning method for tunnel boring machine construction. The method comprises the steps that continuous rock slag images are acquired and preprocessed; inputting the preprocessed rock slag image into the trained multi-task network, and outputting a multi-task result; wherein the task network adopts a unified encoder and multi-head decoder architecture, the multi-head decoder comprises a segmentation head, a detection head, a regression head and an anomaly scoring head, the segmentation head outputs a pixel mask, the detection head outputs a rock slag frame and category, the regression head outputs a particle size distribution parameter, and the anomaly scoring head outputs a frame level anomaly score; synchronously carrying out sliding window statistics based on output results of the regression head and the abnormal score head; and performing multi-level early warning judgment based on a sliding window statistical result. According to the method, multiple functions of rock slag instance segmentation, target detection, particle size distribution regression, flow estimation, anomaly scoring and the like can be realized, and multi-stage early warning and parameter suggestion are realized through time sequence statistics.

The invention discloses a full life cycle management system for an enterprise product supply and demand order contract, relates to the technical field of enterprise supply and demand contract management, and is used for solving the problem of low efficiency of supply and demand performance collaboration. Through multi-source business event collection and unified coding time reference, records dispersed in a business system are integrated into an event stream sorted according to time, a supply and demand event chain and a life cycle unit corresponding to a contract are automatically identified, the performance state is visible and the abnormity is traceable, and the efficiency is improved. The performance terms and the settlement terms are analyzed into computable constraints, cooperative control of a performance task network and resources is driven, resource demand intervals are generated according to uncompleted tasks and constraints, key resource states are compared to identify resource conflicts, and processing control items such as production, delivery, purchase or term adjustment are generated, so that stockout and delay risks are reduced; and the performance stability and the resource utilization efficiency of the enterprise are improved.

A method includes generating a synthetic dataset with a generative model. The method also includes tuning the generative model based on feedback from a task network that receives the synthetic dataset as input. The task network may perform image recognition. The synthetic dataset may be generated based on a set of classes and labels of the classes. The method may iteratively generate the synthetic dataset and tune the generative model, based on feedback from the task network.

The invention discloses an improved multi-modal three-dimensional medical image classification method based on fusion assistance. The method comprises the following steps: 1, inputting multi-modal medical image data into ResNet to extract modal features and fuse the modal features; 2, constructing a multi-branch network, inputting a fusion feature into a main classification branch, and extracting a cross-plane global context and a fine-grained feature in combination with cross-plane key slice selection and Transform; 3, introducing discriminant prior knowledge generated by the main classification branch into a fusion auxiliary branch, and extracting enhanced features; 4, fusing the main branch fine features and the auxiliary branch enhanced features to obtain semantic level fusion features; 5, inputting the fused features into a classifier to output category probabilities, and taking a category corresponding to the maximum value as a diagnosis label; and 6, joint loss is constructed based on prediction and real labels, and the multi-branch multi-task network is trained and optimized. According to the method, complementary information of the multi-modal medical image is fully mined, focus perception is enhanced by combining judgment prior guidance and multi-task collaborative optimization, and the diagnosis accuracy and stability are improved.

The invention discloses a paralytic nursing intelligent early warning system and method based on the Internet of Things technology. The system comprises a multi-modal physiological parameter acquisition module, an edge calculation preprocessing unit, an intelligent data transmission module, a multi-scale time sequence feature extraction module, a space-time diagram convolutional network module, a cross-modal attention fusion module, a multi-task risk prediction module and a model training and optimization module. According to the system, multi-mode data such as electrocardio, blood pressure, blood oxygen, eye movement tracks and voice are collected, preprocessed at an edge end and then transmitted to a cloud end; a multi-scale convolutional network is adopted to extract time sequence features, parameter association is modeled through space-time diagram convolution, cross-modal data fusion is realized by using an attention mechanism, and finally risk classification, anomaly detection and trend prediction are completed through a multi-task network. The early-stage, accurate and explainable early warning of the stroke risk is realized, and the early warning accuracy and clinical practicability are remarkably improved.

The invention relates to substation inspection, in particular to a transmission line multi-modal inspection data fusion method based on deep learning, and the method comprises the steps: constructing a unified space-time coordinate system, mapping multi-modal inspection data to the coordinate system, and carrying out the space-time alignment of the multi-modal inspection data; according to the characteristics of different modes, a special encoder is used for carrying out feature extraction on inspection data of each mode, and the inspection data are projected to a feature space with a unified dimension; performing dynamic weighted fusion on the high-level semantic features among the inspection data in different modes by using a mutual attention mechanism to obtain a global perception vector; inputting the global perception vector into a downstream task network, and optimizing model parameters by using a corresponding loss function; according to the technical scheme provided by the invention, the defect that the multi-modal inspection data of the power transmission line is difficult to effectively fuse in the prior art can be effectively overcome.

The invention relates to a multi-edge task unloading method combined with task pre-sorting, and belongs to the technical field of edge calculation and task unloading. The method comprises the following steps: firstly, establishing a multi-edge, multi-user and multi-task network architecture comprising a cloud layer, an edge layer and a user layer; based on network architecture support, an application model, a transmission model and an execution model for an application task of a user are established, the application model represents any subtask as a triple, the transmission model defines a corresponding transmission rate, and the execution model comprises a local execution model, an edge execution model and an overall cost model; a task pre-sorting mechanism is introduced, and subtasks which can be parallel are sorted in a descending mode according to priority factors; and finally, based on the ordered task sequence after sorting, performing task unloading strategy decision by adopting a deep Q learning network. According to the method, joint optimization of task execution delay and energy consumption can be realized in a multi-user, multi-task and multi-edge node computing environment.

The invention provides a footprint image intelligent analysis method and system based on a multi-task network, and the method comprises the steps: carrying out the key point marking and preprocessing of a footprint image, and obtaining a key point multichannel Gaussian thermodynamic diagram; matching the footprint image with the key point multichannel Gaussian thermodynamic diagram, and inputting the footprint image and the key point multichannel Gaussian thermodynamic diagram into a deep multi-task network to generate a footprint analysis diagram; and obtaining individual attribute information based on the footprint analysis graph, and carrying out footprint matching and identity discrimination to complete footprint image intelligent analysis based on the multi-task network. According to the method, high-precision footprint segmentation can be realized, an analytic graph can be generated and attribute inference and identity matching can be completed at the same time, and compared with an existing single-task method, the robustness and generalization are remarkably improved. The system has the advantages of being high in accuracy, high in interpretability and good in practicability, and is suitable for the fields of security evidence obtaining, identity recognition, motion behavior research, medical health monitoring and the like.

The application discloses a kind of computing power network green space-time task scheduling and mixed energy collaborative optimization method.The steps are as follows: one, establish the computing power network system model of fusing computing resources, network routing resources and mixed energy, construct multi-objective joint optimization problem;Two, model the problem as a Markov decision process, define the state space containing task, network and energy state, and the action space composed of computing power node selection, routing path selection and forwarding time;Three, use the integrated deep reinforcement learning framework, configure multiple intelligent agents with different preference weights for parallel training, to achieve the best balance between quality of service and carbon emissions;Four, according to the integrated strategy of complete training, combined with the real-time input task flow and carbon emission intensity, output the optimal space-time scheduling decision and energy storage charging and discharging strategy.

The application discloses a kind of based on ring network fusion's collaborative task efficiency analysis method and device, it is related to efficiency analysis technical field.The unmanned aerial vehicle collaborative task system is simplified as four types of node types containing OODA ring;Based on the four types of nodes of OODA ring, determine actual connection edge, obtain task network, extract generalized OODA ring from task network, cover collaborative relationship;Generalized OODA ring node attribute quantization framework based on task demand is built, including task-driven node attribute extraction, quantization method and data acquisition path and data normalization processing;Based on generalized OODA ring node attribute quantization framework, nonlinear aggregation is carried out, and the efficiency of task network is comprehensively analyzed based on the probability of task success.The problem that the prior art cannot accurately capture the efficiency change of unmanned aerial vehicle collaborative task system in dynamic confrontation process, and is limited in practical application, is solved.

This invention discloses an adaptive end-to-end network learning-driven polarization descattering depth estimation method. It proposes a physics-driven end-to-end network to simultaneously acquire clear reconstructed images and high-precision depth images in scattering environments. The method first acquires linearly polarized images of the scene in four directions, extracting light intensity and linear polarization feature maps through decoupling calculations. Then, a dual-task network including a polarization-guided fusion module is constructed. Utilizing the physical sensitivity of DOLP to scattering particle concentration and optical path length, it transforms these into a spatial attention mask, physically constraining and enhancing the shallow texture of the intensity map in the feature space. Finally, dehazing and depth estimation branches are encoded and decoded in parallel. This invention overcomes the intensity-depth ambiguity of traditional RGB images, effectively suppressing the interference of non-uniform fog on geometric perception and visual texture, and achieving high-precision scene depth estimation and high-fidelity dehazing imaging under adverse weather conditions.

The application discloses a kind of mine target area prediction method, system, equipment and storage medium, including obtaining the element content chart of the region to be detected, and according to element content chart constructs chemical data set and is calculated to obtain element feature map, element feature map is carried out to obtain element scale feature map by inflation convolution operation, constructs a well-trained mine target area prediction source task network and multiple same structures target convolutional neural network, element scale feature map is input in corresponding target convolutional neural network with selected weight, to control target convolutional neural network in calculation by selected weight and carry out voting election to obtain final prediction result, can be combined multiple scale features to carry out mine target area prediction, while selectively migrate weight parameters in mine target area prediction source task network are used for each scale network training, solve the problem of few data, improve the accuracy and reliability of intelligent prospecting prediction.

The application discloses a knowledge distillation method and device for reactor nuclear accident deduction and fault diagnosis, and comprises the following steps: obtaining nuclear energy simulation device fault data; building a multi-task network; dividing the obtained data set into a training set, a verification set and a test set, and pre-training the multi-task network by using the training set; fixing the parameters in the multi-task network after pre-training, that is, freezing the pre-training model with large parameters as a teacher model; adopting a multi-task proportional penetration knowledge distillation method to transfer the knowledge of the teacher model to a student model; and obtaining physical quantity deduction results and fault results of a target pipeline by reasoning by using the student model. D The application can simultaneously process multiple fault diagnoses and parameter deductions in a reactor nuclear accident by building a multi-task network based on a physical information neural network, fully utilizes the correlation between different tasks, and improves the accuracy of fault diagnosis and the deduction capability for the development process of a nuclear accident.

The invention discloses a hybrid drive bow net off-line interference cascaded H-bridge rectifier fault diagnosis method, which specifically comprises the following steps: establishing a hybrid logic dynamic model containing a bow net off-line arc effect, and deducing a physical residual error; decomposing the physical residual error and the grid-side current by using a variational mode decomposition technology, and constructing a physical-signal double-domain anti-interference feature vector; inputting the physical-signal double-domain anti-interference feature vector into an LSTM auto-encoder and an unsupervised anomaly perception model, and judging whether an anomaly exists or not through a reconstruction error; and when the judgment result is abnormal, inputting the feature vector into a multi-task LSTM network based on an attention mechanism, and synchronously decoupling and identifying an IGBT open-circuit fault and a sensor fault through a multi-head output structure. According to the method, high-frequency arc noise interference is effectively suppressed, and the limitation that a single data driving model is insufficient in generalization ability under non-stationary strong interference is overcome.

The invention belongs to the crossing field of shipbuilding and electrical engineering, and discloses a shipbuilding process flexible scheduling method based on adjustable potential analysis and a related device, and the method comprises the steps: responding to dynamic disturbance information, and updating a state task network diagram and a scheduling diagram of a shipbuilding process; according to the updated state task network graph and the scheduling graph of the shipbuilding process, dynamic adjustable potential of each task of the shipbuilding process is obtained by combining the dynamic adjustable potential model of each task of the shipbuilding process; and based on the scheduling constraint optimization model, according to the updated state task network diagram and scheduling diagram of the shipbuilding process and the dynamic adjustable potential of each task, generating a flexible scheduling result of the shipbuilding process. An optimal scheduling path can be selected in a disturbance scene to dynamically generate an adjustment strategy, so that the toughness of a production system is remarkably improved; meanwhile, based on dynamic adjustable potential analysis, the adjustment level and the flexible boundary of each task can be accurately identified, and real-time scheduling optimization and intelligent configuration of the flexible manufacturing system are supported.

The application discloses a method for reducing gradient deviation in multi-task network quantization training process, and relates to the technical field of image processing, comprising the following steps: in the process of quantization training of the multi-task network with images as input, image shallow features and image deep features that need to be fused are obtained; the image shallow features and the image deep features are multiplied by learnable weights respectively; two multiplication results are taken as inputs of corresponding feature fusion layers, and then fused image features are obtained; the fused image features are normalized on all channels at each spatial position to obtain normalized image features; and the normalized image features are taken as inputs of subsequent structures of the feature fusion layers of the multi-task network. The method can optimize the balanced contribution of shallow branches and deep branches in the multi-task network in the quantization training process, reduce the gradient deviation in the multi-task network quantization training process, and improve the precision of the multi-task network after quantization training.

The invention provides a microblog forwarding prediction method based on multi-task supervised comparative learning. The problem that a classifier researched by the microblog forwarding prediction method has height deviation is solved. A multi-task network is constructed, forwarding prediction and standing field detection of the microblog are achieved at the same time, and the correlation between tasks is utilized; supervised comparative learning is introduced into each task, so that the features of minority classes are learned more fully, and the performance on the minority classes is improved; the two tasks learn a classifier with small deviation according to a course strategy. Experimental results show that the method has a good effect on the microblog forwarding prediction task. However, the forwarding behavior of the user is closely associated with a plurality of behaviors, more related tasks such as interest recognition, sentiment analysis and sentiment classification can be introduced in the future to be combined with the method, the understanding ability of the model can be further enhanced through the extension, and the user experience is improved. And more comprehensive and deep insight is hopefully brought to the social media analysis field.

The invention relates to the technical field of recombined bamboo board processing, in particular to a recombined bamboo board density accurate detection method. Comprising the following steps: S1, collecting a vibration signal and cooperatively suppressing in different frequency bands, and synchronously collecting a dual-band image in a stable window; s2, dynamically determining a spectrum identification wave band based on the slab temperature, and accurately segmenting the bamboo wood body through a multi-task network; s3, performing multi-scale transformation on the image, calculating texture entropy features and mapping the texture entropy features into texture density contribution values; s4, performing multi-factor coupling compensation on the segmented contour to obtain geometric density, and performing Bayesian fusion on the geometric density and the texture density contribution value to obtain optimal estimation density; and S5, performing double-track judgment in combination with the confidence coefficient of the density estimation and the stability of the production process. Through multi-physical field sensing, macro and micro information complementation and error model and probability fusion cooperation, high-precision and high-robustness online density detection and process monitoring are realized.

The embodiment of the invention provides a cooperative task complexity analysis method and system based on task network characteristics, and the method comprises the steps: carrying out the hierarchical decomposition of a command and control task into atomic tasks according to four dimensions, constructing a task network model, traversing the task network model, extracting network scale characteristics, structural relation characteristics and uncertainty characteristics, and carrying out the calculation of the complexity of the atomic tasks. Complexity calculation is carried out on the network scale feature, the structural relationship feature and the uncertainty feature, a corresponding complexity quantification result is determined, and optimization adjustment is carried out on the task network model according to the complexity quantification result, so that the task network complexity of the task network model is reduced, and the task network complexity of the task network model is improved. The accuracy and efficiency of command and control task complexity measurement can be improved.

The invention discloses a semantic segmentation neural network model construction method for a river channel bank collapse monitoring scene, and the method comprises the steps: training a teacher network containing dual-path convolution and a dynamic mask mechanism through a bank collapse image data set, introducing geometric consistency loss in the training, and strengthening boundary features; extracting a collaborative association feature representing a structured dependency relationship between boundary detection and region segmentation tasks in the teacher network; and constructing a single-branch lightweight student network, and guiding student network training by using a distillation loss function containing a collaborative association feature alignment item. According to the method, background redundancy is eliminated through dynamic sparse calculation, double-task priori knowledge is migrated to a single-task network through collaborative distillation guided by function affiliation, the problems that bank collapse boundary detection is fuzzy and edge end model reasoning delay is high are solved, and high-precision and low-power-consumption real-time monitoring is achieved.

The application belongs to the technical field of network attack and defense simulation, and discloses a network attack and defense simulation engine system based on discrete event driving, which comprises an event scheduling module, a simulation network construction module, an attack and defense simulation module, a simulation kernel module and a panoramic situation awareness module; the event scheduling module is constructed based on MITRE ATT&CK as a theoretical base and is used for converting all network activities into a schedulable discrete event sequence; the network activities comprise packet sending, protocol timeout and attack triggering; the simulation network construction module can construct a simulation network containing complete business logic. The application can convert the MITRE ATT&CK tactical intention into an ordered event stream through the event scheduling module, and can simultaneously perform dynamic decomposition and re-planning on a high-level target through a hierarchical task network planner in the attack and defense simulation module, and can also reconfigure an attack chain in real time by using an ATT&CK technology ID.

The invention discloses a gait identity recognition method based on a generative self-supervised visual pre-training model BEiT and DAS technology. The method comprises the following steps: firstly, carrying out scene construction and data preparation, collecting label-free data and label data, extracting data signals for signal preprocessing, and constructing a time frequency characteristic data set; constructing a pre-training image reconstruction network of a self-supervised visual pre-training model BEiT, performing feature extraction on an input image, constructing a deep learning network based on the self-supervised visual pre-training model BEiT as a feature extractor, and training the deep learning network by using unlabeled spatio-temporal data in an experimental scene to obtain weight parameters; and constructing a downstream classification task network, performing classification learning of downstream tasks by using a small amount of data with labels to obtain a new weight, and completing identification and classification of non-label data by using the obtained new weight. According to the method, the demand quantity of the tagged data is reduced, and the accuracy is ensured.

The invention relates to the field of trajectory prediction of automatic driving, in particular to a vehicle trajectory prediction method based on discriminator implicit future interactive learning guidance. Comprising the following steps: step 1, designing a training framework of a track prediction model based on cGAN; 2, designing a sub-task network PRNet for evaluating the future interaction relationship between the vehicles; and step 3, alternately training the generator G and the discriminator D based on a dynamic weight strategy. Experimental results on a public data set and a baseline model show that the method provided by the invention has the advantages that the accuracy and the social compliance of the trajectory are effectively improved, and the universality is realized.

The invention discloses a training method, an analysis method, equipment and a program product of an electroencephalogram analysis model. In the training method, a first network is adopted to extract graph features of electroencephalogram signals to be analyzed as semantic features; using a second network to extract graph features of the to-be-analyzed electroencephalogram signals as domain change features; mapping the semantic features and the domain change features into a Granger causal relationship matrix by adopting a causal representation network; adopting a target task network to execute a target task according to the Granger causality matrix; performing individual classification according to the domain change characteristics by adopting a domain classification network; and updating parameters of the first network, the second network, the causal representation network, the target task network and the domain classification network according to target task loss and individual classification loss. The training method has good field generalization ability.

The invention discloses a driving behavior intelligent analysis method and system based on multi-modal feature fusion, and the method comprises the steps: obtaining a driving behavior data set and a data flow, which comprise a vehicle time sequence signal and image data; respectively decomposing the image data and the vehicle time sequence signal to obtain a small image feature map and a time-frequency feature vector; constructing a multi-task network pre-training model, carrying out interaction and enhancement on the image wavelet feature map and the time-frequency feature vector to obtain a fusion feature vector, carrying out deep extraction on the fusion feature vector to obtain feature output data, and carrying out evaluation and mapping based on the feature output data to obtain a dangerous driving probability and a dangerous driving safety score; training the multi-task network pre-training model to obtain a multi-task network model; and obtaining a corresponding driving behavior label, confidence and a driving behavior score through model reasoning. According to the method, feature extraction and interactive fusion are carried out on the multi-modal data, and accurate analysis of driving behaviors is realized.

The invention relates to an adhesive tape coating regulation and control optimization method and system based on intelligent image analysis, and the method comprises the following steps: S1, obtaining a product specification, current situation data and production line capability, building a quantitative index system and constraint conditions, determining a security domain based on historical data and a physical model, and obtaining a constraint set and the security domain; s2, constructing a multi-modal sensing network based on the constraint set and the security domain; s3, based on a multi-mode sensing network, collecting images and process data of each station, and preprocessing the images and the process data; s4, constructing a multi-task network analysis model, and performing analysis based on the preprocessed image and process data of each station to obtain an analysis result; and S5, based on an analysis result, adopting reinforcement learning to control strategy generation, and obtaining an optimal control sequence. The intelligent quality control efficiency and reliability of film coating are effectively improved.

This application provides a method, apparatus, device, and medium for generating control strategies for virtual objects, relating to the field of game technology. The method includes: obtaining a request for acquiring a target control strategy for a target virtual object; based on the acquisition request, obtaining target character skill network parameters and target task network parameters corresponding to the target virtual object through a basic strategy model; determining an initial network model based on the target character skill network parameters and target task network parameters; training a target network model based on the initial network model, and determining the target control strategy for the target virtual object through the target network model. This method achieves improved efficiency in generating target control strategies by eliminating the need to build an initial network model from scratch, then training the target network model based on it, and finally determining the target control strategy for the target virtual object through the target network model.

This invention discloses a method for constructing a semantic segmentation neural network model for riverbank collapse monitoring scenarios. The method includes: training a teacher network containing dual-path convolution and dynamic masking mechanisms using a riverbank collapse image dataset, introducing geometric consistency loss during training to strengthen boundary features; extracting collaborative correlation features from the teacher network that represent the structured dependencies between boundary detection and region segmentation tasks; constructing a single-branch lightweight student network, and using a distillation loss function containing alignment terms of collaborative correlation features to guide student network training. This invention eliminates background redundancy through dynamic sparse computation and transfers prior knowledge from the dual-task approach to the single-task network through function-attribution-guided collaborative distillation, solving the problems of fuzzy boundary detection and high inference latency at the edge of the model, thus achieving high-precision, low-power real-time monitoring.

The invention discloses an intelligent teaching system integrating college culture and vocational education, particularly relates to the field of intelligent teaching, and is used for solving the problem that existing teaching knowledge points are separated from virtual practical teaching. According to the system, campus culture activity planning is divided into multi-stage subtasks, a task network directed graph model is constructed based on professional roles, students are organized in a virtual or mixed reality environment to form a cross-professional virtual project team, and collaborative practical training execution under culture theme driving is achieved. The system continuously records and verifies student operation behaviors and task output in a task promotion process, dynamically evaluates an overall activity promotion state, and automatically generates a teaching guidance and team reorganization scheme when a cooperation bottleneck appears, thereby realizing intelligent guidance, state evaluation and cooperation optimization of a culture activity practical training process, and improving the training efficiency. And normalization, controllability and effect evaluability of fusion implementation of college and university culture and vocational education are improved.