102results about How to "Fast training" patented technology

The invention belongs to the technical field of image classification neural networks, and discloses a lightweight image classification neural network architecture system based on knowledge distillation. The invention provides a lightweight image classification neural network architecture system based on knowledge distillation. The lightweight image classification neural network architecture system comprises a data set preprocessing module, a teacher model preprocessing module, a subnet model construction module, a personalized training module, a real loss calculation module, a distillation loss calculation module, a total loss calculation module and a student model construction module. The invention provides a lightweight neural network architecture system based on knowledge distillation, and designs a brand new training mode by using the superiority of knowledge distillation for teacher model weight extraction: the knowledge distillation of teacher and student models is combined with a deep compression method; the system for constructing the lightweight network can be used for training various complex image classification neural networks so as to improve the network performance.

The invention discloses an emergency fire fighting question answering method and system based on the AI large language model technology and a storage medium, and relates to the technical field of artificial intelligence, the emergency fire fighting question answering system comprises a front-end module, an AI preprocessing thread pool, a knowledge base module and a large model calculation engine module, the AI preprocessing thread pool receives a service request sent by the front-end module, and sends the service request to the AI preprocessing thread pool; different task threads are distributed, a knowledge base of a corresponding region category is automatically called according to the region category input by a service request to perform content similarity retrieval, knowledge content data obtained through retrieval and the corresponding service request are sorted and then uniformly sent to the large model calculation engine module supporting parallel processing to perform analysis and calculation, and the large model calculation engine module is used for performing analysis and calculation on the knowledge content data and the corresponding service request. By using the technical means of the invention, the user does not need to think what the keyword of the current question is, and can completely organize the question, directly ask the question, automatically query the local professional knowledge base and give professional and accurate answers according to the actual situation of the site, so that real barrier-free communication is realized.

The invention relates to the technical field of image target detection, and discloses a dark light environment small target detection method and system, and the method comprises the steps: collecting original image information; performing dark light enhancement on the original image; using the improved Zero-DCE network to carry out brightness adjustment on the image after the dark light enhancement; and adding a simAM attention mechanism in a backbone feature extraction module of the yov11, and inputting the enhanced image into the improved YOLOv11 network to obtain a small target detection result. A dark light enhancement training data set is not needed, the data acquisition difficulty is reduced, and the occurrence of dark light enhancement model overfitting in the training process is avoided. The model training speed and the reasoning speed are high, and the requirement for computing power resources of equipment is low. Each pixel of the image has a brightness enhancement curve of the data, and the curves are obtained in training, so that the enhanced image is smoother.

The invention relates to the technical field of meteorological element forecasting, in particular to a meteorological element forecasting product intelligent generation method and system. The method comprises the following steps: acquiring meteorological element original data, and performing data assimilation on the meteorological element original data to obtain assimilated product data; combining the meteorological element original data and assimilation product data to obtain a forecast sensitive factor and site feature data; establishing a meteorological element product data set according to the meteorological element original data, the assimilation product data, the forecast sensitive factors and the site feature data; constructing a meteorological element forecast product intelligent generation model, and training and verifying the meteorological element forecast product intelligent generation model by using the meteorological element product data set; and generating a meteorological element forecasting product through the trained meteorological element forecasting product intelligent generation model. According to the method, the meteorological element original data is utilized, and the problem of efficient, accurate and intelligent generation of the meteorological element forecast product of the meteorological station is solved.

The invention provides a pipeline defect assessment method and system based on multi-source data fusion, and the method comprises the following steps: collecting the data of the operation state of a pipeline body through a related collection device, and marking the state and fault type of the pipeline, so as to construct a training set and a test set; and preprocessing the static data of the pipeline, wherein the preprocessing comprises data cleaning, standardization and feature extraction. And optimizing the extreme learning machine by using a bird flock algorithm, analyzing static data and evaluating pipeline defects. Preprocessing the image data of the pipeline body, wherein the preprocessing comprises image conversion, enhancement and denoising; and evaluating the circumferential weld defect based on the pre-trained VGG16 model. And performing decision-level fusion on a defect evaluation conclusion obtained by static data analysis and circumferential weld image analysis by using a D-S evidence reasoning algorithm to obtain a final evaluation result. The method effectively improves the accuracy and reliability of pipeline defect evaluation, and has wide application prospects and important practical value.

The invention discloses a quantum state preparation method and device, equipment and a medium. The method comprises the following steps: setting a target quantum state to be prepared; setting a Hamiltonian, wherein the Hamiltonian is a negative value of an outer product of a left vector and a right vector of the target quantum state; constructing a variable component sub-line including training parameters according to the target quantum state; substituting a preset parameter value as a training parameter into the variable component sub-line so as to output a training quantum state evolved by the variable component sub-line; measuring an expected value of the training quantum state, which is evolved by the variable component sub-line, about the Hamiltonian; training a training parameter of the variable component sub-line through a classical optimizer until the training parameter or an expected value is converged to a set threshold value; comparing the training parameter or the expected value with a set threshold value; and outputting the training quantum state evolved by the variable component sub-line under the convergence condition as a to-be-prepared target quantum state. The preparation of any quantum state can be efficiently and accurately realized.

The invention relates to an innovative bearing heat treatment result prediction technology, and the powerful capability of a deep learning network is fully utilized. The method is characterized in that key parameters, such as quenching time, quenching temperature and other conditions possibly influencing a heat treatment result, of the bearing in the heat treatment process are integrated to serve as a feature set of input data, and the physical or chemical attribute value of the bearing after heat treatment is completed serves as a target variable of model prediction. According to the characteristic that data is high in time sensitivity, a long short-term memory (LSTM) network is selected to learn a complex mapping relation between input features and output results, time sequence dependency in the data is effectively captured, and time sequence features of the data are grasped. The trained model not only can provide an accurate prediction result for an operator, but also can help the operator to understand how different parameters affect the quality of a final product, so that product loss caused by experiments is reduced.

The invention provides a space debris orbit prediction model training method and a space debris orbit prediction method, relates to the technical field of space debris orbit prediction, and is used for solving the technical problem that in the prior art, target information is difficult to effectively extract from sparse data, so that space situation changes cannot be accurately perceived. The space debris orbit prediction model training method comprises the following steps: acquiring historical orbit data of space debris; according to historical orbit data, extracting position residual errors between the prediction state vectors and the reference state vector of the space debris at different moments; generating a multi-dimensional feature data set according to the position residual error; performing data dimension reduction processing on the multi-dimensional feature data set to obtain a sample data set after dimension reduction; and training an initial random forest model through a machine learning algorithm by using the sample data set after dimension reduction, so that the initial random forest model establishes a mapping relationship between the orbit parameters of the space debris and future states, and a space debris orbit prediction model is obtained.

The invention discloses a water supply network water leakage detection method based on an emergence phenomenon and federal learning, and relates to the technical field of intelligent monitoring and management of water supply networks. The method specifically comprises the steps of data acquisition and data preprocessing, emergence feature extraction, federated learning framework, quantum calculation acceleration, artificial intelligence model construction, real-time monitoring and early warning system, data-driven pipe network optimization, test and evaluation and the like. According to the method, efficient processing of data, accurate extraction of water leakage characteristics, rapid training and optimization of a model and collaborative analysis of multi-source data on the premise of protecting data privacy are realized, so that the precision and efficiency of water leakage detection are remarkably improved.

The invention relates to a risk assessment method and device for geological disasters along a power grid and computer equipment. The method comprises the following steps: constructing a sample database according to historical geological disaster data in a target area along a power grid; based on the at least two environmental influence factors and the sample database, constructing a training sample database, and based on preset machine learning method parameters, training the training sample database to obtain at least two geological disaster risk assessment models; comparing the geological disaster risk assessment model to obtain a target risk assessment model; and performing geological disaster assessment on the target area through the target risk assessment model. The method can improve the accuracy of risk assessment of geological disasters along the power grid; the personnel cost and the time cost are obviously reduced, and the risk assessment efficiency is improved.

The invention provides a distributed training method, device, equipment and system for a model and a computer program product, and the method comprises the steps: reading a historical networking link establishment file of a training cluster under the condition that a training task of a target model is submitted again; wherein the historical networking link establishment file is a file which is generated by the training cluster based on the historically submitted training task and is used for recording the communication relationship between the cards; and in the process that the target model carries out distributed training based on the training task, controlling each card to carry out communication according to the historical networking link establishment file. In the distributed training process of the target model based on the training task, each card can be directly controlled to communicate according to the historical networking link establishment file, so that the model training is quickly carried out, and the time consumption of the distributed training is shortened.

The invention discloses a real-time topic tweet classification method and device based on small samples, and the method comprises the steps: S1, obtaining original tweet sample data, and marking the data; s2, dividing the original tweet data marked with the data into a training set and a verification set, and performing preprocessing; s3, establishing a word embedding model, inputting the preprocessed training set data into the model for training, and converting the labeled data into word vector representation; s4, establishing a classifier model, transmitting the training set data into the classifier model, and performing pre-training by taking a mask language modeling task as a target task; s5, obtaining tweet data containing each target topic specific keyword on each large social network site through a web crawler, and performing incremental training on the model; s6, inputting the verification set data into the text classifier to obtain a final topic tweet classification result; s7, repeatedly executing the incremental training process in the step S5 in real time by taking day as a unit; the problem that a traditional text classification method depends on large-scale marked data is solved.

The application relates to the technical field of artificial intelligence optical character recognition, in particular to an end-to-end communication big data journey card recognition method, and the steps of the method comprise the following steps: adjusting and correcting a text box according to size information of the text box, so that the size of the text box meets the required angle and proportion of adjustment; performing separated feature extraction on the corrected text box through a preset neural network to obtain feature data of a font; and obtaining text data corresponding to the feature data through a Bi-GRU recurrent neural network combined with an activation function.

The present invention discloses a method for predicting UAV conflicts based on a safety interval progressive decision tree, including: establishing a UAV safety interval parameter library; establishing a coordinate conversion model parameter library for converting WGS-84 coordinate system and GLONASS-PZ90 coordinate system to CGCS2000 coordinate system, and encapsulating the constructed Bursa model as a coordinate conversion function; parsing flight dynamic data and performing standardized conversion; predicting airspace UAV conflicts, and dividing the prediction results into a data set to be observed, a data set of key attention, and a data set of warnings according to the conflict risk level; calculating the time when the UAV pairs in the data set of key attention may reach the warning level; calculating the time when the conflicting UAV pairs in the warning data set are expected to collide. The present invention can maximize the utilization of airspace space resources, thereby effectively improving the flight flow in the airspace.

A fabric defect classification method and system fusing multi-scale dynamic convolution and graph embedding constraints, aiming at the problems that existing methods are difficult to model irregular defects, similar defects are easy to confuse, and small sample generalization is weak, hierarchical multi-scale features are extracted through Swin Transformer; then a dynamic multi-scale refining module is designed, different receptive field features are fused through context-aware dynamic convolution, scale-dependent is modeled through cross-scale collaborative attention, and high discriminative features are output; the refined features are input into an integrated deep random vector functional connection network, and a boundary-aware graph embedding constraint based on local Fisher discriminant analysis is innovatively introduced in the optimization target, the feature manifold structure is explicitly modeled by constructing same-class intrinsic graph and different-class penalty graph, and the intra-class compactness and inter-class separation degree are strengthened. The present invention significantly reduces the misjudgment rate of similar defects, and meets the needs of industrial real-time detection.

The invention discloses a natural leather defect segmentation method based on an MLP multi-layer neural network, comprising: obtaining a data set of leather images through an image acquisition module, dividing the data set into a training set and a test set; generating a leather color feature map according to different colors; using a feature vector as an input of the MLP multi-layer neural network, completing the construction of a leather defect segmentation model through feature training of the MLP multi-layer neural network; storing the model file; using a traditional image processing method to pre-process the leather image to be segmented, using the model to predict the leather image, and obtaining a defect segmentation map of the leather image; using a traditional image processing method to enrich the details of the defect segmentation map; extracting the outer contour and defect contour of the leather, and converting them into a DXF file. The invention is suitable for leather defect segmentation, can solve the technical problems of low segmentation efficiency and insufficient segmentation, reduce labor costs, and improve production efficiency.

The invention discloses an oil spill drift prediction method based on intelligent correction of an ocean dynamic field, and belongs to the technical field of oil spill drift prediction. A corrector and a discriminator update loss function values of the corrector and the discriminator respectively according to two scores; circularly and repeatedly iterating to obtain a trained model; using the trained model to correct the numerical forecasting ocean dynamic field during the oil spill accident to obtain a corrected ocean dynamic field; the oil spill drifting numerical prediction model takes the corrected ocean dynamic field as a drive to obtain an oil spill drifting prediction result; and continuously updating the corrected ocean dynamic field of the next date to obtain a drift prediction result of the next date. The confrontation time convolution network can effectively resist the interference of noise data or incomplete data, the calculation complexity is greatly reduced, and model training is completed more quickly; and the confrontation time convolution network performs efficient correction on the air flow field data, so that the accuracy of oil spill drift trajectory prediction is remarkably improved.

The application discloses a data processing method, device and equipment and a computer readable storage medium. The method comprises the following steps: obtaining sample voice information and sample text information; processing the sample voice information through an acoustic model in an initial recognition model to obtain acoustic feature information; processing the acoustic feature information and the sample text information through a translation model in the initial recognition model to obtain a first predicted translation result and a second predicted translation result respectively; training the initial recognition model based on the first predicted translation result and the second predicted translation result to obtain a target recognition model; and the K vector and the V vector in the acoustic model are spliced with a prefix vector and / or the K vector and the V vector in the translation model are spliced with a prefix vector. The training efficiency of the target recognition model is improved.

The application provides an RGB-D image feature collaborative fusion method based on transfer learning and width learning, and comprises the following steps: obtaining an RGB-D data set, performing preliminary training through a neural network, and performing retraining in the data set after modifying the structure; after feature extraction, performing correlation analysis and fusion on RGB image features and depth image features; and using width learning to classify and identify the fused features. The application can reasonably fuse the features of RGB images and depth images, ensure that the feature information of color images and depth images can complement each other, improve the running speed of the system by using width learning, and finally make the classification result have higher accuracy and reliability.

The present application relates to the field of text semantic understanding, and more particularly to a text abnormal word recognition method and system based on a neural language model, which is used for recognizing network language, new words and wrong words, and the method comprises the following steps: step 1) collecting text to be recognized; step 2) determining the context word sequence of each word in the text to be recognized in a sliding window manner; step 3) inputting the context word sequence of each word into a pre-established and trained recognition model respectively to obtain the probability of occurrence of each word under the context word sequence; and step 4) comparing the probability of occurrence of each word with a set threshold value to determine whether the word is an abnormal word. The present application does not need to mark abnormal words and can be trained on a large amount of unsupervised data; the prediction probability threshold value of abnormal words and normal words determined by using a large number of positive and negative samples to predict probability statistics has universality; and the selection of negative samples is performed in a negative sampling manner, so that the training speed is accelerated.

An alloy performance prediction method fusing vision-language-process multi-modal data comprises the steps that a multi-modal data set of an alloy material is constructed, and the multi-modal data set comprises three kinds of modal input including process parameters such as temperature and time of solid solution and aging treatment, SEM image visual information and SEM image description text information and corresponding alloy mechanical property true values; training a visual encoder ResNet50 model through comparative learning and training a language encoder BERT model through mask language modeling to obtain an SEM image and vector codes of language description of the SEM image; splicing and fusing process parameters such as temperature and time of solid solution and aging treatment and the vector codes obtained in the second step, and training a random forest regression device according to corresponding alloy mechanical properties; and the random forest regression device obtained through training is used for alloy performance prediction. According to the method, the structured process data, the unstructured SEM image data and the derived text description data are subjected to collaborative fusion and joint modeling for the first time, complementarity among different modal data is fully utilized, and more comprehensive and more three-dimensional digital representation of the alloy state is constructed; the multi-modal fusion framework and the feature learning mechanism are suitable for wide material systems. Meanwhile, the adopted'depth representation + random forest 'hybrid model has the advantage of high training efficiency while ensuring high prediction precision.

The application provides a camera scheduling model training method, a violation event detection method and device. The camera scheduling model training method comprises: obtaining historical sample data of a camera; constructing a first sample probability distribution based on the historical sample data; sampling based on the first sample probability distribution to obtain first sample action data; and training a reinforcement learning model based on the first sample action data to obtain a camera scheduling model. The application uses historical sample data to construct a sample probability distribution, and then uses the sample probability distribution for sampling as the input of the reinforcement learning model to train the reinforcement learning model. Since the input of the reinforcement learning model is generated by the constructed sample probability distribution and does not truly respond to the environment, the problem of long time consumption for obtaining real samples is solved, a large number of samples can be quickly obtained for training, and the training efficiency of the camera scheduling model is improved.

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training an action selection system using reinforcement learning techniques. In one aspect, a method includes, at each of a plurality of iterations: obtaining a batch of experiences, each experience tuple including: a first observation, an action, a second observation, and a reward; for each experience tuple, determining a state value for the second observation, including: processing the first observation using a policy neural network to generate an action score for each action in a set of possible actions; sampling a plurality of actions from the set of possible actions according to the action scores; processing the second observation using a Q neural network to generate a Q value for each sampled action; and determining the state value for the second observation; and using the state value to determine an update to current values of Q neural network parameters.

The invention discloses a CNN-LSTM-based high-voltage user power consumption data prediction method, which takes high-voltage user power consumption data as a research object, cleans historical power consumption data of high-voltage users by using an LOF method, removes unreasonable outlier data, and reduces the influence of the outlier data on the performance of a prediction model. Through data classification and feature engineering, original data are converted into a more representative feature set, which is convenient for model learning. And the volatility of the data is reduced through data normalization, so that the stability and convergence speed of model training are improved. And then constructing a CNN-LSTM hybrid model, that is, adding a convolutional neural network CNN on the basis of a long short-term memory network LSTM. The hybrid model is composed of a CNN part and an LSTM part, the CNN part is used for extracting local features of data, the LSTM part is used for capturing a dependency relationship of a time sequence, and then more accurate power consumption prediction is carried out. After the construction and training of the model are completed, the prediction result of the model is effectively and intuitively evaluated through a mean square error and a graphical method.

The invention discloses a base station energy-saving decision-making method for carrying out memory backtracking on DT based on state machine transfer, a state machine memory module with a backtracking function is designed to be used for a Decision Transformer to carry out long-term stable decision-making on base station energy saving, and the method comprises the following steps: recording past information such as states, actions and environment rewards, and arranging the information into a time sequence according to a specific sequence; accumulating the environment rewards in the trajectory, and calculating an accumulated expected return (return-to-go) in the trajectory; the method comprises the following steps: constructing a decision model, and taking a Decision Transform as a main body model; designing a state machine memory module and a state machine backtracking module; determining a loss function and training a decision model, obtaining a model meeting requirements as the decision model by adjusting a learning rate, an optimizer and regularization parameters, and storing the decision model; and inputting environment historical data, and outputting decision actions by the large model. According to the method, the problem that a traditional offline reinforcement learning method Decision Transform suddenly loses efficacy in a long-term decision that the reasoning track length exceeds the training sample track length is solved, and the stability of the energy-saving decision of the base station is improved.

The intelligent automobile lane-changing decision method and device, the electronic device and the storage medium provided by the embodiments of the present disclosure comprise: constructing a virtual lane-changing scene by using a simulation method; constructing a decision model for determining a mapping function between a state and a reward, obtaining a Q table for evaluating the lane-changing feasibility of an intelligent automobile based on the mapping function, and obtaining a lane-changing decision, i.e., "lane-changing" or "not lane-changing"; based on a greedy algorithm, making the intelligent automobile produce a random lane-changing behavior in the constructed virtual lane-changing scene, storing the state and the reward of the current lane-changing behavior in an experience replay pool, when the number of state-reward pairs in the experience replay pool reaches a minimum sample number, randomly sampling a plurality of state-reward pairs from the experience replay pool for single-step training of the decision model, storing a new state-reward pair generated in the training process in the experience replay pool, repeatedly performing the above training process until a maximum training number is reached, and obtaining a lane-changing decision model. The lane-changing decision generated by the present disclosure is high in safety and wide in applicability.

The invention discloses an asynchronous federal learning method based on grouping management, and the method comprises the steps: uploading a local model parameter w to a server after a client completes local training; after the server receives the local model parameters w uploaded by the clients in a specified number group, the server performs aggregation based on the uploaded local model parameters w, updates the global model parameters W and completes the global training of the current round; the server updates the number M of groups participating in the global training in the next round, and issues the updated global model parameter W to the client of the uploading local model; and updating the local model parameter w into the received global model parameter W by the corresponding client, and starting the next round of local training. A plurality of groups of clients are dynamically selected for aggregation in each round, grouping is increased and differences are averaged in the early stage of training, the phenomenon of model imbalance caused by aggregation of rapid clients by the clients in asynchronous training is improved, grouping is reduced in the later stage of training, the aggregation speed is increased, and unnecessary time consumption caused by model equalization is reduced.

This invention discloses a substation acoustic signature fault diagnosis method based on LSTM-BHPSO. The method involves collecting substation acoustic signature signals and extracting features using MFCC to obtain a feature dataset, which is then divided into training and testing sets. An LSTM-based substation acoustic signature fault diagnosis model is constructed. The optimal initial parameters are obtained by iteratively optimizing the particle positions using the number of feature points extracted by MFCC, the number of batch samples, the number of network layers, and the number of hidden and fully connected layers. These optimal initial parameters are then input into the LSTM-based substation acoustic signature fault diagnosis model. Finally, the trained LSTM-based substation acoustic signature fault diagnosis model is used to diagnose substation acoustic signature faults. This invention offers good processing efficiency, avoids local optima, and facilitates accurate fault diagnosis results.

The present invention relates to a method for detecting the position and size of a double-leaf plug sliding door of a train set, comprising the following steps: detecting the position of the upper middle seam; detecting the position of the upper left and right sides of the line; detecting the position of the left and right sides of the vertical bars; detecting the position of the lower left and right sides of the corners; detecting the position of the lower middle seam. The method for detecting the position and size of the double-leaf plug sliding door of a train set of the present invention can quickly solve the overall error change, the size is within a controllable range, and the air tightness test is ensured to pass smoothly, improve production efficiency, make the adjustment controllable and standardized, and can be used to quickly train learning personnel; it can quickly solve the problem of poor sealing of various parts of the double-leaf plug sliding door, fundamentally clarify the detection method, integrate the beauty of craftsmanship and manufacturing into production, achieve consistency in production process and results, thereby improving the ability to handle the air leakage points of the plug sliding door during the air tightness test, so that the air tightness test can be passed smoothly, and the product quality is improved.

The embodiment of the invention provides a reliability detection and model automatic training method for a sealing structure of a wellhead. The method comprises the steps of obtaining a size data set of the sealing structure; the size data set comprises a plurality of size data; the size data represents information of a physical structure of the sealing structure; performing matching processing on the size data set and a data set in a preset model library to obtain multiple groups of matched data sets; determining an initial model corresponding to the matched data set based on a preset model library; based on a Bayesian optimization algorithm, processing the size data set and each initial model to obtain a reliability detection model; the reliability detection model is used for processing the size data of the sealing structure to obtain a reliability detection result of the sealing structure. The method is used for achieving the effect of accurately and effectively detecting the reliability of the sealing structure.