Intelligent prediction method and device based on Mama deep learning network

Abstract

The invention discloses an intelligent prediction method and device based on a Mama deep learning network, and relates to the field of equipment performance intelligent prediction, and the method comprises the steps: obtaining industrial equipment sensor monitoring data, and carrying out the preprocessing; performing sample generation on the preprocessed sensor monitoring data based on a sliding time window strategy to obtain a time window sample; the method comprises the following steps: constructing a Mamb-GRU parallel network, wherein the Mamb-GRU parallel network comprises a linear marking layer, a Mamb module, a GRU module, a feature fusion module and a full connection layer; training the Mamb-GRU parallel network based on the time window sample to obtain a performance prediction model; and predicting the remaining service life and the health state of the industrial equipment based on the performance prediction model, thereby effectively combining the complementarity of the long-period degradation trend and the short-time dynamic features, improving the multi-scale feature fusion quality, and further improving the prediction precision.

Description

Technical Field

[0001] The present application relates to the field of intelligent prediction of equipment performance, and in particular to an intelligent prediction method and device based on the Mamba deep learning network. Background Art

[0002] As a key technology in the field of equipment prognostics and health management (PHM), equipment RUL prediction or health status prediction technology has received increasing attention in recent years. Existing methods are mainly divided into two categories: physical model-based and data-driven. With the development of Internet of Things and sensor technologies, data-driven methods, especially deep learning-based methods, have become the mainstream methods for equipment RUL prediction or health status prediction. However, traditional deep learning models have significant shortcomings when processing time series data. For example, while methods based on recurrent neural networks (RNNs) can capture local features, their ability to model long-term dependencies is limited, resulting in insufficient prediction accuracy. Transformer-based methods, while capable of processing long-series dependencies, suffer from high computational complexity and slow inference speed.

[0003] To reduce computational complexity and improve model inference speed, Gu et al. first proposed the Mamba model in December 2023. Through a structured state space model (SSM) and a selective scanning mechanism, the Mamba model exhibits the advantage of linear time complexity in long-sequence modeling, allowing it to more efficiently capture complex dependencies in sequence data. Since its introduction, it has garnered widespread attention from researchers. However, while the Mamba model can effectively capture global degradation trends, it still suffers from limited ability to capture local temporal features. Therefore, effectively combining the complementarity of long-term degradation trends with short-term dynamic features to improve the multi-scale feature fusion quality and prediction accuracy of the Mamba model is a technical challenge and a cutting-edge hot topic that needs to be addressed in the field of PHM. Summary of the Invention

[0004] The purpose of this application is to provide an intelligent prediction method and device based on the Mamba deep learning network, which can effectively combine the complementarity of long-term degradation trends and short-term dynamic features, improve the quality of multi-scale feature fusion, and thus improve prediction accuracy.

[0005] To achieve the above objectives, this application provides the following solutions: In a first aspect, the present application provides an intelligent prediction method based on a Mamba deep learning network, the intelligent prediction method based on a Mamba deep learning network comprising: Obtaining industrial equipment sensor monitoring data; Preprocessing the industrial equipment sensor monitoring data; Based on the sliding time window strategy, samples are generated from the pre-processed sensor monitoring data to obtain time window samples; Constructing a Mamba-GRU (Gated Recurrent Unit) parallel network, which includes: a linear labeling layer, a Mamba module, a GRU module, a feature fusion module, and a fully connected layer; Training the Mamba-GRU parallel network based on the time window samples to obtain a performance prediction model; The remaining useful life (RUL) or health status of the industrial equipment is predicted based on the performance prediction model.

[0006] Optionally, preprocessing the industrial equipment sensor monitoring data specifically includes the following steps: Removing abnormal data from the industrial equipment sensor monitoring data; Perform wavelet transform to reduce noise on the sensor monitoring data after removing abnormal data; The sensor monitoring data after wavelet transform denoising is normalized.

[0007] Optionally, the linear marking layer specifically adopts the following formula: ; in, express t The output of the linear labeling layer at each moment, express t The input of the linear labeling layer at each moment, Batch represents batch operation, and represents the parameters of the linear marker layer, L represents the length of the input sequence, d represents the feature dimension, R represents the set of real numbers.

[0008] Optionally, the Mamba module specifically adopts the following formula: ; ; in, represents the state matrix; and represents the projection parameters; x ( t )for tThe output of the linear labeling layer at each moment, Indicates the hidden state of the current input in the Mamba module, y ( t ) represents the output of the Mamba module, Indicates the hidden state of the previous moment in the Mamba module, R represents the set of real numbers, m Represents the state dimension, L Indicates the length of the input sequence.

[0009] Optionally, the GRU module specifically adopts the following formula: ; ; ; ; in, for t The reset gate of time, is the sigmoid activation function, For the GRU module t -1 moment of hidden state, Hidden The weight matrix, x ( t )for t The output of the linear labeling layer at each moment, Input to reset gate x ( t ), for t The door of renewal at all times, Input to the update gate x ( t ), To update the hidden state in the door The weight matrix, For the GRU module t The candidate hidden state at time t, tanh(·) is the tanh activation function, ☉ is the Hadamard product, To reset the weight matrix of the hidden state after gate processing, Hidden The weight matrix, For the GRU module t The output of the moment, that is, the output of the GRU module t The hidden state of the moment.

[0010] Optionally, the data processing process in the feature fusion module is as follows: The long-term degradation trend extracted by the Mamba module is spliced ​​with the short-term dynamic features captured by the GRU module to obtain the spliced ​​features; The concatenated features are calculated using a scaled dot product attention mechanism to obtain two feature scores; Normalize the two feature scores using the softmax function to obtain a dynamic weight distribution; The long-term degradation trend and the short-term dynamic feature are weighted and summed according to the dynamic weight distribution to achieve dynamic fusion and obtain a fusion feature.

[0011] Optionally, the wavelet transform denoising of the sensor monitoring data after removing abnormal data is specifically performed using the following formula: ; ; in, j Indicates scale, k Indicates translation, t Indicates time, f ( t ) is the original signal, c j,k are the coefficients obtained after discrete wavelet transform, is the discrete wavelet basis function, is the mother wavelet function, is the reconstructed signal.

[0012] In a second aspect, the present application provides an intelligent prediction device based on the Mamba deep learning network, wherein the intelligent prediction device for industrial equipment performance includes: Data acquisition module, used to obtain industrial equipment sensor monitoring data; A preprocessing module, used for preprocessing the industrial equipment sensor monitoring data; A time window sample generation module is used to generate samples from the preprocessed sensor monitoring data based on a sliding time window strategy to obtain time window samples; A Mamba-GRU parallel network construction module is used to construct a Mamba-GRU parallel network, wherein the Mamba-GRU parallel network includes: a linear labeling layer, a Mamba module, a GRU module, a feature fusion module, and a fully connected layer; A training module, configured to train the Mamba-GRU parallel network based on the time window samples to obtain a performance prediction model; The performance prediction module is used to predict the RUL or health status of industrial equipment based on the performance prediction model.

[0013] Optionally, preprocessing the industrial equipment sensor monitoring data specifically includes the following steps: Remove abnormal data from sensor monitoring data; Perform wavelet transform to reduce noise on the sensor monitoring data after removing abnormal data; The sensor monitoring data after wavelet transform denoising is normalized.

[0014] Optionally, the linear marking layer specifically adopts the following formula: ; in, express t The output of the linear labeling layer at each moment, express t The input of the linear labeling layer at each moment, Batch represents batch operation, and represents the parameters of the linear marker layer, L represents the length of the input sequence, d represents the feature dimension, R represents the set of real numbers.

[0015] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides an intelligent prediction method and device based on the Mamba deep learning network. Through the parallel structure of the Mamba module and the GRU module, it realizes the complementary capture of long-term degradation trends and short-term dynamic features, breaking through the representation bottleneck of traditional models. It introduces a scaled dot product attention mechanism to calculate the spliced ​​features, realizes multi-scale feature adaptive fusion, effectively suppresses the interference of noise and working condition disturbances, and ultimately achieves efficient fusion and high-precision prediction. Specifically, the linear time complexity of the Mamba module reduces the computational cost, and combined with the local feature capture capability of the GRU module, it significantly improves the prediction accuracy while ensuring computational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 A flowchart of an intelligent prediction method based on the Mamba deep learning network provided in one embodiment of the present application; Figure 2 This is a schematic diagram of the sliding time window processing process provided by this application; Figure 3 Schematic diagram of the overall structure of the Mamba-GRU parallel network provided in this application; Figure 4 Schematic diagram of the linear marking layer provided for this application; Figure 5 A schematic diagram of the structure of the Mamba module provided for this application; Figure 6 A schematic diagram of the structure of the GRU module provided in this application; Figure 7 This is a schematic diagram of the prediction effect of the FD001 test set provided in this application; Figure 8 This is a schematic diagram of the prediction effect of the FD003 test set provided in this application; Figure 9 This is a schematic diagram of the RUL increment prediction effect of a single engine provided in this application; wherein, Figure 9 (a) is the RUL prediction diagram of engine No. 24 in the FD001 test set; Figure 9 (b) is a schematic diagram of RUL prediction for engine No. 76 in the FD001 test set; Figure 9 (c) is the RUL prediction diagram of engine No. 39 in the FD003 test set; Figure 9 (d) is a schematic diagram of the RUL prediction of engine No. 99 in the FD003 test set. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0019] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0020] In an exemplary embodiment, Figure 1 As shown, an intelligent prediction method based on the Mamba deep learning network is provided. The method is executed by a computer device, specifically a computer device such as a terminal or a server, or a terminal and a server. In the embodiment of the present application, the method includes the following steps 101 to 106. Among them: Step 101: Acquire industrial equipment sensor monitoring data.

[0021] Step 102: Preprocess the industrial equipment sensor monitoring data.

[0022] Specifically, the preprocessing process includes feature selection based on sensor signal stability, wavelet transform noise reduction and Min-Max normalization, as follows: S1021: Feature selection based on sensor signal stability, i.e., removing abnormal data from sensor monitoring data.

[0023] In actual operation, sensors exhibit different changing patterns over time under single operating conditions. It can be found that some sensors do not change significantly over time. Obviously, these values ​​are not related to the device RUL and should be removed.

[0024] S1022: Discrete wavelet transform denoising, that is, performing discrete wavelet transform denoising on the sensor monitoring data after removing abnormal data.

[0025] In practice, the signals collected by sensors are subject to noise. To minimize the impact of noise on device RUL prediction results, this application uses the Discrete Wavelet Transform (DWT) to reduce noise on the original signals. The output sensor signal is processed using the following formula.

[0026] (1) (2) in, j Indicates scale, k Indicates translation, t Indicates time, f ( t ) is the original signal, c j,k are the coefficients obtained after discrete wavelet transform, is the discrete wavelet basis function, is the mother wavelet function, is the reconstructed signal.

[0027] S1023: performing minimum-maximum normalization processing on the sensor monitoring data to determine normalized observation data, that is, performing normalization processing on the sensor monitoring data after wavelet transform noise reduction.

[0028] In practical applications, when the sensor monitoring data collected by sensors vary in size and range, directly using them may have an adverse effect on model stability and convergence during training. To address this issue, this application uses the min-max normalization method to scale all selected sensor values ​​to a uniform range of [0, 1], as defined in Equation (3).

[0029] (3) in, Indicates the i The normalized measurement value of each sensor is between [0,1]; Indicates the i The first sensor j measurements, and Respectively represent i The minimum and maximum values ​​of the sensor monitoring data.

[0030] Step 103: Generate samples of the pre-processed sensor monitoring data based on a sliding time window strategy to obtain time window samples.

[0031] While step-by-step input to the model is simple and computationally efficient, it cannot fully capture long-term dependencies and complex dynamic behaviors. To address this limitation, this application adopts a time window input strategy that provides additional contextual information for each time step, enabling the deep learning algorithm to extract high-order features from the data, thereby improving the accuracy and robustness of RUL prediction. Specifically, the following steps are involved: S1031: Based on the time window input strategy, the sensor monitoring data is segmented through the sliding window mechanism to construct time window samples. Figure 2 As shown, n Indicates the number of sensors, S represents the sliding step of the sliding time window, T Indicates the time window length, N Represents the total length of sensor monitoring data. The time window generates samples as ,in, Indicates the first moment in the sliding time window, and the sample label is the last moment in the time window The corresponding true RUL value. To ensure that each sample in the test set has a complete time window, the window length T Must meet ,in, Indicates the first q The life cycle of an engine, , Indicates the total number of engines in the test set.

[0032] S1032: Using the time window samples as input samples of the Mamba-GRU parallel network.

[0033] Step 104: Construct a Mamba-GRU parallel network, which includes a linear labeling layer, a Mamba module, a GRU module, a feature fusion module, and a fully connected layer.

[0034] like Figure 3 As shown in the figure, the Mamba-GRU parallel network connects the Mamba module and the GRU module in parallel, comprehensively utilizing the advantages of the Mamba module and the GRU module to achieve complementary capture of time series features, so that the model can more comprehensively learn the performance degradation features in the time series, and ultimately improve the performance of the model.

[0035] In addition, a feature fusion module was designed, using a scaled dot-product attention mechanism to calculate the correlation between features, resulting in two feature scores. These scores are then normalized using a softmax function to obtain a dynamic weight distribution. This dynamic weight distribution enables adaptive fusion of multi-scale degraded features. Specifically, the Mamba-GRU parallel network consists of a linear labeling layer, a Mamba module, a GRU module, a feature fusion module, and a fully connected layer. Only the Mamba and GRU modules are connected in parallel; all others are connected sequentially. Below, we introduce each module one by one: Linear Marker Layer The linear tokenization layer processes the input time series data through a linear layer, such as Figure 4 As shown, the normalized time series format is performed by a single linear layer in Equation (4), (4) in, express t The output of the linear labeling layer at each moment, express t The input of the linear labeling layer at each moment, Batch represents batch operation, and represents the parameters of the linear marker layer, L represents the length of the input sequence, d represents the feature dimension, R represents the set of real numbers.

[0036] Mamba Module The Mamba module is based on SSM. By integrating the advantages of RNN and Convolutional Neural Network (CNN), it uses a hybrid operation of recursion and convolution to reduce the computational complexity to a linear level and significantly improve the efficiency of long sequence processing. The state space model transforms a one-dimensional function or sequence (represented as x ( t )∈R) is mapped to the hidden state , and finally output .in, m Indicates the size of the state dimension, L is the sequence length, the process can be expressed as linear ordinary differential equations (5) and (6): (5) (6) in, Represents the state matrix , and represents the projection parameters; x ( t )for t The output of the linear labeling layer at each moment, Indicates the hidden state of the current input in the Mamba module, y ( t ) represents the output of the Mamba module, Indicates the hidden state of the previous moment in the Mamba module, R represents the set of real numbers, m Represents the state dimension, L Indicates the length of the input sequence.

[0037] However, state-space models do not use discrete sequences, but take continuous sequences as input and predict output sequences. Mamba discretizes continuous systems to make them more suitable for deep learning scenarios. Specifically, a time scale parameter is introduced and a fixed discretization rule is used to discretize the continuous parameters. A and B Transformed into discrete parameters and The commonly used discretization rule is zero-order hold, which is defined as follows: (7) (8) in, represents the time scale parameter, I represents the identity matrix, A 、 B After discretization, the SSM-based model can be calculated in two ways: linear recursion or global convolution, defined as the following equation: (9) (10) (11) (12) in, L and Represents the input sequence x and the length of the structured convolution kernel.

[0038] The Mamba module is composed of the basic blocks of most SSM architectures such as H3 and the Gated MLP commonly found in modern neural networks such as Transformer. Figure 5 shown.

[0039] GRU module The GRU module is a neural network derived from RNN. Compared to traditional RNNs, the GRU module not only extracts time series information but also effectively solves the vanishing and exploding gradient problems found in traditional RNNs. The core concept of the GRU module is the introduction of a gating mechanism, which consists of two gating units: an update gate and a reset gate. These two gating units control the flow of information within the network, allowing the network to selectively forget or update certain information. This allows the GRU module to better handle temporal dependencies and reduces the number of parameters that require training. The calculation formula for the GRU module is as follows: (13) (14) (15) (16) in, for t The reset gate of time, is the sigmoid activation function, For the GRU module t -1 moment of hidden state, Hidden The weight matrix, x ( t )for t The output of the linear labeling layer at each moment, Input to reset gate x ( t ), for t The door of renewal at all times, Input to the update gate x ( t ), To update the hidden state in the door The weight matrix, For the GRU module t The candidate hidden state at time t, tanh(·) is the tanh activation function, ☉ is the Hadamard product, To reset the weight matrix of the hidden state after gate processing, Hidden The weight matrix, For the GRU module t The output of the moment. The structure of the GRU module is as follows Figure 6 shown.

[0040] Feature fusion module The feature fusion module uses a dynamic feature fusion method based on the scaled dot product attention mechanism. Specifically, it includes the following steps: Step a: First, the long-term degradation trend extracted by the Mamba module is concatenated with the short-term dynamic features captured by the GRU module to obtain the concatenated features. Step b: The concatenated features are then calculated using the scaled dot product attention mechanism to obtain two feature scores: the original attention score is generated by calculating the dot product of the query vector (Query) and the key vector (Key), and divided by the scaling compensation factor of the feature dimension Achieve temperature regulation and effectively suppress the gradient diffusion problem in high-dimensional space. m is the state dimension; Step c: Finally, the two feature scores are normalized using the softmax function to obtain a dynamic weight distribution. The two branch outputs are then weighted and summed according to the weights to achieve dynamic fusion and obtain the fused features. This attention mechanism allows the model to flexibly adjust its dependence on the two branches based on the input data, enhancing the model's expressiveness and adaptability.

[0041] Fully connected layer The fully connected layer implements degraded feature mapping and RUL prediction, and regression prediction is performed through three fully connected layers. First, the input features are mapped through the ReLU activation function; then, they are processed by the second layer of Dropout regularization; finally, the RUL prediction value is output.

[0042] Step 105: Training the Mamba-GRU parallel network based on the time window samples to obtain a lifespan prediction model.

[0043] Specifically, the above training process includes the following steps: S1051: Divide the input features into a training set and a test set.

[0044] S1052: Input the features and RUL in the training set into the Mamba-GRU parallel network to train the Mamba-GRU parallel network.

[0045] S1053: Input the features in the test set into the trained Mamba-GRU parallel network to determine the RUL prediction value.

[0046] S1054: Compare the RUL prediction value with the actual RUL to determine a life comparison result.

[0047] S1055: Selecting a performance evaluation index based on the lifespan comparison result to evaluate the prediction accuracy of the trained Mamba-GRU parallel network.

[0048] The root mean square error (RMSE) and score function were used to evaluate the model prediction performance.

[0049] S1056: According to the prediction accuracy, adjust the trained Mamba-GRU parallel network, and use the adjusted Mamba-GRU parallel network as a RUL prediction model.

[0050] Based on the results of error distribution and robustness analysis, the performance of the Mamba-GRU parallel network in handling different working conditions and complex degradation modes is summarized, and the model design is further optimized.

[0051] Step 106: Predict the RUL of the industrial equipment based on the performance prediction model.

[0052] The Mamba-GRU parallel network constructed in this application was numerically tested on the publicly available C-MAPSS (Commercial Modular Aero-Propulsion System Simulation) engine dataset provided by NASA. The configuration of model parameters directly determines the model's learning ability, generalization capability, and ability to capture degradation patterns in time series data. The training process in this application systematically optimizes the model's hyperparameters using a grid search method. Key hyperparameters such as window size, batch size, dropout rate, regularization coefficient, and learning rate were systematically evaluated through different combinations of training and validation within a predefined parameter range. The optimal configuration was selected based on the model's accuracy in RUL prediction. This process effectively balances training efficiency and generalization performance, ensuring the model's accuracy and stability in identifying degradation patterns in time series data. Table 1 shows the selected hyperparameters and their ranges, using the FD001 sub-dataset as an example. Bold values ​​indicate the best-performing configuration. These results demonstrate that optimized parameter settings significantly contribute to improved model performance.

[0053] Table 1

[0054] The size of the time window determines the length of historical information that the model can capture. A larger window can provide more historical context, enabling the model to better identify long-term trends. Experimental results show that when the time window is 25, the model can produce the best performance. The batch size mainly affects the training efficiency and the stability of parameter updates. Experiments show that the optimal batch size is 256. Dropout is a regularization technique that reduces overfitting by randomly dropping neurons during training. Experiments show that the model performs best at a Dropout rate of 0.3, which is consistent with the complexity of the model and the characteristics of the dataset. In addition, the learning rate also has a great influence on the stability of training. Experiments show that a learning rate of 0.005 often leads to convergence problems, while a learning rate of 0.001 can achieve the best prediction performance. The optimal parameters of the proposed Mamba-GRU parallel network are finally set to: time window length T The training batch size is 25, the dropout rate is 0.3, the learning rate is 0.001, and the Adam optimizer and the mean square error (MSE) loss function are used.

[0055] The test set consists of four sub-datasets, with the number of engines being 100, 259, 100, and 248 respectively. The FD001 and FD003 sub-datasets were selected for the experiment. The RUL prediction results of the two sub-datasets are shown in the figure below. Figure 7 and Figure 8 As shown in the figure, when the true RUL is small, the predicted RUL is very close to the true value with minimal error. Even if the true RUL is large, the predicted value remains highly consistent with the true value.

[0056] In order to simulate the RUL increment prediction during the actual operation of the aircraft, this experiment not only performs the overall RUL prediction for all engines in the test set, but also performs the time series RUL increment prediction for individual engines in different sub-datasets. The prediction results are shown in the figure. Figure 9 As shown. Among them, Figure 9 Part (a) is the RUL prediction diagram for engine No. 24 in the FD001 test set; Figure 9 Part (b) is the RUL prediction diagram for engine No. 76 in the FD001 test set; Figure 9 Part (c) is the RUL prediction diagram for engine No. 39 in the FD003 test set; Figure 9 Part (d) is a schematic diagram of the RUL prediction of engine No. 99 in the FD003 test set.

[0057] As the engine's life cycle approaches its end, prediction accuracy improves significantly. This is because degradation characteristics become more pronounced as the engine approaches failure, and the degradation trend exhibits stronger linearity, which enables the prediction model to learn and capture these characteristics more effectively, making the predicted RUL curve closer to the true linear degradation curve.

[0058] The comparative experimental results of the Mamba-GRU parallel network proposed in this application and the existing deep learning model are shown in Table 2.

[0059] Table 2

[0060] As can be seen from Table 2, on the FD001 sub-dataset, the RMSE of the Mamba-GRU parallel network of this application is second only to CNN-Transformer, and the Score is much better than other methods. On the FD003 sub-dataset, the RMSE and Score of the Mamba-GRU parallel network of this application are much better than the existing methods. Compared with the five methods listed in Table 2, the RMSE is reduced by 18.15%, 16.39%, 14.04%, 12.29% and 7.91% respectively, and the Score is reduced by 49.3%, 46.6%, 44.0%, 42.1% and 70.2% respectively. The experimental results show that the network constructed in this paper has good performance on the C-MAPSS dataset, which further demonstrates the superiority of the Mamba-GRU parallel network proposed in this application in the RUL prediction of industrial equipment.

[0061] Based on the same inventive concept, the embodiments of the present application also provide an intelligent prediction device based on a Mamba deep learning network for implementing the intelligent prediction method based on a Mamba deep learning network mentioned above. The implementation solution provided by the device is similar to the implementation solution described in the above method. Therefore, the specific limitations of one or more embodiments of the intelligent prediction device based on a Mamba deep learning network provided below can be found in the above limitations on the industrial equipment RUL prediction method, and will not be repeated here.

[0062] In an exemplary embodiment, an intelligent prediction device based on a Mamba deep learning network is provided, comprising: Data acquisition module, used to obtain industrial equipment sensor monitoring data; A preprocessing module, used for preprocessing the industrial equipment sensor monitoring data; A time window sample generation module is used to generate samples from the preprocessed sensor monitoring data based on a sliding time window strategy to obtain time window samples; A Mamba-GRU parallel network construction module is used to construct a Mamba-GRU parallel network, wherein the Mamba-GRU parallel network includes: a linear labeling layer, a Mamba module, a GRU module, a feature fusion module, and a fully connected layer; A training module, configured to train the Mamba-GRU parallel network based on the time window samples to obtain a performance prediction model; The performance prediction module is used to predict the RUL or health status of industrial equipment based on the performance prediction model.

[0063] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0064] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. An intelligent prediction method based on Mamba deep learning network, characterized in that: The intelligent prediction method based on the Mamba deep learning network includes: Obtaining industrial equipment sensor monitoring data; Preprocessing the industrial equipment sensor monitoring data; Based on the sliding time window strategy, samples are generated from the pre-processed sensor monitoring data to obtain time window samples; Constructing a Mamba-GRU parallel network, wherein the Mamba-GRU parallel network includes: a linear labeling layer, a Mamba module, a GRU module, a feature fusion module, and a fully connected layer; Training the Mamba-GRU parallel network based on the time window samples to obtain a performance prediction model; The remaining service life or health status of the industrial equipment is predicted based on the performance prediction model.

2. The intelligent prediction method based on the Mamba deep learning network according to claim 1 is characterized in that: Preprocessing the industrial equipment sensor monitoring data specifically includes the following steps: Removing abnormal data from the industrial equipment sensor monitoring data; Perform wavelet transform to reduce noise on the sensor monitoring data after removing abnormal data; The sensor monitoring data after wavelet transform denoising is normalized.

3. The intelligent prediction method based on the Mamba deep learning network according to claim 1 is characterized in that: The linear marking layer specifically adopts the following formula: ； in, express t The output of the linear labeling layer at each moment, express t The input of the linear labeling layer at each moment, Batch represents batch operation, and represents the parameters of the linear marker layer, L represents the length of the input sequence, d represents the feature dimension, R represents the set of real numbers.

4. The intelligent prediction method based on the Mamba deep learning network according to claim 1 is characterized in that: The Mamba module specifically uses the following formula: ； ； in, represents the state matrix; and represents the projection parameters; x ( t )for t The output of the linear labeling layer at each moment, Indicates the hidden state of the current input in the Mamba module, y ( t ) represents the output of the Mamba module, Indicates the hidden state of the previous moment in the Mamba module, R represents the set of real numbers, m Represents the state dimension, L Indicates the length of the input sequence.

5. The intelligent prediction method based on the Mamba deep learning network according to claim 1 is characterized in that: The GRU module specifically adopts the following formula: ； ； ； ； in, for t The reset gate of time, is the sigmoid activation function, For the GRU module t -1 moment of hidden state, Hidden The weight matrix, x ( t )for t The output of the linear labeling layer at each moment, Input to reset gate x ( t ), for t The door of renewal at all times, Input to the update gate x ( t ), To update the hidden state in the door The weight matrix, For the GRU module t The candidate hidden state at time t, tanh(·) is the tanh activation function, ☉ is the Hadamard product, To reset the weight matrix of the hidden state after gate processing, Hidden The weight matrix, For the GRU module t The output of the moment, that is, the output of the GRU module t The hidden state of the moment.

6. The intelligent prediction method based on the Mamba deep learning network according to claim 1 is characterized in that: The data processing process in the feature fusion module is as follows: The long-term degradation trend extracted by the Mamba module is spliced ​​with the short-term dynamic features captured by the GRU module to obtain the spliced ​​features; The concatenated features are calculated using a scaled dot product attention mechanism to obtain two feature scores; Normalize the two feature scores using the softmax function to obtain a dynamic weight distribution; The long-term degradation trend and the short-term dynamic feature are weighted and summed according to the dynamic weight distribution to achieve dynamic fusion and obtain a fusion feature.

7. The intelligent prediction method based on the Mamba deep learning network according to claim 2 is characterized in that: The wavelet transform denoising of the sensor monitoring data after removing abnormal data is specifically performed using the following formula: ； ； in, j Indicates scale, k Indicates translation, t Indicates time, f ( t ) is the original signal, c j,k are the coefficients obtained after discrete wavelet transform, is the discrete wavelet basis function, is the mother wavelet function, is the reconstructed signal.

8. An intelligent prediction device based on the Mamba deep learning network, characterized in that: The intelligent prediction device based on the Mamba deep learning network includes: Data acquisition module, used to obtain industrial equipment sensor monitoring data; A preprocessing module, used for preprocessing the industrial equipment sensor monitoring data; A time window sample generation module is used to generate samples from the preprocessed sensor monitoring data based on a sliding time window strategy to obtain time window samples; A Mamba-GRU parallel network construction module is used to construct a Mamba-GRU parallel network, wherein the Mamba-GRU parallel network includes: a linear labeling layer, a Mamba module, a GRU module, a feature fusion module, and a fully connected layer; A training module, configured to train the Mamba-GRU parallel network based on the time window samples to obtain a performance prediction model; The performance prediction module is used to predict the RUL or health status of industrial equipment based on the performance prediction model.

9. The intelligent prediction device based on the Mamba deep learning network according to claim 8, characterized in that: Preprocessing the industrial equipment sensor monitoring data specifically includes the following steps: Remove abnormal data from sensor monitoring data; Perform wavelet transform to reduce noise on the sensor monitoring data after removing abnormal data; The sensor monitoring data after wavelet transform denoising is normalized.

10. The intelligent prediction device based on the Mamba deep learning network according to claim 8, characterized in that: The linear marking layer specifically adopts the following formula: ； in, express t The output of the linear labeling layer at each moment, express t The input of the linear labeling layer at each moment, Batch represents batch operation, and represents the parameters of the linear marker layer, L represents the length of the input sequence, d represents the feature dimension, R represents the set of real numbers.

Images