Ship power system lightweight fault diagnosis method and device based on ensemble learning

Abstract

The invention discloses a ship power system lightweight fault diagnosis method based on ensemble learning. The method comprises the following steps: S1, ship sensor data acquisition: acquiring real-time operation data of a ship power system based on each measuring point; s2, feature processing: improving redundancy and noise of data, and mapping an original feature matrix formed by an improved data set to a low-dimensional space through a PCA technology; s3, ensemble learning: adopting an ensemble learning device to construct a power system fault diagnosis model, wherein the power system fault diagnosis model comprises two layers of fault diagnosis units: the first layer uses an independent classifier; and the second layer outputs a final fault diagnosis result through the neural network in combination with the integrated learning device. The invention further discloses a ship power system lightweight fault diagnosis device based on ensemble learning. The ship power system fault diagnosis method can effectively overcome the defects of a traditional method in a complex environment, improves the accuracy and robustness of ship power system fault diagnosis, and can be widely applied to the field of ship maintenance.

Description

Technical Field

[0001] The present invention relates to the fields of information processing, data mining, and fault diagnosis, and in particular to a lightweight fault diagnosis method and device for a ship power system based on ensemble learning. The present invention combines the actual application of fault diagnosis services in a health management system. From the perspective of practicality, it receives sensor parameter information of each subsystem of the ship, and through steps such as data preprocessing and feature dimensionality reduction, and uses an ensemble learning model to predict the model fault category, and finally generates a fault diagnosis result process, aiming to form a comprehensive maintenance plan. Background Art

[0002] The stable operation of the ship power system is crucial for the safety of the crew and the operation of the ship. The ship power system usually consists of multiple complex subsystems, including engines, transmission systems, propellers, etc. These systems face various environmental conditions and workloads during long voyages and are easily affected by problems such as mechanical failures, electrical failures, or system failures, which may lead to ship stoppages or even accidents.

[0003] With the development and application of machine learning in the industrial field, a large number of studies have been carried out in the field of ship fault diagnosis. Traditional ship power system fault diagnosis methods are mainly based on rule engines or simple statistical methods, and these methods have obvious problems of low accuracy and high misclassification rate when dealing with complex unbalanced class sample problems. Especially in the large-scale data analysis and fault prediction of power systems, traditional methods are often difficult to effectively handle changing operating conditions and fault modes. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies of the above background art, and provide a lightweight fault diagnosis method and device for a ship power system based on ensemble learning. On the basis of combining ensemble learning and principal component analysis (PCA) technology, it can effectively address the limitations of traditional methods in complex environments, and improve the accuracy and robustness of ship power system fault diagnosis by integrating multiple classifiers and optimizing the processing method of sensor data.

[0005] The present invention provides a lightweight fault diagnosis method for a ship power system based on integrated learning, comprising the following steps: S1, ship sensor data acquisition: various ship sensors acquire data, each acquired data is defined as a measuring point, and real-time operation data of the ship power system is acquired based on each measuring point; S2, feature processing: improving the redundancy and noise of the acquired data and performing label encoding, and mapping the original feature matrix composed of the improved data set to a low-dimensional space through the PCA technology; S3, integrated learning: based on the features and labels after feature processing, an integrated learner is used to construct a power system fault diagnosis model, and the power system fault diagnosis model includes two layers of fault diagnosis units: the first layer uses an independent classifier; the second layer outputs the final fault diagnosis result through a neural network combined with an integrated learner.

[0006] In the above technical scheme, the specific steps of step S2 are as follows: S21, time series data preprocessing: prepare the original data for the next step of analysis and model training through data cleaning, missing value processing, feature extraction and labeling processing methods; S22, feature dimensionality reduction: use PCA technology to extract the most representative features from the data with the maximum variance to achieve feature lightweight, and thereby generate the original feature matrix based on matrix multiplication and project it into a new low-dimensional space.

[0007] In the above technical solution, the specific steps of step S21 are as follows: S211, data cleaning and missing value processing: delete invalid data points and abnormal data, use interpolation methods to fill missing data, ensure data continuity and integrity, and select key system parameters as features; S212, feature extraction: standardize or normalize features of various dimensions and orders of magnitude through feature scaling to ensure that the weight contributed by each feature in constructing a power system fault diagnosis model is relatively balanced; S213, label encoding: classify fault categories according to manual experience and represent each fault state with a digital code.

[0008] In the above technical solution, in step S211, the specific process of selecting key system parameters as features is as follows: considering the correlation and influence between various features, eliminating irrelevant or redundant features to simplify the model and improve prediction performance.

[0009] In the above technical solution, in step S212, the standardization is based on Z-score standardization.

[0010] In the above technical solution, the specific steps of step S22 are as follows. This step uses PCA technology for feature dimensionality reduction: S221. Standardize the data: Perform standardization or normalization on the encoded original feature matrix to ensure that each feature has a similar range in terms of value; S222. Calculate the covariance matrix: Calculate the covariance matrix of the standardized feature matrix; S223. Calculate the eigenvalues and eigenvectors: Perform eigenvalue decomposition or singular value decomposition on the covariance matrix to obtain the eigenvalues and the corresponding eigenvectors; S224. Select the number of principal levels: According to the magnitudes of the eigenvalues, select the top k eigenvalues to determine the number of eigenvectors to be retained as principal components. The number of selected principal components is determined by the cumulative explained variance ratio to account for most of the variance of the data; S225. Generate the selective feature matrix: Use the selected principal components to construct the selective feature matrix, and project the selective feature matrix into a low-dimensional space through matrix multiplication.

[0011] In the above technical solution, the specific steps of step S3 are as follows: S31. Use multiple independent support vector machine classifiers. Each support vector machine classifier makes a diagnosis for a given fault category and outputs the probability or category label belonging to the fault category; S32. Receive the output results from the support vector machine classifiers through a backpropagation neural network. The backpropagation algorithm trains the network propagation layer to minimize the error between the predicted output and the actual output. The network propagation layer fuses the received output results and finally outputs the diagnosis result of the faults in the ship power system.

[0012] In the above technical solution, it further includes step S4. Generate a diagnosis report: Based on the evaluation object of the faults in the ship power system and the fault diagnosis results, retrieve based on the experience in the expert database to generate the corresponding fault diagnosis report and suggestions.

[0013] In the above technical solution, the specific process of step S4 is as follows: The power system fault diagnosis model receives information on the specific fault location, key parameters, and predicted fault type, combines with Sentence - BERT to calculate and match historical cases, and generates maintenance suggestions according to expert rules.

[0014] The present invention also provides a lightweight fault diagnosis device for a ship power system based on ensemble learning, which has a computer program that can execute the lightweight fault diagnosis method for a ship power system based on ensemble learning.

[0015] The lightweight fault diagnosis method and device for a ship power system based on ensemble learning of the present invention have the following beneficial effects: (1) Improved the accuracy and efficiency of fault diagnosis: By collecting ship sensor data and performing precise feature processing and selection, the present invention can effectively clean the data and reduce redundancy, thereby improving the accuracy of fault diagnosis. The use of the integrated learning method combined with a multi-layer model can comprehensively consider the outputs of multiple classifiers, further enhancing the precision and speed of fault diagnosis.

[0016] (2) Reduced maintenance costs and ship downtime: Through real-time monitoring and timely generation of fault diagnosis reports, the present invention can help ship maintenance personnel quickly locate and solve problems, reducing the ship downtime caused by fault repairs. This not only saves maintenance costs but also reduces production losses caused by ship downtime.

[0017] (3) Optimized maintenance strategies and extended equipment life: Through analysis and report generation, the present invention provides detailed fault diagnosis results and suggestions, helping ship maintenance personnel optimize maintenance strategies. Timely maintenance and preventive measures can effectively extend the service life of ship equipment, improving its overall operating efficiency and reliability.

[0018] (4) Enhanced ship safety and reliability: By promptly detecting and handling potential equipment failures, the present invention helps enhance the operational safety and reliability of the ship. It reduces the potential threats of equipment failures to ship operation safety, thus ensuring the safety of the ship and its crew. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is the overall flowchart of the lightweight fault diagnosis method for ship power systems based on integrated learning of the present invention; Figure 2 It is the schematic flowchart of step 2 in the lightweight fault diagnosis method for ship power systems based on integrated learning of the present invention; Figure 3 It is the schematic flowchart of step 3 in the lightweight fault diagnosis method for ship power systems based on integrated learning of the present invention; Figure 4 It is the schematic flowchart of step 4 in the lightweight fault diagnosis method for ship power systems based on integrated learning of the present invention; Figure 5 It is the distribution diagram of sample prediction results of step 42 in the lightweight fault diagnosis method for ship power systems based on integrated learning of the present invention; Figure 6 It is the schematic flowchart of step 5 in the lightweight fault diagnosis method for ship power systems based on integrated learning of the present invention; Figure 7 It is the user interface diagram of the diagnosis result report of step 5 in the lightweight fault diagnosis method for ship power systems based on integrated learning of the present invention; Figure 8Schematic diagram of the architecture of the lightweight fault diagnosis device for ship power systems based on integrated learning of the present invention; Figure 9 Schematic diagram of the technical principle and key points of the lightweight fault diagnosis method and device for ship power systems based on integrated learning of the present invention. Detailed implementation manners

[0020] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments, but the embodiments should not be construed as limiting the present invention.

[0021] In practice, the application field of the present invention is the fault diagnosis and monitoring of ship power systems. As an important maritime transportation tool, the stable operation of the power system of a ship is directly related to navigation safety and the life and property safety of the crew. The purpose of the present invention is to provide a lightweight fault diagnosis method and device for ship power systems based on integrated learning, and to establish a process from data acquisition, feature processing, fault diagnosis to report generation, which can help ship maintenance personnel to timely and accurately detect and handle faults in the power system, and improve the reliability and safety of the ship.

[0022] The technical solutions provided by the present invention will be introduced below with reference to the accompanying drawings and embodiments: Embodiment 1 As Figure 1 shown, the specific implementation manners of the present invention include the following steps: Step 1: Equipment sensor data acquisition The data acquisition communicates with various system devices and intelligent gateways, such as the edge gateways and positions of each device, is compatible with multiple communication protocols, supports large data volume concurrency, and defines data standards based on TCP, including standards in aspects such as data definition, data description, data quality, data transmission, and data processing, to achieve data traceability.

[0023] Receive ship sensors, such as gas turbine status parameter information, including control machine gas temperature, dynamic vortex speed, average gas temperature, starter current, high-pressure vibration, saturated steam output, fuel manifold flow rate, fuel manifold temperature, fuel manifold pressure, fuel pump outlet pressure, fuel pump steam turbine speed, actual fuel consumption parameters.

[0024] See Figure 2 , Step 2: Temporal data preprocessing The data preprocessing stage aims to prepare the raw data to make it suitable for further analysis and model training. In the fault diagnosis of ship power systems, data preprocessing involves cleaning data, feature extraction, and labeling for effective machine learning and pattern recognition.

[0025] Step 21: Data cleaning and missing value processing Data cleaning, delete invalid data points and abnormal data, such as sensor readings beyond reasonable ranges or error data caused by sensor failures. Use interpolation methods to fill in missing data to ensure data continuity and integrity. Based on the domain knowledge and experience of system engineers, select key system parameters as features, such as engine speed, oil pressure, temperature, etc. Consider the correlation and influence of features, and eliminate irrelevant or redundant features to simplify the model and improve prediction performance.

[0026] Step 22: Feature extraction Feature scaling Normalize or standardize features with different dimensions and magnitudes to ensure that the contribution weights of each feature to the model are relatively balanced, based on Z-score standardization.

[0027] Step 23: Label encoding Fault classification labels Classify fault categories according to manual experience, such as normal operation, fuel filter blockage, fuel pump body failure, power system shutdown, injector valve opening error status. Represent each fault state with a digital code: Normal operation: 0, fuel filter blockage: 1, fuel pump body failure: 2, power system shutdown: 3, injector valve opening error status: 4. Manually confirm the key information extracted from Step 2 and clarify the goals and requirements of the processing tasks. These tasks may involve multiple aspects such as rescue, protection, and command, and need to be determined according to specific situations.

[0028] See Figure 3 , Step 3: Feature dimensionality reduction Use PCA technology for feature dimensionality reduction.

[0029] Step 31: Standardize data First, perform standardization or normalization on the original feature matrix to ensure that each feature has a similar numerical range. This is a prerequisite for PCA because PCA calculates the correlation between features based on the covariance matrix.

[0030] Step 32: Calculate the covariance matrix Calculate the covariance matrix of the standardized feature matrix. The covariance matrix describes the linear relationship and variance magnitude between different features.

[0031] Step 33: Calculate eigenvalues and eigenvectors Perform eigenvalue decomposition (or singular value decomposition) on the covariance matrix to obtain eigenvalues (values describing the variance magnitude in the feature matrix) and corresponding eigenvectors (describing the directions of the principal components).

[0032] Step 34: Select the number of principal components Select the top k eigenvalues according to their magnitudes to determine how many principal components to retain. The number of principal components selected should be able to explain most of the variance in the data. Generally speaking, it can be determined by the cumulative proportion of explained variance.

[0033] Step 35: Generate a selective feature matrix Use the selected principal components (eigenvectors) to construct a new feature matrix - the selective feature matrix. These principal components are linear combinations of the original features, and the original feature matrix can be projected onto the new low-dimensional space through matrix multiplication.

[0034] Considering that the system parameters have large differences in dimension and order of magnitude, standardize the feature dataset of the samples, map the sample data to the interval [0, 1] to avoid the influence of too large differences in the dimension units or orders of magnitude of the sample data on the accuracy of model classification. The correlation between parameters leads to redundant stacking of information, which will affect the convergence speed of the model. Use PCA to reduce the dimension of the original feature matrix, and determine the dimension of the feature space after dimensionality reduction by calculating the cumulative contribution rate. In order to better perform data analysis, reduce the original multi-dimensional space to a 5-dimensional space by PCA, which not only avoids redundant information but also maximally retains the information, greatly improving the operation efficiency of the model.

[0035] See Figure 4 , Step 4: Ensemble learning Ensemble learning can obtain better generalization ability and performance than any single classifier by combining the prediction results of multiple base classifiers. This method is particularly suitable for dealing with complex, highly non-linear problems, such as multi-class diagnosis of ship power system faults.

[0036] Step 41: In the first layer, three independent support vector machine (SVM) classifiers are used, and each classifier diagnoses for a given fault category. SVM is a powerful supervised learning algorithm that can effectively handle high-dimensional spaces and complex decision boundaries. SVM is based on mapping samples into a high-dimensional space and classifying by finding the optimal hyperplane. For each fault category, different SVM models are trained and used to classify and predict the test samples. Each SVM classifier will output its prediction results for the test samples, that is, the probability or class label of the predicted sample belonging to each fault category.

[0037] Step 42: On the second layer, a BP (Back Propagation) neural network is designed to integrate the prediction results of the three SVM classifiers in the first layer and generate the final diagnostic result. The BP neural network is a classic feedforward neural network that trains the network through the backpropagation algorithm to minimize the error between the predicted output and the actual output. Here, its role is to take the outputs of the three SVM classifiers as inputs and perform data fusion through the network propagation layer. The input layer of the BP neural network receives the outputs from the three SVM classifiers, and the propagation layer combines this information and finally outputs the diagnostic result for the faults in the ship power system. The distribution of the sample prediction results is as Figure 5 shown.

[0038] The performance of the ensemble learning model is evaluated, including accuracy and recall. After training and testing, the SVM algorithm with the Gaussian radial basis function as the kernel function is selected. When the optimal penalty parameter c = 0.1 and the kernel function parameter g = 0.5, the BP neural network has 2 hidden layers, the input layer length is 15, the output layer is 5, the activation function uses Tanh, the alpha exponent is 0.001, the learning rate is set to 0.01, and the optimizer uses Adma. The comparison of the prediction classification results for the fault data of the ship subsystem is shown in Table 1.

[0039] Table 1 Comparison of Ensemble Learning Algorithms

[0040] Ensemble learning can integrate the advantages of multiple classifiers and improve the overall classification accuracy and stability. For the problem of sample imbalance, by integrating multiple classifiers, the misclassification risk of certain classes can be reduced.

[0041] Example 2 This example is basically the same as Example 1, except that: See Figure 6 , Step 5: Generate a diagnostic result report Once the ensemble learning model completes the fault diagnosis of the ship equipment, the next step is to generate a detailed diagnostic result report. This process includes inputting information such as the fault location, parameters, and fault type, calculating and matching historical cases in combination with Sentence - BERT, and generating maintenance suggestions according to expert rules.

[0042] Step 51: Obtain the results of fault diagnosis from the integrated learning model, including the identified specific fault locations (e.g., main engine lubrication system), key parameters (e.g., lubricating oil pressure, lubricating oil temperature), and predicted fault types (e.g., lubricating oil pump failure). Step 52: Use the Sentence-BERT model to search and match the historical case database. These case databases may contain descriptions of previous similar faults, diagnostic processes, and repair plans. Sentence-BERT compares the text information in the new diagnostic results with the historical cases and calculates the similarity. This helps to find the most relevant historical cases to provide a more accurate reference for the current fault. Step 53: Based on the rules of expert knowledge, it can generate detailed repair suggestions according to the matching degree between the diagnostic results and the historical cases. These rules of expert knowledge may be based on the following aspects: Fault severity: Evaluate its potential impact on ship operation according to the fault type and influence.

[0043] Repair priority: Suggest the priority of repair, such as urgent, important but can be postponed, normal maintenance, etc.

[0044] Specific repair steps: Provide detailed operation guides and steps to ensure that the fault can be repaired effectively and safely.

[0045] Report generation: Combine the above information into the diagnostic result report to form a complete report document.

[0046] See Figure 7 , the report clearly lists the following: Fault location and parameters: Specifically describe the location where the fault occurs and the values of key parameters.

[0047] Fault type: Accurately identify the specific type of the fault and possible causes.

[0048] Historical case matching results: List the most relevant historical cases for the current fault for reference.

[0049] Repair suggestions: Repair suggestions generated according to the expert rule system, including priority and specific operation steps.

[0050] Embodiment 3 See Figure 8 , the lightweight fault diagnosis device for ship power system based on integrated learning of the present invention includes the following parts: Ship sensor data acquisition module: Various ship sensors collect data, and each collected data is defined as a measurement point. Based on each measurement point, the real-time operation data of the ship power system is obtained; Feature processing module: Improve the redundancy and noise of the collected data and perform label encoding, and map the original feature matrix composed of the improved data set to a low-dimensional space through PCA technology; Ensemble learning module: Based on the features and labels after feature processing, use an ensemble learner to build a dynamic system fault diagnosis model. The dynamic system fault diagnosis model includes two layers of fault diagnosis units: the first layer uses independent classifiers; the second layer combines a neural network with the output of the ensemble learner to output the final fault diagnosis result; Diagnostic report generation module: According to the evaluation object and fault diagnosis result of the ship power system fault, retrieve based on the experience in the expert library to generate corresponding fault diagnosis reports and suggestions.

[0051] Technical key points of the present invention: The technical key points of the lightweight fault diagnosis of ships based on ensemble learning in the present invention are as Figure 9 shown.

[0052] S01: Ship sensor data acquisition The present invention relates to a method for diagnosing faults in a ship power system. First, ship sensor data is acquired. Each acquired data is defined as a measurement point, mainly including gas turbine state parameter information. Through various sensors, such as temperature sensors, pressure sensors, etc., the operation data of the ship power system is obtained in real time.

[0053] S02: Feature processing selection The acquired data undergoes a feature processing selection step, including data cleaning and feature dimensionality reduction algorithms, to improve data redundancy and noise. During the data cleaning process, outliers and incorrect data are removed to ensure data quality. Principal component analysis (PCA) is used for feature dimensionality reduction. PCA is a technique that maps the original feature matrix to a low-dimensional space through linear transformation. Its main goal is to reduce the number of features, reduce the data dimension, extract the most representative features, while maintaining the integrity and interpretability of the information, so as to achieve the purpose of lightweight.

[0054] S03: Ensemble learning method Based on the processed features and labels, an ensemble learning method is used to build a dynamic system fault diagnosis model. This method includes two layers of fault diagnosis modules: the first layer uses multiple independent classifiers, that is, multiple machine learning algorithms. In the present invention, 3 support vector machines (SVMs) are used to classify and predict the data; the second layer combines the output results of a neural network with the ensemble learner to perform the final fault category diagnosis.

[0055] S04: Diagnostic report generation Finally, based on the evaluation object and the fault diagnosis results, retrieve the experience from the expert database to generate corresponding fault diagnosis reports and suggestions. The report details the detected fault types, possible cause analysis, as well as recommended repair and preventive measures, helping ship maintenance personnel to respond and solve problems quickly.

[0056] In summary, the present invention provides a lightweight fault diagnosis method and device for ship power systems based on ensemble learning. Through the present invention, the accuracy and efficiency of fault diagnosis for ship power systems can be improved, the suspension of navigation and losses caused by faults can be reduced, while the maintenance strategy is optimized and the service life of equipment is extended.

[0057] Through the above specific implementation manners, the method of the present invention can realize the full-process operation of ship information collection, preprocessing, fault diagnosis, and report generation. The purpose of the present invention is to provide a ship monitoring and fault diagnosis method based on ensemble learning. This method realizes the real-time monitoring and accurate fault diagnosis of ship equipment through steps such as collecting ship sensor data, data preprocessing, feature dimensionality reduction, and ensemble learning, providing strong guarantee for the safe operation of ships. The present invention not only improves the accuracy and efficiency of fault diagnosis, but also provides strong technical support for the safe operation and maintenance of ships by automatically generating detailed diagnosis reports including fault locations, parameters, types, and treatment suggestions. In addition, this method also has flexibility and scalability, and can be updated and optimized regularly with the changes in the state of ship equipment and the emergence of new fault types to ensure long-term high-efficient fault diagnosis capabilities. In summary, the ship monitoring and fault diagnosis method of the present invention is of great significance for improving the safety and reliability of ships, and has broad application prospects and market value.

[0058] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and its equivalent technologies, the present invention also intends to include these changes and modifications.

[0059] The content not detailed in this specification belongs to the well-known prior art of those skilled in the art.

Claims

1. A lightweight fault diagnosis method for ship power systems based on ensemble learning, characterized in that: The steps include: S1. Ship sensor data collection: Various ship sensors collect data. Each collected data is defined as a measurement point. Real-time operation data of the ship power system is obtained based on each measurement point. S2, feature processing: improve the redundancy and noise of the collected data and encode the labels, and map the original feature matrix composed of the improved data set to a low-dimensional space through PCA technology; S3. Ensemble learning: Based on the features and labels after feature processing, an ensemble learner is used to build a power system fault diagnosis model. The power system fault diagnosis model includes two layers of fault diagnosis units: the first layer uses an independent classifier; the second layer outputs the final fault diagnosis result through a neural network combined with an ensemble learner.

2. The lightweight fault diagnosis method for ship power systems based on ensemble learning according to claim 1, wherein: The specific steps of step S2 are as follows: S21. Time series data preprocessing: Prepare the raw data for the next step of analysis and model training through data cleaning, missing value processing, feature extraction and labeling processing methods; S22. Feature dimensionality reduction: Use PCA technology to extract the most representative features from the data with the largest variance to achieve feature lightweight, and then generate the original feature matrix based on matrix multiplication and project it into a new low-dimensional space.

3. The lightweight fault diagnosis method for ship power systems based on ensemble learning according to claim 2, characterized in that: The specific steps of step S21 are as follows: S211, Data cleaning and missing value processing: Delete invalid data points and abnormal data, use interpolation methods to fill missing data, ensure data continuity and integrity, and select key system parameters as features; S212, Feature extraction: Standardize or normalize the features of each dimension and order of magnitude through feature scaling to ensure that the weights contributed by each feature in building the power system fault diagnosis model are relatively balanced; S213, label coding: classify the fault categories according to manual experience and represent each fault state with a digital code.

4. The lightweight fault diagnosis method for ship power systems based on ensemble learning according to claim 3, wherein: In step S211, the specific process of selecting key system parameters as features is as follows: Consider the correlation and influence between each feature and remove irrelevant or redundant features to simplify the model and improve prediction performance.

5. The lightweight fault diagnosis method for ship power systems based on ensemble learning according to claim 4, wherein: In the step S212, the standardization is based on Z-score standardization.

6. The lightweight fault diagnosis method for a ship power system based on ensemble learning according to claim 5, wherein: The specific steps of step S22 are as follows, which uses PCA technology to perform feature dimensionality reduction: S221, Standardized data: Standardize or normalize the encoded original feature matrix to ensure that each feature has a similar range in value; S222, calculating the covariance matrix: calculating the covariance matrix of the standardized feature matrix; S223, calculating eigenvalues ​​and eigenvectors: performing eigenvalue decomposition or singular value decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors; S224, select the number of main layers: according to the size of the eigenvalue, select the first k eigenvalues ​​to determine the number of eigenvectors retained as principal components, and the selected number of principal components determines the variance of most of the data by the cumulative explained variance ratio; S225. Generate a selective feature matrix: construct a selective feature matrix using the selected principal components, and project the selective feature matrix into a low-dimensional space through matrix multiplication.

7. The lightweight fault diagnosis method for ship power systems based on ensemble learning according to claim 6, characterized in that: The specific steps of step S3 are as follows: S31. Use multiple independent support vector machine classifiers. Each support vector machine classifier makes a diagnosis for a given fault category and outputs the probability or category label belonging to the fault category. S32. Receive the output results from the support vector machine classifiers through a backpropagation neural network. The backpropagation algorithm trains the network propagation layer to minimize the error between the predicted output and the actual output. The network propagation layer fuses the received output results and finally outputs the diagnosis results for the faults of the ship power system.

8. The lightweight fault diagnosis method for ship power systems based on ensemble learning according to claim 7, characterized in that: It further includes step S4: generating a diagnosis report. According to the evaluation object of the ship power system fault and the fault diagnosis results, retrieve based on the experience in the expert database to generate the corresponding fault diagnosis report and suggestions.

9. The lightweight fault diagnosis method for ship power systems based on ensemble learning according to claim 8, characterized in that: The specific process of the said step S4 is as follows: The power system fault diagnosis model receives information such as the specific fault location, key parameters, and predicted fault types, calculates and matches historical cases by combining Sentence - BERT, and generates maintenance suggestions according to expert rules.

10. A lightweight fault diagnosis device for a ship power system based on ensemble learning, having a computer program, characterized in that: This computer program can execute the lightweight fault diagnosis method for ship power systems based on ensemble learning as described in any one of claims 1 to 9.

Images