343results about How to "Reduce training time" patented technology

The invention provides a model pre-training method and device for a multi-language task, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining a multi-modal training data set which comprises a plurality of pieces of sample text data and a plurality of pieces of sample image-text pair data with multi-language contents, and covers a universal field and a preset business field; based on a plurality of sample image-text pair data, performing comparative learning training of alignment of visual features and text features on the initial model, freezing model parameters of a decoding module and adjusting model parameters of a visual encoder and a projection module in the training process until a first ending condition is met; based on the multiple pieces of sample image-text pair data and the multiple pieces of sample text data, constraint training of content understanding is carried out on the initial model meeting a first ending condition, and model parameters of a visual encoder, a projection module and a decoding module are adjusted in the training process until a second ending condition is met to obtain a target model; the information extraction capability of the model in the specific field can be remarkably improved.

The invention belongs to the technical field of power distribution network fault recovery, and provides a power distribution network fault recovery method and system based on multi-agent reinforcement learning, and the method comprises the steps: taking a recovered load as a target function based on the load priority of a power distribution network, and taking the operation limitation of each node line in the power distribution network as a constraint condition; constructing a power distribution network fault recovery problem model; based on the power distribution network fault recovery problem model, constructing a multi-agent reinforcement learning agent set, and converting the power distribution network fault recovery problem into a Markov decision process; and a multi-agent reinforcement learning framework is constructed based on the agent set of multi-agent reinforcement learning, multi-agent training is carried out, each agent generates a decision according to respective observation information, and a fault recovery strategy is obtained. The objective function and the constraint condition constructed based on the load priority of the power distribution network and the node line operation limitation can ensure the accuracy and the feasibility of the recovery strategy. And the strategy generation time is shortened through the multi-agent parallel decision-making process.

The invention belongs to the technical field of model training, and particularly relates to a large model increment training method and system based on dynamic sparsification. The method comprises the following steps: S1, evaluating the parameter importance of a large model by adopting a multi-dimensional evaluation method, and reflecting the contribution of parameters to the performance of the large model; s2, performing dynamic sparsification control on the large model structure by adopting a self-adaptive structure adjustment method; s3, performing incremental training management on the large model by constructing a capacity-adjustable empirical playback buffer pool and proposing a multi-objective loss function; and S4, carrying out model compression optimization on the large model by adopting a mixed precision quantification strategy. The method has the characteristics that the network structure can be dynamically adjusted by evaluating the importance of model parameters in real time, so that the efficient training is realized while the model performance is ensured.

The invention relates to a chilled water system of a central air conditioner and an energy-saving optimization control method. The energy-saving optimization method comprises the following steps: S1, establishing an energy consumption model of the chilled water system of the central air conditioner; s2, firstly performing correlation analysis on the selected cooling load influence factors by adopting a grey correlation analysis method, screening out strong correlation influence factors, and establishing a central air conditioner cooling load prediction model based on PSO-LSSVM (Particle Swarm Optimization-Least Square Support Vector Machine); s3, optimizing the energy consumption model parameters of the chilled water system of the central air conditioner while predicting the cooling load value of the next point by using an improved white whale algorithm, and enabling the output result of the energy consumption model to reach the minimum value; the system comprises an energy consumption modeling module, a cooling load prediction module and an energy-saving optimization module. According to the method, the energy consumption model and the cooling load prediction model are established, the cooling load value of the next point is predicted, parameter optimization is performed, and an optimization structure is output; the energy consumption model and the cold load prediction model can be updated periodically, and the problem of performance reduction caused by long-time equipment operation is avoided.

The invention relates to the technical field of network load prediction and scheduling methods, in particular to a mobile communication network load prediction and scheduling method and system based on deep learning, and the method comprises the steps: obtaining data from a plurality of edge computing nodes, including base station data, user behavior data and environment parameter data; constructing a multi-layer model for load prediction and resource scheduling based on cooperative work of the central system and the edge system; carrying out feature extraction by utilizing a dynamic feature embedding transformation module in the edge computing node; recursive spatial-temporal feature processing is carried out based on the optimized features; generating a network load prediction result through the prediction model; and resource scheduling optimization is performed through a reinforcement learning model based on a prediction result, and multi-dimensional features are nested and mapped to a high-dimensional feature space through dynamic feature embedding of original data. And the nested features are further optimized through a time-varying collaborative tensor optimization algorithm, so that the spatial relevance between different base stations can be enhanced while complex nonlinear features are captured.

The invention discloses a multi-source ship target identification method based on two-stage collaborative fusion, and relates to the technical field of signal processing, and the method comprises the steps: firstly collecting target sea area data, including original optical image and radar image data, by means of an optical camera and a radar sensor on an aircraft, and then carrying out the preprocessing of the collected data through Gaussian filtering; performing sea-land segmentation by using a deep learning algorithm, constructing a loss function optimization network parameter to obtain a segmentation mask, and removing a land background area according to a mask result to obtain processed optical and radar images; an improved SURF feature and graph cut algorithm combined registration method is adopted, optimal transformation parameters are obtained by screening feature point pairs and solving the minimum value of an energy function, and image registration is achieved; for the registered images, fusing the images based on an adaptive fusion rule; and finally, extracting corresponding features through manual feature extraction and feature extraction based on deep learning, and fusing the features to obtain a final target recognition result.

The invention discloses a networked measurement and control data transmission task scheduling method for a giant low-orbit satellite constellation, and solves the problem of low completion rate of measurement and control data transmission tasks caused by incapability of global station distribution in the prior art. Obtaining a resource attribute, a ground station state and a low earth orbit satellite state by using the task layer, the resource layer, the equipment state and the constellation state, performing data conversion to obtain input data, inputting the input data into the trained multi-agent network to obtain an action corresponding to the current time slot, and determining an entry low earth orbit satellite and an associated ground station according to the action; obtaining a routing path between the entry low earth orbit satellite and the task low earth orbit satellite according to the inter-satellite shortest path routing table, and determining a scheduling sequence according to the routing path and the associated ground station; the mutual influence between global tasks and resources is considered, the effectiveness of decision making during execution is reduced, and meanwhile, the scheduling efficiency of measurement and control data transmission tasks in a giant low-orbit satellite constellation scene is improved.

The invention provides an adaptive optimization method and system for a training process of a multi-modal large model. The method comprises the following steps: receiving a submitted training execution script and a training strategy of the multi-modal large model; according to the training execution script and the training strategy, utilizing a pre-stored large model training script to execute a distributed training task so as to generate a multi-modal large model; acquiring real-time monitoring data generated in the process of executing the distributed training task; adjusting a training strategy based on the real-time monitoring data, and continuing to execute the distributed training task by utilizing the large model training script according to the large model training execution script and the adjusted training strategy until a trained multi-modal large model is obtained. The training strategy is dynamically adjusted in real time, unnecessary computing resource waste is avoided, the training time of the model is shortened, and the training efficiency is improved; the model can be adaptively optimized according to the characteristics of the monitoring data, and the accuracy and generalization ability of the finally obtained multi-modal large model are improved.

The invention discloses a blood vessel segmentation method and device based on large model fine tuning, equipment and a medium, and the method comprises the steps: obtaining a blood vessel image, carrying out the preprocessing of the obtained blood vessel image, and obtaining a training set; performing fine tuning on a pre-trained image segmentation model by adopting the training set, so that the image segmentation model learns related features of blood vessel segmentation; wherein the low-rank adaptive fine tuning technology is adopted to perform fine tuning on the image segmentation model: two parallel low-rank matrixes are inserted into a Transform layer in the image encoder, the input is processed by the two low-rank matrixes in sequence, and the processing result and the original output are added to obtain the final output. According to the method, the defects of a large model in a blood vessel segmentation task are overcome through efficient parameter fine tuning and a multi-scale feature fusion module. Accurate blood vessel segmentation can be realized under limited annotation data, the difficulty and cost of medical image segmentation tasks are greatly reduced, and the method can be widely applied to the technical field of medical image processing.

The invention provides a Chinese electronic medical record named entity recognition method based on a multi-head attention mechanism and a pulse neural network (SNN), and the method comprises the steps: obtaining a dynamic word vector through a pre-training BERT model, obtaining a text code through a bidirectional long short-term memory network (BiLSTM) model, and screening the features of the code through the SNN network for a named entity in an electronic medical record; the multi-head attention mechanism calculates the dependency relationship between characters, enhances the relevance to the context, and finally outputs a final sequence through a conditional random field (CRF) module to obtain an entity recognition result. According to the method, the time cost is remarkably reduced, the efficiency and precision of identifying the named entities in the medical text are improved, and a more efficient and accurate technical solution is provided for the field of medical information processing.

The invention relates to the technical field of artificial intelligence, and discloses a multi-modal model training fine-tuning optimization method based on artificial intelligence, which comprises the following specific steps: preparing and preprocessing data, collecting a large-scale multi-modal data set, constructing an image text data set containing one million high-quality images and text description accurately corresponding to the images, and performing fine-tuning optimization on the image text data set. According to the method, data are subjected to comprehensive and meticulous cleaning operation, noise data and mistakenly labeled data in the data are strictly removed, so that the accuracy and reliability of the data are ensured, and the effective utilization rate of the data can be increased by at least 20% through data cleaning; wherein the image data is screened, blurred, low-resolution and unclear-content images are removed, it is ensured that each image can clearly convey specific visual information, and text description, grammar error checking, spelling error and semantic ambiguity parts are corrected or deleted.

The invention discloses a large model cloud computing dynamic updating method based on elastic adaptive incremental learning, which comprises the following steps: S1, in a cloud computing platform, acquiring a current version of a large model and a new data stream to be processed; s2, local parameter updating is conducted on the large model according to the new data, and parameters or model layers needing to be updated are dynamically selected according to the characteristics of the data in the updating process; s3, introducing an adaptive learning strategy, and monitoring the change condition of data distribution in real time; s4, in a large model updating process, splitting a large model updating task to a plurality of computing nodes for parallel execution through a multi-node collaborative scheduling mechanism of the cloud; s5, after the large model is updated, the system verifies the performance of the large model; and S6, in the process of updating and deploying the large model, the system dynamically recycles computing resources which are not used any more. According to the method, idle computing resources can be fully utilized in a large-scale cloud computing environment, so that computing tasks are distributed in a balanced manner, and dynamic adjustment of task loads is ensured.

The invention provides a vegetation prediction model construction method, a vegetation prediction method, a vegetation prediction device and vegetation prediction equipment, and relates to the technical field of data processing. The vegetation prediction model construction method comprises the following steps: taking ecological environment data of a target area as an independent variable, taking vegetation spatial distribution data corresponding to the target area as a dependent variable, and training a first random forest model to obtain a vegetation spatial distribution prediction model; taking the output of the vegetation parameter spatial distribution prediction model and the period-by-period meteorological data as independent variables, taking the period-by-period vegetation feature change data as dependent variables, training a second random forest model, and obtaining a vegetation feature time change prediction model; and splicing the vegetation spatial distribution prediction model and the vegetation feature time change prediction model to obtain a vegetation prediction model. According to the technical scheme, the data utilization efficiency can be improved, the model training time is shortened, and meanwhile the adaptability of the vegetation prediction model and the accuracy of vegetation prediction are improved.

The invention discloses an unmanned aerial vehicle autonomous navigation method based on imitation learning and reinforcement learning fusion, and belongs to the technical field of unmanned aerial vehicle processing, and the method comprises the following steps: S1, collecting a data set of an unmanned aerial vehicle, and carrying out normalization processing on the data set to obtain a standard data set; s2, constructing an Actor network by utilizing expert operation data of the standard data set; s3, inputting the depth image and the state information of the standard data set into an Actor network, and generating a pre-judgment action; and S4, inputting the pre-judgment action into a discriminator to complete autonomous navigation. Through introduction of the generative adversarial imitation learning module, the unmanned aerial vehicle can continuously optimize a navigation strategy, and high robustness and adaptability can be maintained even in a complex environment.

The invention provides a large model training data generation method in the medical field, which takes real case data as reference, calls a large language model through an automatic program, enables the large language model to generate problem data, reference data and answer data in batches, and guides the large language model to generate and optimize model fine adjustment data through a set prompt prompt word. Therefore, the specialized model fine adjustment data is generated for the industry model in the medical field, and the training time of the subsequent industry large model is saved. In addition, due to the fact that the case data is high in obtaining difficulty and much in privacy information, compared with direct obtaining of real case information, the technical scheme disclosed by the invention has the advantages of being low in cost and high in privacy.

The invention relates to the technical field of artificial intelligence, in particular to a large model fine tuning method, system and device based on LoRA-MoE and a medium. According to the method, the characteristic that the LoRA technology can use smaller parameters to carry out fine adjustment on a large model is combined with the characteristic that the MoE architecture can enable the model to be higher in generalization ability and better in effect, and the LoRA-MoE technology is adopted to carry out fine adjustment on the large model; according to the method, the problems that a large number of video memories need to be consumed during fine adjustment of the large model and the effect is poor when fine adjustment is conducted through the LoRA technology can be solved, it is ensured that consumption can be reduced in the fine adjustment process of the large model, and a certain fine adjustment effect is ensured.

The invention relates to a TCN and BiGRU-based power load decomposition and identification method and system, and the method comprises the steps: obtaining a data set of the total household electricity utilization power and the electricity utilization power of each electric appliance, carrying out the data preprocessing of the data set, and building a training data set and a test data set; constructing a TCN BiGRU network model comprising a TCN network, a BiGRU network and a GAM attention mechanism; inputting the power sequence into a TCN network to perform preliminary feature extraction, extracting sequence long-time dependence information, and obtaining power sequence shallow layer features; taking the shallow-layer features as input features of a BiGRU network, and extracting bidirectional deep-layer sequence features from the shallow-layer features; taking the bidirectional deep sequence features as input sequence information of a GAM attention mechanism, and outputting a predicted sequence midpoint; training and verifying the TCNBiGRU network model through the training data set and the test data set; and power load decomposition and identification are carried out through the trained TCNBiGRU network model. The method and the system are beneficial to extracting long-time dependence information and improving the power load decomposition and identification precision.

The invention provides a multi-teacher model knowledge distillation method integrating semantic weight soft labels, which comprises the following specific steps: firstly, inputting a data set into each teacher model of a multi-teacher module, and outputting probability distribution of labels corresponding to input data by the teacher models, namely soft labels; secondly, labels output by the teacher model are converted into semantic vector representations through a pre-training language model, similarity values among the labels are calculated through the semantic vector representations of the labels, probability distribution of the soft labels is weighted through the semantic similarity values, and therefore semantic weight soft labels are obtained; finally, in the knowledge distillation process, the student model will distill knowledge from the teacher model through the two types of losses. The two types of loss comprise cross entropy loss and weighted relative entropy loss based on label semantic vector similarity. According to the method, the semantic information of the label can be fused into the probability distribution of the soft label of the multi-teacher model, and the knowledge distillation performance is improved.

The invention relates to the technical field of natural language processing, in particular to a method and system for achieving intelligent number asking based on a large model, and the method comprises the following steps: integrating data from various data sources including a database, a data warehouse, WEB service and Excel files; the data is cleaned, converted, standardized and normalized through the data management center, so that the data quality is ensured, and basic data support is provided for subsequent intelligent question analysis; the beneficial effects are that the method for realizing intelligent number asking based on the large model provided by the invention responds to complex and changeable data requirements in real time, helps an enterprise to quickly make a data-based decision, and reduces the time cost of waiting for the assistance of an IT department or a data analysis team; through effective integration and intelligent search of mass data resources, various data in an enterprise can be fully mined and utilized, and idleness and waste of the data resources are avoided.

The invention relates to a hyperspectral image classification method based on a meta-pseudo-label multi-scale image convolution network, and the method comprises the steps: building a meta-pseudo-label multi-scale feature search image convolution network (MPL-MFSGCN), dividing the HSI into superpixel images of different scales through employing an SLIC superpixel segmentation technology, and converting the superpixel images into image structures; secondly, constructing a multi-scale feature search graph convolutional network MFSGCN architecture, optimizing a superpixel feature weight combination by using a neural network search algorithm, and enhancing the discrimination of feature expression; a semi-supervised teacher-student model sharing the same graph convolutional network is adopted to generate meta-pseudo tags, and a student network is finely adjusted to improve the robustness of the network; the key features can be automatically captured under the small sample condition, the classification precision higher than that of an advanced method is obtained compared with a common HSI data set, and the classification accuracy is effectively improved.

The invention provides an Internet of Things early-stage malicious traffic detection method combining a domain name and DNS traffic characteristics, and the method is divided into four parts: the first part is traffic capturing and filtering; the method comprises the following steps: 1, extracting DNS traffic information, 2, training a detection model, and 4, detecting malicious traffic of the Internet of Things, and specifically comprises the following steps: extracting traffic characteristics such as an Internet of Things DNS domain name and query information; dNS domain name features are learned by using a long and short term memory model with a multi-head attention mechanism, DNS traffic features are extracted by using a convolutional neural network, and detection is performed in the early stage of malicious traffic occurrence. According to the method provided by the invention, the malicious traffic of the Internet of Things can be quickly and efficiently detected in the early stage, the recognition accuracy reaches 98%, and the model training time and the detection time are short. The method for detecting the malicious traffic in the early stage is convenient for a network manager to quickly respond and protect the Internet of Things equipment.

The invention discloses a federated undersampling learning method for unbalanced network intrusion detection, which comprises a federated learning server and a client, and communication connection is established between the federated learning server and the client; all models for network intrusion detection are configured in the server; the server is used for maintaining a global model and deploying the global model to the client for training and learning, and the client adopts a local data set to train a local model downloaded from the server and uploads the local model after training and learning to the server; the server aggregates the local models uploaded by the client to obtain a global model with updated parameters; and when the client adopts the local data set to train the server, undersampling processing is performed on the local data set to obtain a balanced data set to perform learning training on the client. When federated learning is carried out on an intrusion detection network model, the influence of training data imbalance in a client on the model detection performance is solved through an undersampling method.

The invention belongs to the technical field of medical images and deep learning, and particularly relates to a non-paired cross-modal medical image conversion method based on contour guide path regularization, which comprises the following specific steps: constructing a conversion path from a source domain to a target domain; a path regularization method is applied to smooth path transition and ensure the stability in the image conversion process; according to a path regularization technology based on contour guidance, the model can be focused on a key anatomical region, the structural consistency and accuracy of the generated CT image are further improved, and the training time can be remarkably shortened by reducing additional training steps. The invention aims to simplify the unsupervised training normal form and overcome the defects of the traditional GAN-based single-step generation method in the aspect of capturing the complex anatomical structure of the medical image. While the training time and the sampling time are considered, the problem that the quality of the generated medical image is limited is overcome.

The invention provides a model training method and device, electronic equipment and a storage medium, and relates to the technical field of distributed computing. The method comprises the steps of dividing a context into context tasks based on structural analysis of a to-be-trained model; performing dependency relationship analysis on the context task to determine a task dependency graph; configuring priorities for the context tasks according to the task dependency graph, distributing the context tasks with the configured priorities to corresponding computing nodes, training according to an execution sequence, and obtaining load information and task execution information of the computing nodes; and according to the load information and the task execution information of the computing nodes, scheduling context tasks among the computing nodes, adjusting the execution sequence of the context tasks in the computing nodes, and outputting a training result. According to the real-time load condition of each node, task allocation is adjusted, load balancing of the computational nodes is realized, the utilization efficiency of computational resources is improved, a context task scheduling strategy is optimized, the utilization rate of the computational resources is improved, and the training time is shortened.

The invention discloses an intrusion detection method and system based on artificial intelligence, and belongs to the technical field of network security, and the method comprises the steps: firstly extracting the features of network flow data, and carrying out the standardization and dimension reduction processing of the extracted features, and obtaining a feature data set; constructing a joint machine learning model which comprises a plurality of parallel learning models, performing batch training on all the learning models by adopting the data set and combining transfer learning, and fusing prediction results of a plurality of pre-trained learning models by adopting an integrated learning method to obtain a trained joint machine learning model; inputting real-time network flow data into the trained combined machine learning model, and performing intrusion monitoring on the network flow data by the combined machine learning model; according to the method, efficient and real-time intrusion detection is carried out on the network traffic, and the detection accuracy and adaptability are improved.

The invention discloses a voice recognition hybrid model construction method and system for power grid equipment monitoring. Power grid equipment related document information is collected to form a corpus set in the power grid equipment monitoring field, the corpus set is filled into an instruction format template to form a monitoring professional voice recognition corpus, the monitoring professional voice recognition corpus is loaded into a model training algorithm, and incremental training is performed on a power grid equipment monitoring professional model and a hot word model; exporting the power grid equipment monitoring professional model and the universal recognition model as an acoustic model and a language model, respectively inputting a to-be-recognized voice signal segment into the acoustic model and the language model, calculating a word sequence score difference value, if the score difference value is higher than a threshold value, decoding and outputting the model corresponding to the highest score, and outputting the speech signal segment; and otherwise, judging whether the phoneme fragment contains the power grid professional word based on the hot word model, if so, decoding and outputting the phoneme fragment by a monitoring professional model, and if not, decoding and outputting the phoneme fragment by a general model. According to the invention, the recognition accuracy of voice recognition of monitoring service instructions and non-instruction services is realized.

The invention discloses a dynamic space-time diagram flow prediction method and system based on course learning, and relates to the technical field of supply chain logistics data analysis, and the method comprises the steps: building a space-time matrix based on historical multi-source data, generating a dynamic adjacent matrix through learning, and carrying out the smooth fusion through combining a static diagram, and forming a dynamic diagram structure. And then space and time features are respectively extracted by using a graph convolutional network and a gating loop unit, and deep interaction and fusion are realized through a bidirectional cross attention mechanism. A multi-dimensional difficulty estimator is innovatively introduced, the prediction difficulty of each training sample is quantified from three dimensions of space, time and time-space coupling, the selection sequence of the training samples is dynamically adjusted based on an adaptive course scheduler, and progressive learning is realized. And finally, feature representation is obtained through global pooling, and multi-step traffic prediction is realized by adopting a parallel independent decoder, so that error accumulation is avoided. According to the invention, prediction precision and model training efficiency in a complex supply chain logistics scene are effectively improved.

The invention discloses a valve leakage detection method, which comprises the following steps of: acquiring an acoustic emission signal of a pressure pipeline by using an acoustic emission sensor and amplifying the acoustic emission signal; respectively carrying out signal noise reduction processing and leakage signal feature extraction on the amplified acoustic emission signal and the non-leakage acoustic emission signal to correspondingly obtain noise-filtered leakage acoustic emission data and normal acoustic emission data; inputting the normal acoustic emission data into the GRU model to obtain prediction data, and determining a confidence rate value according to a difference value between the prediction data and real data; respectively inputting the normal acoustic emission data and the leakage acoustic emission data into the GRU model, calculating to obtain a normal difference value and a leakage difference value, and respectively comparing the normal difference value and the leakage difference value with confidence rate values to obtain a normal confidence rate and a leakage confidence rate; and calculating a difference value between the normal confidence rate and the leakage confidence rate to obtain a confidence rate difference value, if the confidence rate difference value is smaller than a preset value, judging that leakage does not occur, otherwise, judging that leakage occurs. Real-time and accurate detection of internal leakage of the valve is realized.

The invention is suitable for the technical field of structural design, and provides a concrete interface shear strength prediction method, device and equipment and a storage medium, and the method comprises the following steps: obtaining the shear strength and concrete parameters of a concrete interface; establishing an initial LightGBM model, recording the initial LightGBM model as a first model, and preprocessing shear strength and concrete parameters to obtain a training set and a test set; based on a collaborative group optimization algorithm and a genetic algorithm, optimizing hyper-parameters of the first model to obtain a target hyper-parameter set; applying the target hyper-parameter set to the first model to obtain a second model, and training the second model based on the training set and the test set to obtain a shear strength prediction model; and inputting the concrete parameters of the target concrete interface into the shear strength prediction model to obtain a shear strength prediction value of the target concrete interface. According to the method, the shear strength of the concrete interface can be accurately predicted under the condition of complex working conditions.

The embodiment of the invention provides a diffusion model construction method and device, equipment and a storage medium. The method comprises the steps that a pre-training model is acquired, a target model structure of a target diffusion model is constructed, and the target model structure comprises a basic network and a target up-sampling stack network; performing initialization operation on the target model structure through the pre-training model to obtain an initial diffusion model; sequentially carrying out time sequence modeling processing and text embedding processing on the initial diffusion model to obtain a processed initial diffusion model; and training the modeled initial diffusion model according to a training set to obtain the target diffusion model, the training set comprising a plurality of pieces of text information, and the target diffusion model being used for analyzing the plurality of pieces of text information to obtain corresponding video information. The method aims at generating the diffusion model which is excellent in performance and universal, reduces the construction cost of the diffusion model, and can meet the requirements of scenes such as the financial field.