Parameter prediction method and device based on deep learning hybrid network

Abstract

The embodiment of the invention provides a parameter prediction method based on a deep learning hybrid network, and the method comprises the steps: collecting original data in the operation process of a gun barrel, and the original data at least comprises the shooting number and the corresponding abrasion loss; preprocessing the original data to obtain training data; constructing an encoder and decoder model, and training the encoder and decoder model by using the training data and real data to obtain a trained model; deploying the trained model into a gun barrel operation system, collecting and preprocessing gun barrel operation data in real time to obtain measurement data, processing the test data by using the trained model, and outputting a gun barrel abrasion loss estimation result in real time after performing inverse transformation operation on an output predicted value. On-line, real-time and accurate estimation of the abrasion loss of the gun barrel is realized, and the prediction precision of the abrasion loss of the gun barrel is improved.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the technical field of industrial equipment status monitoring and related technical fields, and specifically, to a parameter prediction method and device based on a deep learning hybrid network. Background Art

[0002] With the continuous development of artillery technology, the barrel, as the core component of the gun, is subjected to the combined effects of high-temperature and high-pressure gas and the squeeze-in of the ammunition belt during the shooting process, which causes the inner bore structure to be continuously ablated and worn, thereby causing the ballistic performance to gradually degrade until the end of the barrel life. Therefore, based on an in-depth study of the ablation and wear laws of the inner bore of the gun barrel, it is of great significance to establish a mathematical model that can accurately reflect its ablation and wear characteristics for quantifying the changes in the ballistic performance of the gun. The researchers proposed a method based on BP neural network and least squares support vector machine to predict the ablation and wear of the inner bore of the barrel. In addition, existing studies have predicted the wear of the gun barrel by establishing a gray linear model and an optimized Verhulst model.

[0003] However, the existing ablation wear prediction methods usually rely on the original data obtained under equal interval shooting. In practical applications, the collected data often have the characteristics of "small sample", "poor information" and "non-equal interval", which seriously restricts the effectiveness and applicability of traditional prediction methods. The ablation wear of the gun barrel is a key factor affecting its service life, which will lead to a series of adverse effects such as reduced muzzle velocity of projectiles, shortened shooting distance, and reduced first shot hit rate and first group coverage.

[0004] Therefore, there is an urgent need for a parameter prediction method based on deep learning hybrid network to solve the above problems. Summary of the invention

[0005] The embodiments described in this article provide a parameter prediction method and device based on a deep learning hybrid network to solve the problems existing in the prior art.

[0006] According to a first aspect of the present disclosure, a parameter prediction method based on a deep learning hybrid network is provided, comprising:

[0007] Collecting raw data during the operation of the gun barrel, the raw data at least including: the number of shots fired and the corresponding amount of wear;

[0008] Preprocessing the original data to obtain training data;

[0009] Constructing an encoder-decoder model, and training the encoder-decoder model using the training data and real data to obtain a trained model;

[0010] Deploy the trained model to the gun barrel operation system, collect gun barrel operation data in real time and perform preprocessing to obtain measurement data, process the measurement data using the trained model, and after performing an inverse transformation operation on the output prediction value, output the estimated result of the gun barrel wear amount in real time.

[0011] In some embodiments of the present disclosure, during the model training process, an adaptive moment estimation optimizer is adopted, where:

[0012] The adaptive moment optimizer designs independent adaptive learning rates for different parameters by calculating the first-order moment estimation and second-order moment estimation of the gradient;

[0013] The parameters of the adaptive moment optimizer include the learning rate, the exponential decay rate of the first-order moment estimation, the exponential decay rate of the second-order moment estimation, and a minimum value to prevent division by zero.

[0014] In some embodiments of the present disclosure, during the model training process, the mean squared error prediction loss and the scale-invariant mean squared error reconstruction loss are adopted as the loss functions, where:

[0015] The mean squared error prediction loss is used to optimize the prediction accuracy of the gun barrel wear amount;

[0016] The scale-invariant mean squared error reconstruction loss is used to optimize the reconstruction accuracy of the input signal;

[0017] The total optimization objective is the weighted sum of the mean squared error prediction loss and the scale-invariant mean squared error reconstruction loss.

[0018] In some embodiments of the present disclosure, the encoder-decoder model includes: a convolutional neural network, a bidirectional gated recurrent unit, and a denoising autoencoder, where:

[0019] The convolutional neural network is used to extract features from the input data;

[0020] The bidirectional gated recurrent unit is used to capture the temporal dependence in the time series data;

[0021] The denoising autoencoder is used to handle data sparsity and enhance the generalization ability of the model;

[0022] The fully connected layer of the model is used to integrate the features extracted from the convolutional neural network, the bidirectional gated recurrent unit, and the denoising autoencoder, and output the final predicted result of the gun barrel wear amount.

[0023] In some embodiments of the present disclosure, the convolutional neural network includes multiple convolutional layers and max pooling layers, where:

[0024] The convolutional layer extracts features through an activation function;

[0025] The max pooling layer is used to downsample the features to further extract key features;

[0026] The bidirectional gated recurrent unit includes a forward propagation gated recurrent unit and a backward propagation gated recurrent unit, where:

[0027] The forward propagation gated recurrent unit is used to process the forward time dependencies in the time series data;

[0028] The backward propagation gated recurrent unit is used to process the backward time dependencies in the time series data;

[0029] The outputs of the forward and backward propagation gated recurrent units are merged through a concatenation operation to enhance the feature representation of the time series data;

[0030] The denoising autoencoder includes an encoder and a decoder, where:

[0031] The encoder consists of two one-dimensional convolutional neural network layers and a bidirectional gated recurrent unit layer, and is used to extract low-dimensional features from the input data;

[0032] The decoder includes a gated recurrent unit layer and two one-dimensional transposed convolutional layers, and is used to reconstruct the input data;

[0033] The denoising autoencoder injects Gaussian noise into the input data during the training process to improve the robustness of the model.

[0034] In some embodiments of the present disclosure, the gated recurrent unit includes an update gate and a reset gate, where:

[0035] The update gate is used to control the information transfer between the previous hidden state and the current candidate hidden state;

[0036] The reset gate is used to determine the degree of forgetting of the previous hidden state;

[0037] The gated recurrent unit alleviates the vanishing gradient problem and improves the computational efficiency of the model through the collaborative action of the update gate and the reset gate.

[0038] In some embodiments of the present disclosure, the inverse transformation operation on the output prediction value specifically includes: converting the prediction value back to the original dimension;

[0039] Wherein, the inverse transformation operation includes the following steps:

[0040] Determine the range of the model output value and the range of the target quantity;

[0041] Design an inverse transformation function based on these two ranges to map the model output values to the range of the target quantity;

[0042] Use the inverse transformation function to transform the predicted values output by the model to obtain the final estimated value of the gun barrel wear.

[0043] According to the second aspect of the present disclosure, there is provided a parameter prediction device based on a deep learning hybrid network, including:

[0044] An acquisition module for acquiring raw data during the operation of the gun barrel, where the raw data at least includes: the number of firing rounds and the corresponding wear amount;

[0045] A preprocessing module for preprocessing the raw data to obtain training data;

[0046] A training module for constructing an encoder-decoder model, and training the encoder-decoder model using the training data and real data to obtain a trained model;

[0047] A real-time estimation module for deploying the trained model to the gun barrel operation system, collecting and preprocessing gun barrel operation data in real time to obtain measurement data, processing the test data using the trained model, and performing an inverse transformation operation on the output predicted values, and then real-time outputting the estimated result of the gun barrel wear amount.

[0048] In some embodiments of the present disclosure, the training module, during the model training process, adopts an adaptive moment estimation optimizer, where:

[0049] The adaptive moment optimizer designs independent adaptive learning rates for different parameters by calculating the first-order moment estimation and second-order moment estimation of the gradient;

[0050] The parameters of the adaptive moment optimizer include the learning rate, the exponential decay rate of the first-order moment estimation, the exponential decay rate of the second-order moment estimation, and a minimum value to prevent division by zero.

[0051] In some embodiments of the present disclosure, the training module, during the model training process, adopts the mean square error prediction loss and the scale-invariant mean square error reconstruction loss as the loss function, where:

[0052] The mean square error prediction loss is used to optimize the prediction accuracy of the gun barrel wear amount;

[0053] The scale-invariant mean square error reconstruction loss is used to optimize the reconstruction accuracy of the input signal;

[0054] The total optimization objective is the weighted sum of the mean square error prediction loss and the scale-invariant mean square error reconstruction loss.

[0055] In some embodiments of the present disclosure, the encoder-decoder model includes: a convolutional neural network, a bidirectional gated recurrent unit, and a denoising autoencoder, where:

[0056] The convolutional neural network is used to extract features from the input data;

[0057] The bidirectional gated recurrent unit is used to capture the temporal dependencies in the time series data;

[0058] The denoising autoencoder is used to handle data sparsity and enhance the generalization ability of the model;

[0059] The fully connected layer of the model is used to integrate the features extracted from the convolutional neural network, the bidirectional gated recurrent unit, and the denoising autoencoder, and output the final prediction result of the gun barrel wear amount.

[0060] In some embodiments of the present disclosure, the convolutional neural network includes a plurality of convolutional layers and max pooling layers, where:

[0061] The convolutional layer extracts features through an activation function;

[0062] The max pooling layer is used to downsample the features to further extract key features;

[0063] The bidirectional gated recurrent unit includes a forward propagation gated recurrent unit and a backward propagation gated recurrent unit, where:

[0064] The forward propagation gated recurrent unit is used to handle the forward temporal dependencies in the time series data;

[0065] The backward propagation gated recurrent unit is used to handle the backward temporal dependencies in the time series data;

[0066] The outputs of the forward and backward propagation gated recurrent units are merged through a concatenation operation to enhance the feature representation of the time series data;

[0067] The denoising autoencoder includes an encoder and a decoder, where:

[0068] The encoder consists of two one-dimensional convolutional neural network layers and a bidirectional gated recurrent unit layer, and is used to extract low-dimensional features from the input data;

[0069] The decoder includes a gated recurrent unit layer and two one-dimensional transposed convolutional layers, and is used to reconstruct the input data;

[0070] The denoising autoencoder injects Gaussian noise into the input data during the training process to improve the robustness of the model.

[0071] In some embodiments of the present disclosure, the gated recurrent unit includes an update gate and a reset gate, where:

[0072] The update gate is used to control the information transfer between the previous hidden state and the current candidate hidden state;

[0073] The reset gate is used to determine the degree of forgetting of the previous hidden state;

[0074] Through the cooperative action of the update gate and the reset gate, the gated recurrent unit alleviates the vanishing gradient problem and improves the computational efficiency of the model.

[0075] In some embodiments of the present disclosure, the real-time estimation module performs an inverse transformation operation on the output prediction value, specifically including: converting the prediction value back to the original dimension;

[0076] Among them, the inverse transformation operation includes the following steps:

[0077] Determine the range of the model output value and the range of the target quantity;

[0078] Design an inverse transformation function according to these two ranges to map the model output value to the range of the target quantity;

[0079] Use the inverse transformation function to transform the prediction value output by the model to obtain the final estimated value of the gun barrel wear amount.

[0080] According to the third aspect of the present disclosure, there is provided a computer device, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps of the method in any one of the above embodiments are implemented.

[0081] According to the fourth aspect of the present disclosure, there is provided a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method in any one of the above embodiments are implemented.

[0082] The parameter prediction method and device based on a deep learning hybrid network provided by the embodiments of the present disclosure first collect the original data during the operation of the gun barrel. The original data at least includes: the number of shootings and the corresponding wear amount. Then, the original data is preprocessed to obtain training data. Next, an encoder-decoder model is constructed, and the encoder-decoder model is trained using the training data and real data to obtain a trained model. Finally, the trained model is deployed into the gun barrel operation system, the gun barrel operation data is collected in real time and preprocessed to obtain measurement data, the trained model is used to process the test data, and after performing an inverse transformation operation on the output prediction value, the gun barrel wear amount estimation result is output in real time. The online, real-time, and accurate estimation of the gun barrel wear amount is realized. Based on the structure of the denoising autoencoder, combining the stacked convolutional neural network with the bidirectional gated recurrent unit within the encoder-decoder framework can effectively capture and represent the complex non-linear features in the degradation data obtained from the measurement source, and enhance the generalization ability of the prediction model for different data sets. At the same time, by incorporating the reconstruction loss into the total loss function, the accuracy and generalization ability of the prediction model are further improved, and the prediction accuracy of the gun barrel wear amount is increased.

[0083] The above description is only an overview of the technical solutions of the embodiments of the present application. In order to be able to understand the technical means of the embodiments of the present application more clearly, it can be implemented according to the content of the description. And in order to make the above and other purposes, features, and advantages of the embodiments of the present application more obvious and understandable, the specific embodiments of the present application are specifically given below. Brief Description of the Drawings

[0084] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments will be briefly described below. It should be understood that the following described drawings only relate to some embodiments of the present disclosure and do not limit the present disclosure, where:

[0085] Figure 1 is a schematic flowchart of a parameter prediction method based on a deep learning hybrid network provided by the embodiments of the present disclosure;

[0086] Figure 2 is a schematic framework diagram of a GRU unit provided by the embodiments of the present disclosure;

[0087] Figure 3 is a schematic architecture diagram of an encoder-decoder network of a DAE provided by the embodiments of the present disclosure;

[0088] Figure 4 is a schematic structural diagram of a parameter prediction device based on a deep learning hybrid network provided by the embodiments of the present disclosure;

[0089] Figure 5It is a schematic structural diagram of a computer device provided by an embodiment of the present disclosure.

[0090] In the drawings, reference numerals with the same last two digits correspond to the same elements. It should be noted that the elements in the drawings are schematic and not drawn to scale. Detailed implementation manners

[0091] In order to make the objectives, technical solutions, and advantages of the embodiments of the present disclosure clearer, the technical solutions of the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some but not all of the embodiments of the present disclosure. All other embodiments obtained by those skilled in the art based on the described embodiments of the present disclosure without creative efforts also belong to the scope of protection of the present disclosure.

[0092] Reference to "embodiment" herein means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application. The phrase "embodiment" appearing in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0093] The term "and / or" herein is merely a description of an association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can represent: the existence of A, the simultaneous existence of A and B, and the existence of B. In addition, the character " / " herein generally represents an "or" relationship between the associated objects before and after.

[0094] In addition, in all embodiments of the present disclosure, terms such as "first" and "second" are only used to distinguish one component (or a part of the component) from another component (or another part of the component).

[0095] In the description of the present application, unless otherwise specified, "a plurality of" means two or more (including two). Similarly, "a plurality of groups" means two or more groups (including two groups).

[0096] Glossary:

[0097] Barrel wear amount: It refers to the degree of ablation and wear of the inner bore structure of the gun barrel due to the action of high-temperature and high-pressure gas and the squeezing of the cartridge belt during multiple shootings.

[0098] Convolutional neural network (CNN): An algorithm of deep learning that can automatically extract features of input data through operations such as convolutional layers and pooling layers.

[0099] Bidirectional Gated Recurrent Unit (BiGRU): An improved Recurrent Neural Network (RNN) structure that can better capture the temporal dependencies in time series data through bidirectional propagation.

[0100] Denoising Autoencoder (DAE): An autoencoder structure that enhances the generalization ability of the model by introducing noise and attempting to reconstruct the original signal.

[0101] Stacking: Stacking multiple neural network layers or models together to build a deeper model, thereby improving the performance of the model.

[0102] To enable those skilled in the art to better understand the solution of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0103] Based on the problems existing in the prior art, Figure 1 is a schematic flowchart of a parameter prediction method based on a deep learning hybrid network provided by an embodiment of the present disclosure. As Figure 1 shown, the specific process of the parameter prediction method based on the deep learning hybrid network includes:

[0104] S110. Collect the original data during the operation of the gun barrel, that is, the gun barrel wear amount dataset, and the gun barrel wear amount dataset includes at least: the number of firing rounds and the corresponding wear amount.

[0105] During the operation of the gun barrel, wear is an inevitable phenomenon, which will directly affect the performance and life of the gun barrel. The gun barrel will be strongly impacted by high temperature, high pressure and projectile friction during firing, resulting in gradual wear of the inner wall. The accumulation of wear amount will reduce the firing accuracy, increase safety hazards, and even may cause the gun barrel to fail.

[0106] In a possible implementation manner, the gun barrel wear amount dataset includes the number of firing rounds and the corresponding wear amount, where

[0107] the number of firing rounds: used to record the cumulative number of firings of the gun barrel during operation. Each firing will cause a certain degree of wear on the inner wall of the gun barrel, so the number of firing rounds is an important indicator for measuring the wear amount.

[0108] the wear amount: used to record the wear degree of the inner wall of the gun barrel after each firing. The wear amount can be quantified by measuring the change in the inner diameter of the gun barrel, the change in surface roughness, or other relevant parameters.

[0109] Optionally, the specific form of the gun barrel wear amount dataset can be a two-dimensional table, and each row represents a measurement record. For example, it can include the following fields:

[0110]

[0111]

[0112] To ensure the accuracy and reliability of the data, the following methods can be used to collect the data on the wear amount of the gun barrel:

[0113] Inner diameter measurement: Use a high-precision inner diameter measuring instrument (such as a laser measuring instrument) to regularly measure the change in the inner diameter of the gun barrel and record the wear amount.

[0114] Surface roughness detection: Detect the change in the surface roughness of the inner wall of the gun barrel through a surface roughness instrument to indirectly reflect the wear condition.

[0115] Optical detection: Use an endoscope or optical imaging technology to observe the wear marks on the inner wall of the gun barrel and perform image analysis.

[0116] Sensor monitoring: Install wear sensors on the inner wall of the gun barrel to monitor the change in the wear amount in real time.

[0117] S120. Preprocess the original data to obtain training data.

[0118] In the specific implementation process, in the data preprocessing stage, input the original data, that is, the gun barrel wear amount data set, and output the standardized data set after resampling and normalization processing.

[0119] The preprocessing process includes cleaning, resampling, normalization processing, and standardization processing of the original data to convert it into a standardized format.

[0120] S130. Construct an encoder-decoder model, and use the training data and real data to train the encoder-decoder model to obtain a trained model.

[0121] Among them, the output generated by the network is compared with the specified target value of the gun barrel wear amount.

[0122] In the model training stage, input the standardized training data and its corresponding labels (i.e., the real wear amount), and output a trained deep learning model.

[0123] In the specific implementation manner, during the model training process, use an optimizer to optimize the parameters of the encoder-decoder model to minimize the loss function;

[0124] Optionally, adopt the Adaptive Moment Estimation (Adam) optimizer. The Adam optimizer combines the ideas of the momentum method and the RMSprop optimizer, and designs independent adaptive learning rates for different parameters by calculating the first-order moment estimate and the second-order moment estimate of the gradient.

[0125] In specific implementation, first initialize the parameters of the Adam optimizer, including the learning rate, β1 (the exponential decay rate of the first moment estimate), β2 (the exponential decay rate of the second moment estimate), and ε (a very small value to prevent division by zero). Then, in each iteration, calculate the first moment estimate and the second moment estimate based on the current gradient and the previous gradient, and then use these estimates to update the weights and biases of the model.

[0126] In a specific embodiment, as an example, during the model training process, the mean squared error prediction loss Lp and the scale-invariant mean squared error reconstruction loss Lrec are used as the loss functions, introducing two optimization objectives:

[0127] 1. The mean squared error prediction loss L P , which is used to optimize the prediction accuracy of the gun barrel wear amount;

[0128]

[0129] Among them, represents the estimated value of the gun barrel wear amount, y i represents the true value of the gun barrel wear amount, N represents the batch size used during training, and i is a natural number.

[0130] 2. The scale-invariant mean squared error reconstruction loss L rec , which is used to optimize the reconstruction accuracy of the input signal;

[0131]

[0132] Among them, X i represents the original input, represents the reconstructed signal, ||.|| 2 represents the L 2 regularization operation, N represents the batch size used during training, 1 N is a unit vector of length N, and i represents a natural number. The scale-invariant mean squared error has better performance in signal reconstruction. By penalizing the differences between element samples without considering the absolute global scale, the performance of signal reconstruction is improved.

[0133] The total optimization objective is the weighted sum of the mean squared error prediction loss and the scale-invariant mean squared error reconstruction loss. The total optimization objective L total can be expressed as:

[0134] L total = L P + λL rec (3)

[0135] Among them, λ represents the weight for weighing the performance of the estimation result, which is used to balance the prediction performance and the reconstruction performance.

[0136] The embodiments of this application introduce two optimization objectives, namely, prediction loss and reconstruction loss, and their combination methods. By optimizing the total loss, the accuracy and generalization ability of the prediction model can be further improved.

[0137] In the proposed model, a reconstructed input signal is introduced, which is similar to building a "closed-loop feedback". The reconstruction loss is used to optimize the capacity-related features extracted by the CNN-BiGRU network, thereby improving the robustness and accuracy of the prediction of the gun barrel wear amount.

[0138] In a specific implementation manner, as another embodiment, the encoder-decoder model includes components such as a convolutional neural network (CNN), a bidirectional gated recurrent unit (BiGRU), and a denoising autoencoder (DAE), where:

[0139] The convolutional neural network CNN is used to extract features from the input data;

[0140] The bidirectional gated recurrent unit BiGRU is used to capture the temporal dependencies in the time series data;

[0141] The denoising autoencoder DAE is used to handle data sparsity and enhance the generalization ability of the model;

[0142] The fully connected layer of the model is used to integrate the features extracted from the convolutional neural network CNN, the bidirectional gated recurrent unit BiGRU, and the denoising autoencoder DAE, and output the final prediction result of the gun barrel wear amount;

[0143] Optionally, the convolutional neural network CNN includes multiple convolutional layers and max pooling layers, where:

[0144] The convolutional layer extracts features through an activation function;

[0145] The max pooling layer is used to downsample the features to further extract key features;

[0146] Specifically, the cascades of the convolutional neural network CNN are systematically arranged to extract deeper sequence information and capture temporal dynamic patterns. The initial data samples are enhanced and fused, and finally the best synthesis is formed. For a specified sample x, the features output from the convolutional layer are as follows:

[0147] C i = f(∑x * k i + b i ) (4)

[0148] where i represents a natural number, k iLet \(W_i\) denote the \(i\)-th convolutional kernel, \(b\) denote the bias, \(f()\) denote the LeakyReLU activation function, and \(x\) denote the specified sample. Meanwhile, the max pooling layer is applied for downsampling to further extract features, which can be calculated as

[0149]

[0150] where \(W\) represents the width of the pooling kernel, \(s\) represents the stride of the pooling kernel, \(l\) represents the length of the feature map, and \(n\) represents a natural number.

[0151] GRU (Gated Recurrent Unit) is used to encapsulate the temporal dependencies inherent in time series data, effectively alleviating the vanishing and exploding gradient problems inherent in the RNN design. Figure 2 Figure shows the architecture framework of the GRU unit.

[0152] The output \(h_t\) at time step \(t\) t-1 along with the contemporaneous input \(x_t\) t is fed into the GRU unit, resulting in a new output \(h_{t + 1}\). t . The learning process of the GRU unit is as follows:

[0153] \(z_t\) t = \(\sigma(W_x x_t + U_h h_{t - 1}+b_z)\) (6) z \(x_t\) t \(U_h\) z \(h_{t - 1}\) t-1 \(b_z\) z ) (6)

[0154] \(r_t\) t = \(\sigma(W_x x_t + U_h h_{t - 1}+b_r)\) (7) r \(x_t\) t \(U_h\) r \(h_{t - 1}\) t-1 \(b_r\) r ) (7)

[0155]

[0156] where \(t\) represents the time step, \(z_t\) t and \(r_t\) t represent the outputs of the update and reset gates at the previous time step \(t\), \(h_t\) t-1 represents the output at time step \(t\), \(x_t\) t represents the contemporaneous input at time step \(t\), the function \(\sigma\) represents the sigmoid activation function, \(W_x\) and \(U_h\) represent the weight matrices of the input and recurrent network elements respectively, \(h\) represents the hidden state vector, \(b\) represents the bias term, \(\odot\) represents the Hadamard product, which performs element-wise multiplication.

[0157] Optionally, the bidirectional gated recurrent unit BiGRU includes a forward propagation gated recurrent unit and a backward propagation gated recurrent unit, which consider previous and subsequent information to enhance the output result. The two GRU units are connected to obtain the output of the BiGRU. Among them:

[0158] The forward propagation gated recurrent unit is used to process the forward time dependence in time series data;

[0159] The backward propagation gated recurrent unit is used to process the backward time dependence in time series data;

[0160] The outputs of the forward and backward propagation gated recurrent units are merged through a concatenation operation to enhance the feature representation of time series data;

[0161] Specifically, BiGRU consists of GRU layers for forward and backward propagation;

[0162]

[0163] Among them, t represents the time step, and h t-1 represents the output at time step t, and x t represents the contemporaneous input at time step t, and represent the forward and reverse hidden states respectively, and con() describes the concatenation operation.

[0164] Optionally, the denoising autoencoder DAE includes an encoder and a decoder, Figure 3 describes the architecture of the encoder-decoder network of the DAE. The DAE can extract low-dimensional features from the initial signal, avoiding cumbersome operations. Among them:

[0165] The encoder consists of two one-dimensional convolutional neural network layers CNN and a bidirectional gated recurrent unit BiGRU layer, which is used to extract low-dimensional features from the input data. To achieve robust data representation, Gaussian noise is injected into the original data X;

[0166] The decoder includes a gated recurrent unit layer and two one-dimensional transposed convolution layers. Among them, the GRU layer is before the sequence of two 1D transposed convolution layers. The decoder is used to reconstruct the input data, denoted as The final output derived from the decoder layer appears as data that has been reconstructed to maintain the same dimensional structure as the initial input (i.e., );

[0167] In addition, the features captured from the stacked CNN and BiGRU are fed into a fully connected layer to achieve prediction.

[0168] Based on the structure of the denoising autoencoder (DAE), combining the stacked convolutional neural network with the bidirectional gated recurrent unit within the encoder-decoder framework can effectively capture and represent the complex non-linear features in the degraded data obtained from the measurement source, and enhance the generalization ability of the prediction model for different datasets.

[0169] In this embodiment, the encoder-decoder model is trained by combining components such as CNN, BiGRU, and DAE, using the Adam optimizer and a specific loss function, and finally achieves high-precision prediction of the gun barrel wear amount. A novel hybrid network structure that combines the convolutional neural network (CNN) and the bidirectional gated recurrent unit (BiGRU) with the denoising autoencoder (DAE) is proposed for predicting the gun barrel wear amount. This hybrid network structure can effectively extract potential high-level features from the original dataset, enhance the generalization ability of the prediction model for different datasets, and improve the prediction accuracy and generalization ability. The GRU unit effectively alleviates the vanishing gradient problem through its gating mechanism, and the BiGRU structure further enhances the model's processing ability for time series data. The DAE structure improves the generalization ability and robustness of the model through denoising and feature extraction.

[0170] S140: Deploy the trained model to the gun barrel operation system for real-time monitoring and prediction of the gun barrel wear amount. Collect the gun barrel operation data in real time and perform the same preprocessing operations as in the training stage on it to obtain the standardized measurement data. Use the trained model to process the test data, output the predicted value of the gun barrel wear amount, perform an inverse transformation operation on the predicted value output by the model, and convert it back to the original dimension to obtain the final estimated result of the gun barrel wear amount.

[0171] In a specific embodiment, in the online stage, input the measurement data of the new gun barrel at different firing numbers. These data need to go through the same preprocessing steps as the training data, including operations such as standardization and denoising, to ensure that the data format is consistent with the training data. Input the preprocessed measurement data into the trained model, and the model processes the data through its encoder-decoder structure (including CNN, BiGRU, and DAE components) and outputs the predicted value of the gun barrel wear amount. The predicted value output by the model is a standardized value, and it needs to be further converted to the actual value through an inverse transformation operation. This actual value has the same dimension as the original measurement data.

[0172] In a specific embodiment, the inverse transformation is an important step in the online test part, and its purpose is to convert the standardized predicted value output by the model back to the original dimension so that it has practical physical significance.

[0173] Optionally, the inverse transformation can be achieved through the following steps: First, determine the range of the model output values (the standardized range) and the range of the target quantity (the range of the original measured data). Then, design an inverse transformation function such as a linear transformation or a non-linear transformation based on these two ranges, which can map the model output values to the range of the target quantity. Finally, in the online testing phase, use the inverse transformation function to convert the predicted values output by the model to obtain the final estimated value of the gun barrel wear

[0174] Optionally, the estimated value of the gun barrel wear after inverse transformation will be output to the monitoring interface of the gun barrel operation system in real time for the operator's reference. The system can compare the prediction result with the actual measured value to further optimize the prediction accuracy of the model. For example, when the deviation between the predicted value and the actual value is large, the system can trigger an alarm or automatically adjust the model parameters

[0175] In a specific implementation manner, the inverse transformation operation on the output predicted value specifically includes: converting the output predicted value back to the original dimension

[0176] Among them, the inverse transformation operation includes the following steps

[0177] Determine the range of the model output values and the range of the target quantity

[0178] Design an inverse transformation function based on these two ranges to map the model output values to the range of the target quantity

[0179] Use the inverse transformation function to convert the predicted values output by the model to obtain the final estimated value of the gun barrel wear

[0180] In this step, the trained model is deployed to the gun barrel operation system. By collecting and preprocessing data in real time and combining the inverse transformation operation, high-precision real-time prediction of the gun barrel wear is achieved. As a key step, the inverse transformation ensures that the prediction result has practical physical significance. It is not only applicable to a single gun barrel system but also can be extended to a multi-gun barrel system, and supports adaptive learning and visual analysis, providing comprehensive technical support for gun barrel maintenance

[0181] The parameter prediction method based on the deep learning hybrid network provided by the embodiments of the present disclosure first collects the original data during the operation of the gun barrel. The original data at least includes the number of shootings and the corresponding wear amount. Then, the original data is preprocessed to obtain training data. Next, an encoder-decoder model is constructed, and the encoder-decoder model is trained using the training data and the real data to obtain a trained model. Finally, the trained model is deployed into the gun barrel operation system, the gun barrel operation data is collected in real time and preprocessed to obtain measurement data, the trained model is used to process the test data, and after performing an inverse transformation operation on the output prediction value, the gun barrel wear amount estimation result is output in real time, realizing the online, real-time, and accurate estimation of the gun barrel wear amount.

[0182] Based on the above embodiments, the embodiments of the present disclosure further provide a parameter prediction device based on the deep learning hybrid network, as Figure 4 shown. The parameter prediction device based on the deep learning hybrid network includes:

[0183] A collection module 210, configured to collect the original data during the operation of the gun barrel. The original data at least includes the number of shootings and the corresponding wear amount;

[0184] A preprocessing module 220, configured to preprocess the original data to obtain training data;

[0185] A training module 230, configured to construct an encoder-decoder model, and use the training data and the real data to train the encoder-decoder model to obtain a trained model;

[0186] A real-time estimation module 240, configured to deploy the trained model into the gun barrel operation system, collect the gun barrel operation data in real time and preprocess it to obtain measurement data, use the trained model to process the test data, and after performing an inverse transformation operation on the output prediction value, output the gun barrel wear amount estimation result in real time.

[0187] In a specific implementation manner, during the model training process, an adaptive moment estimation optimizer is adopted, where:

[0188] The adaptive moment optimizer calculates the first-order moment estimation and the second-order moment estimation of the gradient, and designs independent adaptive learning rates for different parameters;

[0189] The parameters of the adaptive moment optimizer include the learning rate, the exponential decay rate of the first-order moment estimation, the exponential decay rate of the second-order moment estimation, and a minimum value ε for preventing division by zero.

[0190] In a specific embodiment, the loss function of the encoder-decoder model includes a mean squared error prediction loss and a scale-invariant mean squared error reconstruction loss, where:

[0191] The mean squared error prediction loss is used to optimize the prediction accuracy of the gun barrel wear amount;

[0192] The scale-invariant mean squared error reconstruction loss is used to optimize the reconstruction accuracy of the input signal;

[0193] The total optimization objective is the weighted sum of the mean squared error prediction loss and the scale-invariant mean squared error reconstruction loss, where the weights are used to balance the prediction performance and the reconstruction performance;

[0194] The scale-invariant mean squared error reconstruction loss improves the signal reconstruction performance by penalizing the differences between element samples without considering the absolute global scale.

[0195] In a specific embodiment, the encoder-decoder model includes: a convolutional neural network, a bidirectional gated recurrent unit, and a denoising autoencoder, where:

[0196] The convolutional neural network is used to extract features from the input data;

[0197] The bidirectional gated recurrent unit is used to capture the temporal dependencies in the time series data;

[0198] The denoising autoencoder is used to handle data sparsity and enhance the generalization ability of the model;

[0199] The fully connected layer of the model is used to integrate the features extracted from the convolutional neural network, the bidirectional gated recurrent unit, and the denoising autoencoder, and output the final prediction result of the gun barrel wear amount;

[0200] The convolutional neural network includes multiple convolutional layers and max pooling layers, where:

[0201] The convolutional layer extracts features through an activation function;

[0202] The max pooling layer is used to downsample the features to further extract key features;

[0203] The bidirectional gated recurrent unit includes a forward propagation gated recurrent unit and a backward propagation gated recurrent unit, where:

[0204] The forward propagation gated recurrent unit is used to handle the forward temporal dependencies in the time series data;

[0205] The backward propagation gated recurrent unit is used to handle the backward temporal dependencies in the time series data;

[0206] The outputs of the forward and backward propagation gated recurrent units are merged through a concatenation operation to enhance the feature representation of time series data;

[0207] The denoising autoencoder includes an encoder and a decoder, where:

[0208] The encoder consists of two one-dimensional convolutional neural network layers and a bidirectional gated recurrent unit layer, and is used to extract low-dimensional features from the input data;

[0209] The decoder includes a gated recurrent unit layer and two one-dimensional transposed convolutional layers, and is used to reconstruct the input data;

[0210] The denoising autoencoder injects Gaussian noise into the input data during the training process to improve the robustness of the model.

[0211] In a specific embodiment, the model constructs a closed-loop feedback mechanism by introducing the reconstructed input signal during the training process to improve the robustness and accuracy of the prediction of the gun barrel wear amount.

[0212] In a specific embodiment, the gated recurrent unit includes an update gate and a reset gate, where:

[0213] The update gate is used to control the information transfer between the hidden state at the previous moment and the current candidate hidden state;

[0214] The reset gate is used to determine the degree of forgetting of the hidden state at the previous moment;

[0215] The gated recurrent unit alleviates the vanishing gradient problem and improves the computational efficiency of the model through the collaborative action of the update gate and the reset gate.

[0216] In a specific embodiment, the inverse transformation operation on the predicted output value specifically includes: converting the predicted value back to the original dimension;

[0217] Among them, the inverse transformation operation includes the following steps:

[0218] Determine the range of the model output value and the range of the target quantity;

[0219] Design an inverse transformation function according to these two ranges to map the model output value to the range of the target quantity;

[0220] Use the inverse transformation function to transform the predicted value output by the model to obtain the final estimated value of the gun barrel wear amount.

[0221] The parameter prediction device based on the deep learning hybrid network provided by the embodiments of the present disclosure realizes the online, real-time, and accurate estimation of the gun barrel wear amount.

[0222] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the descriptions of the method embodiments. The device embodiments described above are only illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of the present invention. Those of ordinary skill in the art can understand and implement it without creative work.

[0223] The embodiments of the present application also provide a computer device. Specifically, please refer to Figure 5 , Figure 5 which is the basic structural block diagram of the computer device in this embodiment.

[0224] The computer device includes a memory 310 and a processor 320 that communicate with each other through a system bus. It should be noted that only the computer device with components 310-320 is shown in the figure, but it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Among them, those skilled in the art of the present technology can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0225] The computer device can be a desktop computer, a notebook, a palm computer, a cloud server and other computing devices. The computer device can interact with users through means such as a keyboard, a mouse, a remote control, a touchpad or a voice control device.

[0226] The memory 310 includes at least one type of readable storage medium, which includes non-volatile memory or volatile memory, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory, etc.), random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disc, etc. RAM can include static RAM or dynamic RAM. In some embodiments, the memory 310 can be an internal storage unit of the computer device, such as the hard disk or memory of the computer device. In other embodiments, the memory 310 can also be an external storage device of the computer device, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device, etc. Of course, the memory 310 can also include both the internal storage unit and the external storage device of the computer device. In this embodiment, the memory 310 is generally used to store the operating system and various application software installed on the computer device, such as the program code of the above method, etc. In addition, the memory 310 can also be used to temporarily store various data that have been output or will be output.

[0227] The processor 320 is generally used to execute the overall operations of the computer device. In this embodiment, the memory 310 is used to store program code or instructions, and the program code includes computer operation instructions. The processor 320 is used to execute the program code or instructions stored in the memory 310 or process data, such as running the program code of the above method.

[0228] In this document, the bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The bus system can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, only a thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0229] Another embodiment of the present application also provides a computer-readable medium, which can be a computer-readable signal medium or a computer-readable storage medium. A processor in the computer reads the computer-readable program code stored in the computer-readable medium, so that the processor can execute the functional actions specified in each step or the combination of steps in the above method; and generate a device that implements the functional actions specified in each block or the combination of blocks in the block diagram.

[0230] The computer-readable medium includes but is not limited to electronic, magnetic, optical, electromagnetic, infrared memories or semiconductor systems, devices or apparatuses, or any suitable combination of the foregoing. The memory is used to store program code or instructions, and the program code includes computer operation instructions. The processor is used to execute the program code or instructions of the above method stored in the memory.

[0231] For the definitions of the memory and the processor, reference can be made to the description of the foregoing computer device embodiments, and details are not described herein again.

[0232] In several embodiments provided by the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of devices or units can be electrical, mechanical, or other forms.

[0233] In each embodiment of the present application, each functional unit or module can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0234] When an integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods of various embodiments of this application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs.

[0235] Unless the context clearly indicates otherwise, the singular forms of the words used in this specification and the appended claims include the plural, and vice versa. Thus, when referring to the singular, the corresponding plural of the term is usually included. Similarly, the terms "comprising" and "including" will be interpreted as inclusive rather than exclusive. Likewise, the term "including" and "or" should be interpreted as inclusive, unless such an interpretation is explicitly prohibited in this specification. Where the term "example" is used in this specification, especially when it is located after a group of terms, the "example" is merely exemplary and illustrative and should not be considered exclusive or extensive.

[0236] Further aspects and scopes of adaptability become apparent from the description provided herein. It should be understood that the various aspects of this application can be implemented alone or in combination with one or more other aspects. It should also be understood that the description herein and the specific embodiments are for illustrative purposes only and are not intended to limit the scope of this application.

[0237] The above has described several embodiments of the present disclosure in detail. However, obviously, those skilled in the art can make various modifications and variations to the embodiments of the present disclosure without departing from the spirit and scope of the present disclosure. The protection scope of the present disclosure is defined by the appended claims.

Claims

1. A parameter prediction method based on a deep learning hybrid network, characterized in that: include: Collecting raw data during the operation of the gun barrel, the raw data at least including: the number of shots fired and the corresponding amount of wear; Preprocessing the original data to obtain training data; Constructing an encoder-decoder model, and training the encoder-decoder model using the training data and real data to obtain a trained model; The trained model is deployed to the gun barrel operation system, the gun barrel operation data is collected in real time and preprocessed to obtain measurement data, the test data is processed using the trained model, and the output prediction value is inversely transformed to output the gun barrel wear estimation result in real time.

2. The method according to claim 1, characterized in that During the model training process, an adaptive moment estimation optimizer is used, where: The adaptive moment optimizer designs independent adaptive learning rates for different parameters by calculating the first-order moment estimate and the second-order moment estimate of the gradient; The parameters of the adaptive moment optimizer include a learning rate, an exponential decay rate of a first-order moment estimate, an exponential decay rate of a second-order moment estimate, and a minimum value for preventing division by zero.

3. The method according to claim 1, characterized in that During the model training process, the mean square error prediction loss and the scale-invariant mean square error reconstruction loss are used as the loss function, where: The mean square error prediction loss is used to optimize the prediction accuracy of barrel wear; The reconstruction loss of the scale-invariant mean square error is used to optimize the reconstruction accuracy of the input signal; The overall optimization objective is a weighted sum of the mean square error prediction loss and the scale-invariant mean square error reconstruction loss.

4. The method according to claim 1, characterized in that The encoder-decoder model includes: a convolutional neural network, a bidirectional gated recurrent unit and a denoising autoencoder, wherein: The convolutional neural network is used to extract features from input data; The bidirectional gated recurrent unit is used to capture the temporal dependency in time series data; The noise autoencoder is used to handle data sparsity and enhance the generalization ability of the model; The fully connected layer of the model is used to integrate the features extracted from the convolutional neural network, the bidirectional gated recursive unit and the denoising autoencoder, and output the final barrel wear prediction result.

5. The method according to claim 4, characterized in that The convolutional neural network includes multiple convolutional layers and maximum pooling layers, wherein: The convolution layer extracts features through an activation function; The maximum pooling layer is used to downsample the features to further extract key features; The bidirectional gated recursive unit includes a forward propagation gated recurrent unit and a backward propagation gated recurrent unit, wherein: The forward propagation gated recurrent unit is used to process the positive time dependency in the time series data; The back-propagation gated recurrent unit is used to process the reverse time dependency in the time series data; The outputs of the forward and backward propagation gated recurrent units are combined through a concatenation operation to enhance the feature representation of the time series data; The denoising autoencoder comprises an encoder and a decoder, wherein: The encoder consists of two one-dimensional convolutional neural network layers and a bidirectional gated recurrent unit layer for extracting low-dimensional features from input data; The decoder includes a gated recurrent unit layer and two one-dimensional deconvolution layers for reconstructing input data; The denoising autoencoder injects Gaussian noise into the input data during training to improve the robustness of the model.

6. The method according to claim 5, characterized in that The gated recursive unit comprises an update gate and a reset gate, wherein: The update gate is used to control the information transfer between the previous hidden state and the current candidate hidden state; The reset gate is used to determine the degree of forgetting of the hidden state at the previous moment; The gated recursive unit alleviates the gradient vanishing problem and improves the computational efficiency of the model through the synergistic effect of the update gate and the reset gate.

7. The method according to claim 1, characterized in that The inverse transformation operation of the outputted predicted value specifically includes: converting the predicted value back to the original dimension; The inverse transformation operation comprises the following steps: Determine the range of model output values ​​and the range of target quantities; According to these two ranges, an inverse transformation function is designed to map the model output value into the range of the target quantity; The predicted value output by the model is transformed using the inverse transformation function to obtain the final estimated value of the barrel wear.

8. A parameter prediction device based on a deep learning hybrid network, characterized in that: include: A collection module, used to collect raw data during the operation of the gun barrel, wherein the raw data at least includes: the number of shots fired and the corresponding amount of wear; A preprocessing module, used for preprocessing the raw data to obtain training data; A training module is used to construct an encoder-decoder model, and train the encoder-decoder model using the training data and real data to obtain a trained model; The real-time estimation module is used to deploy the trained model to the gun barrel operation system, collect the gun barrel operation data in real time and perform preprocessing to obtain measurement data, use the trained model to process the test data, and after inverse transformation operation on the output prediction value, output the gun barrel wear estimation result in real time.

9. A computer device, characterized in that: include: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images