A method and system for predicting the remaining service life of electromechanical equipment across domains based on dynamic parameter learning

Abstract

The present invention discloses a cross-domain remaining life prediction method and system for electromechanical equipment based on dynamic parameter learning, including: obtaining source domain data and target domain data of electromechanical equipment, and preprocessing the source domain data and the target domain data; using a feature extractor to independently pre-train on labeled source domain data, so that the feature extractor has feature extraction capabilities for source domain data; using a domain adaptive method to train the feature extractor, so that the feature extractor can extract effective high-dimensional data features from both the preprocessed source domain data and the target domain data; in the inference stage, only the feature extractor and the classifier are used to predict the remaining life of the electromechanical equipment. Through real-time remaining life prediction, the present invention enables enterprises to formulate preventive maintenance plans in advance, extend the service life of equipment, reduce failure rates and maintenance costs, and improve overall operational efficiency and equipment reliability.

Description

Technical Field

[0001] The present invention belongs to the field of electromechanical equipment prediction, and specifically relates to a method and system for predicting the cross-domain remaining life of electromechanical equipment based on dynamic parameter learning. Background Art

[0002] With the rapid development of industrial automation and intelligent manufacturing, electromechanical equipment is increasingly used in manufacturing, energy, transportation and other fields. The normal operation of these devices is crucial to production efficiency and safety, so predicting the remaining life of the equipment has become a key task in equipment maintenance management. Traditional remaining life prediction methods usually rely on a large amount of labeled data for model training, which means that the historical operating data of the equipment needs to be recorded and analyzed in detail to extract effective features for model construction. However, these methods face many challenges in practical applications. First, different types of electromechanical equipment have significant differences in operating environment and operating conditions, resulting in changes in data distribution. At the same time, even for the same type of equipment, its operating status may change significantly under different working environments. This diversity of data distribution makes it difficult for a single model to maintain high-precision predictions in all scenarios.

[0003] To address these challenges, researchers have begun to explore effective methods for data migration and model adaptation between different domains to improve the accuracy and reliability of cross-domain remaining life prediction. This involves establishing effective associations between the source domain and the target domain so that the model can quickly adapt to the new environment. Transfer learning and domain adaptation techniques have become important methods to solve this problem. By transferring the knowledge acquired in the source domain to the target domain, the need for labeled data in the target domain is reduced. This method can not only improve the prediction performance of the model in the new domain, but also greatly shorten the development and deployment cycle of the model. However, how to design a robust and efficient transfer learning algorithm that can handle the complexity and uncertainty of the equipment operating environment is still a technical problem that needs to be solved urgently. Therefore, the research on cross-domain remaining life prediction not only has important theoretical value, but also has broad practical application prospects. Summary of the invention

[0004] In view of the data distribution difference problem faced by cross-domain remaining life prediction of electromechanical equipment in the prior art, the present invention proposes a cross-domain remaining life prediction method and system for electromechanical equipment based on dynamic parameter learning, which aims to balance the distribution difference between the source domain and the target domain by adaptively adjusting parameters, thereby improving the generalization ability of the model in the target domain. The present invention is particularly suitable for cross-domain remaining life prediction of electromechanical equipment.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A cross-domain remaining life prediction method for electromechanical equipment based on dynamic parameter learning includes the following steps:

[0007] Acquire source domain data and target domain data of electromechanical equipment, and preprocess the source domain data and the target domain data;

[0008] Use the feature extractor to pre-train independently on labeled source domain data, so that the feature extractor has the ability to extract features from the source domain data;

[0009] Using a domain adaptive method to train a feature extractor, so that the feature extractor can extract effective high-dimensional data features from both the preprocessed source domain data and the target domain data;

[0010] In the inference phase, only feature extractors and classifiers are applied to predict the remaining lifetime of electromechanical devices.

[0011] Preferably, using a domain adaptive method to train a feature extractor so that the feature extractor can extract effective high-dimensional data features from both the preprocessed source domain data and the target domain data includes:

[0012] Adopting an adversarial network structure, the feature discriminator is used to determine whether the high-dimensional data features come from the source domain or the target domain;

[0013] The classifier is used to classify the high-dimensional data features extracted from the source domain, and the classification discriminator determines whether the classification result is correct or not;

[0014] The feature discriminator introduces dynamic parameters λ and influence factor κ through a dynamic parameter learning mechanism, and dynamically adjusts the weight between the feature discriminator and the classifier, so that the feature extractor can extract high-dimensional data features in both the source domain and the target domain.

[0015] Preferably, the feature discriminator introduces dynamic parameters λ and influence factor κ through a dynamic parameter learning mechanism, and dynamically adjusts the weight between the feature discriminator and the classifier, so that the feature extractor can extract high-dimensional data features in both the source domain and the target domain, including:

[0016] Calculate the conditional distribution distance D c ;

[0017] Calculate the marginal distribution distance D m ;

[0018] Based on the conditional distribution distance D c and marginal distribution distance D m , calculate the dynamic parameter λ;

[0019] Calculate the impact factor κ;

[0020] Based on the dynamic parameter λ and the influencing factor κ, a comprehensive loss function is constructed.

[0021] Preferably, the conditional distribution distance D is calculated c include:

[0022]

[0023] In the formula, C represents the total number of categories, P s (X|Y=c) and P t (X|Y=c) represents the conditional distribution of the source domain and the target domain under category c, and the weight w c Defined as: Where N c is the number of samples of category c, N total is the total number of samples in all categories, X represents the feature space, and Y represents the label space.

[0024] Preferably, the marginal distribution distance D is calculated m include:

[0025]

[0026] Where: L represents the number of levels, α l is the weight of each layer, usually determined by empirical rules or automatic learning, P s (X) and P t (X) represents the marginal distribution of source domain and target domain respectively.

[0027] Preferably, based on the conditional distribution distance D c and marginal distribution distance D m , the calculation of dynamic parameter λ includes:

[0028]

[0029] Where α and β are weight parameters used to control the adjustment ratio of early and late λ, T is the total number of training rounds, and t is each round.

[0030] Preferably, calculating the impact factor κ includes:

[0031]

[0032] Where γ and δ are weight parameters, which are used to control the adjustment ratio of early and late κ.

[0033] Preferably, based on the dynamic parameter λ and the influencing factor κ, constructing a comprehensive loss function includes:

[0034] L=L Cls +(1-κλ)·L Dism +λ·L Disc

[0035] Where, L Cls is the classification loss, L Dism is the feature discriminator for the marginal distribution alignment loss, L Disc Conditional distribution alignment loss for the feature discriminator.

[0036] The present invention also provides a cross-domain remaining life prediction system for electromechanical equipment based on dynamic parameter learning, comprising: a data collection and preprocessing module, a feature extraction module, a domain adaptation process module, and a remaining life prediction module;

[0037] The data collection and preprocessing module is used to obtain source domain data and target domain data of electromechanical equipment, and preprocess the source domain data and the target domain data;

[0038] The feature extraction module is used to use the feature extractor to independently perform pre-training on the labeled source domain data, so that the feature extractor has the ability to extract features from the source domain data;

[0039] The domain adaptation process module is used to train the feature extractor using a domain adaptation method, so that the feature extractor can extract effective high-dimensional data features from both the preprocessed source domain data and the target domain data;

[0040] The remaining life prediction module is used to predict the remaining life of the electromechanical equipment by only applying the feature extractor and the classifier during the inference stage.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] The present invention proposes a cross-domain remaining life prediction method for electromechanical equipment based on dynamic parameter learning, which significantly improves the intelligence and precision level of equipment maintenance management. By adopting the adversarial migration algorithm, this method effectively reduces the data distribution difference between the source domain and the target domain, and significantly improves the generalization ability and prediction accuracy of the model in different fields. The introduction of the dynamic parameter learning mechanism enables the model to adaptively balance the alignment strength of the marginal distribution and the conditional distribution, ensuring that the distribution difference is reduced and the classification performance of the source domain is maintained during the cross-domain migration process. The modular design makes the feature discriminators of each component highly flexible and scalable, which is convenient for optimization and upgrading according to actual needs. Through remaining life prediction, enterprises can formulate preventive maintenance plans in advance, extend the service life of equipment, reduce failure rate and maintenance costs, and improve overall operational efficiency and equipment reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0044] Figure 1 A block diagram of a model training system according to an embodiment of the present invention;

[0045] Figure 2 A block diagram of an inference training system according to an embodiment of the present invention;

[0046] Figure 3 A training process block diagram of a cross-domain remaining life prediction system for electromechanical equipment based on dynamic parameter learning according to an embodiment of the present invention;

[0047] Figure 4 The present invention is a flowchart of a method for predicting the remaining useful life of electromechanical equipment across domains based on dynamic parameter learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] Embodiment 1

[0051] In view of the data distribution difference problem faced by cross-domain remaining life prediction of electromechanical equipment in the prior art, this paper proposes a multi-domain adaptive method based on dynamic parameter learning, which aims to balance the distribution difference between the source domain and the target domain by adaptively adjusting parameters, thereby improving the generalization ability of the model in the target domain. This method is particularly suitable for cross-domain remaining life prediction of electromechanical equipment, and specifically includes the following aspects:

[0052] The domain adaptation method model of the present invention is composed of a feature extractor, a feature discriminator, a classifier and a classification discriminator, forming an adversarial network structure.

[0053] During training, the feature extractor is responsible for extracting generalizable feature representations from the data in the source domain and the target domain; the feature discriminator is used to determine whether the feature comes from the source domain or the target domain; the classifier performs supervised learning on the source domain to ensure the high accuracy of the model on labeled data; the classification discriminator is used to determine whether the result output by the classifier is correct.

[0054] During inference, the feature extractor is responsible for extracting features from the data in the target domain, and the classifier is responsible for predicting the remaining life of the device.

[0055] In order to effectively balance the distribution difference between the source domain and the target domain, the present invention adopts a dynamic parameter learning mechanism to achieve adaptive adjustment of the distribution difference between the source domain and the target domain by dynamically adjusting the weights between the feature discriminator and the classifier. The mechanism introduces dynamic parameters λ and influence factor κ to dynamically update the weight of the loss function according to the distribution difference between the source domain and the target domain, thereby ensuring that the classification performance of the source domain is not lost while reducing the distribution difference.

[0056] The specific steps include: Figure 4 As shown,

[0057] Acquire source domain data and target domain data of electromechanical equipment, and preprocess the source domain data and the target domain data;

[0058] Use the feature extractor to pre-train independently on labeled source domain data, so that the feature extractor has the ability to extract features from the source domain data;

[0059] Using a domain adaptive method to train a feature extractor, so that the feature extractor can extract effective high-dimensional data features from both the preprocessed source domain data and the target domain data;

[0060] In the inference phase, only feature extractors and classifiers are applied to predict the remaining lifetime of electromechanical devices.

[0061] In this embodiment, source domain data acquisition: collecting source domain (such as operating data of a certain type of electromechanical equipment under specific working conditions), including time series data collected by multiple sensors and corresponding remaining life labels.

[0062] Target domain data acquisition: Collect sensor time series data of the target domain (such as operating data of different models of equipment or under different working conditions), but no remaining life labels are required.

[0063] Data preprocessing: Perform normalization, denoising, segmentation and other preprocessing operations on the data in the source domain and target domain to ensure data quality and consistency.

[0064] In this embodiment, the physical sign extraction module (feature extractor) is composed of multiple convolutional layers of different sizes, which extracts high-dimensional data features from the time series data of the source domain and the target domain, and is represented by feature A. To ensure the consistency of the features of the source domain and the target domain in the high-dimensional space, the same-weight feature extractor is used to extract features from the data of the source domain and the target domain. Before performing domain adaptation, the physical sign extraction module is first pre-trained independently on the labeled source domain data to enable it to have better feature extraction ability for the source domain data.

[0065] In this embodiment, the domain adaptation method is used to train the feature extractor so that the feature extractor can extract effective high-dimensional data features from the preprocessed source domain data and the target domain data, including:

[0066] As Figure 1 shown in the block diagram of the model training system of the embodiment of the present invention, in the training stage, the ability of the training feature extractor to extract features in the target domain is adopted, and an adversarial network structure is used. The feature discriminator discriminates whether feature A comes from the source domain or the target domain;

[0067] The classifier classifies the feature A extracted from the source domain, and the classification discriminator discriminates whether the classification result is correct or wrong;

[0068] The feature discriminator introduces dynamic parameters λ and influence factor κ through a dynamic parameter learning mechanism, dynamically adjusts the weights between the feature discriminator and the classifier, balances the distribution differences between the source domain and the target domain, so that the feature extractor can be extracted in both the source domain and the target domain. The specific process is as follows:

[0069] S1. Calculation of the conditional distribution distance D c Calculation

[0070] D c represents the conditional distribution distance, which is used to measure the difference in conditional distribution between the source domain and the target domain. For each category y, that is, each remaining life label, the feature distribution probabilities of the source domain and the target domain under this category are calculated. MMD (Maximum Mean Discrepancy) is used to measure the similarity between two probability distributions. To solve the problem of class imbalance, the class weight w c can be introduced so that the contribution of each category to D c is proportional to the number of samples in the source domain and the target domain.

[0071]

[0072] In the formula, C represents the total number of categories, P s (X|Y = c) and P t (X|Y = c) respectively represent the conditional distributions of the source domain and the target domain under category c, and the weight w c is defined as: Where N c is the number of samples of category c, N total is the total number of samples in all categories. X represents the feature space. In the context of predictive maintenance, X may include various sensor data, operating parameters, environmental conditions and other features related to the equipment status. Y represents the label space, i.e., the prediction target. In the remaining service life prediction task, Y usually represents different remaining service life categories.

[0073] S2, marginal distribution distance D m Calculation

[0074] D m Represents the marginal distribution distance, which is used to measure the difference between the source domain and the target domain in the overall data distribution. It is calculated using the hierarchical MMD method. Hierarchical MMD can more carefully capture the distribution differences between the source domain and the target domain at different levels and scales by calculating MMD at multiple levels and scales. The specific implementation is as follows:

[0075]

[0076] Where: L represents the number of levels, α l is the weight of each layer, usually determined by empirical rules or automatic learning, P s (X) and P t (X) represents the marginal distribution of source domain and target domain respectively.

[0077] S3. Calculation of dynamic parameter λ

[0078] The dynamic parameter λ is used to balance the weights of the conditional distribution and the marginal distribution in the loss function. The total number of training rounds T is set. When calculating λ in each round t, the proportion of the current round is considered. This allows λ to be gradually adjusted as the training progresses. Its calculation formula is:

[0079]

[0080] Among them, α and β are weight parameters, which are used to control the adjustment ratio of early and late λ.

[0081] λ reflects the proportion of the conditional distribution in the total distribution difference. c When λ is larger, λ increases, indicating that more attention should be paid to the alignment of conditional distributions; conversely, when λ decreases, more attention should be paid to the alignment of marginal distributions.

[0082] S4. Application of Impact Factor κ

[0083] The influence factor κ is used to quantify the impact of the marginal distribution distance on the conditional distribution distance, and it plays a regulatory role in the loss function. By adjusting κ, the model can control the degree of emphasis on the alignment of marginal distribution and conditional distribution at different stages. The value of K is dynamically adjusted based on the number of training rounds. The total number of training rounds is set to T. When calculating κ in each round t, the proportion of the current round is considered This allows κ to be gradually adjusted as the training progresses.

[0084]

[0085] Among them, γ and δ are weight parameters, which are used to control the adjustment ratio of early and late κ.

[0086] S5. Construction of comprehensive loss function

[0087] Taking into account the alignment of marginal distribution and conditional distribution, the present invention designs the following comprehensive loss function:

[0088] L=L Cls +(1-κλ)·L Dism +λ·L Disc

[0089] in:

[0090] L Cls Represents the classification loss, which is used for supervised learning in the source domain and uses cross entropy loss:

[0091]

[0092] Where N is the number of samples, y i，c is the true label of sample i in category c, is the probability predicted by the model.

[0093] L Dism It represents the loss of the feature discriminator for the marginal distribution, which is used to measure the difference between the source domain and the target domain in the marginal distribution, using the cross entropy loss:

[0094]

[0095] Among them, d i is the domain label of sample i (source domain is 0, target domain is 1), is the predicted probability of the feature discriminator.

[0096] L Disc It represents the loss of the feature discriminator for the conditional distribution, which is used to measure the difference between the source domain and the target domain in the conditional distribution, and also uses the cross entropy loss:

[0097]

[0098] in, represents the domain label of sample i under category c, is the predicted probability of the feature discriminator under category c.

[0099] By adjusting λ and κ, the comprehensive loss function can adaptively balance the alignment strength of the marginal distribution and the conditional distribution, enabling the model to achieve more effective transfer learning between different domains, ensuring that the classification performance of the source domain is not lost while reducing the distribution difference.

[0100] In this embodiment, the method for real-time prediction of the remaining life of electromechanical equipment based on the source domain and the target domain after the distribution difference is balanced includes:

[0101] After the feature extractor is trained in the above process, only the feature extractor and the classifier can be used to predict the remaining life of the equipment. That is, in the inference stage, the output of the classifier is the prediction of the remaining life of the equipment.

[0102] Embodiment 2

[0103] Figure 2 As shown, it is a block diagram of the inference training system of the embodiment of the present invention; during inference, the feature extractor is responsible for extracting features from the data in the target domain, and the classifier is responsible for predicting the remaining life of the equipment.

[0104] like Figure 3 As shown, an embodiment of the present invention provides a training process block diagram of a cross-domain remaining life prediction system for electromechanical equipment based on dynamic parameter learning. The system is mainly composed of a data collection and preprocessing module, a feature extraction module, a domain adaptation process module, a remaining life prediction module, an output and visualization module, a model training module, and a feedback and update module. Each module works together to achieve efficient and accurate cross-domain remaining life prediction.

[0105] 1. Data collection and preprocessing module

[0106] The data collection and preprocessing module is responsible for collecting the operation data of the equipment in real time through a variety of high-precision sensors installed on the electromechanical equipment. Sensor types include temperature sensors, vibration sensors, pressure sensors, current sensors, etc., which can obtain multi-dimensional time series data of the equipment under different operating conditions. The collected raw data is transmitted to the central data storage unit of the system through wired or wireless networks to ensure the real-time and integrity of the data.

[0107] Data preprocessing steps include data normalization, denoising, outlier detection, and data segmentation to improve data quality and consistency. Specifically, normalization ensures that data from different sensors are within the same scale range, denoising eliminates noise interference in the data through filtering algorithms, outlier detection identifies and processes abnormal data points, and segmentation divides continuous time series data into fixed-length subsequences for subsequent feature extraction.

[0108] 2. Feature extraction module

[0109] The preprocessed data is input into the feature extraction module, which uses a deep learning model to extract high-dimensional feature representations from time series data. The feature extractor ensures the consistency of source and target domain data in the high-dimensional feature space by sharing parameters. These high-dimensional features effectively capture the key information of the equipment's operating status, providing a solid foundation for subsequent domain adaptation and remaining life prediction.

[0110] 3. Domain Adaptation Process Module

[0111] To achieve cross-domain transfer learning, the data after feature extraction is further processed by the domain adaptation module. The domain adaptation module adopts an adversarial network structure. The feature discriminator adaptively adjusts the weights between the feature discriminator and the classifier by introducing dynamic parameters λ and influence factors κ to balance the distribution differences between the source domain and the target domain. The specific steps include calculating the conditional distribution distance Dc and the marginal distribution distance Dm. The dynamic parameter λ is adjusted according to the ratio of the two, and the influence factor κ controls the model's emphasis on distribution alignment at different stages. The comprehensive loss function L achieves an adaptive balance between marginal distribution and conditional distribution by adjusting λ and κ:

[0112] L=L Cls +(1-κλ)·L Dism ++λ·L Disc

[0113] Among them, L Cls is the classification loss, L Dism is the marginal distribution alignment loss, L Disc is the conditional distribution alignment loss.

[0114] 4. Remaining life prediction module

[0115] For target domain data, in the inference stage, the target domain data is extracted by the feature extractor, and the remaining life is predicted by the classifier. The prediction results reflect the remaining service life of the equipment under the current operating state, supporting maintenance decisions and the formulation of preventive maintenance plans.

[0116] 5. Output and visualization module

[0117] The prediction results are displayed through the output and visualization module for maintenance personnel to refer to. This module includes a visualization interface and report generation tools, which can intuitively present the prediction results in the form of charts, reports, etc. At the same time, the system supports comparative analysis of prediction results with historical data to help maintenance personnel identify equipment performance trends and potential failure risks.

[0118] The prediction results are displayed through the output and visualization module for maintenance personnel to refer to. This module includes a visualization interface and report generation tools, which can intuitively present the prediction results in the form of charts, reports, etc. At the same time, the system supports comparative analysis of prediction results with historical data to help maintenance personnel identify equipment performance trends and potential failure risks.

[0119] The present invention is achieved by Figure 1-4 The system architecture shown in the figure realizes an efficient and accurate cross-domain remaining life prediction method for electromechanical equipment based on dynamic parameter learning. This method effectively reduces the distribution difference of the model in extracting features in the source domain and the target domain, and improves the generalization ability and prediction accuracy of the model. The modular design makes the system highly flexible and scalable, able to adapt to the ever-changing operating environment and actual needs, and has significant industrial application value and broad market prospects.

[0120] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A cross-domain remaining life prediction method for electromechanical equipment based on dynamic parameter learning, characterized in that: The following steps are involved: Acquire source domain data and target domain data of electromechanical equipment, and preprocess the source domain data and the target domain data; Use the feature extractor to pre-train independently on labeled source domain data, so that the feature extractor has the ability to extract features from the source domain data; Using a domain adaptive method to train a feature extractor, so that the feature extractor can extract effective high-dimensional data features from both the preprocessed source domain data and the target domain data; In the inference phase, only feature extractors and classifiers are applied to predict the remaining life of electromechanical equipment; Using a domain adaptive method to train a feature extractor so that the feature extractor can extract effective high-dimensional data features from both the preprocessed source domain data and the target domain data includes: Adopting an adversarial network structure, the feature discriminator is used to determine whether the high-dimensional data features come from the source domain or the target domain; The classifier is used to classify the high-dimensional data features extracted from the source domain, and the classification discriminator determines whether the classification result is correct or not; The feature discriminator introduces dynamic parameters λ and influence factors κ through a dynamic parameter learning mechanism, dynamically adjusting the weight between the feature discriminator and the classifier, so that the feature extractor can extract high-dimensional data features in both the source domain and the target domain; The feature discriminator introduces dynamic parameters λ and influence factor κ through a dynamic parameter learning mechanism, and dynamically adjusts the weight between the feature discriminator and the classifier, so that the feature extractor can extract high-dimensional data features in both the source domain and the target domain, including: Calculate the conditional distribution distance D c ; Calculate the marginal distribution distance D m ; Based on the conditional distribution distance D c and marginal distribution distance D m , calculate the dynamic parameter λ; Calculate the impact factor κ; Based on the dynamic parameter λ and the influencing factor κ, a comprehensive loss function is constructed; Calculate the conditional distribution distance D c include: In the formula, C represents the total number of categories, P s (X|Y=c) and P t (X|Y=c) represents the conditional distribution of the source domain and the target domain under category c, and the weight w c Defined as: Where N c is the number of samples of category c, N total is the total number of samples in all categories, X represents the feature space, and Y represents the label space; Calculate the marginal distribution distance D m include: Where: L represents the number of levels, α l is the weight of each layer, usually determined by empirical rules or automatic learning, P s (X) and P t (X) represents the marginal distribution of the source domain and the target domain respectively; Based on the conditional distribution distance D c and marginal distribution distance D m , the calculation of dynamic parameter λ includes: In the formula, α and β are weight parameters used to control the adjustment ratio of early and late λ, T is the total number of training rounds, and t is each round; Calculation of the impact factor κ includes: In the formula, γ and δ are weight parameters, which are used to control the adjustment ratio of early and late κ; Based on the dynamic parameter λ and the influencing factor κ, the comprehensive loss function is constructed including: L=L Cls +(1-κλ)·L Dism +λ·L Disc Where, L Cls is the classification loss, L Dism is the feature discriminator for the marginal distribution alignment loss, L Disc Conditional distribution alignment loss for the feature discriminator.

2. A cross-domain remaining life prediction system for electromechanical equipment based on dynamic parameter learning, the system is used to implement the method of claim 1, characterized in that: include: Data collection and preprocessing module, feature extraction module, domain adaptation process module, and remaining life prediction module; The data collection and preprocessing module is used to obtain source domain data and target domain data of electromechanical equipment, and preprocess the source domain data and the target domain data; The feature extraction module is used to use the feature extractor to independently perform pre-training on the labeled source domain data, so that the feature extractor has the ability to extract features from the source domain data; The domain adaptation process module is used to train the feature extractor using a domain adaptation method, so that the feature extractor can extract effective high-dimensional data features from both the preprocessed source domain data and the target domain data; The remaining life prediction module is used to predict the remaining life of the electromechanical equipment by only applying the feature extractor and the classifier during the inference stage.

Images