Few-sample marine engine anomaly detection method based on enhanced element transfer learning

Abstract

The invention relates to a few-sample marine engine anomaly detection method based on enhanced element transfer learning, and belongs to the technical field of marine engine anomaly detection. The method comprises the steps of obtaining historical data of a ship engine, constructing a ship engine operation database through integration, association and storage, and obtaining a ship engine operation data set through data preprocessing according to the ship engine operation database, obtaining a ship engine anomaly detection model through enhanced element transfer learning according to the ship engine operation data set, collecting real-time ship engine data, and obtaining a ship engine anomaly detection result through the ship engine anomaly detection model according to the real-time ship engine data, according to the ship engine anomaly detection result, a guarantee decision is made through a preset ship engine operation and maintenance strategy, few-sample ship engine anomaly detection and timely early warning are achieved, and ship safety is guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of marine engine anomaly detection, and in particular relates to a few-sample marine engine anomaly detection method based on enhanced meta-transfer learning, which can be used to realize engine anomaly detection in a digital simulation scenario of a marine engine. Background Art

[0002] The marine engine is composed of many mechanical, electrical, hydraulic and other subsystems. It is a complex electromechanical device and a nonlinear complex system. The structure and function of each subsystem are very different. There are many interconnections and interactions between subsystems at the same level or different levels. Due to the complexity of the system, marine engines are prone to failure during operation. The relationship between the fault phenomenon and the source of the marine engine failure is very complex. Most of the original faults may produce some kind of distortion during the propagation process, or show completely different fault phenomena, or directly lead to the failure of other subsystems. If you only rely on the operator's experience to make judgments, it is often impossible to quickly and accurately find the root cause of the fault, which may cause the best time for fault handling to be missed, thereby aggravating the severity of the fault.

[0003] With the rapid development of new generation information technologies such as big data, artificial intelligence and the Internet of Things, data has become a digital expression of material flow, information flow and energy flow. By acquiring, analyzing and integrating a large amount of marine engine operation information and data, strong support can be provided for the abnormal detection of marine engines. When a marine engine fails, accurately locating the faulty equipment and quickly analyzing the cause of the failure are of vital importance to ensure navigation safety. At present, there are many studies on marine engines, but they mainly focus on performance optimization, energy efficiency improvement and technological innovation. There are relatively few studies on marine engine anomaly detection algorithms. In addition, marine engine anomaly detection has problems such as few abnormal samples and unbalanced sample numbers between categories. The existing marine engine anomaly detection algorithms are insufficient in identifying a few types of faults, and the accuracy of the diagnostic results is low. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present invention proposes a few-sample marine engine anomaly detection method based on enhanced meta-transfer learning. When an abnormality occurs in the ship's engine, it can be detected and located, assisting maintenance personnel and other relevant personnel to improve work efficiency and ensure safety during navigation.

[0005] The purpose of the present invention can be achieved by the following technical solutions: A few-sample marine engine anomaly detection method based on enhanced meta-transfer learning, comprising: Acquiring historical data of a ship engine, wherein the historical data of the ship engine includes operating parameters of the ship engine and operating status of the ship engine; Building a ship engine operation database by integrating and associating storage according to the ship engine historical data, and obtaining a ship engine operation data set by preprocessing the ship engine operation database; According to the ship engine operation data set, a ship engine anomaly detection model is obtained by enhanced meta-transfer learning, wherein the enhanced meta-transfer learning includes a deep neural network, meta-transfer learning and a classifier; Real-time ship engine data is collected, and ship engine anomaly detection results are obtained through the ship engine anomaly detection model according to the real-time ship engine data, and a security decision is made through a preset ship engine operation and maintenance strategy according to the ship engine anomaly detection results.

[0006] Preferably, the constructing a ship engine operation database by integrating associated storage according to the ship engine historical data, and obtaining a ship engine operation data set by data preprocessing according to the ship engine operation database comprises: By integrating ship engine data from different sensors and subsystems through data fusion, ship engine parameters and equipment information of each ship engine subsystem are matched and associated and classified and stored in the ship engine operation database, and various abnormal state descriptions of the abnormal state of the ship engine are associated and stored in the ship engine operation database; The data preprocessing includes data screening, denoising, and normalization. The data screening excludes irrelevant or low-quality data points. The denoising filters out noise through filtering and signal processing techniques. The noise includes sensor noise, mechanical noise, and environmental noise. The normalization scales the data to a uniform numerical range.

[0007] Preferably, the obtaining of the ship engine anomaly detection model by enhanced meta-transfer learning according to the ship engine operation data set comprises: According to the ship engine operation data set, a ship engine source domain data set and a ship engine target domain data set are obtained by segmentation, wherein the ship engine source domain data set includes a ship engine source domain training set and a ship engine source domain verification set; Performing short-time Fourier transform on the ship engine source domain data set and the ship engine target domain data set to obtain a two-dimensional time-frequency image of the ship engine; According to the ship engine source domain training set, a pre-trained classifier is optimized by deep neural network pre-training to obtain pre-trained classifier parameters; According to the pre-trained classifier parameters, an enhanced meta-transfer model of a ship engine is obtained by meta-training with improved meta-transfer learning, wherein the improved meta-transfer learning uses a deep residual network as a backbone network, and introduces an attention mechanism and domain adaptation to fine-tune the classifier parameters; The ship engine enhanced meta-transfer model is tested and adjusted according to the ship engine target domain data set to obtain a ship engine anomaly detection model.

[0008] Preferably, the pre-trained classifier parameters obtained by pre-training and optimizing the classifier through a deep neural network according to the ship engine source domain training set include: The two-dimensional time-frequency images of the ship engines in the ship engine source domain training set are input into a deep neural network; The deep neural network includes three residual blocks, each of which includes two convolutional layers and two batch normalization layers. The convolutional layers use a 3*3 convolution kernel and the output sizes are 80*80*160, 80*80*320, and 80*80*640 respectively. The mathematical expression of the deep neural network optimization classifier parameters is: , , in, is the classifier, is the feature extractor, is the learning rate, Assign classifier parameters for parameter update, is the cross entropy loss function, is the change in cross entropy loss during parameter optimization, For the The actual labels of the ship engine source domain samples, is the indicator function, To be the first The ship engine source domain samples are classified as The probability of a ship engine labeled sample, is the number of samples in the ship engine source domain training set.

[0009] Preferably, obtaining the enhanced meta-transfer model of the ship engine by improving the meta-transfer learning meta-training according to the pre-trained classifier parameters comprises: The deep residual network is used as the backbone network and the attention mechanism and domain adaptation are introduced for fine-tuning; The domain adaptation adopts the distance measurement method of maximum mean deviation, and the mathematical expression of the domain adaptation loss function is: , in, is the domain adaptive loss function, is the maximum mean deviation loss function, is the classification loss of the labeled source domain, is the hyperparameter used to balance the maximum mean difference loss and classification loss; The fine-tuning designs a scaling shift for a small sample ship engine anomaly detection scenario, that is, a scaling factor and a shift factor are added to each convolution kernel; The mathematical expression of the scaling shift is: , in, SS is the function representation of scaling and shifting, X For input, is element-wise multiplication, S 1 and S 2 is the meta-learner, is the scaling factor, is the shift factor, W is the weight, b is bias; The mathematical expression of the meta-transfer learning optimization classifier is: , in, is the meta-learning rate, is the change of classifier parameters.

[0010] Preferably, the process of optimizing the classifier by meta-transfer learning includes: Forming a plurality of meta-tasks by random sampling according to the ship engine source domain training set and the ship engine source domain verification set; Train on the ship engine source domain training set of each meta-task, and update the meta-learner on the ship engine source domain validation set of each meta-task; Each meta-task updates the scaling and shift factors by calculating the cross entropy loss. When the meta-training reaches the preset number of iterations, that is, all meta-tasks are trained, the meta-transfer learning optimization classifier ends and the ship engine enhanced meta-transfer model training is completed.

[0011] Preferably, the step of testing and adjusting the ship engine enhanced meta-transfer model according to the ship engine target domain data set to obtain the ship engine anomaly detection model comprises: The ship engine target domain data set is input into the ship engine enhanced meta-transfer model to test the anomaly detection accuracy, and the input test data is judged to determine whether the engine is in good operating condition, and possible anomalies are warned, and the equipment of the subsystem where the anomaly is located is located; The anomaly detection accuracy is obtained by comparing the predicted output results of the ship engine enhanced meta-migration model with the real labels of the ship engine operation information. If the anomaly detection accuracy reaches 95%, the classifier parameter adjustment is completed; if the anomaly detection accuracy does not reach 95%, the model parameters are continued to be adjusted until the accuracy meets the standard, and the ship engine anomaly detection model is obtained.

[0012] A system for detecting anomalies of a few samples of marine engines based on enhanced meta-transfer learning, the system being applied to the above-mentioned method for detecting anomalies of a few samples of marine engines based on enhanced meta-transfer learning, comprising a data acquisition preprocessing module, an enhanced meta-transfer learning module, and a marine engine abnormality warning module; The data acquisition preprocessing module is used to obtain historical data of ship engines, build a ship engine operation database by integrating associated storage, and obtain a ship engine operation data set by data preprocessing; The enhanced meta-transfer learning module is used to obtain a ship engine anomaly detection model through enhanced meta-transfer learning according to the ship engine operation data set; The ship engine abnormality warning module is used to collect real-time ship engine data, obtain ship engine abnormality detection results through the ship engine abnormality detection model, and make security decisions and warnings based on the ship engine abnormality detection results through a preset ship engine operation and maintenance strategy.

[0013] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for detecting anomalies of a few-sample marine engines based on enhanced meta-transfer learning is implemented.

[0014] A storage medium containing computer executable instructions, which, when executed by a computer processor, are used to execute the above-mentioned few-sample marine engine anomaly detection method based on enhanced meta-transfer learning.

[0015] The beneficial effects of the present invention are: (1) By pre-training the classifier parameters through a deep neural network, the enhanced meta-transfer model of the ship engine is obtained by improving the meta-transfer learning meta-training. By testing and adjusting the enhanced meta-transfer model of the ship engine, the ship engine anomaly detection model is obtained, which realizes the optimization design for the situation where abnormal samples are scarce. The model can still accurately identify and distinguish various types of anomalies when the abnormal sample data set is small, which can greatly improve the monitoring and maintenance efficiency of marine engines. In addition, due to the effective response ability of the anomaly detection model to the situation with few samples, it can be extended to other fields that require high-precision fault prediction and anomaly detection.

[0016] (2) By improving meta-transfer learning, the classifier is further optimized with the parameters of the deep neural network pre-trained classifier as the initialization. The deep residual network is used as the backbone network, and the attention mechanism and domain adaptation are introduced to fine-tune the classifier parameters, thereby enhancing the model's ability to recognize abnormal features, reducing the classification loss of labeled source domain data, and reducing the domain loss between the unlabeled target domain and the source domain, thereby improving the accuracy of ship engine anomaly detection; (3) Through data fusion, ship engine data from different sensors and subsystems are effectively integrated to provide comprehensive and accurate information, enhance data integrity and analysis value, exclude irrelevant or low-quality data points through data screening, filter out noise through filtering and signal processing technology to ensure data accuracy and reliability, scale the data to a uniform numerical range through normalization to facilitate comparison and analysis between different parameters, enhance data comparability and model training effect, and obtain ship engine time-frequency images through short-time Fourier transform, which helps to accurately monitor the operating status and improve fault diagnosis efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0018] Figure 1 The present invention is a flowchart of a method for detecting anomalies of a few-sample marine engine based on enhanced meta-transfer learning. DETAILED DESCRIPTION

[0019] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0020] See also Figure 1 , a few-sample marine engine anomaly detection method based on enhanced meta-transfer learning, comprising: S1: Acquire historical data of a ship engine, wherein the historical data of the ship engine includes operating parameters of the ship engine and operating status of the ship engine, wherein the operating parameters of the ship engine include speed of the ship engine, temperature of the ship engine, vibration of the ship engine, and pressure of the ship engine, and the operating status of the ship engine includes healthy status of the ship engine and abnormal status of the ship engine; S2: constructing a ship engine operation database by integrating associated storage according to the ship engine historical data, and obtaining a ship engine operation data set by data preprocessing according to the ship engine operation database, wherein the data preprocessing includes data screening, denoising, and normalization; S3: obtaining a ship engine anomaly detection model through enhanced meta-transfer learning according to the ship engine operation data set, wherein the enhanced meta-transfer learning includes a deep neural network, meta-transfer learning and a classifier; S4: collecting real-time ship engine data, obtaining a ship engine abnormality detection result through the ship engine abnormality detection model according to the real-time ship engine data, and making a security decision through a preset ship engine operation and maintenance strategy according to the ship engine abnormality detection result.

[0021] In this embodiment, the ship engine operation database is constructed by integrating associated storage according to the ship engine historical data, and the ship engine operation data set is obtained by data preprocessing according to the ship engine operation database, which is specifically implemented by the following steps: S201: effectively integrating ship engine data from different sensors and subsystems through data fusion, providing comprehensive and accurate information, enhancing data integrity and analysis value, matching and associating ship engine parameters and information of each ship engine subsystem equipment, and storing them in a ship engine operation database by category, and at the same time, associating and storing various abnormal state descriptions of the abnormal state of the ship engine in the ship engine operation database; Specifically, the abnormal state description includes the abnormal location, abnormal characteristics, abnormal data volume, abnormal cause, corresponding operating conditions when the abnormality occurs and corresponding collection time, and performs parameter correlation analysis to facilitate fault location.

[0022] S202: The data screening excludes irrelevant or low-quality data points. The denoising removes noise through filtering and signal processing technology. The noise includes sensor noise, mechanical noise, and environmental noise to ensure the accuracy and reliability of the data. The normalization scales the data to a uniform numerical range to facilitate comparison and analysis between different parameters, thereby enhancing the comparability of the data and the effect of model training.

[0023] In this embodiment, the ship engine anomaly detection model is obtained by enhanced meta-transfer learning according to the ship engine operation data set, specifically by the following steps: S301: obtaining a ship engine source domain data set and a ship engine target domain data set by segmentation according to the ship engine operation data set, wherein the ship engine source domain data set includes a ship engine source domain training set and a ship engine source domain verification set; S302: performing short-time Fourier transform on the ship engine source domain dataset and the ship engine target domain dataset to obtain a two-dimensional time-frequency image of the ship engine; S303: Pre-training and optimizing the classifier through deep neural network according to the ship engine source domain training set to obtain pre-trained classifier parameters; S303-1: inputting the two-dimensional time-frequency image of the ship engine in the ship engine source domain training set into a deep neural network; S303-2: The deep neural network includes three residual blocks, each of which includes two convolutional layers and two batch normalization layers. The convolutional layers use a 3*3 convolution kernel and the output sizes are 80*80*160, 80*80*320, and 80*80*640 respectively. S303-3: The mathematical expression of the deep neural network optimization classifier parameters is: , , in, is the classifier, is the feature extractor, is the learning rate, Assign classifier parameters for parameter update, is the cross entropy loss function, is the change in cross entropy loss during parameter optimization, For the The actual labels of the ship engine source domain samples, is the indicator function, To be the first The ship engine source domain samples are classified as The probability of a ship engine labeled sample, is the number of samples in the ship engine source domain training set; It should be noted that the learning rate is used to control the convergence speed of the model.

[0024] Specifically, a deep neural network is a complex artificial neural network structure composed of multiple layers of neurons. Each neuron is responsible for receiving input, processing, and generating output. A deep neural network includes an input layer, a hidden layer, and an output layer. The input layer is responsible for receiving data. The hidden layer contains several neurons that process the data received by the previous layer and pass the results to the next layer. The output layer is the last layer of the network, and the number of neurons in it depends on the category requirements of the classification task.

[0025] S304: obtaining a ship engine enhanced meta-transfer model through improved meta-transfer learning meta-training according to the pre-trained classifier parameters, wherein the improved meta-transfer learning uses a deep residual network as a backbone network, and introduces an attention mechanism and domain adaptation to fine-tune the classifier parameters; S304-1: Introduce the attention mechanism in the deep residual network to enhance the model's ability to identify abnormal features, thereby improving diagnostic accuracy; introduce domain adaptation technology to reduce the classification loss of labeled source domain data, while reducing the domain loss between the unlabeled target domain and the source domain; Specifically, traditional domain adaptation algorithms map two features to the same feature subspace by reducing the distribution difference between them. KL divergence is generally used to quantify the difference between the two. However, in small sample transfer learning, the amount of data is scarce, and the generalization performance of the data distribution estimated by a small number of samples is poor. Therefore, the distance measurement method of the maximum mean deviation is used; S304-2: The mathematical expression of the domain adaptive loss function is: , in, is the domain adaptive loss function, is the maximum mean deviation loss function, is the classification loss of the labeled source domain, is a hyperparameter used to balance the maximum mean difference loss and classification loss.

[0026] S304-3: The fine-tuning is designed to scale and shift for the small sample ship engine anomaly detection scenario, that is, to add a scaling factor and a shift factor to each convolution kernel; the scaling and shift operation eliminates the need to update all parameters of the model, which can reduce the risk of overfitting of the model. At the same time, the model parameters are transferred from the deep neural network to the meta-transfer learning, which can accelerate the convergence of the network; S304-4: The mathematical expression of the scaling shift is: , in, SS is the function representation of scaling and shifting, X For input, is element-wise multiplication, S 1 and S 2 is the meta-learner, is the scaling factor, is the shift factor, W is the weight, b For bias.

[0027] S304-5: The mathematical expression of the meta-transfer learning optimization classifier is: , in, is the meta-learning rate, is the change of classifier parameters.

[0028] S304-51: forming a plurality of meta-tasks by random sampling according to the ship engine source domain training set and the ship engine source domain verification set; S304-52: Train on the ship engine source domain training set of each meta-task, and update the meta-learner on the ship engine source domain verification set of each meta-task.

[0029] S304-53: Each meta-task updates the scaling and shifting factors by calculating the cross entropy loss. The mathematical expression of the updated scaling and shifting factors is: , , S304-54: When the meta-training reaches the preset number of iterations, that is, when all meta-tasks are trained, the meta-transfer learning optimization classifier ends, and the ship engine enhanced meta-transfer model training is completed.

[0030] S305: testing and adjusting the ship engine enhanced meta-transfer model according to the ship engine target domain data set to obtain a ship engine anomaly detection model; S305-1: The ship engine target domain data set is input into the ship engine enhanced meta-transfer model to test the anomaly detection accuracy, and the input test data is judged to determine whether the engine is in good condition, and possible anomalies are warned, and the equipment of the subsystem where the anomaly is located is located; S305-2: The anomaly detection accuracy is obtained by comparing the predicted output result of the ship engine enhanced meta-migration model with the real label of the ship engine operation information. If the anomaly detection accuracy reaches 95%, the classifier parameter adjustment is completed; if the anomaly detection accuracy does not reach 95%, the model parameters are continued to be adjusted until the accuracy meets the standard, and the ship engine anomaly detection model is obtained.

[0031] Specifically, in practice, the probability of a ship engine producing an abnormality is lower than its healthy state, the sample size of the abnormal data is smaller, and the number of different abnormal samples varies greatly. If an unbalanced data set is used, the model may not be able to recognize certain anomalies during training, affecting the generalization accuracy of the model and thus the accuracy of the diagnosis results. Therefore, the classifier is trained through enhanced meta-transfer learning, and the parameters are continuously adjusted and optimized to improve the accuracy of ship engine anomaly detection.

[0032] It should be noted that the meta-transfer learning is an algorithm that combines meta-learning and transfer learning to perform image classification. The algorithm is divided into two stages: pre-training and meta-training. In the pre-training stage, a large number of labeled data sets are used to train the basic network, and then the basic network is transferred to a small sample data set for meta-learning.

[0033] In this embodiment, the real-time ship engine data is collected, a ship engine abnormality detection result is obtained by using the ship engine abnormality detection model according to the real-time ship engine data, and a guarantee decision is made by using a preset ship engine operation and maintenance strategy according to the ship engine abnormality detection result. Specifically, the following steps are implemented: S401: transmitting the real-time ship engine data to the data processing unit for preprocessing and time-frequency feature extraction, and detecting the engine abnormality through the ship engine abnormality detection model to obtain the ship engine abnormality detection result; It should be noted that the time-frequency feature extraction is to obtain a two-dimensional time-frequency image of the ship engine using short-time Fourier transform.

[0034] S403: The ship engine operation and maintenance strategy includes trying to restore parameters to normal through system control and sending fault information to the terminal. The fault information includes a fault detection report, i.e., possible abnormalities of the ship engine and their probability of occurrence, and the engine subsystem equipment located at the location of the ship engine abnormality.

[0035] Specifically, while trying to restore the parameters to normal through system control, the fault information is sent to the terminal in a timely manner to remind maintenance personnel and other relevant personnel to perform timely repairs, thereby achieving accurate classification and timely repair of engine faults, thereby ensuring ship safety.

[0036] A few-sample marine engine anomaly detection system based on enhanced meta-transfer learning includes a data acquisition preprocessing module, an enhanced meta-transfer learning module, and a marine engine anomaly warning module; The data acquisition preprocessing module is used to obtain historical data of ship engines, build a ship engine operation database by integrating associated storage, and obtain a ship engine operation data set by data preprocessing; The enhanced meta-transfer learning module is used to obtain a ship engine anomaly detection model through enhanced meta-transfer learning according to the ship engine operation data set; The ship engine abnormality warning module is used to collect real-time ship engine data, obtain ship engine abnormality detection results through the ship engine abnormality detection model, and make security decisions and warnings based on the ship engine abnormality detection results through a preset ship engine operation and maintenance strategy.

[0037] The computer storage medium of the embodiment of the present invention may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device.

[0038] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0039] The program code included on the computer readable medium can be transmitted with any appropriate medium, including but not limited to wireless, electric wire, optical cable, RF, etc., or any suitable combination of the above. The computer program code for performing the operation of the present invention can be written in one or more programming languages ​​or their combinations, and the programming language includes object-oriented programming languages-such as Java, Smalltalk, C++, and also includes conventional procedural programming languages-such as "C" language or similar programming languages. The program code can be executed completely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on the remote computer, or completely on the remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0040] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A few-sample marine engine anomaly detection method based on enhanced meta-transfer learning, characterized in that: include: Acquiring historical data of a ship engine, wherein the historical data of the ship engine includes operating parameters of the ship engine and operating status of the ship engine; Building a ship engine operation database by integrating and associating storage according to the ship engine historical data, and obtaining a ship engine operation data set by preprocessing the ship engine operation database; According to the ship engine operation data set, a ship engine anomaly detection model is obtained by enhanced meta-transfer learning, wherein the enhanced meta-transfer learning includes a deep neural network, meta-transfer learning and a classifier; Real-time ship engine data is collected, and ship engine anomaly detection results are obtained through the ship engine anomaly detection model according to the real-time ship engine data, and a security decision is made through a preset ship engine operation and maintenance strategy according to the ship engine anomaly detection results.

2. The method for detecting anomalies of a few-sample marine engine based on enhanced meta-transfer learning according to claim 1, characterized in that: The constructing of a ship engine operation database by integrating and associating storage according to the ship engine historical data, and obtaining a ship engine operation data set by data preprocessing according to the ship engine operation database include: By integrating ship engine data from different sensors and subsystems through data fusion, ship engine parameters and equipment information of each ship engine subsystem are matched and associated and classified and stored in the ship engine operation database, and various abnormal state descriptions of the abnormal state of the ship engine are associated and stored in the ship engine operation database; The data preprocessing includes data screening, denoising, and normalization. The data screening excludes irrelevant or low-quality data points. The denoising filters out noise through filtering and signal processing techniques. The noise includes sensor noise, mechanical noise, and environmental noise. The normalization scales the data to a uniform numerical range.

3. The method for detecting anomalies of a few-sample marine engine based on enhanced meta-transfer learning according to claim 1, characterized in that: The ship engine anomaly detection model obtained by enhanced meta-transfer learning according to the ship engine operation data set includes: According to the ship engine operation data set, a ship engine source domain data set and a ship engine target domain data set are obtained by segmentation, wherein the ship engine source domain data set includes a ship engine source domain training set and a ship engine source domain verification set; Performing short-time Fourier transform on the ship engine source domain data set and the ship engine target domain data set to obtain a two-dimensional time-frequency image of the ship engine; According to the ship engine source domain training set, a pre-trained classifier is optimized by deep neural network pre-training to obtain pre-trained classifier parameters; According to the pre-trained classifier parameters, an enhanced meta-transfer model of a ship engine is obtained by meta-training with improved meta-transfer learning, wherein the improved meta-transfer learning uses a deep residual network as a backbone network, and introduces an attention mechanism and domain adaptation to fine-tune the classifier parameters; The ship engine enhanced meta-transfer model is tested and adjusted according to the ship engine target domain data set to obtain a ship engine anomaly detection model.

4. The method for detecting anomalies of a few-sample marine engine based on enhanced meta-transfer learning according to claim 3 is characterized in that: The pre-trained classifier parameters obtained by pre-training and optimizing the classifier through a deep neural network according to the ship engine source domain training set include: The two-dimensional time-frequency images of the ship engines in the ship engine source domain training set are input into a deep neural network; The deep neural network includes three residual blocks, each of which includes two convolutional layers and two batch normalization layers. The convolutional layers use a 3*3 convolution kernel and the output sizes are 80*80*160, 80*80*320, and 80*80*640 respectively. The mathematical expression of the deep neural network optimization classifier parameters is: ， ， in, is the classifier, is the feature extractor, is the learning rate, Assign classifier parameters for parameter update, is the cross entropy loss function, is the change in cross entropy loss during parameter optimization, For the The actual labels of the ship engine source domain samples, is the indicator function, To be the first The ship engine source domain samples are classified as The probability of a ship engine labeled sample, is the number of samples in the ship engine source domain training set.

5. The method for detecting anomalies of a few-sample marine engine based on enhanced meta-transfer learning according to claim 3 is characterized in that: The method of obtaining a ship engine enhanced meta-transfer model by improving meta-transfer learning meta-training according to the pre-trained classifier parameters comprises: The deep residual network is used as the backbone network and the attention mechanism and domain adaptation are introduced for fine-tuning; The domain adaptation adopts the distance measurement method of maximum mean deviation, and the mathematical expression of the domain adaptation loss function is: ， in, is the domain adaptive loss function, is the maximum mean deviation loss function, is the classification loss of the labeled source domain, is the hyperparameter used to balance the maximum mean difference loss and classification loss; The fine-tuning designs a scaling shift for a small sample ship engine anomaly detection scenario, that is, a scaling factor and a shift factor are added to each convolution kernel; The mathematical expression of the scaling shift is: ， in, SS is the function representation of scaling and shifting, X For input, is element-wise multiplication, S 1 and S 2 is the meta-learner, is the scaling factor, is the shift factor, W is the weight, b is bias; The mathematical expression of the meta-transfer learning optimization classifier is: ， in, is the meta-learning rate, is the change of classifier parameters.

6. The method for detecting anomalies of a few-sample marine engine based on enhanced meta-transfer learning according to claim 5, characterized in that: The process of meta-transfer learning to optimize the classifier includes: Forming a plurality of meta-tasks by random sampling according to the ship engine source domain training set and the ship engine source domain verification set; Train on the ship engine source domain training set of each meta-task, and update the meta-learner on the ship engine source domain validation set of each meta-task; Each meta-task updates the scaling and shift factors by calculating the cross entropy loss. When the meta-training reaches the preset number of iterations, that is, all meta-tasks are trained, the meta-transfer learning optimization classifier ends and the ship engine enhanced meta-transfer model training is completed.

7. The method for detecting anomalies of a few-sample marine engine based on enhanced meta-transfer learning according to claim 3, characterized in that: The step of testing and adjusting the ship engine enhanced meta-transfer model according to the ship engine target domain data set to obtain the ship engine anomaly detection model comprises: The ship engine target domain data set is input into the ship engine enhanced meta-transfer model to test the anomaly detection accuracy, and the input test data is judged to determine whether the engine is in good operating condition, and possible anomalies are warned, and the equipment of the subsystem where the anomaly is located is located; The anomaly detection accuracy is obtained by comparing the predicted output results of the ship engine enhanced meta-migration model with the real labels of the ship engine operation information. If the anomaly detection accuracy reaches 95%, the classifier parameter adjustment is completed; if the anomaly detection accuracy does not reach 95%, the model parameters are continued to be adjusted until the accuracy meets the standard, and the ship engine anomaly detection model is obtained.

8. A few-sample marine engine anomaly detection system based on enhanced meta-transfer learning, characterized in that: The system is applied to the few-sample marine engine anomaly detection method based on enhanced meta-transfer learning as described in any one of claims 1 to 7, comprising a data acquisition preprocessing module, an enhanced meta-transfer learning module, and a marine engine anomaly warning module; The data acquisition preprocessing module is used to obtain historical data of ship engines, build a ship engine operation database by integrating associated storage, and obtain a ship engine operation data set by data preprocessing; The enhanced meta-transfer learning module is used to obtain a ship engine anomaly detection model through enhanced meta-transfer learning according to the ship engine operation data set; The ship engine abnormality warning module is used to collect real-time ship engine data, obtain ship engine abnormality detection results through the ship engine abnormality detection model, and make security decisions and warnings based on the ship engine abnormality detection results through a preset ship engine operation and maintenance strategy.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, it implements the few-sample marine engine anomaly detection method based on enhanced meta-transfer learning as described in any one of claims 1-7.

10. A storage medium containing computer executable instructions, characterized in that: The computer executable instructions, when executed by a computer processor, are used to execute the few-sample marine engine anomaly detection method based on enhanced meta-transfer learning as described in any one of claims 1-7.

Images