Aero-engine infrared radiation characteristic prediction method based on IPOA-BiLSTM-attention
By employing the IPOA-BiLSTM-attention method, combined with BiLSTM and an improved Pelican optimization algorithm, the problems of long prediction time and model instability in existing technologies for infrared radiation features are solved, achieving efficient and accurate prediction of infrared radiation features for aero-engines.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- AECC SICHUAN GAS TURBINE RES INST
- Filing Date
- 2024-10-28
- Publication Date
- 2026-04-28
AI Technical Summary
Existing technologies for predicting infrared radiation characteristics of aero-engines have excessively long computation times and unstable model performance, making it difficult to accurately reflect the nonlinear relationship between input and output. Furthermore, the reliance on experience in setting hyperparameters leads to poor application feasibility.
We employ an IPOA-BiLSTM-attention approach, which uses an inverse Monte Carlo method to acquire training data, combines a BiLSTM model to extract bidirectional sequence features, introduces an attention mechanism to filter key features, and utilizes an improved pelican optimization algorithm to optimize model hyperparameters, thereby improving prediction accuracy and efficiency.
It achieves fast and accurate prediction of infrared radiation intensity, improves the prediction accuracy and efficiency of the model, reduces the influence of redundant information, and enhances the hyperparameter optimization capability and convergence speed.
Smart Images
Figure CN121936249A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of infrared radiation feature prediction technology for aero-engines and artificial intelligence, specifically to a method for predicting infrared radiation features of aero-engines based on IPOA-BiLSTM-attention. Background Technology
[0002] The engine exhaust system contributes over 90% of the total infrared radiation of an aircraft. Developing infrared stealth technology for engine exhaust systems and reducing their infrared radiation characteristics is a crucial issue. With the rapid development of computer technology and computational mathematics, a series of numerical methods for calculating the spatial transmission of infrared radiation energy have emerged, such as the finite volume method, discrete transfer method, Monte Carlo method, and inverse Monte Carlo method. Infrared radiation intensity can be obtained through numerical simulation. However, if the mesh size is large, the computation time becomes excessively long. Furthermore, adjusting the solver's mesh size or related parameters further increases the time required, significantly reducing the efficiency of infrared stealth design. Therefore, a fast and accurate method for predicting infrared radiation intensity is needed.
[0003] With the development of big data and artificial intelligence technologies, data-driven modeling methods have been widely applied in the aerospace field. Machine learning methods can overcome the time-consuming nature of traditional numerical simulations and can learn the implicit nonlinear relationship between the prediction target and input variables. Traditionally, neural networks are used to learn the patterns of infrared radiation intensity for prediction. However, the infrared radiation intensity of aero-engines has 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 of the neural network model significantly affect model performance; however, in most case studies, these hyperparameters are often set empirically, leading to unstable model performance and poor practical application feasibility. Summary of the Invention
[0004] To address the shortcomings of the existing technologies mentioned above, this paper proposes a method for predicting the infrared radiation features of aero-engines based on IPOA-BiLSTM-attention, implemented using the following techniques:
[0005] An infrared radiation feature prediction method for aero-engines based on IPOA-BiLSTM-AM includes the following steps:
[0006] S1: Simulate and calculate the infrared radiation characteristics of a typical aero-engine using the inverse Monte Carlo method to obtain a training dataset;
[0007] S2: Establish a bidirectional long short-term memory (BiLSTM) neural network model to extract sequence features from both the front and back directions and capture the dependency information in the bidirectional sequence;
[0008] S3: Introduce the attention mechanism into the model structure, assign different weights to feature vectors, and highlight key features;
[0009] S4: Improve the pelican optimization algorithm (POA) by combining nonlinear inertia weighting factor and Cauchy mutation mechanism;
[0010] S5: The improved pelican optimization algorithm (IPOA) is used to optimize the initial learning rate, number of hidden layer nodes and regularization coefficient of the model, and the combined model with optimal hyperparameters is used to predict the infrared radiation characteristics of aero-engines.
[0011] 2. The method for predicting the infrared radiation characteristics of aero-engines based on IPOA-BiLSTM-attention as described in claim 1, characterized in that: the infrared radiation characteristics of a typical aero-engine in the 3-5µm band are calculated through simulation, and the data includes the elevation angle (θ) of the detection point. d ), azimuth Distance (r) d and the infrared radiation intensity of the detection point;
[0012] 3. The method for predicting infrared radiation features of aero-engines based on IPOA-BiLSTM-attention as described in claim 1, characterized in that: in step S2, the bidirectional long short-term memory neural network (BiLSTM) processes the input sequence (X=[x1,x2,…,x…)). n The process involves generating hidden sequences in both the forward and reverse directions. and The final output is the concatenation of the forward and reverse hidden outputs. Its expression is:
[0013]
[0014] In the formula, This is the output sequence of the BiLSTM layer.
[0015] In step S2, the basic unit of the Long Short-Term Memory (LSTM) neural network mainly includes 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 of time, as expressed in the following expression:
[0016] i t =σ(w ix x t +w ih h t-1 +b i )
[0017] f t =σ(w fx x t +w fh h t-1 +b f )
[0018]
[0019] o t =σ(w ox x t +w oh h t-1 +b o )
[0020]
[0021] In the formula, w and b represent the weight matrix and bias, respectively; i t f t and ot 分 Specifically, the input gate, forget gate, and output gate; c t and h t These are the unit state and the hidden state, respectively. σ and σ represent the tanh and sigmoid activation functions, respectively.
[0022] 4. The method for predicting infrared radiation features of aero-engines based on IPOA-BiLSTM-attention as described in claim 1, characterized in that: in step S3, the attention mechanism is used to filter out the information most important to the current mission objective, highlighting the role of important features. Its formula is:
[0023] e t =utanh(wh) t )
[0024]
[0025] In the formula, e tThe values of the attention probability distribution are calculated from the BiLSTM output vector, where u and w are weight coefficients, and s is the value of the attention probability distribution. t For attention layer output.
[0026] 5. The method for predicting the infrared radiation characteristics of aero-engines based on IPOA-BiLSTM-AM as described in claim 1, characterized in that: in step S4, the traditional pelican optimization algorithm simulates the behavior and strategies of pelicans during attack and hunting. This is divided into approaching the prey (exploration stage) and water-based flight (development stage).
[0027] The mathematical model for the pelican's approach to its prey phase is as follows:
[0028]
[0029] In the formula, Let p represent the new position of the i-th pelican in the j-th dimension, where I is a random number of 1 or 2. j Let F be the position of the prey in the j-th dimension. p Let it be the value of its objective function.
[0030] During the exploration phase, the mathematical model for updating the location is expressed as follows:
[0031]
[0032] In the formula, This represents the new state of the i-th pelican. This represents the objective function of the pelican during the exploration phase.
[0033] The mathematical model for the pelican's flight phase over the water is as follows:
[0034]
[0035] In the formula, Let R represent the new position of the i-th pelican in the j-th dimension based on stage 2, where R is a constant with a value of 0.2. R represents the radius of the neighborhood of each population member, in order to conduct a local search and converge to a better solution.
[0036] During the development phase, the mathematical model for its update location is represented by the following formula:
[0037]
[0038] In the formula: For the i-th Pelican's new position, Let i be the objective function for the i-th pelican in its new position after the second-stage update.
[0039] The aforementioned nonlinear inertia weighting factor is used to adjust the relationship between the pelican's updated position and its current position during the exploration phase, which can improve both local search capability and convergence speed. Its formula is:
[0040]
[0041] In the formula, T is the maximum number of iterations, and t is the current number of iterations.
[0042] After introducing a nonlinear inertia weighting factor, the mathematical model for the pelican approaching its prey during the predator phase is as follows:
[0043]
[0044] The Cauchy mutation strategy described above is used to increase the diversity of the pelican population during the development phase and improve the optimization capability of the POA.
[0045] After introducing the Cauchy mutation strategy, the mathematical model for the pelican's water-based flight phase is as follows:
[0046]
[0047] 6. The method for predicting the infrared radiation characteristics of an aero-engine based on IPOA-BiLSTM-AM as described in claim 1, characterized in that: the specific steps of step S5 are as follows:
[0048] S501: Initialize the population;
[0049] S502: Introduce a nonlinear inertia weighting factor to update the position of the pelican during the exploration phase;
[0050] S503: Introduces Cauchy mutation strategy to update the position of the Pelicans during the development phase;
[0051] S504: Update the current optimal fitness value;
[0052] S505: Determine whether the constraint conditions are met. If not, return to S502; if met, output the result.
[0053] Compared with the prior art, the present invention has the following beneficial effects:
[0054] 1) BiLSTM can extract sequence features from both the front and back directions, capturing dependency information in bidirectional sequences;
[0055] 2) The attention mechanism highlights information from key historical time points, reducing the impact of redundant or useless information on the neural network;
[0056] 3) The improved Pelican Optimization Algorithm (IPOA) by combining nonlinear inertia weighting factors with Cauchy mutation strategy has stronger search capabilities and faster convergence speed than the traditional Pelican Optimization Algorithm (POA). It can more accurately optimize the hyperparameters of the model, thereby further improving the prediction accuracy and efficiency of the model. Attached Figure Description
[0057] Figure 1 This is a schematic diagram of the method flow of the present invention;
[0058] Figure 2 This is a diagram of the structure of a bidirectional long short-term memory neural network (BiLSTM).
[0059] Figure 3 This is a diagram of the Long Short-Term Memory (LSTM) neural network structure.
[0060] Figure 4 A diagram illustrating the structure of the attention mechanism;
[0061] Figure 5 A performance comparison chart of the hybrid improved pelican optimization algorithm (IPOA) and the traditional pelican optimization algorithm (POA);
[0062] Figure 6 This is a diagram showing the predicted infrared radiation characteristics of an aero-engine, as an example of the present invention. Detailed Implementation
[0063] 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.
[0064] Figure 1 The flowchart of a method for predicting infrared radiation features of aero-engines based on IPOA-BiLSTM-attention, as described in this application, is shown below. Figure 1 The method specifically includes the following steps:
[0065] S1: Simulate and calculate the infrared radiation characteristics of a typical aero-engine in the 3-5µm band using the inverse Monte Carlo method to obtain a training dataset;
[0066] In S1:
[0067] The infrared radiation characteristics of a typical axisymmetric nozzle were calculated using the inverse Monte Carlo method. This dataset is based on the elevation angle (θ) of the detection point. d), azimuth and distance from the detection point (r) d The three parameters are used as inputs, and the infrared radiation intensity of the detection point is used as the output.
[0068] S2: Establish a bidirectional long short-term memory (BiLSTM) neural network model to extract sequence features from both the front and back directions and capture the dependency information in the bidirectional sequence;
[0069] In S2:
[0070] like Figure 2 As shown, the Bidirectional Long Short-Term Memory (BiLSTM) network processes the input sequence (X=[x1,x2,…,x…)). n The process involves generating hidden sequences in both the forward and reverse directions. and The final output is the concatenation of the forward and reverse hidden outputs. Its expression is:
[0071]
[0072] In the formula, This is the output sequence of the BiLSTM layer.
[0073] like Figure 3 As shown, 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 achieves the retention and updating of memories over a longer period of time. Its expression is:
[0074] i t =σ(w ix x t +w ih h t -1+b i )
[0075] f t =σ(w fx x t +w fh h t-1 +b f )
[0076]
[0077] o t =σ(w ox x t +w oh ht-1 +b o )
[0078]
[0079] In the formula, w and b represent the weight matrix and bias, respectively; i t f t and o t These are the input gate, forget gate, and output gate; c t and h t These are the unit state and the hidden state, respectively. σ and σ represent the tanh and sigmoid activation functions, respectively.
[0080] S3: Introduce the attention mechanism into the model structure, assign different weights to feature vectors, and highlight key features;
[0081] In S3:
[0082] like Figure 3 As shown, the attention mechanism is used to filter out the information most important to the current task objective, highlighting key features. Its formula is:
[0083] e t =utanh(wh) t )
[0084]
[0085] In the formula, e t The values of the attention probability distribution are calculated from the BiLSTM output vector, where u and w are weight coefficients, and s is the value of the attention probability distribution. t For attention layer output.
[0086] S4: Improve the pelican optimization algorithm (POA) by combining nonlinear inertia weighting factor and Cauchy mutation mechanism;
[0087] In S4:
[0088] Traditional pelican optimization algorithms simulate the behavior and strategies of pelicans during attack and hunting. They are divided into two phases: approaching prey (exploration phase) and flying over water (development phase).
[0089] The mathematical model for the pelican's approach to its prey phase is as follows:
[0090]
[0091] In the formula, Let p represent the new position of the i-th pelican in the j-th dimension, where I is a random number of 1 or 2.j Let F be the position of the prey in the j-th dimension. p Let it be the value of its objective function.
[0092] During the exploration phase, the mathematical model for updating the location is expressed as follows:
[0093]
[0094] In the formula, This represents the new state of the i-th pelican. This represents the objective function of the pelican during the exploration phase.
[0095] The mathematical model for the pelican's flight phase over the water is as follows:
[0096]
[0097] In the formula, Let R represent the new position of the i-th pelican in the j-th dimension based on stage 2, where R is a constant with a value of 0.2. R represents the radius of the neighborhood of each population member, in order to conduct a local search and converge to a better solution.
[0098] During the development phase, the mathematical model for its update location is represented by the following formula:
[0099]
[0100] In the formula: For the i-th Pelican's new position, Let i be the objective function for the i-th pelican in its new position after the second-stage update.
[0101] The aforementioned nonlinear inertia weighting factor is used to adjust the relationship between the pelican's updated position and its current position during the exploration phase, which can improve both local search capability and convergence speed. Its formula is:
[0102]
[0103] In the formula, T is the maximum number of iterations, and t is the current number of iterations.
[0104] After introducing a nonlinear inertia weighting factor, the mathematical model for the pelican approaching its prey during the predator phase is as follows:
[0105]
[0106] The Cauchy mutation strategy described above is used to increase the diversity of the pelican population during the development phase and improve the optimization capability of the POA.
[0107] After introducing the Cauchy mutation strategy, the mathematical model for the pelican's water-based flight phase is as follows:
[0108]
[0109] like Figure 5 As shown, the test results of IPOA and traditional POA on the CEC-2022 function, which are improved by a hybrid of nonlinear inertia weighting factor and Cauchy mutation strategy, show that IPOA has stronger search ability and faster convergence speed, and can more accurately optimize model hyperparameters.
[0110] S5: The improved pelican optimization algorithm (IPOA) is used to optimize the initial learning rate, number of hidden layer nodes and regularization coefficient of the model, and the combined model with optimal hyperparameters is used to predict the infrared radiation characteristics of aero-engines.
[0111] In S5:
[0112] The specific steps are as follows:
[0113] S501: Initialize the population;
[0114] S502: Introduce a nonlinear inertia weighting factor to update the position of the pelican during the exploration phase;
[0115] S503: Introduced the Cauchy mutation strategy to update the position of the Pelican during the development phase;
[0116] S504: Update the current optimal fitness value;
[0117] S505: Determine whether the constraint conditions are met. If not, return to S502; if met, output the result.
[0118] In S5:
[0119] This application utilizes an infrared spectral radiation intensity dataset of a typical axisymmetric nozzle model based on inverse Monte Carlo simulation. The prediction results of the IPOA-BiLSTM-attention model in this application are then analyzed. Figure 6 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 can maintain good performance in different bands, demonstrating the strong reliability of the method of this application.
[0120] 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.
[0121] 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 infrared radiation features of aero-engines based on IPOA-BiLSTM-attention, characterized by: This can be achieved through the following steps: S1: Obtain a training dataset by simulating and calculating the infrared radiation characteristics of a typical aero-engine; S2: Establish a bidirectional long short-term memory (BiLSTM) neural network model to extract sequence features from both the front and back directions and capture the dependency information in the bidirectional sequence; S3: Introduce the attention mechanism into the model structure, assign different weights to feature vectors, and highlight key features; S4: Improve the pelican optimization algorithm (POA) by combining nonlinear inertia weighting factor and Cauchy mutation mechanism; S5: The improved pelican optimization algorithm (IPOA) is used to optimize the initial learning rate, number of hidden layer nodes and regularization coefficient of the model, and the combined model with optimal hyperparameters is used to predict the infrared radiation characteristics of aero-engines.
2. The method for predicting infrared radiation features of aero-engines based on IPOA-BiLSTM-attention as described in claim 1, characterized in that: The infrared radiation characteristics of a typical aero-engine are calculated through simulation.
3. The method for predicting infrared radiation features of aero-engines based on IPOA-BiLSTM-attention as described in claim 1, characterized in that: In step S2, the bidirectional long short-term memory neural network (BiLSTM) processes the input sequence (X = [x1, x2, ..., x...]). n The process involves generating hidden sequences in both the forward and reverse directions. and 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. In step S2, the basic unit of a long short-term memory (LSTM) neural network mainly includes 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 of time, as expressed in the following expression: i t =σ(w ix x t +w ih h t-1 +b i ) f t =σ(w fx x t +w fh h t-1 +b f ) o t =σ(w ox x t +w oh h t-1 +b o ) In the formula, w and b represent the weight matrix and bias, respectively; i t f t and o t These are the input gate, forget gate, and output gate; c t and h t These are the unit state and the hidden state, respectively. σ and σ represent the tanh and sigmoid activation functions, respectively.
4. The method for predicting infrared radiation features of aero-engines based on IPOA-BiLSTM-attention as described in claim 1, characterized in that: In step S3, the attention mechanism is used to filter out the information most important to the current task objective, highlighting key features. Its formula is: yes t =utanh(wh t ) In the formula, e t The values of the attention probability distribution are calculated from the BiLSTM output vector, where u and w are weight coefficients, and s is the value of the attention probability distribution. t For attention layer output, y attention This is the final output.
5. The method for predicting infrared radiation features of aero-engines based on IPOA-BiLSTM-attention as described in claim 1, characterized in that: In step S4, the Mandarin Duck optimization algorithm simulates the behavior and strategies of the Mandarin Duck during attack and hunting. This is divided into two phases: approaching the prey (exploration phase) and flying over water (development phase). The mathematical model for the approaching prey phase of the mandarin duck is as follows: In the formula, Let p represent the new position of the i-th eagle in the j-th dimension, where I is a random number of 1 or 2. j Let F be the position of the prey in the j-th dimension. p Let it be the value of its objective function. During the exploration phase, the mathematical model for updating the location is expressed as follows: In the formula, This represents the new state of the i-th pelican. This represents the objective function of the pelican during the exploration phase. The mathematical model for the pelican's flight phase over the water is as follows: In the formula, Let R represent the new position of the i-th pelican in the j-th dimension based on stage 2, where R is a constant with a value of 0.
2. R represents the radius of the neighborhood of each population member, in order to conduct a local search and converge to a better solution. During the development phase, the mathematical model for its update location is represented by the following formula: In the formula: For the i-th Pelican's new position, Let i be the objective function for the i-th pelican in its new position after the second-stage update. The aforementioned nonlinear inertia weighting factor is used to adjust the relationship between the pelican's updated position and its current position during the exploration phase, which can improve both local search capability and convergence speed. Its formula is: In the formula, T is the maximum number of iterations, and t is the current number of iterations. After introducing a nonlinear inertia weighting factor, the mathematical model for the pelican approaching its prey during the predator phase is as follows: The Cauchy mutation strategy described above is used to increase the diversity of the pelican population during the development phase and improve the optimization capability of the POA. After introducing the Cauchy mutation strategy, the mathematical model for the pelican's water-based flight phase is as follows:
6. The method for predicting infrared radiation features of aero-engines based on IPOA-BiLSTM-attention as described in claim 1, characterized in that: The specific steps of step S5 are as follows: S501: Initialize the population; S502: Introduce a nonlinear inertia weighting factor to update the position of the pelican during the exploration phase; S503: Introduces Cauchy mutation strategy to update the position of the Pelicans during the development phase; S504: Update the current optimal fitness value; S505: Determine whether the constraint conditions are met. If not, return to S502; if met, output the result.