Engine residual life prediction method, model training method, device and equipment

Abstract

The embodiment of the application provides a kind of engine remaining life prediction method, model training method, device and equipment, the method comprises: obtaining multiple first engine sample data, determines first sample feature based on encoding module, respectively determines the type label corresponding to each first sample feature based on clustering module and label mapping relationship, respectively input each first sample feature into the prediction module associated with corresponding type label, determines prediction loss according to each sample remaining life and corresponding remaining life label, according to prediction loss, the joint training of encoding module and prediction module is carried out, obtains the remaining life prediction model after training, then the engine target data obtained is input into the remaining life prediction model after training, determines the target remaining life of engine target data based on the remaining life prediction model after training.The embodiment of the application can improve the robustness and generalization performance of life prediction model, improve the accuracy of remaining life prediction result.

Description

Technical Field

[0001] This application relates to, but is not limited to, the field of machine learning technology, and in particular to a method for predicting engine remaining life, a model training method, an apparatus, and equipment. Background Technology

[0002] In recent years, Predictive Maintenance and Health Management (PHM) has played a very important role in the engine field, especially in engines for the aviation, automotive, and marine industries. Predicting the remaining life of an engine is a key technology in predictive maintenance and health management, which is of great significance for improving engine reliability and reducing engine life cycle costs.

[0003] Currently, the remaining life prediction model first extracts sample features from engine sample data during training, then predicts life based on the sample features, and then updates the remaining life prediction model through model loss. However, traditional remaining life prediction models have poor robustness and generalization performance, and cannot accurately extract features, resulting in low accuracy of remaining life prediction results. Summary of the Invention

[0004] The following is an overview of the subject matter described in detail herein. This overview is not intended to limit the scope of the claims.

[0005] This application provides an engine remaining life prediction method, model training method, apparatus, and equipment, which can improve the robustness and generalization performance of the life prediction model, thereby improving the accuracy of the remaining life prediction results.

[0006] To achieve the above objectives, a first aspect of this application proposes an engine remaining life prediction method, comprising: acquiring multiple first engine sample data, wherein the first engine sample data carries a remaining life label; determining first sample features of each first engine sample data based on an encoding module, and determining type labels corresponding to each first sample feature based on a clustering module and a label mapping relationship, wherein the label mapping relationship of a later clustering round is determined by the label mapping relationship of the previous clustering round, and the type label is used to characterize the engine fault type; inputting each first sample feature into a prediction module associated with the corresponding type label, and determining the sample remaining life of the first engine sample data based on the prediction module; determining a prediction loss based on each sample remaining life and the corresponding remaining life label, and jointly training the encoding module and the prediction module based on the prediction loss to obtain a trained remaining life prediction model; acquiring engine target data, inputting the engine target data into the trained remaining life prediction model, and determining the target remaining life of the engine target data based on the trained remaining life prediction model.

[0007] In some embodiments, determining the type label corresponding to each of the first sample features based on the clustering module and the label mapping relationship includes: obtaining multiple initial centroids and the label mapping relationship corresponding to each initial centroid, wherein the initial centroids are centroids determined in the previous clustering round or preset; inputting each of the first sample features into the clustering module, performing clustering processing on each of the first sample features based on the clustering module to obtain multiple first feature clusters; traversing the first feature clusters, determining the label mapping relationship corresponding to each first feature cluster based on the distance between the current centroid of the first feature cluster and each of the initial centroids; and determining the type label corresponding to each of the first sample features based on the label mapping relationship corresponding to each of the first feature clusters.

[0008] In some embodiments, determining the prediction loss based on the remaining lifetime of each sample and the corresponding remaining lifetime label, and jointly training the encoding module and the prediction module based on the prediction loss, includes:

[0009] The remaining lifetime of each sample predicted by the prediction module is subtracted from the corresponding remaining lifetime label, and the square of the subtraction result is used as the prediction distance value; the prediction loss corresponding to the prediction module is determined based on the average of the prediction distance values ​​corresponding to the prediction module; the prediction modules are traversed, and the encoding module and the prediction module are jointly trained based on the prediction loss corresponding to the prediction module.

[0010] In some embodiments, before determining the first sample features of each of the first engine sample data based on the encoding module, the method further includes: acquiring multiple second engine sample data and multiple sample pairs, wherein the sample pair includes two third engine sample data and the sample pair carries a similarity label; determining the second sample features of each of the second engine sample data based on the encoding module, and determining the first reconstructed data of each of the second sample features based on the decoding module; determining a first target loss based on the first reconstructed data and the corresponding second engine sample data; determining the predicted similarity between the two third engine sample data in the sample pair based on the encoding module; determining a second target loss based on the predicted similarity and the corresponding similarity label; and pre-training the encoding module alternately based on the first target loss and the second target loss.

[0011] In some embodiments, after determining the first target loss based on the first reconstructed data and the corresponding second engine sample data, the method further includes: inputting each of the second sample features into a clustering module, performing clustering processing on each of the second sample features based on the clustering module to obtain multiple second feature clusters; determining a clustering loss based on the distance between each of the second sample features and the current centroid of the corresponding second feature cluster, and adjusting the first target loss based on the clustering loss.

[0012] In some embodiments, after determining the second target loss based on the predicted similarity and the corresponding similarity label, the method further includes: determining second reconstructed data for each of the third sample features based on the decoding module; determining a second sub-loss based on the second reconstructed data and the corresponding third engine sample data; and adjusting the second target loss based on the second sub-loss.

[0013] To achieve the above objectives, a second aspect of this application proposes a model training method, comprising: acquiring multiple first engine sample data, wherein the first engine sample data carries a remaining life label; determining first sample features of each first engine sample data based on an encoding module, and determining type labels corresponding to each first sample feature based on a clustering module, wherein the type labels are used to characterize engine fault types; inputting each first sample feature into a prediction module associated with the corresponding type label, and determining the sample remaining life of the first engine sample data based on the prediction module; determining a prediction loss based on each sample remaining life and the corresponding remaining life label, and jointly training the encoding module and the prediction module based on the prediction loss to obtain the trained remaining life prediction model.

[0014] To achieve the above objectives, a third aspect of this application proposes an engine remaining life prediction device, comprising: an acquisition module for acquiring multiple first engine sample data, wherein the first engine sample data carries a remaining life label; a determination module for determining first sample features of each first engine sample data based on an encoding module, and determining type labels corresponding to each first sample feature based on a clustering module, wherein the type labels are used to characterize engine fault types; a first prediction module for inputting each first sample feature into a prediction module associated with the corresponding type label, and determining the sample remaining life of the first engine sample data based on the prediction module; a training module for determining a prediction loss based on each sample remaining life and the corresponding remaining life label, and jointly training the encoding module and the prediction module based on the prediction loss to obtain a trained remaining life prediction model; and a second prediction module for acquiring engine target data, inputting the engine target data into the trained remaining life prediction model, and determining the target remaining life of the engine target data based on the trained remaining life prediction model.

[0015] To achieve the above objectives, a fourth aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the engine remaining life prediction method described in the first aspect, or to implement the model training method described in the second aspect.

[0016] To achieve the above objectives, a fifth aspect of the present application provides a storage medium, which is a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the engine remaining life prediction method described in the first aspect or the model training method described in the second aspect.

[0017] The embodiments of this application include at least the following beneficial effects: First sample features of the first engine sample data are determined by the encoding module, and then type labels are determined by the clustering module, which is equivalent to performing the task of determining the engine fault type. The first sample features are then input into the prediction module associated with the corresponding type label. The prediction module is used to predict the remaining life of the first sample features, which is equivalent to performing the task of predicting the remaining life. The encoding module and prediction module are jointly trained using prediction loss, realizing joint modeling of the tasks of determining the engine fault type and predicting the remaining life. This improves the robustness and generalization performance of the life prediction model, enhances the feature extraction performance of the encoding module, and improves the accuracy of the prediction module, thereby improving the accuracy of the remaining life prediction result. Sharing underlying features between different tasks reduces the complexity of the remaining life prediction model and achieves implicit data augmentation and knowledge transfer. Subsequently, the target remaining life of the engine target data can be obtained based on the trained remaining life prediction model, thereby effectively improving the accuracy of the target remaining life.

[0018] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the description, claims and drawings. Attached Figure Description

[0019] The accompanying drawings are used to provide a further understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0020] Figure 1 This is a schematic flowchart of an optional engine remaining life prediction method provided in an embodiment of this application.

[0021] Figure 2 A schematic diagram of an optional process for determining a type label provided in an embodiment of this application;

[0022] Figure 3 A schematic diagram of an optional joint training process provided for an embodiment of this application;

[0023] Figure 4 A schematic diagram of an optional pre-training process provided for an embodiment of this application;

[0024] Figure 5 A schematic diagram of an optional process for adjusting the first target loss provided in an embodiment of this application;

[0025] Figure 6A schematic diagram of an optional process for adjusting the loss of the second target provided in an embodiment of this application;

[0026] Figure 7 A schematic diagram of an optional model training method provided in an embodiment of this application;

[0027] Figure 8 A schematic diagram of an optional architecture for determining the first target loss in the first training sub-stage provided in an embodiment of this application;

[0028] Figure 9 A schematic diagram of an optional architecture for determining the second target loss in the first training sub-stage provided in an embodiment of this application;

[0029] Figure 10 A schematic diagram of an optional architecture for determining the first target loss in the second training sub-stage provided in an embodiment of this application;

[0030] Figure 11 A schematic diagram of an optional architecture for determining clustering loss in the second training sub-stage provided in an embodiment of this application;

[0031] Figure 12 A schematic diagram of an optional process for adjusting the first target loss provided in an embodiment of this application;

[0032] Figure 13 A schematic diagram of an optional architecture for determining the reconstruction loss and the second target loss in the second training sub-stage provided in an embodiment of this application;

[0033] Figure 14 A schematic diagram of an optional process for adjusting the loss of the second target provided in an embodiment of this application;

[0034] Figure 15 A schematic diagram of an optional architecture for the third training sub-stage provided in an embodiment of this application;

[0035] Figure 16 A schematic diagram of an optional architecture for joint training in the third training sub-stage provided in an embodiment of this application;

[0036] Figure 17 A schematic diagram illustrating an optional effect of the remaining lifetime prediction model provided in this application embodiment;

[0037] Figure 18 A schematic diagram of an optional architecture for an application scenario provided in this application embodiment;

[0038] Figure 19 This is a schematic diagram of an optional structure of the engine remaining life prediction device provided in this embodiment.

[0039] Figure 20 This is a schematic diagram of an optional hardware structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0041] In the description of this application, "several" means one or more, "multiple" means two or more, "greater than", "less than", "exceeding" etc. are understood to exclude the number itself, and "above", "below", "within" etc. are understood to include the number itself.

[0042] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, or the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0043] It should be noted that in various specific embodiments of this application, when processing data related to the characteristics of the target object, such as target object attribute information or attribute information sets, is required, the permission or consent of the target object will be obtained first. Furthermore, the collection, use, and processing of this data will comply with relevant laws, regulations, and standards. The target object can be a user. In addition, when embodiments of this application need to obtain target object attribute information, separate permission or consent from the target object will be obtained through pop-ups or redirection to a confirmation page. Only after obtaining the target object's separate permission or consent will the necessary target object-related data for the normal operation of the embodiments of this application be obtained.

[0044] To facilitate understanding of the technical solutions provided in the embodiments of this application, some key terms used in the embodiments of this application will be explained below:

[0045] Cloud technology refers to a hosting technology that unifies hardware, software, and network resources within a wide area network (WAN) or local area network (LAN) to achieve data computation, storage, processing, and sharing. Based on the cloud computing business model, cloud technology encompasses network technology, information technology, integration technology, management platform technology, and application technology. It can form resource pools, providing flexible and convenient on-demand access. Cloud computing technology will become a crucial support. Backend services of technical network systems require substantial computing and storage resources, such as video websites, image websites, and many portal websites. With the rapid development and application of the internet industry, every item may have its own identification mark in the future, requiring transmission to backend systems for logical processing. Data at different levels will be processed separately, and various industry data will require robust system support, which can only be achieved through cloud computing.

[0046] Artificial Intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to have perception, reasoning, and decision-making capabilities. AI technology is a comprehensive discipline involving a wide range of fields, encompassing both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, pre-trained model technology, operating / interactive systems, and mechatronics. Pre-trained models, also known as large models or foundational models, can be widely applied to downstream tasks in various areas of AI after fine-tuning. AI software technologies mainly include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0047] Machine learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and learn-by-doing.

[0048] In related technologies, the remaining life prediction model first extracts sample features from engine sample data during training, then predicts life based on the sample features, and then updates the remaining life prediction model through model loss. However, traditional remaining life prediction models have poor robustness and generalization performance, and cannot accurately extract features, resulting in low accuracy of remaining life prediction results.

[0049] To address the issue of low accuracy in remaining service life prediction, this application provides a method, model training method, apparatus, and device for predicting engine remaining service life. The method includes: acquiring multiple first engine sample data, each carrying a remaining service life label; determining first sample features for each first engine sample data based on an encoding module, and determining type labels corresponding to each first sample feature based on a clustering module and a label mapping relationship, wherein the label mapping relationship for a later clustering round is determined by the label mapping relationship for the previous clustering round, and the type labels are used to characterize engine fault types; inputting each first sample feature into a prediction module associated with its corresponding type label, and determining the sample remaining service life of the first engine sample data based on the prediction module; determining a prediction loss based on the remaining service life of each sample and its corresponding remaining service life label, and jointly training the encoding module and the prediction module based on the prediction loss to obtain a trained remaining service life prediction model; acquiring engine target data, inputting the engine target data into the trained remaining service life prediction model, and determining the target remaining service life of the engine target data based on the trained remaining service life prediction model. According to the scheme provided in the embodiments of this application, the first sample features of the first engine sample data are determined by the encoding module, and then the type label is determined by the clustering module, which is equivalent to performing the task of determining the engine fault type. Then, the first sample features are input into the prediction module associated with the corresponding type label, and the prediction module is used to predict the remaining life of the first sample features, which is equivalent to performing the task of predicting the remaining life. The encoding module and the prediction module are jointly trained by prediction loss, realizing the joint modeling of the task of determining the engine fault type and the task of predicting the remaining life. This can improve the robustness and generalization performance of the life prediction model, improve the feature extraction performance of the encoding module and improve the accuracy of the prediction module, thereby improving the accuracy of the remaining life prediction result. The underlying features are shared between different tasks, which reduces the complexity of the remaining life prediction model and realizes implicit data augmentation and knowledge transfer. Subsequently, the target remaining life of the engine target data can be obtained based on the trained remaining life prediction model, thereby effectively improving the accuracy of the target remaining life.

[0050] The engine remaining life prediction method, model training method, apparatus and equipment provided in the embodiments of this application are specifically described through the following embodiments. First, the engine remaining life prediction method in the embodiments of this application is described.

[0051] The engine remaining life prediction method provided in this application relates to the field of computer technology. The engine remaining life prediction method provided in this application can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the engine remaining life prediction method, but is not limited to the above forms.

[0052] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0053] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0054] The embodiments of this application will be further described below with reference to the accompanying drawings.

[0055] like Figure 1 As shown, Figure 1This is a schematic flowchart of an optional engine remaining life prediction method provided in an embodiment of this application. The engine remaining life prediction method can be executed by a server, or by a terminal, or by a server in conjunction with a terminal. The engine remaining life prediction method includes, but is not limited to, the following steps S110 to S150:

[0056] Step S110: Obtain sample data of multiple first engines;

[0057] Step S120: Determine the first sample features of each first engine sample data based on the encoding module, and determine the type label corresponding to each first sample feature based on the clustering module and the label mapping relationship;

[0058] Step S130: Input each first sample feature into the prediction module associated with the corresponding type label, and determine the remaining lifespan of the first engine sample data based on the prediction module.

[0059] Step S140: Determine the prediction loss based on the remaining lifetime of each sample and the corresponding remaining lifetime label, and jointly train the encoding module and the prediction module based on the prediction loss to obtain the trained remaining lifetime prediction model.

[0060] Step S150: Obtain engine target data, input the engine target data into the trained remaining life prediction model, and determine the target remaining life of the engine target data based on the trained remaining life prediction model.

[0061] The first engine sample data carries a remaining life label. Based on historical experience or simulation software, the current operating cycle corresponding to the first engine sample data can be determined. Then, the actual remaining life can be obtained by subtracting the current operating cycle from the maximum operating cycle of the sample. The actual remaining life is used as the remaining life label. For example, a sliding window with a size of 50*16 is used, where 50 represents the length of the time series, 16 represents the number of features, and the sliding step size is 1. The remaining life of the last time step of the window is taken as the remaining life label. To avoid the time series length of some samples being too short and affecting the use of the model, when the time series length of a sample is less than 50, the time series length is adjusted to 50 through padding.

[0062] Taking aero-engines as an example, the first engine sample data or engine target data can include engine degradation parameters such as flight altitude, Mach number, throttle resolver angle, difference between ambient temperature and ISA, ram pressure recovery, fuel flow rate, ram shaft speed, high-pressure shaft speed, high-pressure turbine inlet temperature, engine pressure ratio, and thrust. The first engine sample data can be input into simulation software to simulate engine performance. Typically, the flight altitude input range is 0 to 15000, the Mach number input range is 0 to 0.90, and the throttle resolver angle input range is... The input range is 20 to 100; the input range for the difference between ambient temperature and ISA is -20 to 20; the input range for ram pressure recovery is 0.99 to 1; the input range for fuel flow is 0.55 to 1.5; the input range for low-pressure shaft speed is 0 to 4750; the input range for high-pressure shaft speed is 7300 to 13280; the input range for high-pressure turbine inlet temperature is 900 to 1400; the input range for engine pressure ratio is 1.2 to 1.8; and the input range for thrust is 15000 to 230000.

[0063] Specifically, multiple sensors can be embedded in different components of the aircraft engine to collect first engine sample data.

[0064] In this process, the label mapping relationship of the subsequent clustering round is determined by the label mapping relationship of the previous clustering round. The label mapping relationship refers to the mapping relationship between the clustering result and the type label. The type label is used to characterize the engine fault type. Since the differences between the first sample features corresponding to different engine fault types are usually large, while the differences between the first sample features corresponding to the same engine fault type are usually small, after clustering each first sample feature using the clustering module, it can be considered that each first sample feature located in the same cluster corresponds to the same engine fault type, and each first sample feature located in different clusters corresponds to different engine fault types. Therefore, the engine fault type corresponding to the first sample feature can be determined based on the clustering result of the first sample feature, that is, the type label corresponding to the first sample feature can be determined.

[0065] For example, the type label can be 1, 2, 3, etc. When the type label corresponding to the first sample feature is 1, it means that the first sample feature corresponds to the first type of engine failure. When the type label corresponding to the first sample feature is 2, it means that the first sample feature corresponds to the second type of engine failure. When the type label corresponding to the first sample feature is 3, it means that the first sample feature corresponds to the third type of engine failure.

[0066] Taking aero engines as an example, the type of engine failure can be adjusted by changing the relevant engine degradation parameters. Engine failure types can include fan outflow / efficiency reduction, fan inflow / efficiency reduction, high-pressure compressor flow / efficiency reduction, high-pressure turbine flow / efficiency reduction, low-pressure turbine flow / efficiency reduction, nozzle area changes, and so on.

[0067] Specifically, when the remaining lifetime prediction model converges or the number of training rounds reaches a preset threshold, the remaining lifetime prediction model can be considered to have finished training, and there is no need to combine the encoding module and the prediction module. The convergence of the remaining lifetime prediction model can refer to the prediction accuracy of the remaining lifetime prediction model exceeding a preset accuracy threshold. In this application embodiment, the specific value of the number threshold or accuracy threshold is not limited.

[0068] Based on this, the first sample features of the first engine sample data are determined by the encoding module, and then the type label is determined by the clustering module, which is equivalent to performing the task of determining the engine fault type. Then, the first sample features are input into the prediction module associated with the corresponding type label, and the prediction module is used to predict the remaining life of the first sample features, which is equivalent to performing the task of predicting the remaining life. The encoding module and the prediction module are jointly trained by prediction loss, realizing the joint modeling of the tasks of determining the engine fault type and predicting the remaining life. This can improve the robustness and generalization performance of the life prediction model, improve the feature extraction performance of the encoding module and the accuracy of the prediction module, thereby improving the accuracy of the remaining life prediction results. The underlying features are shared between different tasks, which reduces the complexity of the remaining life prediction model and realizes implicit data augmentation and knowledge transfer. Subsequently, the target remaining life of the engine target data can be obtained based on the trained remaining life prediction model, thereby effectively improving the accuracy of the target remaining life.

[0069] In one possible implementation, the prediction module can consist of two fully connected layers. During the training of the remaining lifetime prediction model, the shape of the input tensor of the remaining lifetime prediction model can be set to (30, 50, 16), that is, 30 batches of sample data are input each time, the time series length of each batch is 50, and the number of features is 16. The Adam optimizer is used to update the parameters of the encoding module and the prediction module, and the initial learning rate is set to 0.005. To prevent overfitting, a dropout layer can be added to the fully connected layer, and the dropout rate is set to 0.2.

[0070] Additionally, refer to Figure 2 In one embodiment, Figure 1 Step S120 in the illustrated embodiment includes, but is not limited to, the following steps:

[0071] Step S210: Obtain multiple initial centroids and the label mapping relationship corresponding to each initial centroid;

[0072] Step S220: Input each first sample feature into the clustering module, and perform clustering processing on each first sample feature based on the clustering module to obtain multiple first feature clusters;

[0073] Step S230: Traverse the first feature cluster and determine the label mapping relationship corresponding to each first feature cluster based on the distance between the current centroid of the first feature cluster and each initial centroid.

[0074] Step S240: Based on the label mapping relationship corresponding to each first feature cluster, determine the type label corresponding to each first sample feature.

[0075] The initial centroids are determined or preset in the previous clustering round. In the first training round, each initial centroid needs to be preset. The number of initial centroids is the same as the number of clusters obtained by clustering. The number of clusters also needs to be preset. For example, the number of clusters obtained by clustering is preset to three, representing that there are three types of engine faults. In other training rounds, the centroids of the clusters obtained by clustering in the previous training round can be used as the initial centroids of the current training round.

[0076] Based on this, in the current training round, the clustering module can cluster each first sample feature into multiple first feature clusters. Since the previous training round updates the parameters of the encoding module and the prediction module based on the prediction loss, and the first engine sample data used in the current training round may not be the same as the first engine sample data used in the previous training round, the centroids of the clusters obtained in the current training round and the centroids of the clusters obtained in the previous training round usually do not coincide. Therefore, it is necessary to traverse the first feature clusters to determine the initial centroid closest to the current centroid of the first feature cluster, and use the label mapping relationship corresponding to the closest initial centroid as the label mapping relationship corresponding to the first feature cluster. This achieves centroid alignment between different training rounds and can effectively determine the label mapping relationship corresponding to the current centroid of each first feature cluster. This ensures that the prediction module can predict the remaining life of the first sample features of the same engine fault type in different training rounds. Each prediction module corresponds to one type of engine fault. After the remaining life prediction model is trained, it can accurately predict the target remaining life of the engine target data under various engine fault types.

[0077] Additionally, refer to Figure 3 In one embodiment, Figure 1 Step S140 in the illustrated embodiment includes, but is not limited to, the following steps:

[0078] Step S310: Subtract the remaining lifetime predicted by the prediction module from the corresponding remaining lifetime label, and use the square of the subtraction result as the prediction distance value.

[0079] Step S320: Determine the prediction loss corresponding to the prediction module based on the average value of each prediction distance value corresponding to the prediction module;

[0080] Step S330: Traverse the prediction modules and perform joint training on the encoding module and the prediction module according to the prediction loss corresponding to the prediction module.

[0081] Based on this, different prediction modules will input the first sample features under different engine failure types. Therefore, different prediction modules will predict the remaining life of samples under different engine failure types. After calculating the prediction distance value using the sample remaining life and the corresponding remaining life label, it is necessary to use the prediction distance value corresponding to each prediction module to calculate the prediction loss corresponding to each prediction module. The prediction loss is used to jointly train the encoding module and the corresponding prediction module. For example, the remaining life prediction model includes prediction module a and prediction module b. The prediction loss a corresponding to prediction module a is used to jointly train the encoding module and prediction module a. The prediction loss b corresponding to prediction module b is used to jointly train the encoding module and prediction module b. When all prediction modules have completed training, it means that the remaining life prediction model has completed training. In addition, through joint training, different tasks can share underlying features, which reduces the complexity of the remaining life prediction model and realizes implicit data augmentation and knowledge transfer. Subsequently, the target remaining life of engine target data can be obtained based on the trained remaining life prediction model, thereby effectively improving the accuracy of the target remaining life.

[0082] Specifically, the formula for calculating the prediction loss for any prediction module is as follows:

[0083]

[0084] Where RMSE is the prediction loss corresponding to the prediction module, n is the number of all first sample features input to the prediction module, and y i Let be the remaining lifetime of the sample corresponding to the i-th feature of the first sample. Let the remaining lifetime label be the feature corresponding to the i-th first sample. This is for predicting distance values.

[0085] As can be seen, during the training process of the remaining life prediction model, since a smaller prediction loss indicates a higher prediction accuracy, the gradients of the encoding module and the prediction module can be calculated simultaneously based on the prediction loss. When the prediction loss is larger, the gradients of the encoding module and the prediction module should be set to be larger, and vice versa. Then, the parameters of the encoding module and the prediction module can be jointly optimized through backpropagation of the gradients, thereby improving the feature extraction performance of the encoding module and the accuracy of the prediction module.

[0086] Additionally, refer to Figure 4 In one embodiment, Figure 1 Before step S120 in the illustrated embodiment, the following steps may also be included, but are not limited to:

[0087] Step S410: Obtain multiple second engine sample data and multiple sample pairs;

[0088] Step S420: Determine the second sample features of each second engine sample data based on the encoding module, and determine the first reconstructed data of each second sample feature based on the decoding module;

[0089] Step S430: Determine the first target loss based on the first reconstructed data and the corresponding second engine sample data;

[0090] Step S440: Based on the encoding module, determine the third sample features of the two third engine sample data in the sample pair respectively, and determine the predicted similarity between the two third sample features;

[0091] Step S450: Determine the second target loss based on the predicted similarity and the corresponding similarity label;

[0092] Step S460: Alternately pre-train the encoding module based on the first target loss and the second target loss.

[0093] In this process, after inputting the second sample features into the decoding module, the first reconstructed data of the second sample features can be determined based on the decoding module. Then, the first target loss can be determined by the first reconstructed data and the corresponding second engine sample data. The smaller the first target loss, the smaller the difference between the first reconstructed data and the corresponding second engine sample data, which is equivalent to the better the feature extraction effect of the encoding module. Therefore, the parameters of the encoding module can be optimized by the first target loss to improve the feature extraction performance of the encoding module.

[0094] The sample pair includes two third engine sample data. Each sample pair carries a similarity label, which represents the true similarity between the features of the two third samples. Generally, when two third engine sample data belong to the same engine fault type, the similarity between them will be high. Conversely, when two third engine sample data belong to different engine fault types, the similarity between them will be low. Therefore, we can determine whether the third engine sample data belong to the same engine fault type and then set the similarity label for the sample pair. Then, by predicting the similarity and the corresponding similarity label, we can determine the second objective loss. The smaller the second objective loss, the smaller the difference between the predicted similarity and the corresponding similarity label, which means the feature extraction effect of the encoding module is better. Therefore, we can optimize the parameters of the encoding module through the second objective loss to improve the feature extraction performance of the encoding module.

[0095] Based on this, the encoding module is pre-trained alternately according to the first target loss and the second target loss. This means that after the encoding module is pre-trained according to the first target loss in the current pre-training round, it needs to be pre-trained according to the second target loss in the next pre-training round. This can better optimize the parameters of the encoding module, gradually improve the feature extraction performance of the encoding module, and accelerate the convergence speed of the remaining life prediction model when training based on the pre-trained encoding module in the future.

[0096] Specifically, the formula for calculating the loss of the first objective is as follows:

[0097]

[0098] Among them, y i For the i-th first reconstructed data, x i Let be the i-th sample data of the second engine, n be the number of sample data of the second engine, and Loss1 be the first target loss.

[0099] Specifically, the first engine sample data, the second engine sample data, and the third engine sample data can be sampled from the same sample dataset or from different sample datasets. For example, 30 engine sample data can be randomly sampled from the sample dataset and then combined in pairs to obtain 435 sample pairs. The two engine sample data in each sample pair are used as the third engine sample data. After determining the sample pairs, two weakly supervised sample sets can be created. One weakly supervised sample set is used to store the previous third engine sample data of each sample pair, and the other weakly supervised sample set is used to store the next third engine sample data of each sample pair.

[0100] Since the higher the similarity between two corresponding third-sample features, the smaller the distance between them, the similarity between two corresponding third-sample features can be represented by the distance between them. Accordingly, a distance label is used to represent the similarity label. The formula for calculating the distance between two corresponding third-sample features is as follows:

[0101]

[0102] Among them, f i Let s be the third sample feature corresponding to the previous third engine sample data of the i-th sample pair stored in one of the weakly supervised sample sets. i The third sample feature corresponding to the third engine sample data of the i-th sample pair stored in another weakly supervised sample set, where n is the amount of data of the third engine sample data in the weakly supervised sample set;

[0103] The results of the first and second sub-loss functions can be added together to obtain the second target loss. The formula for the second target loss is as follows:

[0104] Loss2 = Loss 2_1 +Loss 2_2

[0105] Where Loss2 is the second objective loss, Loss 2_1 The result of calculating the first sub-loss function, Loss 2_2 This is the result of calculating the second sub-loss function;

[0106] The formula for the first sub-loss function is as follows:

[0107] Loss 2_1 =0.5*(1-atch_y)*d W

[0108] Where, d W t_y represents the distance between the two corresponding third sample features, where t_y is the distance label of the sample pair;

[0109] The formula for the second sub-loss function is as follows:

[0110] Loss 2_2 =0.5*batch_y*max(0,margin-d) W )

[0111] Where, d WLet be the distance between the corresponding two third sample features, batch_y be the distance label of the sample pair, and margin be the preset decision threshold. Margin can be set to 1, or max(0, margin-d). W ) is to take 0 and margin-d W The maximum value between;

[0112] As can be seen, when the two third engine sample data within a sample pair are in the same engine fault type, the distance label of the sample pair is set to 0; when the two third engine sample data within a sample pair are in different engine fault types, the distance label of the sample pair is set to 1. This is equivalent to the following: when the two third engine sample data within a sample pair are in the same engine fault type, the second target loss of the same type of engine sample data can be calculated using the first sub-loss function; when the two third engine sample data within a sample pair are in different engine fault types, the second target loss of the different type of engine sample data can be calculated using the second sub-loss function.

[0113] In one possible implementation, after sampling first engine sample data, second engine sample data, or third engine sample data from the sample dataset, since the first engine sample data, second engine sample data, and third engine sample data can be a set of engine data, for example, a set of engine data including 16 types of data, it is necessary to normalize the sampled set of engine data to eliminate the dimensional differences between different types of engine data. Specifically, the normalization formula is as follows:

[0114]

[0115] Where x is the target engine data in a set of engine data, x min x is the minimum value in a set of engine data. max Let x be the maximum value in a set of engine data, and x′ be the normalized result of the target engine data.

[0116] In one possible implementation, the encoding module may include a bidirectional gated recurrent unit (GRU). The GRU can reduce the dimensionality of the second engine sample data through multiple time steps. It can also use attention weighting to weight the hidden states corresponding to each time step, and then sum the weighted results to obtain the second sample features of the second engine sample data. Specifically, the formula for calculating the weights corresponding to each time step is as follows:

[0117] e t =u tanh(wH t +b)

[0118]

[0119] Among them, e t Let H be the attention distribution value at time step t, where u is the first weight, w is the second weight, and H is the second weight. t Let be the hidden state output by the bidirectional GRU at time step t, b be the bias term, exp be an exponential function with the natural constant e as the base, and α be the value of α. t This represents the weight corresponding to the t-th time step.

[0120] Additionally, refer to Figure 5 In one embodiment, Figure 4 Following step S430 in the illustrated embodiment, the following steps may also be included, but are not limited to:

[0121] Step S510: Input each second sample feature into the clustering module, and perform clustering processing on each second sample feature based on the clustering module to obtain multiple second feature clusters;

[0122] Step S520: Determine the clustering loss based on the distance between each second sample feature and the current centroid of the corresponding second feature cluster, and adjust the first target loss based on the clustering loss.

[0123] Based on this, the clustering loss and the unadjusted first target loss can be added together to obtain the adjusted first target loss, so that the first target loss includes the loss information of the clustering results. After training the encoding module with the first target loss, the second sample features encoded by the encoding module can be close to the current centroid of the corresponding second feature cluster, which further improves the feature extraction performance of the encoding module. This allows the remaining lifetime prediction model to better distinguish the clustering results, reduces the possibility of misclassification, and enables the remaining lifetime prediction model to accurately input the features extracted by the encoding module into the corresponding prediction module.

[0124] Specifically, the formula for calculating clustering loss is as follows:

[0125]

[0126] Among them, c i It is the i-th feature of the second sample, e i It is C i The current centroid of the corresponding second feature cluster, where n is the number of second sample features, rec is the clustering loss, and c i -e i This is equivalent to calculating the distance between the current centroid of the second sample feature and the corresponding second feature cluster.

[0127] Additionally, refer to Figure 6 In one embodiment, Figure 4 Following step S450 in the illustrated embodiment, the following steps may also be included, but are not limited to:

[0128] Step S610: Determine the second reconstructed data for each third sample feature based on the decoding module;

[0129] Step S620: Determine the reconstruction loss based on the second reconstruction data and the corresponding third engine sample data, and adjust the second target loss based on the reconstruction loss.

[0130] Based on this, after inputting the third sample features into the decoding module, the second reconstructed data of the third sample features can be determined based on the decoding module. Then, the second target loss can be determined by the second reconstructed data and the corresponding third engine sample data. The smaller the second target loss, the smaller the difference between the second reconstructed data and the corresponding third engine sample data, which is equivalent to the better the feature extraction effect of the encoding module. The reconstruction loss and the unadjusted second target loss are added together to obtain the adjusted second target loss. The parameters of the encoding module can be optimized by the adjusted second target loss to further improve the feature extraction performance of the encoding module.

[0131] Additionally, refer to Figure 7 , Figure 7 This is an optional flowchart of a model training method provided in an embodiment of this application. The model training method can be executed by a server, or by a terminal, or by a server in conjunction with a terminal. The model training method includes, but is not limited to, the following steps S710 to S740.

[0132] Step S710: Obtain multiple first engine sample data;

[0133] Step S720: Determine the first sample features of each first engine sample data based on the encoding module, and determine the type label corresponding to each first sample feature based on the clustering module;

[0134] Step S730: Input each first sample feature into the prediction module associated with the corresponding type label, and determine the remaining lifespan of the first engine sample data based on the prediction module.

[0135] Step S740: Determine the prediction loss based on the remaining lifetime of each sample and the corresponding remaining lifetime label, and jointly train the encoding module and the prediction module based on the prediction loss to obtain the trained remaining lifetime prediction model.

[0136] The first engine sample data carries a remaining life label. The label mapping relationship of the subsequent clustering round is determined by the label mapping relationship of the previous clustering round. The type label is used to characterize the engine failure type.

[0137] The aforementioned model training method and the engine remaining life prediction method are based on the same inventive concept. Therefore, this model training method determines the first sample features of the first engine sample data through the encoding module, and then determines the type label through the clustering module, which is equivalent to performing the task of determining the engine fault type. Then, the first sample features are input into the prediction module associated with the corresponding type label, and the prediction module is used to predict the remaining life of the first sample features, which is equivalent to performing the task of predicting the remaining life. By jointly training the encoding module and the prediction module through prediction loss, the joint modeling of the tasks of determining the engine fault type and predicting the remaining life is realized. This can improve the robustness and generalization performance of the life prediction model, improve the feature extraction performance of the encoding module and improve the accuracy of the prediction module, thereby improving the accuracy of the remaining life prediction result. The underlying features are shared between different tasks, which reduces the complexity of the remaining life prediction model and realizes implicit data augmentation and knowledge transfer. Subsequently, the target remaining life of the engine target data can be obtained based on the trained remaining life prediction model, thereby effectively improving the accuracy of the target remaining life.

[0138] The detailed principles of steps S710 to S740 can be found in the previous explanation of steps S110 to S140, and will not be repeated here.

[0139] The following is a detailed explanation of the entire process of the engine remaining life prediction method.

[0140] refer to Figures 8 to 18 , Figure 8 This is a schematic diagram of an optional architecture for determining the first target loss in the first training sub-stage provided in an embodiment of this application. Figure 9 This is a schematic diagram of an optional architecture for determining the second target loss in the first training sub-stage provided in an embodiment of this application. Figure 10 This is a schematic diagram of an optional architecture for determining the first target loss in the second training sub-stage provided in an embodiment of this application. Figure 11 This is a schematic diagram of an optional architecture for determining the clustering loss in the second training sub-stage provided in an embodiment of this application. Figure 12 This is a schematic diagram of an optional process for adjusting the first target loss provided in an embodiment of this application. Figure 13 This is a schematic diagram of an optional architecture for determining the reconstruction loss and the second target loss in the second training sub-stage provided in an embodiment of this application. Figure 14 This is a schematic diagram of an optional process for adjusting the loss of the second target provided in an embodiment of this application. Figure 15 This is a schematic diagram of an optional architecture for the third training sub-stage provided in an embodiment of this application. Figure 16 This is a schematic diagram of an optional architecture for joint training in the third training sub-stage provided in an embodiment of this application. Figure 17 This is an optional effect diagram illustrating the prediction performance of the remaining lifetime prediction model provided in this application embodiment. Figure 18 This is a schematic diagram of an optional architecture for an application scenario provided in the embodiments of this application.

[0141] First, the training phase will be explained in detail below. Three training sub-phases can be completed sequentially during the training phase to obtain the trained remaining lifetime prediction model.

[0142] I. The first training sub-phase:

[0143] Refer again Figure 8 and Figure 9 First, multiple second engine sample data and multiple sample pairs are obtained, where each sample pair includes two third engine sample data and carries a similarity label.

[0144] Then, based on the encoding module, the second sample features of each second engine sample data are determined respectively, and based on the decoding module, the first reconstructed data of each second sample feature are determined respectively;

[0145] Then, based on the first reconstructed data and the corresponding second engine sample data, the first target loss is determined;

[0146] Then, based on the encoding module, the third sample features of the two third engine sample data in the sample pair are determined, and the predicted similarity between the two third sample features is determined.

[0147] Then, based on the predicted similarity and the corresponding similarity labels, the second objective loss is determined;

[0148] Then, the encoding module is pre-trained alternately based on the first target loss and the second target loss.

[0149] II. The second training sub-phase:

[0150] Refer again Figures 10 to 14 First, multiple second engine sample data and multiple sample pairs are obtained again. Among them, the sample pairs include two third engine sample data and the sample pairs carry similarity labels.

[0151] Then, based on the encoding module, the second sample features of each second engine sample data are determined respectively, and based on the decoding module, the first reconstructed data of each second sample feature are determined respectively;

[0152] Then, based on the first reconstructed data and the corresponding second engine sample data, the first target loss is determined;

[0153] Then, each second sample feature is input into the clustering module, and the clustering module performs clustering processing on each second sample feature to obtain multiple second feature clusters;

[0154] Then, based on the distance between each second sample feature and the current centroid of the corresponding second feature cluster, the clustering loss is determined, and the first target loss is adjusted based on the clustering loss;

[0155] Then, based on the encoding module, the third sample features of the two third engine sample data in the sample pair are determined, and the predicted similarity between the two third sample features is determined.

[0156] Then, based on the predicted similarity and the corresponding similarity labels, the second objective loss is determined;

[0157] Then, the second reconstructed data for each third sample feature is determined based on the decoding module;

[0158] Then, based on the second reconstruction data and the corresponding third engine sample data, the reconstruction loss is determined, and the second target loss is adjusted according to the reconstruction loss;

[0159] Then, the encoding module is pre-trained alternately based on the first target loss and the second target loss.

[0160] III. The third training sub-phase:

[0161] Refer again Figures 15 to 16 First, multiple first engine sample data are obtained, where the first engine sample data carries a remaining life tag;

[0162] Then, the first sample features of each first engine sample data are determined based on the encoding module;

[0163] Then, obtain multiple initial centroids and the label mapping relationship corresponding to each initial centroid. The initial centroids are either centroids determined in the previous clustering round or preset.

[0164] Then, each first sample feature is input into the clustering module, and the clustering module performs clustering processing on each first sample feature to obtain multiple first feature clusters;

[0165] Then, the first feature cluster is traversed, and the label mapping relationship corresponding to each first feature cluster is determined based on the distance between the current centroid of the first feature cluster and each initial centroid.

[0166] Then, based on the label mapping relationship corresponding to each first feature cluster, the type label corresponding to each first sample feature is determined, wherein the type label is used to characterize the engine fault type;

[0167] Then, each first sample feature is input into the prediction module associated with the corresponding type label, and the remaining lifespan of the first engine sample data is determined based on the prediction module.

[0168] Then, the prediction loss is determined based on the remaining lifetime of each sample and the corresponding remaining lifetime label. The encoding module and the prediction module are jointly trained based on the prediction loss to obtain the trained remaining lifetime prediction model.

[0169] Refer again Figure 17 By inputting the engine sample data from the test set into the trained remaining life prediction model, the remaining life of each engine sample data can be predicted. As shown in the figure, the curves of the actual remaining life and the curves of the predicted remaining life are highly similar, indicating that the prediction accuracy of the remaining life prediction model is high.

[0170] Next, the model application phase will be described in detail below.

[0171] Refer again Figure 18 First, acquire the engine target data;

[0172] Then, the engine target data is input into the trained remaining life prediction model, and the target sample features of the engine target data are determined based on the encoding module of the trained remaining life prediction model.

[0173] Then, based on the clustering module and label mapping relationship of the trained remaining lifetime prediction model, the target type label corresponding to the target sample feature is determined;

[0174] Then, the target sample features are input into the target prediction module associated with the target type label;

[0175] Then, the target remaining life of the engine target data is determined based on the target prediction module.

[0176] Based on this, the training phase first employs two training sub-stages to enable the remaining life prediction model to effectively reduce the dimensionality of the input data and cluster the resulting features. This significantly improves the training performance of the third training sub-stage. The encoding module determines the first sample features of the first engine sample data, and the clustering module determines the type label, essentially performing the task of determining the engine fault type. The first sample features are then input into the prediction module associated with the corresponding type label, and the prediction module uses these features to predict the remaining life, essentially performing the task of predicting the remaining life. The encoding and prediction modules are jointly trained using prediction loss, achieving joint modeling of the tasks of determining the engine fault type and predicting the remaining life. This improves the robustness and generalization performance of the life prediction model, enhances the feature extraction performance of the encoding module, and increases the accuracy of the prediction module, thereby improving the accuracy of the remaining life prediction results. Sharing underlying features across different tasks reduces the complexity of the remaining life prediction model while enabling implicit data augmentation and knowledge transfer. Subsequently, the target remaining life of the engine target data can be obtained based on the trained remaining life prediction model, effectively improving the accuracy of the target remaining life.

[0177] Additionally, refer to Figure 19 This application also provides a model training device 1900, comprising:

[0178] The acquisition module 1901 is used to acquire multiple first engine sample data, wherein the first engine sample data carries a remaining life tag;

[0179] The determination module 1902 is used to determine the first sample features of each first engine sample data based on the encoding module, and to determine the type label corresponding to each first sample feature based on the clustering module. The label mapping relationship of the later clustering round is determined by the label mapping relationship of the previous clustering round. The type label is used to characterize the engine fault type.

[0180] The first prediction module 1903 is used to input each first sample feature into the prediction module associated with the corresponding type label, and determine the remaining lifespan of the first engine sample data based on the prediction module.

[0181] Training module 1904 is used to determine the prediction loss based on the remaining lifetime of each sample and the corresponding remaining lifetime label. The encoding module and the prediction module are jointly trained based on the prediction loss to obtain the trained remaining lifetime prediction model.

[0182] The second prediction module 1905 is used to acquire engine target data, input the engine target data into the trained remaining life prediction model, and determine the target remaining life of the engine target data based on the trained remaining life prediction model.

[0183] The aforementioned engine remaining life prediction device 1900 and engine remaining life prediction method are based on the same inventive concept, and will not be described in detail here.

[0184] Additionally, refer to Figure 20 , Figure 20 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes:

[0185] The processor 2001 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0186] The memory 2002 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 2002 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 2002, and the processor 2001 calls and executes the engine remaining life prediction method of the embodiments of this application, for example, executing the above-described... Figure 1 Method steps S110 to S150 Figure 2 Method steps S210 to S240, Figure 3 Method steps S310 to S330, Figure 4 Method steps S410 to S460 Figure 5 Method steps S510 to S520 Figure 6 Alternatively, the method steps S610 to S620 can be performed, or the model training method of the embodiments of this application can be executed, for example, the method described above. Figure 7 Method steps S710 to S740;

[0187] Input / output interface 2003 is used to implement information input and output;

[0188] The communication interface 2004 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0189] Bus 2005 transmits information between various components of the device (e.g., processor 2001, memory 2002, input / output interface 2003, and communication interface 2004);

[0190] The processor 2001, memory 2002, input / output interface 2003 and communication interface 2004 are connected to each other within the device via bus 2005.

[0191] This application embodiment also provides a storage medium, which is a computer-readable storage medium for computer-readable storage. The storage medium stores one or more programs, which can be executed by one or more processors to implement the above-described engine remaining life prediction method, for example, executing the above-described... Figure 1 Method steps S110 to S150 Figure 2 Method steps S210 to S240, Figure 3 Method steps S310 to S330, Figure 4 Method steps S410 to S460 Figure 5 Method steps S510 to S520 Figure 6 Alternatively, the method steps S610 to S620 can be performed, or the model training method of the embodiments of this application can be executed, for example, the method described above. Figure 7 Method steps S710 to S740.

[0192] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0193] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0194] It will be understood by those skilled in the art that Figures 1 to 7 The technical solutions shown do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0195] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0196] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0197] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0198] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0199] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0200] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0201] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0202] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0203] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for predicting the remaining life of an engine, characterized in that, include: Acquire multiple first engine sample data, wherein the first engine sample data carries a remaining life tag; The encoding module determines the first sample features of each first engine sample data, and the clustering module and the label mapping relationship determine the type label corresponding to each first sample feature. The label mapping relationship of the later clustering round is determined by the label mapping relationship of the previous clustering round. The type label is used to characterize the engine fault type. Each of the first sample features is input into the prediction module associated with the corresponding type label, and the remaining lifespan of the first engine sample data is determined based on the prediction module. The prediction loss is determined based on the remaining lifetime of each sample and the corresponding remaining lifetime label. The encoding module and the prediction module are jointly trained based on the prediction loss to obtain the trained remaining lifetime prediction model. Obtain engine target data, input the engine target data into the trained remaining life prediction model, and determine the target remaining life of the engine target data based on the trained remaining life prediction model. The step of determining the type label corresponding to each feature of the first sample based on the clustering module and the label mapping relationship includes: Obtain multiple initial centroids and the label mapping relationship corresponding to each initial centroid, wherein the initial centroids are centroids determined in the previous clustering round or preset; Each of the first sample features is input into the clustering module, and the clustering module performs clustering processing on each of the first sample features to obtain multiple first feature clusters; Traverse the first feature cluster, and determine the initial centroid closest to the current centroid of the first feature cluster based on the distance between the current centroid of the first feature cluster and each of the initial centroids. Use the label mapping relationship corresponding to the closest initial centroid as the label mapping relationship corresponding to the first feature cluster. Based on the label mapping relationship corresponding to each first feature cluster, the type label corresponding to each first sample feature is determined.

2. The method according to claim 1, characterized in that, The step of determining the prediction loss based on the remaining lifetime of each sample and the corresponding remaining lifetime label, and jointly training the encoding module and the prediction module based on the prediction loss, includes: The remaining lifetime of each sample predicted by the prediction module is subtracted from the corresponding remaining lifetime label, and the square of the subtraction result is used as the prediction distance value. The prediction loss corresponding to the prediction module is determined based on the average value of each prediction distance value corresponding to the prediction module. The prediction modules are traversed, and the encoding module and the prediction module are jointly trained according to the prediction loss corresponding to the prediction module.

3. The method according to claim 1, characterized in that, Before determining the first sample features of each of the first engine sample data based on the encoding module, the method further includes: Acquire multiple second engine sample data and multiple sample pairs, wherein the sample pair includes two third engine sample data and the sample pair carries a similarity label; The second sample features of each second engine sample data are determined based on the encoding module, and the first reconstructed data of each second sample feature are determined based on the decoding module. Based on the first reconstructed data and the corresponding second engine sample data, the first target loss is determined; Based on the encoding module, the third sample features of the two third engine sample data in the sample pair are determined respectively, and the predicted similarity between the two third sample features is determined. Based on the predicted similarity and the corresponding similarity label, determine the second target loss; The encoding module is pre-trained alternately based on the first target loss and the second target loss.

4. The method according to claim 3, characterized in that, After determining the first target loss based on the first reconstructed data and the corresponding second engine sample data, the method further includes: Each of the second sample features is input into the clustering module, and the clustering module performs clustering processing on each of the second sample features to obtain multiple second feature clusters; The clustering loss is determined based on the distance between each second sample feature and the current centroid of the corresponding second feature cluster, and the first target loss is adjusted based on the clustering loss.

5. The method according to claim 3, characterized in that, After determining the second target loss based on the predicted similarity and the corresponding similarity label, the method further includes: The second reconstructed data for each of the third sample features is determined based on the decoding module; Based on the second reconstruction data and the corresponding third engine sample data, the reconstruction loss is determined, and the second target loss is adjusted based on the reconstruction loss.

6. A model training method, characterized in that, include: Acquire multiple first engine sample data, wherein the first engine sample data carries a remaining life tag; The encoding module determines the first sample features of each first engine sample data, and the clustering module and the label mapping relationship determine the type label corresponding to each first sample feature. The label mapping relationship of the later clustering round is determined by the label mapping relationship of the previous clustering round. The type label is used to characterize the engine fault type. Each of the first sample features is input into the prediction module associated with the corresponding type label, and the remaining lifespan of the first engine sample data is determined based on the prediction module. The prediction loss is determined based on the remaining lifetime of each sample and the corresponding remaining lifetime label. The encoding module and the prediction module are jointly trained based on the prediction loss to obtain the trained remaining lifetime prediction model. The step of determining the type label corresponding to each feature of the first sample based on the clustering module and the label mapping relationship includes: Obtain multiple initial centroids and the label mapping relationship corresponding to each initial centroid, wherein the initial centroids are centroids determined in the previous clustering round or preset; Each of the first sample features is input into the clustering module, and the clustering module performs clustering processing on each of the first sample features to obtain multiple first feature clusters; Traverse the first feature cluster, and determine the initial centroid closest to the current centroid of the first feature cluster based on the distance between the current centroid of the first feature cluster and each of the initial centroids. Use the label mapping relationship corresponding to the closest initial centroid as the label mapping relationship corresponding to the first feature cluster. Based on the label mapping relationship corresponding to each first feature cluster, the type label corresponding to each first sample feature is determined.

7. An engine remaining life prediction device, characterized in that, include: The acquisition module is used to acquire multiple first engine sample data, wherein the first engine sample data carries a remaining life tag; The determination module is used to determine the first sample features of each first engine sample data based on the encoding module, and to determine the type label corresponding to each first sample feature based on the clustering module and the label mapping relationship. The label mapping relationship of the later clustering round is determined by the label mapping relationship of the previous clustering round, and the type label is used to characterize the engine fault type. The first prediction module is used to input each of the first sample features into the prediction module associated with the corresponding type label, and determine the remaining lifespan of the first engine sample data based on the prediction module. The training module is used to determine the prediction loss based on the remaining lifetime of each sample and the corresponding remaining lifetime label, and to jointly train the encoding module and the prediction module based on the prediction loss to obtain the trained remaining lifetime prediction model. The second prediction module is used to acquire engine target data, input the engine target data into the trained remaining life prediction model, and determine the target remaining life of the engine target data based on the trained remaining life prediction model. The step of determining the type label corresponding to each feature of the first sample based on the clustering module and the label mapping relationship includes: Obtain multiple initial centroids and the label mapping relationship corresponding to each initial centroid, wherein the initial centroids are centroids determined in the previous clustering round or preset; Each of the first sample features is input into the clustering module, and the clustering module performs clustering processing on each of the first sample features to obtain multiple first feature clusters; Traverse the first feature cluster, and determine the initial centroid closest to the current centroid of the first feature cluster based on the distance between the current centroid of the first feature cluster and each of the initial centroids. Use the label mapping relationship corresponding to the closest initial centroid as the label mapping relationship corresponding to the first feature cluster. Based on the label mapping relationship corresponding to each first feature cluster, the type label corresponding to each first sample feature is determined.

8. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the engine remaining life prediction method according to any one of claims 1 to 5, or the model training method according to claim 6.

9. A storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the engine remaining life prediction method according to any one of claims 1 to 5, or the model training method according to claim 6.

Images