CN-121997052-A - Model training method and device, electronic equipment and storage medium

CN121997052ACN 121997052 ACN121997052 ACN 121997052ACN-121997052-A

Abstract

The application relates to a model training method, a model training device, an electronic device and a storage medium, wherein a training sample set is constructed by acquiring a fault signal data set of ship machinery and performing empirical mode decomposition on preprocessed original signal data; and carrying out iterative optimization training on the ship fault detection model to be trained based on the parallel feature extraction network architecture in the fault detection model to be trained until the ship fault detection model converges to obtain a trained ship fault detection model. The method can improve the processing capacity of nonlinear and nonstationary signals, avoid the limitation of artificial feature design, enhance the capture of time dependence relation of dynamic signals, and improve the robustness and the multi-fault recognition capacity of the model in a multi-noise environment through a parallel structure and feature fusion mechanism.

Inventors

DENG CHANGYI
LI LANQING
LI MINGSHI

Assignees

国家工业信息安全发展研究中心

Dates

Publication Date: 20260508
Application Date: 20260302

Claims (12)

1. A method of model training, the method comprising: acquiring a fault signal data set of the ship machinery, and preprocessing original signal data in the fault signal data set; Performing empirical mode decomposition on the preprocessed original signal data to obtain an eigenmode function component group corresponding to the original signal data; constructing a training sample set based on the intrinsic mode function component group corresponding to each original signal data in the fault signal data set; The training sample set is input into a ship fault detection model to be trained, wherein the ship fault detection model synchronously extracts spatial features and time sequence features of training samples through a parallel feature extraction network, and fuses the extracted spatial features and the time sequence features to obtain fusion features; and carrying out iterative optimization training on the ship fault detection model based on the difference between the classification prediction result and the real fault label of the corresponding training sample until the ship fault detection model converges, so as to obtain a trained ship fault detection model.
2. The method of claim 1, wherein the preprocessing of the raw signal data in the fault signal data set comprises: and (3) performing at least one preprocessing operation of resampling processing and normalization processing on the original signal data in the fault signal data set.
3. The method of claim 2, wherein resampling the raw signal data in the fault signal data set comprises: The original signal data is sampled in a segmented mode by adopting a sliding window with a preset overlapping rate, and a plurality of resampled signal segments are obtained; And constructing a sample data table based on the resampled signal segments, wherein each row in the sample data table corresponds to one resampled signal segment, and each row is attached with a corresponding fault tag.
4. The method according to claim 1, wherein the performing empirical mode decomposition on the preprocessed raw signal data to obtain the set of eigenmode function components corresponding to the raw signal data includes: determining the original signal data as a current processing signal; Determining local maximum values and local minimum values of the current processing signal, and fitting an upper envelope curve and a lower envelope curve of the current processing signal by an interpolation method based on the local maximum values and the local minimum values; determining an envelope mean value of the current processing signal according to the upper envelope line and the lower envelope line, and subtracting the envelope mean value from the original signal data to obtain a residual signal of the current processing signal; determining a standard deviation of the residual signal; If the standard deviation of the residual signal does not meet the preset threshold condition, taking the residual signal as a new current processing signal, and returning to execute the steps of determining the local maximum value and the local minimum value of the current processing signal and then, if the standard deviation of the residual signal meets the preset threshold condition, determining the residual signal as an intrinsic mode function component; Subtracting the sum of all extracted eigenmode function components from the original signal data to obtain a residual signal; If the residual signal is not a monotonic function or the amplitude is less than a set threshold, taking the residual signal as a new current processing signal, and returning to the steps of determining the local maximum value and the local minimum value of the current processing signal and then; And if the residual signal is a monotonic function or the amplitude is smaller than a set threshold value, stopping empirical mode decomposition of the original signal data, and forming an eigenmode function component group corresponding to the original signal data by all eigenmode function components which are extracted currently.
5. The method of claim 1, wherein the constructing a training sample set based on the set of eigenmode function components for each of the raw signal data in the fault signal data set comprises: The following processing is respectively executed for the eigenvalue function component group corresponding to each original signal data: determining a correlation coefficient between each eigenmode function component in the set of eigenmode function components and the raw signal data; screening effective components of the eigenmode functions from the eigenmode function component group according to the correlation coefficient to form an effective component group; Performing alignment processing on the effective component groups from different original signal data; And constructing a training sample set based on the aligned effective component groups.
6. The method of claim 5, wherein said aligning the set of active components from different raw signal data comprises: The method comprises the steps of (1) filling effective component groups with the number of effective components being less than that of targets by adopting a zero-value filling strategy to make up so that the number of the effective components in each effective component group is consistent; and/or the number of the groups of groups, And reducing the effective component groups with the effective component number being greater than the target number by adopting a component merging strategy, so that the effective component numbers in the effective component groups are consistent.
7. The method of claim 1, wherein the parallel feature extraction network comprises a parallel convolutional neural network module and a gated loop unit network module; the convolutional neural network module is used for extracting spatial features of the training samples, and the gating cycle unit network module is used for extracting time sequence features of the training samples.
8. The method of claim 1, wherein the fusing the extracted spatial features with the temporal features to obtain fused features comprises: and splicing the output of the last time step of the spatial feature and the output of the last time step of the time sequence feature in the channel dimension to obtain a fusion feature.
9. A method of detecting a marine vessel failure, the method comprising: Acquiring real-time operation signal data of the ship machinery to be detected, and preprocessing the real-time operation signal data; Performing empirical mode decomposition on the preprocessed real-time operation signal data to obtain an intrinsic mode function component group corresponding to the real-time operation signal data; Constructing data to be detected based on the intrinsic mode function component group; The method comprises the steps of inputting data to be detected into a trained ship fault detection model, synchronously extracting spatial features and time sequence features of a training sample by the ship fault detection model through a parallel feature extraction network, and fusing the extracted spatial features and the time sequence features to obtain fusion features; And determining the fault type of the ship machinery according to the classification prediction result.
10. A marine vessel failure detection apparatus, the apparatus comprising: The data preprocessing module is used for acquiring real-time operation signal data of the ship machinery to be detected and preprocessing the real-time operation signal data; The system comprises an intrinsic mode function component group determining module, a real-time processing module and a real-time processing module, wherein the intrinsic mode function component group determining module is used for carrying out empirical mode decomposition on the preprocessed real-time operation signal data to obtain an intrinsic mode function component group corresponding to the real-time operation signal data; The to-be-detected fault data set construction module is used for constructing to-be-detected data based on the intrinsic mode function component group; The system comprises a training sample, a data input module to be detected, a classification module and a classification prediction module, wherein the training sample is used for training a ship fault detection model, the data input module to be detected is used for inputting the data to be detected into the trained ship fault detection model, the ship fault detection model synchronously extracts spatial features and time sequence features of the training sample through a parallel feature extraction network, and fuses the extracted spatial features and the time sequence features to obtain fusion features; And the fault type determining module is used for determining the fault type of the ship machinery according to the classification prediction result.
11. An electronic device comprising a processor and a memory, the processor being configured to execute a ship fault detection control program stored in the memory, to implement the ship fault detection method of claim 9.
12. A storage medium storing one or more programs executable by one or more processors to implement the method of detecting a marine vessel failure of claim 9.

Description

Model training method and device, electronic equipment and storage medium Technical Field The present application relates to the field of computer technologies, and in particular, to a model training method, apparatus, electronic device, and storage medium. Background With the increase of the types and complexity of the mechanical equipment of the ship, fault detection has become a key for guaranteeing the safety and the high-efficiency operation of the ship. Traditional manual inspection and periodic maintenance methods are inefficient and prone to error. In recent years, fault detection methods based on sensor technology and machine learning are becoming mainstream, but challenges in signal processing, feature extraction, timing modeling and the like are still faced. The existing fault detection method based on machine learning has limited time sequence modeling capability, still relies on manual design for feature extraction, has poor adaptability, and has an unsatisfactory simultaneous detection effect on multiple types of faults. Disclosure of Invention The application provides a model training method, a model training device, electronic equipment and a storage medium, which are used for solving the technical problems that the existing fault detection method depends on manpower and has poor signal processing capability and fault recognition capability. In a first aspect, the present application provides a model training method, the method comprising: acquiring a fault signal data set of the ship machinery, and preprocessing original signal data in the fault signal data set; Performing empirical mode decomposition on the preprocessed original signal data to obtain an eigenmode function component group corresponding to the original signal data; constructing a training sample set based on the intrinsic mode function component group corresponding to each original signal data in the fault signal data set; The training sample set is input into a ship fault detection model to be trained, wherein the ship fault detection model synchronously extracts spatial features and time sequence features of training samples through a parallel feature extraction network, and fuses the extracted spatial features and the time sequence features to obtain fusion features; and carrying out iterative optimization training on the ship fault detection model based on the difference between the classification prediction result and the real fault label of the corresponding training sample until the ship fault detection model converges, so as to obtain a trained ship fault detection model. In a possible implementation manner, the preprocessing the original signal data in the fault signal data set includes: and (3) performing at least one preprocessing operation of resampling processing and normalization processing on the original signal data in the fault signal data set. In a possible implementation manner, the resampling processing of the original signal data in the fault signal data set includes: The original signal data is sampled in a segmented mode by adopting a sliding window with a preset overlapping rate, and a plurality of resampled signal segments are obtained; And constructing a sample data table based on the resampled signal segments, wherein each row in the sample data table corresponds to one resampled signal segment, and each row is attached with a corresponding fault tag. In a possible implementation manner, the performing empirical mode decomposition on the preprocessed raw signal data to obtain the set of eigenmode function components corresponding to the raw signal data includes: determining the original signal data as a current processing signal; Determining local maximum values and local minimum values of the current processing signal, and fitting an upper envelope curve and a lower envelope curve of the current processing signal by an interpolation method based on the local maximum values and the local minimum values; determining an envelope mean value of the current processing signal according to the upper envelope line and the lower envelope line, and subtracting the envelope mean value from the original signal data to obtain a residual signal of the current processing signal; determining a standard deviation of the residual signal; If the standard deviation of the residual signal does not meet the preset threshold condition, taking the residual signal as a new current processing signal, and returning to execute the steps of determining the local maximum value and the local minimum value of the current processing signal and then, if the standard deviation of the residual signal meets the preset threshold condition, determining the residual signal as an intrinsic mode function component; Subtracting the sum of all extracted eigenmode function components from the original signal data to obtain a residual signal; If the residual signal is not a monotonic function or the amplitude is less than a set threshold, taki