8 results about "Time series classification" patented technology

This invention discloses a lightweight adaptive optimization method for high-dimensional hyperparameters of convolutional neural networks (CNNs) for non-stationary time-series signal classification. It aims to address the technical challenges of performance degradation in time-series signal classification models and the reliance on expensive real-world evaluations for hyperparameter configuration under non-stationary perturbation scenarios. This method uses a deep convolutional neural network as the core classification carrier, treating the hyperparameter combinations within the deep convolutional neural network as decision variables to be optimized. With robust classification error rate, computational complexity, and training time as core optimization objectives, it constructs a closed-loop collaborative optimization mechanism of "perception-evaluation-decision" and utilizes a meta-learning dual-branch convolutional polynomial surrogate-assisted evolutionary algorithm (MetaDCP-SAEA) to achieve efficient configuration. This method requires no manual intervention; the convolutional neural network used for classifying non-stationary time-series signals can automatically search for the optimal hyperparameter combination. In simulated non-stationary noise environments, the reduction in classification accuracy can be controlled within 9.17%. It is suitable for robust classification scenarios of non-stationary time-series signals such as industrial IoT monitoring and medical signal diagnosis. It helps to lower the engineering threshold of artificial intelligence technology, promotes the large-scale application of automatic machine learning in complex environments, and has broad market prospects and application value.

This invention discloses a multivariate time series classification method based on wavelet-enhanced dual-branch fusion. Addressing the issues of coexistence of multi-scale periodicity and transient abrupt changes in time series and susceptibility to non-discriminatory perturbations such as baseline drift, this method designs an adaptive periodicity discovery mechanism guided by stationary wavelet to identify the dominant period, and constructs a heterogeneous dual-branch encoder module to model frequency-domain periodic patterns and time-domain transient dynamics respectively. Furthermore, a symmetric mutual-enhancing attention module is designed to achieve interaction between dual-domain features. Finally, a graph neural network classification head based on Kendall's rank correlation coefficient is proposed to achieve end-to-end training and effectively capture morphological similarities between samples. This invention significantly improves classification robustness and accuracy while maintaining model interpretability, making it suitable for high-reliability scenarios such as medical monitoring and industrial equipment diagnostics.

This invention provides a multivariate time series classification method based on a hierarchical image stream multimodal large model, comprising the following steps: acquiring multivariate time series data; converting the multivariate time series data into an image stream visual sequence and a statistically driven text summary through a multimodal enhancement module; encoding the multivariate time series data through a time series source module to extract inherent temporal features as time center anchors; progressively aligning and fusing the image stream visual sequence, the statistically driven text summary, and the temporal features through a hierarchical fusion module to obtain a multimodal fused representation; inputting the multimodal fused representation into a classifier to output the classification result. This invention transforms multivariate time series into an image stream visual sequence and a statistically driven text summary, and achieves progressive alignment and fusion of visual, text, and temporal modalities through a hierarchical co-encoding mechanism, achieving efficient and accurate multivariate time series classification under long sequence and limited annotation conditions.

This invention discloses a non-stationary multivariate time series classification method based on hierarchical constrained multi-domain graph networks. It constructs a collaborative adaptive multi-domain graph network model, SAMGNet, and addresses the problems of incomplete single-domain representation information, insufficient graph structure adaptation capability, and weak robustness under distribution shifts in non-stationary multivariate time series classification through a progressive collaborative mechanism involving multi-domain time series graph construction, multi-domain collaborative fusion encoding, adaptive graph structure modeling, and multi-level consistency comparison learning. The method synchronously maps time series signals to the time-frequency domain, evolution domain, and phase angle domain, and achieves cross-domain semantic alignment through dual-path encoding based on difference perception and consistency constraints, and bidirectional collaborative fusion. This invention constructs an adaptive adjacency matrix by integrating static topological priors and sample-level dynamic correlations. It also tracks the dynamic evolution of non-stationary temporal dependencies through multi-scale message propagation. By constructing a stable semantic space robust to distribution shifts through dual consistency constraints within and across layers, the proposed method achieves a classification accuracy of 98.26% on the UORED-VAFCLS industrial bearing fault diagnosis dataset and 65.66% on the ADFTD clinical EEG dataset. This performance surpasses current mainstream baseline methods and is suitable for non-stationary multivariate temporal classification scenarios such as industrial equipment condition monitoring and medical physiological signal analysis, showing promising application prospects.

The application relates to the technical field of computer data processing, and discloses an RNN time sequence classification method based on a deep variational information bottleneck, which comprises the following steps: acquiring multivariate time sequence data, processing the multivariate time sequence data through a recurrent neural network to generate a complete hidden state sequence; based on the sequence, running an optimal time segmentation algorithm to calculate optimal segmentation points, and then dividing the sequence into two parts of an early stage and a late stage; performing information bottleneck coding on the aggregated representation of the early stage sequence and the final hidden state of the complete sequence respectively to generate first and second latent representations; after fusing the two latent representations, processing through a decoder network to obtain a final classification result; calculating a composite loss function according to the classification result and a real label, and updating all trainable parameters in the model through a back propagation algorithm. The application adopts a data-driven optimal time segmentation algorithm, can adaptively identify and segment key time points at which information characteristics in the sequence change, and improves the pertinence of feature extraction.

Embodiments of the present application provide a time series classification prediction method and device, a server and a computer readable storage medium, relating to the technical field of machine learning. A time series feature sequence is input into a time series classification model for processing to obtain a time series classification result corresponding to each time point; the time series classification model obtains a model input vector by concatenating an original feature vector of each sample time series feature sequence at a time point and a feedback input vector at a previous time point, and inputs the model input vector into a pre-constructed initial time series classification model for training; the feedback input vector is selected from a real label and a model prediction value at the previous time point by a sampling probability corresponding to a current training round, the sampling probability representing a probability of taking the real label at the previous time point as the feedback input vector, and the sampling probability dynamically decays with an increase in the training round, so as to solve the exposure bias problem and improve the accuracy and stability of the time series classification result.

A real-time data-driven online intelligent geological sensing method and system includes the following steps: (1) establishing a database of MWD parameters and corresponding core samples for standardizing the data processing flow; (2) constructing a time series classification model to establish a mapping relationship between MWD parameters and the excavated lithology in a reasonable and reliable manner; (3) converting the time series classification algorithm into an online learning version to make it more efficient and scalable, and continuously learning from the input observations to iteratively optimize the classification model. By implementing the above method, the ultimate goal is to improve the efficiency and safety of drilling operations while enhancing the intelligence level of HDD.

This invention discloses a generative statistical semantic guidance processing method for time series classification. Addressing the issues of attention diffusion in long sequences and overfitting in small samples, this method projects multivariate time series to generate morphological token sequences. It extracts high-order statistical features such as skewness and kurtosis from the tokens to construct statistical semantic cue vectors, calculates importance scores based on these vectors, and performs Top-K soft sparse reweighting. Simultaneously, redundant tokens are adaptively compressed into global context features. Finally, the fused features are input into a multi-scale conditional expert routing module for dynamic calculation, and a large temporal model is introduced during training for manifold distillation. This invention effectively eliminates computational redundancy while preserving global coherence without loss, significantly improving classification accuracy and robustness in small sample and noisy environments.