A method and device for predicting remaining service life of an engine based on a hybrid model
Through a hybrid model-based method, combined with stacking integration technology and neural network model, the problem of low prediction accuracy of engine residual service life is solved, and high-precision and reliable prediction results are achieved.
Patent Information
- Application Number
- CN202210698982.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-06-20
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2042-06-20
AI Technical Summary
In the prior art, the engine residual service life prediction accuracy is low, making it difficult to accurately predict the remaining service life of an aircraft engine.
Using a hybrid model-based method, by establishing an initial hybrid model and linear regression model, using stacking integration technology, attention mechanism, convolutional neural network and other technologies, iterative training and cross-validation are improved to improve the credibility and accuracy of the prediction results.
High-precision prediction of the remaining service life of the engine is realized, reducing the dependence on the prediction results of a single model, and improving the reliability and accuracy of the prediction results.
Smart Images

Figure CN115017819B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of aircraft engine fault prediction and health management, and in particular to a method and device for predicting the remaining service life of an engine based on a hybrid model. Background Art
[0002] With the maturity of manufacturing industry and industrial system, various mechanical equipment and devices play a very important role in various industries. As a key component of aircraft, if an aircraft engine suddenly fails during operation, it will cause a huge disaster. Predictive health management technology (PHM) of equipment is a good management solution for equipment. The remaining service life prediction of aircraft engines is a very important type of PHM technology.
[0003] In order to obtain the remaining service life of the equipment, the commonly used remaining service life prediction methods are roughly divided into three types: methods based on traditional physical models; data-driven methods; and methods that mix the two. Since it is difficult for methods based on physical models to give accurate prediction results and hybrid methods are difficult to implement, data-driven methods are often used. However, existing data-driven methods are mainly based on deep learning models, which learn to process sensor data to obtain prediction results, and the prediction accuracy is limited.
[0004] Therefore, the prior art has the problem of low accuracy in predicting the remaining useful life of the engine. Summary of the invention
[0005] In view of this, it is necessary to provide a method and device for predicting the remaining useful life of an engine based on a hybrid model to solve the problem of low accuracy in predicting the remaining useful life of an engine in the prior art.
[0006] In order to solve the above problems, the present invention provides a method for predicting the remaining service life of an engine based on a hybrid model, comprising:
[0007] Acquire multiple sets of sensor data samples of a sample engine;
[0008] Establishing an initial hybrid model, wherein the initial hybrid model includes a first model and a second model; inputting multiple sets of sensor data samples into the first model / the second model respectively, taking the first remaining life sample / the second remaining life sample as output, obtaining the first predicted remaining life sample / the second predicted remaining life sample of the multiple sets of sensor data samples by stacking and integrating, iterating the training to a first preset number of times, and determining a fully trained target hybrid model;
[0009] Inputting the first predicted remaining life sample and the second predicted remaining life sample into the initial linear regression model, taking the remaining service life of the sample engine as output, iteratively training to a second preset number of times, and obtaining a fully trained target linear regression model;
[0010] Multiple sets of sensor data of the engine to be tested are obtained, and the remaining service life of the engine to be tested is determined based on a target hybrid model and a target linear regression model.
[0011] Further, the multiple groups of sensor data samples are respectively input into the first model / the second model, and the first remaining life sample / the second remaining life sample is output, and the first predicted remaining life sample / the second predicted remaining life sample of the multiple groups of sensor data samples are obtained by stacking and integrating, and iterative training is performed to a first preset number of times to determine a fully trained target hybrid model, including:
[0012] Divide the multiple groups of sensor data samples into K parts, and input them into the first model and the second model respectively, wherein K-1 parts use the corresponding first remaining life samples / second remaining life samples as outputs, and the remaining part uses the initial first predicted remaining life samples / initial second predicted remaining life samples as outputs, and performs K-fold cross validation through stacking integration, iterates calculation to a first preset number of times, and determines the initial first predicted remaining life samples / initial second predicted remaining life samples corresponding to the multiple groups of sensor data samples;
[0013] According to the initial first predicted remaining life sample / initial second predicted remaining life sample, a first predicted remaining life sample and a second predicted remaining life sample of the plurality of groups of sensor data samples are determined by an averaging operation.
[0014] Furthermore, the first model includes a first input layer, a first attention mechanism module, a temporal convolutional neural network module, a first fully connected layer and a first output layer which are connected in sequence.
[0015] Furthermore, the second model includes a second input layer, a convolutional neural network module, a bidirectional gated recurrent neural network module, a second attention mechanism module, a second fully connected layer and a second output layer connected in sequence.
[0016] Furthermore, multiple sets of sensor data samples of the sample engine are obtained, including:
[0017] Acquire multiple groups of initial samples of sensor data of the sample engine, remove abnormal values, and obtain multiple groups of sensor feature samples;
[0018] The multiple groups of sensor feature samples are normalized to obtain multiple groups of processed sensor data samples.
[0019] Further, the first predicted remaining life sample / the second predicted remaining life sample of the plurality of groups of sensor data samples are obtained by stacking and integrating, including:
[0020] The first predicted remaining life initial sample / the second predicted remaining life initial sample are processed by a piecewise line function to obtain the first predicted remaining life sample / the second predicted remaining life sample of multiple groups of sensor data samples.
[0021] Furthermore, the trained target linear regression model is obtained, which also includes:
[0022] Establish evaluation indicators to evaluate the test results of the linear regression model;
[0023] According to the evaluation results, the relevant parameters of the linear regression model are adjusted to obtain a fully trained target linear regression model.
[0024] In order to solve the above problems, the present invention also provides an engine remaining service life prediction device based on a hybrid model, comprising:
[0025] A sample acquisition module, used to acquire multiple sets of sensor data samples of a sample engine;
[0026] A hybrid model acquisition module is used to establish an initial hybrid model, wherein the initial hybrid model includes a first model and a second model; input multiple sets of sensor data samples into the first model / second model respectively, take the first remaining life sample / second remaining life sample as output, obtain the first predicted remaining life sample / second predicted remaining life sample of the multiple sets of sensor data samples by stacking and integration, iterate training to a first preset number of times, and determine a fully trained target hybrid model;
[0027] A linear regression model acquisition module, used for inputting the first predicted remaining life sample and the second predicted remaining life sample into an initial linear regression model, taking the remaining life of the sample engine as output, iteratively training to a second preset number of times, and obtaining a fully trained target linear regression model;
[0028] The remaining service life determination module is used to obtain multiple sets of sensor data of the engine to be tested, and determine the remaining service life of the engine to be tested based on a target hybrid model and a target linear regression model.
[0029] In order to solve the above problem, the present invention also provides a storage medium, which stores computer program instructions. When the computer program instructions are executed by a computer, the computer executes the engine remaining service life prediction method based on the hybrid model as described above.
[0030] The beneficial effect of adopting the above technical solution is: the present invention provides a method and device for predicting the remaining service life of an engine based on a hybrid model, the method comprising: training the first model and the second model in the initial hybrid model according to the sensor data sample of the sample engine, and determining the fully trained target hybrid model; then, through the stacking integration technology, obtaining the first predicted remaining life and the second predicted remaining life of the sensor data sample of the sample engine; next, inputting the first predicted remaining life and the second predicted remaining life into the initial linear regression model respectively, and obtaining the fully trained target linear regression model after training; finally, based on the target hybrid model and the target linear regression model, predicting the remaining service life of the engine to be tested. In the present application, by setting a hybrid model composed of multiple models and using the stacking integration technology to predict the remaining service life of the engine, the remaining service life of the engine is predicted from multiple angles and multiple directions, and the credibility of the remaining service life of the engine is improved. Then, the linear regression model organically merges the multiple prediction results in the hybrid model, and the weight of the prediction result of each single model is redistributed, which reduces the dependence on a certain prediction result and improves the accuracy of the prediction result. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 A schematic flow chart of an embodiment of a method for predicting the remaining useful life of an engine based on a hybrid model provided by the present invention;
[0032] Figure 2 A schematic structural diagram of a first model embodiment provided by the present invention;
[0033] Figure 3 A schematic structural diagram of a second model 1 embodiment provided by the present invention;
[0034] Figure 4 A schematic diagram of the structure of the engine remaining service life prediction device based on the hybrid model provided by the present invention;
[0035] Figure 5 This is a structural block diagram of an embodiment of an electronic device provided by the present invention. DETAILED DESCRIPTION
[0036] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.
[0037] Before describing the embodiments, the attention mechanism, convolutional neural network, temporal convolutional neural network, and recurrent neural network are described:
[0038] The attention mechanism is a mechanism that enables a neural network to focus on a subset of its inputs (or features), enabling the selection of specific inputs. Attention can be applied to any type of input. In the case of limited computing power, the attention mechanism is a resource allocation solution that is the main means of solving the problem of information overload, allocating computing resources to more important tasks.
[0039] Among them, attention is generally divided into two types: one is top-down conscious attention, called focused attention, which refers to attention that is intentional, task-dependent, and actively and consciously focused on an object; the other is bottom-up unconscious attention, called saliency-based attention, which is driven by external stimuli, does not require active intervention, and is irrelevant to the task. If the stimulus information of an object is different from the surrounding information, an unconscious "winner-take-all" or gating mechanism can turn attention to this object. Regardless of whether this attention is intentional or unintentional, most human brain activities rely on attention, such as memorizing information, reading or thinking.
[0040] Convolutional neural network (CNN) is a deep neural network with a convolutional structure. The convolutional structure can reduce the amount of memory occupied by the deep network. Its three key operations are local receptive field, weight sharing, and pooling layer, which can effectively reduce the number of parameters of the neural network and alleviate the overfitting problem of the model.
[0041] Temporal Convolutional Networks start from a general architecture for convolutional sequence prediction. It is not an architecture, but a type of architecture. The characteristics of TCN networks include: 1. The relationship in the convolution is causal, which means that there is no information from the future to the past. 2. This architecture can use sequences of arbitrarily long lengths as input and map them to outputs of the same length. 3. Use residual layers and dilated convolution techniques.
[0042] A recurrent neural network (RNN) is a type of recursive neural network that takes sequence data as input, performs recursion in the direction of sequence evolution, and all nodes (recurrent units) are connected in a chain.
[0043] In the gradient calculation method of recurrent neural networks, when the number of time steps is large or the time step is small, the gradient of the recurrent neural network is more likely to decay or explode. Therefore, it is difficult for recurrent neural networks to capture dependencies with large time step distances in time series in practice.
[0044] The introduction of gated recurrent neural network can better capture the dependencies with large time step distance in time series. It controls the flow of information through learnable gates. Among them, gated recurrent unit (GRU) is a commonly used gated recurrent neural network.
[0045] At present, in order to predict the remaining service life of an engine, a deep learning method is generally used to train a neural network model, and then the trained neural network model is used to predict the remaining service life of the engine.
[0046] However, since each single deep learning model has a different structure and focuses on different data features, the prediction accuracy in the existing technology is limited, and there is a problem of low accuracy in predicting the remaining useful life of the engine.
[0047] In order to solve the problem of low accuracy of engine remaining service life prediction using a single deep learning neural network in the prior art, the present invention provides an engine remaining service life prediction method, device, electronic device and storage medium based on a hybrid model, which are described in detail below.
[0048] like Figure 1 As shown, Figure 1 A flow chart of an embodiment of a method for predicting the remaining useful life of an engine based on a hybrid model provided by the present invention includes:
[0049] Step S101: Acquire multiple groups of sensor data samples of a sample engine.
[0050] Step S102: Establish an initial hybrid model, wherein the initial hybrid model includes a first model and a second model; input multiple groups of sensor data samples into the first model / second model respectively, take the first remaining life sample / second remaining life sample as the output, obtain the first predicted remaining life sample / second predicted remaining life sample of the multiple groups of sensor data samples by stacking and integrating, iterate the training to a first preset number of times, and determine a fully trained target hybrid model.
[0051] Step S103: Input the first predicted remaining life sample and the second predicted remaining life sample into the initial linear regression model, take the remaining life of the sample engine as output, iterate training to a second preset number of times, and obtain a fully trained target linear regression model.
[0052] Step S104: Acquire multiple sets of sensor data of the engine to be tested, and determine the remaining service life of the engine to be tested based on the target hybrid model and the target linear regression model.
[0053] In this embodiment, first, multiple groups of sensor data samples of the sample engine are obtained. In order to improve the training level of the model, the number of sensor data samples should be as large as possible; secondly, an initial hybrid model is established, wherein the initial hybrid model includes a first model and a second model; multiple groups of sensor data samples are respectively input into the first model / second model, and the first remaining life sample / second predicted life sample is used as the output to train the first model and the second model, and then the first predicted remaining life sample and the second predicted remaining life sample corresponding to the multiple groups of sensor data samples are obtained by stacking and integration, and the training is iterated to a first preset number of times to obtain the first model and the second model that are fully trained, thereby determining the fully trained target hybrid model; then, the first predicted remaining life sample and the second predicted remaining life sample are input into the initial linear regression model, and the remaining service life of the sample engine is used as the output, and the training is iterated to a second preset number of times to obtain the fully trained target linear regression model; finally, in order to determine the remaining service life of the engine to be tested, multiple groups of sensor data of the engine to be tested are input into the fully trained target hybrid model to obtain the first predicted remaining life and the second predicted remaining life, respectively; then, the first predicted remaining life and the second predicted remaining life are respectively input into the target linear regression model, so that the remaining service life of the engine to be tested can be determined.
[0054] It can be understood that the stacking integration technology is used in the above-mentioned embodiment to obtain the corresponding first predicted remaining life and second predicted remaining life based on multiple groups of sensor data of the engine to be tested. Since the stacking integration technology traverses all sensor data and iterates multiple times during use, and also averages the results of multiple predictions, the first predicted remaining life and the second predicted remaining life obtained by stacking integration have higher accuracy; further, the target linear regression model is used to determine the final remaining service life of the engine to be tested based on the first predicted remaining life and the second predicted remaining life, which not only effectively utilizes the prediction effects of the first model and the second model, but also organically distributes and optimizes the prediction effects of the first model and the second model, thereby improving the accuracy of the remaining service life of the engine to be tested that is finally obtained.
[0055] In summary, in order to improve the accuracy of the prediction results, the first predicted remaining life / second predicted remaining life obtained in this embodiment based on the stacking integration technology has high reliability; in addition, the prediction results of the two models are weighted again using the target linear regression model, which reduces the dependence on a certain prediction result and improves the accuracy of the prediction results.
[0056] As a preferred embodiment, in step S101, obtaining multiple sets of sensor data samples of a sample engine includes:
[0057] Firstly, the historical failure data of the sample engine is obtained, and the outliers are eliminated to obtain the sensor characteristic change samples; then, the sensor characteristic change samples are normalized to obtain multiple groups of sensor data samples.
[0058] In a specific embodiment, historical failure data of a sample engine is obtained, all sensor data are listed, and sensor data that has not changed is eliminated to obtain a sensor characteristic change sample; that is, sensor data that has not changed in the historical failure data is eliminated to obtain a sensor characteristic change sample.
[0059] Furthermore, it is necessary to normalize the sensor feature change samples to obtain sensor data samples. In this embodiment, the Min-Max normalization method is used to scale the feature data to the range of [0, 1]. The specific formula is as follows:
[0060]
[0061] in is the normalized data, is the value in the ith feature sequence, is the minimum value in the feature sequence, is the maximum value in the feature sequence.
[0062] In this embodiment, by normalizing the sensor characteristic change samples, not only can the adverse effects caused by singular sample data be eliminated, but also the difficulty of calculation can be reduced, which helps to improve the prediction accuracy.
[0063] As a preferred embodiment, in step S102, after multiple groups of sensor data samples are input into the first model and the second model respectively, K-fold cross-validation is performed through stacking integration to determine the first predicted remaining life sample and the second predicted remaining life sample.
[0064] That is, multiple groups of sensor data samples are divided into K parts, which are input into the first model / second model respectively, wherein K-1 parts have the corresponding first remaining life sample / second remaining life sample as output, and the remaining part has the initial first predicted remaining life sample / initial second predicted remaining life sample as output, and K-fold cross validation is performed through stacking integration, and iterative calculation is performed to a first preset number of times to determine the initial first predicted remaining life sample / initial second predicted remaining life sample corresponding to the multiple groups of sensor data samples;
[0065] According to the initial first predicted remaining life sample / initial second predicted remaining life sample, a first predicted remaining life sample and a second predicted remaining life sample of the plurality of groups of sensor data samples are determined by an averaging operation.
[0066] In a specific embodiment, the value of K may be 5. In other embodiments, K may theoretically be any non-zero positive integer.
[0067] Furthermore, after obtaining the first predicted remaining life sample and the second predicted remaining life sample with higher reliability, since the remaining service life of the engine has a threshold range, in order to meet actual problems, the first predicted remaining life sample and the second predicted remaining life sample need to be corrected.
[0068] In a specific embodiment, the predicted remaining service life sample is corrected by a piecewise function, and the maximum value of the remaining service life of the engine is set to 130. The piecewise line function formula is as follows:
[0069]
[0070] Among them, RUL is the revised predicted remaining life sample, is the true value of the predicted remaining life sample, and 130 is the maximum value.
[0071] Preferably, the first model includes a first attention mechanism, a temporal convolutional neural network and a first fully connected layer. Figure 2 As shown, Figure 2 This is a structural schematic diagram of an embodiment of the first model provided by the present invention. The first model 200 includes a first input layer 201, a first attention mechanism module 202, a temporal convolutional neural network module 203, a first fully connected layer 204 and a first output layer 205 connected in sequence.
[0072] Furthermore, the temporal convolutional neural network module 203 includes an extended causal convolution, a weight initialization, a Relu activation layer, a discard layer and a convolution layer, and the convolution layer is a 1*1 convolution layer; the first fully connected layer 204 includes two fully connected layers.
[0073] In a specific embodiment, multiple groups of sensor data samples are input into the first input layer 201. First, after being processed by the first attention mechanism module 202, attention is paid to key sensor data sample features to improve the robustness of the first model 200. Secondly, the temporal convolutional neural network module 203 is mainly composed of a residual network and an extended causal convolution. After the extended causal convolution, feature extraction is performed on the sensor data samples, and the residual module increases the receptive field of the model and improves the learning ability of the model. By learning the sensor features in the sensor data samples and extracting the sensor features of the sensor data samples, the first model 200 can better learn the dependency between the sensor features and the remaining service life of the engine, wherein the activation function of the temporal convolutional neural network module 203 is the relu function. Next, after the first fully connected layer 204, the learning result is mapped out, that is, the predicted value of the remaining service life of the sample engine, wherein the activation function of the first fully connected layer 204 is the relu function, and the activation function of the second fully connected layer 204 is the linear function. Finally, the first output layer 205 outputs the predicted value of the remaining service life of the sample engine.
[0074] In a specific embodiment, the calculation formula of the temporal convolutional neural network module 203 is:
[0075]
[0076] Where X∈R n is a one-dimensional sequence input, X∈R n f : {0,...,k-1} → R is a filter, is the expansion factor, is the filter size, Explains the direction of the past.
[0077] Preferably, the second model includes a convolutional neural network, a bidirectional gated recurrent neural network, a second attention mechanism, and a second fully connected layer. Figure 3 As shown, Figure 3 This is a structural schematic diagram of an embodiment of the second model provided by the present invention. The second model 300 includes a second input layer 301, a convolutional neural network module 302, a bidirectional gated recurrent neural network module 303, a second attention mechanism module 304, a second fully connected layer 305 and a second output layer 306 connected in sequence.
[0078] Furthermore, the convolutional neural network module 302 includes a convolutional layer and a pooling layer; the bidirectional gated recurrent neural network module 303 includes a forward BLSTM and a reverse BLSTM, and the second fully connected layer 305 includes two fully connected layers.
[0079] In a specific embodiment, multiple groups of sensor data samples are input into the second input layer 301. First, the sensor features are extracted through the convolution layer and pooling layer of the convolutional neural network module 302. Secondly, the bidirectional gated recurrent neural network module 303 can quickly learn the relationship between the sensor features and the labels based on the sensor features, and clarify the time dependency of the sensor data sequence. Then, after being processed by the second attention mechanism module 304, attention is paid to features that are more relevant to predicting the remaining service life of the engine. Finally, after being processed by the second fully connected layer 305, the result of the mapping learning, that is, the predicted value of the remaining service life of the sample engine, is mapped. Finally, the second output layer 306 outputs the predicted value of the remaining service life of the sample engine.
[0080] Among them, the activation function used by the convolutional neural network module 302 is the tanh function, the activation function used by the bidirectional gated recurrent neural network module 303 is the relu function, the activation function of the first second fully connected layer 305 is the relu function, and the activation function of the second second fully connected layer 305 is the linear function.
[0081] In this embodiment, based on the convolutional neural network module 302, it is possible to quickly extract the sensor data features of the sample engine, based on the bidirectional gated recurrent neural network module 303, it is possible to achieve rapid convergence, and based on the second attention mechanism module 304, it is possible to improve the robustness of the prediction results. Therefore, the second model 300 can obtain a better prediction result of the remaining service life of the sample engine.
[0082] In a specific embodiment, the calculation formula of the convolutional neural network module 302 is:
[0083]
[0084] in, is the generated feature, are the input features, represents the convolution kernel, is the bias term, is the activation function.
[0085] In a specific embodiment, the calculation formula of the bidirectional gated recurrent neural network module 303 is:
[0086]
[0087] in, is the internal network function, are the weights of the forward BLSTM output and the reverse BLSTM output at time t, is the bias of the output layer, represents the unit state at time t, represents the input at time t, represents the forward unit state at time t-1, Represents the backward unit state at time t-1.
[0088] In a specific embodiment, the calculation formula of the second attention mechanism module 304 is:
[0089]
[0090] in, is the weight matrix, is the bias term, Represents input, Represents the normalized result of the input, c t Represents the attention mechanism.
[0091] After the first model and the second model are fully trained, the first predicted remaining life sample and the second predicted remaining life sample can also be corrected. Generally, since the first model and the second model are not completely the same, the first predicted remaining life sample and the second predicted remaining life sample are not equal.
[0092] In other embodiments, the hybrid model may also include other numbers of multiple models of other types, and simultaneously predict the remaining life of the engine, thereby obtaining multiple corresponding prediction results.
[0093] Furthermore, in order to effectively utilize the first model and the second model, unify the remaining life prediction results of the engine, and improve the accuracy of the prediction results, the present application also uses a linear regression model as a learner, inputs the first predicted remaining life sample and the second predicted remaining life sample into the initial linear regression model, and uses the remaining service life of the sample engine as the output, iteratively trains the linear regression model to a second preset number of times, and obtains a fully trained target linear regression model.
[0094] In order to test the accuracy of the prediction results based on the linear regression model, an evaluation index is also established to evaluate the prediction results. The evaluation index is as follows:
[0095]
[0096] Where n represents the total number of sample engines, Represents the error between the predicted value and the true value of the i-th sample engine.
[0097] Finally, in order to clearly illustrate in detail the process of obtaining the remaining service life of the engine to be tested based on multiple sets of sensor data by training a complete target hybrid model and a linear regression model, the present application is illustrated through the following specific embodiments.
[0098] In this embodiment, the C-MPASS dataset for aircraft engines provided by NASA is selected for illustration. The C-MPASS dataset is a turbofan engine dataset widely used for RUL prediction. It has four sub-datasets (FD001 to FD004), each of which has historical data of turbofan engines under different operating conditions and failure modes. This embodiment uses the first sub-dataset (FD001), which has degradation data of 100 engines.
[0099] First, the changes in the sensor data in the above data set are listed, including 21 sensor data and 3 operating parameters. The sensor data numbered 1, 5, 10, 16, 18, and 19 have not changed, and the third operating parameter has not changed either, so these parameters are discarded and the current operating cycle of the engine is selected as one of the features.
[0100] Secondly, the Min-Max normalization method is used to scale the feature data to the range of [0, 1]. The segmented line function is used to correct the remaining service life of the engine, and the segmented line function threshold is set to 130.
[0101] Then, according to the operating cycle of the engine to be tested, the sensor data is cut to construct sliding time windows of different sizes. In this embodiment, 30, 60, 90, and 120 are selected as the cutting standards for the time window size to divide the sensor data into different data sets.
[0102] Next, the data set is input into the first model and the second model respectively to obtain the first predicted remaining life and the second predicted remaining life respectively. The first model and the second model are K-fold cross-validated through the stacked integration framework in the integrated learning to obtain new training set and test set.
[0103] Finally, the linear regression model is used as the final learner, and the new training set and test set are input into the final learner to obtain the remaining service life of the engine to be tested.
[0104] Furthermore, the effect of this method can be evaluated by comparing the prediction accuracy of this method with other methods. The RMSE and Score evaluation results of this method and other methods are shown in the following table:
[0105]
[0106] In summary, the engine remaining service life prediction method based on the hybrid model provided by the present invention can accurately predict the remaining service life of the engine. As can be seen from the above table, the performance index of this method is much better than other methods. The present invention further improves the accuracy of the prediction results by constructing multiple models to learn the data and then re-learning the results after model learning.
[0107] In order to solve the above problems, the present invention also provides a device for predicting the remaining service life of an engine based on a hybrid model, such as Figure 4 As shown, Figure 4 The present invention provides a schematic diagram of the structure of the engine remaining service life prediction device based on the hybrid model. The engine remaining service life prediction device 400 based on the hybrid model includes:
[0108] The sample acquisition module 401 is used to acquire multiple sets of sensor data samples of a sample engine;
[0109] A hybrid model acquisition module 402 is used to establish an initial hybrid model, wherein the initial hybrid model includes a first model and a second model; input multiple sets of sensor data samples into the first model / second model respectively, take the first remaining life sample / second remaining life sample as output, obtain the first predicted remaining life sample / second predicted remaining life sample of the multiple sets of sensor data samples by stacking and integration, iterate training to a first preset number of times, and determine a fully trained target hybrid model;
[0110] The linear regression model acquisition module 403 is used to input the first predicted remaining life sample and the second predicted remaining life sample into the initial linear regression model, take the remaining life of the sample engine as the output, iterate the training to a second preset number of times, and obtain a fully trained target linear regression model;
[0111] The remaining service life determination module 404 is used to obtain multiple sets of sensor data of the engine to be tested, and determine the remaining service life of the engine to be tested based on the target hybrid model and the target linear regression model.
[0112] The present invention also provides an electronic device, such as Figure 5 As shown, Figure 5 This is a structural block diagram of an embodiment of an electronic device provided by the present invention. The electronic device 500 can be a computing device such as a mobile terminal, a desktop computer, a notebook, a palm computer, and a server. The electronic device 500 includes a processor 501 and a memory 502, wherein the memory 502 stores an engine remaining service life prediction program 503 based on a hybrid model.
[0113] In some embodiments, the memory 502 may be an internal storage unit of a computer device, such as a hard disk or memory of the computer device. In other embodiments, the memory 502 may also be an external storage device of the computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device. Further, the memory 502 may also include both an internal storage unit of the computer device and an external storage device. The memory 502 is used to store application software and various types of data installed in the computer device, such as program codes installed in the computer device. The memory 502 may also be used to temporarily store data that has been output or is to be output. In one embodiment, the engine remaining service life prediction program 503 based on the hybrid model may be executed by the processor 501, thereby realizing the engine remaining service life prediction method based on the hybrid model of each embodiment of the present invention.
[0114] In some embodiments, the processor 501 may be a central processing unit (CPU), a microprocessor or other data processing chip, used to run program codes or process data stored in the memory 502, such as executing an engine remaining useful life prediction program based on a hybrid model.
[0115] This embodiment also provides a computer-readable storage medium, on which a hybrid model-based engine remaining useful life prediction program is stored. When the computer program is executed by a processor, the hybrid model-based engine remaining useful life prediction method as described in any of the above technical solutions is implemented.
[0116] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
[0117] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.
Claims
1. A method for predicting the remaining useful life of an engine based on a hybrid model. It is characterized in that include: Acquire multiple sets of sensor data samples of a sample engine; Establishing an initial hybrid model, wherein the initial hybrid model includes a first model and a second model; inputting the multiple sets of sensor data samples into the first model / the second model respectively, taking the first remaining life sample / the second remaining life sample as output, obtaining the first predicted remaining life sample / the second predicted remaining life sample of the multiple sets of sensor data samples by stacking and integrating, iterating training to a first preset number of times, and determining a fully trained target hybrid model; Inputting the first predicted remaining life sample and the second predicted remaining life sample into an initial linear regression model, taking the remaining service life of the sample engine as output, iteratively training to a second preset number of times, and obtaining a fully trained target linear regression model; Acquire multiple sets of sensor data of the engine to be tested, and determine the remaining service life of the engine to be tested based on the target hybrid model and the target linear regression model; The step of inputting the plurality of sensor data samples into the first model / the second model respectively, taking the first remaining life sample / the second remaining life sample as output, obtaining the first predicted remaining life sample / the second predicted remaining life sample of the plurality of sensor data samples by stacking and integrating, iterating the training to a first preset number of times, and determining a fully trained target hybrid model includes: Divide the multiple groups of sensor data samples into K parts, and input them into the first model and the second model respectively, wherein K-1 parts use the corresponding first remaining life samples / second remaining life samples as outputs, and the remaining part uses the initial first predicted remaining life samples / initial second predicted remaining life samples as outputs, and performs K-fold cross validation through stacking integration, iterates calculation to a first preset number of times, and determines the initial first predicted remaining life samples / initial second predicted remaining life samples corresponding to the multiple groups of sensor data samples; Determine, according to the initial first predicted remaining life sample / the initial second predicted remaining life sample, a first predicted remaining life sample and a second predicted remaining life sample of the plurality of groups of sensor data samples by an averaging operation; The first model includes a first input layer, a first attention mechanism module, a temporal convolutional neural network module, a first fully connected layer and a first output layer connected in sequence; The second model includes a second input layer, a convolutional neural network module, a bidirectional gated recurrent neural network module, a second attention mechanism module, a second fully connected layer and a second output layer connected in sequence.
2. The method for predicting the remaining useful life of an engine based on a hybrid model according to claim 1, It is characterized in that The step of obtaining multiple sets of sensor data samples of a sample engine includes: Acquire multiple groups of initial samples of sensor data of the sample engine, remove abnormal values, and obtain multiple groups of sensor feature samples; The multiple groups of sensor feature samples are normalized to obtain multiple groups of processed sensor data samples.
3. The method for predicting the remaining useful life of an engine based on a hybrid model according to claim 1, It is characterized in that The method of obtaining the first predicted remaining life sample / the second predicted remaining life sample of the plurality of groups of sensor data samples by stacking and integrating includes: The first predicted remaining life initial sample / the second predicted remaining life initial sample are processed by a piecewise line function to obtain the first predicted remaining life sample / the second predicted remaining life sample of the plurality of groups of sensor data samples.
4. The method for predicting the remaining useful life of an engine based on a hybrid model according to claim 1, It is characterized in that The obtained fully trained target linear regression model further includes: Establishing evaluation indicators to evaluate the test results of the linear regression model; According to the evaluation results, the relevant parameters of the linear regression model are adjusted to obtain a fully trained target linear regression model.
5. An engine remaining useful life prediction device based on a hybrid model, It is characterized in that include: A sample acquisition module, used to acquire multiple sets of sensor data samples of a sample engine; A hybrid model acquisition module is used to establish an initial hybrid model, wherein the initial hybrid model includes a first model and a second model; the multiple groups of sensor data samples are respectively input into the first model / the second model, and the first remaining life sample / the second remaining life sample is output, and the first predicted remaining life sample / the second predicted remaining life sample of the multiple groups of sensor data samples is obtained by stacking and integrating, and iterative training is performed to a first preset number of times to determine a fully trained target hybrid model; A linear regression model acquisition module, used for inputting the first predicted remaining life sample and the second predicted remaining life sample into an initial linear regression model, taking the remaining service life of the sample engine as output, iteratively training to a second preset number of times, and obtaining a fully trained target linear regression model; A remaining service life determination module, used for acquiring multiple sets of sensor data of the engine to be tested, and determining the remaining service life of the engine to be tested based on the target hybrid model and the target linear regression model; The step of inputting the plurality of sensor data samples into the first model / the second model respectively, taking the first remaining life sample / the second remaining life sample as output, obtaining the first predicted remaining life sample / the second predicted remaining life sample of the plurality of sensor data samples by stacking and integrating, iterating the training to a first preset number of times, and determining a fully trained target hybrid model includes: Divide the multiple groups of sensor data samples into K parts, and input them into the first model and the second model respectively, wherein K-1 parts use the corresponding first remaining life samples / second remaining life samples as outputs, and the remaining part uses the initial first predicted remaining life samples / initial second predicted remaining life samples as outputs, and performs K-fold cross validation through stacking integration, iterates calculation to a first preset number of times, and determines the initial first predicted remaining life samples / initial second predicted remaining life samples corresponding to the multiple groups of sensor data samples; Determine, according to the initial first predicted remaining life sample / the initial second predicted remaining life sample, a first predicted remaining life sample and a second predicted remaining life sample of the plurality of groups of sensor data samples by an averaging operation; The first model includes a first input layer, a first attention mechanism module, a temporal convolutional neural network module, a first fully connected layer and a first output layer connected in sequence; The second model includes a second input layer, a convolutional neural network module, a bidirectional gated recurrent neural network module, a second attention mechanism module, a second fully connected layer and a second output layer connected in sequence.
6. An electronic device, It is characterized in that It comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the method for predicting the remaining useful life of an engine based on a hybrid model as described in any one of claims 1 to 4 is implemented.
7. A storage medium, It is characterized in that The storage medium stores computer program instructions, and when the computer program instructions are executed by a computer, the computer is caused to execute the engine remaining service life prediction method based on a hybrid model according to any one of claims 1 to 4.
