Active decision-making tree integrated diagnosis device for health state of shipborne electric power energy supply system

Abstract

The invention discloses an active decision-making tree integrated diagnosis device for the health state of a shipborne electric power energy supply system, and belongs to the technical field of health management of shipborne electric power energy supply systems. The device is applied to a power supply side, a power utilization side and a propulsion side of a shipborne power supply system, and comprises a data set construction module used for obtaining a health state data set of the shipborne power supply system; the model construction module is used for constructing an active decision-making tree model based on the ship-borne power supply system health state data set, and the active decision-making tree model comprises a plurality of independent decision-making trees; the data acquisition module is used for acquiring health state data of the shipborne power supply system to be diagnosed in real time; and the diagnosis module is used for inputting the collected health state data of the shipborne power supply system into the active decision tree model and outputting a diagnosis result. According to the method, an active decision tree integration model is adopted, the method does not depend on a network topology structure, and robustness is higher; the model training and diagnosis process does not need complex feature calculation, and high precision and low calculation complexity are both considered.

Description

Technical Field

[0001] This invention relates to the field of health management technology for shipborne power supply systems, and in particular to an integrated diagnostic device for the active decision tree of the health status of shipborne power supply systems. Background Technology

[0002] To ensure safe navigation and stable operation of equipment, the health of the ship's onboard power supply system, which is the "nervous system" of the ship, is directly related to the safety of navigation, stable operation of equipment, and the protection of personnel's lives.

[0003] Currently, the health diagnosis of shipboard power supply systems faces severe practical challenges: First, ships are exposed to harsh environments such as high salt spray, high humidity, and vibration for extended periods, leading to accelerated degradation of the insulation performance of electrical equipment and making fault points extremely difficult to detect (e.g., grounding faults may be caused by corrosion or insulation layer damage). The insulation characteristics of non-metallic ships, in particular, significantly increase the difficulty of fault location, extending the average troubleshooting time by 40% compared to metallic ships, making conventional diagnostic methods inefficient. Second, frequent load fluctuations during ship navigation (such as power system start-up and shutdown, and high-power equipment switching) can easily trigger instantaneous overloads or voltage fluctuations. Traditional static diagnostic methods cannot effectively capture such transient anomalies, and electrical system faults often exhibit cascading effects, requiring simultaneous collection and comprehensive analysis of multiple parameters such as voltage, current, and insulation resistance for accurate diagnosis. This places higher demands on the multi-parameter processing capabilities of diagnostic devices. Although real-time diagnostics can detect potential faults such as insulation aging, short circuits, and overloads in advance, thus avoiding major accidents such as power outages and propulsion system malfunctions caused by power interruptions, and can also optimize maintenance cycles, reduce unplanned downtime, and support integrated power system energy efficiency management, the traditional methods currently used for diagnosing the health status of power supply systems all have obvious shortcomings and are difficult to meet actual needs.

[0004] Specifically, the traditional methods for solving the health status diagnosis problem of power supply systems can be mainly divided into the following three types: (1) Power flow analysis based on the topology of the ship power supply system, although it is a core technology for power system analysis, can analyze power flow through system topology information, equipment parameters and operating conditions, calculate node voltage, phase angle and line power, and confirm the fault location and type based on the changes in voltage, current or power flow after the fault. However, it has a large amount of calculation, high computational complexity, strong dependence on topology and poor dynamic adaptability, and cannot quickly respond to the dynamic changes of the ship power supply system.

[0005] (2) The health status diagnosis method based on impedance method calculates the fault location by measuring the impedance between the fault point and the monitoring point, and calculates the impedance by combining current and voltage data. Although the calculation process is simple and has strong real-time performance, in complex shipboard power supply systems, the diagnostic accuracy is low due to interference from multiple devices, and it is difficult to accurately locate the fault.

[0006] (3) The health status diagnosis method based on phasor method calculates the fault location by measuring the phase difference and amplitude difference of the phasors under fault conditions by measuring the node voltage and current phasors (complex form). Although it has the advantages of high accuracy and strong adaptability, it has the disadvantage of poor real-time performance, similar to the power flow analysis method, and cannot meet the needs of ships for rapid fault diagnosis.

[0007] In summary, while existing traditional diagnostic methods have certain feasibility and local advantages, in practical applications they are difficult to balance low computational complexity, high accuracy, high adaptability, high real-time performance, and low dependence on models, and cannot adapt to the complex operating conditions of shipboard power supply systems. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention introduces an active decision tree ensemble model, which can effectively alleviate the application barriers of traditional methods in the health status diagnosis of shipboard power supply systems and provide reliable technical support for the health management of shipboard power supply systems.

[0009] To achieve the above objectives, the present invention provides an active decision tree integrated diagnostic device for the health status of a shipborne power supply system, applicable to the power supply side, power consumption side, and propulsion side of a shipborne power supply system, comprising: The dataset construction module is used to obtain the health status dataset of the shipboard power supply system, including the health status data of the shipboard power supply system and the corresponding health status categories. The health status categories include normal status labels and various fault status types. The model building module is used to build an active decision tree model based on the health status dataset of the shipboard power supply system. The active decision tree model includes multiple independent decision trees, and the corresponding independent datasets are generated by using the bootstrap method to generate multiple independent decision trees. The data acquisition module is used to collect real-time health status data of the shipboard power supply system to be diagnosed. The diagnostic module is used to input the collected health status data of the shipboard power supply system into the active decision tree model and output the diagnostic results.

[0010] Furthermore, the acquisition of health status test data of the shipborne power supply system and the construction of a health status test dataset include: Preprocessing the health status test data of the shipboard power supply system includes cleaning noise and processing missing values ​​in the health status test data. The preprocessed health status test data of the shipborne power supply system is normalized to eliminate the dimensional differences of different characteristic factors.

[0011] Furthermore, the active decision tree model is constructed using the aforementioned health status test dataset as follows: (2.1) Based on the preset number of independent decision trees in the active decision tree model, a corresponding number of independent datasets are constructed by randomly sampling with replacement from the health status test dataset using the bootstrap method, and each dataset corresponds to a separate independent decision tree. (2.2) For each independent decision tree, its corresponding independent dataset is used as the root node data; (2.3) Calculate the information gain or Gini coefficient of each feature in the independent dataset, evaluate the fault discrimination of each feature, and select the feature with the largest information gain or Gini coefficient as the split point of the independent decision tree corresponding to the independent dataset for splitting; (2.4) Divide the independent dataset layer by layer, recursively calculate the child nodes until the stopping condition is met, and generate the construction of the independent decision tree model; (2.5) Repeat steps (2.2)-(2.5) to generate a preset number of independent decision trees; (2.6) The active decision tree ensemble model is formed by integrating all the independent decision trees.

[0012] Furthermore, the self-help method specifically refers to: For the fault boundary samples in the health status test dataset, a boundary weight is set, wherein the boundary weight is greater than 1; The number of samples in the normal state and the number of samples in the faulty state in the health status test dataset are counted. When the difference between the two exceeds a threshold, a weight is assigned to all samples in the health status test dataset corresponding to the faulty state based on the difference in the proportion. The weight is greater than 1. The independent dataset is obtained by sampling with replacement based on the health status test dataset after the above weight adjustment.

[0013] Furthermore, the fault boundaries include at least critical overload, early stage of insulation aging, voltage overshoot threshold, frequency fluctuation threshold, and harmonic distortion threshold.

[0014] Furthermore, the stopping condition is when the child node samples belong to the same category or the tree reaches the maximum preset depth.

[0015] Furthermore, the integration of all independent decision trees specifically involves obtaining the final diagnostic result by combining the outputs of each independent decision tree through voting or weighted summation.

[0016] Furthermore, the display module is used to display real-time health status data, diagnostic results, and alarm information based on the diagnostic results of the shipboard power supply system.

[0017] This invention also provides an active decision tree integrated diagnosis method for the health status of a shipborne power supply system, comprising the following steps: (1) Obtain the health status dataset of the shipboard power supply system, including the health status data of the shipboard power supply system and the corresponding health status categories, wherein the health status categories include normal status labels and various fault status types; (2) Based on the health status dataset of the shipboard power supply system, an active decision tree model is constructed. The active decision tree model includes multiple independent decision trees, and the corresponding independent dataset is generated by the bootstrap method to generate multiple independent decision trees. (3) Real-time collection of health status data of the shipboard power supply system to be diagnosed; (4) Input the collected health status data of the shipboard power supply system into the active decision tree model and output the diagnostic results.

[0018] The beneficial effects of this invention are: (1) Breaking through the limitations of traditional algorithms and improving diagnostic performance: Compared with the diagnostic module using traditional algorithms, the active decision tree integration model of the present invention does not require the construction of a mathematical model of the power supply system and does not depend on the network topology. It is more robust when facing complex and dynamically changing ship power supply systems. At the same time, the model training and diagnostic process does not require complex feature calculations, avoiding the inefficiency problem caused by a large amount of computation in traditional algorithms. It balances high accuracy and low computational complexity, and is suitable for the real-time diagnostic needs of ships.

[0019] (2) Data-driven and adaptable to complex scenarios with stronger scalability: The model of this invention is based on data-driven and can continuously update the power supply system operation data through the data acquisition unit. The model training unit iteratively optimizes the active decision tree integration model based on new data, continuously learns the fault patterns under new operating conditions, and easily handles problems such as chain faults and transient anomalies that are difficult to deal with by traditional methods. Moreover, when facing large-scale data, the module distributed computing unit can process data and training models in parallel, greatly improving the efficiency of diagnostic decision-making. At the same time, it supports subsequent functional expansion (such as adding fault type discrimination) to adapt to the diagnostic needs of power supply systems of different sizes of ships. Attached Figure Description

[0020] Figure 1 This is a schematic diagram showing the installation location of the active decision tree integrated diagnostic device for the health status of the shipborne power supply system, according to an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the active decision tree integrated diagnosis method for the health status of a shipborne power supply system according to an embodiment of the present invention.

[0022] Figure 3 This is a schematic diagram of the active decision tree model in an embodiment of the present invention.

[0023] Figure 4 This is the verification process for the active decision tree ensemble model in an embodiment of the present invention.

[0024] Figure 5 This is a schematic diagram of the various performance indicators output by the active decision tree on the test set in an embodiment of the present invention.

[0025] Figure 6 This is a schematic diagram of the confusion matrix of the model in an embodiment of the present invention on the test set.

[0026] Figure 7 This is a schematic diagram showing the distribution of the importance of each feature in an embodiment of the present invention. Detailed Implementation

[0027] The present invention will be further explained and described below with reference to the accompanying drawings and embodiments.

[0028] like Figure 1 As shown in the figure, the embodiment of the present invention provides an active decision tree integrated diagnostic device for the health status of a shipborne power supply system, which can be installed on the power supply side, the power consumption side, and the propulsion side.

[0029] like Figure 2 As shown, the device includes: The dataset construction module 101 is used to obtain the health status dataset of the shipboard power supply system, including the health status data of the shipboard power supply system and the corresponding health status categories. The health status categories include normal status labels and various fault status types.

[0030] The health status data of the shipboard power supply system includes all characteristics related to the health status of the shipboard power supply (all of which are key parameters collected in real time by the module), such as: voltage, current, load, temperature, frequency, switching action time, insulation medium, etc. This embodiment of the invention uses voltage, current, load, and temperature as examples for analysis.

[0031] First, the health status test data of the shipboard power supply system is preprocessed, including cleaning up noise and removing missing values ​​from the data.

[0032] Then, the preprocessed health status test data of the shipboard power supply system is normalized to eliminate differences in the units of measurement for different features. Standardization is performed on the four types of feature data—voltage, current, load, and temperature—converting each type of feature data into a standard format with a mean of 0 and a standard deviation of 1. This operation eliminates the interference of differences in the units of measurement for different features (e.g., voltage is in volts (V) while current is in amperes (A)) on model training, ensuring that each feature plays a reasonable role in model training, thereby improving the diagnostic accuracy of the module.

[0033] The health status categories include normal status labels and various fault status types, including short circuit faults, insulation aging, three-phase imbalance faults, electrical component faults, abnormal system parameters, and control and regulation faults.

[0034] The model building module 102 is used to build an active decision tree model based on the health status dataset of the shipboard power supply system. The active decision tree model includes multiple independent decision trees.

[0035] This invention employs an active decision tree ensemble model as the core machine learning algorithm to accurately identify faults in shipboard power supply systems. Compared to machine learning algorithms such as artificial neural networks, support vector machines, and Naive Bayes, which are prone to overfitting to noisy data, have slow training processes, require large amounts of training data, and have complex parameter tuning, the active decision tree ensemble model performs better and is more adaptable to the diagnostic needs under complex ship operating conditions.

[0036] Active decision tree ensemble models belong to ensemble learning methods. In this embodiment of the invention, multiple independent decision trees are constructed to perform classification tasks, such as... Figure 3 As shown, each decision tree is trained based on a different subset of data, and the final diagnostic result is determined by integrating the prediction results of all decision trees. Simultaneously, this embodiment of the invention significantly enhances generalization ability and reduces the risk of overfitting by actively selecting key training samples and features and combining them with a bootstrap method to generate diverse training sets, ensuring diagnostic stability in scenarios such as high salt spray and load fluctuations on ships. Figure 4 As shown, the specific process is as follows: (1) Based on the preset number of independent decision trees in the active decision tree model, the corresponding number of independent datasets are constructed by randomly sampling with replacement from the health status test dataset using the bootstrap method, and each dataset corresponds to a separate independent decision tree.

[0037] (1.1) Set boundary weights for fault boundary samples in the health status test dataset, where the boundary weights are greater than 1.

[0038] Prioritizing samples at the fault boundary (such as critical overload, early insulation aging data, etc.) improves the model's ability to identify minor faults and solves the problem of "balanced but inefficient samples" in traditional ensemble models.

[0039] (1.2) Count the number of samples in the normal state and the number of samples in the faulty state in the health status test dataset. When the difference between the two exceeds a threshold, assign weights to all samples in the faulty state in the health status test dataset according to the difference in proportions. The weights are greater than 1.

[0040] When the difference between the proportion of normal and faulty samples exceeds a threshold, the weight of minority class samples is automatically increased to avoid the model biased in predicting the majority class (normal state), thus solving the problem of "high false negative rate" in traditional decision tree models.

[0041] (1.3) Independent datasets are obtained by sampling with replacement based on the health status dataset after the above weight adjustment.

[0042] Each decision tree corresponds to an independent dataset. If n training sets are generated, then n decision trees are matched to ensure the diversity of training data and improve the robustness of the model.

[0043] (2) For each independent decision tree, its corresponding independent dataset is used as the root node data.

[0044] Input the independent dataset into the corresponding independent decision tree, using the independent dataset as the root node data.

[0045] (3) Calculate the information gain or Gini coefficient of each feature in the independent dataset, evaluate the fault discrimination of each feature, and select the feature with the largest information gain or Gini coefficient as the split point of the independent decision tree corresponding to the independent dataset for splitting.

[0046] The optimal feature is selected by calculating the information gain or Gini coefficient of each feature.

[0047] Let the independent dataset be D , K This represents the total number of system health status categories (including normal status and various types of power faults). The total number of samples in the dataset. For dataset D The middle belongs to the first k Number of samples in the health status category For the first k The percentage of samples in a state of near-health (empirical probability).

[0048] Based on independent datasets, calculate feature information entropy ( )for: Among them: the agreement is hour, The physical meaning of information entropy is the dataset. D The uncertainty in the distribution of health status in China, with a range of values ​​of [value missing]. ; The closer to 0, the more concentrated the distribution of health status in the dataset (the higher the purity). The more disordered the distribution, the lower the purity.

[0049] Feature conditional entropy ( ),set up A For any monitoring feature in the health status test dataset, V Features A The number of different values ​​(or the number of discretized intervals). Features A Take the first v The value (or belonging to the first value)v A subset of samples from (intervals), For subset The number of samples, For subset Weights in the total dataset D, subsets The entropy is ,(in , For subset The middle belongs to the first k If the number of samples in the health state category is given, then the conditional entropy corresponding to feature A is: Its physical meaning is the residual uncertainty in the distribution of health status in dataset D after knowing the values ​​of feature A. The smaller the value, the stronger the explanatory power of feature A in terms of health status. Information gain ( The entropy is defined as the difference between information entropy and conditional entropy, i.e.: Its physical meaning is through characteristics A The degree to which the uncertainty of the health status distribution is reduced after splitting the dataset; The larger the value, the stronger the feature. A The stronger the ability to distinguish the health status of the system, the more suitable it is as a node splitting feature of an independent decision tree.

[0050] The Gini coefficient, or Gini impurity, is: Its physical meaning is the probability that two samples randomly drawn from dataset D have different health states, and its value ranges from 1 to 2. ; The closer to 0, the more concentrated the distribution of the health status of the dataset (higher purity); the closer to 1, the more chaotic the distribution (lower purity).

[0051] (4) Divide the independent dataset layer by layer, recursively calculate the child nodes until the stopping condition is met, and generate the construction of the independent decision tree model.

[0052] The stopping condition is when a subset of samples are classified into the same category (fault type or normal state) or the tree reaches its maximum preset depth. This embodiment of the invention does not perform pruning during training, preserving the original discriminative ability of the decision tree, while actively reducing redundant information interference through feature selection.

[0053] (5) Repeat steps (2)-(5) to generate a preset number of independent decision trees.

[0054] (6) Integrate all independent decision trees to form an active decision tree ensemble model.

[0055] The final diagnostic result is obtained by weighting and summing the outputs of each independent decision tree based on either voting or error.

[0056] All decision trees are integrated to form an active decision tree ensemble model, which is stored in a model library for subsequent use in diagnostics.

[0057] The data acquisition module 103 collects real-time health status data of the shipborne power supply system of the node to be diagnosed for real-time health status diagnosis.

[0058] This module integrates the TI INA238 high-precision power monitoring chip and has a built-in temperature sensor. It can accurately collect data such as voltage, current and temperature of the shipboard power supply system to be diagnosed, and calculate the power in real time based on the collected voltage and current values ​​to characterize the system load status.

[0059] The diagnostic module 104 is used to input the collected health status data of the shipborne power supply system into the active decision tree model and output the diagnostic results.

[0060] During diagnosis, real-time health status data of the shipboard power supply system of the node to be diagnosed will be collected and input into the active decision tree ensemble model. The model randomly selects a subset of key features from all features, and each decision tree independently outputs category labels (such as "short circuit fault", "insulation aging", "normal" etc.) based on the subset data. Finally, through a voting mechanism, the category label with the most votes is output as the diagnosis result, so as to achieve accurate fault identification.

[0061] In the scenario of diagnosing the health status of shipboard power supply systems, through the above process, relying on the active decision tree ensemble model to learn the patterns of massive historical operating data, it is possible to quickly determine whether a node is faulty and the type of fault based on the real-time characteristic data of any node, thus meeting the ship's requirements for real-time and accurate diagnosis.

[0062] The display module is used to display the health status data, diagnostic results, and alarm information based on the diagnostic results of the ship's power supply system in real time.

[0063] This module uses a PowerView 380 LCD display screen, which can display system health status data such as voltage, current, temperature and load power output by the data acquisition module in real time, and simultaneously present the system health diagnosis results output by the diagnostic module. When the data acquisition module triggers an alarm signal or the diagnostic module determines that the system is abnormal, alarm or shutdown prompt information can be output through audio and visual linkage.

[0064] This invention also provides an active decision tree integrated diagnostic method for the health status of a shipborne power supply system, comprising the following steps: S201. Obtain the health status dataset of the shipboard power supply system, including the health status data of the shipboard power supply system and the corresponding health status categories. The health status categories include normal status labels and various fault status types.

[0065] S202. Based on the health status dataset of the shipboard power supply system, construct an active decision tree model. The active decision tree model includes multiple independent decision trees, and use the bootstrap method to generate corresponding independent datasets and generate multiple independent decision trees.

[0066] S203. Real-time acquisition of health status data of the shipboard power supply system to be diagnosed.

[0067] S204. Input the collected health status data of the shipborne power supply system into the active decision tree model and output the diagnostic results.

[0068] This invention employs 1000 data points related to the health status diagnosis of power supply system nodes for preliminary verification and testing of the model, ensuring that the module can fully learn data patterns. The dataset contains four core features (all key parameters collected in real time by the module): voltage, current, load, and temperature; the target variable is fault type.

[0069] To ensure that the active decision tree ensemble model can fully learn data features and effectively test its generalization ability, the module divides 1000 data points into a training set and a test set in an 8:2 ratio: the training set contains 800 samples, used for the training process of the active decision tree ensemble model, providing sufficient data support for the model to learn the characteristics and patterns of the power supply system's health status; the test set contains the remaining 200 samples, used to evaluate the model's predictive performance on new data after training, objectively verify the model's accuracy on untouched data, avoid model overfitting, and ensure the reliability of the module's diagnostics.

[0070] To ensure the reliability and repeatability of the training and testing results of the active decision tree ensemble model, the module presets the following core parameters during training and testing to ensure stable model performance: (1) Maximum depth of decision tree: The parameter is set to 5. This value is used to limit the complexity of each decision tree in the algorithm. If the maximum depth of the decision tree is too large, it will cause the decision tree to contain too many split nodes, which will easily cause the model to overfit. If the depth is too small, the decision tree structure will be too simple, which may cause the model to underfit and not be able to fully learn the data features. Setting it to 5 can achieve a balance between complexity and fitting effect.

[0071] (4) Select the number of features: The parameter setting rule is "sqrt" (i.e., the square root of the total number of features in the dataset). Since the total number of features in the dataset used in this module is 4, each decision tree will randomly select 2 features to form a feature subset during the node splitting process. This setting can further reduce the correlation between decision trees and improve model diversity.

[0072] (5) Minimum number of samples for leaf nodes: The parameter is set to 2. This parameter controls the minimum number of samples that a leaf node in a decision tree needs to contain. If the number of samples is less than 2, the node will no longer be a leaf node and needs to be split. This parameter can prevent the decision tree from overfitting due to too few samples in the leaf nodes.

[0073] (6) Minimum number of samples for splitting: The parameter is set to 2. This parameter controls the minimum number of samples required for each node in the decision tree to perform a splitting operation. If the number of samples contained in a node is less than 2, the node will not be split and will be directly used as a leaf node. This setting can balance the complexity and generalization ability of the decision tree.

[0074] (7) Number of decision trees: The parameter is set to 50. This parameter determines the total number of decision trees in the active decision tree ensemble model. When the model generates 50 independent decision trees, the training process will automatically stop. 50 decision trees can ensure the diagnostic accuracy of the model while avoiding the reduction in training efficiency caused by too many decision trees, thus meeting the real-time diagnostic needs of the module.

[0075] After training, the performance metrics output by the active decision tree ensemble model on the test set are as follows: Figure 5 As shown, the diagnostic effectiveness of the module is fully verified. Support represents the actual number of samples appearing in the test set for each category. Therefore, in a test set of 200 samples, the model had 100 faults and 100 no faults. The model achieved 100% accuracy and recall on this test set, demonstrating high performance in predicting and identifying the positive class.

[0076] To further analyze the model's classification performance, a confusion matrix of the model on the test set was plotted, as shown below. Figure 6 As shown: The confusion matrix shows that the model can classify both faulty samples (label 1) and fault-free samples (label 0) 100% correctly without any false alarms, proving that the active decision tree ensemble model has extremely high performance on the test set after being trained on a dataset of 1000 samples.

[0077] In addition to the confusion matrix, the model also outputs a feature importance distribution map, such as... Figure 7 As shown: Feature importance represents the contribution of each feature to the model's predictions. In tree models such as active decision tree ensembles and gradient boosting trees, feature importance is typically calculated by measuring the reduction in error each feature reduces during tree splitting. A higher feature importance value indicates that the feature plays a more significant role in the decision tree splitting process. Among the four features—voltage, current, load, and temperature—current has the highest feature importance, meaning that current plays a dominant role in the model's predictions during active decision tree ensemble training. Voltage has the lowest feature importance, so low as to be negligible, meaning its contribution to the model's predictions is very small.

[0078] according to Figure 7 (Feature Importance Distribution Chart) It can be seen that for the dataset in this case, temperature and voltage are of low importance, while current is the most important feature. Changes in current can better help distinguish different types of faults. By analyzing feature importance, we can clarify which features significantly contribute to the performance of the active decision tree ensemble model. This provides a clear direction for further model optimization (such as focusing on core features to improve acquisition accuracy and simplifying the acquisition process of secondary features). It can also help the diagnostic module reduce the data processing load while ensuring diagnostic accuracy, making it more suitable for complex ship operating conditions.

[0079] To further verify the long-term operational stability of the device of the present invention under actual ship operating conditions, a 72-hour continuous health status test was conducted on a shipboard power supply system. The test covered the three core sub-modules of the power supply side, power consumption side, and propulsion side. The monitoring characteristics were four types of parameters: current, voltage, load, and temperature, which are the key inputs of the model. The test results are shown in Table 1.

[0080] Table 1 As shown in Table 1, the device of the present invention, through the collaborative work of the acquisition module and the active decision tree integrated diagnostic model, achieves 100% accurate identification and status determination of the four core parameters of the three major modules within a 72-hour continuous monitoring period. All monitoring indicators are within the qualified range, with no missed or false alarms. This solves the technical pain point of traditional diagnostic devices being unable to operate reliably for a long time and ensures the safety of the power system during long-term ship voyages.

[0081] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the principles and spirit of the present invention should be included within the protection scope of the present invention.

Claims

1. A proactive decision tree integrated diagnostic device for the health status of a shipborne power supply system, applied to the power supply side, power consumption side, and propulsion side of a shipborne power supply system, characterized in that... include: The dataset construction module is used to obtain the health status dataset of the shipboard power supply system, including the health status data of the shipboard power supply system and the corresponding health status categories. The health status categories include normal status labels and various fault status types. The model building module is used to build an active decision tree model based on the health status dataset of the shipboard power supply system. The active decision tree model includes multiple independent decision trees, and the corresponding independent datasets are generated by using the bootstrap method to generate multiple independent decision trees. The data acquisition module is used to collect real-time health status data of the shipboard power supply system to be diagnosed. The diagnostic module is used to input the collected health status data of the shipboard power supply system into the active decision tree model and output the diagnostic results.

2. The shipborne power supply system health status active decision tree integrated diagnostic device according to claim 1, characterized in that, The acquisition of health status test data of the shipborne power supply system and the construction of a health status test dataset include: Preprocessing the health status test data of the shipboard power supply system includes cleaning noise and processing missing values ​​in the health status test data. The preprocessed health status test data of the shipborne power supply system is normalized to eliminate the dimensional differences of different characteristic factors.

3. The shipborne power supply system health status active decision tree integrated diagnostic device according to claim 1, characterized in that, The active decision tree model is constructed using the health status test dataset as follows: (2.1) Based on the preset number of independent decision trees in the active decision tree model, a corresponding number of independent datasets are constructed by randomly sampling with replacement from the health status test dataset using the bootstrap method, and each dataset corresponds to a separate independent decision tree. (2.2) For each independent decision tree, its corresponding independent dataset is used as the root node data; (2.3) Calculate the information gain or Gini coefficient of each feature in the independent dataset, evaluate the fault discrimination of each feature, and select the feature with the largest information gain or Gini coefficient as the split point of the independent decision tree corresponding to the independent dataset for splitting; (2.4) Divide the independent dataset layer by layer, recursively calculate the child nodes until the stopping condition is met, and generate the construction of the independent decision tree model; (2.5) Repeat steps (2.2)-(2.5) to generate a preset number of independent decision trees; (2.6) The active decision tree ensemble model is formed by integrating all the independent decision trees.

4. The shipborne power supply system health status active decision tree integrated diagnostic device according to claim 3, characterized in that, The self-help method is specifically as follows: For the fault boundary samples in the health status test dataset, a boundary weight is set, wherein the boundary weight is greater than 1; The number of samples in the normal state and the number of samples in the faulty state in the health status test dataset are counted. When the difference between the two exceeds a threshold, a weight is assigned to all samples in the faulty state in the health status test dataset according to the difference in the proportion. The weight is greater than 1. The independent dataset is obtained by sampling with replacement based on the health status test dataset after the above weight adjustment.

5. The shipborne power supply system health status active decision tree integrated diagnostic device according to claim 4, characterized in that: The fault boundaries include at least critical overload, early stage of insulation aging, voltage overshoot threshold, frequency fluctuation threshold, and harmonic distortion threshold.

6. The shipborne power supply system health status active decision tree integrated diagnostic device according to claim 3, characterized in that, The stopping condition is when the child node samples are classified into the same category or the tree reaches the maximum preset depth.

7. The shipborne power supply system health status active decision tree integrated diagnostic device according to claim 3, characterized in that, The integration of all independent decision trees specifically involves obtaining the final diagnostic result by combining the outputs of each independent decision tree through voting or weighted summation.

8. The shipborne power supply system health status active decision tree integrated diagnostic device according to claim 1, characterized in that, Also includes: The display module is used to display the health status data, diagnostic results, and alarm information based on the diagnostic results of the ship's power supply system in real time.

9. A proactive decision tree integrated diagnostic method for the health status of a shipborne power supply system, characterized in that, Includes the following steps: (1) Obtain the health status dataset of the shipboard power supply system, including the health status data of the shipboard power supply system and the corresponding health status categories, wherein the health status categories include normal status labels and various fault status types; (2) Based on the health status dataset of the shipboard power supply system, an active decision tree model is constructed. The active decision tree model includes multiple independent decision trees, and the corresponding independent dataset is generated by the bootstrap method to generate multiple independent decision trees. (3) Real-time collection of health status data of the shipboard power supply system to be diagnosed; (4) Input the collected health status data of the shipboard power supply system into the active decision tree model and output the diagnostic results.

Images