Aero-engine infrared radiation curve prediction method and device and electronic equipment

Abstract

The invention provides an aero-engine infrared radiation curve prediction method and device, and relates to the technical field of aircraft design, and the method comprises the steps: obtaining a training data set through the simulation calculation of the infrared radiation characteristics of a typical aero-engine; sequentially establishing a time domain convolutional network layer, a bidirectional long and short term memory neural network layer, an attention mechanism layer and a full connection layer to generate a prediction model; a variable spiral strategy and a sparrow warning mechanism are combined to improve a salmons group optimization algorithm; optimizing the initial learning rate, the hidden layer node number and the regularization coefficient of the prediction model by using an improved salat swarm optimization algorithm and the training data set, and determining an optimal hyper-parameter to generate a trained prediction model; the trained prediction model is adopted to predict the infrared radiation characteristics of the aero-engine, an infrared radiation characteristic curve is constructed through an interpolation method based on prediction results of different wavebands, the obtained result precision is higher, and the generalization ability is higher.

Description

Technical Field

[0001] This invention relates to the field of aircraft design technology, specifically to a method and device for predicting the infrared radiation curve of an aero-engine, and electronic equipment. Background Technology

[0002] Engine exhaust systems emit a significant portion of infrared radiation. Developing infrared stealth technology for engine exhaust systems to reduce their infrared radiation characteristics is a crucial and pressing issue. With the rapid advancements in computer technology and computational mathematics, a series of numerical methods have been developed for calculating infrared radiation energy during spatial transmission, such as the finite volume method, discrete transfer method, Monte Carlo method, and inverse Monte Carlo method. These methods can obtain infrared radiation intensity through numerical simulation. However, the computation time increases significantly with a large mesh size. Furthermore, adjusting the solver's mesh size or related parameters further extends the computation time, undoubtedly drastically reducing the efficiency of infrared stealth design. Therefore, it is necessary to develop a fast and accurate method for predicting infrared radiation intensity.

[0003] With the continuous development of machine learning technology, data-driven modeling methods have been widely applied in the aerospace field. Machine learning methods can effectively overcome the shortcomings of traditional numerical simulations, which require long processing times, and learn the implicit nonlinear relationship between the prediction target and input variables. Traditional methods typically use neural networks to learn the patterns of infrared radiation intensity for prediction. However, the infrared radiation intensity of aero-engines exhibits significant time-series characteristics, making it difficult for a single neural network to accurately reflect the mapping relationship between input and output. Furthermore, the hyperparameter settings in neural network models have a significant impact on model performance, but in most case studies, these hyperparameters are often set based on experience, leading to unstable model performance and poor feasibility in practical applications. Summary of the Invention

[0004] To address the shortcomings of the prior art, this invention provides a method, apparatus, and electronic device for predicting the infrared radiation curve of an aero-engine.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a method for predicting the infrared radiation curve of an aero-engine, wherein the method includes: A training dataset was obtained by simulating and calculating the infrared radiation characteristics of a typical aero-engine. A prediction model is generated by sequentially establishing a temporal convolutional network layer, a bidirectional long short-term memory neural network layer, an attention mechanism layer, and a fully connected layer. The temporal convolutional network layer is used to extract the time-series information of the input feature sequence. The bidirectional long short-term memory neural network layer is used to extract the input feature sequence from both the preceding and following directions, capturing the dependency information in the bidirectional sequence. The attention mechanism layer is used to assign different weights to each vector in the input feature sequence and highlight key features. An improved sand cat swarm optimization algorithm is proposed by combining a variable spiral strategy with a sparrow warning mechanism. The initial learning rate, number of hidden layer nodes, and regularization coefficient of the prediction model are optimized using an improved sand cat swarm optimization algorithm and the training dataset to determine the optimal hyperparameters and generate the trained prediction model. The trained prediction model is used to predict the infrared radiation characteristics of the aero-engine, and infrared radiation characteristic curves are constructed by interpolation based on the prediction results of different bands.

[0006] Optionally, the step of obtaining a training dataset by simulating the infrared radiation characteristics of a typical aero-engine includes: The infrared radiation characteristics of a typical aero-engine are calculated by simulating using the inverse Monte Carlo method to obtain a training dataset.

[0007] Optionally, the temporal convolutional network layer includes: causal convolution, dilated convolution, and residual modules; the dilated convolution utilizes a spreading factor. Expanding the convolutional field of view, the output features after dilated convolution for: ;in, Given the input feature sequence, For filters, For filter size, As the expansion factor, when At times, dilated convolution and causal convolution have the same structure; The residual module consists of two sets of causal dilated convolutional layers, weight normalization layers, ReLU activation function layers, and Dropout layers, which are connected sequentially from bottom to top to obtain the output features.

[0008] Optionally, when the bidirectional long short-term memory neural network layer processes the input feature sequence, it generates a first hidden sequence in the forward direction and a second hidden sequence in the reverse direction; the first hidden sequence and the second hidden sequence are concatenated as the output sequence. The long short-term memory neural network layer includes an input gate, a forget gate, and an output gate, which control the flow of information to maintain and update memories over a long period of time.

[0009] Optionally, the bidirectional long short-term memory neural network layer generates the output sequence using the following formula: The input feature sequence is: The first hidden sequence is The second hidden sequence is , This is the output sequence of a bidirectional long short-term memory neural network layer.

[0010] Optionally, the implementation formula for the attention mechanism layer is: ;in, The value of the attention probability distribution is calculated from the output sequence of the bidirectional long short-term memory neural network layer. and These are the weighting coefficients. For attention layer output, This is the final output of the attention mechanism layer.

[0011] Optionally, the step of improving the sand cat swarm optimization algorithm by combining a variable spiral strategy and a sparrow vigilance mechanism includes: Initialize the population; Update the optimal solution, worst solution, and individual optimal solution of the population, and calculate the preset relevant parameters; A variable spiral strategy is introduced to update the sand cat's position during the prey-hunting phase; Introduce a sparrow alert mechanism to update the current location of the sand cat; Update the current optimal fitness value; Determine whether the constraints are met. If not, return to the steps of updating the optimal solution, worst solution, and individual optimal solution of the population, and calculating the preset relevant parameters. If the constraints are met, determine that the improvement of the sand cat group optimization algorithm is completed, and output the improved sand cat group optimization algorithm.

[0012] Optionally, the step of optimizing the initial learning rate, number of hidden layer nodes, and regularization coefficient of the prediction model using the improved sand cat swarm optimization algorithm and the training dataset to determine the optimal hyperparameters for generating the trained prediction model includes: The training dataset obtained from the simulation is normalized and divided into a training set and a test set. Establish an objective function model, wherein the objective function is the root mean square error; The improved sand cat swarm optimization algorithm is used to optimize the parameters of the prediction model, including the initial learning rate, the number of hidden layer nodes, and the regularization coefficient; the optimal hyperparameters are selected by judging the value of the fitness function or the maximum number of iterations. Establish prediction models for each infrared band corresponding to the optimal hyperparameter combination, and generate the trained prediction models.

[0013] This invention also provides an infrared radiation curve prediction device for an aero-engine, wherein the device includes: The acquisition module is used to obtain a training dataset by simulating and calculating the infrared radiation characteristics of a typical aero-engine. The model generation module is used to sequentially build a temporal convolutional network layer, a bidirectional long short-term memory neural network layer, an attention mechanism layer, and a fully connected layer to generate a prediction model. The temporal convolutional network layer is used to extract the time-series information of the input feature sequence. The bidirectional long short-term memory neural network layer is used to extract the input feature sequence from both the preceding and following directions, capturing the dependency information in the bidirectional sequence. The attention mechanism layer is used to assign different weights to each vector in the input feature sequence and highlight key features. An improved module is used to improve the sand cat swarm optimization algorithm by combining a variable spiral strategy with a sparrow warning mechanism; The training module is used to optimize the initial learning rate, number of hidden layer nodes, and regularization coefficient of the prediction model using an improved sand cat swarm optimization algorithm and the training dataset, and to determine the optimal hyperparameters to generate the trained prediction model. The prediction module is used to predict the infrared radiation characteristics of the aero-engine using the trained prediction model, and to construct infrared radiation characteristic curves based on the prediction results of different bands through interpolation.

[0014] This invention provides an electronic device, which includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, they implement the steps of any of the above-described methods for predicting the infrared radiation curve of an aero-engine.

[0015] This invention provides a readable storage medium storing a program or instructions, which, when executed by a processor, implement the steps of any of the above-described methods for predicting the infrared radiation curves of an aero-engine.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The infrared radiation curve prediction scheme for aero-engines provided in this embodiment of the invention obtains a training dataset by simulating the infrared radiation characteristics of a typical aero-engine; it then sequentially establishes a temporal convolutional network layer, a bidirectional long short-term memory neural network layer, an attention mechanism layer, and a fully connected layer to generate a prediction model; it improves the sand cat swarm optimization algorithm by combining a variable spiral strategy and a sparrow warning mechanism; it optimizes the initial learning rate, number of hidden layer nodes, and regularization coefficient of the prediction model using the improved sand cat swarm optimization algorithm and the training dataset to determine the optimal hyperparameters and generate the trained prediction model; and it uses the trained prediction model to predict the infrared radiation characteristics of the aero-engine, constructing infrared radiation feature curves based on the prediction results of different bands through interpolation. Through this scheme, on the one hand, the temporal convolutional network layer can effectively extract the time-series information of the input features; on the other hand, the bidirectional long short-term memory neural network layer can extract sequence features from both directions, capturing the dependency information in the bidirectional sequence. Thirdly, the attention mechanism can highlight information at key historical time points and reduce the impact of redundant or useless information on the neural network. Fourthly, the sand cat swarm optimization algorithm, which is improved by combining the variable spiral strategy and the sparrow vigilance mechanism, has stronger search capabilities and faster convergence speed compared to the sand cat swarm optimization algorithm. It can more accurately optimize the hyperparameters of the model, improve the prediction accuracy and efficiency of the prediction model, and thus improve the stability and feasibility of the prediction model in practical applications. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of a method for predicting the infrared radiation curve of an aero-engine according to an embodiment of the present invention. Figure 2 This is a schematic diagram of another method for predicting the infrared radiation curve of an aero-engine according to an embodiment of the present invention; Figure 3 This is a diagram of the temporal convolutional network structure. Figure 4 This is a diagram of a bidirectional long short-term memory neural network structure. Figure 5 This is a diagram of the long short-term memory neural network structure. Figure 6 For attention mechanism structure diagram; Figure 7 Performance comparison of the hybrid improved sand cat swarm optimization algorithm with other optimization algorithms; Figure 8 This is a schematic diagram of the prediction results of the infrared radiation curve of an aero-engine in an example of the present invention. Figure 9 This is a structural block diagram of an aircraft engine infrared radiation curve prediction device according to an embodiment of the present invention. Detailed Implementation

[0018] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Furthermore, it should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other. To make the objectives, technical solutions, and advantages of the present invention clearer, the present application will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] Figure 1 A flowchart illustrating a method for predicting the infrared radiation curve of an aero-engine according to an embodiment of this application is shown. (Refer to...) Figure 1 The method specifically includes the following steps: Step 101: Obtain a training dataset by simulating and calculating the infrared radiation characteristics of a typical aero-engine.

[0020] In one optional embodiment, the training dataset can be obtained by simulating the infrared radiation characteristics of a typical aero-engine. This can be achieved by simulating the infrared radiation characteristics of a typical aero-engine using the inverse Monte Carlo method.

[0021] In practical implementation, a typical aero-engine can be simulated and calculated. The infrared radiation characteristics of the band, and the training dataset obtained include the elevation angle of the detection point. Azimuth ,distance And the infrared radiation intensity value of the corresponding detection point.

[0022] Step 102: Sequentially build a temporal convolutional network layer, a bidirectional long short-term memory neural network layer, an attention mechanism layer, and a fully connected layer to generate a prediction model.

[0023] The TCN layer extracts the time-series information of the input feature sequence; the Bi-directional Long Short-Term Memory (BiLSTM) layer extracts the input feature sequence from both directions, capturing the dependency information in the bi-directional sequence; and the attention (AM) layer assigns different weights to each vector in the input feature sequence and highlights key features. The generated prediction model can be characterized as a TCN-BiLSTM-AM model.

[0024] In one optional embodiment, the temporal convolutional network layer, i.e., the TCN layer, includes: causal convolution, dilated convolution, and residual modules; the dilated convolution utilizes a spreading factor. Expanding the convolutional field of view, the output features after dilated convolution for: ;in, Given the input feature sequence, For filters, For filter size, As the expansion factor, when At times, dilated convolution and causal convolution have the same structure.

[0025] The residual module consists of two sets of causal dilated convolutional layers, a weight normalization layer, a ReLU activation function layer, and a Dropout layer, which are connected sequentially from bottom to top to obtain the output features. Among them, the ReLU (Rectified Linear Unit) layer is a commonly used non-linear activation function, and Dropout is a regularization technique used to prevent overfitting in deep neural networks. It increases the generalization ability of the model by randomly shutting down some neurons during training.

[0026] Bidirectional Long Short-Term Memory (BiLSTM) neural network layers process the input feature sequence ( During processing, a first hidden sequence is generated in the forward direction and a second hidden sequence is generated in the reverse direction; the first hidden sequence and the second hidden sequence are concatenated as the output sequence; wherein, the long short-term memory neural network layer includes an input gate, a forget gate and an output gate, which controls the flow of information to maintain and update memory over a long period of time.

[0027] The bidirectional long short-term memory neural network layer generates the output sequence using the following expression: The input feature sequence is: The first hidden sequence is The second hidden sequence is , This is the output sequence of a bidirectional long short-term memory neural network layer.

[0028] The basic units of a Long Short-Term Memory (LSTM) neural network mainly include an input gate, a forget gate, and an output gate. By controlling the flow of information, it enables the retention and updating of memories over a longer period. Its expression is: In the formula, and Let these represent the weight matrix and the bias, respectively. , and These are the input gate, the forget gate, and the output gate; and These are the unit state and the hidden state, respectively. and These represent the tanh and sigmoid activation functions, respectively.

[0029] The implementation formula for the attention mechanism layer can be: ; in, The value of the attention probability distribution is calculated from the output sequence of the bidirectional long short-term memory neural network layer. and These are the weighting coefficients. For attention layer output, This is the final output of the attention mechanism layer.

[0030] Step 103: Improve the sand cat swarm optimization algorithm by combining the variable spiral strategy and the sparrow warning mechanism.

[0031] The traditional sand cat swarm optimization (SCSO) algorithm simulates the behavior and strategies of sand cats when searching for and attacking prey, and is divided into a prey-searching phase, also known as the exploration phase, and a prey-attacking phase, also known as the development phase.

[0032] Hunting begins once the sand cat colony is initialized. The conversion factor is used... The decision is made between the search phase and the attack phase. The mathematical model is as follows: ;in, Let be the sensitivity coefficient of the sand cat population. Its mathematical model is: ;in, This is the current iteration. This is the hearing coefficient, which simulates hearing characteristics. It is generally set to 2, but other values ​​can be set according to the actual problem being solved.

[0033] when At that time, the sand cat colony entered the search phase. The mathematical model for updating the location of each sand cat is: ; in, For the first The optimal position of the sand cat in the population at the next iteration. For the first The sensitivity coefficient of a sand cat is mathematically modeled as follows: ; when At this point, the sand cat swarm enters its attack phase. The mathematical model for updating the location of each sand cat is: ; in, for Simulates the attack behavior of a sand cat by using random angles within a certain range.

[0034] For random positions, the mathematical model is:

[0035] The variable spiral strategy allows sand cat swarms to search for prey in a spiral pattern, expanding their ability to explore unknown areas and increasing the algorithm's ability to escape local optima. Its mathematical model is as follows: ;in, The maximum number of iterations, This represents the current iteration number.

[0036] After introducing the variable spiral strategy, the mathematical model for the position update of the sand cat swarm during the prey-hunting phase is as follows: The sparrow watch mechanism is used to improve the convergence speed of SCSO. Its mathematical model is as follows: ; in, It is the globally optimal position; , ,and These are the current, best, and worst fitness values, respectively; when This indicates that sand cats are near the edge of their population and vulnerable to natural predators; when When this happens, it means that the sand cats in the group are aware of the threat and need to get close to other sand cats to avoid being caught; This indicates the step control parameters and the direction in which the sparrow moves.

[0037] An alternative approach to improve the sand cat swarm optimization algorithm by combining a variable spiral strategy with a sparrow vigilance mechanism includes the following sub-steps: Sub-step 1: Initialize the population; Sub-step 2: Update the optimal solution, worst solution, and individual optimal solution of the population, and calculate the preset relevant parameters; Sub-step 3: Introduce a variable spiral strategy to update the sand cat's position during the prey-hunting phase; Sub-step 4: Introduce a sparrow alert mechanism to update the current location of the sand cat; Sub-step 5: Update the current optimal fitness value; Sub-step 6: Determine whether the constraints are met. If not, return to sub-step 2 to update the optimal solution, worst solution, and individual optimal solution of the population, and calculate the preset relevant parameters. If the constraints are met, determine that the improvement of the sand cat swarm optimization algorithm is completed, and output the improved sand cat swarm optimization algorithm (ISCSO).

[0038] Step 104: Optimize the initial learning rate, number of hidden layer nodes, and regularization coefficient of the prediction model using the improved sand cat swarm optimization algorithm and training dataset to determine the optimal hyperparameters and generate the trained prediction model.

[0039] Optionally, an improved sand cat swarm optimization algorithm and training dataset can be used to optimize the initial learning rate, number of hidden layer nodes, and regularization coefficient of the prediction model to determine the optimal hyperparameters for generating the trained prediction model. This can include the following sub-steps: Sub-step 1: Normalize the training dataset obtained from the simulation and divide it into a training set and a test set; In practice, the max-min normalization method can be used to normalize the data in the training dataset to [0, 1].

[0040] Sub-step 2: Establish the objective function model, where the objective function is the root mean square error; Sub-step 3: Optimize the initial learning rate, number of hidden layer nodes, and regularization coefficient in the prediction model using the improved Sand Cat Group Optimization Algorithm (ISCSO); select the optimal hyperparameters by judging the value of the fitness function or the maximum number of iterations. Sub-step 4: Establish the prediction model corresponding to each infrared band with the optimal hyperparameter combination, and generate the trained prediction model.

[0041] Since the prediction model (TCN-BiLSTM-AM model) is trained based on ISCSO, the trained prediction model can be called the ISCSO-TCN-BiLSTM-AM model.

[0042] Step 105: Use the trained prediction model to predict the infrared radiation characteristics of the aero-engine, and construct the infrared radiation characteristic curves based on the prediction results of different bands by interpolation.

[0043] The method for predicting the infrared radiation curve of an aero-engine provided in this invention involves simulating the infrared radiation characteristics of a typical aero-engine to obtain a training dataset; sequentially establishing a temporal convolutional network layer, a bidirectional long short-term memory neural network layer, an attention mechanism layer, and a fully connected layer to generate a prediction model; improving the sand cat swarm optimization algorithm by combining a variable spiral strategy and a sparrow warning mechanism; optimizing the initial learning rate, number of hidden layer nodes, and regularization coefficient of the prediction model using the improved sand cat swarm optimization algorithm and the training dataset to determine the optimal hyperparameters and generate the trained prediction model; using the trained prediction model to predict the infrared radiation characteristics of the aero-engine, and constructing the infrared radiation feature curve based on the prediction results of different bands through interpolation. Through this method, on the one hand, the temporal convolutional network layer can effectively extract the time-series information of the input features; on the other hand, the bidirectional long short-term memory neural network layer can extract sequence features from both directions, capturing the dependency information in the bidirectional sequence. Thirdly, the attention mechanism can highlight information at key historical time points and reduce the impact of redundant or useless information on the neural network. Fourthly, the sand cat swarm optimization algorithm, which is improved by combining the variable spiral strategy and the sparrow vigilance mechanism, has stronger search capabilities and faster convergence speed compared to the sand cat swarm optimization algorithm. It can more accurately optimize the hyperparameters of the model, improve the prediction accuracy and efficiency of the prediction model, and thus improve the stability and feasibility of the prediction model in practical applications.

[0044] The following is combined Figures 2-8 The method for predicting the infrared radiation curve of an aero-engine provided in this application is illustrated with a specific example.

[0045] This specific example uses the prediction of infrared radiation curves of aero-engines based on the ISCSO-TCN-BiLSTM-AM model as an illustration. Figure 2 The flowchart of an infrared radiation profile prediction method for aero-engines based on ISCSO-TCN-BiLSTM-AM is shown in this application. The method specifically includes the following steps: S1: Simulation calculation to construct an infrared radiation feature dataset.

[0046] In this step, a typical aero-engine can be simulated and calculated using the reverse Monte Carlo method. The infrared radiation characteristics of the band are used to obtain the training dataset, i.e., the infrared radiation characteristic dataset.

[0047] The infrared radiation characteristics of a typical axisymmetric nozzle were simulated and calculated in S1 using the inverse Monte Carlo method. This dataset is based on the elevation angle of the detection point. Azimuth ,distance One parameter is taken as input, and the infrared radiation intensity value of the corresponding band of the detection point is taken as output; among them, the elevation angle is taken as input. and azimuth The range is degrees, with intervals of 10 degrees; detection range is The interval is; the spectral range is ; spectral range is Spectral resolution is A total of 41 bands.

[0048] S2: Establish a prediction model.

[0049] The specific process of building a prediction model is as follows: Figure 2 As shown in the dashed box on the left, the prediction model is a network model. The construction of the network model includes: establishing a TCN layer, establishing a BiLSTM layer, introducing an AM layer into the model, and establishing a fully connected layer. The following is a detailed explanation of each layer.

[0050] 1) Temporal convolutional networks (TCNs) extract bidirectional time-series information from input features; TCN for input sequence Processing is performed. In TCN, dilated convolution utilizes a spreading factor. To expand the field of view of convolution. The output after dilated convolution. for: ; In the formula, , for filters, For filter size, is the expansion factor. When At times, dilated convolution and causal convolution have the same structure.

[0051] like Figure 3 As shown in the temporal convolutional network structure diagram, TCN uses a general residual module, which consists of two sets of dilated convolutional layers, weight normalization layers, ReLU activation function layers, and Dropout layers, connected sequentially from bottom to top to finally obtain the output features.

[0052] 2) The bidirectional long short-term memory (BiLSTM) neural network model extracts sequence features from both the front and back directions to capture the dependency information in the bidirectional sequence; like Figure 4The diagram shows the structure of a bidirectional long short-term memory (LSTM) neural network. The basic units of an LSTM network mainly include an input gate, a forget gate, and an output gate. By controlling the flow of information, it achieves the retention and updating of memories over a longer period. Its expression is: In the formula, and Let these represent the weight matrix and the bias, respectively. , and These are the input gate, the forget gate, and the output gate; and These are the unit state and the hidden state, respectively. and Do not represent the tanh and sigmoid activation functions.

[0053] like Figure 5 As shown in the diagram of the Long Short-Term Memory Neural Network (BiLSTM), the BiLSTM processes the input sequence... Processing is performed to generate hidden sequences in both the forward and reverse directions. The final output is the concatenation of the forward and reverse hidden outputs. Its expression is: ; In the formula, This is the output sequence of the BiLSTM layer.

[0054] 3) Introduce the attention mechanism (AM) into the model structure, assign different weights to the feature vectors, and highlight key features; like Figure 6 As shown in the diagram of the attention mechanism, the attention mechanism is used to filter out the information most important to the current task objective, highlighting important features. Its formula is: ; In the formula, The values ​​are the attention probability distributions calculated from the BiLSTM output vector. and These are the weighting coefficients. For attention layer output.

[0055] S3: The sand cat swarm optimization algorithm is improved by combining the variable spiral strategy and the sparrow warning mechanism. The initial learning rate, number of hidden layer nodes and regularization coefficient of the TCN-BiLSTM-AM model are optimized using ISCSO to obtain the optimal hyperparameters and input them.

[0056] The specific implementation process of S3 is as follows: Figure 2 As shown in the dashed box on the right.

[0057] Traditional sand cat swarm optimization algorithms simulate the behavior and strategies of sand cats when searching for and attacking prey. This is divided into two phases: searching for prey (exploration phase) and attacking prey (development phase).

[0058] Hunting begins once the sand cat colony is initialized. The conversion factor is used... The decision is made between the search phase and the attack phase. The mathematical model is as follows: ;in, Let be the sensitivity coefficient of the sand cat population. Its mathematical model is: ;in, This is the current iteration. This is the hearing coefficient, which simulates hearing characteristics. It is generally set to 2, but other values ​​can be set according to the actual problem being solved.

[0059] when At that time, the sand cat colony entered the search phase. The mathematical model for updating the location of each sand cat is: ; in, for Simulates the attack behavior of a sand cat by using random angles within a certain range. For random positions, the mathematical model is: ; The described variable spiral strategy allows sand cat swarms to search for prey in a spiral pattern, expanding their ability to explore unknown areas and increasing the algorithm's ability to escape local optima. Its mathematical model is as follows: ; in, The maximum number of iterations, This represents the current iteration number.

[0060] After introducing the variable spiral strategy, the mathematical model for the position update of the sand cat swarm during the prey-hunting phase is as follows: The aforementioned sparrow warning mechanism is used to improve the convergence speed of SCSO. Its mathematical model is as follows: ; in, It is the globally optimal position; , ,and These are the current, best, and worst fitness values, respectively; when This indicates that sand cats are near the edge of their population and vulnerable to natural predators; when This means that the sand cats in the group are aware of the threat and need to get close to other sand cats to avoid being caught; This indicates the step control parameters and the direction in which the sparrow moves.

[0061] The specific implementation of improving the sand cat swarm optimization algorithm by combining a variable spiral strategy and a sparrow vigilance mechanism may include the following sub-steps: Sub-step 1: Initialize the population; Sub-step 2: Update the population's optimal solution, worst solution, and individual optimal solution, and calculate relevant parameters. , , and ; Sub-step 3: Introduce a variable spiral strategy to update the sand cat's position during the prey-hunting phase; Sub-step 4: Introduce a sparrow alert mechanism to update the position of the sand cat in the current iteration; Sub-step 5: Update the current optimal fitness value; Sub-step 6: Determine whether the constraints are met. If not, return to sub-step 2; if met, output the result.

[0062] like Figure 7 The performance comparison of the hybrid improved Sand Cat Swarm Optimization (ICSSO) algorithm with other optimization algorithms is shown in the figure. The results of the test on the CEC-2022 function show that the ISCSO algorithm, which is improved by a hybrid variable spiral strategy and sparrow warning mechanism, has stronger search capabilities and faster convergence speed, and can more accurately optimize the hyperparameters of the model.

[0063] S4: Training the model.

[0064] In this step, the infrared radiation feature dataset constructed based on simulation calculations is used for model training to establish a prediction model for the corresponding infrared band based on the optimal parameter combination.

[0065] The process of training a model may include the following sub-steps: Sub-step 1: Divide the simulation dataset into training and testing sets, and normalize the data to [0, 1] using the max-min normalization method. The mathematical expression for this is: ;in, This is the actual value. and They are respectively The maximum and minimum values, This is the normalized value.

[0066] Sub-step 2: Establish the objective function model. The objective function is the root mean square error (RMSE), and its mathematical expression is: ; in, This is the actual value. For predicted values, This represents the number of samples.

[0067] Sub-step 3: Use ISCSO to optimize the initial learning rate, number of hidden layer nodes, and regularization coefficients in the TCN-BiLSTM-AM model; select the optimal parameter combination by judging the value of the fitness function or the maximum number of iterations.

[0068] Sub-step 4: Establish a prediction model for the corresponding infrared band based on the optimal parameter combination.

[0069] S5: Prediction of infrared radiation characteristics and construction of infrared radiation curves.

[0070] In this step, a combined model with optimal hyperparameters is used to predict the infrared radiation characteristics of the aero-engine. Then, based on the prediction results of different bands, infrared radiation characteristic curves are constructed by interpolation.

[0071] This application utilizes inverse Monte Carlo simulation to calculate an infrared spectral radiance dataset for a typical axisymmetric nozzle model. The prediction results obtained using this application's ISCSO-TCN-BiLSTM-AM and other reference models (BiLSTM, LSTM, BPNN) are as follows: Figure 8 As shown in the figure, the results indicate that the prediction results of this application are closer to the simulation results, proving that the method of this application has high prediction accuracy and maintains good performance in different bands, demonstrating the strong generalization ability of the method of this application. Furthermore, when local abrupt changes occur in the curve, the ISCSO-TCN-BiLSTM-AM model performs better than other models in fitting multiple local abrupt change sites of the curve, demonstrating a stronger ability to reconstruct the local features of the curve.

[0072] Figure 9 The structural block diagram of the aircraft engine infrared radiation curve prediction device according to the embodiments of this application is shown.

[0073] The aircraft engine infrared radiation curve prediction device provided in this application includes the following functional modules: The acquisition module 901 is used to obtain a training dataset by simulating and calculating the infrared radiation characteristics of a typical aero-engine. The model generation module 902 is used to sequentially build a temporal convolutional network layer, a bidirectional long short-term memory neural network layer, an attention mechanism layer, and a fully connected layer to generate a prediction model. The temporal convolutional network layer is used to extract the time-series information of the input feature sequence. The bidirectional long short-term memory neural network layer is used to extract the input feature sequence from both the preceding and following directions, capturing the dependency information in the bidirectional sequence. The attention mechanism layer is used to assign different weights to each vector in the input feature sequence and highlight key features. Improved module 903 is used to improve the sand cat swarm optimization algorithm by combining a variable spiral strategy and a sparrow warning mechanism; Training module 904 is used to optimize the initial learning rate, number of hidden layer nodes and regularization coefficient of the prediction model using the improved sand cat swarm optimization algorithm and the training dataset, and to determine the optimal hyperparameters to generate the trained prediction model. The prediction module 905 is used to predict the infrared radiation characteristics of the aero-engine using the trained prediction model, and to construct infrared radiation characteristic curves based on the prediction results of different bands by interpolation.

[0074] Optionally, the acquisition module is specifically used to: calculate the infrared radiation characteristics of a typical aero-engine through inverse Monte Carlo simulation to obtain a training dataset.

[0075] Optionally, the temporal convolutional network layer includes: causal convolution, dilated convolution, and residual modules; the dilated convolution utilizes a spreading factor. Expanding the convolutional field of view, the output features after dilated convolution for: ;in, Given the input feature sequence, For filters, For filter size, As the expansion factor, when At times, dilated convolution and causal convolution have the same structure; The residual module consists of two sets of causal dilated convolutional layers, weight normalization layers, ReLU activation function layers, and Dropout layers, which are connected sequentially from bottom to top to obtain the output features.

[0076] Optionally, when the bidirectional long short-term memory neural network layer processes the input feature sequence, it generates a first hidden sequence in the forward direction and a second hidden sequence in the reverse direction; the first hidden sequence and the second hidden sequence are concatenated as the output sequence. The long short-term memory neural network layer includes an input gate, a forget gate, and an output gate, which control the flow of information to maintain and update memories over a long period of time.

[0077] Optionally, the bidirectional long short-term memory neural network layer generates the output sequence using the following formula: The input feature sequence is: The first hidden sequence is The second hidden sequence is , This is the output sequence of a bidirectional long short-term memory neural network layer.

[0078] Optionally, the implementation formula for the attention mechanism layer is: ; in, The value of the attention probability distribution is calculated from the output sequence of the bidirectional long short-term memory neural network layer. and These are the weighting coefficients. For attention layer output, The final output of the attention mechanism layer.

[0079] Optionally, the improved module includes: The first submodule is used to initialize the population; The second submodule is used to update the optimal solution, worst solution, and individual optimal solution of the population, and to calculate preset relevant parameters; The third submodule is used to introduce a variable spiral strategy to update the position of the sand cat during the prey-hunting phase. The fourth submodule is used to introduce a sparrow alert mechanism and update the current location of the sand cat; The fifth submodule is used to update the current optimal fitness value; The sixth submodule is used to determine whether the constraints are met. If not, it returns to the second submodule to update the optimal solution, worst solution, and individual optimal solution of the population, and calculate the preset relevant parameters. If the constraints are met, it is determined that the improvement of the sand cat group optimization algorithm is completed, and the improved sand cat group optimization algorithm is output.

[0080] Optionally, the training module includes: The seventh submodule normalizes the training dataset obtained from the simulation and divides it into a training set and a test set. The eighth submodule establishes the objective function model, wherein the objective function is the root mean square error; The ninth submodule uses the improved sand cat swarm optimization algorithm to optimize the initial learning rate, number of hidden layer nodes, and regularization coefficient in the prediction model; it selects the optimal hyperparameters by judging the value of the fitness function or the maximum number of iterations. The tenth submodule establishes prediction models for each infrared band corresponding to the optimal hyperparameter combination, and generates the trained prediction models.

[0081] In the embodiments of this application Figure 9 The infrared radiation curve prediction device for aero-engines shown can be a device with an operating system. This operating system can be Android, iOS, or other possible operating systems; this application does not specifically limit its implementation.

[0082] The embodiments provided in this application Figure 9The infrared radiation curve prediction device for aero-engines shown can achieve Figure 1 The various processes implemented in the method implementation examples will not be described again here to avoid repetition.

[0083] Optionally, embodiments of this application also provide an electronic device, including a processor, a memory, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, they implement the processes performed by the aforementioned aircraft engine infrared radiation curve prediction device and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0084] It should be noted that the electronic device in this application embodiment includes the server described above.

[0085] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0086] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0087] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting the infrared radiation curve of an aero-engine, characterized in that, The method includes: A training dataset was obtained by simulating and calculating the infrared radiation characteristics of a typical aero-engine. A prediction model is generated by sequentially establishing a temporal convolutional network layer, a bidirectional long short-term memory neural network layer, an attention mechanism layer, and a fully connected layer. The temporal convolutional network layer is used to extract the time-series information of the input feature sequence. The bidirectional long short-term memory neural network layer is used to extract the input feature sequence from both the preceding and following directions, capturing the dependency information in the bidirectional sequence. The attention mechanism layer is used to assign different weights to each vector in the input feature sequence and highlight key features. An improved sand cat swarm optimization algorithm is proposed by combining a variable spiral strategy with a sparrow warning mechanism. The initial learning rate, number of hidden layer nodes, and regularization coefficient of the prediction model are optimized using an improved sand cat swarm optimization algorithm and the training dataset to determine the optimal hyperparameters and generate the trained prediction model. The trained prediction model is used to predict the infrared radiation characteristics of the aero-engine, and infrared radiation characteristic curves are constructed by interpolation based on the prediction results of different bands.

2. The method for predicting the infrared radiation curve of an aero-engine according to claim 1, characterized in that, The step of obtaining a training dataset by simulating the infrared radiation characteristics of a typical aero-engine includes: The infrared radiation characteristics of a typical aero-engine are calculated by simulating using the inverse Monte Carlo method to obtain a training dataset.

3. The method for predicting the infrared radiation curve of an aero-engine according to claim 1, characterized in that, The temporal convolutional network layer includes: causal convolution, dilated convolution, and residual modules; the dilated convolution utilizes a spreading factor. Expanding the convolutional field of view, the output features after dilated convolution for: ;in, Given the input feature sequence, For filters, For filter size, As the expansion factor, when At times, dilated convolution and causal convolution have the same structure; The residual module consists of two sets of causal dilated convolutional layers, weight normalization layers, ReLU activation function layers, and Dropout layers, which are connected sequentially from bottom to top to obtain the output features.

4. The method for predicting the infrared radiation curve of an aero-engine according to claim 1, characterized in that: When the bidirectional long short-term memory neural network layer processes the input feature sequence, it generates a first hidden sequence in the forward direction and a second hidden sequence in the reverse direction; the first hidden sequence and the second hidden sequence are concatenated as the output sequence. The long short-term memory neural network layer includes an input gate, a forget gate, and an output gate, which control the flow of information to maintain and update memories over a long period of time.

5. The method for predicting the infrared radiation curve of an aero-engine according to claim 4, characterized in that, The bidirectional long short-term memory neural network layer generates the output sequence using the following formula: The input feature sequence is: First hidden sequence The second hidden sequence is , This is the output sequence of a bidirectional long short-term memory neural network layer.

6. The method for predicting the infrared radiation curve of an aero-engine according to claim 1, characterized in that: The implementation formula for the attention mechanism layer is as follows: ; in, The value of the attention probability distribution is calculated from the output sequence of the bidirectional long short-term memory neural network layer. and These are the weighting coefficients. For attention layer output, This is the final output of the attention mechanism layer.

7. The method for predicting the infrared radiation curve of an aero-engine according to claim 1, characterized in that, The steps for improving the sand cat swarm optimization algorithm by combining a variable spiral strategy and a sparrow vigilance mechanism include: Initialize the population; Update the optimal solution, worst solution, and individual optimal solution of the population, and calculate the preset relevant parameters; A variable spiral strategy is introduced to update the sand cat's position during the prey-hunting phase; Introduce a sparrow alert mechanism to update the current location of the sand cat; Update the current optimal fitness value; Determine whether the constraints are met. If not, return to the steps of updating the optimal solution, worst solution, and individual optimal solution of the population, and calculating the preset relevant parameters. If the constraints are met, determine that the improvement of the sand cat group optimization algorithm is completed, and output the improved sand cat group optimization algorithm.

8. The method for predicting the infrared radiation curve of an aero-engine according to claim 1, characterized in that, The step of optimizing the initial learning rate, number of hidden layer nodes, and regularization coefficient of the prediction model using the improved sand cat swarm optimization algorithm and the training dataset to determine the optimal hyperparameters and generate the trained prediction model includes: The training dataset obtained from the simulation is normalized and divided into a training set and a test set. Establish an objective function model, wherein the objective function is the root mean square error; The improved sand cat swarm optimization algorithm is used to optimize the parameters of the prediction model, including the initial learning rate, the number of hidden layer nodes, and the regularization coefficient; the optimal hyperparameters are selected by judging the value of the fitness function or the maximum number of iterations. Establish prediction models for each infrared band corresponding to the optimal hyperparameter combination, and generate the trained prediction models.

9. A device for predicting the infrared radiation curve of an aero-engine, characterized in that, The device includes: The acquisition module is used to obtain a training dataset by simulating and calculating the infrared radiation characteristics of a typical aero-engine. The model generation module is used to sequentially build a temporal convolutional network layer, a bidirectional long short-term memory neural network layer, an attention mechanism layer, and a fully connected layer to generate a prediction model. The temporal convolutional network layer is used to extract the time-series information of the input feature sequence. The bidirectional long short-term memory neural network layer is used to extract the input feature sequence from both the preceding and following directions, capturing the dependency information in the bidirectional sequence. The attention mechanism layer is used to assign different weights to each vector in the input feature sequence and highlight key features. An improved module is used to improve the sand cat swarm optimization algorithm by combining a variable spiral strategy with a sparrow warning mechanism; The training module is used to optimize the initial learning rate, number of hidden layer nodes, and regularization coefficient of the prediction model using an improved sand cat swarm optimization algorithm and the training dataset, and to determine the optimal hyperparameters to generate the trained prediction model. The prediction module is used to predict the infrared radiation characteristics of the aero-engine using the trained prediction model, and to construct infrared radiation characteristic curves based on the prediction results of different bands through interpolation.

10. An electronic device, characterized in that, The electronic device includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions are executed by the processor to perform the steps of any one of the aircraft engine infrared radiation curve prediction methods according to claims 1-8.

Images