Ship power distribution system health assessment method, system and product

Abstract

The invention discloses a ship power distribution system health assessment method, system and product, and the method comprises the steps: firstly obtaining all equipment operation parameters and historical data of a ship power distribution system, carrying out the data preprocessing, and adding a working condition mode identifier for the preprocessed data according to different working conditions of the data; extracting preprocessed data features through a deep learning model, and identifying a working condition label; obtaining a characteristic parameter matrix lambda and a characteristic weight matrix W; then, on the basis of the characteristic parameter matrix lambda, reference matrixes of health data characteristics under different working conditions are calculated, and a multi-working-condition health reference library is constructed; respectively calculating the health degree score under each working condition through the feature weight matrix W and the feature deviation degree based on the health benchmark; and finally, dividing power distribution system health degree scores, and determining the current system health degree. According to the method, the multi-working-condition adaptability and the feature expression capability are remarkably improved, and a full-chain solution from real-time state sensing to health assessment is provided for the ship power distribution system.

Description

Technical Field

[0001] The present invention belongs to the field of ship safety technology, and specifically relates to a ship power distribution system health assessment method, system and product. Background Art

[0002] Ship power distribution systems are a key component in ensuring a stable and reliable power supply. Compared to conventional power systems, ship power distribution systems face challenges such as frequent load fluctuations and complex operating mode switching. Traditional monitoring methods based on threshold alarms struggle to effectively capture potential fault characteristics, while regular maintenance strategies suffer from delayed response times and high operational and maintenance costs. Therefore, dynamic and refined health assessments are urgently needed.

[0003] In recent years, data-driven health assessment technologies have been increasingly applied in the power sector, with the integration of multidimensional feature mining and intelligent algorithms becoming a research hotspot. For example, a research team from China Huaneng Group proposed a photovoltaic power generation health assessment method based on principal component analysis and Mahalanobis distance (Wang Jing, Ren Xin, Liu Xingwei, et al. Photovoltaic Power Generation Health Assessment Method and Device [P]. Chinese Patent: CN202311428291 A: 2024-01-23). ​​This method uses dimensionality reduction to construct a feature mapping model, enabling quantitative classification of component status. A research team from Nanjing University of Information Science and Technology also developed a unit health assessment system based on a GRU network (Xiong Liang, Sun Chengzhi, Song Jianfeng, et al. A Unit Health Assessment Method Based on Deep Learning [P]. Chinese Patent: CN202510408872 A: 2025-05-02). This system uses a time series prediction model combined with the quartile method to assess the overall health of equipment.

[0004] However, existing technologies still have significant limitations: first, traditional dimensionality reduction methods are insufficient in extracting dynamic correlation features of multi-operating condition data, resulting in increased evaluation deviation when operating conditions switch; second, a single model is difficult to take into account the feature differences of the system under different operating conditions, and lacks feature extraction of key features and dynamic weight allocation mechanism. Summary of the Invention

[0005] To address the shortcomings of existing research, this paper provides a method, system, and product for assessing the health of a ship's power distribution system. By incorporating domain knowledge-guided operating condition classification and combining it with a deep learning network model, deep feature extraction from operational data is performed to deeply explore the system's health status. A self-attention layer dynamically weights the characteristics of different operating conditions, improving the accuracy of health assessments for ship power distribution systems.

[0006] The technical solution adopted by the method of the present invention is: a method for health assessment of a ship power distribution system, comprising the following steps: Step 1: Obtain the operating parameters and historical data of various equipment in the ship's power distribution system; Step 2: Pre-process the operating parameters of various equipment in the ship's power distribution system, and add operating mode identifiers to the pre-processed data according to the different operating conditions to which the data belongs; Step 3: Extract the pre-processed data features through the deep learning network model, identify the working condition label; obtain the feature parameter matrix l , feature weight matrix W; Step 4: Based on the characteristic parameter matrix l , calculate the benchmark matrix of health data characteristics under different working conditions and build a multi-working condition health benchmark library; Step 5: Calculate the health score under each working condition using the feature weight matrix W and the feature deviation based on the health benchmark; Step 6: Assign a health rating level to the distribution system and determine the current system health.

[0007] Preferably, in step 1, the operating parameters of various equipment in the ship's DC power distribution system include operating parameters of generators, distribution boards, power transformers from the DC distribution boards to the AC distribution boards, converters, and propulsion motors.

[0008] Preferably, in step 2, the data preprocessing includes data denoising, missing value filling and data normalization.

[0009] Preferably, the adding of working mode identification to the pre-processed data is to add corresponding working condition labels to the data under different working conditions, and the labels are divided according to the load rate status of the ship at that time.

[0010] Preferably, the deep learning network model includes a transformer branch and an LSTM branch arranged in parallel, and a self-attention layer, a fully connected layer and a Softmax activation layer; The transformer branch adds position encoding to the input data and then passes through the multi-head attention layer and feedforward network in sequence to extract global features; In the LSTM branch, input data passes through the LSTM unit group and the timing gating operation layer in sequence to extract local data; The global features and local features are fused by a feature fusion layer using a weighted splicing method, and then sequentially pass through the self-attention layer, the fully connected layer and the Softmax activation layer, and the final classification result is output through the softmax classifier.

[0011] Preferably, in step 4, in the multi-operating condition health assessment, the health benchmark for each operating condition is composed of the following statistics: (1) Mean vector ， ; Among them, the mean vector mIndicates the characteristic distribution center of the health status under the current working conditions; N is the number of samples of health data under the current working conditions, For the i The feature vector of healthy samples; ,in d is the dimension of the feature vector; (2) Covariance matrix , ; Among them, the covariance matrix represents the correlation between the characteristics in the health state, , diagonal elements Indicates the i The variance of the features, off-diagonal elements Indicates the i and j The covariance of the features; (3) Inverse covariance matrix From the covariance matrix Taking the inverse, we get , used to calculate the Mahalanobis distance and eliminate redundant correlations between features; Finally, an independent benchmark library for each working condition is formed .

[0012] Preferably, in step 5, the feature deviation is calculated by combining the feature weight matrix W and the Mahalanobis distance to obtain the current device health score; The specific calculation process is as follows: (1) Calculate vector difference ; (2) Calculate the product of the dynamic weight matrix W and the inverse covariance matrix ; ; in, Indicates the weight matrix d Rank d The elements of the column represent the d The autocorrelation weight of each feature; (3) Calculate the weighted difference vector ; ; in, Indicates the current sample i The difference between the feature and the corresponding working condition health benchmark mean, that is, i The vector difference of the features ; (4) Calculate the final deviation D The smaller it is, the closer it is to a healthy state; DThe bigger it is, the worse the health status is; (5) Departure D Obtain the current device health score S by normalizing and mapping it to a health score; ; in, They represent the maximum deviation and the minimum deviation respectively.

[0013] Preferably, in step 6, the health rating of the power distribution system is graded based on expert knowledge and a threshold value determined by cluster analysis in combination with various operating states of the system and historical operating data.

[0014] The technical solution adopted by the system of the present invention is: a ship power distribution system health assessment system, comprising: one or more processors; A storage device is used to store one or more programs, which, when executed by the one or more processors, enable the one or more processors to implement the ship power distribution system health assessment method.

[0015] The technical solution adopted by the product of the present invention is: a ship power distribution system health assessment product, including computer program instructions, when the computer program instructions are run on a computer, the computer executes the ship power distribution system health assessment method.

[0016] Compared with the prior art, the beneficial effects of the present invention include: (1) This invention constructs a multi-operating condition labeling system based on the system load rate, which solves the evaluation distortion problem caused by changes in operating conditions; (2) The present invention constructs a transformer-LSTM deep learning network model to extract key system features. Through different working condition labels, it performs differentiated benchmark calculations for each working condition feature and constructs a multi-working condition feature health benchmark library. This effectively solves the evaluation distortion problem caused by changes in working conditions. (3) The present invention integrates the Mahalanobis distance deviation calculation and the dynamic weight allocation of the self-attention layer to calculate the system health score; the health score calculated by combining the dynamic weight allocation is more consistent with the impact difference of the features and better reflects the degree of deviation of the system from health; (4) The present invention constructs a five-level health mapping rule that combines health scores with expert knowledge, converts data feature deviations into system health scores for system health rating, and more intuitively reflects the health status of the system; (5) The present invention has significantly improved the adaptability to multiple working conditions and the ability to express characteristics, providing a full-chain solution from real-time status perception to health assessment for the ship's power distribution system. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The technical solution of the present invention is further illustrated below using embodiments and specific implementation methods. In addition, some drawings are also used in the process of illustrating the technical solution. For those skilled in the art, other drawings and the intention of the present invention can be obtained based on these drawings without making any creative efforts.

[0018] Figure 1 is a method principle diagram of an embodiment of the present invention; Figure 2 It is a structural diagram of the transformer-LSTM deep learning network model in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to facilitate ordinary technicians in this field to understand and implement the present invention, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the implementation examples described herein are only used to illustrate and explain the present invention and are not used to limit the present invention.

[0020] Please see Figure 1 This embodiment provides a method for evaluating the health of a ship power distribution system, comprising the following steps: Step 1: Obtain the operating parameters and historical data of various equipment in the ship's power distribution system; In one embodiment, the operating parameters of the various equipment in the ship's power distribution system mainly include operating parameters of generators, distribution boards, power transformers from DC distribution boards to AC distribution boards, converters, propulsion motors, etc.

[0021] Generator parameters include: stator status such as winding inter-turn insulation status, rotor status such as rotor balance, inter-turn insulation status, eccentricity status, bearing wear status, excitation device and automatic voltage regulator excitation and voltage regulation capabilities, permanent magnet demagnetization phenomenon, etc.

[0022] The distribution board parameters include: the opening and closing status of the busbar disconnect solid-state switch, the opening and closing status of each panel circuit breaker, the busbar power supply quality, such as voltage, current, power, harmonics, etc., the busbar insulation status, the suppression capacitor and inductor working status, such as charging and discharging action, temperature, etc.

[0023] The parameters of the power transformer from the DC distribution board to the AC distribution board include: winding working status.

[0024] Converter parameters include: working status of power device modules.

[0025] The propulsion motor parameters include: stator status such as winding inter-turn insulation status, rotor status such as rotor balance, inter-turn insulation status, eccentricity status, bearing wear status, etc.

[0026] Step 2: Pre-process the operating parameters of various equipment in the ship's power distribution system, and add operating mode identifiers to the pre-processed data according to the different operating conditions to which the data belongs; In one embodiment, data preprocessing includes data cleaning tasks such as data denoising, missing value filling, and data normalization of the original data to ensure the accuracy and consistency of the data and provide a reliable foundation for subsequent analysis and model training.

[0027] According to the different ship load rates, it is divided into 4 working condition labels: (1) Working condition 1: Waiting at the port, the load rate is 0.

[0028] (2) Working condition 2: low-load sailing, load rate 1-40%.

[0029] (3) Working condition three: medium load navigation, load rate 41%-80%.

[0030] (4) Working condition 4: high-load navigation, load rate 81%-100%.

[0031] The ship is divided into four working condition labels according to its load rate. Different features are extracted for different working conditions in the feature extraction stage to improve the model's ability to capture working condition-sensitive features.

[0032] Step 3: Extract the pre-processed data features through the deep learning model, identify the working condition label, and obtain the feature parameter matrix l , feature weight matrix W; Please see Figure 2 In one embodiment, the deep learning network model includes a transformer branch and an LSTM branch arranged in parallel, and a self-attention layer, a fully connected layer and a Softmax activation layer; The transformer branch adds position encoding to the input data and then passes through the multi-head attention layer and feedforward network in sequence to extract global features; In the LSTM branch, input data passes through the LSTM unit group and the timing gating operation layer in sequence to extract local data; The global features and local features are fused by a feature fusion layer using a weighted splicing method, and then sequentially pass through the self-attention layer, the fully connected layer and the Softmax activation layer, and the final classification result is output through the softmax classifier.

[0033] The deep learning network model is an improved transformer-LSTM hybrid model. Unlike the classic model where the two parts of the structure run sequentially, the parallel structure uses two branches to extract complementary features, enhancing the model's ability to express complex data patterns.

[0034] This embodiment uses parallel branches to extract features. At the feature fusion layer, the two features are fused by weighted splicing. The fusion formula is as follows:

[0035] in, Indicates splicing by channel, is the feature after splicing, Extract features for the transformer branch, Extract features for LSTM branches.

[0036] This embodiment adds a self-attention layer after the feature fusion layer, assigns weights to the feature matrix after feature fusion through the self-attention mechanism, and strengthens key features.

[0037] In this embodiment, the weighted features of the self-attention layer are connected to the fully connected layer, and then the final classification result is output through the softmax classifier.

[0038] In one embodiment, the deep learning network model is a trained model that ensures a high level of recognition accuracy for operating conditions on a validation set. After the model converges (accuracy is considered converged when it stabilizes and fluctuates less than 1%), model features are extracted for subsequent calculations.

[0039] Step 4: Based on the characteristic parameter matrix l , calculate the benchmark matrix of data characteristics of health data under different working conditions (health data must be collected when the system is in a fault-free and normal operating state, and verified by experts to exclude potential abnormalities) and build a multi-working condition health benchmark library; In one embodiment, the health baseline is calculated for each feature group based on the different working condition labels. In the multi-working condition health assessment, the health baseline of each working condition is composed of the following statistics: (1) Mean vector m; ; ; Mean vector m Represents the characteristic distribution center of the health status under the current working conditions. N is the number of samples of health data under the current working conditions, For the i The feature vector of healthy samples. ,in d is the dimension of the feature vector.

[0040] (2) Covariance matrix ; ; ; Covariance matrix Indicates the correlation between various characteristics under health status. , where the diagonal elements Indicates the i The variance of the features. Off-diagonal elements Indicates the i and j The covariance of the features.

[0041] (3) Inverse covariance matrix ; Inverse covariance matrix From the covariance matrix Taking the inverse, we get , used to calculate the Mahalanobis distance and eliminate redundant correlations between features.

[0042] Finally, an independent benchmark library for each working condition is formed: .

[0043] Step 5: Calculate the health score under each working condition using the feature weight matrix W and the feature deviation based on the health benchmark; In one embodiment, the feature weight matrix W and Mahalanobis distance are combined to calculate the feature deviation to obtain the current device health score; The specific calculation process is as follows: (1) Calculate vector difference ; (2) Calculate the product of the dynamic weight matrix W and the inverse covariance matrix ; ; in, Indicates the weight matrix d Rank d The elements of the column represent the d The autocorrelation weight of each feature.

[0044] (3) Calculate the weighted difference vector ; ; in, Indicates the current sample i The difference between the feature and the corresponding working condition health benchmark mean, that is, i The vector difference of the features .

[0045] (4) Calculate the final deviation D The smaller it is, the closer it is to a healthy state; D The bigger it is, the worse the health status is; (5) Departure D Obtain the current device health score S by normalizing and mapping it to a health score; ; in, They represent the maximum deviation and the minimum deviation respectively.

[0046] Step 6: Assign a health rating level to the distribution system and determine the current system health.

[0047] In one embodiment, the health rating of the power distribution system is based on expert knowledge and is determined by cluster analysis based on thresholds determined by combining the system's operating status and historical operating data. The health rating is divided into five levels: healthy, sub-healthy, abnormal, faulty, and scrapped. The specific health ratings are as follows: Health level 0.9-1.0: The system is in a healthy state, operating normally, with no abnormal signs.

[0048] Health level 0.8-0.9, the system is in sub-healthy state, operating normally, with slight abnormal signs.

[0049] Health level is 0.7-0.8. The system is in an abnormal state, system performance has degraded, and there are minor faults. Troubleshooting and repair are required.

[0050] Health level is 0.4-0.7, the system is in a faulty state, system performance is poor, and system shutdown for maintenance is required.

[0051] If the health level is below 0.4, the system is in a failed state and cannot operate, and needs to be completely replaced.

[0052] This embodiment also provides a ship power distribution system health assessment system, including: one or more processors; A storage device is used to store one or more programs, which, when executed by the one or more processors, enable the one or more processors to implement the ship power distribution system health assessment method.

[0053] This embodiment also provides a ship power distribution system health assessment product, including computer program instructions. When the computer program instructions are run on a computer, the computer executes the ship power distribution system health assessment method.

[0054] It should be understood that the embodiments described above are only some of the embodiments of the present invention, rather than all of the embodiments. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention may be arbitrarily combined with each other to form a feasible technical solution. Such combination is not restricted by the order of steps and / or structural composition mode, but must be based on the ability of ordinary technicians in this field to implement it. When the combination of technical solutions is mutually inconsistent or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0055] It should be understood that the above description of the preferred embodiment is relatively detailed and cannot be regarded as limiting the scope of protection of the patent of the present invention. Under the guidance of the present invention, ordinary technicians in this field can also make substitutions or modifications without departing from the scope of protection of the claims of the present invention, which all fall within the scope of protection of the present invention. The scope of protection requested by the present invention shall be based on the attached claims.

Claims

1. A method for health assessment of a ship power distribution system, characterized in that: The following steps are involved: Step 1: Obtain the operating parameters and historical data of various equipment in the ship's power distribution system; Step 2: Pre-process the operating parameters of various equipment in the ship's power distribution system, and add operating mode identifiers to the pre-processed data according to the different operating conditions to which the data belongs; Step 3: Extract pre-processed data features through the deep learning network model and identify the working condition label; Get the characteristic parameter matrix λ , feature weight matrix W; Step 4: Based on the characteristic parameter matrix λ , calculate the benchmark matrix of health data characteristics under different working conditions and build a multi-working condition health benchmark library; Step 5: Calculate the health score under each working condition using the feature weight matrix W and the feature deviation based on the health benchmark; Step 6: Assign a health rating level to the distribution system and determine the current system health.

2. The method for health assessment of a ship power distribution system according to claim 1, characterized in that: In step 1, the operating parameters of various equipment in the ship's DC power distribution system include the operating parameters of the generator, the switchboard, the power transformer from the DC switchboard to the AC switchboard, the converter, and the propulsion motor.

3. The method for health assessment of a ship power distribution system according to claim 1, wherein: In step 2, the data preprocessing includes data denoising, missing value filling and data normalization.

4. The method for health assessment of a ship power distribution system according to claim 1, wherein: In step 2, the adding of working mode identification to the pre-processed data is to add corresponding working condition labels to the data under different working conditions, and the labels are divided according to the load rate status of the ship at that time.

5. The method for health assessment of a ship power distribution system according to claim 1, wherein: In step 3, the deep learning network model includes a transformer branch and an LSTM branch set in parallel, as well as a self-attention layer, a fully connected layer and a softmax activation layer; The transformer branch adds position encoding to the input data and then passes through the multi-head attention layer and feedforward network in sequence to extract global features; In the LSTM branch, input data passes through the LSTM unit group and the timing gating operation layer in sequence to extract local data; The global features and local features are fused by a feature fusion layer using a weighted splicing method, and then sequentially pass through the self-attention layer, the fully connected layer and the Softmax activation layer, and the final classification result is output through the softmax classifier.

6. The method for health assessment of a ship power distribution system according to claim 1, characterized in that: In step 4, in the multi-condition health assessment, the health benchmark for each condition is composed of the following statistics: (1) Mean vector ， ; Among them, the mean vector μ Indicates the characteristic distribution center of the health status under the current working conditions; N is the number of samples of health data under the current working conditions, For the i The feature vector of healthy samples; ,in d is the dimension of the feature vector; (2) Covariance matrix , ; Among them, the covariance matrix represents the correlation between the characteristics in the health state, , diagonal elements Indicates the i The variance of the features, off-diagonal elements Indicates the i and j The covariance of the features; (3) Inverse covariance matrix From the covariance matrix Taking the inverse, we get , used to calculate the Mahalanobis distance and eliminate redundant correlations between features; Finally, an independent benchmark library for each working condition is formed .

7. The method for health assessment of a ship power distribution system according to claim 1, characterized in that: In step 5, the feature deviation is calculated by combining the feature weight matrix W and the Mahalanobis distance to obtain the current device health score; The specific calculation process is as follows: (1) Calculate vector difference ; (2) Calculate the product of the dynamic weight matrix W and the inverse covariance matrix ; ； in, Indicates the weight matrix d Rank d The elements of the column represent the d The autocorrelation weight of each feature; (3) Calculate the weighted difference vector ; ； in, Indicates the current sample i The difference between the feature and the corresponding working condition health benchmark mean, that is, i The vector difference of the features ; (4) Calculate the final deviation D The smaller it is, the closer it is to a healthy state; D The bigger it is, the worse the health status is; (5) Departure D Obtain the current device health score S by normalizing and mapping it to a health score; ； in, They represent the maximum deviation and the minimum deviation respectively.

8. The method for health assessment of a ship power distribution system according to any one of claims 1 to 7, characterized in that: In step 6, the health score of the distribution system is graded based on expert knowledge and the threshold value determined by cluster analysis in combination with the system's operating status and historical operating data.

9. A ship power distribution system health assessment system, characterized in that: include: one or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, enables the one or more processors to implement the ship power distribution system health assessment method according to any one of claims 1 to 8.

10. A ship power distribution system health assessment product, comprising computer program instructions, characterized in that: When the computer program instructions are executed on a computer, the computer is caused to execute the ship power distribution system health assessment method according to any one of claims 1 to 8.

Images