86 results about "Nonlinear feature extraction" patented technology

The invention provides an underground water environment automatic monitoring super station state supervision method and system. The method comprises the following steps: collecting a multi-source dynamic time sequence data set and carrying out time-space alignment; generating a dynamic characteristic index set by using a nonlinear dynamic characteristic extraction method; based on the index set, constructing a self-adaptive cooperative measurement network to perform anomaly monitoring, and generating an anomaly detection result and a cooperative control instruction; generating real-time monitoring and early warning information by using a dynamic threshold adjustment algorithm; and generating an adaptive control instruction and a dynamic resource allocation scheme by using a neural network adaptive control strategy and a resource scheduling optimization algorithm. According to the method, a dynamic characteristic index set is generated, a C-C method is adopted to reconstruct a high-dimensional phase space, a Wolf algorithm and a G-P algorithm are combined to calculate related indexes and dimensions, and a multi-scale fractal mode is analyzed through an R / S analysis method and wavelet transform. The methods capture complex dynamic behaviors of the underground water system, and solve the problem that the traditional method is insufficient in non-linear feature extraction capability.

The invention relates to a non-linear feature extraction-based roadside end radar and camera fused three-dimensional target detection method and device, and a medium. The three-dimensional target detection method comprises the steps of synchronously completing data enhancement and downsampling of an image and a point cloud; extracting high-dimensional nonlinear features by using an encoder improved by a Kolmogov-Arnod network, and projecting the high-dimensional nonlinear features to a unified aerial view space; establishing cross-modal dependence by using multi-head cross attention; carrying out weight fusion by using nonlinear convolution to generate an integrated aerial view feature map; and the decoder and the detection head output the three-dimensional coordinate, the size and the category of the target. Compared with the prior art, the Kolmogorov-Arnod network is introduced to perform nonlinear enhancement on the image and the point cloud encoder, the cross-modal weight is dynamically calculated through multi-head cross attention in the aerial view space, and finally, the weighted fusion is completed by using the KANs convolution. Therefore, the three-dimensional detection precision in a complex traffic scene is remarkably improved while the consistency of the receptive field is ensured.

The invention discloses a steam turbine fault diagnosis method and system based on a knowledge graph, and the method comprises the following steps: S1, synchronously collecting the vibration signal, the iron abrasive particle concentration, the oil viscosity and working condition parameters of a steam turbine bearing in a steam turbine, and relates to the technical field of steam turbine fault diagnosis. In order to solve the problems that a turbine bearing serving as a core rotating part is prone to abrasion faults, and early abrasion such as surface microcracks and oil film degradation concealment is high, vibration signals, iron abrasive particle concentration, oil viscosity and working condition parameters are synchronously collected to generate multi-source data streams with aligned timestamps, nonlinear feature extraction is combined, and a multi-source data stream with aligned timestamps is obtained. By constructing the initial knowledge graph, mapping the extracted features to the initial knowledge graph and generating the dynamic knowledge graph based on the graph neural network, time sequence reasoning of the bearing abrasion state is achieved, so that the early abrasion features of the turbine bearing can be obtained, and early warning is conducted according to the early abrasion features of the turbine bearing.

The invention discloses a large model dynamic batch processing method based on sequence splicing. The method comprises the following steps: receiving input sequences of a plurality of users in a batch; converting the input sequence into a corresponding token sequence; carrying out heterogeneous splicing on all token sequences along a sequence length dimension to form a unified joint token; the spliced joint tokens pass through a normalization layer, standardization operation is executed on the joint tokens, and data distribution is unified; carrying out linear projection on the joint token through a shared linear transformation layer to generate a joint query vector, a joint key vector and a joint value vector, and splitting the joint query vector, the joint key vector and the joint value vector into sub-vector groups corresponding to each user; executing multi-head attention calculation to obtain attention output of the user; and carrying out linear transformation on the attention output, and inputting a transformed result into a shared MLP to carry out nonlinear feature extraction and enhancement so as to obtain a final output corresponding to each user. According to the method, the problems of efficiency bottleneck and resource consumption when a large model processes mass data are effectively solved.

The invention relates to the field of computer big data, and discloses a self-adaptive parallel computing high-dimensional data dimension reduction and classification optimization system which comprises a quantum tensor decomposition module, a dynamic routing module, a heterogeneous computing classification module and a feedback regulation and control module. The quantum tensor decomposition module realizes nonlinear feature extraction of high-dimensional data through super-adjacency tensor modeling and entanglement entropy constraint; the dynamic routing module generates an adaptive communication path based on pulse time sequence weight optimization and quantum key verification; the heterogeneous calculation classification module dynamically schedules a quantum processor according to data characteristics, and FPGA and GPU resources execute mixed gradient aggregation; and the feedback regulation and control module guarantees the robustness of the system through multi-dimensional parameter closed-loop correction and a multi-stage fault-tolerant mechanism. All the modules cooperate to form a closed loop of dimension reduction, transmission, classification and regulation, the problems of feature extraction distortion, low resource utilization rate and poor dynamic environment adaptability in the prior art are solved, and the real-time performance and reliability of high-dimensional data processing are remarkably improved.

The invention discloses a nuclear power plant equipment fault diagnosis method, and the scheme can comprise the steps: firstly, collecting normal and fault data of 100% and 80% working conditions, and covering a plurality of crevasse fault types; state monitoring is carried out through KPCA, data are mapped to a high-dimensional space in a non-linear mode through a kernel function, and non-linear features are effectively captured through kernel matrix calculation, feature decomposition and anomaly judgment based on T2 and SPE statistics. A CNN-Bi-LSTM model is adopted in the fault diagnosis link, the CNN extracts local space features, and the Bi-LSTM captures bidirectional time sequence dependence; for a small sample working condition, model parameters trained by a source working condition (sufficient data) are migrated to a target working condition through migration learning, and fine adjustment adaptation is carried out. According to the scheme, nonlinear feature extraction, space-time modeling and transfer learning are fused, the accuracy and robustness of fault recognition under complex working conditions are improved, and the method is particularly suitable for data imbalance and sample scarcity scenes.

The invention discloses an intelligent tea mat tasting system based on multispectrum and deep learning, and belongs to the technical field of artificial intelligence, the system comprises a tea mat body, a multispectral sensor, a PLC and a terminal device, the multispectral sensor is arranged at the bottom of the tea mat body and comprises two light sources and two light sensing elements, and the PLC is connected with the terminal device; the PLC receives light signals, converts the light signals into electric signals and transmits the electric signals to the terminal device, and the terminal device comprises a multi-channel signal processing module, a peak value feature extraction module, a chaotic dynamics analysis module, a quality mapping module, a multi-standard comparison module, a probabilistic reasoning module, a recommendation generation module and a display module. A tea soup quality model is established through nonlinear feature extraction of multi-channel spectral signals in combination with chaos dynamics analysis, and tea quality evaluation is realized by utilizing three-dimensional flavor topological mapping and multi-standard library comparison, so that objective quantitative evaluation of tea quality is realized.

The invention discloses a method for detecting the content of salidroside in a rhodiola rosea extracting solution based on artificial intelligence. The method comprises the following steps: S1, performing spectral scanning of different wavebands on an extracting solution sample in a plurality of near-infrared wavelength ranges; s2, performing multi-dimensional interference suppression processing and signal purification; s3, carrying out spectral feature compression calculation to obtain a salidroside feature intensity factor; s4, carrying out salidroside characteristic signal-to-noise ratio enhancement and purity index calculation; s5, predicting the content of salidroside and outputting a content prediction score; and S6, mapping the content prediction score into the final salidroside content. According to the method, spectral signal purification, nonlinear feature extraction, dynamic noise enhancement, artificial intelligence prediction modeling and a secondary recheck mechanism are utilized to realize adaptive analysis of extracting solutions from different sources, so that the detection accuracy and stability are remarkably improved, and rapid and traceable salidroside content detection can be realized under the condition that a large precise instrument is not needed.

The invention discloses an intelligent positioning method and system for a microseism event. Intelligent positioning of the microseism event is realized by combining a convolutional neural network and probability density function mapping. Comprising the following steps: according to a work area speed model, generating four-dimensional training data representing detector coordinates and seismic phase arrival time and a three-dimensional label representing a seismic source position through Poisson disk sampling and a Gaussian probability density function; designing a convolutional neural network structure, optimizing network weight through back propagation, and constructing a micro-seismic event intelligent positioning model; and inputting actual data into the positioning model to obtain probability density distribution of the seismic source in a three-dimensional space, and finally outputting a high-precision seismic source positioning result through peak value extraction. According to the method, the observation system is creatively integrated into network input, and the strong nonlinear feature extraction capability of the convolutional neural network is combined, so that the generalization capability of the positioning model to different observation systems and the robustness to detector coordinate deviation and arrival time error are remarkably improved.

The invention discloses an anomaly detection method and device based on the receiving power of a low-orbit satellite-borne GNSS receiver, and the method comprises the steps: obtaining a source receiving power sequence of the low-orbit satellite-borne GNSS receiver, and carrying out the enhancement processing based on Doppler frequency shift compensation and ionospheric noise filtering, and then generating a to-be-detected receiving power sequence; performing feature extraction on the to-be-detected receiving power sequence and then generating a receiving power feature vector based on a multi-scale entropy feature and a chaotic feature; and comparing the received power feature vector with a pre-constructed dynamic anomaly threshold to generate an anomaly detection result. According to the method, time sequence mismatching caused by frequency shift is overcome, energy distortion caused by ionosphere disturbance is also remarkably weakened, so that the accuracy and sensitivity of a subsequent nonlinear feature extraction process are guaranteed, and particularly, the robustness and effectiveness of GNSS receiving power anomaly detection can be remarkably improved in a complex dynamic space environment.

The embodiment of the invention relates to the technical field of computer security authentication, in particular to a security authentication method and system based on a secure computer. The method comprises the following steps: carrying out sensing array deployment and data acquisition on a secure computer to obtain calibrated time-space synchronization data; performing temperature and current fusion processing on the calibrated time-space synchronization data to obtain a thermoelectric coupling characteristic spectrum; carrying out point location time sequence correlation analysis on the thermoelectric coupling characteristic spectrum to obtain a point location cross-correlation matrix; performing nonlinear feature extraction on the point location cross-correlation matrix to obtain a nonlinear dynamic feature set; performing dynamic behavior analysis on the nonlinear dynamic feature set to obtain a security calculation behavior dynamic phase spectrum; and performing phase difference calculation on the dynamic phase spectrum of the security calculation behavior to obtain a phase difference vector. According to the method, the early detection capability of threats which are difficult to reproduce and are in a novel hardware level is improved through real-time monitoring and anomaly recognition of computer microcosmic physical characteristics.

The invention provides a multi-scale context aggregation and dynamic supervision medical image segmentation method and application thereof, and belongs to the technical field of medical image processing. In order to solve the problems of weak non-linear feature fitting ability, global context missing and unstable training convergence in the prior art, the ResUKAN + network is constructed. According to the method, a residual KAN convolution module is embedded in a full level of an encoder, and nonlinear feature extraction is enhanced by using a B-spline function; a multi-scale context aggregation module is arranged on a bottleneck layer, and dynamic pyramid pooling and a double attention mechanism are fused to capture global dependency; a dynamic auxiliary supervision head is introduced at the tail end of a decoder, complementary features are extracted through a heterogeneous receptive field, and loss calculation is optimized in combination with a dynamic weight mechanism which is exponentially attenuated along with a training period. According to the method, the segmentation precision and robustness of the fuzzy boundary and the multi-scale focus are remarkably improved, and the method is suitable for medical image intelligent diagnosis.

The invention belongs to the technical field of data science and physical field simulation, and discloses an AUV structure stress real-time prediction method based on a convolution auto-encoder. The method comprises the following steps: carrying out nonlinear feature extraction on high-dimensional structure stress field data by adopting a trained convolution auto-encoder to obtain low-dimensional representation; establishing a mapping relation between an input working condition and the low-dimensional representation by adopting a deep learning neural network, and predicting a low-dimensional feature under a new working condition; and reconstructing the predicted low-dimensional features by using a decoder part of the convolutional auto-encoder to obtain high-dimensional structure stress field data. According to the method, the nonlinear characteristics of the flow field can be adaptively extracted, and the accurate mapping relation between the input working condition and the full-field stress response is established, so that high-precision and high-efficiency real-time prediction of the stress field of the AUV structure is realized, and the calculation efficiency and the model generalization ability are remarkably improved.

The invention belongs to the technical field of electromechanical system simulation, and particularly relates to an electromechanical simulation model evaluation method based on AI analysis, which comprises the following steps: data preprocessing, AI evaluation engine construction, dynamic index evaluation and closed-loop optimization. An electromechanical simulation model evaluation system based on AI analysis comprises the following layered architecture: a data preprocessing layer, an AI evaluation engine layer, a dynamic evaluation layer and a closed-loop optimization layer, one of kernel principal component analysis or a variational auto-encoder is introduced to carry out nonlinear feature extraction, and game Nash equilibrium is introduced to carry out model weight dynamic allocation. And sliding window anomaly detection is introduced to dynamically adjust the threshold. According to the invention, a'physical rule + AI reasoning 'dual-drive evaluation architecture of the electromechanical simulation model is initiated; the dynamic weight distribution algorithm improves the evaluation efficiency compared with a traditional method; and the three-level indexes converge step by step, so that comprehensive evaluation from local performance to global performance is realized.

The application relates to a code detection method and device, computer equipment and a storage medium. The method comprises the following steps: obtaining a target code to be detected; performing string division on the target code to obtain a target substring sequence; obtaining a substring vector corresponding to each target substring in the target substring sequence to form a vector sequence; performing linear feature extraction on the vector sequence to obtain first extracted features, and performing nonlinear feature extraction on the vector sequence to obtain second extracted features; performing fusion processing on the first extracted features and the second extracted features to obtain fusion features; and performing code detection based on the fusion features to obtain a code detection result corresponding to the target code. A cloud server can use an artificial intelligence-based malicious code detection model to implement the code detection method of the application, thereby achieving the purpose of reducing network attacks. The method can improve the accuracy of malicious code detection.

The invention relates to a data detection method for a distributed photovoltaic system and an energy storage system thereof, and relates to the technical field of photovoltaic energy storage data detection. According to the method, a high-precision distributed photovoltaic and energy storage system data detection model is constructed by combining the powerful nonlinear feature extraction and state reconstruction capability of the deep belief network with the precise setting of key parameters by a second-order oscillation firefly optimization algorithm. Therefore, the system can predict the key operation state of equipment or positions where sensor nodes are not directly installed according to the actually measured data of a limited number of intelligent fusion terminals in the area, so that the deployment number of physical sensors and communication modules is greatly reduced on the hardware level, and the equipment investment and the later maintenance cost are directly reduced.

The present invention discloses a real-time super-resolution reconstruction method and system based on multi-frame fusion. The method includes: constructing a network model; constructing the network model includes: performing linear feature extraction on the current frame to obtain a linear feature map of the current frame; performing non-linear feature extraction and feature fusion on the current frame and historical frames to obtain a non-linear feature map after feature fusion; the historical frames are the first N frames of the current frame, where N is a natural number; based on the non-linear feature map after feature fusion and the linear feature map of the current frame, an image with the target resolution of the current frame is obtained. The present invention effectively solves the defect of the existing super-resolution technology in terms of real-time performance, and improves the quality and visual effect of image reconstruction on the premise of ensuring real-time performance.

The invention discloses a parallel double-hop deep residual BP neural network-based distribution network ground screen corrosion detection method and device, and is used for solving the technical problem of poor corrosion detection accuracy caused by manual detection in most of existing distribution network ground screen corrosion detection methods. The method comprises the following steps: acquiring a plurality of to-be-detected distribution network earth screen images; inputting each to-be-detected distribution network earth screen image into a plurality of preset parallel double-hop deep residual BP neural networks, wherein each preset parallel double-hop deep residual BP neural network comprises a double-hop deep residual model and a BP neural network model; each double-hop depth residual model performs shallow low nonlinear feature extraction on input to-be-detected distribution network earth screen images, and outputs target shallow low nonlinear features corresponding to each to-be-detected distribution network earth screen image; and each BP neural network model performs corrosion detection on the input target shallow low-nonlinearity feature to generate a target distribution network earth screen corrosion detection result corresponding to each to-be-detected distribution network earth screen image.

The invention provides an electroencephalogram signal-based dysmnesia assessment system, which comprises a signal acquisition module for acquiring an electroencephalogram signal of a target object; the preprocessing module is used for performing 0.5-40Hz band-pass filtering on the electroencephalogram signal of the target object; the signal segmentation module is used for segmenting each channel and determining a plurality of non-overlapping Epoches corresponding to each channel; the feature extraction module is used for performing linear feature extraction and nonlinear feature extraction on each Epoch in each channel, and determining electroencephalogram signal fusion features of the target object based on the linear feature and the nonlinear feature of each Epoch; and the dysmnesia assessment module is used for inputting the electroencephalogram signal fusion features into a trained dysmnesia assessment model, and performing dysmnesia assessment on the target object by using the dysmnesia assessment model to obtain a dysmnesia assessment result of the target object. According to the scheme, dysmnesia evaluation of the target object can be efficiently realized.

The present application relates to a kind of electroencephalogram classification method based on physical information residual polynomial network, design to be trained network, first with polynomial feature extraction layer carries out nonlinear feature extraction, then through LSTM network capture long-term time series dependence of electroencephalogram, then realize the neural network architecture of E / I path separation, respectively to interface excitatory pathway and inhibitory pathway, then by the feature correlation analysis layer of embedding Wilson-Cowan neural population dynamics equation carries out physical constraint, finally in succession fusion layer, classification layer, classification layer completes the construction of to-be-trained network, then based on each sample formed by each multi-channel electroencephalogram, for to-be-trained network training, obtains electroencephalogram classification model, while guaranteeing electroencephalogram classification prediction accuracy, significantly enhance biological explainability.

The present application relates to a kind of lithium battery SOH and RUL collaborative prediction method based on nonlinear enhancement LSTM, belong to battery life prediction technical field, solve the problem of large amount of electrochemical data demand, single and insufficient feature extraction, waste of computing resources and deep, weak nonlinear feature extraction capability and low SOH and RUL prediction accuracy in prior art.The present application is based on data-driven, through multi-channel convolutional neural network multi-dimensional feature extraction, and based on nonlinear enhancement LSTM model carries out battery sequence data feature extraction, in-depth mining nonlinear feature of battery parameter, SOH evaluation result is used as the input of RUL prediction, using neural network method carries out RUL prediction, the prediction accuracy of SOH and RUL of lithium battery is high;Data feature is fully extracted, and the amount of electrochemical data demand is reduced;Using the same pre-training nonlinear enhancement LSTM model shares input, and the computing resource is greatly saved.

The invention relates to an electroencephalogram signal classification method based on a physical information residual polynomial network, and the method comprises the steps: designing a to-be-trained network, carrying out the nonlinear feature extraction through a polynomial feature extraction layer, capturing the long-term time sequence dependence relation of electroencephalogram signals through an LSTM network, and achieving the neural network architecture of E / I path separation. The method comprises the following steps: respectively connecting an excitability pathway and an inhibitory pathway, then carrying out physical constraint by a feature correlation analysis layer embedded with a Wilson-Cowan neural population kinetic equation, finally, sequentially connecting a fusion layer and a classification layer in series, and completing the construction of a to-be-trained network by the classification layer, and further constructing a to-be-trained network based on each sample formed by each multi-channel electroencephalogram signal. According to the method, the to-be-trained network is trained, the electroencephalogram signal classification model is obtained, and the biological interpretability is remarkably enhanced while the electroencephalogram signal classification prediction accuracy is guaranteed.

The present application relates to the technical field of medical monitoring, in particular to a vascular hemodynamic abnormality monitoring system and a pulse wave analysis method thereof, the core of the present application lies in that a flexible piezoelectric sensor array is used to capture human vascular pulse wave signals, the signals are transmitted to a wearable monitoring device for preliminary preprocessing, and then transmitted to a data processing host through a wireless manner, a pulse wave analysis module in the host adopts a multi-level analysis method, including multi-scale signal decomposition and reconstruction, nonlinear feature extraction and pattern recognition, multi-parameter adaptive judgment standard, deep learning feature automatic extraction and multi-site information fusion decision, ensuring the accuracy and comprehensiveness of the analysis, finally, the processing result is transmitted to an application platform to generate a vascular health report, realizing non-invasive continuous monitoring, this innovation effectively fills the gap of the prior art in continuous monitoring, and has important significance for real-time monitoring of vascular health status and prevention of vascular diseases.

The invention discloses a high-reliability planetary gearbox fault diagnosis method, system, medium and equipment, and the method comprises the steps: carrying out the feature extraction and feature selection based on a random forest algorithm, and carrying out the nonlinear feature extraction based on multi-scale permutation entropy, performing multi-scale coarse graining processing on each vibration signal, calculating permutation entropy values of coarse graining sequences under different scales, forming a multi-dimensional phase space, and obtaining a feature matrix of permutation entropy; performing feature dimension reduction and feature embedding based on kernel principal component analysis; according to the fault diagnosis based on the bidirectional long-short term memory neural network, a classifier of the bidirectional long-short term memory neural network is used, a feature set is divided into a training set and a test set to serve as input layers, and then a classification result is output through two LSTM units of a hidden layer, a full connection layer and Softmax function mapping. And high-reliability planetary gearbox fault diagnosis is realized.

The invention discloses a text generation method and device, equipment, a storage medium and a product. The method comprises the following steps: inputting a text vector of a text to be inquired into a text generation model, and carrying out encoding processing on the text vector by utilizing an encoder in the text generation model to obtain a first encoding vector and a second encoding vector; performing linear feature extraction on the first coding vector and the second coding vector to obtain a first fusion vector corresponding to the first coding vector and a second fusion vector corresponding to both the first coding vector and the second coding vector; performing nonlinear feature extraction on the first fusion vector and the first coding vector to obtain a first feature vector; performing cross fusion on the first feature vector and the second fusion vector to obtain a second feature vector; and decoding the second feature vector to obtain a reply text matched with the to-be-inquired text. According to the embodiment of the invention, the text generation accuracy can be improved when the text is generated.

This invention discloses a multi-view, multi-manifold classifier that preserves both local and global structure. In the feature space obtained through multiple empirical kernel mappings, this invention uses nonlinear feature extraction to preserve the inherent low-dimensional embeddings of the original data from multiple views. Furthermore, this invention utilizes inter-class graphs representing multi-manifold information and intra-class graphs representing sub-manifold information to constrain the training of the discriminative hyperplane. This invention overcomes the deficiency of existing multi-view classifiers that neglect the manifold structure of the original samples, improving the classification performance of multi-view data by utilizing the local and global geometric information surrounding each view's data point.

The invention discloses a digital economic high-frequency transaction fluctuation prediction method based on a large model, and relates to the technical field of agricultural Internet, and the method comprises the following steps: S1, obtaining agricultural high-frequency transaction data from a transaction large model in real time; s2, performing time sequence modeling on the preprocessed agricultural high-frequency transaction data by using an autoregression variational inference model; s3, performing nonlinear feature extraction modeling on the agricultural high-frequency transaction data by using a large-scale deep neural network; s4, constructing a digital economic agriculture high-frequency transaction fluctuation prediction model; s5, dynamically updating model parameters by using the latest agricultural high-frequency transaction data according to the digital economic agricultural high-frequency transaction fluctuation prediction model to form a real-time feedback mechanism; and S6, predicting a future market fluctuation trend based on the updated digital economic agriculture high-frequency transaction fluctuation prediction model, and outputting a predicted fluctuation probability and a corresponding uncertainty interval. The prediction stability of the model under extreme market conditions is improved.

The invention discloses a ship classification and identification method and device based on underwater acoustic uniform linear array data features, and relates to the field of underwater acoustic signal processing. A uniform linear array is constructed by constructing the relation between array structure parameters and phase-space reconstruction parameters in the nonlinear feature calculation process; according to the method, nonlinear feature extraction can be directly carried out on array data on underwater sound time-domain signals which are simultaneously collected by a uniform linear array and come from a plurality of channels, an additional phase space reconstruction step is not needed, and the method can be used for ship classification and identification. The method not only improves the calculation efficiency, but also can make full use of multi-channel information, improves the stability and robustness of feature extraction, enables the method to more accurately represent the dynamic characteristics of signals in a complex underwater environment, and can accurately carry out the classification and recognition of ship targets.

The application discloses a wind turbine generator unit variable flow system performance degradation identification method and related device, comprising: collecting the temperature of IGBT module in the variable flow system, the DC bus voltage, the output power harmonic distortion rate, and constructing a data set based on the same; performing data processing on the data set to obtain a processed data set; performing nonlinear feature extraction on the data set to obtain a dimension-reduced principal component score vector; constructing a hidden Markov model; and calculating the probability of the wind turbine generator unit variable flow system being in each degradation state based on the hidden Markov model and the dimension-reduced principal component score vector. The method and related device can realize performance degradation identification of the variable flow system, and have lower requirements on data quality.

The invention provides a milling chatter prediction method and system based on time distribution and mixed attention, and belongs to the field of milling state monitoring. The method comprises the following steps: collecting a multi-channel original time sequence signal in a milling process, and pre-processing the multi-channel original time sequence signal; inputting the preprocessed signal into a time sequence distribution convolution feature extraction module, and carrying out nonlinear feature extraction and dimension reduction to obtain a multi-channel time sequence feature sequence; inputting the multi-channel time sequence feature sequence into a global time sequence attention modeling unit, and outputting global dynamic evolution features; inputting the global dynamic evolution characteristics into a time dimension self-attention module and a channel attention module, and performing key time slice screening and multi-sensor channel importance re-calibration to obtain fusion characteristics; and inputting the fusion features into a classifier, and outputting the chatter probability in a future time window. According to the method, accurate capture and early warning of the chatter initiation precursor are realized, and an effective technical means is provided for active control of the stability of the machining process.