9 results about "Classifier fusion" patented technology

The invention relates to the technical field of nuclear power plants, in particular to a nuclear power equipment operation defect grading method and device, electronic equipment and a storage medium. The invention provides a nuclear power equipment operation defect grading method, and aims to realize grading of nuclear power equipment operation defects through a systematic and multi-level fusion decision-making mechanism. The problems that rule grading lacks flexibility, graph grading is influenced by keyword sensitivity, large model grading field knowledge is insufficient, and precision and adaptability are limited due to the fact that a traditional fusion strategy is single in the prior art are effectively solved. The core of the method is that lexical meta processing and a value matrix mechanism are introduced, dynamic and fine-grained fusion of multi-base classifier output is achieved, and therefore the overall performance of a grading system is improved. Through lexical element processing, value matrix construction and dynamic updating and multi-classifier fusion decision making, the nuclear power equipment defect grading method with high precision, high adaptability and high reliability is realized, and many limitations in related technologies are effectively solved.

The application provides a car data document classification method and device, equipment and medium, relates to the technical field of data document classification, and obtains a text feature vector, a similarity feature vector and an initial classification feature vector corresponding to a to-be-classified car data document; generates a fusion classification feature vector corresponding to the to-be-classified car data document according to the text feature vector, the similarity feature vector and the initial classification feature vector; inputs the fusion classification feature vector and the similarity feature vector into a preset document classification model obtained by training a multi-Fisher classifier according to the fusion classification feature vectors of historical car data documents and corresponding classification labels; and obtains a classification result of the to-be-classified car data document according to an output result of the preset document classification model. The car data document classification model of the multi-classifier fusion is constructed, so that the generalization ability of the traditional classification model is improved, and the classification precision and efficiency are improved.

The embodiment of the invention provides a generative classifier robust retraining method and system for medical federated learning, and relates to the technical field of machine learning, and the method comprises the steps: each participating client generates an adversarial network based on local data training federated conditions, learns global data distribution, and generates intermediate layer features; implementing a knowledge transfer mechanism, and transferring the feature extraction capability of the global classification model to a local discriminator; generating an adversarial network based on the trained federal conditions, and generating a class-balanced synthetic feature data set at a server side; and a feature fixed retraining strategy is adopted, and only the classifier part of the global model is retrained. According to the method, a federated condition generative adversarial network is utilized at a server side to generate class balanced intermediate features, a classifier specially used for retraining a global model is adopted, and the condition generative adversarial network and a knowledge transfer mechanism are fused, so that high-quality balanced feature representation can be generated according to global data distribution; and the problems of non-independent identical distribution and long-tail data distribution are effectively solved.

A direct current partial discharge signal time domain feature classification method, system and device based on ensemble learning and a medium, the method comprises the following steps: preprocessing a partial discharge original signal, and dividing the partial discharge original signal into a training set and a test set according to a preset proportion; using a voting classifier to fuse the limit gradient lifting, the random forest and the support vector machine into an integrated model through a soft voting strategy; inputting the training set into the integrated model for training, and performing performance evaluation on the trained integrated model by using the test set; preprocessing a to-be-detected partial discharge original signal, inputting the to-be-detected partial discharge original signal into the trained integrated model, and obtaining a classification result in combination with a soft voting strategy; a training set and a test set with high signal-to-noise ratio and high feature representation capability are constructed through preprocessing, and the training set and the test set are utilized to perform training and performance evaluation on the integrated model fused by the voting classifier, so that the integrated model can accurately predict classification results of multiple types of typical insulation defects of the to-be-detected partial discharge original signals.

The invention provides a rolling bearing small sample open set cross-working-condition fault diagnosis method based on graph learning, and relates to the technical field of rotating machine fault diagnosis. According to the invention, through constructing a dynamic adaptive graph convolution multi-classifier fusion network, designing an unknown fault category discrimination joint strategy and adopting an N-way K-shot element learning mode, the model can realize effective identification of known fault types across working conditions and can effectively identify unknown fault types under the constraint of small samples, so that the method has the advantages of high efficiency and high reliability. Therefore, the accuracy of fault diagnosis of the rolling bearing is improved.

This invention provides an ECG signal classification system based on deep learning neural networks. It includes a common feature extraction module using a convolutional neural network (CNN) to extract shallow features from ECG signals; a domain-invariant branch module using a CNN to extract features from the shallow features and perform feature matching for the source domain by minimizing the classification loss, obtaining a prediction; a domain-specific branch module using a CNN with LMMD loss to perform feature matching between each pair of source and target domains for the input shallow features, obtaining n predictions; and a fusion module assigning different confidence levels to each sub-branch and combining the n+1 predictions to obtain the final prediction. The multi-branch network and multi-source unsupervised domain adaptation of this invention effectively learn the general features of ECG signals, enabling the knowledge learned from the source dataset to be applied to the unknown target dataset. Furthermore, by using a prior classifier-based fusion strategy, multiple predictions are organically combined to obtain the final result, further improving the model's generalization performance.

This invention discloses a method for ultra-narrow pulse radar target recognition using prior knowledge-assisted multi-classifier fusion, belonging to the field of radar target recognition technology. This method introduces prior conceptual information about the target into the automated construction process of a tree-like hierarchical structure, compensating for the knowledge deficiencies of data-based automated construction methods. The difficulty of distinguishing between classification nodes in the constructed tree structure is significantly reduced. Therefore, even in recognition scenarios with numerous categories and insufficient data, this invention's method still exhibits good performance in refined target recognition.

The invention provides a mobile malicious software detection method based on heterogeneous flow fusion, realizes more effective capture of malicious behaviors in combination with semantic information of heterogeneous flows, and belongs to the technical field of network security. The method comprises the following steps: firstly, explicitly modeling a relationship between entities from different flows by adopting a heterogeneous information network (HIN), and retaining semantic relevance of heterogeneous flows; secondly, a meta-path group is constructed for each stream view, each group comprises content-oriented meta-paths and action-oriented meta-paths, and semantic correlation between applications is established; then, the flow2vec distinguishes HIN entity semantics of different streams based on context constraints; and finally, fusing semantic embedding of each view through a DNN classifier based on channel attention, and weighting contribution of the semantic embedding to realize accurate detection. According to the method, semantics of multiple stream types are integrated, the complementary advantages of the semantics are expected to be utilized, understanding of application program behaviors is enhanced, and malicious information hidden in application program codes is found.

The application provides a downhole drilling working condition recognition method and device based on data driving. The method comprises the following steps: before downhole drilling, the importance of different working condition characteristics is quantified by introducing contribution value, and the output of the working condition classifier of each working condition is subjected to explainability analysis; according to the contribution value of different working condition characteristics to each working condition, a judgment matrix of each working condition is constructed, thereby obtaining a weight vector of each working condition, and an objective weight distribution method driven by data is established; during downhole drilling, downhole measurement while drilling data is collected, and downhole drilling parameter data can be directly obtained; according to the weight vector of each working condition and the working condition index value of each working condition, the working condition of each window is decided through multi-classifier fusion; when a preset number of different working conditions is cumulatively output, the feature vector of each window and the working condition index threshold of each working condition are updated, and a double dynamic updating strategy is established to adapt to the complex and changeable drilling environment. The method provided by the application improves the accuracy and real-time performance of working condition recognition.